The Cognitive Penetrability of Perception: New Philosophical Perspectives [Hardcover ed.] 0198738919, 9780198738916

Citation preview

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

The Cognitive Penetrability of Perception

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

The Cognitive Penetrability of Perception New Philosophical Perspectives

edited by

John Zeimbekis and Athanassios Raftopoulos

1 i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

3

Great Clarendon Street, Oxford, OX DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © the several contributors  The moral rights of the authors have been asserted First Edition published in  Impression:  All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press  Madison Avenue, New York, NY , United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number:  ISBN –––– Printed and bound by CPI Group (UK) Ltd, Croydon, CR YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

This book is for Fred Dretske and Jonathan Lowe

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

Preface and Acknowledgements The idea for this book emerged after a book symposium on Athanassios Raftopoulos’s Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? organized by John Zeimbekis at the University of Grenoble in June . Most papers at the symposium argued for or against the cognitive penetrability of perception. For some time, it had been evident that how cognitive states influence perception affects the outcomes of debates on high-level perceptual contents, the nature of perceptual warrant, and issues surrounding non-conceptual content, cognitive phenomenology, and the phenomenology of late vision. This was a new set of questions compared to those posed by the computationalist theories of mind that originally formulated hypotheses about penetrability and impenetrability. The new questions were connected to the old ones, and many of the original problems were still being discussed, especially modularity and the question of how cognitive architecture affects issues in epistemology. At the same time, the number of philosophers working on cognitive penetrability was getting larger and more diversified. The time seemed ripe for a book to connect the multiple strands of the cognitive penetrability debate—the recent philosophical developments, the psychological findings, the computationalist background, and the epistemological consequences of cognitive architecture—by gathering papers from as many as possible of the most important players in the field. In contacting potential contributors, we sought to cover as many areas as possible in which the ramifications of cognitive penetrability were being discussed, in order to properly represent the emerging philosophical agenda of the debate. During the time it took us to put together the final version, the project was overshadowed by the loss of Fred Dretske and Jonathan Lowe. This book is dedicated to them. Fred Dretske immediately responded to our request to contribute a chapter, sending us his contribution to the book early on and subsequently revising it. Jonathan Lowe was among the original Grenoble participants and an enthusiastic supporter of the project; his contribution to this volume is a revised version of the paper he presented at that meeting. We are very fortunate that this book includes a chapter on the perception/cognition distinction by Fred Dretske, and a chapter on perception, realism, and the individuation of objects by Jonathan Lowe. We would like to thank the contributors for sending us original material, even before it was clear that the book would come to fruition; Peter Momtchiloff for his encouragement and handling of the project; and the readers at Oxford University Press for their thought-provoking and constructive comments. Thanks are due to the Department of Philosophy of the University of Grenoble for hosting the original workshop; to the French National Research Foundation (CNRS) and the Rhône-Alpes

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

viii preface and acknowledgements regional authority for funding at early stages of the project; and to our wives, Alex and Ali, for putting up with us throughout and in general. The editors and publisher gratefully acknowledge the permission granted to reproduce the copyright material in this book. Every effort has been made to trace copyright holders and to obtain their permission for the use of copyright material. The publisher apologizes for any errors or omissions and would be grateful if notified of any corrections that should be incorporated in future reprints or editions of this book.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

Contents List of Figures List of Contributors The Cognitive Penetrability of Perception: An Overview Athanassios Raftopoulos and John Zeimbekis

Part I. Definitional and Methodological Issues

xi xiii 



. Cognitive Penetrability: A No-Progress Report Edouard Machery



. Towards a Consequentialist Understanding of Cognitive Penetration Dustin Stokes



Part II. Modularity, Encapsulation, and Impenetrability



. Unencapsulated Modules and Perceptual Judgment Jack C. Lyons



. Perceptual Integration, Modularity, and Cognitive Penetration Daniel C. Burnston and Jonathan Cohen



. Multisensory Perception and Cognitive Penetration: The Unity Assumption, Thirty Years After Ophelia Deroy

Part III. Substantive Impenetrability and Penetrability Claims





. Perception versus Conception: The Goldilocks Test Fred Dretske



. Cognitive Penetration and the Reach of Phenomenal Content Robert Briscoe



. Cognitive Penetration of the Dorsal Visual Stream? Brad Mahon and Wayne Wu



. Attention and Cognitive Penetration Christopher Mole



i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

x contents

Part IV. Cognitive Penetrability and the Phenomenology of Perception . ‘Looks the Same but Feels Different’: A Metacognitive Approach to Cognitive Penetrability Jérôme Dokic and Jean-Rémy Martin

 

. Cognitive Penetrability and Consciousness Athanassios Raftopoulos



. Seeing, Visualizing, and Believing: Pictures and Cognitive Penetration John Zeimbekis



Part V. Cognitive Penetrability and Nonconceptual Content



. Cognitive Penetration and Nonconceptual Content Fiona Macpherson



Part VI. Cognitive Penetrability and Realism



. Perceptual Content, Cognitive Penetrability, and Realism Jonathan Lowe . Cognitive (Im)Penetrability of Vision: Restricting Vision versus Restricting Cognition Costas Pagondiotis





Afterword: Epistemic Evaluability and Perceptual Farce Susanna Siegel



Author Index Subject Index

 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

List of Figures . A shared assumption: modules correspond to non-overlapping encapsulated processing units, and their outputs feed directly into the central system



. Two ways in which multisensory audiovisual processing can be reconciled with modularity



. Triangulation



. Duck/rabbit



. Figure–ground ambiguity



. Adapting texture density versus adapting numerosity. Reproduced, with permission, from Durgin ()



. Faces and their corresponding anti-faces. Reproduced, with permission, from Leopold et al. ()



. Category task paradigm, illustration of the stimuli and task conditions. Originally published as fig.  in Kravitz and Behrmann (). Reproduced with kind permission from Springer Science and Business Media



. Visually driven picture perception



. J. M. W. Turner, Approach to Venice, oil on canvas, ; detail. Courtesy National Gallery of Art, Washington, DC



. Opaque cube



. Ambiguous experience of volume



. John Pugh, Siete Punto Uno. Reproduced by kind permission of the artist: artofjohnpugh.com



. Topology of two-dimensional surfaces



i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

List of Contributors Robert Briscoe is Associate Professor of Philosophy at Ohio University. His research focuses on topics in the philosophy of cognitive science, perception, and mind. Special interests include the role of action in perception, spatial representation, mental imagery, depiction, and pictorial experience. Daniel C. Burnston is a PhD candidate in Philosophy and the Interdisciplinary Cognitive Sciences program at the University of California, San Diego. His work focuses on issues surrounding functional decomposition and explanation in cognitive science, neuroscience, and biology. Jonathan Cohen is Professor in the Department of Philosophy at the University of California, San Diego. He is the author of The Red and the Real: An Essay on Color Ontology (Oxford University Press, ). Though he has published in many areas of philosophy, his principal interests these days are in philosophy of perception and philosophy of language, and especially in questions in these areas that interact with the cognitive sciences. Ophelia Deroy is the associate director of the Institute of Philosophy at the University of London and a researcher at the Centre for the Study of the Senses. She specializes in philosophy of mind and cognitive neurosciences, and collaborates with scientists in Oxford, Paris, and London. She has published widely on issues related to multisensory perception, sensory deficits, and synaesthesia. Jérôme Dokic is Professor of Cognitive Philosophy at the École des hautes études en sciences sociales and a member of the Institut Jean-Nicod in Paris. He has published many essays on perception, memory, imagination, and more recently epistemic or noetic feelings and metacognition. Fred Dretske (–) was Professor Emeritus in Philosophy at Stanford University and Senior Research Scholar in Philosophy at Duke University. He was the author of Seeing and Knowing (University of Chicago Press, ), Knowledge and the Flow of Information (MIT Press, ), Explaining Behavior: Reasons in a World of Causes (MIT Press, ), and Naturalizing the Mind (MIT Press, ). Jonathan Lowe (–) was Professor of Philosophy at Durham University. He was the author of Kinds of Being: A Study of Individuation, Identity and the Logic of Sortal Terms (Blackwell, ), Subjects of Experience (Cambridge University Press, ), The Possibility of Metaphysics (Oxford University Press, ), Locke (Routledge, ), The Four-Category Ontology: A Metaphysical Foundation for Natural Science

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

xiv

list of contributors

(Oxford University Presss, ), Personal Agency (Oxford University Press, ) and Forms of Thought: A Study in Philosophical Logic (Cambridge University Press, ). Jack C. Lyons is Professor of Philosophy at the University of Arkansas. His research is mainly in epistemology, cognitive science, and philosophy of mind. He has published a recent book with Oxford University Press, entitled Perception and Basic Beliefs, and is editor of the journal Philosophical Topics and an associate editor of Episteme: A Journal of Individual and Social Epistemology. Edouard Machery is Professor in the Department of History and Philosophy of Science at the University of Pittsburgh. He is the author of Doing Without Concepts (Oxford University Press, ) as well as the editor of The Oxford Handbook of Compositionality (Oxford University Press, ), La philosophie expérimentale (Vuibert, ), Arguing about Human Nature (Routledge, ), and Current Controversies in Experimental Philosophy (Routledge, ). Fiona Macpherson is Professor and Head of Philosophy and Director of the Centre of the Study of Perceptual Experience at the University of Glasgow. She is CoDirector of the Centre for the Study of the Senses, Institute of Philosophy, University of London. She has been Visiting Professor at the Universities of London and Umeå, and a Research Fellow at the Australian National University, Rosamund Chambers Research Fellow at Girton College, Cambridge, and a Research Fellow at Harvard. Brad Mahon is an Assistant Professor in the Departments of Brain and Cognitive Sciences and Neurosurgery at the University of Rochester, and faculty in the Center for Language Sciences and Center for Visual Sciences. He received his BS () and PhD () from Harvard University. His research uses functional neuroimaging to study how vision and language are organized in the brain and how the brain reorganizes after damage. Jean-Rémy Martin has a PhD in Cognitive Science (UPMC, Paris) obtained at the Institut Jean-Nicod (IEC-ENS, Paris, France). He is currently a postdoctoral fellow at the University of Sussex. His research and related publications cover the areas of cognitive psychology, cognitive psychopathology (particularly with respect to schizophrenia), and philosophy of perception. He is currently working on hypnosis from a theoretical and empirical point of view. Christopher Mole teaches in the Department of Philosophy, and in the Programme in Cognitive Systems, at the University of British Columbia. He is the author of Attention Is Cognitive Unison (Oxford University Press, ), and co-editor, with Declan Smithies and Wayne Wu, of Attention: Philosophical and Psychological Essays (Oxford University Press, ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

list of contributors

xv

Costas Pagondiotis is Assistant Professor in Philosophy of Mind at the University of Patras, Greece. His publications and research are mainly in the areas of philosophy of mind, philosophy of cognitive science, and contemporary epistemology. He is currently writing a book on perception as embodied and embedded in which he defends a version of direct realism and the conceptuality of perceptual experience. Athanassios Raftopoulos is Professor of Epistemology and Cognitive Science in the Department of Psychology at the University of Cyprus. He is the author of Cognition and Perception: How Do Psychology and Neural Sciences Inform Philosophy? (MIT Press, ), and co-author with Philippos Kargopoulos of The Science of Logic and the Art of Thinking (Equinox, ). He is editor of Cognitive Penetrability of Perception: Attention, Action, Planning, and Bottom-up Constraints (Nova Science, ), co-editor with Peter Machamer of Perception, Realism, and the Problem of Reference (Cambridge University Press, ), and co-editor with Andreas Demetriou of Cognitive Developmental Change: Theories, Models and Measurement (Cambridge University Press, ). He has published over  papers on the philosophy of science, cognitive science, perception, epistemology, and the philosophy of mind. Susanna Siegel is Edgar Pierce Professor of Philosophy at Harvard University. She is author of The Contents of Visual Experience (Oxford University Press, ) and of numerous articles in the philosophy of perception and epistemology. Dustin Stokes is Assistant Professor in the Department of Philosophy at the University of Utah. He works generally in the areas of philosophy of mind and cognitive science. His current research concerns the cognitive penetrability of perceptual experience, relations between sense modalities, and the role of imagination and imagery in practical and theoretical reasoning. He also works on the cognition and perception of artworks. Wayne Wu is Associate Professor in and Associate Director of the Center for the Neural Basis of Cognition, Carnegie Mellon University. He works on various topics including different aspects of attention, agency, schizophrenia, consciousness, and perception, by integrating the perspectives of philosophy and cognitive science. John Zeimbekis is Assistant Professor of Philosophy at the University of Patras. He works on the philosophy of perception, especially demonstrative thought, the metaphysics of qualities, and the relations between thought, perception, imagery, memory, and pictures. He has published papers on perception in Noûs and Philosophical Studies. He also works on topics in aesthetics and is the author of a book on aesthetic value, Qu’est-ce qu’un jugement esthétique? (Vrin, ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

The Cognitive Penetrability of Perception An Overview Athanassios Raftopoulos and John Zeimbekis

The hypothesis that perception is cognitively penetrable holds that cognitive states such as beliefs, desires, and possibly other states can causally influence perceptual processing in such a way that they end up determining subjects’ perceptual contents or experiences. The philosophical significance of penetrability is easy to grasp: if perception is cognitively penetrable, then what we think literally affects how we see the world. This book elucidates the nature of the cognitive penetrability hypothesis, assesses its plausibility, and explores what the philosophical consequences would be if perception turned out to be cognitively penetrable or cognitively impenetrable. Section 1 provides a brief outline of the origins and history of the cognitive penetrability debate. Section  disambiguates key concepts and explains their roles in the debate: impenetrability, modularity, and informational encapsulation; early vision and late vision; hard-wired perceptual formation principles; perceptual learning; theory-ladenness of observation; the role of attention in cognitive penetration; and the perception/cognition distinction itself. Section  presents existing definitions of the cognitive penetrability of perception. Section  explains the relevance of the penetrability debate to a range of philosophical topics: the contents of perception; cognitive phenomenology and epistemic feelings; nonconceptual content; awareness and attention; perceptual warrant; realism and representationalism; and action. The contributed chapters are summarized in the sections of the introduction that deal with the corresponding topics.2 1 We cross-refer to sections within the Introduction by e.g. ‘Section ’ or simply ‘see ’. Where we refer to a section in another chapter in the volume, we use the symbol ‘§’. 2 We often abbreviate ‘cognitive penetrability of perception’ and ‘cognitive impenetrability of perception’ to ‘penetrability’ and ‘impenetrability’, and sometimes to ‘CP’ and ‘CI’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

 Cognitive Penetrability: History and Recent Developments . The computationalist background The concepts of cognitive penetrability and impenetrability were originally deployed in the framework of a computationalist theory of mind by Zenon Pylyshyn (; ) and Jerry Fodor (). Pylyshyn and Fodor used the concepts to define the limits of a computationalist account of how the brain represents and manipulates the contents of propositional attitudes. In that framework, cognitively penetrable states and processes are those sensitive to the information involved in computations that perform thought and reasoning. Cognitively impenetrable processes either involve different computations, which are not sensitive to those that perform thought, or else are outside the scope of the computationalist approach to mind. As part of the hypothesis that perception is cognitively impenetrable, Pylyshyn (; ) sought to isolate a set of early perceptual processes from thought and to describe them as distinct items in a hierarchical cognitive architecture. This was the first systematic empirical hypothesis to the effect that perception is cognitively impenetrable. Specifically, it claimed that part of visual perception was cognitively impenetrable. Pylyshyn calls this part of vision ‘early vision’. It includes the set of visual processes from stimulus onset up to the construction of viewer-centred (egocentric) volumetric representations of object surfaces. At this stage of vision, according to Pylyshyn, the scene should also be segmented into primitive visual objects, usually called ‘proto-objects’ (see Pylyshyn and Storm ; Pylyshyn ). Simultaneously, Pylyshyn offered a precisely formulated hypothesis about the penetrability of visual perception: he held that the remainder of vision—‘late vision’, which includes object recognition and identification—is performed jointly with contributions from longterm memories, semantic information (such as conceptually encoded information about kinds), agentively focused attention, and even conscious hypothesizing by the perceiving subject. For more information about Pylyshyn’s accounts of early and late vision, and a comparison with David Marr’s account of vision, see Section .. Fodor’s hypothesis that perception is modular is closely related to Pylyshyn’s hypothesis that perception is cognitively impenetrable. Fodor (: –) uses the concept of cognitive penetrablility to describe the kinds of influences on perception that the modularity hypothesis denies, and Pylyshyn (: ch. ) uses Fodor’s concept of modularity to describe impenetrable visual processes. Fodor’s modules are information-processing brain mechanisms that perform circumscribed, partial tasks on limited and exclusive ranges of inputs. The role of perceptual modules is to yield outputs in a suitable format for use by cognitive centres by drawing only on perceptual inputs and perceptual processing resources. For example, Fodor favours a model for vision in which ‘the final stage of visual input analysis involves accessing a “form-concept” dictionary which, in effect, pairs D sketches with basic categories’ (: ). On such a model, brain representations of three-dimensional objects are

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



generated from retinal stimulations by vision alone, without any help from memories or semantic information (such as conceptually encoded information associating generic shapes with kinds). Fodor enumerates a number of properties to describe modular systems, but a system cannot be modular unless it is informationally encapsulated (Fodor : ; : ). For a computational system to be encapsulated, its informational resources have to be restricted to its proprietary database of information. This amounts to an absence of informational exchange between systems: ‘X is informationally encapsulated from Y if and only if X cannot use information from Y in its computations’ (Mahon and Wu, Chapter  of this volume). If a perceptual information-processing system is modular in Fodor’s sense, it will be informationally encapsulated from other systems, including cognitive ones. It will not be sensitive to the information involved in the computations that perform thought and reasoning, and will thus be cognitively impenetrable. However, the cognitive impenetrability of a perceptual module does not imply its informational encapsulation. On the differences between cognitive impenetrability and perceptual modularity, see section . on encapsulation, and section . on early vision.

. Cognitive impenetrability and New Look psychology In addition to their positive role in computationalist accounts of cognitive architecture, the concepts of impenetrability and modularity were enlisted for criticism of competing views of the mind based on experimental findings in the ‘new look in perception’ movement in psychology. Those findings were interpreted as evidence that ‘all perceptual experience is necessarily the end product of a categorization process’, that ‘perception is a process of categorization in which organisms move inferentially from cues to category identity’ (Bruner and Goodman ), and that ‘the perceptual effect of a stimulus is necessarily dependent upon the set or expectancy of the organism’ (Bruner and Postman , cited by Mole, Chapter ). Pylyshyn, in turn, held that the view of the mind offered by Bruner and Goodman meant that ‘values and needs determine how we perceive the world, down to the lowest levels of the visual system’ (: ; our emphasis). A programmatic statement of the New Look movement and its views on perception can be found in Bruner and Goodman (). For detailed overviews, see Gregory (; ) and Rock (). In this volume, see the contributions by Machery and Mole (Chapters  and ). A classic illustration of the relevant findings is provided by Bruner and Postman () (for detailed discussion see Mole, Chapter ). In Bruner and Postman’s experiment, when subjects who expected to see items belonging to certain categories (playing cards) were presented with the objects at sufficiently short intervals, they reported that the objects had the expected colours or a compromise between the expected and actual colour, even when the objects had anomalous colours (e.g. a red six of spades). Bruner and Postman explained the reports by claiming that expectations influence perception of the stimulus. If the expectations were determined by conceptual memory colours,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

then conceptually encoded information or beliefs causally influence the mechanisms that generate perceived colour. Comparable findings on colour perception were reported by Delk and Fillenbaum (), and recently by Hansen et al. () and Olkkonen et al. (); Hansen et al. concluded that their ‘results show a high-level cognitive effect on low-level perceptual mechanisms’. There are many similar findings in other areas such as auditory recognition of sentences, visual recognition of words, or dependence of size perception on value. If the interpretations are correct, the findings meet a condition for cognitive penetration of early vision given by Pylyshyn: ‘if a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, i.e., it can be altered in a way that bears some logical relation to what the person knows’ (: ). Pylyshyn (; ) used the CI hypothesis to argue against such interpretations of the experiments. He presented clinical findings (such as Humphreys and Riddoch’s  work on visual agnosia) which suggested the dissociation of part of visual perception from cognition. He adduced psychophysical evidence (Sperling and Melchner ; Tanner and Swets ) that early vision resolves ambiguities in the scene autonomously from rational inference, arguing that many putative cases of cognitive influence on visual interpretation of stimuli were in fact hard-wired visual constraints impenetrable to cognitive states. Finally, he sought to show that ‘many apparent examples of cognitive effects in vision arise either from a postperceptual decision process or from a preperceptual attention-allocation process’ (Pylyshyn : ).

. Cognitive architecture and epistemology A third strand of the original cognitive penetrability debate, on which Fodor in particular focused, are the consequences of hypotheses about cognitive architecture for epistemology. New Look accounts of the nature of perception were used by philosophers of science to give a psychological defence of the idea that scientific observation is theory-laden (Hanson ; Kuhn ; Feyerabend ). (See Mole, Chapter : Kuhn used Bruner and Postman’s  study, while Hanson drew on the phenomenology of perceiving ambiguous figures, which was popular at the time thanks to work by Gestalt psychologists and to Wittgenstein’s influence.) Duhem () had argued that observations are framework-relative, and that observational data are never uninterpreted, theory-neutral descriptions of events; to this, Hanson, Kuhn, and Feyerabend added that what we see is already an interpretation of incoming information based on the perceiver’s theoretical commitments and conceptual frameworks—in other words, that perception is cognitively penetrated. This allowed them to challenge the view that scientific theories can be empirically tested and compared to other theories, since it left no theory-neutral ground on which to make a rational choice among alternative theories based on experimental outcome, making scientific theories potentially incommensurable. This outcome has been defended by constructivists in the philosophy of science who deny that scientific theories relate

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



observers to mind-independent objects (see Kukla ; Kitcher ). On the other hand, Brown (), Brewer and Lambert (; ), and Kitcher () hold that perception is theory-laden, but simultaneously seek to alleviate the consequences of conceptual relativism by arguing that strong bottom-up components of perception provide empirical constraints sufficient to prevent constructivism. Fodor dealt directly with the epistemological consequences of choosing a theory of cognitive architecture. Paul Churchland () had argued that observation is theoryladen by relying on new findings about the plasticity of perception from cognitive neuroscience, connectionism, and vision science. In seeking to rebut Churchland’s claims, Fodor explicitly stated what he took to be the epistemological relevance of the impenetrability and encapsulation hypotheses: ‘The question at issue is: What are the psychological conditions under which differences among the theories that observers hold are not impediments to perceptual consensus among observers. Cognitive encapsulation seems to be an empirically necessary condition for this’ (Fodor : ). Like Pylyshyn, Fodor used the thesis (inspired by Marr ; see section .) that putative instances of theory-ladenness were in fact hard-wired modular constraints on visual processing: perceptual modules apply general assumptions inherent to their functional architecture to transform proximal stimuli into brain representations which can be exploited cognitively. For observation to be theory-laden, Fodor pointed out, perceptual processing would have to access the contents of propositional attitudes, and this appears to be refuted by the impermeability of perceptual illusions to the contents of beliefs. Two points made by Churchland () in reply to Fodor were to prove particularly resilient: (a) that the existence of abundant top-down neural pathways from higher cognitive centres to the circuits of low-level vision can be explained only if one assumes that those pathways allow a transfer of information from cognitive areas in the brain to perceptual processing sites; (b) that perceptual plasticity of the brain in general, and the areas dedicated to perceptual processing in particular, refutes Fodor’s claim that perception is performed by encapsulated modules.

. Recent developments The hypothesis that early vision is cognitively impenetrable has been defended extensively by Raftopoulos (b; ; ; ; ). Raftopoulos () provided a review of recent accounts of attention, arguing that they confirm Pylyshyn’s thesis that attentional influence on perception is restricted to late vision and the pre-perceptual allocation of attention. He also presented an analysis of recent neuroscientific evidence about the latencies at which visual processes occur (esp. Lamme ; ; ; Lamme and Roelfsema ) in order to determine whether the outputs of early vision are representations of which subjects can be phenomenally aware and can be reported by subjects (see Block a; b)—a point that needs to be settled if the outputs of putatively impenetrable states are to be epistemically useful. An account of the relations between cognitive impenetrability and the hypothesis that perception has nonconceptual content was offered by Raftopoulos and Muller ()

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

and Raftopoulos (; ), who claimed that the CP of early vision is a necessary condition for perception to have nonconceptual content. This position was disputed by Bermúdez and Cahen () and is discussed in Macpherson’s contribution to this volume (Chapter ). Finally, Raftopoulos (a; b; ) has used the CI hypothesis to argue against theory-ladenness and in favour of scientific realism. (On the role of different forms of attention, see Section .; on Lamme and Roelfsema’s work, see .; on awareness and consciousness, see .; on nonconceptual content, see .; on the theory-ladenness of observation, see ..) Siegel (; ; ; ) has shown that questions about cognitive penetrability, and its potential impact on justification, are entangled with the questions of whether kinds are represented in perception and whether it is possible to distinguish visual from cognitive phenomenology. Siegel’s (; ) thesis that visual experience is affected by concept possession and that the contents of perception include high-level properties such as kinds implies a form of cognitive penetration of perception, although it is not clear whether early or late perceptual processes would be affected. Siegel () describes ‘illicit feedback loops’ from cognitive to perceptual states which have the potential to vitiate the perceptual justification of beliefs: if a perceptual experience has been formed through processes affected by beliefs, and the affecting beliefs are related to the belief that is being evaluated, justification of that belief by the perceptual experience will be epistemically suspect. (See Section . on CP and the contents of perception, and . on CP and perceptual justification of belief.) Lyons (; ; ; and Chapter  in this volume) has made substantive proposals about the nature of perceptual modularity and studied the epistemological consequences of cognitive penetrability and perceptual unencapsulation. On the moderate conception of modularity that he puts forward, a system can be modular even if it is not informationally encapsulated (see Section . on modularity; Lyons, Chapter ; and Lyons ). Lyons denies that cognitive penetration of perception would necessarily vitiate perceptual warrant, arguing that some forms of cognitive penetration are epistemically benign and others vicious, depending on whether penetration increases or diminishes the reliability of perception. For instance, cognitive influence resulting from slow learning will tend to be triggered by ‘enduring, stable features of the world’ and thus to be more reliable, unlike top-down effects which can be generated as a result of recent and local knowledge. (On the consequences of CP for perceptual warrant, see ..) Experimental findings on colour (Delk and Fillenbaum ; Hansen et al. ; Levin and Banaji ; Perky ) were used by Macpherson (, and Chapter ) to argue that colour experience is cognitively penetrable. On her account of the experiments, the way we categorize objects determines which colour experience we have of them. Macpherson specifically argues that it is the phenomenal looks of objects that are affected by memory colours (which are conceptually encoded states), not our beliefs or judgments about the colours of the objects. She also describes a mechanism

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



which could allow the phenomenal character of nonperceptual, doxastic states, such as imagining, to interact with the phenomenal character of perceptual experiences. Rebuttals of the claim that colour perception is cognitively penetrable argue that the experimental findings can be explained at the level of perceptual judgment (Lyons ; Zeimbekis ) or appeal to the presence of early visual shape–colour associations which do not entail the influence of conceptually encoded memory traces (Deroy ; Lyons ; Raftopoulos forthcoming). Much work currently under way on cognitive penetrability asks how it would affect the outcomes of existing philosophical debates, especially concerning perceptual warrant, nonconceptual content, perceptual and cognitive phenomenology, forms of awareness and consciousness, the role of attention in perception, realism and representationalism, and action. We explain the relevance of the penetrability debate to each of those topics in section . Before doing so, we have to describe some of the central concepts proper to the penetrability debate (Section ) and review existing definitions of CP (Section ). We end this historical outline with a methodological consideration. While the concepts of cognitive penetrability and impenetrability were originally introduced in order to defend the impenetrability of perception in the framework of a computationalist theory of the mind, much recent work on the topic tends to argue in favour of the penetrability of perception. It does so either by citing experimental findings and phenomenological considerations, or by loosening the definition of penetrability to make it more inclusive (see Section ). Scepticism about this reversal is expressed in the contribution by Edouard Machery (Chapter , ‘Cognitive Penetrability: A NoProgress Report’). While Machery does not make any substantive claims in favour of impenetrability, he seeks to ‘rekindle the doubts against the cognitive penetrability hypothesis’ on methodological grounds. He draws a historical comparison between the problems afflicting original cognitive penetrability claims, based on New Look psychology in the s and s, and shortcomings in the interpretations of experimental data offered by recent defences of cognitive penetrability. Among the recurrent flaws pointed out in the findings and their interpretations is an inability to determine the locus of cognitive penetration. For example, in Delk and Fillenbaum’s () colour experiments, there was no way to control whether it was perceptual experience or judgment that was affected by colour memories; but only an influence of memory on perception would count as CP. Another key set of recent findings for the penetrability debate, from Balcetis and Dunning’s () studies of distance perception, is based on estimates that subjects made from memory, so it is not clear whether the subjects reported their perceptions or their memories. Yet Balcetis and Dunning conclude from their findings that ‘the impact of motivational states extends from social judgment down into perceptual processes’. In fact, Machery points out that New Look psychologists consciously dismissed the problem of knowing whether it is perceptual experiences or judgments that are influenced by cognitive states (‘it is not particularly important to find out whether the stimuli are really “seen” as larger or smaller’—Tajfel

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

, cited by Machery).3 Another recurrent problem, according to Machery, is failure to replicate key experiments. One of the most paradigmatic New Look studies, Bruner and Goodman’s () experiment on how value affects size perception, failed to be replicated on several attempts (in , , and ); the same applies to recent experiments by Proffitt et al. () on distance estimates. The replication problem is compounded by the protocols used in several studies, which did not guarantee normal viewing conditions or involved stimuli that were ambiguous and degraded. Machery concludes that recent empirical research and theoretical arguments in favour of cognitive penetrability ‘suffer from the exact same kind of problems that in the first place resulted in widespread doubts’ about the findings of New Look psychology and their interpretation.

 Key Concepts of the Cognitive Penetrability Debate The concepts of modularity, informational encapsulation, early vision, late vision, hard-wired visual processing, theory-ladenness, perceptual learning, different concepts of attention, and the perception/cognition distinction itself, are constitutive of the cognitive penetrability debate. In this section, we explain those concepts and distinctions and their roles in the debate.

. Cognitive impenetrability, modularity, and informational encapsulation As explained in Section ., if a perceptual system is modular in Fodor’s sense, it will be cognitively impenetrable, but the cognitive impenetrability of a perceptual module does not imply its informational encapsulation. A perceptual information-processing system can be impermeable to cognitive influence but open to influence from other perceptual modules, including those that process information from other sensory modalities. Pylyshyn’s hypothesis that early vision is cognitively impenetrable is consistent with the occurrence of such interactions between different sensory modalities during perception, and this makes it more resilient than Fodor’s hypothesis about perceptual encapsulation. An example of crossmodal interaction is the McGurk effect (McGurk and MacDonald ): /ba/sounds paired with lip movements appropriate to the utterance of /ga/result in perceptions of the sound as /da/, suggesting that auditory processing receives inputs from vision. If information from sensory modality A affects which outputs sensory modality B generates, then B is not informationally encapsulated from A and is not modular in Fodor’s sense: its information-processing resources are not restricted to its proprietary database of information. However, as long as B is not influenced by a cognitive process, it is not cognitively penetrable.

3 Machery also considers, and rejects, Pylyshyn’s () proposal for resolving the locus problem by using signal detection theory. Pylyshyn sought to pin findings about penetration on changes in subjects’ response bias parameters, which would mean that the locus of penetration was judgment, not perception.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



Daniel Burnston and Jonathan Cohen (Chapter , ‘Perceptual Integration, Modularity, and Cognitive Penetration’) outline a concept of modularity which is compatible with evidence that perceptual processes are unencapsulated with respect to one another. They outline an integrative model of perception designed to deal with interactions across and within sensory systems; then, instead of rejecting modularity, they replace the criterion of informational encapsulation with anisotropy. Fixation of belief is a characteristic example of an isotropic process: the holistic nature of the abductive inferences that the brain uses (Fodor ) means that information stored anywhere in the system may affect the formation of a belief. According to Burnston and Cohen, anisotropic systems are modular ‘just because, and in so far as, there is a delimited range of parameters to which their processing is sensitive’. On this account, if photoreceptors turn out not to be informationally encapsulated (see Broackes ), they will nevertheless be modular, while systems in charge of representing certain higher-order, multimodal spatial relations (Koechlin et al. ) could turn out to be non-modular. On the other hand, processes that are standardly thought of as cognitive—such as recognizing scenes of chasing (Gao and Scholl ) and certain social cues (Langton et al. )—might turn out to be modular if they rely on ‘a significantly delimitable range of input parameters’. (Such cases of recognition are discussed again in Section ..) Apart from modularity, Burnston and Cohen argue that a perception/cognition distinction, and the cognitive impenetrability of certain perceptual systems, can also be preserved in an integrativist framework. Lyons (Chapter ) proposes a concept of modularity based on the criteria of isolability, unitariness, and specialization to information-processing systems. The key difference between his definition of modules and Fodor’s can be grasped through Lyons’s criterion of isolability, which is weaker than informational encapsulation. For information-processing systems S and S and task T, ‘S is isolable from S with respect to T iff S computes T and could do so even if S didn’t compute any functions’. On this definition, S is an unencapsulated system (it can draw on S’s information-processing resources), yet it is modular. Lyons argues that this definition of perceptual modules can secure much of what Fodor required of informational encapsulation when it comes to safeguarding perception from epistemically harmful cognitive influences. Both Lyons’s and Burnston and Cohen’s weakening of the criteria for modularity can be used to reject Fodor’s encapsulation hypothesis in favour of the hypothesis that early vision is cognitively impenetrable. This move also clearly limits the scope of criticisms of perceptual modularity (such as Prinz ) to encapsulation, as distinct from cognitive impenetrability. Ophelia Deroy (Chapter , ‘Multisensory Perception and Cognitive Penetration: The Unity Assumption, Thirty Years After’) studies interactions between sensory modalities that seem not to be restricted to low-level perceptual processing, but to be mediated by cognitive information. Deroy points out that, while input modules are usually seen to ‘all fit at the same place in the hierarchy’ between sensory systems

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

and systems that implement thought-like processes, there may also be multisensory modules—for instance, visual and auditory spatial information may ‘feed into a multisensory system which generates representations of audio-visual localization’. If such modules were appropriately influenced by semantic information, perception would be cognitively penetrated. That appears to be the case when multisensory interaction exhibits the phenomenon of ‘semantic congruence’, in which background assumptions about particular objects or combinations of sensory features determine whether or not information from different sensory sources will be integrated into a single object. For example, kettle shapes and whistling sounds are integrated into a single audiovisual object, while kettle shapes and barking sounds are not (Jackson ). Such assumptions apparently involve stored conceptual representations of objects, suggesting that their influence amounts to cognitive penetration. Deroy reconstructs and assesses the empirical reasoning underlying the interpretation of congruence as a semantic effect, concluding that the evidence does not show congruence to be a top-down causal influence. She also argues that if a top-down causal explanation of congruence turned out to be preferable, it would still need to be supplemented by an argument showing that the top-down influences are cognitive ones specifically. An alternative explanation of the evidence is that it results from ‘crossmodal correspondences’: nonrational representations of congruence which do not involve conceptual representations, such as the documented association between high pitches and bright surfaces (Spence ).

. Cognitive penetrability, early vision, and late vision The distinction between early vision and late vision became decisive for the cognitive penetrability debate when Pylyshyn used it to isolate the part of perception that he considers cognitively impenetrable. Note that the early/late vision distinction does not coincide with the distinction between cognition and perception, which is the subject of Section .. Fodor, as we saw earlier (.), holds that the informationally encapsulated part of vision encompasses all of the processes that lead up to visual object recognition. But that is unlikely to be the case; it is widely thought that object recognition is secured by an interaction of visual information with conceptual information and longterm memories (see Peterson  for a review of object-recognition hypotheses). In keeping with this, Pylyshyn (; ) isolated an early stage of vision that excludes recognitional processes and extends no further than the visual representations that Marr () calls the /D sketch. He claimed that only this early stage of vision is cognitively impenetrable. The distinction between early and late vision is central to Marr’s account of vision. According to Marr, vision first extracts information about the edges of objects in a scene, mainly from the way the scene reflects light (: ch. ). From that information jointly with further information about lighting, and with processes that include stereopsis, parallax, the hard-wired interpretation of monocular depth cues, and

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



geometric and topological principles for generating volumetric shape representations, the visual system constructs a viewer-centred (egocentric) representation of volumes and depth relations in the scene (Marr : chs  and ). Marr calls this type of visual representation the / D sketch; it is no less volumetric than a D sketch, but being a spatial representation of a three-dimensional scene from a limited viewpoint, without additional semantic interpretation, it is restricted to visible surfaces. The invisible parts of objects are only represented subsequently by what Marr calls the D model, which is suitably formatted for matching with memorized object representations. Marr uses the terms ‘early vision’, and sometimes ‘pure perception’, to describe the kinds of processes that go into constructing the / D visual sketch: ‘the recovery of surface information by purely data-driven processes without the need for particular hypotheses about the nature, use, or function of the objects being viewed’ (Marr : ). He thus sees vision up to that point as involving no processing that we could call semantic. Among the evidence used by Marr to support this position were findings in clinical neurology. Warrington and Taylor (; ) found that right parietal lesions caused difficulty in recognizing objects only when they were viewed from angles which foreshortened their natural axes (a geometric property exploited by early vision to construct depth from contours). Marr (: ) saw this as evidence that shape is constructed by the visual system independently of semantic input about the identity of objects (information about what kinds of objects they are). Generally, his position was that viewer-centred volumetric shape can be determined by vision alone even in difficult circumstances, in which one may have expected that vision draws on semantic information to supplement or disambiguate visual output (: ). Marr contrasted his account of vision with models given by computer vision scientists (e.g. Freuder ), whose object-recognition programs used semantic information to determine shapes and segment objects in a scene. Such models suggested that visual object recognition is determined throughout by semantic information. (Similar criticism of models provided by computer vision scientists can be found in Pylyshyn; see Pylyshyn : .) Instead, Marr thought that semantic information only intervenes to assist object recognition at a later stage of vision which presupposes the / D sketch (because to reach that stage, vision first has to map the / D sketch’s viewer-centred coordinate frame to an object-centred coordinate system; see Marr : –). To distinguish early vision from late vision, Pylyshyn (: ) drew on clinical evidence of functional dissociations of visual and cognitive functions; on the persistence of visual illusions; on the independence of principles of visual organization from principles of inference (e.g. in Kanizsa’s visual completions); and on psychophysical evidence of the attenuation and gating of visual signals. Humphreys and Riddoch’s () work on visual agnosia has shown that there can be deficits in object recognition despite the preservation of both memory and visual computations. In their words, this ‘supports the view that “perceptual” and “recognition” processes are separable’, and that the perceptual representation used in recognition ‘can be “driven” solely by

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

stimulus information, so that it is unaffected by contextual knowledge’ (Humphreys and Riddoch : ; cited by Pylyshyn : ). Pylyshyn also discusses the dissociation between recognition and visually guided action (Milner and Goodale ; Weiskrantz ; Holmes ), concluding that the outputs of the dorsal stream are not ‘available to the rest of the mind/brain’ (: ). (For an explanation of what the dorsal and ventral visual streams are, see Mahon and Wu, Chapter , §.) Pylyshyn’s description of the limits of early vision is fundamentally similar to Marr’s because it is based on Marr’s hierarchic conception of vision. Early vision is described as extending no further than viewer-centred representations, and the feedback of semantic information to early visual processes is admitted only after early vision has already assigned shape. The outputs of early vision are described as consisting of ‘shape representations involving at least surface layouts, occluding edges (where these are parsed into objects), and other details sufficiently rich to allow looking up parts of the stimulus in a shape-indexed memory for identification’ (: ). Pylyshyn adds two key features to Marr’s account. One is the thesis that early vision constructs primitive object-like representations (Pylyshyn ; Pylyshyn and Storm ), as distinct from generating a viewer-centred representation of the scene without differentiation into objects. The second is the thesis that attention does not directly modulate early visual processing; we discuss this in Section . on attention and cognitive penetration. Recent neuroscientific work on the timing of neural processes in the visual system allows us to form a working hypothesis for defining early vision in terms of processing latencies (Lamme ; ; ; Lamme and Roelfsema ; Roelfsema et al. ; Roelfsema ). That research suggests the following three-stage picture of vision. () Signal transmission from the retina through the visual areas of the brain up to the inferior temporal cortex lasts for about  ms after stimulus onset. This flow of information is described by Lamme and Roelfsema as a ‘feedforward sweep’ because it is a form of signal transmission immune to feedback from areas further downstream; the reason for this is that neurons in each area fire at a much slower frequency than neurons in the earlier area, ‘leaving no time for lateral connections and no time for feedback connections to exert their effect’ (Lamme and Roelfsema : ). Thus, for as long as the feedforward sweep of information through the visual areas lasts, it is a bottom-up form of signal transmission. The feedforward sweep is unconscious; it determines the classical receptive field of neurons and their basic tuning properties, extracts information which is subsequently useful for categorization, and results in some initial feature detection. () Feedback (‘recurrent’) connections and lateral connections between neurons begin to take place at the end of the feedforward sweep, culminating at about  ms; information is fed back to earlier areas (such as V; see Lamme and Roelfsema ), resulting in interaction between information distributed along the visual stream. At this stage, the visual system forms representations of transducable features (spatiotemporal properties, surface properties, viewer-centred shape, colour, texture, orientation,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



motion, etc.) and performs primitive feature binding and object segregation. This form of feedback is not thought to involve signals from cognitive centres but is considered to be confined to interactions within the visual areas of the brain; this is why Lamme and Roelfsema call it ‘local recurrent processing’.4 () Finally, at – ms, signals from frontal and prefrontal areas and mnemonic circuits begin to intervene and modulate perceptual processing in the visual cortex; eventually, the recurrent interactions with areas outside the visual stream make storage in visual working memory possible and give rise to global recurrent processing. (See Lamme , . These findings are supported by the ‘global neuronal workspace model’: Dehaene et al. ; Dehaene et al. ; Dehaene and Changeux ; Sergent et al. ). Visual processing, which up to this point comprised bottom-up processes, lateral connections, and local top-down effects from one visual area to another, can now be modulated by the activation of information stored in long-term memory as the synaptic weighting of neurons. Thus, global recurrent processing potentially marks the onset of conceptualization of perceptual content; this stage of visual processing is conceptually modulated and cognitively penetrated. Raftopoulos () holds that the feedforward sweep and local recurrent processing jointly form a suitable candidate for the role assigned to early vision by Pylyshyn, and that global recurrent processing in visual areas corresponds to late vision. For early vision to be cognitively impenetrable, either it has to receive no top-down signals from cognitive centres, or, if it does, the signals must not affect it in a ‘semantically coherent way’ or in a way that alters the content of early vision (see Section  on definitions of cognitive penetrability). This would have to apply equally during global recurrent processing. According to Raftopoulos, the feedforward sweep and local recurrent processing are impenetrable in this sense. Raftopoulos also argues that Lamme and Roelfsema’s distinction between local and global recurrent processing is consistent with Pylyshyn’s account of the outputs of early vision; and that, in keeping with Pylyshyn’s account, visual processing during the feedforward sweep and local recurrent processing are not directly influenced by cognitively driven attention.

. Cognitive penetrability and hard-wired ‘perceptual knowledge’ A top-down flow of information is the transmission of signals from a later stage of an information processing stream in the brain back to an earlier stage. Churchland describes such effects when, citing evidence by Wolter and Lund (), he writes that ‘there is some evidence that fully  percent of the axonal fibers in the human optic nerve are descending projections from the LGN [lateral geniculate nucleus] back to 4 Contextual or horizontal modulation and its effects are also well documented; see Lamme ; Lamme and Spekreijse ; and Raftopoulos : –, for further references. There is ample evidence that the early vision module consists of a set of interconnected processes for orientation, shape, colour, motion, stereo, and luminance that cooperate within it (Cavanagh ; Zeki ; ; and Raftopoulos : , for further references).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

the retinal surface’ (Churchland : ). This evidence has been confirmed by Van Essen () and more controversially by Zeki (). Generally, local feedback signals are thought to be essential to vision from an early stage (see also Zeki ; Cavanagh ; Lamme and Spekreijse ). Evidence of top-down effects on vision is often used to argue that vision is cognitively penetrable, and this is how Churchland was using it in the passage cited. However, for a top-down effect to amount to cognitive penetration of perception, a perceptual process has to be affected by another process downstream in the cognitive areas of the brain. For example, the fact that there are top-down effects on area V from V does not mean that there is cognitive penetration of vision, because V is a visual area which does not contain any information we could plausibly count as cognitive, such as long-term memories, concepts, or the contents of propositional attitudes. Many such forms of local top-down visual processing can be mistaken for cognitive influences on early vision, because the computations they perform are consistent with assumptions about the physical world. Vision has evolved to function according to those assumptions and they are built into the neural circuits that subserve visual processing. For example, Marr shows that the design of physiological visual mechanisms implements Bayesian priors to transform light differentials into edges, calculate depth from stereopsis or curvature from smooth variations in illumination, or implement the epipolar constraint. He describes the assumptions as ‘hard-wired’ into the visual system (: ) and as reflecting ‘some kind of a statistical rule of the universe’.5 According to Marr and other vision scientists, the assumptions are necessary because the retinal image under-determines both the distal object and the percept; unless vision was constrained by them, it would not construct perceptual representations of distal objects. Visual processing is constrained by such principles, which modulate information processing, at every level. Cavanagh (: ) holds that vision does not rely on purely bottom-up analyses drawing on retinal information, but instead completes construction of the percept by drawing on ‘object knowledge’, a set of rules that guide or constrain visual processing in order to resolve underdetermination. Rulebased extensions from partial data would constitute a form of inference. Spelke () holds that perception is constrained by a number of domain-specific principles about material objects and some of their properties. These constraints involve attentional bias toward particular inputs and a certain number of principled predispositions constraining the computation of those inputs; their operation is described by Spelke as inferential and akin to thinking.

5 Perhaps because Fodor () uses the expression ‘hard-wired’ to describe modular processing, the expression has become almost synonymous with ‘modular’ and would thus appear to imply cognitive impenetrability. But we are not making the point that principles like those Marr describes for visual processing are modular and thus impenetrable; our point is that knowledge of the principles is irrelevant to their application to stimuli by vision.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



However, such ‘hard-wired’ principles, ‘formation principles’ (Burge ), or ‘operational constraints’ (Raftopoulos ) are not available to introspection; they function outside the realm of consciousness and their operations cannot be attributed as acts to perceivers. When perception parses objects according to the Spelke’s principles that objects move in continuous paths and are rigid, subjects do not have to believe those principles; instead, they constrain perception’s modus operandi. Unlike theoretical commitments, they are only used by perception and are not available for a wide range of cognitive tasks. The constraints cannot be overridden since they are not under the perceiver’s control; subjects cannot substitute them with another body of constraints even when they know that they lead to errors. One way to account for the status of such apparent assumptions in perception is to hold that while perception operates in accordance with them, they are not represented even at a sub-personal level. They could be design-level properties that do not require the organism to token any representations of rules (see Dennett ). In that case, a neural state is formed through the spreading of activation and its modification as it passes through synapses; hard-wired constraints are implemented by computational processors, and computational principles describe the transformations from one state to another. Although the states produced by transformations have contents, the mathematical principles are not states of the system and do not have to be represented by it. This view has been defended by Burge: For many philosophers, the notion of computational states or explanations is theory-laden in a way that I do not intend. When I call states or explanations ‘computational’, I do not mean that there are transformations on syntactical items, whose syntactical or formal natures are independent of representational content [of the computed states]. I also do not mean that the principles governing transformation are instantiated in the psychology, or ‘looked up’, even implicitly in the system . . . principles governing perceptual transformations . . . are not the representational content of any states in the system, however unconscious. (Burge : )

Alternatively, the rules or generalizations could be realized by computational processors in the brain as a form of ‘tacit knowledge’ not realized by attitude states. Davies () holds that such knowledge is not inferentially integrated with attitude states but is subdoxastic and exists in special-purpose, separate subsystems (see also Stich ). The constituent concepts of attitude states such as beliefs must be concepts possessed by the believer, while the contents of tacit states are not conceptualized and not even accessible to the subject. Similarly, Stich and Nichols (: ) describe a ‘tacit theory’ as one which is not sententially encoded. In neither case would the principles reflected by operational constraints on perception be rules of inference that the visual system looks up to perform transformations, or premises used in such inferences. Not only are operational constraints not conceptually and propositionally structured mental contents, but they do not even seem to be contentful states of the perceptual system. Finally, the fact that perception relies on such constraints does not entail that perception is affected by concepts, since

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

on any interpretation of the principles implemented by operational constraints, those principles are not conceptually encoded.

. Cognitive penetrability and perceptual learning Even when one perceptual information-processing system draws on another perceptual system’s resources in a way that is infra-perceptual and clearly non-cognitive, it is still possible to argue that its outcome is the same as that of cognitive penetration. One such form of influence, which occurs without any contribution from personal-level or doxastic states, is the effect of perceptual learning on perceptual processing, in which the subject’s past experiences partly determine how her visual states process incoming information. There is evidence that visual memories acquired in the course of perceptual learning affect the way we perceive the world. The perceptual tasks affected are those that psychologists call object classification, identification, and categorization. The uses of the terms in psychology do not coincide with their uses in philosophy (see Section ., and Lowe, Chapter ). The task that psychologists call object classification is a coarsegrained form of classification which occurs at very short latencies: – ms and – ms after stimulus onset for classifying items as animals and faces respectively. Object identification and categorization are lengthy processes; they start about  ms after stimulus onset and last until  and even  ms (Johnson and Olshausen ). When the effects of familiarity on visual processing occur at – ms, which they often do, they are post-sensory, affecting the semantic information and processing required for identification and categorization (Delorme et al. ). As such they could be used to claim the cognitive penetrability only of late vision. However, Liu et al. () and Peterson and Enns () have shown that familiarity with objects or scenes built through repeated exposure, and sometimes through a single presentation, facilitates visual search, affects figure–ground segmentation, and speeds up object classification (in addition to identification). Crouzet et al. () have also shown that object classification can be affected by familiarity and repetition memory. These findings could suggest two potential sources of cognitive penetration. The first is that, since classification occurs at very early latencies, such effects could not be considered post-sensory and would potentially penetrate early vision. This threat would materialize if the classification process required either semantic information to intervene, or the representations of objects in working memory to be activated, since that too would amount to conceptual involvement in the visual process. However, the effects of familiarity on early classification seem to result from low-level visual areas: from V to V (Kirchner and Thorpe ) and perhaps a bit more upstream from the posterior IT cortex (Peterson ) and lateral occipital complex (Grill-Spector et al. ). Besides, if the classifications affected by visual memories required semantic information or activation of object memories, they would not be as fast as they are. (For further studies confirming that the effects are

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



low-level, see Chaumon et al. , Grill-Spector et al. , Peterson , Ullman et al. , Delorme et al. , and VanRullen and Thorpe .) Ullman et al. () show that statistical differences in physical properties of different subsets of images are detected very early by the visual system before any top-down semantic involvement, and suggest that early visual areas store implicit associations representing fragments of objects and shapes; according to Peterson, early visual memory stores ‘edge complexes’ as opposed to whole objects and object-shapes. The second potential source of cognitive penetration is that even if visual memories affected early processing without the mediation of either concepts or personal-level states, they would still mean that our past experiences shape the way we see the world, constituting what Stokes (Chapter ) calls ‘diachronic cognitive penetration’. That what we see is already an interpretation of incoming information is also true of the hardwired formation principles discussed in Section .; but in those cases, what we see is a function not of past personal experiences but of the way the visual system evolved to reflect high-level generalities about the distal objects presented to perception. In the case of visual memories which are proper to early vision and affect its processing of incoming data, individuals with different experiences see the world differently. Thus, although the findings about visual memory do not actually show influences on perception from cognitive states, they could be used to defeat epistemological objectives associated with denials of cognitive penetrability. As explained in Sections . and ., the cognitive impenetrability and perceptual modularity hypotheses were designed to rebut epistemological views flowing from certain interpretations of psychological findings, so the indirect form of perceptual bias caused by perceptual learning would strike at the heart of both hypotheses. Stokes (Chapter ) explicitly incorporates considerations about theory-neutrality and perceptual justification in his definition of cognitive impenetrability; and like Fodor (; ), Raftopoulos (; ) sees claims about cognitive architecture as inseparable from theses in the epistemology of perception. This takes us to our next topic, the relations between cognitive penetrability and theory-ladenness.

. Cognitive penetrability and the theory-ladenness of observation Bogen () recently distinguished three forms of observational theory-ladenness: (i) theories affect perceptual processes so that the percept is partially determined by our theoretical frameworks; (ii) observations cannot be described in a theoryneutral way and the meaning of observational terms is determined by theoretical presuppositions; (iii) theories make certain observations more salient than others. The concept of theory-ladenness used in the CP literature is the first one, which affects perception. As we saw in section , Hanson () and Kuhn () supported their claims that observation is theory-laden by citing psychological findings which suggest that perception is penetrable, so they thought that penetrability makes observation theory-laden. Simultaneously, defenders of the impenetrability hypothesis deny that observation is theory-laden (Fodor ; Raftopoulos ; ). However, they

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

do not proceed by denying theory-ladenness in order to block the conclusion that perception is cognitively penetrable; instead, Fodor, Raftopoulos, and sometimes Pylyshyn () start by trying to show that perception is impenetrable and that therefore observation is not theory-laden. A potential danger for that strategy lies in the fact that cognitive impenetrability may not suffice to keep theory-ladenness at bay: as Bogen’s other senses of the concept of theory-ladenness show, observation could be theory-laden in some other sense even if perception was cognitively impenetrable. A similar conclusion can be drawn from a concept of theory-ladenness recently proposed by Lyons (), on which the formation of perceptual beliefs is conceptually driven. If perception is taken to include perceptual belief, which is a doxastic state, then part of perception is theory-laden. But that part is unlikely to be an early part of perception; and if the CI hypothesis is restricted to early perceptual processes, CI is likely to be compatible with the form of theory-ladenness described by Lyons. Whether observation is theory-laden, and whether it is theory-laden as a consequence of cognitive penetration of perception, depends on what it is for a subject to possess a theory and on whether the influence of the theory can amount to cognitive penetration. Any concept or belief which affects how perceptual experience is generated counts as a penetrating state, irrespective of whether it is part of a scientific theory, a folk theory, or not essential to any particular theory. However, having a theory does not always mean having a set of beliefs and concepts. Theories may be tacit as opposed to ‘internally represented knowledge structures that [invoke] explicit rules or explicit sentence-like principles’ (Stich and Nichols ). This is likely to apply to our folk physical theory about the behaviour of everyday objects, and may or may not apply to the brain’s diachronic adaptation to wearing inverted lenses (described by Churchland ). It is not known whether the effects, if any, of folk physics on perception would constitute cognitive penetration. Pagondiotis (Chapter ; see Section .) argues that visual experience is penetrated by practical non-propositional knowledge. It is sometimes claimed that folk psychology (construed as a theory) is modular, which would make it impenetrable to other beliefs (Jackendoff ). This position has been criticized on the plausible grounds that social reasoning does not draw on limited resources or even limited sets of beliefs (Currie and Sterelny )— i.e. that social reasoning is isotropic. Finally, the concept of theory-ladenness should not be used to describe the fact that a perceptual system applies the kinds of formation principles or operational constraints described in Section .. For the reasons pointed out there, while the functioning of perceptual systems can be described as if they were applying a theory, there is no theory, and perhaps there are even no representations, involved in those transformations; and in any case not a theory known or applied by subjects. An interesting form of theory-ladenness is that which potentially stems from the influence of perceptual learning on perceptual processing. As seen in the section on perceptual learning (.), although the findings about visual memory do not show

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



influences on perception from cognitive states specifically, they could still be used to defeat epistemological uses of the cognitive impenetrability hypothesis. Suppose that through perceptual learning by means of repeated visual experiences in her field, a scientist’s specific professional needs have shaped her perceptual sensitivity so that she can recognize patterns that others cannot. She has learned which dimensions of visual analysis to attend to, and this process has reshaped her basic sensors by selecting the output of certain feature detectors. Suppose that this learning has induced changes in the circuitry of her early vision altering her visual perception; and that her theoretical commitments allow her to focus on specific locations of the visual array in her field and synthesize the picture in a way that is unavailable to others who do not share her visual training and memories. It is true that scientists (e.g. radiologists) and other experts (e.g. birdwatchers) who are trained to discern certain patterns and have memorized them can perceive patterns that other subjects cannot. Similarly, the scientist can detect patterns that others cannot. If this scenario is accurate, it has significant consequences for the role of theoryladenness in the penetrability debate. It suggests that non-cognitive, clearly perceptual influences on incoming visual information can be indirect bearers of the kinds of theoretical commitments that we usually think of as the content of conceptually couched theories. The outcome would vindicate Churchland’s belief that theoryladenness of observation flows from the brain’s plasticity. It would also refute the implicit assumption, described above, that the cognitive impenetrability of perception prevents theory-ladenness: here, the influences on vision are not cognitive, so impenetrability can be upheld; yet perception ends up being theory-laden. Since what is at stake are the epistemological consequences of discoveries about perceptual processing, it is worth asking whether the scenario described, if it was realistic, would affect scientists’ assessments of experiential evidence in the context of theory evaluation. One would expect perceptual learning, which is a honing or improvement of perceptual discernment in a specific domain, to be a benign influence on perception and not to lead to incommensurability of the learner’s percepts with those of other perceivers. Yet Kuhn argued that theories are incommensurable precisely on the grounds that observers presented with the same stimuli can see different things if their brain circuits are shaped differently by perceptual learning. Perhaps one could argue against Kuhn that, since both slow and fast perceptual learning are data-driven and not cognitively driven, if two subjects developed perceptual skills under different Kuhnian paradigms but were trained with the same data, they would end up seeing the same things. In that case, perceptual learning would not entail the epistemological consequences usually expected of theory-ladenness. A similar question has been asked about cognitive penetration by Lyons (Chapter , and Lyons ). He argues that in a reliabilist framework, penetrability can sometimes be an epistemologically benign influence because it can enhance the rapid classification of stimuli. While Lyons has in mind synchronic contexts, not the diachronic ones of perceptual learning, his points could apply equally well to the enhancement of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

classification by perceptual learning, thus confirming the expectation that perceptual learning should be epistemologically virtuous. If perceptual learning is epistemologically virtuous despite making perception theory-laden (due to brain plasticity), then some of the parties to the initial cognitive penetrability debate may have been talking at cross-purposes. According to Fodor, encapsulation is an empirically necessary condition for there to be perceptual consensus between observers despite differences among the theories they hold, and failure of encapsulation entails theory-ladenness (see .). But if the dependence of vision on visual memories due to perceptual learning is epistemologically benign, then it may not prevent the perceptual consensus that Fodor sought to secure, despite the fact that it may be a form of theory-ladenness. (Whether the information contained in visual memories can count as a theory of sorts is an issue that we cannot settle here, so we shall leave the question open.)

. Cognitive penetrability and attention The concept of attention has played an important role since the early days of the cognitive penetrability debate. Fodor () first used the ‘attention-shift argument’ to rebut Churchland’s contention that ambiguous figures, like Jastrow’s rabbit/duck drawing, illustrate the dependence of visual contents on ‘higher cognitive assumptions’ (Churchland : ). Churchland’s idea was that, depending on which concept one applies, one can see the figure differently each time. Fodor argued that in such cases of visual ambiguity, attention focuses visual resources on different parts of the scene each time; those resources then perform visual processing without any further contribution from attention or higher cognitive assumptions; and consequently, the output of the visual processing differs each time, but not due to any direct influence of concepts on perception. In similar spirit to Fodor, Pylyshyn sought to restrict the influence of cognitive states on perception to ‘the allocation of attention to certain locations or certain properties prior to the operation of early vision’, and the operation of late recognitional procedures which can involve endogenous attention, especially when the scene is ambiguous or sensory exploration is required to understand it (: ; : ). This classical impenetrabilist view of the role of attention in perception has been amended and developed by Raftopoulos (), who distinguished several forms of attention and granted that some of them are part of early visual processing. The classical view is challenged by Mole (Chapter ). To understand the shape of the current debate, we have first to explain some of the background literature on attention. Attentional mechanisms are needed because visual scenes typically contain more information than vision can process at any given time; the visual system selects one or a few objects at a time for more thorough processing, as inattentional blindness strikingly illustrates. Attention can bias the competitive interactions among stimuli by increasing the activity of neurons processing some of the stimuli (Desimone and Duncan ), for instance by lowering firing thresholds for salient feature detectors or by increasing their activity to enhance their output (Egeth et al. ; Kahneman et al.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



; Ungerleider and Haxby ). Numerous other specific roles have been assigned to attention (see Carrasco  for a review). But two distinctions concerning attention are particularly important for the penetrability debate: the difference between exogenous and endogenous attention, and the difference between the ‘spotlight’ and the ‘biased competition’ views of attention. Exogenous attention is drawn by the stimulus, or some early processing of the stimulus, and peaks at – ms; endogenous attention is cognitively driven and takes approximately  ms to deploy (Carrasco ). Both forms of attention can be overt, covert, spatial, or feature-oriented. (An example of endogenous spatial attention is expecting something to appear at a location and focusing attention on that location; watching out for snake-like shapes on a hike would be feature-based endogenous attention.) Endogenous attention is a cognitive phenomenon on any available criterion: it is personal-level, agentive, and takes so long to deploy that the scene attended to has already been processed by vision beyond the point of object recognition. Exogenous attention, however, can occur in early vision (see e.g. Desimone and Duncan ; Vecera ; Kravitz and Behrmann ). Endogenous effects on early visual processing would amount to CP, while exogenous effects can be compatible with impenetrability of the visual processes in which they are involved. The ‘spotlight’ (or ‘lens’) conception of attention (Posner ; Eriksen and Yeh ) construes attention as a distinct faculty or process that intervenes to resolve perceptual competition between representations and select some of them for further processing. The ‘biased competition’ account instead construes attention as the result of a perceptual competition which has been won by some set of stimuli and the way they were processed (Desimone ; Duncan ; Desimone and Duncan ): some properties of the scene just become salient in the course of visual processing. This concept of attention lends itself to use by theories which argue against the direct influence of endogenous attention on early visual processing and locate the influence of attention at an early visual stage potentially isolated from cognitive influence. However, the biased competition account does not have to be used in this way; it does not explicitly preclude unconscious cognitive influences from being among the factors that make certain properties of the scene salient during visual processing. In this sense, the distinction between these two conceptions of attention is potentially orthogonal to the previous one between endogenous and exogenous attention. While the classical impenetrabilist view of attention, just described, is that attention focuses visual resources and those resources then perform their processing without any further contribution from attention, the opposite view is taken by Christopher Mole (‘Attention and Cognitive Penetration’, Chapter ). Mole holds that ‘perceptual effects that are owing to what Bruner and Postman call “set or expectancy” can more intuitively be described as effects of attention’. For this, Bruner and Postman’s () perceptual set would have not just to be capable of directing the focus of attention to some set of stimuli already pre-packaged by visual processing, but to contribute to the way in which that processing yields its outputs. This is just what Mole argues,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

by using Kravitz and Behrmann’s () study of the role of feature-based attention in vision to show that attention is inextricably involved in early visual processing. In other words, attention is not ‘held off ’ while vision does its processing, as the attentionshift argument has it. Mole also enlists Desimone and Duncan’s biased-competition theory of attention against Pylyshyn’s view of the role of attention in perception. He sees Pylyshyn’s view as excluding attention from visual processing, but the competition theory as integrating it into vision: ‘The integrated competition theory should not be thought of as a theory that tells us what happens to the perceptual processing of objects as a consequence of attention having been directed at that object. It should, instead, be thought of as a theory of what it is for attention to be directed at an object.’ For example, Mole describes findings by Kravitz and Behrmann that semantic information can affect feature attention in early visual processing; on his reading of the biased competition account, this means that semantic information can unconsciously affect the competition for attention by early visual resources. The biased competition account of attention has been used in a different way by Raftopoulos (). He argues that feature attention can be part of early visual processing (not something extraneous to visual processing as Pylyshyn holds), but that in those cases it is not endogenous, nor necessarily unconsciously directed by concepts. In that case, early feature attention could be part of an integrativist cognitive architecture, like that proposed by Burnston and Cohen (Chapter ), in which perceptual states interact without their interaction entailing cognitive penetration. Raftopoulos made extensive use of the idea that feature attention can be restricted to early visual processing, arguing from Desimone and Duncan’s biased-competition account, Rensink’s (a; b) coherence field theory of attention, and Vecera’s () distinction between object segregation and object individuation, to the conclusion that attention does not directly affect early visual processing. For example, compare the following two scenarios concerning the role of attention in constructing visual objects. If objects were individuated in perception by having attention consciously focused on them, that would mean that object individuation is cognitively driven and cognitively penetrated. But if, when we consciously scan a scene, perception stops at objects which have already been segregated by visual processing in which feature attention plays an unconscious role (Vecera ), then object individuation is exogenously driven and could be cognitively impenetrable. A clear-cut case for the cognitive penetration of vision by attention could be made if we could show that agentively driven attention directly affects processes defined as part of early vision by Pylyshyn. Zeimbekis (Chapter ) argues that consciously driven spatial attention can affect visual processes that construct volumetric representations from monocular cues. Another attentionally mediated form of cognitive penetration would occur if semantic information determined the focus of attention unconsciously on expected features or locations. For example, Tye (: ) holds that differences in the phenomenal content of perception caused by visually ambiguous figures, such as Jastrow’s duck/rabbit figure, may be unconsciously caused by the conceptual abilities of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



viewers. In that case, it could be argued that feature attention is unconsciously focused due to concept possession, making visual content cognitively penetrable through the mediation of attention. However, what Tye has in mind is not always easy to distinguish from cognitive influence on late vision. For instance, Raftopoulos (: ) cites evidence of the penetration of object-recognition processes in which, following an unconscious analysis of the semantic content or gist of the scene (in the parietal cortex at about  ms), visual resources are directed to the sites in the visual scene where useful information is most likely to be found. Here, conceptually encoded information about the characteristic properties of kinds directs the visual search to enhance and accelerate object recognition, but the effect is on late vision. Finally, another attention-borne indirect effect of cognition on early vision is precueing. Pre-cueing occurs when a perceiver expects either an unspecified kind of stimulus to appear at a certain location of her visual field, or a specific kind of stimulus to appear anywhere in her visual field. The expectations influence the baseline activations of neurons in the visual areas whose receptive fields fall within the cued location, or of neurons that encode the features of the cued stimulus. These biases are anticipatory and occur before stimulus presentation (Nobre et al. : ). They are top-down biases because the cue is stored in working memory and activates the relevant neurons in the visual areas through a top-down flow of information. To deny that this amounts to cognitive penetration of early vision, one could try arguing that pre-cueing does not affect visual processing in a direct, online way, but just sets the initial values of certain parameters for subsequent computations. This is the kind of argument that Mole’s contribution, described above, seeks to rebut. (The question of whether indirect cognitive influences on early vision should count as cognitive penetration is discussed in Section , ‘Definitions of Cognitive Penetrability’.)

. Perception and cognition: is there a distinction? For causal relations between two states to count as cognitive penetration of perception, it has to be possible to maintain a distinction between perceptual and cognitive states. The same distinction is presupposed by claims that perception is not, or cannot be, penetrated by cognitive states. As Dretske (Chapter ) writes, ‘the first step . . . is to carefully distinguish perception from cognition.’ An alternative view would be to see the penetrability question as primitive: the cognitive impenetrability of perception would distinguish perception from cognition, but penetrability would mean that perception cannot be separated from cognition. We think this second approach should be rejected, for the following reasons. Although statements which describe evidence of cognitive penetration could perhaps be formulated as denials of the perception/cognition distinction, they do not have to be formulated in this way and rarely are today. For example, when describing putative effects of long-term memory on visual colour processing, Hansen et al. (: ) state that their ‘results show a high-level cognitive effect on low-level

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

perceptual mechanisms’. To reformulate this as a claim that there is no distinction between the states, Hansen et al. would have had to say that the effects of long-term memory on visual colour processing mean that long-term memory and visual colour processing cannot be distinguished. Instead, they construe the events as effects of one system on another distinct system, keeping the systems apart as relata. A precise proposal about how to preserve distinctions between systems in cases of penetration is proposed by Lyons (Chapter ): admitting cognitive penetrability would not obliterate the distinctions between information-processing systems because the systems could still be defined on the criteria of isolability, unitariness, and specialization (see .). It’s worth noting that in the s, evidence about what we today call ‘cognitive penetrability of perception’ was sometimes interpreted in a way that suggested obliteration of the perception-cognition distinction: ‘all perceptual experience is necessarily the end product of a categorization process’ (Bruner and Goodman ); ‘the perceptual effect of a stimulus is necessarily dependent upon the set or expectancy of the organism’ (Bruner and Postman , cited by Mole, Chapter ). Subsequent reactions against such accounts of perception in Marr’s account of vision, and Pylyshyn’s denial of the continuity of cognition and perception—to the extent that it is based on Marr’s hierarchic conception of vision—may be partly responsible for the shift in how penetrability claims are formulated. This takes us to our second point: Granting the cognitive impenetrability of perception could at best be part of some way of securing a perception/cognition distinction. If an information-processing system is cognitively impenetrable, that implies that it functions autonomously from certain other systems and can be defined without reference to them. But to secure the perception/cognition distinction, some explanation needs to be given of why the first system deserves to be called perceptual and the others cognitive. So we cannot see impenetrability as primitive and make the perception/cognition distinction depend on it; if anything has chances of being primitive in the debate, it is the status of some processes as perceptual and others as cognitive. Briefly, whether one defends CP or CI, one needs to keep perceptual and cognitive states distinct; defending CI additionally requires that certain kinds of relations do not obtain between the states. So we propose to take the perception/cognition distinction as more basic than questions about penetrability and impenetrability. The perception/cognition distinction is not immune to doubt. An extreme way to doubt it would be to adopt a view of the brain as devoid of any functional hierarchy (see Lashley ), but nobody today appears to be tempted by this position. Interactionists, who emphasize the existence of feedback from areas activated later by incoming information to areas which were activated earlier, do not deny that cognition has a functional cognitive hierarchy (Churchland et al. ; McClelland and Rumelhart ; Vecera and O’Reilly ). As Peterson (: ) describes them, ‘these models maintain a hierarchical structure in that lower-level processes must at least be initiated before higher-level processes are initiated’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



More moderately, one might worry, like Shea (), that perception is hard to individuate as a category because (i) it depends on top-down influences, Bayesian priors, and cross-modal effects, and (ii) there are borderline states that appear to be neither perceptual nor cognitive. On the first point, we have seen (Section .) that early vision can be distinguished from cognition on functional grounds by using evidence of dissociations from neurology, and that even late visual processes can be mapped functionally by using such evidence. (For an explanation of how such evidence is used to distinguish items in cognitive architecture, see Mahon and Wu, Chapter , §.) Warrington and Taylor’s (; ) work on left/right parietal lesions was used by Marr () to show that processes which precede viewer-centred brain representations of visual scenes are dissociable from the processes that assist object recognition with semantic feedback. Humphreys and Riddoch’s () work on visual agnosia was used by Pylyshyn to show that template matching, which is a late visual process, is dissociable from both early visual computations and semantic information. Mahon and Wu (Chapter ) construe cognition as a set of semantic information-processing systems which survive damage to specific modalities. Caramazza et al. () argue from clinical evidence for a functional distinction between (a) semantic information which is accessed by object-recognition processes across modalities and even by word recognition and (b) information-processing that is specific to sensory modalities. As Mahon and Wu write, ‘In experimental cognitive science, “semantics” is generally operationalized as the information that mediates the mapping from input to output systems, where input and output systems are modality-specific in terms of both their content and their format.’ In other words, cognitive science has a working hypothesis about how to distinguish perception from cognition. There is also a working hypothesis for defining an important part of perception, early vision, by using the timing of neural processes as they spread from sensory surfaces into areas of the brain that function autonomously from areas further downstream. The set of processes that Lamme and Roelfsema call the feedforward sweep and local recurrent processing qualifies as non-cognitive due to the priority of those processes relative to doxastic and personal-level processes, and as visual due to the traceability of the stimulus to the eyes. These seem good enough reasons to call them perceptual. Shea’s second point is that there are borderline states which are hard to classify as perceptual or cognitive. These include Carey’s () account of the representation of agency on perceiving animated figures, Michotte’s () accounts of perceiving causal relations, and Dienes and Perner’s () epistemic feelings. Such borderline cases are the subject of several chapters in this book. Burnston and Cohen argue that systems in charge of representing certain higher-order, multimodal spatial relations (Koechlin et al. ) may turn out to be non-modular and isotropic. In that case, it could be argued that such systems count as cognitive. However, on Burnston and Cohen’s account, other processes which are standardly thought of as cognitive, such

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

as recognizing scenes of chasing and recognizing certain social cues (Gao and Scholl ; Langton et al. ) might turn out to be modular, anisotropic, and thus arguably noncognitive.6 The most prominent borderline cases are perceptual contents. If perceptual experience represents high-level properties such as kinds (Siegel ), concepts may either be part of perceptual states or else determine those states’ contents. In that case, can perceptual contents—for instance, conscious visual contents—legitimately be counted as perceptual and not cognitive? One way to respond to this challenge is to try to deny Siegel’s thesis. Dretske (Chapter ) devises a thought experiment which serves as a criterion to determine what should be excluded from ‘cognitively unspoiled’ perceptual experience. He applies the criterion to Siegel’s claim that kinds are represented in perceptual experience, and concludes that while low-level properties that characterize kinds are represented in perception, kinds themselves are not. (For detailed discussion see Section .. Elsewhere, Dretske has argued at length against the idea that perception and cognition are ‘merely different stages in a more or less continuous informationhandling process’; see Dretske : .) Another possible response (to the challenge to the perception/cognition distinction from high-level perceptual contents) would be to argue that if perception represents kinds, it does so in some nonconceptual way, and that this prevents conscious perceptual contents from qualifying as cognitive states, construed as thought-like, conceptual states. For example, semantic information could influence vision in a way that made the features which characterize kinds more salient in visual experience, yet without making those contents conceptual (Tye  arguably makes such a proposal). Macpherson (Chapter ) argues that the cognitive penetration of perception is compatible with the thesis that perception has nonconceptual content; if Macpherson is right, then perhaps perceptual experience can represent kinds and still not be a cognitive state. Dokic and Martin (Chapter ) examine another borderline case which sits between perceptual and cognitive states: epistemic feelings during perception. They argue that feelings of familiarity depend on recognitional abilities, and that such feelings can alter the overall phenomenology of perception without affecting perceptual content. They also hold that epistemic feelings of confidence can explain the memory colour effect (Delk and Fillenbaum ; Hansen et al. ) as occurring at the level of judgment, not sensory experience. If Dokic and Martin are right, epistemic feelings experienced during perception are cognitive states, not perceptual ones. We do not think that the borderline cases threaten the perception/cognition distinction, because they mainly affect the classification of late visual processes as perceptual or cognitive. Since there are well-established working hypotheses about how to keep early vision distinct from cognition, it seems more reasonable to expect that perception 6 For more on cross-modal effects, see Section .. We have already explained why another potential set of borderline cases—hard-wired formation principles and Bayesian priors—do not count as cognitive processes or states (Section .).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



and cognition can be distinguished if only on the grounds provided by those hypotheses; and to continue to make statements about the cognitive penetrability or impenetrability of perception in the expectation that the distinction will be upheld. But if the borderline cases do not refute the perception/cognition distinction, they are definitely an issue for the penetrability debate. Suppose that definitions of early vision in functional terms or in terms of processing latencies succeed in isolating a set of non-borderline perceptual processes; then key issues for the penetrability debate will be played out from the stage that Marr called ‘late vision’ onwards, where the perception/cognition distinction is much harder to trace. While functional dissociations do allow us to isolate some late visual processes (like template matching in recognition—Pylyshyn : ), late vision is much harder to distinguish from cognitive states like propositional attitudes, concepts, long-term memories, imagery, or cognitively driven attention. Processes at these stages are involved in object recognition, perceptual belief, demonstrative thoughts and their verification, long-term visual recall, and visual imagery. They are important for epistemological reasons, and their cognitive penetrability would still affect perception’s experiential outputs, to which subjects have access for reports and judgments.

 Definitions of Cognitive Penetrability A necessary condition for the cognitive penetrability of information-processing systems was put forward by Pylyshyn by using the concept of semantic coherence: ‘if a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, i.e., it can be altered in a way that bears some logical relation to what the person knows’ (: ). Applied to perception, semantic coherence would be a relation between the contents of perceptual and cognitive systems. Macpherson (Chapter ) provides a clarification of the nature of such a relation: while the content of a visual state may be nonconceptual and the content of a cognitive state conceptual, a semantic relation between contents can hold if concepts that specify the content of the conceptual state are required for a correct specification of the content of the visual state.7 In addition to the content-based condition, Pylyshyn placed a requirement on vehicle states: for there to be CP, the perceptual system has to draw directly on the informational resources of a cognitive system in performing its computations. This condition is implicit in Pylyshyn’s denial that attentionally mediated cognitive influences on perception qualify as CP (: ; : ). It can be used to disqualify several key case types from counting as CP even though they meet the content condition: 7 Stokes (Ch. ) considers an attenuated interpretation of Pylyshyn’s semantic criterion, which loosens it to ‘representational coherence’: ‘This is the essence of what we mean by cognitive penetration: it is an influence that is coherent . . . when the meaning of the representation is taken into account’ (Pylyshyn : , cited by Stokes).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

(a) Concept possession can determine the focus of spatial or feature attention, with the result that which stimuli are processed visually depends causally and counterfactually on a cognitive state in a way that is semantically coherent with the content of that state. Such selective allocation of attention can occur either following stimulus onset or prior to stimulus onset as a result of pre-cueing (see Section .). For example, conceptually encoded information about the shapes or familiar orientations of kinds can affect figure–ground segregation, or which visual experience we have of an ambiguous figure. Pylyshyn denies that this dependence amounts to cognitive penetration on the grounds that visual processing is only affected indirectly by cognition, through attentional selection of the stimuli to be processed, not by drawing directly on the cognitive state’s computational resources. (b) As explained in Sections . and ., local top-down influences on incoming visual information can be indirect bearers of theoretical commitments through perceptual learning. Here again, the semantic coherence criterion is met but the cognitive influence is mediated by an external historical causal link between cognitive and perceptual vehicle states. The link is more attenuated than in Pylyshyn’s attentionally mediated cases, so on Pylyshyn’s account it should not count as a form of CP. Commentators who are prepared to call cognitive influence through perceptual learning a form of CP call it ‘diachronic penetration’ (Churchland ; Stokes, Chapter ). (c) The indirect causal link between cognitive and perceptual vehicle states could also be internal and still not secure cognitive penetration on the vehicle criterion, as suggested by an example from Macpherson (Chapter ):8 Suppose that Murdo believes that aliens are attacking Earth. This belief causes stress, which induces a migraine. Suppose that whenever Murdo has a migraine, he experiences flashing lights in the top half of his visual field. So suppose that Murdo visually experiences flashing lights in the sky on account of having his migraine. The content of this experience bears a semantic relation to the belief that caused it—the belief about the alien attack—but one might want to deny that this is a case of cognitive penetration.

Wu () similarly places both semantic and vehicle-based requirements on CP. According to Wu, failure of informational encapsulation would be entailed by a conjunction of the following three conditions: () Internal causal link: S’s visual experience V with content p is causally dependent on a nonvisual system Y via an internal causal link. () Computational condition: The influence of Y on V(p) makes the visual content intelligible owing to the computations that underwrite p using information in Y as a resource. () No explanatory defeaters: The resulting p is not explained by changes (i) in the proximal stimulus, (ii) the state of the eyes, (iii) the locus of attention. 8 This case, ‘Murdo’s migraine’, should be distinguished from the migraine described in Macpherson (), which does not preserve semantic coherence.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



The computational condition is semantic as well as being about vehicle states: the computations that support the visual state have to be able to exploit the conceptual representational resource. (This entails that late vision is cognitively penetrable since, for example, object recognition processes use conceptual information.) Jointly, () and () exclude perceptual learning from counting as cognitive penetration, since in perceptual learning the causal link is external. With (.iii), they exclude attentional effects from being cases of cognitive penetration—but with the exception of precueing: pre-cueing causally affects perceptual processing through an internal mental link, and would count as cognitive penetration on Wu’s account. Vehicle-based and semantic conditions are also required by Macpherson (). Macpherson describes cognitive penetrability as a situation in which ‘that which is perceived, the viewing conditions, and the state of the sensory organ are held fixed’; ‘the location of one’s attentional focus is held fixed’; different perceptual experiences are caused on account of differences between the states of cognitive systems; and the difference between cognitive states ‘makes this difference intelligible’ (Macpherson : , ). Intelligibility satisfies Pylyshyn’s content criterion, and the condition that fixes attentional focus precludes at least external indirect causal connections. Nevertheless, Macpherson finds an important case type that satisfies the definition: the influence of memory colour and typical colours represented by concepts (Delk and Fillenbaum ; Hansen et al. ; Perky ; Segal and Fusella ). Other commentators weaken the semantic condition: they do not require any relation of semantic intelligibility to hold between the contents of penetrating and penetrated states as long as a cognitive state causes changes to perceptual content. For instance, Siegel () defines cognitive penetrability as the nomological possibility that cognitive or affective states can cause a change in the visual contents that are or would be experienced while seeing and attending to the same distal stimuli under the same external conditions. According to Stokes (), a ‘perceptual experience E is cognitively penetrated if and only if () E is causally dependent upon some cognitive state C and () the causal link between E and C is internal and mental’. Both definitions include a requirement that the content of penetrated perceptual states be different to what it would have been without penetration, and both include a vehicle-based requirement: Siegel’s through attention-fixing, Stokes’s by excluding external causal links. But neither requires that the content of penetrating states make the content of penetrated states intelligible, or that they be semantically coherent. One motivation for dropping the semantic intelligibility condition is that some apparently paradigmatic cases of cognitive penetration do not seem to satisfy it (see Stokes, Chapter ). Consider wishful seeing (Balcetis and Dunning ), in which subjects see a gift card with a $ balance as being closer to them than one with the $ balance. A conceivable explanation of wishful seeing is that, as Lyons puts it (Chapter ), ‘thinking they’re going to get money makes subjects happy, and this

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

causes biochemical changes that affect perception in systematic ways, making objects look closer, hills less steep’. It is not clear how such a relation between penetrating and penetrated states could be described as semantically coherent. Another reason to drop the condition is that penetrating states may not always be conceptual states. If and when they are not, it may not be possible to define a relation of semantic intelligibility between penetrating and penetrated states. Consider the following cases: () If knowing what I am eating directly affects its taste, that is a case of concepts penetrating perception; but if knowing what I am eating causes feelings (such as disgust—see Nanay ) which affect taste, then the penetrating state is an affective one which may not be conceptual. () Cognitively driven spatial attention (see . on attention) is a potential source of cognitive penetration that does not involve concepts: while it is frequently driven by conceptual information (e.g. when we focus attention spatially as a function of conceptual expectations about the orientation of objects), perhaps endogenous spatial attention can also be driven by habits of sensory exploration or by some form of perceptual activity that involve no specific concepts. However, it is worth noting that the semantic intelligibility condition does apply to many key cases of penetrability currently under discussion. Cases on which much of Siegel’s work has focused, in which possessing a recognitional or a sortal concept affects which visual experience one has, would preserve semantic coherence. Whether those effects are unconscious effects on early vision or conscious ones during late vision, they alter visual phenomenology in ways that depend on semantic information about the characteristic features of kinds. An intelligibility condition is central to Macpherson’s (Chapter ) account of why cognitively penetrated perceptual states can still have nonconceptual contents: the nonconceptual contents of penetrated states are specifiable in terms of concepts that specify the contents of the penetrating states. Finally, Macpherson’s () thesis about the penetrability of colour perception also preserves semantic coherence: it claims that penetrated colour experiences represent actual colours as being closer to typical colours—and information about typical colour is semantic information. Another recent tendency is to doubt the need for a direct relation between vehicle states. For example, Macpherson (Chapter ) makes the vehicle condition optional, thus potentially admitting as cases of CP all the attention-shift cases that preserve the intelligibility condition (enumerated earlier in this section: pre-cueing, ambiguous figures, perceptual learning, and internal contrived cases like Murdo’s migraine). Stokes () sets a condition on vehicle states that excludes attention-shifts from counting as cases of cognitive penetrability by stipulating that the penetration relation must be internal and ‘involve mental states and processes’; however, he attenuates the condition by adding ‘but with no restriction on how long that [causal] chain is’. Stokes () also drops the intelligibility requirement. This qualifies as cognitive penetration the wishful-seeing cases, certain indirect forms of orectic and affective influence, and

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



possibly pre-cueing (which could be counted as internal because it is a lowering of the firing thresholds of neurons to make them more receptive to certain stimuli). What fuels these doubts about the vehicle condition, and the ease with which it disqualifies attention-shift cases, is a recent consequentialist conception of cognitive penetrability (the expression is used by Stokes, Chapter ). Deniers of the cognitive penetrability of perception constantly rely on the attention-shift argument, but their opponents increasingly point out that this tactic has an epistemological cost: applying the vehicle condition succeeds in defining parts of visual processing that are isolated from direct cognitive influence, but indirect cognitive influence can be just as harmful epistemologically. It can also defeat one of the main objectives of defending the hypothesis in the first place. For example, Lyons (Chapter ) compares two potential explanations of wishful seeing. On one explanation, wishful seeing is an attentional effect; on the other, it is ‘the direct cognitive penetration of vision by desire: subjects want the item to be closer, and this makes it look closer.’ Direct penetration minus the attentional mediation would satisfy Pylyshyn’s vehicle condition and ensure transfer of information between vehicles. But Lyons concludes that whether there is direct penetration or indirect attention-mediated influence, the outcome is the same epistemologically, especially for the perceptual justification of belief. This suggests that tightening the definition of penetrability to exclude attentionally mediated cognitive influences on perception may technically secure impenetrability, but still admit the epistemological consequences that the impenetrability hypothesis was meant to avoid. (For more on this topic see Section ..) Dustin Stokes (Chapter , ‘Towards a Consequentialist Understanding of Cognitive Penetration’) takes the disputes over the content and vehicle conditions as his starting point, and proposes to settle them by offering a consequentialist definition of cognitive penetrability. He argues that there is no consensus today on what cognitive penetrability is, with the result that the penetrability and impenetrability hypotheses have in effect become empirically untestable. Stokes shows that two standard criteria for penetrability yield contradictory and ambiguous verdicts in cases whose interpretation should be decisive for the debate. The first is Pylyshyn’s semantic coherence; the second is Stokes’s own () criterion of internal causal influence. The definitions yield conflicting verdicts on key case types: the influence on size perception of beliefs about the value of perceived objects, and diachronic influence on perception which enables perceptual representation of kinds according to Siegel. The first cases do not amount to cognitive penetration on Pylyshyn’s criterion, while they do on Stokes’s () earlier criterion. On the other hand, the perception of kinds does amount to cognitive penetration on Pylyshyn’s criterion, while other cases of diachronic influence on perception may not count as penetration on the weaker criterion of internal causal influence. Stokes then puts forward a consequentialist definition of the cognitive penetrability of perception based on the existing consensus about its presumed consequences: theory-ladenness of empirical observation, vitiation of the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

epistemic role of perception, and denial that perception is modular. On the consequentialist definition, cognitive influence on perception through attention-shifting qualifies as cognitive penetration for epistemological reasons, and perceptual learning amounts to cognitive penetration because it lends a degree of theory-ladenness to observation.9 To summarize, two conditions were defined for the cognitive penetrability of perception in the original computationalist framework: () the perceptual system draws on the informational resources of a cognitive system to perform its computations (the vehicle condition); () it generates output representations that it would not generate if it did not draw on the cognitive resources, and those representations stand in a relation of semantic coherence to the penetrating informational resources (the content condition). Gradually, proposals have emerged to weaken or drop these conditions. In particular, effects of cognitive influence on perceptual content, but without semantic coherence, are now widely seen to qualify as cases of penetrability. This is despite the fact that apparently good cases of cognitive penetrability can be found even by keeping the coherence condition.

 The Philosophical Relevance of Cognitive Penetrability The concepts outlined in the previous sections are constitutive of the topic of cognitive penetrability. In this section, we explain how hypotheses about penetrability and impenetrability affect the outcome of philosophical debates in several areas: the contents of perception, perceptual and cognitive phenomenology, nonconceptual content, consciousness and awareness, representationalism and realism, perceptual warrant, and action.

. High-level and low-level perceptual contents Debates about the contents of perception oppose the view that perceptual experience represents low-level properties (Dretske ; Tye ; ; Clark ; Raftopoulos ) to the view that it represents high-level properties and especially kinds (Siegel ; ; Bayne ; Masrour ). Part of this debate uses the methodology of phenomenal contrast cases, in particular for patients with associative agnosia, who cannot classify objects under sortals even though their ability to perceive low-level properties remains intact. Bayne () has argued that such patients, unlike normal subjects, suffer from an inability to perceive kinds. Use of the contrast methodology 9 A contrasting view is held by Raftopoulos (a; b; ; ; forthcoming), who holds that indirect effects on perception are not epistemologically damaging in the way that direct effects are. In that case, if the term ‘cognitive penetrability of perception’ is reserved for cognitive influences that admit consequences such as theory-ladenness, Kuhnian incommensurability or constructivism, only direct cognitive effects on perception should count as cases of cognitive penetrability.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



(Kriegel ; Siegel ) in cases like the auditory perception of sentences (Strawson ), the perception of kinds, or cases of agnosia also suggests that perceptual contents can be partly a function of semantic information about the objects of perception. (A synopsis of claims about high-level content and their methodology is provided by Briscoe, Chapter , §.) High-level perceptual content would imply cognitive penetration in some form, whether of early vision or late vision. Thus, one way to defend low-level content is to argue that perception is not penetrable in the ways that would be required for subjects to have high-level perceptual contents. This would involve showing that mechanisms which generate perceptual representations of low-level properties (orientation, movement, texture, shape, brightness, pitch, and so forth) function independently of semantic information about kinds; a position defended by Marr () as well as by proponents of the impenetrability of early vision. Additionally, the content generated by encapsulated processes would have to be a level of content that subjects can be aware of, since otherwise it cannot be the content low-level theorists have in mind. Defenders of high-level content, on the other hand, want to know whether their theses imply that perception is cognitively penetrated and whether that would make their account of perceptual contents incompatible with certain accounts of perceptual warrant (see the contributions to this volume by Lyons and Siegel (Chapter  and Afterword; Siegel ; ; Lyons ). Reciprocally, the question of which properties are represented in perception affects the penetrability debate. Evidence from agnosia for high-level perceptual content, or support for it from the contrast methodology, obliges defenders of impenetrability to show that high-level content is caused by cognitive penetration only of late vision. This in turn raises a question for both the impenetrability hypothesis and the low-level view of perceptual content: how to account for later visual processes like object recognition. On the impenetrability hypothesis, late visual states are affected by concepts; but being visual, they must also qualify as perceptual states. So how can the contents of perception be low-level properties? One way to deny that perception represents kinds and still account for Siegel’s () contrast cases may be to argue that concept possession causally affects early perceptual processing, but in a way that keeps conscious perceptual content lowlevel. For instance, it could be argued (see Tye : ) that when we speak a language, the contents of our auditory perception of sentences include salient word boundaries which would not be represented if we did not speak the language, but without including recognition of the words. Some accounts of low-level content are closely connected to a defence of the cognitive impenetrability hypothesis. Dretske’s () defence of low-level content is part of his denial that perception and cognition are ‘merely different stages in a more or less continuous information-handling process’, that they are ‘mutually interdependent and cannot be separated except by arbitrary rules’, and that ‘recognition, identification, and classification . . . occur at every phase of the perceptual

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

process’ (: , ). If the contents of perception are counterfactually dependent on causal cognitive influences, then it seems that low-level contents result from classification and that there is continuity between perception and cognition after all. In his contribution to this volume, Fred Dretske (Chapter , ‘Perception versus Conception: The Goldilocks Test’) argues against high-level perceptual content by devising a thought experiment, the ‘Goldilocks test’. This thought experiment is deceptively simple: the more one thinks about it, the more robust it turns out to be. In the thought experiment, subjects endowed with idealized painting skills can share their perceptual experiences by painting them. Thus, subjects can have each others’ visual perceptions and even superimpose them for comparison. This imaginary scenario is then applied to Siegel’s (; ) main contrast case, in which Siegel claims that acquiring the recognitional concept for pine trees alters the visual experience caused by pine trees. Call the subject who possesses the recognitional concept pine tree ‘E’ (for expert), and the subject who does not possess the concept ‘N’ (for novice). Now, imagine that the subjects paint a pine tree according to their visual experiences of it, and thus display and superimpose their visual experiences for comparison. If concept possession alters visual experience, N’s picture should be different to E’s picture. There are two ways this could happen: () N’s picture does not cause the visual experience of pine-tree-ness because the picture contains too little information. In that case, E should be able to add the information to N’s picture. But this is excluded by the thought experiment’s premises, which state that E and N have the same stimuli and saccades when they look at the pine tree. One may ask if subjects’ saccades and attention do operate identically when only one of them possesses a recognitional concept for the object perceived. But the thought experiment legitimately brackets this difference on the grounds that the stimuli are available to be saccaded or attended by subjects who possess or do not possess the concept alike. To deny this premise of the thought experiment, the defender of high-level content would have to commit herself to the unlikely view that if you don’t possess the concept, then you cannot visually experience certain shapes and colours. () N’s picture does not cause the visual experience of pine-tree-ness because the picture contains too much information. In that case, the picture would represent the features of a kind more specific than being-a-pine-tree, such as being-a-White-Pine specifically. But, as Dretske points out, ‘If one is going to see kind properties, one should be able to see them despite seeing details distinctive of more specific kinds’, and this makes it hard to explain how N’s picture could fail to cause an experience of pine-tree-ness in E by having too much information. Dretske concludes that N’s picture will have no more, and no less, information than E’s picture, but the same information. Since the pictures of the thought experiment are the subjects’ visual experiences of the object, their visual experiences will also have the same information. What does differ, according to Dretske, is that N just doesn’t know that arrangement of coloured shapes is a realization, a perceptually

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



determinate form, of the property pine-tree-ness. Thus, the Goldilocks test works like a destructive dilemma which forces us to either deny the experience of kinds or restrict it to a non-visual, cognitive phenomenology. (Cognitive phenomenology is the topic of Section ..) Robert Briscoe (Chapter , ‘Cognitive Penetration and the Reach of Phenomenal Content’) analyzes the three methods which have been used to reach the rich content view: phenomenal contrast (Siegel ); application of the contrast method to patients suffering from associative agnosia (Bayne ); and emprical evidence from attention, detection times, and visual adaptation (used by Fish ). He concludes that none of the three methodologies has succeeded in showing that visual experience can come to represent high-level properties as a result of influence from cognitive systems involved in object recognition. For instance, while visual searches are influenced cognitively through saliency maps (Wolfe et al. ), this affects the appearance of low-level properties without implying that high-level properties become part of phenomenal contents. Briscoe points out that Rubens and Benson’s () accounts of functional deficits in associative agnosia do not imply changes to the phenomenal character of experience (as claimed by Bayne ). (Remember (.) that work on visual agnosia has also been used to deny cognitive influence on vision; see Humphreys and Riddoch :  and Pylyshyn : .) When Fish appeals to evidence about detection times, he mistakenly assumes that the detection of high-level properties means that the properties have been experienced in the phenomenal character of perception even when the exposure time is extraordinarily short ( ms). Briscoe concludes that, for now, there is no evidence that cognitive systems involved in object recognition penetrate vision causing high-level properties to be represented in phenomenal contents. The visual experience of low-level properties can be affected by information originating from elsewhere outside the visual system, according to Briscoe. He describes psychophysical and neuroscientific evidence for what he calls ‘synchronic low-level informational penetration’ (SLIP): the effect of haptic information about slant on the visual appearance of surfaces (Ernst et al. ), the ‘sound-induced flash illusion’ (Shams et al. ), and the proprioceptive scaling of size (originally described by Malebranche). Finally, Briscoe describes what appears to be a genuine case of cognitive penetration that affects visual experience. He argues that findings by Vishton et al.  on the Ebbinghaus illusion, jointly with the immediate way in which the dorsal visual stream is thought to operate, imply that conscious distal intentions penetrate visual size perception. (This claim is discussed again in ..)

. Cognitive phenomenology, perceptual phenomenology, and epistemic feelings The distinction between cognitive and sensory phenomenology can be used to defend cognitive impenetrability by restricting the phenomenology of representing highlevel properties to cognitive, non-sensory processes. According to Marr (), object recognition is subject to top-down cognitive influences, while the visual processes

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

that lead up to the /D sketch are not (see .). Suppose, for the sake of argument, that Marr is right. Then, the absence of top-down influences on the earlier visual processes would allow us to claim that their phenomenology does not represent kinds. But that would not preclude there being a distinctive way it feels to represent kinds. This distinct phenomenology could supervene on later visual processes; it would be part of what Lyons () describes as ‘late experiential effects’ which do not affect the outputs of sensory processes. This would constitute a case of what Kriegel (: ) calls ‘phenomenological inflationism’ (the tendency to claim that higher-level properties as such are phenomenologically manifest): the mental representation of kinds, associated with recognitional concepts, would have its own phenomenology in addition to the phenomenology of perceiving low-level properties. In that case, the inability of patients suffering from pure associative agnosia (discussed by Bayne ) to represent kinds would be due to a deficit of the recognitional processes that subserve a cognitive phenomenology; this would be consistent with the fact that the visual processes which generate representations of low-level properties remain intact in those patients. Another way to seek to deny that experiences affected by cognition are sensory experiences would be to show that they are non-sensory epistemic feelings. Jerome Dokic and Jean-Remy Martin (Chapter , ‘Looks the Same but Feels Different: A Metacognitive Approach to Cognitive Penetrability’) use recent findings about metaperceptual feelings (e.g. feelings of presence or reality, familiarity, and confidence) to argue that while the possession of recognitional concepts does affect perceptual phenomenology globally, it does not influence visual contents, and that therefore the contrast cases do not imply the cognitive penetrability of vision. Metaperceptual feelings are presented through pathological cases (derealization disorder, Capgras syndrome, and Fregoli delusion) in which the feelings are either missing or overgenerated. (For reviews of the literature on metacognition see Dokic and Martin, Chapter ; Koriat ; Dokic .) Dokic and Martin argue that the experience of kinds can be explained by metacognitive epistemic feelings correlated with concept possession, but generated without the concepts entering the contents of perception, and without tokening of the concepts in beliefs. If their hypothesis is correct, then the part of perceptual experience that changes as a function of concept possession is not a representational part that affects the contents of perception, and should instead be factored into psychological modes adopted towards perceptual contents. Siegel () has considered the possibility that the experience of high-level properties in perception is due to epistemic feelings of familiarity, whether nonrepresentational ones which do not alter visual contents, or those that accompany commitment-involving doxastic attitudes. She objects to the first alternative on the grounds that feelings of familiarity emerge in connection with some content; and to the second on the grounds that the experience of kinds emerges independently of whether a commitment-involving attitude is adopted toward the object of perception. Dokic and Martin, citing recent work on metacognition (especially the work of Koriat

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



; ), argue that the feeling of familiarity is ‘grounded on the subpersonal monitoring of the quality of perceptual processes’. In that case, while occurrence of the feelings would presuppose possession of the recognitional concept (or at least repeated perceptions of members of the same kind), it would not require the relevant kind to be represented in perceptual contents. (Koriat  similarly describes as ‘contentless’ the mnemonic cues that affect the fluency of information retrieval and encoding.) Dokic and Martin also apply findings about metacognitive feelings to claims that colour perception is affected by memory colour (Delk and Fillenbaum ; Hansen et al. ; Macpherson ). In these experiments, subjects are placed in situations of what Tversky and Kahneman () call judgmental uncertainty (see Zeimbekis ). Dokic and Martin argue that epistemic feelings of confidence play a decisive role in the outcome of matching or colour categorization tasks. Consequently, whether perceived colour is cognitively penetrated will depend on whether such feelings can influence judgment while leaving experience intact. To show that any experiences affected by cognitive states are not sensory but cognitive, one also has to deal with the existence of cognitive states whose phenomenology is hard to distinguish from sensory phenomenology, such as visualizing and mental imagery. Both sides of the imagery debate consider visualizing and mental imagery to be cognitive states (Tye ; Pylyshyn ; ; Kosslyn ; see also Marr ), yet the phenomenology of those states has distinctly visual ingredients. The effect of conscious acts of visualizing on the visual experience of perceived shape is studied by John Zeimbekis (Chapter , ‘Seeing, Visualizing, and Believing: Pictures and Cognitive Penetration’). Zeimbekis describes two forms of visual ambiguity that involve visualizing. Visually ambiguous figures, especially pictures, have always been used to test the impenetrability hypothesis (Fodor ; Churchland ; Pylyshyn ; ; Macpherson ; Raftopoulos ; ). But two forms of ambiguity remain unexplored in the penetrability literature and pose a significant threat to the impenetrability hypothesis: the ability to visually experience a picture surface as flat after it has caused nonconceptual contents representing volumetric shapes; and the ability to use a surface initially perceived as flat to visualize three-dimensional scenes. In both cases, the visual processes which extract viewer-centred volumetric shapes (equivalent to Marr’s /D sketch) have to rely solely on monocular depth cues in the absence of parallax and stereopsis. Zeimbekis argues that those processes can be cognitively penetrated by acts of visualizing, including ones that draw on conceptual information about kinds. However, he holds that the penetrability of the visual processes does not weaken their ability to provide perceptual warrant and justification. When pictures support /D visual experiences through cognitively driven acts of visualizing, the penetrated visual states do not cause beliefs because they are consciously sustained acts which have a different phenomenology to visual perception. On the other hand, naturalistic pictures with good monocular depth cues, which generate /D contents without any need for visualizing, cause visual experiences which are harder to distinguish from object perceptions. Zeimbekis argues that such

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

states differ from visual object perceptions because they represent volumetric shape without either stereopsis or parallax, and do not engage the motion-guiding dorsal visual stream (see Matthen ). A comparison with monocular and binocular object perception suggests that such picture perceptions also differ phenomenologically from object perceptions (for reasons pointed out by Sacks  and Briscoe ). Thus, while the visual ambiguity caused by penetration of the visual processes that assign shape can lead to states with contradictory visual contents, at most one of those states can ground a perceptual belief and be taken into account in perceptual justification.

. Nonconceptual content The hypotheses that early perceptual processes are cognitively impenetrable or informationally encapsulated were formulated independently of the hypothesis that perceptual content is nonconceptual. (See Evans ; Peacocke ; a similar thesis is developed in Dretske ; . For an overview of nonconceptual content, see Bermúdez and Cahen  and Macpherson, Chapter .) Yet the concepts of cognitive impenetrability and nonconceptual content appear to be closely related. At first sight, it appears that the cognitive impenetrability of perception could suffice for perception to have nonconceptual content: if perceptual representational states are generated by cognitively impenetrable perceptual processes, then they should not be conceptual states. Conversely, one might ask whether the cognitive penetration of perception would make perceptual states conceptual. Both questions are important because, if impenetrability was a necessary condition for perception to have nonconceptual content, and if the penetrability of perception was established, then the contents of perception would not be nonconceptual. Macpherson (Chapter ), who argues that perception is cognitively penetrable (Macpherson ) but also defends nonconceptualism about perception (, and Chapter ), seeks to show that the two theses are compatible. On the other hand, Raftopoulos (; Raftopoulos and Muller ) has argued that cognitive impenetrability is necessary for perception to have nonconceptual content. Raftopoulos () reasons as follows. On Pylyshyn’s hypothesis, memory and conceptual information can influence recognitional processes, and visual processing is impenetrable up to and including viewer-centred object representations. Recent evidence from visual neuroscience and studies of awareness and attention (reviewed in Raftopoulos ) suggests that the cognitively impenetrable part of vision does not go beyond the viewer-centred representations which are thought to constitute the content of the representational brain states of which we can be phenomenally aware. The upshot is that, on current defences of the impenetrability hypothesis, the impenetrable part of vision is that whose outputs are representations corresponding to descriptions of perceptual nonconceptual content (viewer-centred volumetric representations, according to Evans  and Peacocke ). Raftopoulos construes nonconceptual contents to include both subpersonal computational states and personal-level states

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



(like Bermúdez ). Since impenetrability would be the absence of transmission of conceptual information to the perceptual processes that yield phenomenal content, it would guarantee the nonconceptual character of a certain part of perception. This is especially important for nonconceptual content conceived as a solution to the grounding problem and as part of the explanation of concept acquisition, since what does the grounding and explaining of conceptual states in these cases cannot be a conceptual state. It also has repercussions for the epistemological questions related to penetrability (see Sections . and .). Whether impenetrability is a necessary condition for nonconceptual content depends on two sets of factors. The first has to do with the nature of cognitive influences on perception. It has been argued that some perceptual contents are generated partly under cognitive influence but remain nonconceptual (e.g. figure– ground segregations; see Tye ; Peterson ; Zeimbekis, Chapter ). Such cases have been used to argue that ‘it does not seem to be a necessary condition on the nonconceptual content of representations in general that they be insulated from the propositional attitudes’ (Bermúdez and Cahen ). The second set of factors are the definitions one adopts of mental content in general, of perceptual content, and of nonconceptual content in particular. Fiona Macpherson (Chapter , ‘Cognitive Penetration and Nonconceptual Content’) sets out to determine whether impenetrability is a necessary condition for perception to have nonconceptual content, by taking into account concept-based and non-concept-based accounts of content, four definitions of nonconceptual content, and two definitions of cognitive penetrability. One of the aims of Macpherson’s chapter is to show that the mechanism of cognitive penetration described in Macpherson () allows us to preserve the thesis that perception has nonconceptual content. In what Macpherson calls ‘classic cognitive penetration’ of perception, the subject has a perceptual state with a content that she could not have if her perception was not causally affected by some cognitive state. In what Macpherson calls ‘cognitive penetration lite’, the subject has a perceptual state with a content that she could have even if her perception was not causally affected by some cognitive state. Macpherson argues that the second form of penetrability is compatible with standard definitions of nonconceptual content. For example, according to the state view of nonconceptual content (Heck ), a mental state is nonconceptual when it is possible for the subject to have that state even if she does not possess the concepts that would be involved in a correct specification of the state’s contents. Thus, if one of the content-specifying concepts did causally influence perceptual mechanisms to yield the same content, it would still remain true of the content that we could have had it without possessing the concept, satisfying the state definition of nonconceptual content. But in that case, Macpherson’s definition of ‘cognitive penetration lite’ would also be satisfied. According to another view of nonconceptual content, the content view, mental contents are nonconceptual to the extent that they do not count concepts as constituents. As Macpherson points out, one way to meet this condition for penetrated perceptual

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

contents is to adopt a possible-worlds account of mental contents, on which no mental content has concepts as constituents. In fact, Macpherson’s () account of the mechanism of penetration suggests that in a range of cases, the contents of penetrated states are of the same kind as the contents of non-penetrated states; this applies to cases in which visual imagery and perception contribute jointly and indistinguishably to the phenomenology of an experience (Perky ; Segal ; Segal and Fusella ; Chand and Murthy ). Macpherson’s ‘cognitive penetration lite’ is incompatible with one interpretation of Bermúdez and Cahen’s () account of nonconceptual content, and with Raftopoulos and Muller’s () account of nonconceptual content, according to which a content is nonconceptual only if it is not cognitively penetrated. On the other hand, what Macpherson calls ‘classic cognitive penetration’ (in which the subject has a perceptual state with a content that she could not have if her perception was not causally affected by some cognitive state) is incompatible with all four accounts of nonconceptual content. Thus, it turns out that whether the contents of perception are nonconceptual depends on whether all forms of cognitive penetration function in the way Macpherson’s ‘classic cognitive penetration’ does.

. Awareness and attention The core questions concerning cognitive penetrability and attention were explained earlier (Section .): the timing of different kinds of attention has to be understood before we can know which stages of visual processing they affect and whether they amount to cognitive penetration of perception. Beyond these questions, attention has been assigned a number of philosophically important tasks: accounting for consciousness (Campbell ; Lamme ; Kriegel ; Block b; Mole ), explaining how we acquire coherent object representations (Treisman ; Vecera ), verifying demonstrative thoughts, securing conscious awareness of objects (Dehaene et al. ; Lamme ), and others. For instance, if the individuation of objects was determined through mediation of feature attention or spatial attention driven by sortal concepts, that would support the form of sortalism Campbell calls the ‘delineation thesis’. Distinguishing different kinds of awareness and attention, the latencies at which they occur, and the processes and states which are thought to be their neural correlates is also crucial for an assessment of hypotheses about the contents of perception and cognitive phenomenology. The question of whether awareness as a form of consciousness presupposes attention is inextricably tied to the question of whether early vision is affected by attention. On one view, local recurrent processing (see .) gives rise to phenomenal awareness, marking off unconscious brain representations from representations of which we can have some form of consciousness (Lamme ; ; Block a; b; Super et al. ). For this to be consistent with denials that early vision is attentionally modulated, awareness should not presuppose attention. Traditionally it was thought that awareness does presuppose attention (for recent defences see Dehaene et al.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



; Prinz ). If that is so, then attention must intervene in early vision, potentially modulating it. On the other hand, there is now also empirical evidence to suggest the contrary (van Boxtel et al. ; Koch and Tsuchiya ; Lamme ; ; ). These questions about awareness and attention directly affect the outcome of the debate outlined in Section . about the contents of perception, since those contents are typically thought to be divulged phenomenally. Pylyshyn (; ) is sceptical about how well introspection can inform us of the contents of different levels of perception, claiming that we introspectively confuse the panoramic conception of the visual scene which is the result of late vision with earlier stages. Raftopoulos, applying Lamme’s distinction between phenomenal and report awareness, holds that ‘the phenomenal content of perception differs from the content of the same experiences when consciously accessed’ (: ). The contrast methodology, used to settle disputes based on introspective reports, may not be in a position to overcome the host of objections raised by Pylyshyn and may not cut as finely as accounts based on the neural correlates of different levels of awareness. For instance, a case has been made that phenomenal awareness, as understood by Block (), is the output of local recurrent processing (described in .). Unless neuroscientific findings show that such processing can interact with long-term memory, the differences in phenomenology captured by contrast cases may supervene on differences in doxastic processes which manipulate the outputs of those processes without altering them. Athanassios Raftopoulos (Chapter , ‘Cognitive Penetrability and Consciousness’) analyzes theories of phenomenal awareness and conscious access awareness in order to clarify the nature of the contents of late vision. He argues that studies of the attentional modulation of perceptual processing (in particular Carrasco et al. ; Epstein and Broota ) show that late vision is cognitively penetrable. But he also claims that this cognitive influence on late vision alters the visual phenomenology of perception, with the result that scenes look different before and after the onset of cognitively driven attention. According to Raftopoulos, most accounts of perceptual awareness preclude such awareness of the visual states which result from the penetration of late vision, because on those accounts, concepts do not alter the visual phenomenology of perceptual states but instead figure alongside visual information in hybrid conceptual and visual states. Raftopoulos argues that this applies to Burge’s () concept of phenomenal consciousness, Dretske’s () concept of thing awareness, Tye’s () account of the phenomenal character of perceptual experience, and Jackendoff ’s () distinction between visual phenomenology and visual understanding. On one possible construal, Block’s () distinction between phenomenal awareness and conscious access awareness could accommodate the form of awareness that Raftopoulos describes, although Block does not explicitly identify such a form of perceptual awareness. Raftopoulos calls the states of perceptual awareness resulting from the cognitive penetration of late vision ‘conceptually modulated visual awareness’. Such cognitive penetration of late vision means that the content of perception is

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

not low-level content, and that the resulting content and phenomenology of perception are not due to penetration of early vision. On the other hand, it also means that the cognitive impenetrability of early vision does not prevent the final outputs of perception, which are accessible to the subject, from being the outputs of cognitively penetrated states. This shifts the interest of the penetrability debate to the epistemological consequences of what happens in late vision.

. Representationalism and realism In addition to his arguments against modularity (outlined in .), Churchland () challenged Fodor’s use of modularity to counter antirealism in the philosophy of science, arguing that a modular account of perception would not secure the neutrality of perception sought by Fodor () but a form of ‘universal dogmatism’ (: ). Churchland’s criticism of modularity and his criticism of the idea of using modularity to argue against antirealism are distinct: he holds that even if perception was modular, it would still not secure any form of realism. This fundamental question is the subject of Jonathan Lowe’s contribution (Chapter , ‘Perceptual Content, Cognitive Penetrability, and Realism’). Lowe does not see questions of realism as independent of theories of perception, but he holds that the argument from perception to realism cannot begin from a representationalist theory of perception. For instance, he holds that it would be circular to appeal to objects and three-dimensional space prior to making an abductive inference to them, and that the argument for realism cannot proceed by ‘appeal to antecedently assumed empirical scientific theories’. Therefore, the cognitive impenetrability hypothesis is disqualified from being the starting point from which to defend realism: it is both an empirical scientific hypothesis and a representationalist theory of perception. Nonconceptualist representationalist theories like those of Dretske and of Tye, and Raftopoulos (), which is the target of Lowe’s discussion, are also disqualified from playing this role. However, those theories have another role to play in the defence of realism, as Lowe later explains. Lowe claims that there is a non-representational, ‘introspectively accessible nonconceptual content of perception’ which is topologically two-dimensional—despite our ability to also experience that space as three-dimensional. We cannot explain this two-dimensional content physiologically (from the shape of retina) to argue against antirealists for the reasons already explained, but ‘the phenomenological fact itself [. . .] is unassailable, since it is ascertainable by introspection’. From this datum we can abductively infer the existence of three-dimensional objects. By using the two-dimensional phenomena as the first step in the defence of realism, we avoid the criticism that we have inferred the world of three-dimensional objects from our subjective representations of the world. According to Lowe, representationalists have no ground from which to answer this criticism. However, representationalist theories based on empirical hypotheses—such as nonconceptualist representationalism and Fodor’s modularity hypothesis—do have an important role to play in the framework

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



of inferences to the best explanation: ‘the realist’s hypothesis has a genuine and substantive explanatory role to play in accounting for the structure of our perceptual experience and can be shown to fulfill this role without begging any question against the antirealist.’ In addition to the foundational issue of how to defend realism by using the cognitive impenetrability hypothesis, Lowe makes a substantive contribution to the details of the penetrability debate. He argues that the level of perceptual content isolated by the most recent version of the impenetrability hypothesis (Raftopoulos ) cannot play the role just outlined for representationalist theories. According to Lowe, that content does not suffice to individuate the objects of perception but only to discriminate their spatial boundaries. For example, Tibbles the cat shares spatial boundaries with a hunk of matter, but the matter can change while Tibbles remains the same (see Lowe ). Lowe calls hunks of matter ‘Spelke-objects’ after Spelke (), and argues that they are only one among many sorts of objects that perception primitively distinguishes. He argues that identity criteria for objects have to be relativized to broad, primitive sortals (like animal, hunk of matter, material artefact); these sortal-relative identity criteria are of a very broad kind and could thus be well attuned to real distinctions in nature. Direct realism is committed to the view that subjects are perceptually aware of objects and of some of their properties directly, not by first being aware of representations of the objects and properties. In that case, the contents of perception have to be object-centred, D visual representations of distal objects. This view can be broken down into two claims about vision. First, perceptual contents include expectations about how shape changes as we move relative to objects. On standard accounts of vision, such expectations require longterm memories and specifically semantic information to be applied to objects. On one hand, this suggests that direct realist accounts of perception require cognitive penetration of perception; sensorimotor knowledge without semantic information is unlikely to be adequate to produce the relevant expectations. (A suitable form of penetration may be described by Mahon and Wu (Chapter ), who hold that functional concepts for artefacts can affect processing in the dorsal stream.) On the other hand, it means that direct realists have to explain which status they confer on the experience of objects as described by both nonconceptualists (egocentric volumetric representations, according to Evans  and Peacocke ) and theories of vision (viewer-centred representations). If the first object representations of which we are aware are anything like those described by current theories of phenomenal awareness (see .), then D distal objects are represented only after we have an initial egocentric visual experience of the spatial layout of the scene; in other words, they are not represented directly but indirectly. Here, the threat to naive realism is from the earliness of awareness (the awareness of stages of visual processing prior to D representations). Secondly, direct realists hold that outline shapes are literally objects of vision in the world (Noe : ). This claim can be doubted on the grounds that such content

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

occurs too early for us to be aware of it, by using theories of vision, nonconceptual content, and awareness. For example, Briscoe () has argued that conscious representations of outlines are likely to be cognitively driven achievements. Costas Pagondiotis (Chapter , ‘Cognitive (Im)penetrability of Vision’) asks whether the portion of perception defined as impenetrable by Pylyshyn allows perception to play the role of anchoring the subject to the world. He argues that in this respect, the cognitive impenetrability hypothesis has led to a philosophically undesirable dilemma: either vision is not dependent on cognition and consequently vision is alienated and disconnected from thought, or else vision is dependent upon cognition, with the result that the distinction between vision and thought is blurred. He suggests two ways out of this dilemma. The first remains within the framework of the representationalist–computationalist theory of mind. It involves defining the cognitive penetrability of early vision as its dependence on other systems which involve either non-propositional representations or propositional but domain-specific representations. In either case, vision would not be ‘promiscuous’ like thought is, so this could provide the distinction between vision and thought which, according to Pagondiotis, is not provided by Pylyshyn’s account. The second way out of the dilemma, favoured by Pagondiotis, draws on theories of perception that reject the representationalist–computationalist theory of mind and support an antiintellectualist approach to perception, stressing the dependence of perception on practical non-propositional knowledge (Dreyfus ; Haugeland ; Hurley ; Noë ; O’Regan ). Pagondiotis argues that vision is directly penetrated by practical non-propositional knowledge, and that as a result, visual content is objectdependent and inextricably connected to an embodied agent, instead of being an isolated atomic given disconnected from the perceiver. According to Pagondiotis, taking visual content to be penetrated by practical nonpropositional knowledge paves the way for a defence of the view that perception brings us into direct contact with the world, and for a form of realism based on the attunement between perceptual content and world. That attunement is exemplified by the dependence of perceptual content on sensorimotor knowledge and anticipation. While Pagondiotis sees practical non-inferential knowledge as a form of Churchland’s diachronic penetration, he holds that perception’s dependence on such knowledge is rational without being either inferential or propositional. On this basis, he argues that practical non-propositional knowledge does not constitute a form of theory-ladenness and provides ‘a sufficient ground for scientists and ordinary observers to communicate and to find out what is or is not veridical’—satisfying the epistemological objective set for modularity by Fodor (see .).

. Action Apart from the actions involved in interactive accounts of perception (see previous section, and Pagondiotis, Chapter ), hypotheses about cognitive penetrability and

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



impenetrability are relevant to theories in the philosophy of action that distinguish different levels of action like distal, proximal and motor intentions (Pacherie ), or prior intentions and intentions-in-action; and to accounts of how actions inherit the goals of prior intentions, such as Israel et al.’s () concept of ‘belief-how’. Pylyshyn () has argued that motor systems are cognitively impenetrable, citing evidence about the dorsal ‘vision-for-action’ stream (Milner and Goodale ; Ungerleider and Mishkin ) and findings which suggest its autonomy from the processes that subserve object recognition (: ). Evidence that the final stages of vision and first stages of motor control share information in the same format, and that conscious planning impairs detection and response (Hommel ), suggest direct interactions between dorsal system and motor systems without cognitive mediation; and much recent work in psychology shows direct ties between perception and action (Beets et al. ; Humphreys and Riddoch ; ; Hommel ; Klatzky et al. ; Goodale and Milner ; Riddoch et al. ). However, Briscoe (Chapter ) argues that visual processes which compute size for perception are penetrated by distal intentions. Vishton et al. () not only found that grasping affects the visual experience of size, but that even ‘listening to a description of a reaching task’ affected size perception. As Briscoe points out, this implies that forming the intention to act suffices to influence size perception. If he is right, then there are cases in which action determines perceptual content in a much more intimate way than is usually thought to occur, for instance by sensory exploration. One could think of sensory exploration as an activity in which we probe the visual scene by directing visual resources and letting them process the scene visually, but that clearly does not apply to the case just described. Brad Mahon and Wayne Wu (Chapter , ‘Cognitive Penetration of the Dorsal Visual Stream?’) argue that the contents of intentions, semantic information about functional kinds, or both, penetrate the computations of motor response in the dorsal visual stream. They defend this thesis in a framework which sees action as a problem to be solved by the brain: intentions represent actions, actions are specific input– output mappings, and the brain must produce the mappings. The problem can be split into components on the basis of dissociations between reaching to grasp (impaired in optic ataxia) and object use (impaired in apraxia). Reaching can be determined by the dorsal stream’s visual computations, but information about which manipulations are appropriate for object use constitute stored knowledge; they are comparable to lexical representations of words (Rothi et al. ), and count as semantic information. Thus, prima facie, the dorsal system cannot solve the entire problem of producing an action plan. (One solution could be provided by visual-motor memories—Allport ; Gallese and Lakoff . Mahon and Wu argue against this solution by adducing evidence from object-use apraxia and optic ataxia.) Mahon and Wu show that the action is highly unlikely to be produced by an influence of semantic information on the inputs to, or outputs from, the dorsal-stream processing. For instance, if the ventral stream analyzed the properties of artefacts

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

and highlighted their relevant parts for action, that would still underdetermine the action plan. At output level, it could be thought that the dorsal stream provides a set of action plans and then one is selected. What militates against this proposal is that dorsal computations are thought to be restricted to immediate manipulation, while the grasping component of reach-to-grasp movements anticipates manipulation. Thus, patients with object-form agnosia are unimpaired for reaching but impaired for grasping objects in the right way to perform the appropriate action. They are also impaired for integrating information about the surface properties of an object, such as whether the handle was slippery, into their initial grasp. This suggests that the action plan cannot be determined by the dorsal stream without help from semantic information about functional kinds and other properties of the object to be acted upon. Mahon and Wu consider two possible sources of such semantic information. One is that representations of objects in the contents of intentions to act (on those objects) directly transfer information for computation to the dorsal stream. For example, representing how one intends to use a hammer will determine how one reaches to grasp it. Alternatively, functional concepts for objects may transfer information over which the ventral stream computes; this is consistent with deficits in semantic dementia, which affects cognition of the function of artifacts.

. Perceptual warrant According to Siegel (), cognitive penetration that determined which high-level properties are represented in perception would vitiate the perceptual justification of belief: if Jill visually experiences Jack’s face as angry because she believes that Jack is angry, then visually experiencing Jack’s face as angry can no longer justify her belief that Jack is angry. Such epistemically illicit feedback from beliefs to perception would threaten the internalist principle of justification that Lyons () calls ‘seemings internalism’: ‘one’s belief that x is F is prima facie justified any time it is based on one’s having a perceptual experience that x is F.’ The dependence of perceptual experience on conceptual states that Siegel has in mind is meant to apply even when eye fixation and attention are constant, so the attention-shift argument cannot be used to deny that it is cognitive penetration specifically (see Section ) that damages perceptual justification. Alternatively, one might seek to deny the inference from cognitive penetration to vitiation of justification: the threat to justification may not materialize if the effect of semantic information is restricted to cognitive phenomenology, understood as distinct from the sensory phenomenology to which the epistemic subject appeals during internal justification. However, this route is also blocked because Siegel () holds that the representation of high-level properties affects sensory phenomenology specifically. A way to deal with potential threats to perceptual warrant from cognitive penetration is proposed by Jack C. Lyons (Chapter , ‘Unencapsulated Modules and Perceptual Judgment’). As explained in Section ., Lyons holds that informationprocessing systems can be modular even if they are not informationally encapsulated,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



including from cognitive processes. But he also denies that penetrability would necessarily damage perceptual warrant by arguing that cognitive penetrability does not diminish the overall reliability of perception. Justificatory circularity would require a basing relation, not just a causal relation, to obtain between penetrating and penetrated states and between perceptions and resulting beliefs. In Siegel’s example, the content of the perception is not based on long-term memories, although it is caused by them, and the content of the belief is not based on the content of the perceptual experience although it may be caused by it. Lyons (Chapter , and ) also argues that penetrability can be an epistemologically benign influence in a reliabilist framework, for instance when it enhances the rapid classification of stimuli. If one is on a hike and watching out for snakes, cognitive penetration may allow such rapid classification and even make certain stimuli more salient. Another example is perceptual learning (if construed as a form of ‘diachronic penetration’; see .): as an improvement of perceptual discernment in a specific domain, it appears to be an epistemologically benign influence on perception. The association of impenetrability with epistemological virtuousness can also be doubted. Lyons () considers three possible explanations of a case-type of cognitive penetrability, the penetration of colour perception: () subjects’ current, occurrent belief that bananas are yellow has genuinely topdown effects on the perceptual state, making the banana look more yellow; () subjects’ longstanding belief that bananas are yellow has the top-down effect; and () subjects’ history with yellow bananas has produced an associative connection whereby low-level perceptual features of bananas (and not the banana identification itself) prime yellow in the colour detection system, producing a lateral, rather than top-down, effect on the perceptual state. Explanation () amounts to cognitive penetration, while () does not (see Section .). Yet there is no difference in the epistemological value of () and () in a reliabilist framework according to Lyons—despite the fact that () is currently used in the literature as a denial of cognitive penetration, while () is considered an admission of cognitive penetration. Moreover, () is epistemologically more pernicious than ()—a difference that does not show up on existing definitions of cognitive penetrability.

References Allport, D. A. (). Distributed memory, modular subsystems and dysphasia. In S. K. Newman and R. Epstein (eds), Current Perspectives in Dysphasia, –. New York: Churchill Livingstone. Balcetis, E., and Dunning, D. (). Cognitive dissonance and the perception of natural environments. Psychological Science : –. Balcetis, E., and Dunning, D. (). Wishful seeing: desired objects are seen as closer. Psychological Science : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

Bayne, T. (). Perception and the reach of phenomenal content. Philosophical Quarterly : –. Beets, I. A. M., Rosler, F., and Fiehler, K. (). Non-visual motor training affects visual motion perception: evidence from violating the two-thirds power law. Journal of Neurophysiology : –. Bermúdez, J. L. (). Nonconceptual content: from perceptual experience to subpersonal computational states. Mind and Language (): –. Bermúdez, J., and Cahen, A. (). Nonconceptual mental content. In E. N. Zalta (ed.), Stanford Encyclopedia of Philosophy: . Block, N. (). How can we find the neural correlate of consciousness? Trends in Neuroscience : –. Block, N. (a). Two neural correlates of consciousness. In Consciousness, Function, and Representation: Collected Papers, vol. , –. Cambridge, Mass.: MIT Press. Block, N. (b). Consciousness, accessibility, and the mesh between psychology and neuroscience. Brain and Behavioral Sciences : –. Bogen, J. (). Theory and observation in science. In E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy (Spring  edn): Brewer, W., and Lambert, B. (). The theory ladenness of observation: evidence from cognitive psychology. In Proceedings of the Fifteenth Annual Conference of the Cognitive Science Society, –. Hillsdale, NJ: Erlbaum. Brewer, W., and Lambert, B. (). The theory-ladenness of observation and the theoryladenness of the rest of the scientific process. Philosophy of Science : S–. Briscoe, R. (). Vision, action, and make-perceive. Mind and Language : –. Broackes, J. (). What do the color-blind see? In J. Cohen and M. Matthen (eds), Color Ontology and Color Science, –. Cambridge, Mass.: MIT Press. Brown, H. I. (). Observation and Objectivity. New York: Oxford University Press. Bruner, J., and Goodman, C. (). Value and need as organizing factors in perception. Journal of Abnormal and Social Psychology : –. Bruner, J., and Postman, L. (). On the perception of incongruity: a paradigm. Journal of Personality : –. Burge, T. (). Origins of Objectivity. Oxford: Clarendon Press. Campbell, J. (). Reference and Consciousness. Oxford: Clarendon Press. Caramazza A., Hillis, A., Rapp, B., and Romani, C. (). The multiple semantics hypothesis: multiple confusions? Cognitive Neuropsychology : –. Carey, S. (). The Origin of Concepts. Oxford: Oxford University Press. Carrasco, M. (). Visual attention: the past  years. Vision Research : –. Carrasco, M., Ling, S., and Read, S. (). Attention alters appearance. Nature Neuroscience : –. Cavanagh, P. (). Pathways in early vision. In Z. Pylyshyn (ed.), Computational Processes in Human Vision: An Interdisciplinary Perspective, –. Norwood, NJ: Ablex. Cavanagh, P. (). Visual cognition. Vision Research : –. Chand, P. K., and Murthy, P. (). Understanding a strange phenomenon: Lilliputian hallucinations. German Journal of Psychiatry : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



Chaumon, M., Drouet, V., and Tallon-Baudry, C. (). Unconscious associative memory affects visual processing before  ms. Journal of Vision (): –. Churchland, P. M. (). Scientific Realism and the Plasticity of Mind. Cambridge: Cambridge University Press. Churchland, P. M. (). Perceptual plasticity and theoretical neutrality: a reply to Jerry Fodor. Philosophy of Science : –. Churchland, P. S., Ramachandran, V., and Sejnowski, T. (). A critique of pure vision. In C. Koch and J. Davis (eds), Large-Scale Neuronal Theories of the Brain, –. Cambridge, Mass.: MIT Press. Clark, A. (). A Theory of Sentience. Oxford: Oxford University Press. Crouzet, S. M., Kirchner, H., and Thorpe, S. J. (). Fast saccades toward faces: face detection in just  ms. Journal of Vision (): –. Currie, G., and Sterelny, K. (). How to think about the modularity of mind-reading. In J. L. Bermúdez (ed.), Philosophy of Psychology: Contemporary Readings, –. New York: Routledge. Davies, M. (). Tacit knowledge and subdoxastic states. In A. George (ed.), Reflections on Chomsky, –. Oxford: Blackwell. Davis, M. (). Tacit knowledge and subdoxastic states. In C. MacDonald and G. Macdonald (eds), Philosophy of Psychology: Debates on Psychological Explanation. Oxford: Blackwell. Dehaene, S., and Changeux, J.-P. (). Ongoing spontaneous activity controls access to consciousness: a neuronal model for inattentional blindness. PLoS Biology: Dehaene, S., Changeux, J.-P., Naccache, L. Sackur, J., and Sergent, C. (). Conscious, preconscious, and subliminal processing: a testable taxonomy. Trends in Cognitive Science (): –. Dehaene, S., Kerszberg, M., and Changeux, J.-P. (). A neuronal model of global workspace in effortful cognitive tasks. Proceedings of the National Academy of Sciences : –. Dehaene, S., Sergent, C., and Changeux, J.-P. (). A neuronal network model linking subjective reports and objective physiological data during conscious perception. Proceedings of the National Academy of Sciences (): –. Delk, J. L., and Fillenbaum, S. (). Differences in perceived colour as a function of characteristic colour. American Journal of Psychology (): –. Delorme, A., Rousselet, G., Mace, M., and Fabre-Thorpe M. (). Interaction of bottom-up and top-down processing in the fast visual analysis of natural scenes. Cognitive Brain Research (): –. Dennett, D. (). Intentional systems. Journal of Philosophy (): –. Deroy, O. (). Object-sensitivity versus cognitive penetrability of perception. Philosophical Studies : –. Desimone, R. (). Visual attention mediated by biased competition in extrastriate visual cortex. Philosophical Transactions of the Royal Society B : –. Desimone, R., and Duncan, J. (). Neural mechanisms of selective visual attention. Annual Review of Neurosciences : –. Dienes, Z., and Perner, J. (). Unifying consciousness with explicit knowledge. In A. Cleeremans (ed.), The Unity of Consciousness: Binding, Integration, and Dissociation, –. Oxford: Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

Dokic, J. (). Seeds of knowledge: noetic feelings and metacognition. In M. Beran, J. L. Brandl, J. Perner, and J. Proust (eds), Foundations of Metacognition, –. Oxford: Oxford University Press. Dretske, F. (). Seeing and Knowing. Chicago: University of Chicago Press. Dretske, F. (). Knowledge and the Flow of Information. Cambridge, Mass.: MIT Press. Dretske, F. (). Conscious experience. Mind : –. Repr. in A. Noë and E. Thompson (eds), Vision and Mind. Cambridge, Mass.: MIT Press, . Dreyfus, H. (). What Computers Still Can’t Do: A Critique of Artificial Reason. Cambridge, Mass.: MIT Press. Duhem, P. (). The Aim and Structure of Physical Theory, nd edn, trans. P. P. Wiener. New York: Atheneum. Duncan, J. (). Converging levels of analysis in the cognitive neuroscience of visual attention. Philosophical Transactions of the Royal Society B : –. Egeth, H. E., Virzi, R. A., and Garbart, H. (). Searching for conjunctively defined targets. Journal of Experimental Psychology : –. Epstein, W., and Broota, K. (). Automatic and attentional components in perception of sizeat-a-distance. Perception and Psychophysics (): –. Eriksen, C., and Yeh, Y. (). Allocation of attention in the visual field. Journal of Experimental Psychology: Human Perception and Performance : –. Ernst, M., Banks M., and Bulthoff, M. (). Touch can change visual slant perception. Nature Neuroscience : –. Evans, G. (). The Varieties of Reference. Oxford: Oxford University Press. Feyerabend, P. (). Explanation, reduction, and empiricism. In H. Feigl and G. Maxwell (eds), Minnesota Studies in the Philosophy of Science, vol. , –. Minneapolis: Minnesota University Press. Fish, W. (). High-level properties and visual experience. Philosophical Studies : –. Fodor, J. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, J. (). A reply to Churchland’s ‘Perceptual plasticity and theoretical neutrality’. Philosophy of Science (): –. Fodor, J. A. (). The Mind Does Not Work That Way: The Scope and Limits of Computational Psychology. Cambridge, Mass.: MIT Press. Freuder, E. C. (). A computer vision system for visual recognition using active knowledge. Cambridge, Mass.: MIT AI Lab Tech. Rep. . Gallese, V., and Lakoff, G. (). The brain’s concepts: the role of the sensory-motor system in conceptual knowledge. Cognitive Neuropsychology : –. Gao, T., and Scholl, B. J. (). Chasing vs. stalking: interrupting the perception of animacy. Journal of Experimental Psychology: Human Perception and Performance : –. Goodale, M., and Milner, D. (). Sight Unseen. Oxford: Oxford University Press. Gregory, R. (). The Intelligent Eye. New York: McGraw-Hill. Gregory, R. (). Concepts and Mechanisms of Perception. New York: Scribner’s. Grill-Spector, K., Henson, R., and Martin, A. (). Repetition and the brain: neural models of stimulus-specific effects. Trends in Cognitive Sciences : –. Grill-Spector, K., Kushnir, T., Hendler, T., Edelman, S., Itzchak, Y., and Malach, R. (). A sequence of object-processing stages revealed by fMRI in the human occipital lobe. Human Brain Mapping : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



Hansen, T., Olkkonen, M., Walter, S., and Gegenfurtner, K. (). Memory modulates colour appearance. Nature: Neuroscience (): –. Hanson, N. R. (). Patterns of Discovery. Cambridge: Cambridge University Press. Haugeland, J. (). Mind embodied and embedded. In Having Thought: Essays in the Metaphysics of Mind, –. Cambridge, Mass.: Harvard University Press. Heck, R. G. (). Nonconceptual content and the space of reasons. Philosophical Review : –. Holmes, G. (). Disturbances in visual orientation. British Journal of Ophthalmology : –. Hommel, B. (). Event files: feature binding in and across perception and action. Trends in Cognitive Sciences (): –. Humphreys, G., and Riddoch, J. (). To See But Not to See: A Case Study of Visual Agnosia. Hillsdale, NJ: Erlbaum. Humphreys, G., and Riddoch, J. (). Detection by action: neuropsychological evidence for action-defined templates in search. Nature: Neuroscience (): –. Humphreys, G., and Riddoch, J. (). How to define an object: evidence from the effects of action on perception and attention. Mind and Language (): –. Hurley, S. (). Consciousness in Action. Cambridge, Mass.: Harvard University Press. Israel, D., Perry, J., and Tutiya, S. (). Executions, motivations and accomplishments. Philosophical Review (): –. Jackendoff, R. (). Consciousness and the Computational Mind. Cambridge, Mass.: MIT University Press. Jackendoff, R. (). Languages of the Mind. Cambridge, Mass.: MIT Press. Jackson, C. V. (). Visual factors in auditory localization. Quarterly Journal of Experimental Psychology : –. Johnson, J. S., and Olshausen, B. A. (). The earliest EEG signatures of object recognition in a cued-target task are postsensory. Journal of Vision : –. Kahneman, D., Treisman, A., and Gibbs, B. (). The reviewing of object files: object specific integration of information. Cognitive Psychology : –. Kirchner, H., and Thorpe, S. J. (). Ultra-rapid object detection with saccadic movements: visual processing speed revisited. Vision Research : –. Kitcher, P. (). Real realism: the Galilean strategy. Philosophical Review (): –. Klatzky, R. L., Wu, B., and Stetten, G. (). The disembodied eye: consequences of displacing perception from action. Vision Research : –. Koch, C., and Tsuchiya, N. (). Attention and consciousness: two distinct brain processes. Trends in Cognitive Science (): –. Koechlin, E., Ody, C., and Kouneiher, F. (). The architecture of cognitive control in the human prefrontal cortex. Science (): –. Koriat, A. (). The feeling of knowing: some metatheoretical implications for consciousness and control. Consciousness and Cognition : –. Koriat, A. (). Metacognition and consciousness. In P. D. Zelazo, M. Moscovitch, and E. Thompson (eds), Cambridge Handbook of Consciousness, –. New York: Cambridge University Press. Kosslyn, S. M. (). Image and Brain. Cambridge, Mass.: MIT Press. Kravitz, D. J., and Behrmann, M. (). Space-, object-, and feature-based attention interact to organize visual scenes. Attention, Perception, and Psychophysics (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

Kriegel, U. (). Naturalizing subjective character. Philosophy and Phenomenological Research (): –. Kriegel, U. (). The phenomenologically manifest. Phenomenology and the Cognitive Sciences : –. Kuhn, T. (). The Structure of Scientific Revolutions. Chicago: University of Chicago Press. Kukla, A. (). Social Constructivism and the Philosophy of Science. London: Routledge. Lamme, V. A. F. (). The neurophysiology of figure–ground segregation in primary visual cortex. Journal of Neuroscience : –. Lamme, V. A. F. (). Neural mechanisms of visual awareness: a linking proposition. Brain and Mind : –. Lamme, V. A. F. (). Why visual attention and awareness are different. Trends in Cognitive Sciences (): –. Lamme, V. A. F. (). Independent neural definitions of visual awareness and attention. In A. Raftopoulos (ed.), The Cognitive Penetrability of Perception: An Interdisciplinary Approach, –. Hauppauge, NY: NovaScience. Lamme, V. A. F., and Roelfsema, P. R. (). The distinct modes of vision offered by feed forward and recurrent processing. Trends in Neuroscience : –. Lamme, V. A. F., and Spekreijse, H. (). Contextual modulation in primary visual cortex and scene perception. In M. Gazzaniga (ed.), The New Cognitive Neurosciences, nd edn, –. Cambridge, Mass.: MIT Press. Langton, S. R. H., Watt, R. J., and Bruce, V. (). Do the eyes have it? Cues to the direction of social attention. Trends in Cognitive Sciences (): –. Lashley, K. (). Brain Mechanisms and Intelligence: A Quantitative Study of Injury to the Brain. Chicago: Chicago University Press. Levin, D., and Banaji, M. (). Distortions in the perceived lightness of faces: the role of race categories. Journal of Experimental Psychology: General : –. Liu, H., Agam, Y., Madsen, J., and Krelman, G. (). Timing, timing, timing: fast decoding of object information from intracranial field potentials in human visual cortex. Neuron : –. Lowe, E. J. (). Sortals and the individuation of objects. Mind and Language : –. Lyons, J. C. (). Carving the mind at its (not necessarily modular) joints. British Journal for the Philosophy of Science : –. Lyons, J. (). Perception and Basic Beliefs. New York: Oxford University Press. Lyons, J. C. (). Circularity, reliability, and the cognitive penetrability of perception. Philosophical Issues : –. Macpherson, F. (). Ambiguous figures and the content of experience. Noûs (): –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. Marr, D. (). Vision: A Computational Investigation into Human Representation and Processing of Visual Information. San Francisco, Calif.: Freeman. Repr.  by MIT Press. Masrour, F. (). Is perceptual phenomenology thin? Philosophy and Phenomenological Research (): –. Matthen, M. (). Seeing, Doing, and Knowing: A Philosophical Theory of Sense Perception. Oxford: Clarendon Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



McClelland, J., Rumelhart, D., and the PDP research group (). Parallel Distributed Processing: Explorations in the Microstructure of Cognition, vol. . Cambridge, Mass.: MIT Press. McGurk, H., and MacDonald, J. (). Hearing lips and seeing voices. Nature : –. Michotte, A. (). The Perception of Causality. Oxford: Basic Books. Milner, A., and Goodale, M. (). The Visual Brain in Action. New York: Oxford University Press. Mole, C. (). Attention Is Cognitive Unison. New York: Oxford University Press. Nanay, B. (). Perceptual phenomenology. Philosophical Perspectives : –. Nobre, A. C., Rohenkhol, G., and Stokes, M. G. (). Nervous anticipation: top-down biasing across space and time. In M. Posner (ed.), Cognitive Neuroscience of Attention, nd edn, –. New York: Guilford Press. Noe, A. (). Action in Perception. Cambridge, Mass.: MIT Press. Olkkonen, M., Hansen, T., and Gegenfurtner, K. (). Color appearance of familiar objects: effects of object shape, texture, and illumination changes. Journal of Vision (): –. O’Regan, K. (). Why Red Doesn’t Sound Like a Bell. Oxford: Oxford University Press. Pacherie, E. (). The phenomenology of action: a conceptual framework. Cognition : –. Peacocke, C. (). A Study of Concepts. Cambridge, Mass.: MIT Press. Perky, C. W. (). An experimental study of imagination. American Journal of Psychology : –. Peterson, M. A. (). Overlapping partial configurations in object memory. In M. A. Peterson and G. Rhodes (eds), Perception of Faces, Objects, and Scenes, –. New York: Oxford University Press. Peterson, M. (). Object perception. In B. Goldstein (ed.), Blackwell Handbook of Sensation and Perception, –. Oxford: Blackwell. Peterson, M., and Enns, J. (). The edge complex: implicit memory for figure assignment in shape perception. Perception and Psychophysics (): –. Posner, M. I. (). Orienting of attention. Quarterly Journal of Experimental Psychology : –. Prinz. J. J. (). Is the mind really modular? In R. J. Stainton (ed.), Contemporary Debates in Cognitive Science, –. Oxford: Blackwell. Prinz, J. J. (). When is perception conscious? In B. Nanay (ed.), Perceiving the World, –. Oxford: Oxford University Press. Proffitt, D., Stefanucci, J., Banton, T., and Epstein, W. (). The role of effort in perceiving distance. Psychological Science : –. Pylyshyn, Z. (). Computation and cognition: issues in the foundation of cognitive science. Behavioral and Brain Sciences : –. Pylyshyn, Z. (). Computation and Cognition. Cambridge, Mass.: MIT Press. Pylyshyn, Z. (). Is vision continuous with cognition? Behavioral and Brain Sciences : –. Pylyshyn, Z. (). Seeing and Visualizing: It’s Not What You Think. Cambridge, Mass.: MIT Press. Pylyshyn, Z. (). Things and Places: How the Mind Connects with the World. Cambridge, Mass.: MIT Press. Pylyshyn, Z., and Storm, R. W. (). Tracking multiple independent targets: evidence for a parallel tracking mechanism. Spatial Vision : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

Raftopoulos, A. (a). Is perception informationally encapsulated? The issue of the theoryladenness of perception. Cognitive Science : –. Raftopoulos, A. (b). Re-entrant pathways and the theory-ladenness of observation. Philosophy of Science : –. Raftopoulos, A. (). Defending realism on the proper ground. Philosophical Psychology (): –. Raftopoulos, A. (). Cognition and Perception. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Ambiguous figures and representationalism. Synthese (): –. Raftopoulos, A. (). The cognitive impenetrability of the content of early vision is a necessary and sufficient condition for purely nonconceptual content. Philosophical Psychology (): –. Raftopoulos, A. (forthcoming). The cognitive impenetrability of perception and theoryladenness. Journal of General Philosophy of Science. Raftopoulos, A., and Muller, V. (). Nonconceptual demonstrative reference. Philosophy and Phenomenological Research (): –. Rensink, R. A. (a). The dynamic representation of scenes. Visual Cognition : –. Rensink, R. A. (b). Seeing, sensing, and scrutinizing. Vision Research : –. Riddoch, J. M, Humphreys, G. W., Edwards, S., Baker, T., and Wilson, K. (). Seeing the action: neuropsychological evidence for action-based effects on object selection. Nature: Neuroscience (): –. Rock, I. (). The Logic of Perception. Cambridge, Mass.: MIT Press. Roelfsema, P. R. (). Elemental operations in vision. Trends in Cognitive Science (): –. Roelfsema, P. R., Lamme, V. A. F., and Spekreijse, H. (). The implementation of visual routines. Vision Research : –. Rothi, L. J., Ochipa, C., and Heilman, K. M. (). A cognitive neuropsychological model of limb praxis. Cognitive Neuropsychology : –. Rubens, A., and Benson, D. (). Associative visual agnosia. Archives of Neurology : –. Sacks, O. (). The Mind’s Eye. New York: Knopf. Segal, S. J. (). Processing of the stimulus in imagery and perception. In S. J. Segal (ed.) Imagery: Current Cognitive Approaches, –. New York: Academic Press. Segal, S. J., and Fusella, V. (). Effects of images in six sense modalities on detection (d’) of visual signal from noise. Psychonomic Science : –. Sergent, C., Baillet, S., and Dehaene, S. (). Timing of the brain events underlying access to consciousness during the attentional blink. Nature: Neuroscience : –. Shams, L., Kamitani., Y, Shimojo, S. (). What you see is what you hear. Nature : . Shea, N. (). Distinguishing top-down from bottom-up effects. In S. Biggs, M. Matthen, and D. Stokes (eds), Perception and Its Modalities, –. Oxford: Oxford University Press. Siegel, S. (). Which properties are represented in perception? In T. S. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Siegel, S. (). The Contents of Visual Experience. Oxford: Oxford University Press. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Siegel, S. (). The epistemic impact of the etiology of experience. Philosophical Studies : –. Spelke, E. S. (). Principles of object perception. Cognitive Science : –. Spence, C. (). Crossmodal correspondences: a tutorial review. Attention Perception & Psychophysics (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

introduction



Sperling, G., and Melchner, M. J. (). The attention operating characteristic: examples from visual search. Science (): –. Stich, S. (). Beliefs and subdoxastic states. Philosophy of Science : –. Stich, S., and Nichols, S. (). Folk psychology: simulation or tacit theory? Mind and Language (): –. Stokes, D. (). Perceiving and desiring: a new look at the cognitive penetrability of experience. Philosophical Studies (): –. Strawson, G. (). Mental Reality. Cambridge, Mass.: MIT Press. Super, H., van der Togt, C., Spekreijse, H., and Lamme, V. A. (). Internal state of monkey primary visual cortex (V) predicts figure–ground perception. Journal of Neuroscience : –. Tanner, W. P., and Swets, J. A. (). A decision-making theory of human detection. Psychological Review : –. Treisman, A. (). The Perception of features and objects. In A. Baddeley and L. Weiskrantz (eds), Attention: Selection, Awareness, and Control, –. Oxford: Oxford University Press. Tversky, A., and Kahneman, D. (). Judgment under uncertainty: heuristics and biases. Science : –. Tye, M. (). The Imagery Debate. Cambridge, Mass.: MIT Press. Tye, M. (). Ten Problems of Consciousness. Cambridge, Mass.: MIT Press. Tye, M. (). Consciousness, Colour, and Content. Cambridge, Mass.: MIT Press. Tye, M. (). Consciousness Revisited: Materialism without Phenomenal Concepts. Cambridge, Mass.: MIT Press. Ullman, S., Vidal-Naquet, M., and Sali, E. (). Visual features of intermediate complexity and their use in classification. Nature: Neuroscience (): –. Ungerleider, L., and Haxby, J. (). ‘What’ and ‘where’ in the human brain. Current Opinion in Neurobiology : –. Ungerleider, L., and Mishkin, M. (). Two cortical visual systems. In D. Ingle, M. Goodale, and R. Mansfield (eds), Analysis of Visual Behavior, –. Cambridge, Mass.: MIT Press. van Boxtel. J. J. A., Naotsugu Tsuchiya, N., and Koch, C. (). Consciousness and attention: on sufficiency and necessity. Frontiers in Psychology :. doi: ./ fpsyg... Van Essen, D. C. (). Functional organization of primate visual cortex. In A. Peters and E. Jones (eds), Cerebral Cortex, vol. : Visual Cortex, –. New York: Plenum. VanRullen, R., and Thorpe, S. J. (). The time course of visual processing: from early perception to decision making. Journal of Cognitive Neuroscience : –. Vecera, S. (). Toward a biased competition account of object-based segregation and attention. Brain and Mind : –. Vecera, S. P., and O’Reilly, R. C. (). Figure–ground organization and object recognition processes: an interactive account. Journal of Experimental Psychology: Human Perception and Performance : –. Vishton, P., Stephens, N., Nelson, L., et al. (). Planning to reach for an object changes how the reacher perceives it. Psychological Science : –. Wagner, L., and Carey, S. (). -month-old infants represent probable endings of motion events. Infancy (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos and j. zeimbekis

Warrington, E. K., and Taylor, A. M. (). The contribution of the right parietal lobe to object recognition. Coflex : –. Warrington, E. K., and Taylor, A. M. (). Two categorical stages of object recognition. Perception : –. Weiskrantz, L. (). Blindsight: not an island unto itself. Current Directions in Psychological Science (): –. Wolfe, J., Horowitz, T., Kenner, N., Hyle, M., and Vasan, N. (). How fast can you change your mind? The speed of top-down guidance in visual search. Vision Research : –. Wolter, J., and Lund, O. (). Reaction of centrifugal nerves in the human retina. American Journal of Ophthalmology : –. Wu, W. (). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies : –. Zeimbekis, J. (). Color and cognitive penetrability. Philosophical Studies (): –. Zeki, S. (). The mapping of visual function in the cerebral cortex. In Y. Katsuki, R. Norgren, and S. Masayasu (eds), Brain Mechanisms of Sensation, –. Chichester: Wiley. Zeki, S. (). A Vision of the Brain. Blackwell.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

PA RT I

Definitional and Methodological Issues

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Cognitive Penetrability A No-Progress Report Edouard Machery

Recent empirical research1 and philosophical discussion2 have revived interest in the once influential hypothesis that perception is cognitively penetrable—roughly, that what one believes and what one desires influence what one perceives. Psychologists and philosophers extensively debated the cognitive penetrability hypothesis put forward by New Look psychologists such as Jerome Bruner, and this debate led to widespread (though not universal) doubts about its truth in the s and s. These doubts are now fading, and an increasing number of psychologists and philosophers are embracing this hypothesis. I will call this new movement the ‘New New Look’. In this chapter, my goal will be to rekindle the doubts concerning the cognitive penetrability hypothesis: I will argue that current empirical research on, and current theoretical arguments for, this hypothesis suffer from the exact same kind of problems that in the first place resulted in widespread doubts. I will not argue that the cognitive penetrability hypothesis is likely to be false or that there are strong arguments and persuasive empirical findings against it. As far as I know, this hypothesis might well be true, but I will argue that arguments and findings allegedly justifying a second look at the cognitive penetrability hypothesis are vitiated by well-known problems. Here is how I will proceed. In Section , I will clarify the cognitive penetrability hypothesis and distinguish several formulations of this hypothesis. In Section , I will discuss a traditional theoretical problem that nowadays, just like thirty years ago, makes it difficult to support this hypothesis (a problem I will call ‘the locus problem’): many studies are ill-equipped to ascertain where exactly in the information processing flow emotions, desires, beliefs, motivations, etc., exert their causal influence. 1 See e.g. Bhalla and Proffitt ; Balcetis and Dunning ; ; ; Hansen et al. ; Alter and Balcetis ; Proffitt et al. . 2 Raftopoulos ; ; ; Macpherson ; Siegel ; Stokes ; ; Wu ; Zeimbekis .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

In Sections  and , I will turn toward the empirical findings alleged to support the cognitive penetrability hypothesis. In Section , I will highlight methodological concerns with some empirical studies about cognitive penetrability thirty years ago as nowadays. In Section , I will emphasize the replication problem that vitiates the past and present empirical literature. Before proceeding, I must say a few words about the methodology followed in this chapter. To illustrate the methodological problems that plague past and present research on cognitive penetration, I will focus on a few articles. One could perhaps be concerned that the problems discussed below are only found in these and perhaps a few other articles, but do not plague a whole field. This concern can be set aside for three reasons. The articles discussed below are among the most cited and among the most influential. If they suffer from crippling methodological flaws, the philosophical and psychological conclusions drawn from them are unjustified. Furthermore, these articles are typical of the kind of research done on cognitive penetration, and the flaws in them are found in much of this research. Finally, I will also cite other articles where similar flaws are found.

 What is Cognitive Penetration? To clarify the cognitive penetrability hypothesis, it is useful to organize the discussion around three questions: . What is doing the penetrating? . What counts as penetration? . What is penetrated?

. What is doing the penetrating? The cognitive penetrability hypothesis is true if propositional attitudes, in particular beliefs but also desires (Stokes ), penetrate perception. In contrast, the penetration of perception by other perceptual states (e.g. the penetration of vision by auditory states) does not make it true. So, the McGurk effect (McGurk and MacDonald ) does not provide support for the cognitive penetrability hypothesis. Propositional attitudes like beliefs and desires and perceptual states do not exhaust the whole gamut of mental states: These also include emotions and moods (supposing that emotions and moods are not reducible to propositional attitudes), subdoxastic states of various kinds, including the mental states tapped into by the implicit measures of biases such as the Implicit Association Test (for discussion of the nature of these states, see Machery forthcoming), and the unconscious valuations of events or possible courses of action. Whence the question: should the penetration of perception by mental states other than beliefs and desires count as a form of cognitive penetrability (see e.g. Stefanucci et al.  for a study of the effect of fear on slant perception)?

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



While the answer to this question may well vary across these kinds of mental state, it is affirmative for many of them. To a large extent, we care about the penetrability of perception for epistemological reasons: prima facie, at least, perceptual states would be a poor guide for determining what to believe and consequently how to act if what we believe and desire influenced what we perceive. This point is as true of propositional attitudes such as beliefs and desires as it is of emotions, moods, and at least some types of sudoxastic state. So, if these states—for instance, the states tapped into by implicit measures of bias—influence perception, this should count as a form of cognitive penetrability. What about the influences of personality, cognitive style, experience, learning, education, and development on perception? Do these count as forms of cognitive penetrability? Revisiting old (and controversial) data sets about the non-universality of the Müller–Lyer illusion, McCauley and Henrich () have recently argued that experience with vertical lines during the development of the perceptual systems influences perception—a phenomenon they call ‘diachronic penetrability’. Nisbett and Miyamoto () have reviewed the evidence that perception is influenced by the cognitive style of the perceiver. Finally, Goldstone () has shown that learning can influence perceptual discrimination: dimensions that are relevant in a learning task become more salient and more easily discriminable in follow-up tasks. On the one hand, the epistemological consideration just mentioned militates for treating these alleged causal influences as instances of cognitive penetration. In addition, New Look psychologists were interested in all kinds of influence on perception, including personality, age, and motivation (e.g. Bruner and Goodman ; Erdelyi ). On the other hand, these causal influences may differ from the kind of causal relations that are required for an influence on perception to count as penetration (Section .), or they may influence judgments rather than perception (Section .). In particular, the influence of cognitive style on vision may either be due to patterns of attention or of inspection of visual scenes that vary across cognitive styles, or it may boil down to the influence of cognitive style on perceptual judgment instead of perception. The evidence reviewed by Nisbett and Miyamoto is entirely consistent with these two hypotheses, but unfortunately they do not address them. For these reasons, in what follows I will overlook these forms of influence on perception.

. What counts as penetration? Cognitive penetration is a causal relation: the cognitive penetrability hypothesis is true only if beliefs, desires, or some other nonperceptual states causally influence what we perceive. But not every type of causal relation between these states and perception counts as cognitive penetration. Beliefs and desires influence what we perceive because they influence where we look at, what we touch, what we listen to, etc. The cognitive penetrability hypothesis would be trivially true if such influences counted as instances of cognitive penetration. So, only a specific kind of causal influence on perception counts as cognitive penetration.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

There have been several attempts to identify which kind of causal relation is required for a causal influence to count as an instance of cognitive penetration (e.g. Pylyshyn ; Macpherson ; Stokes ), but no consensus has emerged. For instance, according to Stokes (: ): ‘A perceptual experience E is orectically penetrated iff () E is causally dependent upon some orectic state D and () the causal link between E and D is internal’, while, according to Pylyshyn (: ): ‘If a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs.’ So, Stokes refers, while Pylyshyn does not, to an internal causal relation, and Pylyshyn insists on the existence of a semantic relation (whose nature is unfortunately left vague and obscure) between, on the one hand, beliefs and desires and, on the other, perceptual experience (see also Macpherson ), while Stokes does not. Be it as it may, philosophers and some psychologists tend to insist that cases where beliefs, desires, etc. influence what we perceive because they influence what we attend to are not instances of cognitive penetration.3 Thus, according to Pylyshyn’s definition quoted, attention-mediated influence does not count as an instance of perceptual penetration. The rationale for this exclusion is twofold. Counting such cases as instances of cognitive penetration would trivialize the cognitive penetrability hypothesis. In addition, the influence of mental states on perception through attention is similar to their influence on perception through the direction of gaze, the location of touch, etc., and the latter does not count as an instance of cognitive penetration. In that respect, philosophers seem to differ from many New Look psychologists, who were interested in the many different ways in which propositional attitudes, emotions, motivation, cognitive style, personality, etc. (Section .) influence perception. For instance, Erdelyi (: ) includes the influence of motivation and values on gaze fixation as an instance of the phenomenon of interest, and Haber () views the effect of attention on perceptual experience (‘perceptual enhancement’) as one of the central claims made by New Look theorists. The same is true of some New New Look psychologists, such as Aarts et al. (), who investigate the influence of bodily perception (e.g. thirst) on perceptual attention (which, following the New Look, they call ‘perceptual readiness’). In this chapter, I will not attempt to specify in more detail what is required for a causal influence to count as an instance of cognitive penetration, but I will be following the philosophical consensus, and I will not count influences mediated by attention as instances of cognitive penetration.4

3

On the other hand, Stokes’s definition above does not exclude this as a form of penetration. It is worth noting a puzzling feature of the philosophical discussion of cognitive penetrability. If philosophers are primarily interested in the cognitive penetrability hypothesis for epistemological reasons, as suggested, it is strange to set aside the attention-mediated causal influence of beliefs, desires, emotions, etc. on perception, since the latter raises exactly the same kind of epistemological problems as a causal influence that would not be mediated by attention. 4

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



. What is penetrated? There are at least two distinct ways of unpacking the claim that perception is penetrated by mental states such as beliefs, desires, and emotions. On the one hand, philosophers are typically interested in the causal influence of beliefs, desires, or other mental states on perceptual experience, i.e. on what we consciously see, hear, smell, etc. On this view, which is illustrated by Stokes’s definition above, the cognitive penetrability hypothesis is true if and only if what we experience perceptually is causally influenced by what we believe, desire, etc. It is worth noting that this view remains insufficiently specified, since beliefs, desires, etc. could influence the semantic content of perceptual experience, its phenomenological character, or both.5 On the other hand, for Pylyshyn (; ) and others, the cognitive penetrability hypothesis bears on the possible causal influence of beliefs, desires, or other mental states on the processes involved in early perception, in particular in early vision. Because these processes probably take place before the processes that determine our perceptual experience, the causal influence of beliefs, desires, or other mental states on perceptual experience falls short of establishing that early perception—particularly early vision—is cognitively penetrable. In any case, both versions of the cognitive penetrability hypothesis distinguish between perception (either perceptual experience or the unconscious processes involved in early perception) and perceptual judgments—judgments about what is currently perceived. Evidence that perceptual judgments are influenced by beliefs, desires, emotions, biases, etc. provides no support for the cognitive penetrability hypothesis.

 The Locus Problem . The locus problem in New Look studies Consider the classic experiment by Delk and Fillenbaum (), an experiment still cited approvingly in the recent philosophical and psychological literature (e.g. Macpherson ; Stokes ). In this experiment, participants are asked to match the perceived color of ten two-dimensional cut-outs from the same sheet of orange-red cardboard to that of a color-mixer (i.e. a patch whose color can be adjusted from light to dark red). The color-mixer was set behind the figures, and served as a ground for the figures. The color-mixer and the cut-outs could be seen through an aperture in a black screen that was covered with a waxed paper. They were ‘quite dimly’ illuminated by a

5 Although I will not press this point further here, the discussion of the cognitive penetrability hypothesis in philosophy relies on a naïve view of the nature of experience. It assumes that perceptual experience is a single state (instead of a conglomerate of several states) with a definite (instead of a vague) content and phenomenology that is introspectively accessible (instead of opaque). For early discussion of this issue, see Erdelyi ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

fluorescent light attached to the screen (the lights of the lab were turned off, and the shades were drawn). In the red-object condition, the reddish cut-outs had the shape of typically red objects (e.g. a heart or a pair of lips); in the non-red object condition, they had the shape of typically non-red objects (e.g. a mushroom); and in the control condition they had a geometrical form (e.g. a circle). Delk and Fillenbaum reported that participants chose a darker red for the three typically red objects than for the three geometrical figures and for the three typically non-red objects (there was no difference between those last two conditions). They conclude (: ) that ‘past association of color and form does in some way influence perceived color’. Taking Delk and Fillenbaum’s research, for the time being, at face value (but see Section .), their results do suggest that people’s knowledge of the typical color of objects influences the color they report perceiving, but what they do not clearly show is whether people make these reports because they are perceiving differently the color of cut-outs when these have geometrical shapes or the shapes of red objects, or because they use their knowledge of the objects represented by the D cutouts (e.g. the typical color of a heart) to estimate the color of the perceived cutouts and to thus determine the appropriate color of the color-mixer (for further discussion of this body of research, see Zeimbekis ). One would expect people to take into account their background beliefs about typical colors to assess the color of perceived objects, since failure to take into account the typical color of a kind of object to assess the color of an object of this kind would be an instance of the base-rate neglect illustrated by Kahneman and Tversky’s research (e.g. ). Perhaps one may respond that Delk and Fillenbaum’s () results clearly show that knowledge influences the experience of color rather than perceptual judgment, since subjects were asked to match two colors together: they just needed to compare two perceptual experiences, and they could thus bypass assessing or making a perceptual judgment about the color of the perceived shape. However, the difficulty of the perceptual task (Section .) probably primed participants to think about the color of the cut-outs. In addition, even in a matching task, participants are likely to try to classify what it is that they are perceiving and to use this classification to solve the matching task. To shine a crude light on the problem at hand, consider the following situation. You are hiking in the Australian bush, knowing that very nasty snakes are often found on the trails. If you’re like me, you spontaneously lower the threshold for the recognition of a snake (a phenomenon known as ‘perceptual vigilance’): any object that vaguely looks like a snake (e.g. any object with an elongated shape or any shiny object), anything that moves around the trail, etc., causes you to judge that you are seeing a snake. This is an instance of a general process: in everyday life as in science, we require more or less evidence before making a judgment depending on what is at stake. What the snake recognition example shows is that this process need not be intentional, but can be automatic. The concern, then, is that alleged cognitive-penetration effects are in fact instances of this well-known, commonsensical process. Or, at

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



least, that the evidence put forward fails to distinguish between them being genuine instances of cognitive penetration or mere judgmental effects. The locus of the effect is underdetermined by the available evidence (Garner et al. ; Bolles et al. ; Haber ; Pylyshyn; : Section ; Macpherson ). Furthermore, the locus problem does not simply bear on the contrast between perceptual experience and perceptual judgment. It is sometimes unclear whether beliefs, desires, and other mental states influence the memories of perceptual experiences rather than the experiences themselves. For instance, in an early formulation of the locus problem, Haber (: ) wrote: ‘There is a crucial difference between reporting one’s experience and reporting the attributes of stimuli that one remembers seeing. One task is perceptual, while the other deals with a memorial process.’ Sometimes, it is also unclear whether experience or the encoding of the experience is influenced (Haber ). Signal detection theory was used to sharpen the concern discussed in this section. According to signal detection theory, a perceptual judgment depends on two criteria, called d’, the sensitivity parameter, and β, the response bias parameter. d’ measures the standardized distance between the noise distribution and the signal-plus-noise distribution; the more sensitive the measuring instrument (e.g. a sense), the less spread-out the noise distribution, and the larger d’ is. β measures how large the measured signal must be for the subject to decide to report that she is perceiving something. The more cautious she is, the larger β will be. Signal detection theory was applied to solve the locus problem under the assumption that, if the occurrence of the penetration effect depends on how subjects set up β, then the locus of the effect has to be the decision made by subjects or their perceptual judgment. Reviews suggest that some effects allegedly supporting the cognitive penetrability hypothesis, but not all, are due to a change in β (Erdelyi ; Pylyshyn : Section ). Unfortunately, the relation between, on the one hand, β and perceptual judgment and, on the other, d’ and perception is more controversial than was assumed by proponents and critics of New Look psychology (Pylyshyn : Section ), and signal detection theory does not provide a simple, handy tool to solve the frustrating locus problem. Remarkably, at least some New Look psychologists were aware that their studies did not unambiguously provide evidence for the penetration of perceptual experience in contrast to the causal influence of motivation or values on perceptual judgments; in addition, they judged that the issue of whether motivation influenced perceptual judgment or perceptual experience was not particularly important. Tajfel wrote (: ), ‘it is not particularly important to find out whether the stimuli are really ‘seen’ as larger or smaller. They are reproduced as such,’ adding (p. ): ‘Once again, this does not necessarily mean that the subjects “see” the stimuli in the value series as being more different from each other, or larger, or smaller. They respond as if they perceived them as such, as this is as far as the phenomenology of it can go for the time being.’ Regrettably, contemporary philosophers who appeal to these results do not show similar cautious restraint in interpreting them.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

. The locus problem in New New Look studies One would think that New New Look psychology would improve on the New Look psychology by dealing head-on and convincingly with the well-known locus problem, but this would be a mistake: New New Look suffers from this problem too (e.g. Durgin et al. ). Consider for instance Balcetis and Dunning’s () two well-known studies. Participants are asked to cross a quad while wearing an embarrassing costume (study ) or to climb a hill while kneeling on a skateboard (study ). Cognitive dissonance is manipulated by influencing whether participants feel that they have freely chosen to complete the task. They are then asked to estimate distances or slants. Balcetis and Dunning report that people with high cognitive dissonance (i.e. those who freely chose to complete the task) estimated the space as shorter and the hill as less steep, perhaps as a way to decrease their cognitive dissonance. They conclude (: ) that ‘[t]hese studies suggest that the impact of motivational states extends from social judgment down into perceptual processes’. However, in Balcetis and Dunning’s study , participants’ estimates were made from memory: participants reported the length of the quad after having crossed it. As a result, the locus of the effect is unclear. Is perceptual experience influenced? Or rather is the memory of the distance influenced? Only the former counts as an instance of cognitive penetration. Fortunately, Balcetis and Dunning’s second study asked participants to estimate the slant of the hill before the experiment (see Section . for other concerns with their studies). While this assuages the concern that the study examines the influence of cognitive dissonance on memory, it does little to assuage the concern that judgment, rather than experience, is what cognitive dissonance influences.

 Methodological Concerns . Methodological concerns with New Look studies New Look research was intensely controversial, and was subjected to an extraordinary amount of scrutiny. In particular, the experimental designs used by New Look psychologists were regularly criticized or found faulty (see e.g. Garner et al. ; Bolles et al.  on color perception). A common feature of the experiments meant to support the cognitive penetrability hypothesis is the use of degraded or hard-to-perceive stimuli. In Delk and Fillenbaum (), the color-mixer and the cut-outs are seen through waxed paper in a dimly illuminated room. Experiments on whether being given a set of possible responses influences perceptual experience (and on how this influence manifests itself) used stimuli that were flashed for a very short period of time (Haber : ). Using degraded or hard-to-perceive stimuli gives rise to at least three problems. First, as noted, it exacerbates the locus problem, since uncertainty about one’s visual experience

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



is likely to prime people to think about what they are actually perceiving. Second, it limits the psychological significance of the experimental findings because it is entirely unclear whether similar effects would be found for non-ambiguous, non-degraded, easy-to-perceive stimuli or, more generally, for objects perceived in normal conditions. Third, it limits the philosophical significance of these findings. As noted above, philosophers are largely concerned with the cognitive penetrability hypothesis because it seems to deprive perceptual experience of its distinctive role in the justification of beliefs. But, since it is dubious that degraded and ambiguous perceptual experience has any such role, the influence of beliefs, desires, emotions, etc. on this kind of experience is of little philosophical relevance. An even more serious problem is the obvious demand characteristics in the experiments meant to support the cognitive penetrability hypothesis. For instance, in Delk and Fillenbaum (), participants are shown in succession ten different cut-outs. Since they are asked to identify the color of the cut-outs by matching it to the color of a background (the color-mixer), it is reasonable for them to assume that not all cut-outs are of the same color (what would be the point otherwise?), and it is also reasonable to assume that the shape of the cut-outs is a cue to its actual color—for instance, that the heart cutout is redder than the square cut-out. Such demand characteristics would explain the results reported by Delk and Fillenbaum ().

. Methodological concerns in New New Look studies Unfortunately, New New Look psychological studies sometimes suffer from similar problems. Contemporary psychologists do not use degraded stimuli as often as New Look psychologists (but see the use of ambiguous figures in Balcetis and Dunning ); but they often use unfamiliar and thus difficult perceptual tasks. People are not expert at estimating the slant of a hill or of an inclined plane, and distance estimation is also very noisy. Whether the experimental effects found in these tasks generalize to routine perceptual situations is unclear. Durgin et al. () have shown that demand characteristics play an important role in the studies run by the heirs to the New Look psychology. Focusing on the famous backpack studies of Proffitt and colleagues (in particular, Bhalla and Proffitt , described in Section ), Russell and Durgin () as well as Durgin et al. () showed that the experimenters’ goals were transparent to subjects. Furthermore, both studies examined directly the role of demand characteristics on the occurrence of the experimental effect. The alleged cognitive penetration effect was found in the control condition, in which participants had no reasonable justification for the experimental manipulation, but not in the experimental condition, in which participants were provided with a plausible justification. Demand characteristics may vitiate other contemporary studies such as Balcetis and Dunning’s () studies discussed above (for other demand characteristics problems, see experiment  in Woods et al.’s () discussion of Witt et al. ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

Francis (a) has highlighted another problem with the recent revival of the New Look. Balcetis and Dunning () reported five studies suggesting that desirable objects (where desirability was manipulated in various ways) were viewed as being closer than undesirable objects (where distance was evaluated in various ways). Francis (a) noted that Balcetis and Dunning’s studies had a low power, and that it was consequently unlikely that Balcetis and Dunning had only obtained positive effects. Rather, their article plausibly suffers from a publication bias: experiments resulting in positive findings had been selectively reported (for further discussion, see Balcetis and Dunning  and Francis b).6 Recently, Firestone and Scholl () have emphasized another shortcoming of some famous New New Look studies: the ‘El Greco effect’. If experience of length, color, etc. is penetrated, penetration should influence both the property target (e.g. a color to be identified, a slant to be assessed, or a length to be evaluated) and the property to which the perceived object is matched. For instance, if widths are viewed as shorter, both the width of the target object and the width of any object participants are asked to match the width of the target object to should be viewed as shorter. Or, if reds are viewed as redder, both the red of the target object and the red of the color-mixer participants are asked to match the red of the target object to should be viewed as redder. As a result, there should be no difference between the conditions where penetration is manipulated and those where it is not. However, Firestone and Scholl report finding systematic differences between these conditions (see also Gross et al. ). Other studies suffer from idiosyncratic problems. Levin and Banaji () is cited approvingly by philosophers of psychology (Macpherson ; Siegel ; Stokes ). They attempt to show that expectations about blacks’ and whites’ skin color influence the perceptual experience of people’s actual skin color. In experiment , two base faces with stereotypical African American and Caucasian features were created. Importantly (Levin and Banaji : ), ‘these base images were then matched for both mean luminance and contrast, as measured by the mean and standard deviation, respectively, of the gray-level histogram of the faces as wholes.’ Lighter and darker versions of these two stereotypical faces were then created. Participants were repeatedly presented with stereotypically black or white faces and had to match their shading to that of a (stereotypically black or white) face whose shading could be manipulated. Levin and Banaji report that stereotypically black faces were matched to darker faces than stereotypically white faces. There is much to admire in Levin and Banaji’s careful experimental design, but the stimuli they use (: fig. , p. ) are problematic (Michal De-Medonsa, personal communication).7 While the luminance and contrast of the two base faces may be 6 In addition, two of their five studies reported a p-value of . for their main findings, which is unlikely to have occurred by sheer chance. Finally, Durgin, DeWald, Lechich, Li, and Ontiveros () investigate the demand characteristics of one of their studies. 7 I am assuming here that figure  reports the stimuli used in the experiment. Chaz Firestone made a similar remark at the annual meeting of the Society for Philosophy and Psychology in June .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



identical, the luminance is not similarly distributed in the bottom half of the two faces. Because we are likely to look at, or to pay attention to, different features of the stereotypically black and white faces, this difference in the distribution of luminance plausibly explains Levin and Banaji’s findings. Participants did have different perceptual experiences in experiment  because they looked at different features or parts of the stereotypically black and white faces, and because these were not equally illuminated.8

 RepliGate . Failure to replicate New Look studies Many classic studies meant to support the cognitive penetrability hypothesis turned out to be difficult to replicate reliably. Consider first the classic experiment by Bruner and Goodman (), an experiment still cited approvingly in recent philosophical and psychological literature (e.g. Stokes ). Coins of various values (penny, nickel, dime, quarter, and half-dollar) or cardboard disks of corresponding sizes were visually presented to -year-old children. Children were told to match the (modifiable) size of a patch of light to the size of the coins or of the disks, which they could do by turning a knob. Bruner and Goodman () reported that, in contrast to the size of the cardboard disks, the size of the coins was overestimated, and that the overestimation increased with the value of the coin (except for the half-dollar coin). In addition, in this group of participants, poor children overestimated the size of the coins more than wealthy children. Numerous exact and conceptual replications have been attempted, and many of these have failed (for a brief review, see the beginning of Landis et al. ). To cite only a few notable studies, Carter and Schooler () reported being unable to replicate the difference between poor and rich children’s assessment of the size of the coins as well as the finding that the size of coins is distinctively overestimated. Furthermore, the size of the coins, rather than their value, was the main determinant of over- and underestimation. Landis et al. () reported that some coins were underestimated, while others were overestimated, compared to disks. More recently, Furnham and SpencerBowdage () failed to find any relation between income and estimation of the size of coins, and they also found that coin size was underestimated, not overestimated as reported by Bruner and Goodman. Replication issues also hamper New Look color experiments. Delk and Fillenbaum’s () studies are part of a replication attempt after two similar color experiments had given inconsistent results (Harper ; Fisher et al. ). Bolles et al. () also showed that previous color studies did not replicate when the experimental 8 The other experiments reported in Levin and Banaji () suffer from more subtle problems (David Bitter, presentation at the conference ‘Concepts and Perception’, Riga, May ; Bitter ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

instruments were slightly modified. Similarly, the research on whether being given a set of possible responses influences subjects’ perceptual experience also suffered from unreliable attempts to replicate influential findings (e.g. Haber : –).

. Failure to replicate New New Look studies The New New Look research suffers from replication problems too. Proffitt et al. () is one of the most influential recent studies in this literature, and it belongs to an extensive and apparently successful research program. In the first experiment, subjects were asked to estimate distance from an egocentric point of view. Subjects in the experimental condition wore a heavy backpack, while subjects in the control condition did not. Proffitt and colleagues report that subjects’ estimates were larger in the experimental than in the control condition. Bhalla and Proffitt () reported a similar effect for the slant of a hill. However, Hutchison and Loomis (a) and Woods et al. () were unable to replicate Proffitt et al.’s () first experiment (for further discussion, see Proffitt et al.  and Hutchison and Loomis b). As noted, Durgin and colleagues have examined in great detail the possible source of the effect found by Proffitt and colleagues, arguing that it is due to demand characteristics (see also Durgin et al. ). Astonishingly, these replication issues, and more generally these methodological problems, are often overlooked when proponents of the cognitive penetrability hypothesis report Proffitt’s studies, which are treated as uncontroversial findings (e.g. Balcetis and Dunning ). The famous backpack studies are not the only ones to suffer from replication issues. In the first experiment of Witt et al. (), participants were asked to throw either a light or a heavy ball, and then to estimate distances verbally. They found that estimates were higher in the latter than in the former condition. The second experiment replicated the first, except for the fact that subjects were asked to match a distance between two poles (that could be moved further apart by the experimenters when instructed to do so by subjects) instead of making a verbal report. Witt et al. reported a result similar to that of experiment . Unfortunately, the second and third experiments of Woods et al. () failed to replicate Witt et al.’s results. Other studies that apparently provide support for the cognitive penetrability hypothesis have some replication issues, although these are much more minor than the issues just discussed. In a study reminiscent of Delk and Fillenbaum (), Hansen et al. () asked participants to adjust the color of images of fruits (e.g. a banana and a strawberry) until they looked gray (see also Olkkonen et al. ). Participants overshot the neutral gray point in the direction of the color opposite to the typical color of the fruit. For instance, they adjusted the color of the banana toward blue, suggesting, according to Hansen and colleagues, that the banana still appeared yellow at the neutral gray (a ‘memory color effect’). By contrast, when participants were presented with colored patches, they were accurate in identifying the neutral gray. Olkkonen et al. (: ) concluded that ‘the color appearance of familiar objects is affected by memory colors in a stimulus-dependent and an illumination-independent

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



way’. On the other hand, they found no effect of the typical color of objects when the stimuli were line drawings rather than realistic pictures—a finding that constitutes a failure to replicate their original findings conceptually. In addition, the memory color effect was not found for all stimuli: strawberries were an exception. While these issues do not definitely undermine this research project on the cognitive penetrability of color experience, they are nonetheless puzzling (for further discussion, see Zeimbekis ).9

 Conclusion The New Look psychology died in the midst of empirical and theoretical controversies. The nature of the observed effects was unclear (the locus problem), methodological issues were repeatedly raised, and many studies did not replicate reliably. The situation in the contemporary research on the cognitive penetrability hypothesis is perhaps not as dire, and it includes some interesting and suggestive lines of research. So, perhaps, evidence will accumulate in support of this hypothesis. On the other hand, contemporary research suffers from issues that are at times eerily reminiscent of those that killed the New Look. It is often unclear where desires, beliefs, emotions, etc. exert their influence; methodological problems, including demand characteristics, are found in many studies, and replication is too often unreliable. It is too early to say whether these ills will kill the New New Look, but at the very least philosophers and psychologists should be cautious in drawing conclusions from contemporary studies.

References Aarts, H., Dijksterhuis, A., and De Vries, P. (). On the psychology of drinking: being thirsty and perceptually ready. British Journal of Psychology : –. Alter, A. L., and Balcetis, E. (). Fondness makes the distance grow shorter: desired locations seem closer because they seem more vivid. Journal of Experimental Social Psychology : –. Balcetis, E., and Dunning, D. (). See what you want to see: motivational influences on visual perception. Journal of Personality and Social Psychology : –. Balcetis, E., and Dunning, D. (). Cognitive dissonance and the perception of natural environments. Psychological Science : –. Balcetis, E., and Dunning, D. (). Wishful seeing: more desired objects are seen as closer. Psychological Science : –. Balcetis, E., and Dunning, D. (). A false-positive error in search in selective reporting: a refutation of Francis. i-Perception .

9 In addition, the findings reported by Olkkonen et al. may well be within-modality effects instead of instances of cognitive penetration.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

Bhalla, M., and Proffitt, D. R. (). Visual-motor recalibration in geographical slant perception. Journal of Experimental Psychology: Human Perception and Performance : –. Bitter, D. (). Is low-level visual experience cognitively penetrable? The Baltic International Yearbook of Cognition, Logic and Communication : –. Bolles, R. C., Hulicka, I. M., and Hanly, B. (). Colour judgment as a function of stimulus conditions and memory colour. Canadian Journal of Psychology/Revue canadienne de psychologie : –. Bruner, J. S., and Goodman, C. C. (). Value and need as organizing factors in perception. Journal of Abnormal and Social Psychology : –. Carter, L. F., and Schooler, K. (). Value, need, and other factors in perception. Psychological Review : –. Delk, J. L., and Fillenbaum, S. (). Differences in perceived color as a function of characteristic color. American Journal of Psychology : –. Durgin, F. H., Baird, J. A., Greenburg, M., Russell, R., Shaughnessy, K., and Waymouth, S. (). Who is being deceived? The experimental demands of wearing a backpack. Psychonomic Bulletin and Review : –. Durgin, F. H., DeWald, D., Lechich, S., Li, Z., and Ontiveros, Z. (). Action and motivation: measuring perception or strategies? Psychonomic Bulletin and Review : –. Durgin, F. H., Klein, B., Spiegel, A., Strawser, C. J., and Williams, M. (). The social psychology of perception experiments: hills, backpacks, glucose, and the problem of generalizability. Journal of Experimental Psychology: Human Perception and Performance : –. Erdelyi, M. H. (). A new look at the new look: perceptual defense and vigilance. Psychological Review : –. Firestone, C., and Scholl, B. J. (). ‘Top-down’ effects where none should be found: the El Greco fallacy in perception research. Psychological Science : –. Fisher, S. C., Hull, C., and Holtz, H. (). Past experience and perception: memory color. American Journal of Psychology : –. Francis, G. (a). The same old New Look: publication bias in a study of wishful seeing. i-Perception : –. Francis, G. (b). Some clarity about publication bias and wishful seeing. i-Perception .

Furnham, A., and Spencer-Bowdage, S. (). Inflation and the estimated size of withdrawn notes and coins. Journal of Socio-Economics : –. Garner, W. R., Hake, H. W., and Eriksen, C. W. (). Operationism and the concept of perception. Psychological Review : –. Goldstone, R. L. (). The role of similarity in categorization: providing a groundwork. Cognition : –. Gross, S., Chaisilprungraung, T., Kaplan, E., Menendez, J. A., and Flombaum, J. (). Problems for the purported cognitive penetration of perceptual color experience and Macpherson’s proposed mechanism. The Baltic International Yearbook of Cognition, Logic and Communication : –. Haber, R. N. (). Nature of the effect of set on perception. Psychological Review : –. Hansen, T., Olkkonen, M., Walter, S., and Gegenfurtner, K. R. (). Memory modulates color appearance. Nature Neuroscience : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a no-progress report



Harper, R. S. (). The perceptual modification of colored figures. American Journal of Psychology : –. Hutchison, J. J., and Loomis, J. M. (a). Does energy expenditure affect the perception of egocentric distance? A failure to replicate Experiment  of Proffitt, Stefanucci, Banton, and Epstein (). Spanish Journal of Psychology : –. Hutchison, J. J., and Loomis, J. M. (b). Reply to Proffitt, Stefanucci, Banton, and Epstein. Spanish Journal of Psychology : –. Kahneman, D., and Tversky, A. (). Subjective probability: a judgment of representativeness. Cognitive Psychology : –. Landis, D., Jones, J. M., and Reiter, J. (). Two experiments on perceived size of coins. Perceptual and Motor Skills : –. Levin, D., and Banaji, M. (). Distortions in the perceived lightness of faces: the role of race categories. Journal of Experimental Psychology: General : –. Machery, E. (Forthcoming). De-Freuding implicit attitudes. In Michael Brownstein & Jennifer Saul (Eds.), Implicit Bias and Philosophy. Oxford: Oxford University Press. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research : –. McCauley, R. N., and Henrich, J. (). Susceptibility to the Müller–Lyer illusion, theory neutral observation, and the diachronic cognitive penetrability of the visual input system. Philosophical Psychology : –. McGurk, H., and MacDonald, J. (). Hearing lips and seeing voices. Nature : –. Nisbett, R. E., and Miyamoto, Y. (). The influence of culture: holistic versus analytic perception. Trends in Cognitive Sciences : –. Olkkonen, M., Hansen, T., and Gegenfurtner, K. R. (). Colour appearance of familiar objects: effects of object shape, texture and illumination changes. Journal of Vision : –. Proffitt, D. R., Stefanucci, J., Banton, T., and Epstein, W. (). The role of effort in perceiving distance. Psychological Science : –. Proffitt, D. R., Stefanucci, J., Banton, T., and Epstein, W. (). Reply to Hutchison and Loomis. Spanish Journal of Psychology : –. Pylyshyn, Z. W. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences : –. Pylyshyn, Z. W. (). Seeing and Visualizing: It’s Not What You Think. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Is perception informationally encapsulated? The issue of the theoryladenness of perception. Cognitive Science : –. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Late vision: processes and epistemic status. Frontiers in Psychology : . Russell, R., and Durgin, F. H. (). Demand characteristics, not effort: the role of backpacks in judging distance. Journal of Vision : a. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Stefanucci, J. K., Proffitt, D. R., Clore, G. L., and Parekh, N. (). Skating down a steeper slope: fear influences the perception of geographical slant. Perception : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



e. machery

Stokes, D. (). Perceiving and desiring: a new look at the cognitive penterability of experience. Philosophical Studies : –. Stokes, D. (). Cognitive penetrability of perception. Philosophy Compass : –. Tajfel, H. (). Value and the perceptual judgment of magnitude. Psychological Review : –. Witt, J. K., Proffitt, D. R., and Epstein, W. (). Perceiving distance: a role of effort and intent. Perception : –. Woods, A. J., Philbeck, J. W., and Danoff, J. V. (). The various perceptions of distance: an alternative view of how effort affects distance judgments. Journal of Experimental Psychology: Human Perception and Performance : –. Wu, W. (). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies : –. Zeimbekis, J. (). Color and cognitive penetrability. Philosophical Studies : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Towards a Consequentialist Understanding of Cognitive Penetration Dustin Stokes

If cognitive penetration of perception occurs, it is one interesting pathway by which concepts and perception can relate. There are at least two ways this might be characterized. First, the content of cognitive states like belief are, on one plausible account, structured by concepts. And on one account of cognitive penetration, the content of one’s cognitive states affects, in some intelligible way, the content of perceptual experience. Thus, in a naturally described way, one’s concepts—understood as mentally represented Fregean senses—partly determine one’s perceptual experiences. Second, one can think of concepts as recognitional capacities. And some have recently argued that one’s experiences change by virtue of the acquisition of new recognitional capacities. This involves a kind of diachronic cognitive penetration, where one’s concepts— understood as abilities—affect how one perceives the world. Of course things are not so simple. What is clear is that cognitive penetration, if it occurs, would be of significant philosophical and scientific importance (and in addition to the possibilities outlined just above). But whether the phenomenon occurs is debated. Much of this debate is unfortunately rooted in lack of consensus on just what the phenomenon is or would be. There is little or no agreement, on either side of the debate, regarding a definition or analysis of the target phenomenon. For example, the first characterization mentioned above invokes a debatable semantic criterion for cognitive penetration. And the second characterization suggests long-term cognitive effects on perception, and some theorists claim, sometimes only implicitly, that cognitive penetration must be a direct, synchronic relation. This lack of conceptual consensus has important consequences. Perhaps most importantly, if cognitive penetration is to be empirically testable, then there must be some agreement on just what one is testing for (and thus how experiments should be designed and controlled). As it stands, empirical data are interpreted

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

differently by different theorists, and by appeal to different criteria for what ‘counts’ as a case of cognitive penetration. A variety of sceptical (i.e. non-cognitive penetration) interpretations are invoked, and without an uncontroversial definition of cognitive penetration, adjudication is difficult if not impossible. Indeed, how could one claim that the sceptical interpretations are less plausible, and that an observed mental phenomenon is best explained as cognitive penetration, if the latter notion is not clearly defined or, better, not at least defined in a way that is agreeable to both parties of the debate? This chapter outlines a methodological strategy for resolution and one that will, hopefully, encourage progress on an important (possible) aspect of the human mind. The simple prescriptive thesis is this: cognitive penetration (of perception) should be understood (if not defined) in terms of its consequences.1 This consequentialism about cognitive penetration exploits the clearest point of agreement in the debate, namely, the importance of a phenomenon like cognitive penetration. The discussion proceeds as follows. Section  briefly outlines two extant definitions of cognitive penetration. Section  clarifies how these definitions yield divergent verdicts on a pair of empirical case types. Section  turns to the alleged consequences of cognitive penetration. And Section  characterizes and argues for the consequentialist strategy.

 Two Definitions The term ‘cognitively penetrable’ was first coined by Zenon Pylyshyn (), and his most recent definition is the one most often cited in the literature. Here is its clearest statement: ‘[I]f a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, that is, it can be altered in a way that bears some logical relation to what the person knows’ (Pylyshyn : ). This partial definition identifies a necessary condition for cognitive penetration that we might call the ‘semantic criterion’ (see Macpherson ; Stokes ). The criterion as presented by Pylyshyn—call it (SC)—is ambiguous. In some places, Pylyshyn seems to have in mind a logical, inference-supporting relation. So a cognitive state like a belief penetrates a perceptual experience only if the content of the belief could support an inference to the content of the resulting experience. This makes the semantic criterion a rationality criterion. One worry here is that, at least as discussed in the literature, it is unclear that worries about cognitive penetration are worries about a rational relation. Indeed, most theorists are concerned with cognitive penetration qua non-rational relation—going back to the concerns about 1 Throughout this discussion it is assumed that cognitive penetration concerns a human perceptual phenomenon. No attempt is made to analyze other possible phenomena, e.g. cognitive penetration of cognition or emotion. The latter may be genuine phenomena (indeed, they very probably are), but as a matter of fact these are simply not the subjects of all relevant literature on cognitive penetration.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



theory-ladenness of the s and s, and up through the epistemic circularity worries current in literature today (Siegel ; Lyons ; Raftopoulos ).2 Elsewhere, Pylyshyn ostensibly takes the criterion to require only representational coherence: ‘This is the essence of what we mean by cognitive penetration: it is an influence that is coherent . . . when the meaning of the representation is taken into account’ (Pylyshyn : , n. ). This weakened semantic criterion requires that one could, perhaps under idealized conditions, identify how the content of the penetrating state—say, a belief—affected the content of the resultant perceptual experience. Here again, however, one may worry about the motivation for the criterion thus interpreted, since it imposes a kind of operationalist condition on cognitive penetration. The motivation for Pylyshyn’s semantic criterion, for either of the above interpretations, is ultimately clear. The problem is that this motivation is insufficiently theoryneutral. To see this, one can trace Pylyshyn’s thinking back to his earlier articulations of the concept of cognitive penetrability. The broader goal for Pylyshyn’s earlier relevant work was to introduce a criterion that marked off the proper subject matter for cognitive science. Nearly two decades earlier, Pylyshyn wrote: Much of the paper elaborates various conditions that need to be met if a literal view of mental activity as computation is to serve as the basis for explanatory theories. The coherence of such a view depends on there being a principled distinction between functions whose explanation requires that we posit internal representations and those that we can appropriately describe as merely instantiating causal physical or biological laws. In this paper the distinction is empirically grounded in a methodological criterion called the cognitive impenetrability condition. (Pylyshyn : ; emphasis added)

Thus initially, Pylyshyn took ‘cognitive penetrability’ to distinguish cognitive phenomena (understood computationally) from ‘cognitively impenetrable’ functional architecture (understood biologically). This picture was then amended a few years later, where cognitively penetrable systems are influenced by systems the explanation of which requires terms of rules and representation, and explanation of that influence will also require terms of rules and representations (Pylyshyn ). It was here that something like Pylyshyn’s current semantic criterion emerged; and its application placed perception in the category of biological, functional architecture. One odd consequence, one might think, is that this puts perception outside the purview of cognitive science. And this is for the reason that cognitive science, understood in Pylyshyn’s way, involves a commitment to the computationalist doctrine. It is this doctrine that ultimately motivates a semantic criterion, regardless of its particular form across the nearly twenty years that Pylyshyn has written on the topic. The central worry here is that the computationalist theory is controversial, and most certainly not one to which all parties to the cognitive penetration debate commit. Indeed, a number of philosophers as well as theorists who count themselves ‘cognitive scientists’ 2

These are discussed in detail in Section .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

resist or reject commitment to this form of computationalism, including theorists who otherwise have interest in and have written on whether cognitive penetration occurs. So if Pylyshyn’s notion of cognitive penetration centres around the semantic criterion, and the only motivation for this criterion is the computationalist one, then this definition fails, ironically, for reasons of theory-ladenness. A recent alternative definition follows Pylyshyn by maintaining that cognitive penetration is not merely a causal relation between cognition and perception, while excluding his semantic criterion. (CP) A perceptual experience E is cognitively penetrated if and only if () E is causally dependent upon some cognitive state C and () the causal link between E and C is internal and mental. This definition accomplishes a few things. First, it makes explicit that the phenomenon of interest involves (at least partly) an effect on perceptual experience. Thus on today’s orthodox understanding, the phenomenal character of a sensory experience is affected by a cognitive state like belief. Clause () thus ensures that instances where states like belief or memory are affected by antecedent cognitive states do not count as cognitive penetration. Clause () maintains that this relation must be an internal one, and that the causal chain must involve mental states or processes (but with no restriction on how long that chain is). This clause ensures that instances where one’s beliefs or other cognitive states cause an action of some kind which then causes a (change in) perceptual experience do not count as cognitive penetration. Much could be said about (CP). Suffice it to say that like most definitions, it is not without its problems.3 (Indeed, problems of application of the definition will emerge in Section .) So instead of any extended analysis on those scores, it will prove more instructive, for the purposes of this chapter, to instead identify the ways in which (SC) and (CP) deliver divergent verdicts on different sets of cases.

 Divergent Verdicts Defining cognitive penetration, as a theoretical task, is of little final importance to theorists involved in this debate. A definition is instead subservient to answering empirical questions about the mind: namely, whether human cognition and perception relate in some specially important way/s. So getting the definition right is only important to the degree that it assists in performing and interpreting the results of experimentation and reflection. And it is for this reason that an uncontroversial—i.e. generally agreed upon—definition is needed. Without this agreement, theorists apply distinct criteria to the same data, and come away with diverging verdicts on whether that data evidences cognitive penetration. Two kinds of case make this clear.

3

(CP) is taken from Stokes (; ). See Stokes () for one possible counterexample to (CP).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



The first set of cases involves apparent effects of evaluative attitudes on visual size perception. In a now famous experiment, Bruner and Goodman () asked child subjects to report the size of American coins. In both control and experimental groups, subjects were placed before a wooden box with a glass face. Subjects were to turn a knob, located on the bottom right corner of the box face, in order to adjust the circumference of a patch of light to match the presented targets. Targets were placed in the subject’s left hand, and held six inches to the left of, and on the same horizontal plane as, the adjustable light patch. Target stimuli for experimental subjects were ordinary coins of values ranging from  cent to  cents. Target stimuli for controls were cardboard cutout analogues of the varying coins. After an initial training period with the apparatus, subjects then took as much time as they wished to match the light patch to the respective target stimuli. Control subjects reported the size of the cardboard cutouts with near-perfect accuracy. Experimental subjects consistently overestimated the size of the coins, and by differences (by comparison with controls) as high as %. In a second variation of the study, subjects were separated into groups of ‘poor’ and ‘rich’ children. Here both groups overestimate the size of the coins, but poor children by as much as %. Bruner and Goodman interpret the data to suggest that subjects’ values or desires for money influence how the coins are perceived. Put simply, the high value of money results in seeing money as bigger (than it actually is). This early New Look study was followed by a barrage of similarly spirited experimentation, some of it apparently supportive of the central New Look claim—that perception and cognition are ‘continuous’—some of it not. A number of theorists, both in philosophy and psychology, have revived attention to the New Look approach.4 One very recent example involves valenced images and size perception (van Ulzen et al. ). Here subjects were presented, on a computer screen, with circles containing either negatively, positively, or neutrally valenced images (e.g. respectively, an aimed gun, kittens, mushrooms). The task was then to adjust a report circle, located on the bottom right corner of the computer screen, by pushing or pulling a computer mouse. Leaving out various details, subjects consistently reported circles containing negatively valenced images to be larger than circles containing images of positive or neutral valence (where the actual size of the image-containing circle is static across trials). On the face of it, these are cases where evaluative attitudes held by subjects (perhaps a fear of or aversion to the objects depicted in the negative images) are influencing visual size perception. These data are certainly compelling. But the question to be asked here is whether they are to be explained as instances of cognitive penetration, or something else. Consider, in turn, the two definitions discussed in the previous section. Recall that Pylyshyn’s (SC) requires, at least, an identifiable representational coherence between the casually antecedent mental state and the resultant perception. In the studies both by Bruner and Goodman and by van Ulzen et al., the antecedent mental state is, 4

See Balcetis and Dunning (); van Ulzen et al. (); Stokes ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

plausibly, some orectic attitude (e.g. a desire for money in the first case and a negative affective attitude towards images involving guns and spiders in the second). In both studies, it appears that size perception is affected: subjects see coins or circles with negative images as bigger. However, application of (SC) undercuts this interpretation: it is far from clear that the semantic coherence criterion is satisfied. Consider the Bruner and Goodman results. Here, Bruner and Goodman conjecture, there is some background desire or value for money. Candidate cognitive states/contents would include S desires that [I have money] or S evaluates that [Money is good]. And the (allegedly) resultant experience is one where coins are experienced as bigger (than they, objectively, are). Put crudely, and leaving aside for the moment important questions about the content of experience, suppose that this experience should be specified as something like: S sees that [the coin is size n]. (Here it does not matter how the variable, n, is filled; it simply placeholds a size-specific slot in the content of the subjects experience. And as per the experimental data, this quantification is such that the coin is experienced as bigger than it, objectively, is.) Here then is the trouble: there is no clear way to specify how contents of the first sort would cohere with a content of the second sort. As it is sometimes put in discussion of perceptual content and/or perceptual justification, there is nothing in the first content to ‘hook up’ with the content of the (allegedly) resultant experience. One way to put this is in terms of inference, one could not infer [the coin is size n], for any n, from, say, [Money is good]. And this is true no matter the attitude taken towards the latter content. (For example, the inference would be no better if the attitude was doxastic rather than orectic.) Even on the weaker, non-inferential, interpretation of (SC), the verdict is the same: the content of the antecedent state (the ‘meaning’, as Pylyshyn puts it) bears no intelligible connection to the content of the second state. Therefore, these data do not provide evidence for cognitive penetration of perception. This data still calls for explanation. And so an advocate of (SC) will typically invoke one of the following alternative interpretations to explain this (and similar) data. Some have suggested, in considering these and other New Look(ish) experiments, that the background mental states have an effect on how the perceptual stimulus is remembered, but not how it is perceived. Call this the ‘memory interpretation’.5 Similarly, one might argue that the subjects across control and experimental circumstances have normal, veridical visual experiences (as of the sizes of the target stimuli), but then make judgments about these perceived stimuli in a way informed by the background mental states. Like the previous interpretation, according to this ‘judgment interpretation’ we have, at most, evidence for cognitive effects on (other) cognitive states. And this is uncontroversial.6 Finally, we might allow that the perceptual experiences of experimental subjects are affected (by contrast to control subjects), but argue that this effect is enabled by active changes in attention. Thus, as a result of the subject’s background mental states (say, the subject’s desire for money), she attends 5

See McCurdy ().

6

See Pylyshyn ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



differently to the target stimulus, and in ways that alter experience. This, the critic will urge, is similar in kind to instances where one knows how attending differently will alter experience. Thus one sees a figure as a duck or as a rabbit, by attending to different parts of the figure. And this attentional shifting depends upon ‘knowing the trick’, as it were.7 But this is not the phenomenon of interest, since the effect on perception is indirect: attentional acts mediate between background mental states and the resultant experience. Call this the ‘attention-shift interpretation’. One or other of these interpretations, an advocate of (SC) might urge, better explains the experimental data. So we have, not instances of cognitive penetration, but something else: either cognition affecting cognition, or action-guided perceptual change. What of the second definition, (CP), and its application to the New Look cases? Recall that this definition eschewed the semantic criterion, maintaining instead that cognitive penetration requires both an effect on the phenomenal character of perceptual experience and an internal, mental causal link between antecedent mental state (a belief, value, desire, etc) and resultant experience. This definition has the advantage of incompatibility with the above-named alternative interpretations: if a phenomenon or case meets (CP), then it is not an instance of mere cognitive effects on cognition (by appeal to clause () of (CP)) nor an instance of action-guided perceptual change (by appeal to clause () of (CP)). And plausibly, these New Look cases do meet the conditions of (CP). Taking again the Bruner and Goodman experiment as the example, these subjects have a desire or value for money which affects, directly, the perceptual experience of the size of coins (by contrast with the control stimuli, where no such effect is recorded). Given the online nature of the task—subjects inspect the stimulus as they adjust the light patch to match—the memory interpretation is implausible. Similarly, there is little reason to think that the judgment interpretation is apt, since it would require a consistent mismatch between veridical visual experience of the coins and an erroneous, online report of the size of the coins. Subjects give no indication (e.g. surprise or confusion) that their reports deviate in this way from their current experience. Finally, the attention-shift interpretation is implausible, since there seems to be no relevant attentional difference between controls and experimental subjects, and thus no attentional explanation of the respective differences between these subjects reports. This, anyway, is how an advocate of (CP) might defend its application to the case. The critic may be unpersuaded by this line of reasoning. (For example, one might find one of the above alternative interpretations just as compelling as a (CP) interpretation.) No matter (for now). The important point for the present discussion is that, as some have argued, appeal to (CP) plausibly yields a pro-cognitive penetration verdict on this set of data. Result: by appeal to one definition, (SC), a set of data is judged not to evidence cognitive penetration. By appeal to another definition, (CP), the same data 7 See Fodor (; ). See also Macpherson () and Stokes (; ) for discussion of these interpretive strategies.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

is judged as good evidence for cognitive penetration. Put simply: (SC) plausibly yields a NO verdict, (CP) a YES verdict. This is a clear case of theoretical cross-talk. And no matter which side of the debate one is on, this scenario does not bode well for theory adjudication. A second type of case further reveals the problem. Some recent philosophers of perception have argued that perceptual experience represents ‘high-level’ properties. If vision, for example, is representational, then it represents basic low-level properties like colour and shape. Traditionally, however, properties such as being of a natural or artificial kind, being caused/a cause, and being expressive of emotion are understood as the result of post-perceptual cognitive processes like judgment or belief; these high-level properties, tradition has it, are not ‘picked up’ by perception. Both the phenomenology of experience and empirical evidence provide reason to doubt the traditional view. Susanna Siegel () argues from phenomenal contrast. It is introspectively plausible that one’s overall experience of an object or event of kind K, after one learns to recognize instances of K, changes. Taking one of Siegel’s central examples, as one learns to recognize pine trees and thereby learns to pick them out in an array of different types of tree, the phenomenology of one’s overall experience changes (when seeing pine trees). So one’s overall experience when seeing pine trees is a part of that overall experience, and eventually differs phenomenally from overall experiences (involving pine-tree seeings) prior to acquisition of the recognitional capacity. Siegel then claims that this contrast in overall phenomenology is best explained as a difference in the representational content of perceptual experience, the latter of which is a part of the overall experience. Finally, the change in perceptual representation is best explained as a change in high-level property representation. It is not that one’s experience of colours or shapes changes upon acquisition of the capacity to recognize pine trees. Instead, one has visual experiences as of pine trees; one perceptually represents the relevant objects as pine trees. The argument concludes, then, by identifying kind-properties, as well as other high-level properties, as admissible contents of perceptual experience.8 High-level perceptual representation is particularly plausible in contexts of perceiving art. It is clear that different perceivers of artworks make different judgments, and also that judgments often enough co-vary with the art-knowledge of those perceivers. Additionally, one might think that it is not only judgments that co-vary with knowledge in this way, but in fact that experiences co-vary. A perceiver S with sufficient knowledge of abstract expressionism will know that (nearly) monochromatic colours and rough rectangular shapes are standard features for many of Mark Rothko’s paintings (what are known as his ‘multiform’ works), while there will be some variability within this narrow range of colour and shape. Suppose another perceiver, T, lacks any such knowledge of abstract expressionism or Rothko, knowing only that standard 8 Tim Bayne () argues to the same conclusion by appeal to empirical research on subjects with associative agnosia. See also van Gulick (), Siewert (), Siegel (), and Hawley and Macpherson ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



for paintings, as such, is paint on canvas, while shapes and colours can vary widely. Suppose that S and T both perceive one of Rothko’s multiform pieces (the reader might consider, for example, Four Darks in Red, ). Given T’s lack of knowledge, he is likely to aesthetically judge the work to be dull or lifeless, since it is so colourless and simple in its shapes relative to the broad category of painting. S, however, has more precise knowledge about this kind of work, and will perceive the work not just as a painting, but in the category of Rothko’s multiforms. Again, standard for this category are nearly monochromatic colours and rough rectangles, but these features do vary across various Rothko pieces. S will likely attribute different aesthetic properties; he might judge the work to be striking or vivacious. Now, one might maintain that these aesthetic property attributions are merely post-perceptual judgments. But the following possibility is sufficiently plausible for serious consideration: these perceivers visually perceive the works differently, by virtue of their differing art-knowledge. S, for example, does not just judge the work to be striking or vivacious, she perceptually represents the work as instantiating these high-level aesthetic properties. And this is a consequence of learning and knowing about art.9 These possibilities are both examples of what we might call diachronic cognitive effects on perception. On the face of it, as one learns to recognize certain kinds of things—be it pine trees or Rothko paintings—one’s phenomenal experience of these things changes. And this change takes time. Here is one final example of a case of diachronic effects on perception, this one not requiring any appeal to high-level property representation. Early research on inverting lenses—goggles or lenses that, in short, turn one’s world upside down—revealed remarkable human capacity for relatively quick visual adaptation. Initially, subjects wearing the lenses are radically disoriented, failing at both tasks of descriptive judgment and vision-guided action. But after a training period, often a week or so, subjects adjust, performing the same tasks in statistically successful ways (Stratton ; Kottenhoff ; Taylor ). Paul Churchland () interprets these data as an application of learning—namely, about novel relations between movement and resultant experience, about expectations and what experience actually delivers, and so on—to seeing. In short, the subjects have distinct visual experiences—of motion, orientation, shape—before and after learning how to ‘use’ the inverting lenses successfully.10 Each of these cases involves some learning—how to spot pine trees, how to spot categories of art or aesthetic properties, how to use a radically distorting visual 9 See Walton (). See Stokes () for discussion of both cognitive penetration and the perception of art, and a defence of diachronic cases as genuine examples of cognitive penetration. 10 There are worries, however, about whether the evidence is sufficient to suggest changes in the phenomenology of subjects experiences, rather than just a cognitive adjustment to a stable (but upside down) environment (see Prinz  and Schwitzgebel forthcoming). For present purposes, however, this detail is of no consequence. Grant that there is an apparent change in the phenomenology of experience, the question is whether the diachronic nature of this type of case, and of the high-level property cases, is incompatible with cognitive penetration. Distinct definitions deliver distinct verdicts here, or at least, verdicts for different reasons.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

apparatus. And, apparently, this learning affects the phenomenology of perceptual experience. What is crucial to note is that this effect (if it is a genuine effect) takes place over time. This is what one would expect, insofar as the learning also takes place over time. So, plausibly, the look of pine trees, say, gradually changes as one better learns to recognize pine trees. The question to ask now is whether this diachronic feature of all of the relevant cases, whatever their differences may be, excludes such cases from being instances of cognitive penetration. Consider once more the two definitions articulated in the previous section, (SC) and (CP). Upon first glance, one might apply (SC) to get the verdict that these diachronic cases are indeed cases of cognitive penetration. Specifying the relevant contents here, as is often the case, is a complicated (if not strained) matter. But things might go as follows. Take the first example of high-level property representation. Here a subject S gradually learns that (or, if one prefers, forms a belief that) [Pine trees have features F], where F placeholds the perceivable shape and colour features typically possessed by pine trees. According to the high-level content theory, and simplifying, S’s experience of pine trees will also gradually change, where eventually S will (in the presence of pine trees) token an experience of the type: S sees (or sees that) [There is a tree with features F]. Alternatively, S sees (or sees that) [There is a pine tree], where the high-level property of being a pine tree is at least partly characterized by certain low-level colour and shape properties, F. Here the coherence is much clearer than in the previous set of cases: one learns about pine-tree looks and the content of these concepts or beliefs informs, in a coherent way, the consequent perceptual experiences of pine trees. A similar story could be told about the cases involving perception of art. And finally, a similar story may be told about the inverting-lens cases. One simple way to characterize these cases concerns the shift in the apparent orientation of the world, as experienced by subjects. Suppose that ‘right side up’ denotes the way the world actually is, and appears to be, when we have normal visual experience. And ‘upside down’ denotes the way the world would be if the initial experiences of inverting-lens wearers were accurate. As it is sometimes put in discussions of perceptual contents, ‘right side up’ designates a proper part of the accuracy conditions for normal human visual perception; ‘upside down’ designates a proper part of the accuracy conditions for lens-inverted human visual perception. These designations give the content, or part of it, of the relevant perceptual experiences. Now, the inverting-lens wearer will initially experience the world as upside down but will also, because she understands the nature of the apparatus, have a belief that [The world is right side up]. (The apparatus is vision-inverting, but not world-inverting! And subjects understand this.) The same subject also acknowledges that her current perceptual experience is, so to speak, at tension with this general belief regarding world-orientation: things initially appear upside down. And this belief (that the world is right side up) is maintained and applied as the subject better and better copes with her disoriented visual array. Eventually, there is a shift in the subject’s perceptual

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



experience, part of the content of which would be specified as [The world is right side up]. Here the coherence between background cognitive state and resultant perceptual experience could not be clearer. As it turns out, advocates of (SC), or something like it, resist the claim that these diachronic cases are genuine cases of cognitive penetration. This is not entirely unreasonable, since semantic coherence as specified by Pylyshyn () is merely a necessary condition. So perhaps there are other conditions required but not met by these cases. Alternatively, an advocate might maintain that these cases do not even meet (SC) as baldly characterized, since (SC) requires a coherence relation between a cognitive state and perception. And, one might maintain, the cases in question do not involve a relation between belief, say, and perception, but instead just some kind of noncognitive perceptual change. Jerry Fodor gives this reply to Churchland’s discussion of the inverting lenses. For there are, after all, good ecological reasons why you might expect plasticity of this sort . . . what needs to be kept open for re-calibration is whatever mechanisms compute the appropriate motor commands for getting to (or pointing to, or grasping) a visible object on the basis of its perceived location. Adaptation to inverted (and otherwise spatially distorting) lenses is plausibly an extreme case of this sort of recalibration. (Fodor : )

So this intraperceptual interpretation may well be invoked for the diachronic cases, which alleges that the changes in perceptual systems are made by perceptual systems, over time. So, no cognitive penetration because one necessary relatum, a cognitive state, is absent in the relevant causal story. It is thus less than perfectly clear what verdict (SC) delivers. On some interpretations of the cases, semantic coherence obtains, and obtains between a cognitive state and perception. But on different interpretations, there is coherence, but not between the appropriate mental states. How are these same cases treated by (CP)? Grant for the moment that these diachronic cases involve a cognitive effect on the phenomenal character of perceptual experience. Learning to recognize pine trees or Rothko works, for example, is a cognitive achievement and one that results in changes in perceptual representation. Grant, then, that these cases meet clause () of (CP). However, one may worry that the diachronic nature of the cases precludes them from meeting clause () of the definition. This condition requires that the causal link between cognitive state C and resultant experience E is entirely internal to the perceiving agent and, moreover, mental. But one might worry that various actions mediate, over time, between C (or Cs) as it takes hold and changes, and E (or Es) as it changes in correlative ways. For example, as one hones the capacity to recognize pine trees, one will actively use attention and explore one’s environment. One will look at pine trees from different angles, distances, and under various viewing conditions, perhaps asking questions of the trained pine-tree spotter along the way. And these actions are required, one might infer, to bring about the relevant changes in perception of pine trees. If this is the appropriate characterization of these cases, then the causal

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

chain from C to E violates clause (): some events in that chain are external to the perceiver and, therefore, non-mental. Interpreted this way, diachronic cases are not cognitive penetration per (CP). Of course, things may not be this simple. Another plausible characterization of the diachronic cases locates the just-mentioned attentional and exploratory acts in the (stages of) acquisition and honing of the recognitional capacity, but not in the causal story about how that changing capacity influences perceptual experience (in ways that change, accordingly, over time). So one might grant that learning to spot pine trees requires physically exploring one’s space and actively using attention, while maintaing that the resultant changes in perception are internal, mental causal consequents of the pine-tree spotting capacity. In short, learning the look of pine trees as such involves action. But perceptually representing the property of pine trees is a (relatively) direct effect of that learning (or, once learnt, of the execution of that capacity). On this interpretation, then, clause () of (CP) is also satisfied, and in spite of the diachronicity of the cases. As with (SC), then, a verdict from (CP) on diachronic cases is less than straightforward. Here is not the place for adjudication or extended analysis on how these cases should be understood in terms of one or both definitions. Again, the task here is not to make a conclusive case that these cases are (or are not) instances of cognitive penetration.11 Instead, the lesson to be gleaned is that application of the definitions is challenging, and the two definitions may again yield divergent verdicts. Or, they may yield the same verdict, but for diverging reasons. One reply to this situation is to conclude that (SC) and (CP), and the theorists who advocate these definitions (or something like them), target distinct definienda. If the definitions are used in such divergent ways, or deliver clearly distinct verdicts on the same observed phenomenon, then perhaps they are simply not talking about the same thing. There is some truth to this reply: it acknowledges what may be a mistake endemic to the cognitive penetration literature. Attempts to define or characterize the phenomenon have erred towards defining cognitive penetration as such, and in a way that has lost sight of the supposed consequences of the (possible) phenomenon. That is to say, there are reasons that philosophers and cognitive scientists began discussing the possibility of something like cognitive penetration, and those reasons all concern the consequences that a phenomenon would have if that phenomenon occurred (and, perhaps, with some frequency). These consequences are, it will now be argued, what is of common interest to both parties of the cognitive penetration debate, and no matter what definitions such parties offer. The fix, then, is to understand the target phenomenon in terms of its consequences and, in turn, to abandon essentialist definitions of cognitive penetration. 11 As it stands, the diachronic cases appear mostly disregarded, sometimes only implicitly, as not cognitive penetration; but, as will become clear in the discussion of the two sections that follow, this dismissal may be unmotivated.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



 Consequences of Cognitive Penetration There are three central consequences of interest. They concern, respectively, theoryladenness of empirical observation, the epistemic role of perception, and (modular) architectures of the mind. These are briefly discussed in turn. Traditional empiricist models of scientific inquiry came under fire in the middle of the th century. (Empiricism here includes, but is not exclusive to, logical positivists and logical empiricists.) A number of philosophers challenged the traditional assumption that scientific observation is theory-neutral in a way that would support rational theory choice (Hanson ; ; Kuhn ; Feyerabend ). At least one version of this worry understands the relevant observation as, simply, perceptual observation. The corresponding worry is epistemic: if perceptual observation is laden with theory (in particular the theory or theories being tested), then it will not provide a means for rationally choosing or adjudicating between scientific theories. So two theories—say, an earth-centred theory vs. a sun-centred theory of our galaxy—may be equally successful in terms of elegance, parsimony, internal coherence, and other (non-observational) criteria of theoretical success. The plausible empirical method of adjudicating between these competing theories is to test them against perceptual observations of the world. However, if the observations made by the respective theorists are imbued with theoretical commitment, then those observations fail to provide a neutral form of adjudication. Extending the toy example, the earthcentred theorist might report, upon watching the sunrise, that he sees a sun moving across a stationary horizon. And, predictably, the sun-centred theorist reports seeing a horizon moving to expose a stationary sun. Accordingly, both theorists will report empirical corroboration of their respective predictions, and no progress is made in motivating a rational choice between theories. This is, on one characterization, a perceptual phenomenon. Differences in perception potentially undermine the rational, theoretical role of empirical observation.12 Although theory-ladenness is typically presented as problematizing the empiricist picture of science and, more generally, the role of perception in scientific theory choice, it is worth noting possible instances of ‘good’ theory-laden observation. Kuhn, discussing the effects of ‘normal science’ on its practitioners, writes: What were ducks in the scientist’s world before the revolution are rabbits afterwards . . . Transformations like these, though usually more gradual and almost always irreversible, are common concomitants of scientific training. Looking at a contour map, the student sees lines on paper, the cartographer a picture of a terrain. Looking at a bubble-chamber photograph, the student sees lines on paper, the physicist a record of familiar subnuclear events. Only after a number of such transformations of vision does the student become an inhabitant of the scientist’s world seeing what the scientist sees. (Kuhn : )

12

This example is modeled on Hanson (: –).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

These kinds of perceptual achievements, supposing for the moment that they could occur, might be beneficial in some way. The cartographer more efficiently (perceptually) identifies the map’s features, the sonograph technician, the sex of a foetus. This kind of accuracy and efficiency gives the expert in science and medicine an advantage over the non-expert. And this advantage would depend upon the background theoretical and technological understanding of the perceivers. Questions remain here, of course, about whether this theory-ladenness would be epistemically good. By many criteria for epistemic normativity, theory-ladenness remains an epistemic bad even if it sometimes produces good results. So, for example, just as unreliable belief-forming mechanisms can sometimes produce true beliefs, theory-laden observation may occasionally produce more efficient or accurate perceptual observation. But a true belief formed by unreliable mechanisms is in no way justified by the mere fact that it is true. Analogously, one might maintain, theoryladenness is generally epistemically pernicious even if it sometimes improves observation and testing. Suffice it to say, then, that there are open questions about the epistemic status of theory-laden observation. What is clear, nonetheless, is that if observation is theory-laden, this is of epistemic import. Theory-ladenness, no matter how the open questions are answered, is an important possible consequence of cognitive effects on perception.13 The second consequence is also epistemic, generalizing from the first consequence. Perceptual experience, on the most intuitive picture, provides us with knowledge about the world. But this epistemic role for perception is threatened if there are circumstances where background cognitive states influence perceptual experience that, in turn, influence belief formation. The most obviously pernicious cases are one’s that involve a rough causal schema of the form: Belief that P → Experience that P → Belief that P Here the causal history of the consequent belief involves a circularity that, plausibly, undermines reason for that very belief (Siegel ; forthcoming). The problem with this circularity, or anything like it, is clear given the supposed role for perception and action. Cognitive impenetrability theorists are no less clear on this point. As Fodor suggests, the ‘function of perception is to deliver to thought a representation of the world’ (: ). And since this representation is supposed to inform belief and action (about or in response to the here-and-now), it should track not what the agent believes, wants, or otherwise thinks about the non-present; it should instead track the present environment, here-and-now. Indeed, it is for this very reason that Fodor argues that perceptual systems are cognitively impenetrable, claiming that ‘isolation of perceptual analysis from certain effects of background belief and set . . . has implications for both the speed and objectivity of perceptual integration’ (Fodor : ; emphasis 13 For more recent discussion see e.g. Churchland (; ); Fodor (; ); Brewer and Lambert (); Raftopoulos (; ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



added; see also Pylyshyn ). Whether perceptual processing is isolated (and thus, cognitively impenetrable) is an empirical question, and up for current debate. But the consequence here is clear: if perception is influenced by background cognitive states in these ways, then the objectivity of perceptual representation is threatened. Here too there are important open questions. One such question concerns the scope of the epistemic consequence. If a belief has a circular etiology (following the schema above), does this result in a mere case of local unjustified belief, or does some kind of global scepticism follow? Jack Lyons frames the question this way: Suppose . . . that top-down influence is merely probabilistic in the sense that theory-consonant observations are more likely than they would otherwise be but that objective facts are still significant determinants of what is observed . . . Many factors keep our perceptual access to the world from being infallible—poor observation conditions, camouflage, distraction, sleepiness, etc.—why should prior beliefs be more than another such factor? (Lyons )

Lyons is suggesting that these other factors are insufficient to motivate global scepticism, so why should something like cognitive penetration motivate global scepticism? A response might go: factors like camouflage, distraction, and the like are perceptually muddying, but they are epistemically neutral with respect to some proposition P. Belief etiologies like that schematized above, however, present a perceptual bias towards some P. In some of the alleged cases, the top-down influence is biasing in ways that raise the probability that the subject infers that P. So it is not the confounding of infallibility—the muddiness—that is epistemically pernicious; it is the apparent biasing effect towards experiences that encourage certain inferences. But this brings Lyons’s point regarding local versus global results to the fore: are these biasing effects sufficiently frequent to motivate a general epistemic problem, or are they just another example (even if slightly different) of human fallibility?14 The first consequence, then, is epistemic, concerning the scientific-theoretical role of perception. The second consequence concerns, as we might say, the ‘everyday’ knowledge-providing role of perception. This general epistemic consequence is importantly relevant to the third possible consequence of cognitive penetration, namely, architectures of the mind. Modularity theorists (at least of Fodorian strength) claim that perceptual systems are informationally encapsulated. A core motivation (perhaps the primary motivation) for this claim is that the processing of such systems would be immune to error introduced by the broader cognitive system. Since computations performed by these modules are supposed to be insensitive to what the organism knows, expects, or wants, the resulting perceptual representations more reliably inform the organism about its environment (Fodor : –; see also Pylyshyn ). The modularity thesis is an empirical one, and one that has set substantial

14 Another open question concerns whether the general epistemic consequences are, by necessity, bad. Just as with theory-ladenness, one can conceive of everyday scenarios where expertise or knowledge improves perceptual accuracy.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

research agendas in cognitive science. So, if it turns out that cognitive penetration of perceptual systems occurs (i.e. perceptual systems are unencapsulated relative to cognitive systems), then this alleged feature of modules, even if epistemically desirable, is not actual. Although this last consequence is often put this way, it is not exclusively a concern about perceptual processing.15 If perceptual experience is directly influenced by background cognitive states, modularity is no less threatened. Here are two reasons. First, although higher-level effects on some components of perceptual processing do not imply the cognitive penetration of experience, and although higher-level effects on perceptual experience do not imply (by themselves) that a particular stage of perceptual processing is penetrated, higher-level effects on perceptual experience imply an effect on perceptual processing at some stage. At least this follows if we assume any kind of physicalism, according to which experience is identified with, constituted by, (metaphysically) determined by, or the output of perceptual processes. Second, as mentioned above, the modularity thesis is largely motivated by epistemic concerns. The idea is that, since the envisioned encapsulated perceptual module rigidly performs its computational function and with no interference from extraneous higher-level information—what the organism knows, expects, or wants—the resulting perceptual representations more reliably inform the organism about its environment. And a concern with the reliability or accuracy of perception is a concern with perceptual representation or experience, not merely processing. One forms beliefs on the basis of what one sees, hears, and so on. So modularity theorists have to be concerned with perceptual experiences, not just processing.16 These three consequences are substantial, concerning the epistemology of science, everyday reasoning and rationality, and broad cognitive scientific theories about the architecture of the mind. Some of the details of such consequences require further analysis, and it is unclear which alleged cases of cognitive penetration imply which, if any, of the consequences. But what is clear is that agreement converges on these consequences. Fine details aside, theorists on either side of the cognitive penetration debate agree that the importance of the possible phenomenon—cognitive penetration—consists in its bearing these consequences. And so, it will now be argued, the question ‘Is perception cognitively penetrable?’ should be revised (or revived) accordingly. We should instead ask: does cognition affect perception in such a way that one or more of these consequences is realized?

15 e.g. Fodor emphasizes computational mechanisms of perceptual ‘input systems’ as informationally encapsulated from ‘central cognitive processors’ (Fodor ). And Pylyshyn defends (typically) the cognitive impenetrability of perception with evidence for the impenetrability of early vision: a functionally defined component in visual processing that computes D shape descriptions of objects (Pylyshyn ). 16 For more on relations between cognitive penetration and modularity, see Deroy (; forthcoming); Wu (); Stokes and Bergeron (forthcoming).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



 Consequentialism The basic prescriptive thesis is this: any analysis of cognitive penetration should be constrained by its consequences. Therefore, an analysis (or, if one prefers, a definition) of cognitive penetrability will be successful just in case and to the degree that it describes a phenomenon (or class of phenomena) that has implications for: the rationality of scientific theory choice; the epistemic role of perception; mental architecture.17 Call this the ‘consequentialist constraint’ on analyses of cognitive penetration. The reader will note that, as stated, the consequences are presented as a list; the logical relation between the analysis (or definition) and the three general consequences are in need of elucidation. And indeed the constraint could be satisfied in a number of ways. The remainder of the chapter will articulate three such options, and then give reasons for favouring the third. The resulting characterization of cognitive penetration will then be applied to the cases discussed in Section . Finally, the definitions from Section  will be reconsidered in the light of the consequentialist constraint. The first option is to stick with traditional conceptual analysis. One constructs an essential definition that describes a phenomenon which results in one or more of the relevant consequences. Here the conditions specified in the definition need not make any one consequence explicit, but instead just describe the relevant mental and causal structures. So long as any one satisfaction of these conditions is co-extensive with a phenomenon that bears at least one of the relevant consequences, then the consequentialist constraint is met.18 This option may be tempting, but there are reasons to doubt both the need for and probable success of an essentialist definition. Contemporary philosophy is rife with examples of controversial if not failed attempts at conceptual analysis. So although it might be ideal for some purposes, we have reason to be sceptical that cognitive penetrability is amenable to this kind of definition. Perhaps more importantly, if we can do the work by appeal to the consequences, then why bother with the conceptual analysis and corresponding counterexample game? The second option is what we might call ‘conjunctive consequentialism’. It says that ψ is cognitive penetration if and only if ψ is a cognitive-perceptual relation that implies consequences for theory-ladenness and the epistemic role of perception and mental architecture. In some ways, this is an improvement over the previous option, since it makes the consequences explicit. However, this is no less an instance of traditional conceptual analysis and so any worries that applied to this logical feature of the first option will apply here as well. More substantively, the conjunctivist option makes an assumption: cognitive penetration is a phenomenon that has all (and only)

17 Mental architecture concerns, in this context, more specifically, the nature and structure of perceptual systems and how they relate to nonperceptual systems. 18 A distinct gloss on this option is to make implication of all consequences necessary. This could be done non-explicitly, where non-consequentialist conditions are specified, and any satisfaction of those conditions is coextensive with implication of all three consequences. Or it might be done explicitly, as represented by the second option discussed shortly.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

three general consequences. Accordingly, it rules out the possibility that there are distinct phenomena of interest that imply distinct consequences. This deviates from the spirit of the consequentialist proposal. Each consequence matters and, by the very nature of the consequences themselves, for different reasons. And it is not implausible that there are distinct cognitive-perceptual phenomena that imply one consequence but not the other. Here is one such possible scenario. Strong-modularity theorists maintain that perceptual modules are encapsulated from other systems in the overall organism. As such, these modules are supposed to perform their computational functions in a way that is independent from, and thus not penetrated by, information in other systems. So, cognitive impenetrability is supposed to be a categorical feature of perceptual modules. Therefore, a single genuine case of cognition affecting, say, visual processing would provide a counterexample to encapsulated modules. Now suppose such a case (or even several of them) is found, but the higher-level information tokened by the cognitive state is entirely theoretically neutral. In this scenario, one consequence is implied but not another: mental architecture is not as the modularity theorist claims, while visual observation is, for all we know, theory-neutral. The conjunctivist option rules out this scenario by definition. Finally, a third option is ‘disjunctive consequentialism’. It says that ψ is cognitive penetration if and only if ψ is a cognitive-perceptual relation that implies consequences for theory-ladenness or the epistemic role of perception or mental architecture. This best captures the spirit of the consequentialist constraint. Theorists on both sides of the debate (or involved in related but distinct debates) are interested in these consequences. Moreover, it leaves open the possibility that there are distinct phenomena that meet distinct disjunctive conditions. Of course, one may maintain that what unifies the phenomena is simply that they bear one consequence or more; thus, the consequences unify. Accordingly, one can call any phenomenon that satisfies the cluster condition cognitive penetration. A virtue of disjunctive consequentialism is that it makes no commitment regarding unified explananda beyond the appeal to consequences. This makes explicit the most important issues of concern (namely, the consequences of concern to both parties of the debate) but without commitment to a single unified mental phenomenon of interest. As one may have predicted given some of the discussion above, one does not simply apply the disjunctive analysis to any one case to deliver a verdict. Instead, the inferential procedure remains abductive. For any given case, one considers alternative interpretations and asks what best explains the data. So, one asks, is this data best interpreted in accordance with the memory interpretation, judgment interpretation, attention-shift interpretation, intraperceptual interpretation, and so on, or as a phenomenon that satisfies the cluster condition specified by disjunctive consequentialism? It is important to note that these alternatives are alternatives to any interpretation that characterizes a phenomenon as implying one or more of the relevant consequences. So, for example, if a phenomenon is best interpreted as involving an

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



overt shift in attention, or a cognitive effect on memory (but not perception), then there are no relevant consequent concerns about the epistemic role of perception nor about mental architecture vis-à-vis perceptual systems. The standard alternatives thus preclude the consequences. Making this relationship explicit is another virtue of the account. This will become clearer upon application of the consequentialist account to the two empirical cases types discussed in Section . It should be noted at the outset that the goal here is not to employ the consequentialist account to deliver conclusive verdicts. This is of course the final goal, and for any one case or set of data, extended analysis would be required. But here the ambition is more modest: to offer a sketch of the debate-neutral value of the consequentialist line. Accordingly, discussion of the two types of cases will be brief. Consider first the New Look and New Look-inspired cases involving apparent effects of evaluative attitudes on perception. Here the question is whether this apparent effect is actual: is perception affected by value in such a way that more than one of the relevant consequences follow? And, recall, if the empirical results are best explained in one of the alternative ways, then these consequences do not follow. Working through these alternatives, then, begin with the memory and judgment interpretations. In both control and experimental circumstances, these experiments are online: the perceptual target is available while reports (of a match) are made. Accordingly, it is implausible that the reports are just of subjects’ memories. Although more plausible than the memory interpretation, it also seems unlikely that perception is wholly unaffected (e.g. coins are seen accurately) while judgments consistently are erroneous (e.g. coins are reported as significantly larger), since size judgments are made on the basis of current perceptual experience. If this were the effect, we would expect some kind of confusion on the part of the subjects, since they would on this (judgment) interpretation be seeing the two objects (e.g. coin and matching light patch) veridically (where, given the results, the adjusted light patch is significantly larger than the coin), while reporting that the two objects match. No indication of this kind of confusion or surprise shows up in the relevant experiments. There also seem to be insufficient differences between control and experimental circumstances for an explanation involving overt shifts in attention. In the Bruner and Goodman studies, for example, the task is relevantly the same, with target (coin or cardboard cutout analogues) and report circle displaced by  inches along the horizontal plane. Any attentional performance that occurs in the experimental case would presumably occur in the control case. Finally, the target stimuli in these experiments—coins, or things like guns in the van Ulzen experiments—are not likely those for which we have a mere perceptually learned response or some kind of perceptual evolution or plasticity. Instead, we learn about these artificial kinds. And this higher-level learning, and the accordant values of those kinds, may affect how we perceive such kinds. Although only a brief treatment, this data looks very much like evidence for a genuine cognitive (or, if one prefers, higher-level) effect on perception. The final question is whether this effect is of relevant consequence.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

If the brief analysis just given is apt, then the cases in question do appear to bear the relevant consequences. If non-doxastic states like desires or values influence experience in these ways—by contrast with the alternative interpretations—then both epistemic consequences follow. Perceptual observation in scientific consequences will not be theory-neutral, and the general knowledge-providing role of perception will be threatened. Put more neutrally: at the very least, the epistemic role of perception is cast in a new light, and will accordingly require new analysis. The same goes for mental architecture. Strong-modularity theorists like Fodor are emphatic on at least this point: the computational processing of perceptual modules is supposed not to be influenced by the organism’s goals and values (in addition to beliefs). And, plausibly, this is precisely what is occurring in the relevant cases. The consequence then is that any strong form of perceptual modularity is threatened. Consider the second kind of case, involving some apparent diachronic change to cognitive-perceptual systems. Here one’s perceptual experience apparently changes, but over time, as one learns in some way (e.g. one learns how pine trees look). Two points should be stressed here. Appeal to the two considered definitions offers neither clear verdicts on these cases nor verdicts driven by the same considerations. Second, an often implicit assumption made in the current literature is that these cases are not cases of cognitive penetration simply by virtue of their diachronic nature. But the first point suggests that this assumption is not obviously motivated by appeal to the definitions. So at the very least, the diachronic cases deserve careful consideration. The alternative interpretations most relevant here are the intraperceptual interpretation and an attention-based interpretation. As discussed above, Fodor favours an intraperceptual interpretation of the inverting-lens data. And he might plausibly invoke a similar story for high-level perceptual content changes, perhaps appealing to a story about perceptual learning where, on one gloss, repeated exposure to a stimulus kind improves one’s ability (in some way) to perceive that kind. Thus repeated exposure to pine trees results in an increased capacity for distinguishing pine trees from other kinds of stimuli. This explanation can be given entirely, the thought goes, in terms of perceptual systems.19 Or, one may think that what happens in the diachronic cases is simply that one learns how to better attend to the environment. This kind of change will plausibly often not involve overt attentional action; thus the attentionshift interpretation is not appropriate. However, it may be that attentional mechanisms change in their sensitivities as one acquires new information about kinds of objects like pine trees. And so one, rather naturally, attends to those objects differently and as a result of the acquisition of new information (say, concepts or beliefs about pine trees). We might call this, simply, an ‘attention-based interpretation’. How can the consequentialist approach help in these interpretive contexts? 19 Perceptual learning is, like cognitive penetration, a commonly used concept but not a singularly defined one. Much could be said here, but space will not allow it in this chapter. See Gibson (); Goldstone (); Goldstone et al. ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



If Fodor is right about the inverting-lens cases, and this intraperceptual explanation can be generalized to the other diachronic cases, then disjunctive consequentialism delivers a clear verdict: lacking a cognitive influence on perception, none of these cases implies the relevant consequences for epistemology or architectures of perception. Note, however, that this verdict has nothing to do with the diachronicity of the cases. It is less clear, however, that more nuanced attention-based interpretation precludes cognitive penetration. Indeed, a very important question, and one insufficiently discussed in the literature, is whether all kinds of attentional mediation between cognitive state and experience counts against cognitive penetration, or whether there is some kind (or kinds) of attentional mediation that is neutral with respect to cognitive penetration. A conclusive verdict is not forthcoming absent substantial further analysis of attention and its bearing on cognitive penetration. But some indication of how consequentialism might be applied can nonetheless be offered. Suppose then, for the moment, that some attention-based interpretation of diachronic cases is correct; do any of the relevant consequences follow? On the attention-based interpretation, new concepts or beliefs determine, in an at least partly sub-personal and non-active way, how one attends to stimuli. This can be understood in at least a couple of ways. It may be that when one learns about pine trees, one becomes more attuned to certain perceptual cues. These cues ‘grab’ attention and aid in distinguishing pine trees from other stimuli. Or, put slightly differently, it may be that attentional selection mechanisms change, such that a subconscious mechanism (rather than the agent) selects some features as more important than others, and experience changes accordingly. The first thing to note is that the basic condition of disjunctive consequentialism is satisfied by this picture: there is a cognitive-perceptual relation. The question is whether the mediating attentional mechanisms preclude cognitive penetration. And here the answer should ultimately be given by appeal to the consequences. There is nothing in this picture that ensures negative epistemic consequences— for example a problematic bias towards background scientific theory or towards one’s general desires or values about the world. However, if indeed experience is malleable in these ways, albeit through the mediation of attentional changes, then epistemic consequences remain on the table. Perceptual modalities, simply, are on this view more susceptible to cognitively driven (rather than purely stimulus-driven) change than traditionally theorized. And this at least leaves open the possibility that scientific observation is theory-laden, and that the way that perception provides knowledge is less clean than usually supposed. Neither consequence need be negative—for example, perhaps perception generally becomes more reliable as a result of these attentional changes—but they are plausible consequences nonetheless. More analysis is needed here, but the simple point is that the attention-based interpretation (of diachronic cases) is not, unlike the standard alternative interpretations, clearly incompatible with cognitive penetration, so long as the latter is understood by its consequences.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

The same is true for the mental architecture consequence. Modularity has received countless characterizations since it was first given voice by Fodor in the early s. And so whether modularity is threatened by these diachronic cases, interpreted in terms of basic attentional changes, will depend upon the working notion of modularity. Two features that Fodor took to typify perceptual modules is that they were informationally encapsulated and innately specified. Perhaps only implicitly, it is the former feature that may be invoked to dismiss the diachronic cases, since there seems to be no immediate effect on the information processing of perceptual modules in such cases. But it is not clear why an immediate effect should be the only one of concern. For example, if the affected attentional mechanisms are part of the perceptual module, then their being changed over time is still relevant. Put simply, a change to the computational processing of a module is no less a change for its taking place over time. And if the information-processing activity of modules change in these ways, then the secondmentioned feature is threatened: modules do not appear innately specified in the sense of being ‘hard-wired’, as it is often put. The modularity theorist could of course defend himself at various places here. But the simple lesson is the same as just above: although more analysis is required, there is nothing in the diachronicity of these cases, even when interpreted as involving attention, that precludes the relevant consequence. Here are two concluding lessons on the two kinds of empirical cases considered, and how they are to be treated by consequentialism. First, as stated at the start of this discussion, neither kind of case has been given a conclusive verdict by disjunctive consequentialism. Instead, a clear debate-neutral strategy has been countenanced, and additional points for analysis have been identified. For example, a complete analysis of the diachronic cases should involve some further discussion of perceptual learning and attention, with one eye always on how the cases, thus interpreted, do or do not bear the relevant consequences. Second, it should be emphasized that the consequentialist line does deliver the following clear result: whatever else one says about either kind of case, it is not the distinguishing feature of the kind that drives the verdict on cognitive penetration. In both the evaluative cases and the diachronic cases, interpretations are offered whereby the cases come out as cognitive penetration in spite of, respectively, the penetrating state being non-doxastic and the mental effect being a diachronic one. This is progress: it reveals that certain kinds of cases are still relevant to the debate. And they are relevant to the debate not because they clearly satisfy some extant essential definition, but instead because they have prima facie bearing on the debate-neutral consequences. Consequentialism can claim this result as a virtue. One final way to motivate consequentialism is to see how it relates to existing definitions of cognitive penetration. Recall Pylyshyn’s semantic criterion (SC). It claims that cognitive penetration must be an inferentially or representationally coherent relation between penetrating state and penetrated experience. Consequentialism encourages us to reject this criterion, since it is not motivated by the consequences. The epistemic concerns certainly do not require this kind of coherence. Indeed, in the extreme cases,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



one can imagine scenarios where background cognitive states affect experience in wholly semantically incoherent ways and, accordingly, the epistemic worries would be all the more pressing. Put simply, at least some of the epistemic concern is that perception may be influenced in ways that are irrational, and so a rationality constraint on the phenomenon is ill-placed. The same is true for the mental architecture consequence. Focusing again on modularity, the question is whether perceptual systems are modular in the sense that their information processing is not affected by the processing of independent, cognitive systems. And although it may be difficult for us to understand this kind of situation, perceptual processing could be affected in non-modular ways without this effect being one that is semantically coherent. So here again, (SC) is simply not motivated by the consequences of concern. Of course one may, as Pylyshyn apparently does, motivate (SC) by appeal to a broad computationalism about the mind. But this motivation is independent of, and insufficiently neutral with respect to, the concerns of existing debate in cognitive science. Consequentialism helps us see this and, hopefully, better enables progress on the topic.

Conclusion The virtue of this consequentialist approach is simple: once such an analysis is in hand, empirical studies can be devised, executed, and interpreted accordingly. Testing for cognitive penetration becomes testing for a phenomenon that bears the relevant consequences for the epistemology and architecture of mind. This (re-)identifies the phenomenon of interest as one common to both sides of the debate. And it provides a unified metric for assessing the relevance of particular mental phenomena. The motivation for consequentialism, then, is twofold. It provides a clearer alternative to data interpretation—like the two kinds considered in Sections  and —than extant definitions. And secondly, it is debate-neutral, sensitive to the concerns of all relevant theorists. Indeed, early in the cognitive penetration literature, there is fairly clear precedent for an appeal to consequences. [I]f you consider the sort of background information that penetrates perception (according to modularity theory), it turns out that perception is neutral, de facto, with respect to most of the scientific (and, for that matter, practical) disagreements that observation is called upon to resolve. According to standard versions of modularity theory . . . perceptual processing has access only to background information about certain pervasive features of the relations between distal layouts and their proximal projections. . . . reliance on such information constitutes a perceptual bias . . . But this bias leaves perception neutral with respect to almost all theoretical disputes, so it couldn’t ground any general argument for the unreliability of observation. (Fodor : )

Here Fodor is appealing to epistemic consequences to disarm certain instances of perceptual processing as non-cognitive penetration (some of the very cases that

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

Churchland  provides as examples of cognitive penetration). Although not labelled as such, this is consequentialism at work.20 , 21 Similarly, the second definition discussed in Section  is not entirely uninformed by consequences. As it was originally introduced (Stokes ), the definition was constructed to describe a phenomenon that is not well interpreted in any of the standard alternative ways given by critics. And one way to understand these interpretations as alternatives is that if they are apt for a given case, then that case implies none of the relevant consequences. For example, a case best interpreted as a cognitive effect on judgment implies no relevant consequences for theories of perception and knowledge, or for modularity of perceptual systems. So although the prescription here is to abandon such essentialist definitions, some of them are more informed than others by consequentialist concerns. None of this, of course, settles the debate. This was not the ambition of this paper. The ambition was instead to motivate a conceptual strategy for refining theoretical and empirical research on cognitive-perceptual relations of interest to a variety of theorists of mind. The hope is that this strategy—consequentialism—will enable new progress on an important set of questions about the human mind.

Acknowledgments Thanks first to the editors for organizing this volume, and providing feedback on this paper. Thanks to Vince Bergeron, Fiona Macpherson, and Wayne Wu for discussion of these and related ideas. Thanks to the Utah Philosophy Department, where this paper was given as a department colloquium in Winter ; in particular, thanks to Steve Downes, Matt Haber, Eric Hutton, Kim Johnston, Lucas Matthews, Lex Newman, Anya Plutynski, Jonah Schupbach, Cindy Stark, Jim Tabery, and Blake Vernon. Finally, thanks to audiences at the  Annual Meeting of the European Society for Philosophy and Psychology, the  Annual Meeting of the Cognitive Science Society, and the  Riga International Summer School in Cognitive Sciences and Semantics.

20 Note that here Fodor is employing the two epistemic consequences to dismiss certain kinds of data. At the same time, however, he is acknowledging some kind of influence on perceptual processing. So whether consequences for modularity follow will importantly depend upon the strength of the posited modules— e.g. whether they must be informationally encapsulated (with respect to the mentioned background information). 21 Similarly, Raftopoulos (; ) appeals to an epistemic criterion to evaluate certain arguments and empirical evidence for cognitive penetration. Briefly, he argues that early vision is unaffected by anything sufficiently theoretical. And perceptual experience (later vision, if one likes) is affected in a top-down way, but via the control of spatial attention. Because this latter effect is indirect, Raftopoulos argues, the epistemic consequences do not follow. Here the first and second of the three consequences of disjunctive consequentialism—theory-ladenness and the general epistemic consequence—are collapsed into one and used as the criterion for cognitive penetration.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

a consequentialist understanding



References Balcetis, E., and D. Dunning (). Wishful seeing: desired objects are seen as closer. Psychological Science : –. Bayne, T. (). Perception and the reach of phenomenal content. Philosophical Quarterly : –. Brewer, W. F., and Lambert, B. L. (). The theory-ladenness of observation and the theoryladenness of the rest of the scientific process. Philosophy of Science : –. Bruner, J. S., and Goodman, C. C. (). Value and need as organizing factors in perception. Journal of Abnormal and Social Psychology : –. Churchland, P. M. (). Scientific Realism and the Plasticity of Mind. Cambridge: Cambridge University Press. Churchland, P. M. (). Perceptual plasticity and theoretical neutrality: a reply to Jerry Fodor. Philosophy of Science : –. Deroy, O. (). Object-sensitivity or cognitive penetration of perception: the case of coloured shapes. Philosophical Studies : –. Deroy, O. (forthcoming). Modularity of perception. In M. Matthen (ed.), Oxford Handbook of Philosophy of Perception. Oxford: Oxford University Press. Feyerabend, P. (). Explanation, reduction, and empiricism. In H. Feigl and G. Maxwell (eds), Minnesota Studies in Philosophy of Science vol. , –. Minneapolis: University of Minnesota Press. Fodor, J. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, J. (). Observation reconsidered, Philosophy of Science : –. Fodor, J. (). Precis of The Modularity of Mind. Behavioural and Brain Sciences : –. Fodor, J. (). A reply to Churchland’s ‘Perceptual plasticity and theoretical neutrality’. Philosophy of Science : –. Gibson, E. J. (). Principles of Perceptual Learning and Development. East Norwalk, Conn.: Appleton-Century-Crofts. Goldstone, R. (). Perceptual learning. Annual Review of Psychology : –. Goldstone, R. L., Landy, D. H., and Son, J. Y. (). The education of perception. Topics in Cognitive Science : –. Hanson, N. R. (). Patterns of Discovery. Cambridge: Cambridge University Press. Hanson, N. R. (). Perception and Discovery: An Introduction to Scientific Inquiry. San Francisco, Calif.: Freeman and Cooper. Hawley, K., and Macpherson, F. (). The Admissible Contents of Experience. Oxford: WileyBlackwell. Kottenhoff, H. (). Situational and personal influences on space perception with experimental spectacles. Acta Psychologica : –. Kuhn, T. (). The Structure of Scientific Revolutions. Chicago: University of Chicago Press. Lyons, J. (). Circularity, reliability, and cognitive penetrability of perception. Philosophical Issues : –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research : –. McCurdy, H. G. (). Coin perception studies and the concept of schemata. Psychological Review : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. stokes

Prinz, J. (). Putting the brakes on enactivism. Psyche : –. Pylyshyn, Z. (). Computation and cognition: issues in the foundations of cognitive science. Behavioral and Brain Sciences : –. Pylyshyn, Z. (). Computation and Cognition. Cambridge, Mass.: MIT Press. Pylyshyn, Z. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences : –. Raftopoulos, A. (). Reentrant pathways and the theory-ladenness of perception. Philosophy of Science (Proceedings) : –. Raftopoulos, A. (). Defending realism on the proper ground. Philosophical Psychology : –. Schwitzgebel, E. (forthcoming). The problem of known illusion and the problem of unreportable illusion. Philosophy of Science. Siegel, S. (). Which properties are represented in perception? In T. Szabo Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Siegel, S. (). The visual experience of causation. Philosophical Quarterly : –. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Siegel, S. (). The epistemic impact of the etiology of belief. Philosophical Studies : –. Siewert, C. (). The Significance of Consciousness. Princeton, NJ: Princeton University Press. Stokes, D. (). Perceiving and desiring: a new look at the cognitive penetrability of experience. Philosophical Studies : –. Stokes, D. (). The cognitive penetrability of perception. Philosophy Compass : –. Stokes, D. (). Cognitive penetration and the perception of art. Dialectica : –. Stokes, D., and Bergeron, V. (forthcoming) Modular architectures and informational encapsulation: a dilemma. European Journal for Philosophy of Science. Stratton, G. M. (). Vision without inversion of the retinal image. Psychological Review (): –. Taylor, J. (). The Behavioral Basis of Perception. New Haven, Conn.: Yale University Press. Van Gulick, R. (). Deficit studies and the function of phenomenal consciousness. In G. Graham and G. L. Stephens (eds), Philosophical Psychopathology, –. Cambridge, Mass: MIT Press. van Ulzen, N. R., Semin, G. R., Oudejans R., and Beek, P. (). Affective stimulus properties influence size perception and the Ebbinghaus illusion. Psychological Research : –. Walton, K. (). Categories of art. Philosophical Review : –. Wu, W. (). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

PA RT II

Modularity, Encapsulation, and Impenetrability

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Unencapsulated Modules and Perceptual Judgment Jack C. Lyons

To what extent are cognitive capacities, especially perceptual capacities, informationally encapsulated and to what extent are they cognitively penetrable? And why does this matter? (I will suppose that a penetrable system is just one that isn’t encapsulated, and conversely.) There are a number of reasons we should care about penetrability/encapsulation, but I will focus on two: (a) encapsulation is sometimes held to be definitional of modularity, and (b) penetrability has epistemological implications independent of modularity (in fact, it’s sometimes held that if modularity has epistemological implications, it’s because of encapsulation (Fodor , though see Lyons  for a contrary view)). My main concern is with (b), but I begin with a discussion of (a). I argue that modularity does not require encapsulation; that modularity may have epistemological implications independently of encapsulation; and that the epistemological implications of the cognitive penetrability of perception are messier than is sometimes thought.

 Modularity and Encapsulation Recent discussions of modularity owe a lot to Fodor (). Fodor deliberately declines to define ‘module’ and explicitly insists that modularity comes in degrees, but at the same time he offers a set of nine diagnostic features that, if taken as definitional of modularity, provide a quite demanding theory of modularity. The most important of these features are speed, involuntariness, innateness, domain specificity, introspective opacity, and informational encapsulation. Though few endorse Fodor’s view in its totality, one of the major theses of the book was that cognitive capacities that have some of these properties—‘to some interesting extent’ (p. )—tend to have the others— again, to an interesting extent. This is still an important insight, despite the vagueness of ‘to an interesting extent’; and it is, as far as I know, widely regarded as true, despite what Fodor’s opponents and his more recent self have done to obscure this

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

contribution by focusing on a sharper but more extreme version of the original proposal. Even if learning informs the development of perceptual systems, for example, they’re still innately constrained ‘to some interesting extent’; even if they are cognitively penetrable, such cognitive penetration is not entirely rampant; etc. Fodor starts out careful to deny that the nine features are either categorical (e.g. a system need only be innate to an interesting extent to count as modular) or definitional of modularity, though he sometimes (here in , but even more so in  and ) acts as if they are both. Fodor’s work on modularity embodies several distinguishable theses; three are worth singling out here. I have already mentioned a claim we might call the Cluster Thesis, which holds that capacities exhibiting some of the aforementioned properties tend to exhibit them all. A second assertion, the Input Thesis, claims that all and only the input systems are modular, ‘central’ systems allegedly lacking these nine properties. These are both distinct from the Plurality Thesis, which claims that the mind is not a single, indivisible Cartesian/Lashleyan whole, but a collection of relatively independent systems. The first and third theses, especially, are fairly uncontroversial, although it is easy to read Fodor as endorsing implausibly radical versions of all three claims, by insisting (a) that the nine diagnostic features constitute necessary and sufficient conditions for modularity, (b) that these features must be present to more than just some interesting extent (e.g. that modularity requires a level of innateness that precludes genuine perceptual learning), and (c) that systems failing to satisfy these very strict criteria are therefore radically Quinean (Fodor ) and inseparably intermingled. I don’t claim that this is, in fact, Fodor’s view, but only that it’s not hard to read this into him. Just as a suitably understood version of the Cluster Thesis is pretty uncontroversial, so too there is some version of the Plurality Thesis that should be widely acceptable. This is important, because the Cluster and Input theses presuppose the Plurality Thesis, but not vice versa. A weak doctrine of modularity—or a doctrine of weak modularity— is committed to the Plurality Thesis, and it uses the term ‘module’ to refer to these relatively independent systems. This doctrine of weak modularity is one that I’ve articulated elsewhere (Lyons ), but it’s worth reiterating some of the highlights here. This modularity doctrine doesn’t require innateness, even to an interesting extent; it doesn’t require speed, introspective opacity, etc. It does require something like domain specificity and something superficially like informational encapsulation, though it turns out that the differences between these and what the doctrine does require are deeper than the similarities. When we talk about a ‘system for face recognition’ or the like, we are talking, to a first approximation, about a unified and separable entity that performs the task of face recognition. Although we name tasks by reference to their outputs, it is better, I argue, to think of tasks as input–output functions; and it is convenient for expository purposes to adopt an extensional understanding of functions: as sets of ordered (input–output) pairs. Modularity is a partly implementational concept, and we need

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



to think about the mechanisms, or substrates, that compute these tasks. Suppose a substrate S computes a function. I say that S specializes in T iff T is an exhaustive specification of the input–output function that S computes. This is merely a restriction on the naming of systems (one that is frequently violated without much harm); strictly speaking, it’s not a face recognition system if it’s also involved in recognizing individual cows, bird species, etc. S is isolable with respect to task T iff S computes T and could do so even if no other substrate computed any functions. That is, S, if given one of the inputs of T, is capable by itself of producing the appropriate output, without the assistance of other substrates. Isolability is thus a counterfactual issue about the computational capacities of a substrate. Finally, we need to distinguish parts of tasks from subtasks. A subtask is a task that is computed by a mechanism on the way to computing something else; a part of T is simply a subset of the input–output pairs that constitute T. S is unitary with respect to T iff no proper part of S specializes in and is isolable with respect to any proper part of T. Unitariness ensures that our substrates will be the smallest mechanisms needed for the computation (my leftinferior-temporal-cortex-plus-the-doorknob doesn’t compute anything that left IT doesn’t compute by itself) and that the substrates and tasks are non-gerrymandered (if there’s a system for visual face recognition and one for auditory melody recognition, there will be a disjunctive substrate that computes face-recognition-or-auditorymelody-recognition, but the substrate won’t be unitary with respect to this task). This gives us a theory about cognitive systems: S realizes a system for T iff S is isolable with respect to T, is unitary with respect to T, and specializes in T. I think these cognitive systems are what people mean by the term ‘module’ these days,1 so I’ll simply call them ‘modules’ (in fact, it’s what I’ll mean by ‘module’ henceforth), and we can come up with another term, like ‘F-modules’, for the ones that satisfy Fodor’s criteria. Now perhaps many modules, even in this weak, nonFodorian sense, happen to be domain-specific and informationally encapsulated ‘to some interesting extent’. But notice how different task specificity is from full-blown domain specificity, isolability from informational encapsulation. Anything that computes a function is trivially task-specific, but domain specificity—whatever exactly that is—is surely intended to be harder to come by. A system that specialized in the firstorder predicate calculus would be task-specific but would presumably not count as domain-specific. It is hard to know, as the notion of domain specificity has never, to my knowledge, been spelled out in nearly as much clarity as has the notion of computation of functions. One not insignificant advantage of my view over some other views (e.g. Fodor ; Coltheart ) is that it doesn’t require us to figure out what counts as domain specificity. 1 In Lyons (), I presented the view as an alternative to a modularity view, although this choice was entirely terminological. I had assumed, probably incorrectly, that ‘module’ would retain enough ties with the Fodorian view that we would need some other term for the components posited by the Plurality Thesis. I think the term is now used in a less restrictive way, although I don’t care much if I am mistaken about usage. Ordinary language philosophy is bad enough when it’s about ordinary language; how much worse when it’s about terms of art!

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

More importantly, isolability and encapsulation should not be confused. My brain is isolable from your brain (each can compute even without the aid of the other), but the fact that we’re communicating means that they aren’t informationally encapsulated from each other.2 Isolability is about system boundaries, about what is required to have an intact computational device; encapsulation is a matter of where a given system gets its inputs from. I assume that S is encapsulated from S iff S does not receive any inputs from S . This strikes me as the natural view of encapsulation, although Carruthers () offers a surprising alternative. For some reason—perhaps he is thinking of encapsulation as a monadic property—he starts by defining encapsulation as a mechanism’s being unable to draw on outside information in addition to its input. Because the natural way to understand input is just as whatever information a mechanism draws on, there is an obvious threat of trivialization, which Carruthers handles by defining ‘input’ in a more restricted way. These complications vanish on a two-place relational understanding of encapsulation, as just described. If we want a monadic conception of encapsulation (outright) as well, we can say that an encapsulated system is one that doesn’t receive inputs from any other system. Obviously, the only systems that might satisfy the monadic conception would be ‘input’ systems that take their inputs from sensory transducers, rather than other cognitive systems. Many ‘central systems,’ however, may be encapsulated from each other and from various input systems. Fodor has argued (; ) that encapsulation has a special role to play—that even if the rest of the nine criteria are optional, encapsulation really is a necessary condition for modularity, presumably at least partly on the grounds that ‘it is a point of definition that distinct functional components cannot interface everywhere on pain of their ceasing to be distinct’ (: ). But this isn’t right. Isolability suffices for distinctness, even though isolability imposes no restrictions at all on information exchange: a and b could be isolable even if they shared everything, so long as that sharing was unnecessary. If a could go on without b, then a is isolable from b. Consider again my brain and yours; if we had perfect telepathy, the brains would ‘interface everywhere’, but this would not imply that they were not distinct (this is especially obvious if the telepathic communication were voluntary). It is also important to point out that isolability is different from dissociability, at least if the latter is read as indicating an ability to produce the normal outputs in the absence of other mechanisms. Stokes and Bergeron, in an unpublished paper, point out that my understanding of modularity is superior to Carruthers’s (), at least as they understand him.3 They view Carruthers (see : ) as holding that dissociability is a hallmark of modularity, and although he is not fully explicit about 2 I have defined isolability (with respect to a given task) as a monadic property of a substrate, but a twoplace relation of isolability-from is easily defined: S is isolable from S with respect to T iff S computes T and could do so even if S didn’t compute any functions. 3 Stokes and Bergeron (forthcoming). This isn’t how they put it, and it may not be quite their intention, but credit for recognizing this belongs to them.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



this, as holding that S is dissociable from S iff S could operate normally even if S were removed. This would be the same as my understanding of isolability, if normal operation were understood in terms of performing input–output mappings, but it is a very different notion if understood in terms of producing the normal outputs. S might receive indispensable inputs from S , in which case the removal of the latter would prevent the former from operating normally, i.e. from producing its normal outputs. S and S might, however, still be isolable from each other, in that even though S needs inputs from S , S is still capable of performing its input–output mapping without S ; i.e. S could, if given the inputs it would normally have received from S , produce the appropriate outputs without further assistance from S . I’m not sure that this is how Carruthers intends dissociability; perhaps he has something more like my isolability in mind. In any case, this contrast illustrates the proper understanding of isolability and the theory of modularity that incorporates it. Suppose perception is cognitively penetrated by beliefs, desires, and the like. Then the perceptual systems are receiving input from higher cognitive mechanisms and are therefore not encapsulated. This is not by itself any threat to the existence of perceptual modules, however, for it is no threat to the isolability of the substrates responsible for perception. It may be that the reason beliefs influence perception is that there is really no distinction between the module responsible for those beliefs and the module responsible for perceptual states, i.e. that belief production and percept formation are both part of the task of a single, indivisible module. But another potential reason is that, although perceptual modules are distinct from belief-forming systems, the former receive inputs from the latter. The mere fact of top-down influence (i.e. on perception, from beliefs, etc.) is compatible with either possibility. Even if the inputs were indispensable, in the sense that the perceptual systems would be incapable of producing percepts without input from higher cognition, this would not threaten the distinctness of the perceptual systems; their isolability requires only that they be able to compute a certain perceptual function if given certain inputs, in this case inputs from higher cognition; it is the mapping, not the output, that they must be independently capable of. F-modular systems are by definition encapsulated, and this has important consequences for the frame problem, locality of computation, and other issues of tractability, as Fodor (e.g. ; ) has rightly pointed out. Consequently, we should want to know how many modules in the present, weaker sense are also F-modules. But this weaker understanding of modularity is by itself sufficient to flesh out the Plurality Thesis, and that thesis remains a substantive and insightful claim. It is, as I mentioned earlier, quite plausible, especially in light of the various neuropsychological dissociations, but it is far from trivial. These dissociations are, of course, quite surprising from a commonsense perspective and the architectural thesis they support is highly revisionary of our pretheoretic assumptions.4 4 Prinz () holds that the Plurality Thesis is uncontroversial. This hardly makes detailed articulation of that thesis a worthless endeavour, however. Nor is this thesis trivial simply because it’s uncontroversial

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

 Penetration and Encapsulation Before turning to the implications of encapsulation, penetrability, and modularity for the epistemology of perception, I should say a bit more about cognitive penetrability and its relation to encapsulation. When a module (or system, or capacity) is described as being cognitively penetrable, something more is being claimed than mere failure of encapsulation. Talk about cognitive penetrability is usually intended to indicate a topdown influence—in particular, an influence by the occurrent (and perhaps fleeting) beliefs, desires, fears, goals, etc. of the cognizer. The McGurk effect (McGurk and McDonald ) is a nice illustration of penetrability without cognitive penetrability. Vision influences audition (seeing/ga/makes you hear/da/instead of/ba/), so this auditory system is not encapsulated from vision. However, the effect is a classic example of cognitive impenetrability; knowing the trick behind the illusion does nothing to dispel it, any more than knowing about the Müller–Lyer illusion affects the relative apparent lengths of those lines. What we know isn’t influencing what we hear, but what we see is. I’ll call this ‘lateral penetration’ to distinguish it from cognitive penetration. The latter is a species of top-down influence, while the former involves influence from one system to another that is at least approximately at the same ‘level’ as itself.5 Lateral penetration and cognitive penetration can interact in interesting ways, making certain modules indirectly cognitively penetrable. If vision were cognitively penetrated in McGurk cases, and cognitively influenced visual states went on to laterally influence auditory experience, these auditory experiences would count as cognitively penetrated, even though the connection was indirect. Yet ‘receives input from’ is not transitive. It’s not the case that if module B receives input from A and C receives input from B, then C receives input from A, because it might be that, of all B’s outputs that were responses to inputs from A, none of these is ever fed to C, and so C never receives any information from B that was influenced by A. In such a case, C might remain encapsulated from A (all else being equal). It is tempting to claim that if some of B’s outputs that were responses to inputs from A are ever fed to C, then C is receiving inputs from A and is thereby penetrated by A. Maybe this works, but unless we’re smuggling a lot into the notion of inputs, it will have to be more complicated than this. Consider several cases: . Fodor’s heart rate: Fodor once () joked about his heart rate being cognitively ‘penetrable’ on the grounds that an intention to do calisthenics results in his doing calisthenics, which results in increased heart rate. Heart rate obviously isn’t a psychological phenomenon, but the general point is clear enough. among cognitive scientists. Although we’re all used to it by now, it is really a shocking discovery, and one that turns common sense on its head. 5 For the present purposes (but only for these), I will suppose that any system lower than the level of beliefs and desires etc. counts as approximately at the same level. Thus, if late vision feeds back into early vision, this will count as lateral penetration for the present purposes. For other purposes, of course, we would want more than just two kinds of penetration: lateral and cognitive.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



. Change in fixation: The same moral applies when the ‘penetrated’ capacity really is psychological. I deliberately move my eyes, or turn around, thus altering what I fixate on and thus what I see. . Change in attention: Without moving my eyes, I change the locations or objects to which I am devoting attention, which affects my visual experience (e.g. the Necker cube shifts, or the subject in the old woman/young woman drawing now looks like the young woman). . Oculomotor efference copy: Your eyes are paralysed, but you try to move them to the left, thus causing an apparent shift or relocation of the objects in your field of view (Kornmueller ; Whitham et al. ). Let’s suppose that the way this works is that the intention causes the motor areas to send not only a signal to the oculomotor muscles but also an efference copy to the visual system so that the visual system can update a post-retinotopic representation accordingly. Thus the desire indirectly influences the post-retinotopic representation and the resulting visual experience. . Effort and distance: Subjects who intend to throw a heavy weight to a target judge the distance to that target to be greater than do subjects who have no such intention (Witt et al. ). Suppose it works as follows: the intention to throw the weight causes an activation of motor readiness routines, where the readiness reflects the expected required effort. The visual system takes the degree of readiness as a cue to distance, with the result that more effort-requiring action plans lead to perception of longer distances. This hypothetical account is similar to another, perhaps more familiar, one: . Mind-reading and covert mimicry: Suppose a visuomotor system for mimicking facial expressions feeds into a mind-reading system, thusly: perception of facial features activates motor plans for making the same expressions, thus sending (usually sub-threshold) signals to one’s own facial muscles. At the same time, the somatosensory systems are informed to expect the relevant facial movements, and this information feeds into the mind-reading system, which then attributes the emotion that corresponds to that expression to the person being perceived (Adolphs et al. ; Goldman ). I think that cases – are pretty clearly not instances of cognitive penetration, as it is usually understood in the field, and that case  pretty clearly is. Case  is not, in part because there is no belief, desire, or goal influencing the mindreading system. Consider a variant, however: . Mind-reading and mimicry II: Suppose instead of (possibly unconscious) perception of facial features activating motor plans for mimicry, it’s the perceiver’s beliefs about facial expressions that produce signals to her own facial muscles. The rest of the story is the same.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

Now we have a case of cognitive penetration of the mind-reading judgment, even though the mind-reading system itself is not directly penetrated. Cases  and  are easy to handle, since it is obvious that the perceptual systems are not receiving inputs from the motor systems. Their inputs change as a result of the outputs of the motor systems, but that’s not the same as taking inputs. I’m not sure how to make this more precise, but I think that the difference between cases – and case  is that in the former case, the causal chain is mediated by overt (, ) or covert (, ) action, or behaviour, on the part of the cognizer. Notice that the covert behaviour in case  is something the agent is sometimes aware of engaging in, while in case  it is completely unconscious. There can be activation of motor systems without the execution of actions, which is what is happening in cases –, though again, I’m not sure how to make the notion of action more precise.6 We can incorporate this into our understanding of penetrability in a few different ways; I’ll do it by including the absence of behavioural mediation as a restriction on inputs. (I think this is more of a terminological than a substantive matter.) B receives input from A only if A’s influence on B is not behaviourally mediated in the sense just glossed. Now, I think, we can safely endorse the earlier suggestion that if some of B’s outputs that were responses to inputs from A are ever fed to C, then C is receiving inputs from A and is thereby penetrated by A. This gives us a fairly restrictive working understanding of penetration and an even more restrictive understanding of cognitive (as opposed to lateral) penetration. Cognitive penetration occurs when a system takes beliefs, desires, goals, or other similar person-level states as input, where ‘input’ can be direct or indirect, but cannot be behaviourally mediated.7 I will use ‘influence’ to characterize the broader set of causal relations: cases , , , , and  are all instances of cognitive influence on perception, even though , , and  are not cases of cognitive penetration.

 Epistemological Implications of Cognitive Penetration The classical literature on the epistemological implications of cognitive penetrability (e.g. Fodor ; ; Churchland ) and even some of the recent literature on

6 I intend cases  and  to be understood in such a way that the somatosensory systems receive their inputs whether or not a motor command ever activates the facial muscles; otherwise the signal is behaviourally mediated. 7 Two caveats: first, this isn’t meant as a piece of conceptual analysis; cognitive penetration is a highly theoretical concept. Still I think this articulates the understanding of cognitive penetration most standardly employed in the field; and whatever terminology we use, it is important to distinguish relevantly different phenomena. Second, I’m leaving somewhat open the question of which states count as person-level and cognitive in the relevant senses. I don’t think they need to be conscious to count as cognitive penetrators, but they may need to be the sort of thing one could change one’s mind about, fairly quickly, as the result of practical or theoretical reasoning.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



the subject (Raftopoulos ; Siegel ) lead one to suspect that some or all of the following are common assumptions: • that cognitive penetration of perception is epistemically worse than lateral penetration; • that cognitive penetration is epistemically worse than other forms of cognitive influence on perception; • that cognitive penetration of perception is epistemically worse than cognitive penetration of perceptual judgment; • that replying to the epistemological worries generated by the possibility of cognitive penetration requires defence of encapsulated perceptual systems. I won’t try to pin these views on anyone; whether or not anyone has explicitly held them, it is worthwhile to show that they are all suspect.

. Is cognitive penetration worse than lateral penetration? In some sense, I agree that cognitive penetration is epistemically worse than lateral penetration, though not for what I take to be the usual reasons. The usual worries about penetration seem to focus on (a) the failure of convergence among perceivers with different theoretical preconceptions, and (b) an illegitimate kind of self-corroboration, akin to epistemic circularity. I have argued elsewhere (Lyons ) that (a) doesn’t have any obvious bearing on an individual’s being epistemically justified in believing what she seems to perceive, and that (b) fails to capture what is wrong with the bad cases of cognitive penetration. When a belief or desire influences perception, this is not a matter of the perceptual belief ’s being epistemically based on the penetrating belief or desire; if it were, all such basing would seem to have to have the same structure and hence the same epistemic status. But not all cognitive penetration is vicious. If my fear of snakes primes certain high-level object recognition templates, facilitating matching and thus making me better at detecting whatever snakes are in the woods around me, this would be an epistemically innocuous—perhaps even virtuous—kind of cognitive penetration of perception. If the fear makes me more likely to mistake sticks for snakes, however, or otherwise render me less sensitive to my actual environment, then this is an epistemically pernicious sort of penetration. In addition, if the problem with cognitive penetration were about circularity, it would be hard to see what’s wrong with penetration from desires and fears, as opposed to beliefs. There is perhaps something vaguely (though, for reasons just mentioned, not strictly) circular about believing that p, which causes you to see that p, which causes you to believe that p. But desires are quite different. Furthermore, at least in the anecdotal cases, desires can have opposite effects: sometimes the desire that p makes me more likely to see that p, while other times a very strong fear that p (and thus desire that not-p) makes me more likely to see that p. It is hard to see how either could be a matter of circularity, given that they’re not beliefs, but even harder to see how they both could, given their opposite natures.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

Nothing about the cognitive nature of the penetrator makes any difference here. What matters is whether the penetration—wherever it might originate—makes me a better or a worse perceiver. I argue (Lyons ) that ‘better’ and ‘worse’ are best understood in terms of a reliabilist epistemology, though there are other ways one might try to understand them.8 It might, as a matter of fact, turn out that lateral penetration is more likely to improve perception than cognitive penetration. But even if that is true, it makes the specifically cognitive nature of the penetrator incidental; what matters is whether the penetration makes for improvement or not.

. Is cognitive penetration worse than other kinds of cognitive influence? What about other kinds of influence, besides penetration? Fodor (), Pylyshyn (), and Raftopoulos () are all insistent that Change of fixation and Change of attention do not involve cognitive penetration. Certainly it is true that different forms of cognitive influence on perception are importantly different, and that using ‘cognitive penetration’ in a restrictive way is a good means of registering those important differences. It is not obvious that these important differences are epistemically important differences. A couple of paragraphs back, I described a hypothetical means by which fear of snakes might cognitively penetrate snake perception. Suppose it worked pre-perceptually, instead, by priming attentional mechanisms in such a way that attention is more likely to be captured by snake-like items in the environment. It still seems—to me, anyway—that what matters for the epistemology is whether this cognitive influence makes me better or worse at spotting snakes. If the former, then the influence is epistemically good; if the latter, then it’s epistemically bad. Balcetis and Dunning () describe the kind of ‘wishful seeing’ that recent epistemological discussions have taken to be the paradigm of epistemically pernicious cognitive influence. Thirsty subjects see the water bottle as closer than do nonthirsty subjects; subjects see the gift card with what they are told is a $ balance as nearer than the one with the $ balance. Suppose these results are genuine. These distance judgments seem to be less justified than distance judgments that are not thus contaminated by desire. Does it matter, for this verdict, whether the effect in question is genuinely a case of cognitive penetration, in our narrow sense? Here are three possible mechanisms for the effect: a) The direct cognitive penetration of vision by desire: subjects want the item to be closer, and this makes it look closer. 8 Siegel () claims that the difference between the innocuous and the pernicious cases of penetration is a matter of the rationality or irrationality of the etiology, which is a matter of being sufficiently analogous to a rational or irrational doxastic etiology. Whether this will work out, and whether it will offer different verdicts than my view, will depend on whether the details concerning analogy and rationality can be worked out. It’s not obvious that the reliabilist can automatically get the cases right; whether she can depends in part on how we are to individuate processes, which is obviously a vexed issue for reliabilism. I make some initial stabs at this problem in Lyons (in preparation). See McGrath () for another internalist approach to distinguishing good and bad cases of penetration.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



b) A physiological effect: thinking they’re going to get money makes subjects happy, and this causes biochemical changes that affect perception in systematic ways, making objects look closer, hills less steep, etc. c) An attentional effect (Alter and Balcetis ): desired objects demand greater attention, and this makes them appear closer. Some of these mechanisms might make us feel better or worse about the prospects for human perceptual judgment in general—an attentional effect, for example, might be easier to deliberately counteract. But the epistemic status in the present case doesn’t seem to depend on which mechanism is in play. If distance judgments are contaminated by desire, then this is epistemically bad for those judgments, regardless of the means of contamination. We will rightly care about the means for other reasons—curiosity about psychological mechanisms, concern for the contamination of other judgments—but those reasons seem disconnected from the epistemological verdict about the individual case. Mechanism (b), just considered, would be an instance where the mediation isn’t even entirely psychological. Consider a similar possibility that is another variant of Mind-reading and covert mimicry. It is known that manipulating facial expressions affects subjects’ assessments of their own moods (Strack et al. ), and if simulation plays any role in mindreading, something like the following seems possible: . Perverse mimicry: I’m holding a pencil in my mouth sideways, which makes my facial muscles approximate a smile, and this makes me more likely to attribute positive emotions to others than I would have been without the pencil. My intention to put the pencil in my mouth clearly influences my third-person mental state attributions, and this influence is patently nothing like cognitive penetration, given the highly indirect and extramental nature of the causal link. Nevertheless, does this decision affect the justification of these attributions? Suppose, for further specification of the case, that I’m holding the pencil in a deliberate attempt to positively influence my own mood, while oblivious to the possibility that this might affect my judgments of others.9 Epistemologically speaking, this case seems to me to be relevantly similar to cases of cognitive penetration. I’m not in any sense blameworthy for the effect my action has on my judgments (nor am I in the standard cases of pernicious cognitive penetration), but still, these judgments are being influenced in such a way that I have temporarily become a worse judge concerning the emotional states of others. One interpretation of the situation is that, although the people around me look happy, they shouldn’t look happy; that is, this perceptual or quasi-perceptual state 9 I don’t think it matters whether this is my motivation or whether I just like chewing on pencils. Obviously, if I do it knowing or suspecting that it might affect my mind-reading judgments, then this adversely affects the status of those judgments. That’s not an interesting case, however, so we should assume I’m unaware of this possibility.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

I’m in—of others looking happy to me—is a state that I should not, epistemically, be in. Another interpretation foregoes evaluation of the ‘looks’ state and claims merely that the perceptual or quasi-perceptual judgment is to some extent ill-formed because it was influenced by factors it should not have been influenced by. In any case, the kinds of accounts that would be given for the standard cases of pernicious cognitive penetration seem equally appropriate here, even though the case is far from being an instance of cognitive penetration.10

. Is cognitive penetration of perception worse than cognitive penetration of perceptual belief? In discussions of cognitive penetration, the term ‘perception’ tends to have a narrower meaning than usual. This makes room for the possibility that perceptual belief— in some more or less ordinary sense of the term—is cognitively influenced even if perception—in this narrow sense—is not. What is this narrow sense? There is almost certainly not just one, but let’s start by looking at a much discussed line of research by Proffitt and colleagues (e.g. Proffitt ), who claim to find that wearing a heavy backpack makes hills look steeper and that holding a heavy weight makes the target to which that weight is to be thrown look farther away. Suppose this is a cognitive rather than a lateral effect: it’s not the mere haptic sense of weight that affects the visual appearance, but the expectation that the subject may have to climb the hill or throw the weight, or an imaginative rehearsal of doing so. One line of criticism of this research (Durgin et al. ; Durgin et al. ) argues that the effect results from a response bias due to experimental demand characteristics; the idea is that subjects can guess that the experimenter wants to show that the presence of the backpack will make hills look steeper, and they— unconsciously and unintentionally—comply by judging the hills to be steeper.11 This would certainly be a cognitive effect, but it’s not cognitive penetration of perception, because the effect is entirely post-perceptual. The ensuing debate has been rather inexplicit about what counts as perceptual vs. post-perceptual in this context (see also Witt ; Woods et al. ). Elsewhere, one encounters two main views about the perceptual/post-perceptual distinction. The first (confined for ease of discussion to the visual modality) maps the perceptual/postperceptual distinction onto the distinction between early vision and late vision (Pylyshyn ; Raftopoulos ). Early vision comprises the processing up to about the level of Marr’s () /-D sketch; it ‘extends to the construction of the sort of percept that we might have of a totally unfamiliar scene . . . where nothing is recognized 10 Here it is important that, as mentioned above and argued in Lyons (), the problem with cognitive penetration is not one of circularity. There’s clearly nothing circular about the current case. 11 It is important for the present purposes that this be a matter of judgment, rather than insincerely accommodating report, on the part of the subjects, i.e. that they really believe the hills to be steeper and aren’t just saying so to please the experimenter. This seems to be how Durgin et al. () interpret things, and it will be how I understand the case.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



as familiar’ (: ); it includes information about ‘spatiotemporal [properties], spatial relations, surface shading, orientation, color, binocular stereopsis, size, shape and movement’ (Raftopoulos and Müller : ). One reason for drawing the line here is that perceptual processing up to this point seems to be encapsulated, while further processing seems not to be. A second way of drawing the distinction locates the perceptual/post-perceptual divide at the point where visual (perceptual) experience and its associated sensory phenomenology gives way to something else (Siegel ). Sometimes this something else is cognition without phenomenology, sometimes it’s cognitive phenomenology. Though it’s hard to make this very clear, we seem to have a rough intuitive grasp on the notion of perceptual experience, as something distinct from belief and which has a rich sensory phenomenology. The first way of drawing the distinction embodies an architectural approach, while the second embodies a phenomenological approach. It is unlikely that these two approaches converge. Proponents of the latter approach often endorse a liberal, or rich-content, view about the content of perceptual experience, allowing it to represent high-level contents, not just things like dog and cat (which are already richer than early vision) but perhaps even semantic and causal relations (Siegel ; ), highly subordinate category membership (e.g. being a  Cadillac Eldorado), etc. I will suppose that our perceptual judgments (or perceptual beliefs—I will use these terms interchangeably) include concept-level attributions of properties to objects and identification of particulars and of category members at the basic or entry level, judgments like (depending on the perceiver and circumstances) ‘the dog is on the sofa’, ‘that’s a giraffe’, ‘it’s raining’, ‘this pumpkin pie is chewy’, and plausibly ‘there’s Martha Jones in a  Eldorado’. There is quite a gap between such judgments and the outputs of early vision. The phenomenological approach sees a much tighter relationship between perception (i.e. perceptual experience) and perceptual judgment, especially if it takes a liberal view about the contents of perceptual experience. My own view (Lyons a: b; ) is something of a compromise between these two approaches. I endorse a different kind of architectural account, though my target is perceptual belief, rather than perception in one of the narrower senses. The account starts with a view about what a perceptual module is—it is one that takes inputs from sensory transducers and none of the other inputs to which are under the direct voluntary control of the larger organism—and defines perceptual beliefs as the doxastic outputs of perceptual modules. This gives us a fairly liberal view about the contents of perceptual belief (on the plausible assumption that ‘the dog is on the sofa’, ‘there’s Martha Jones in a  Eldorado’, and the like will satisfy the definition), while remaining agnostic about—and thus compatible with a very conservative view regarding—the contents of perceptual experience. I take the agnosticism, if not the conservatism, to be a good thing, in part because it is very difficult to tell where sensory phenomenology leaves off and cognitive phenomenology begins. When the expert and I both look at a pileated woodpecker,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

it looks like a bird to both of us, but it also looks to her (and not me) like a pileated woodpecker. Our overall experiences are different, but is there any reason to think that there must be a difference in visual experience, in sensory phenomenology? Obviously, the expert has a firm sense that it’s a pileated woodpecker, but why should we not think that this is merely a matter of cognitive phenomenology? You and I look at the same painting, and I am struck with the conviction that I saw this painting yesterday. There’s an introspectable, phenomenological difference between us, but it doesn’t seem to be a visual phenomenological difference. Why think the woodpecker case is any different? Consider a case of spectral inversion without error: you and I have inverted colour qualia, but we associate these qualia with different colour concepts, in such a way that our colour judgments all come out to be the same.12 You and I both look at a green thing in standard lighting conditions. In some sense, we want to say that it looks the same (green) to both of us—whether or not we actually believe it to be green (we might currently be worried about skeptical scenarios). In another sense, however, and ex hypothesi, it looks very different to us: our visual experiences are different. Now, you look at a green thing and I look at a red thing. What you’re looking at looks green to you, and what I’m looking at looks red to me, even though our visual experiences are identical. I don’t know if this is the right way to describe what’s going on—I’m staying agnostic about the contents of experience, after all—but it strikes me as sensible and plausible. One who holds a certain type of liberal view about the contents of perceptual experience will, of course, deny that our perceptual experiences are really identical. Perhaps only some component of the experiences is identical. But it’s hard to see what’s motivating this denial other than the liberal theory about the contents of experience. My view is not, however, agnostic about the contents of perceptual judgments. I hold that you and I make the same perceptual judgments about colours, whatever the nature of our visual experiences. I hold that the expert’s belief that it’s a pileated woodpecker is a perceptual belief, even if her experiential state is the same as mine. It is at least as likely that perceptual judgment is cognitively penetrated as that perception (i.e. early vision, or visual experience, or whatever perception in the narrow sense amounts to) is, regardless of how we understand the latter, since if penetration of perception is possible, then penetration of perceptual judgment is, but not vice versa. Suppose the expert’s pileated-woodpecker identification is cognitively penetrated in such a way that her expectation that she might see some in these woods has made her better at spotting them. This seems have all the hallmarks of an epistemically good case of cognitive penetration, and it doesn’t seem to matter whether this penetration occurs early (it penetrates perception narrowly construed) or late (i.e. ‘post-perceptually’), on either approach to the perception/perceptual judgment distinction. Return to the debate about the perception of distances and steepness of hills. Proffitt and colleagues think these are cases of cognitive penetration of perception; Durgin and colleagues think the effects are post-perceptual. On my theory of perceptual belief, it 12

I recognize that not all philosophers will allow that this is possible; I won’t try to argue that it is.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



is plausible that the beliefs are perceptual beliefs in either case. Then, on either view, these are cases of the cognitive penetration of perceptual belief; on Proffitt’s but not Durgin’s view, they are also cases of the cognitive penetration of perception. (I will follow convention and proceed in terms of ‘post-perceptual’ penetration, though it is post-perceptual penetration of perceptual belief that I am concerned with, here and in what follows.) The epistemic situation seems to be the same regardless of where the penetration occurs. In neither case is the subject aware of what’s happening; in neither case is the subject in a good position to stop it from happening; but in both cases the resulting belief is influenced by factors that should not influence perceptual beliefs. This is not, of course, to claim that the beliefs have no degree of justification, just that they have less than they would had they not been penetrated in these ways. Whichever way the influence works, it makes the subjects worse at judging distances and slopes. If distance judgments really result from the subjects’ desire to help the experimenter, the epistemic status of these judgments is threatened in just the same way as if they result from the subjects’ anticipation of effort. Surely the overall epistemic consequences might be different. Response bias might introduce more serious distortions (threaten reliability in a worse way) than effort anticipation does; on the other hand, response bias relegates the bad epistemic news to those rare cases where there is an actual experimenter around to please, while leaving our judgments intact in most real-world applications. Nevertheless, the epistemic worries that arise from cognitive penetration at the perceptual level arise from cognitive penetration at the post-perceptual level as well. And in any case, if there is a difference in epistemic implications here, it doesn’t seem to generalize past the particular details of this case and the candidate mechanisms (i.e. response bias vs. anticipation of effort). There is no lesson here regarding early vs. late cognitive penetration in general; which is worse will have to be decided on a case-by-case basis. The cognitive impenetrability of the strictly perceptual is sometimes defended with vehemence that suggests that things would be worse, epistemically speaking, if cognitive penetration happened at the perceptual level than if it happened at the postperceptual level—that is, if, for some given case of cognitive penetration, it turned out to be perceptual rather than post-perceptual penetration (see e.g. Fodor ; Raftopoulos ). Perhaps it is thought that post-perceptual penetration would be somehow more corrigible than perceptual penetration, that if cognition penetrated early vision, or perceptual experience, we’d be stuck with that influence; we would have no way, aside from learning a lot of perceptual psychology, to detect and counteract these effects. And insofar as cognitive penetration makes us worse at judging perceptual matters, that would be a bad thing. However, a salient feature of the present cases is that even if the subject were to somehow know that her perceptual belief were cognitively penetrated, she would be in little or no position to tell whether the effect were perceptual or post-perceptual. This is an instance of a general truth with pervasive epistemological implications: we are bad judges of where perceptual experience leaves

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

off and perceptual judgment takes over. More generally, introspection and other nonempirical means are poor guides to drawing a line between the perceptual and the post-perceptual. (Obviously, I favour the architectural over the phenomenal method for drawing the perceptual/post-perceptual distinction.) So even if the penetration is a post-perceptual effect on perceptual belief, we seem to be just as stuck with it as if it had occurred at the perceptual level. In fact, to the extent that there’s an epistemological problem here, it is one that generalizes beyond even perceptual belief. Many of our belief-forming processes are ‘intuitive’ in the psychologist’s sense of the term: they are perception-like, System- (Kahneman and Frederick ) processes that operate quickly, automatically, and without the aid of conscious deliberation. Some of these are clearly cognitively penetrable. The belief bias (Evans et al. ), for example, occurs when our antecedent belief in the conclusion of an argument affects our intuitive judgment concerning the validity of that argument. This is obviously not cognitive penetration of perception, but it is obviously a species of the same genus. And it is problematic in just the same way. What makes this case worrisome, and what makes the perceptual case worrisome, is not just the absence of encapsulation, but the presence of another of Fodor’s diagnostic features: introspective opacity. We don’t have introspective access to the inner workings of our perceptual or intuitive processes, so when cognitive penetration does occur, we have no way of knowing that or counteracting it. This is true, however, not just of perception in one of the narrow senses but of perceptual belief and of intuitive judgment more generally.

. Do perceptual systems need to be encapsulated? I mentioned just now that the vigour with which the encapsulation of perception is sometimes defended suggests that responding to the epistemological worries of cognitive penetration requires that perception be encapsulated. But does it also require that perception be encapsulated? I won’t take a lot of time with this, as it isn’t very surprising, but the answer is no. Being encapsulated ‘to an interesting extent’ is good enough. Again, if we think that cognitive penetration has epistemological implications, we need to know what it is about it that makes it have epistemological implications. If the problem of penetration is one of epistemic circularity, then perhaps complete encapsulation would be necessary to guard against it. But as I’ve argued above and in more detail elsewhere (Lyons ), this isn’t what is epistemologically threatening about cognitive penetration. Rather, what is most epistemically worrisome about the very idea of cognitive penetration of perception is that it would threaten to cut us off from the world, to render us less sensitive to the world around us. In this respect, cognitive penetration is more of a danger, at least in general, than lateral penetration. In part, this is because cognition is already less directly linked to the world than perception. If I were simply to fall in with the wrong social crowd, they might, without even trying,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



convince me to believe and desire all kinds of crazy things. It would be relatively hard for them, even if they were doing it on purpose, to get me to see and hear crazy things. Whether cognitive penetration involves the influence of doxastic states like beliefs, or conative states like desires and fears, whether believing that p makes us more likely to perceive that p or less likely, we don’t want our perceptual beliefs to be endogenously determined. The whole point of perception is to put us in touch with the world as it actually is, and there’s little room here for endogenous determination. Or at least, putting us in touch with the world is the point of perception epistemologically speaking. If the point of perception evolutionarily speaking was to enable us to interact with our environments, and if the pursuit of this goal was sometimes better served by certain kinds of systematic falsehoods than by objectivity and truth, then the evolutionary and the epistemic aims of perception are sometimes at odds with each other. This is why Proffitt’s studies raise a credible threat. It is not absurd to think that our distance and slope judgments and the like might be influenced by considerations of effort; if I judge the distance to be longer when I’m tired or encumbered, then I don’t have to explicitly take encumbrance or tiredness into account in planning routes or predicting timetables, and the reasoning task is thereby simplified. Insofar as the penetration here is cognitive rather than lateral, we have an endogenous influence on our belief, one that reduces our contact with the world. Although this is a credible threat, it’s not a very threatening threat, for it doesn’t diminish our contact very much. We certainly don’t want our perceptual beliefs to be entirely endogenously determined, but degree of endogenous causation can vary, and the degree of epistemic degradation brought about by cognitive penetration can vary accordingly. It is a commonplace that perceptual beliefs are the result of the brain’s forming a best guess about the distal environment on the basis of a number (sometimes a large number) of different cues, none of which entails the presence of the inferred environmental features, and some of which have individually fairly low predictive value. Cognitive penetration might influence some of these cues without influencing others. If my intention to throw a given weight penetrates my motor anticipations, and these anticipations influence my visual estimates of distance, then these visual judgments are cognitively penetrated. But they are far from being entirely endogenously caused. The intention influences the value of one cue. But there are several other entirely visual cues to distance (e.g. texture gradients, aerial perspective, binocular disparity, eye convergence, angular projection of familiar items), and we have no reason to think that my intention to throw has any influence on any of these cues. Consequently, we might expect that the cognitive penetration doesn’t have a very dramatic influence on the distance estimate; it might influence it without influencing it much. Thus, penetration might make us worse at judging distances, without making us much worse. And in fact, our epistemological intuitions about the cases seems to be that, though the fact of penetration results in a lower degree of justification for the relevant beliefs, it may not be a very much lower degree of justification.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

Once again, we need to take things on a case-by-case basis. The belief bias strikes me as a far more pervasive and insidious problem of penetration than any of the perceptual cases I’m aware of. And the problems it poses are practical ones that are visible to the naked eye, as anyone who has paid attention to public discourse about politics or other matters of policy can easily attest. Yet there is room for optimism, at least in the perceptual realm, even though perception—perceptual belief, anyhow—is clearly cognitively penetrable.

Conclusion Perceptual systems might well be modular without being F-modular, i.e. without satisfying Fodor’s strict criteria for modularity. In particular, there is nothing incoherent about unencapsulated modules. The cognitive penetrability of perceptual modules has little direct epistemological consequence, as the penetration can be either epistemically advantageous or pernicious, depending on the details of the case. To the extent that perceptual modules have or approximate F-modularity, it is introspective opacity that threatens to render perceptual errors intractable, not penetrability.13

References Adolphs, R., Damasio, H., Tranel, D., Cooper, G., and Damasio, A. R. (). A role for somatosensory cortices in the visual recognition of emotion as revealed by three-dimensional lesion mapping. Journal of Neuroscience (): –. Alter, A. L., and Balcetis, E. (). Fondness makes the distance grow shorter: desired locations seem closer because they seem more vivid. Journal of Experimental Social Psychology (): –. Balcetis, E., and Dunning, D. (). Wishful seeing: desired objects are seen as closer. Psychological Science : –. Carruthers, P. (). The Architecture of the Mind: Massive Modularity and the Flexibility of Thought. Oxford: Oxford University Press. Churchland, P. M. (). Perceptual plasticity and theory neutrality: a reply to Jerry Fodor. Philosophy of Science : –. Coltheart, M. (). Modularity and cognition. Trends in Cognitive Sciences : –. Durgin, F. H., Baird, J. A., Greenburg, M., Russell, R., Shaughnessy, K., and Waymouth, S. (). Who is being deceived? The experimental demands of wearing a backpack. Psychonomic Bulletin and Review (): –. Durgin, F. H., Klein, B., Spiegel, A., Strawser, C. J., and Williams, M. (). The social psychology of perception experiments: hills, backpacks, glucose and the problem of generalizability. Journal of Experimental Psychology: Human Perception and Performance : –. 13 A version of the paper was presented at the Southern Society for Philosophy and Psychology in Austin, TX in . Thanks to the audience there, especially James Genone and Fred Dretske. Special thanks to Philippe Chuard for comments on a written draft.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

unencapsulated modules



Evans, J., Barston, J. L., and Pollard, P. (). On the conflict between logic and belief in syllogistic reasoning. Memory and Cognition : –. Fodor, J. A. (). Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, J. A. (). Observation reconsidered. Philosophy of Science : –. Fodor, J. A. (). Modules, frames, fridgeons, sleeping dogs, and the music of the spheres. In J. L. Garfield (ed.), Modularity in Knowledge Representation and Natural-Language Understanding, –. Cambridge, Mass.: MIT Press. Fodor, J. A. (). A reply to Churchland’s ‘Perceptual plasticity and theory neutrality’. Philosophy of Science : –. Fodor, J. A. (). The Mind Doesn’t Work That Way. Cambridge, Mass.: MIT Press. Goldman, A. I. (). Simulating Minds. New York: Oxford University Press. Kahneman D., and S. Frederick (). Representativeness revisited: attribute substitution in intuitive judgement. In T. Gilovich, D. Griffin, and D. Kahneman (eds), Heuristics and Biases: The Psychology of Intuitive Judgment, –. Cambridge: Cambridge University Press. Kornmueller, A. E. (). Eine experimentelle anastesie der ausseren augenmusckeln am menschen und ihre auswirkungen. Journal für Psychologie und Neurologie, Leipzig : –. Lyons, J. C. (). Carving the mind at its (not necessarily modular) joints. British Journal for the Philosophy of Science : –. Lyons, J. C. (a). Clades, Capgras, and perceptual kinds. Philosophical Topics : –. Lyons, J. C. (b). Perceptual belief and nonexperiential looks. Philosophical Perspectives : –. Lyons, J. C. (). Perception and Basic Beliefs. New York: Oxford University Press. Lyons, J. C. (). Circularity, reliability, and the cognitive penetrability of perception. Philosophical Issues : –. Lyons J. C. (in preparation). Cognitive processes for epistemologists. Marr, D. (). Vision. London: Allen and Unwin. McGrath, M. (). Siegel on the epistemic impact for epistemological internalism. Philosophical Studies (): –. McGurk, H., and Macdonald, J. (). Hearing lips and seeing voices. Nature : –. Prinz, J. J. (). Is the mind really modular? In R. J. Stainton (ed.), Contemporary Debates in Cognitive Science, –. Malden, Mass.: Blackwell. Proffitt, D. R. (). Embodied perception and the economy of action. Perspectives on Psychological Science : –. Pylyshyn, Z. W. (). Seeing and Visualizing: It’s Not What You Think. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press. Raftopoulos, A., and Müller, V. C. (). The phenomenal content of experience. Mind and Language : –. Siegel, S. (). The phenomenology of efficacy. Philosophical Topics : –. Siegel, S. (). Which properties are represented in perception? In T. S. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Siegel, S. (). The epistemic impact of etiology of experience. Philosophical Studies : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. c. lyons

Stokes, D., and Bergeron, V. (forthcoming) Modular architectures and informational encapsulation: a dilemma. MS. Strack, F., Martin, L., and Stepper, S. (). Inhibiting and facilitating conditions of the human smile: a nonobtrusive test of the facial feedback hypothesis. Journal of Personality and Social Psychology : –. Whitham, E. M., Fitzgibbon, S. P., Lewis, T. W., Pope, K. J., De Los Angeles, D., Clark, C. R., Lillie, P., Hardy, A., Gandevia, S. C., and Willoughby, J. O. (). Visual experiences during paralysis. Frontiers in Human Neuroscience : –. Witt, J. K. (). Action’s effect on perception. Current Directions in Psychological Science : –. Witt, J. K., Proffitt, D. R., and Epstein, W. (). Perceiving distance: a role of effort and intent. Perception : –. Woods, A. J., Philbeck, J. W., and Danoff, J. V. (). The various perceptions of distance: an alternative view of how effort affects distance judgments. Journal of Experimental Psychology: Human Perception and Performance (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Perceptual Integration, Modularity, and Cognitive Penetration Daniel C. Burnston and Jonathan Cohen

Imagine a computational system with a proprietary . . . database. Imagine that this device operates to map its characteristic inputs onto its characteristic outputs . . . and that, in the course of doing so, its informational resources are restricted to what its proprietary database contains. That is, the system is ‘encapsulated’ with respect to information that is not in its database. . . . That’s what I mean by a module. In my view, it’s informational encapsulation, however achieved, that’s at the heart of modularity. (Fodor : )

On the best evidence, there seems to be significant interaction between information sources, at all levels of perception: for better or worse, the picture of perception as a collection of independent, non-interacting, exclusively bottom-up feature extractors must be replaced by a picture of perception that more centrally incorporates informational interaction. The question we want to raise here is what impact this lesson should have for our thinking about claims concerning the modularity and the cognitive penetrability of perception. Many authors have taken the news that perception involves substantial informational integration to require seriously weakening or abandoning ideas about modularity and impenetrability. We believe this is an overreaction. For, we claim, an integrative conception of perception is compatible with robust forms of both modularity and impenetrability. That said, we believe that standard conceptions of the criteria for individuating modules sit uneasily with the integrative view of perception—that the standard view involves associating processes with univocal informational functions, and then looking for exchange of information (or the lack of it) between distinct processes, in a way that is hard to reconcile with the integrative view. Adopting the latter will require some rethinking of our standards for drawing modular boundaries. This paper is entirely collaborative; authors are listed in alphabetical order.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

In what follows we’ll be offering just such a reconception of the notion of modularity in terms of what Fodor () calls ‘isotropy’. We claim that the isotropybased criterion both comports better with the integrative view and better reflects the underlying purposes for which the notion of modularity was introduced, so is well worth adopting. Moreover, consideration of the isotropy-based criterion has significant implications for standard understandings of cognitive penetration, on which that notion is interdefined with the failure of modularity. For the isotropy-based understanding advocated here allows that non-modularity and cognitive penetration can dissociate in important and interesting ways, and suggests that standard ways of arguing from one to the other are illegitimate. Here is our plan. We’ll begin by reviewing, briefly, the case for an integrative conception of perception (Section ). Next, we’ll turn to the literature on modularity, and argue that both foes and ostensible friends of modularity have overestimated the threat posed to the notion by facts about informational integration (Section ). We’ll then pose a dilemma about the formulation of modularity in an integrativist framework, and suggest abandoning some of the standard criteria for modularity in favor of an alternative based on the notion of isotropy (Section ). Then we’ll show how the resulting, integrativist-friendly reformulation of modularity applies in particular cases, how it can be used to make fruitful distinctions with respect to both perception and cognition, and what this tells us about cognitive penetrability and perception (Section ). Finally, we’ll conclude (Section ). We hope that our discussion will bring out more clearly what is at stake in debates about perception, modularity, and cognitive penetration, and will offer a new and improved way of thinking about these topics in light of the lessons of perceptual integration.

 The Unencapsulation of Perception . A (too) simple view Consider the following logically possible, coherent, cartoon conception of perception, which we can call ‘the non-integrative view’.1 On this view, perception consists of a series of separate, dedicated, and informationally encapsulated modules, each devoted to the extraction of a single feature of the distal layout from the impinging energy array. Perhaps there is, for example, a shape extractor that produces as output a representation with the content that the object of perception is square; a color extractor that produces as output a representation with the content that the object of perception is green; a motion extractor that produces as output a representation with the content 1 We emphasize that the view we are sketching here is a caricature that should not be identified with the views of any actual thinker (though of course there may be points of overlap). We offer this deliberately overdrawn cartoon in order to locate its vulnerabilities easily (Section .) and motivate a more sophisticated alternative (Section .).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



that the object of perception is moving along a particular path with a particular velocity; and so on. Once these extractors have done their jobs, they hand off their output to a higher-level module that parses the scene into visual objects on the basis of information from the feature extractors about how features are distributed throughout the scene.2 The non-integrativist’s stipulation that the feature extractors are mutually informationally encapsulated is the stipulation that each does its computational work independently—that the inputs to the computations carried out by the shape extractor do not include the outputs of the color extractor (as it might be), nor vice versa. The view does allow for informational/computational dependence of some higher-level stages of perceptual representation—say, for object representation—on the output of the feature extractors. But, crucially, the non-integrativist views this dependence as asymmetric: the story would be that object representations are computed from representations about the exemplification of features, but not vice versa. Besides being an attractively simple story about the general architecture of perception, the non-integrative view has the virtue of smoothing the way to an extremely simple and direct verdict on the question of whether perception is cognitively penetrated. For, because the view is constructed so as to guarantee that individual perceptual faculties are encapsulated from all other representations, it follows immediately that they are encapsulated from cognitive representations in particular. Moreover, as we have recounted elsewhere (Burnston and Cohen ), the non-integrative view has played an important role in thinking about perception in philosophy, computational psychology, and neuroscience. Given all these merits of the non-integrative view, it is worth asking: is it true?

. Perceptual interaction, within and between modalities Alas, it turns out, it is not. There is abundant and converging evidence, drawn from many levels of organization in many different perceptual systems, that the architecture of perceptual computation is significantly more interactive than the non-integrative view allows, both within and between sensory modalities.3

2 Here and below we use the term ‘object’ in a thin sense which picks out (roughly) bounded and connected things that trace out continuous paths in spacetime, but that needn’t retain their kind affiliations to count as persisting over time. Minimally, we want it to be true of objects that they can be distal, that they can bear features that perceptual systems can be sensitive to, that they can serve as the loci for binding of feature representations, and that they can be served up by perception qua the referents of demonstrative thought and language on at least some occasions. Beyond that, we invite readers to rely on their own preferred understanding of what thin objects amount to. 3 A further reason for dissatisfaction with the non-integrative view revolves around its insistence that the psychological representation of objects is always posterior to the psychological representation of features. As we have argued elsewhere (Burnston and Cohen ), there are reasons to think that perception can represent the presence and location of objects in ways that outstrip the evidence drawn from its representations of exemplified features; and there are cases in which it seems that the perceptual representation of features depends on the perceptual representations of objects, contrary to the nonintegrativist’s preferred direction of representational dependency. We put these complications to one side in the following.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

A first kind of evidence of intramodal interactions comes from Field et al. (), who show that rod photoreceptors, which ordinarily function under scotopic conditions, provide input to blue/yellow opponent pathways in the macaque retina. A related idea, suggested (speculatively) by Broackes (), is that human deuteranopes employ dynamic input from their intact S-cone receptors to extract color information that is ordinarily subserved in trichromats by the output of L-cones or M-cones. If correct, these proposals reveal a level of unencapsulation at the photoreceptor level: they suggest that blue/yellow extractors are not informationally encapsulated from rod scotopic output, and that red/green extractors are not encapsulated from S-cone output.4 Nor do interactions within color vision stop with retinal processing: a variety of interactive effects occur in the process of assigning colors to surfaces, and in the computation of color constancies (see Cohen forthcoming). Once again, if there is this sort of integration, this tells against the view that there is any single informational pathway for the extraction of surface color. Further examples come from a variety of perceptual illusions, which, when manipulated experimentally, can tease apart inputs that affect perceptual processing of scene elements. A particularly revealing set of examples comes from the study of apparent motion.5 A variety of results show that apparent motion is affected by both luminance and contrast, and moreover that the speed of motion, its direction, and motionrelated object perception are all affected, depending on the perceptual situation. For instance, Anstis (b; ) demonstrates that the speed of a square patch moving at a constant velocity against a black/white grated background seems to change as the leading edge contrast changes. When the contrast is low, the square appears to slow down; when contrast is increased, it appears to speed up. When a square or diamond is moved diagonally across a grated or plaid surface, the different speed perceptions produced by the different contrasts of its sides (i.e. against the white or the black background) modify its perceived direction, so that even an object moving on a straight trajectory will appear to swerve back and forth. Similarly, in the ‘crossover illusion’, subjects are shown four rectangular shapes in the shape of a larger rectangle, where the top rectangles are ordered white/black and the bottom are ordered 4 One might object that counting these effects (and other cases discussed in this section) as encapsulation failures is too liberal, and that, on the resulting, highly permissive view, every input to a process (even its canonical ones) will count as encapsulation failures for that process. This worry is well taken: we do not wish to endorse an excessively rigid construal of the canonical input domains for particular processes, relative to which encapsulation will turn out to be trivially impossible. In a way, however, this is just the point we are making with this and our other examples: the non-integrative view of Section . itself sets overrigid boundaries for particular processes, with the result that there turn out to be any number of influences coming from outside them. Thus, our point is that these cases are counterexamples to unencapsulation given the way of drawing boundaries accepted by the non-integrativist. It is compatible with this point that we should decide not to accept the non-integrativist’s way of associating domains and inputs at the end of the day. (Thanks to Athanassios Raftopoulos for urging the need for clarification on this point.) 5 Similar conclusions can be drawn from illusions involving e.g. luminance and color, or from findings concerning illusory contours.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



black/white (or vice versa). Each rectangle then repeatedly switches with the one horizontally adjacent to it. In this setup, subjects perceive one (black or white) object moving diagonally; but, crucially, which color the object is seen as exemplifying is determined by which has the higher contrast compared to the background. Further, in these illusions both contrast and overall luminance can affect the strength of the effect (measured by time until breakdown; Anstis a). Overall, these studies suggest that motion perception draws on temporal sequences of both form and luminance information.6 Of course, this is just to deny that there is any single, informationally encapsulated pathway for the extraction of motion. Multimodal interaction effects are also common at all perceptual levels. Thus, there is neuroscientific evidence of multimodal effects at both the subcortical (Stein and Stanford ) and primary sensory cortical (Kayser and Logothetis ) levels, as well as in areas of temporal cortex implicated in categorical perception (James et al. ).7 Moreover, ongoing investigation into such well-known effects as the flash-lag illusion (MacKay ; Nijhawan ) and the ventriloquist illusion (Pick et al. ; Vroomen and de Gelder ; ) have shown not only that information from visual and auditory modalities can interpenetrate, but that such intermodal interactions can have reliable and positive effects on perceptual learning (Shams et al. ). Similarly, the cutaneous rabbit illusion (Geldard and Sherrick ) and other cases of sensory substitution (Bach y Rita et al. ) are standardly taken to show interaction between tactile and visual pathways that would preclude their mutual encapsulation.8 In short, while there remain many unanswered questions about the nature of these interactions, the prevalence and range of interactive effects within and across modalities is hard to square with the non-integrativist’s picture of independent and mutually informationally encapsulated pathways for the extraction of individual features.

. An integrative alternative Given the sorts of findings discussed in Section ., we need a way of thinking of perception as much more extensively integrated and interactive than the nonintegrative view would allow. In other work (Burnston and Cohen ), we have offered such a conception— henceforth, the integrative view—and argued that it is supported by a wide range of broadly empirical considerations. In outline, the integrative view represents the perceptual system as recovering structure about the distal world from an initially 6 This lesson is reinforced by related studies of visual form detection, which have shown that subjects can accurately compare (without attentional switching) several different kinds of form definition, including motion-defined, luminance-defined, and cyclopean (stereoscopic) form, even on opposite sides of irrelevant distracter forms (Kohly and Regan ). 7 While there is no consensus on the correct interpretation of these effects, they put significant pressure on the traditional view of unimodal cortices that are combined in higher-level ‘convergence’ or ‘association’ areas (Mesulam ; Meyer and Damasio ). 8 For discussion of such multimodal interactions and their general significance for philosophy of perception, see O’Callaghan ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

unstructured transduced signal by deploying a variety of techniques, or ‘perceptual strategies’, each operating on a characteristic range of information, and each delivering a working model of the arrangement of the particular features and objects that fall within its domain. As the process continues, perception integrates the different working models to produce a net perceptual representation as a kind of package deal.9 Among its other benefits, and as its name suggests, the integrative view is especially well suited to accommodate the sorts of perceptual integration described in Section .. Thus, for example, the integrative view easily accommodates the apparent fact that the assignment of colors to surfaces (etc.) involves integrating the output of a range of different computational techniques, each of which can be thought of as a perceptual strategy that delivers a working model from a canonical range of inputs, and whose output working model must be integrated with those of whatever other perceptual strategies have outputs that speak to the description of distal color. Similarly, the integrative view makes sense of the possibility of interaction between perceptual strategies directed at distinct features within or between modalities.10 Of course, on the integrative view, perception must at some stage somehow combine the outputs of multiple perceptual strategies, possibly operating over distinct spatiotemporal scales. Presumably the goal of such integrations is to arrive at a value that is consistent with (or as consistent as possible with) the range of evidence supplied by the different perceptual strategies at work. And different strategies may deliver inconsistent verdicts (particularly, but not exclusively, in laboratory settings designed specifically to pull them apart). When this happens, the perceptual system must have some way of managing the conflicts. And there is evidence of exactly the sorts of conflict management strategies one might expect. Thus, for example, sometimes perception selectively prefers one strategy over others (as in our reaction to the Ames room, where perspective information about depth dominates over our expectations about the relative size of objects). In other cases perception combines the information from distinct perceptual strategies in some more complex way, by sending one or more of the strategies back to run on the input again in the hope of obtaining new answers that won’t conflict, or by giving up (as e.g. we do with impossible Escher figures). The observation of such effects further supports the view that ‘the’ perceptual representation of the world is in fact an interaction effect arising from the integration of multiple perceptual strategies. 9 Caveat: Though, in what follows, we present the integrative view as an alternative general architecture for perception, we don’t want to commit to the claim that it is a correct description of every process occurring in perception. While we find the evidence supporting it to be wide-ranging and compelling, it is obviously an empirical question, not to be prejudged from the armchair, just how much of perception can be understood in terms of perceptual strategies and their integration. 10 Indeed, the view also accommodates the idea that the input to a perceptual strategy may be at least partly endogenous, and may encode ‘internalized regularities’ about the environments in which they are employed. On this general explanatory strategy, which is now fairly ubiquitous in theorizing about perception, see Shepard (); for representative instances, see e.g. Ullman (); Lotto and Purves (); Purves et al. ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



 Modularity Redux? If the non-integrative view were the only interesting theoretical motivation for the position that aspects of perception are impenetrable/modular, then learning that the former view is false would make the position (at least as classically construed) look pretty hopeless. And, as it happens, many thinkers have endorsed versions of this reasoning (though the responses they have proposed to it have differed). We believe these conclusions are too quick, and that there remain interesting senses in which impenetrability/modularity remain very much on the table. However, before we can spell this out (Sections  and ), we want to urge that both foes and friends of modularity have misconstrued the lessons integration holds for the notion. On one side, several foes of modularity (particularly those who have focused on the details of cognitive and neural processing) have taken the evidence of interactions between cognitive/perceptual processes to show that modularity is a lost cause. Thus, after discussing both neural plasticity and the sharing of information between neural systems, Buller and Hardcastle (: ) conclude that ‘there is little sense in which we can maintain that even our most basic processes are distinct in the way that [the] modularity hypothesis implies’. Similarly, Prinz () cites a wide range of interactive effects in perceptual and cognitive processes (including some of the multi-modal effects we mentioned in Section .) and concludes that virtually none of cognition— including early perception—is modular. On the other side, ostensible friends of modularity (in particular, defenders of massive modularity or evolutionary psychology more broadly speaking) have often taken evidence of the kinds of integration discussed in Section  as reason for weakening or rejecting encapsulation as a criterion of modularity.11 Effectively, these theorists accept with the anti-modularists the idea that the evidence of informational integration refutes modularity qua classically conceived, and go on to ‘save’ modularity by replacing it with something weaker. Thus, for example, Coltheart () abandons encapsulation in favor of a loose notion of domain specificity. Sperber () hopes to save modularity by distinguishing senses of domain-specificity relevant for understanding function from those that might be affected by interaction. Carruthers () makes a similar move in offering a ‘weak’ sense of encapsulation that can maintain functional specificity despite interaction effects from other processes. And Barrett and Kurzban () distinguish between the information a module has ‘access’ to and the information in processes.12 We suggest that both of these reactions are unjustified.

11 Alternatively, such theorists have sometimes attempted to save modularity by playing down the importance of integration in (early) perception, and thereby have characterized perception in terms much closer to the feature extractors we caricatured in §. than to the complex and integrative strategies we propose in §. (Barrett and Kurzban ; Samuels ). 12 Indeed, some have held that giving defining characteristics of modularity at all obscures the very point of functional analysis of mental processing (Barrett and Kurzban ; Sperber ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

Consider the pro-modularists first. One reason for dissatisfaction with the defenses they have proposed for modularity is that the weakened notions of domain specificity and weak encapsulation on which they rest have so far remained somewhat loose and metaphorical; indeed, we are sympathetic with the complaint that they may lapse into triviality (Samuels ). Second, and more importantly in our view, weakened notions of modularity sacrifice what we take to have been one of the important hallmarks of Fodor’s original account of the notion—viz. the advantages modularity carries for purposes of implementing a physically/computationally tractable theory of cognition. Fodor () famously argued that we could understand how cognitive processing might possibly work only to the extent that we could implement it in modular terms. Whether or not Fodor’s overall pessimism is justified, the conditions he places on modularity strike us as a potentially useful tool in our collective attempts to understand cognition and perception. Correlatively, giving up these conditions, even to the point of abstracting away from what is known about actual processing altogether, threatens to drain much of the interest from the notion of modularity itself.13 As against the anti-modularists, we suggest that interactions between different types of information within perception do not inevitably conflict with modularity. Crucially, and as we have pointed out elsewhere (Burnston and Cohen ), the very notion of modularity makes sense only relative to a basic functional and informational vocabulary specifying what counts as inside and outside a module. Only relative to such a specification of boundaries does it make sense to think of a particular sort of information exchange as a violation of modularity. Now, if you specify boundaries in a vocabulary tied to individual features, then featural integration in perception (e.g. of the sorts reviewed in Section .) will look like boundary-crossing, hence like modularity violation. But that vocabulary for drawing the boundaries is not mandatory. Indeed, given the extensive evidence of featural integration in perception, we suggest that the perceptual strategies implicated in perceptual processing make for better, and more empirically motivated, candidates for the kinds in terms of which our theories should be cast. But relative to a (in our view, better motivated) vocabulary that locates modular boundaries in terms of perceptual strategies, instances of informational integration in perceptual processing need not count as crossing a modular boundary. For if we can think of each instance of informational integration as occurring within a single perceptual strategy, we can count it as occurring within the boundaries of a single module.14 13 A further concern is that several of these weakened modularity views rely on a not-obviously principled distinction between the information proprietary to a cognitive/neural process and the information that is secondary to it. Proponents have sometimes attempted to meet this worry by appeal to evolutionary teleology or neurophysiological evidence. Suffice to say that it is possible to doubt whether these sorts of evidence can sustain the burden of compelling controversial views of cognitive architecture (Young et al. ; Buller ). 14 Viewed from this perspective, the appeal to integrative effects as evidence against modularity is a strange argumentative strategy; for it accepts the basic classificatory vocabulary of a subset of modular views (e.g. dedicated feature extraction views) only to deny it by citing an overwhelming abundance of evidence against it.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



This shows that, on our preferred way of thinking of modularity, the evidence of informational integration is at least possibly consistent with the claim that perceptual systems are modular. That said, it doesn’t follow immediately from what we have said that modularity is true: this will depend on the (empirical) questions of whether and to what extent perceptual systems operate on stereotyped inputs and perform stereotyped operations. If this is right, then what evidence constitutes support for or against modularity needs to be rethought in light of our conception of basic perceptual strategies.

 Modularity for Integrativists: The Isotropy Criterion Even if we are right that an integrative view about perception and modularity is possible, this doesn’t show that such a view is unavoidable. Deciding the answer to that question requires a criterion for deciding whether particular perceptual processes/strategies count as modular. Unfortunately, we now want to suggest, the standard criterion for determining whether a perceptual process is modular—pitched in terms of integration across informational boundaries—cannot be straightforwardly employed by an integrativist. On the standard ‘boundary-crossing’ criterion, employed in virtually all discussions of modularity in the literature, one assesses the putative modularity of perceptual systems in two sequentially ordered steps. First, one draws a functional line around a system in terms of the (unique) feature it purportedly extracts. And second, one asks whether there are genuine informational incursions across that functionally/ featurally defined boundary. If so, then the system is non-modular; if not, then it is modular. Alas, use of this criterion sits quite uneasily with the integrative view of perception. For, on the integrativist view, discovering perceptual strategies means delineating the particular informational interactions that underlie a perceptual process, and describing the computational principles such integrations follow. Given an integrativist conception of perception, it doesn’t make sense to attempt to circumscribe some set of features as the genuinely, functionally distinguished, set of inputs, and then regard any use of other inputs as crossing a modular boundary. Rather, the integrationist regards featural boundary crossing as part-and-parcel of the standard functional operation of perceptual processes. This means that, contrary to the sequence of steps in the standard criterion, integrativists want to draw the functional boundaries after discovering interactions, not before. On the other hand, if we draw the functional boundaries in a way that makes room for such interactions, as the integrativist prefers, then appeal to the boundary-crossing criterion for modularity threatens to render every process trivially modular. For, the worry goes, there’s no amount of interaction that will require giving up modularity— any interactions at all, in any amounts, can instead be counted as taking place inside

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

the (interactive) modular functional boundaries of the perceptual strategies to which they contribute.15 These considerations suggest a dilemma for integrativists. Horn one: if we define boundaries prior to detailed investigation of the interactive strategies perception employs, then the discovery of informational interactions will result inevitably in the denial of modularity. Horn two: if we wait until discovering all of the interactions, and then draw the boundaries so as to include relevant informational interactions, then modularity is inevitable. Both horns of this dilemma seem uncomfortable, since both trivialize the question of modularity. What’s an integrativist to do? Assuming we do not wish to give up the integrative view, the only method we can see for avoiding the unpalatable options reviewed is to abandon the standard boundary-crossing criterion for modularity. In its place, we propose to return to one of the key ideas Fodor () uses to frame the issue of modularity: the distinction between isotropic and anisotropic processes. The motivating idea here is that what makes modular processes modular, detachable, and in some sense separable from the rest of mentation is that they interface with other aspects of mental processing in a circumscribed number of ways. That is, modular processes are modular just because, and insofar as, there is a delimited range of parameters to which their processing is sensitive. Fodor initially motivates and elaborates the notion of isotropy by the example/analogy of ordinary and scientific rational belief fixation (though he clearly intends to be offering a more widely applicable criterion; cf. Fodor, : ). What qualifies rational belief fixation as isotropic, Fodor claims, is that there is no circumscribed set of inputs that exhausts the set that is of possible relevance for the updating and revision of any particular belief in the cognitive system, so you have to be prepared to look at information stored anywhere else in the system when fixing belief. Thus, scientific belief fixation is isotropic because ‘everything that the scientist knows is, in principle, relevant to determining what else he ought to believe. In principle, our botany constrains our astronomy, if only we could think of ways to make them connect’ (Fodor : ). Generalizing beyond the case of belief fixation, the idea we’re proposing is that what qualifies a process as modular is its not being 15 A related objection comes from the worry that the integrativist-friendly individuation procedure will result in an unconstrained, merely disjunctive characterization of mental processes in terms of literally everything they do—one that will render arbitrary processes trivially modular (Prinz : ). We agree that the procedure in question would trivialize modularity if it depended on utterly unconstrained, merely disjunctive characterizations of mental processes. But our contention is that the integrativist’s interactionincluding characterizations are not unconstrained—that they are, instead, the kinds implicated in the best scientific descriptions of perception. It is worth bearing in mind that whether a particular characterization of a kind is disjunctive depends on the primitive vocabulary. Though round or red is disjunctive relative to a vocabulary of primitives such as round and red, it is also true that round is disjunctive relative to a vocabulary of primitives such as round-and-red and round-and-triangular. As such, just how seriously we take the threat of disjunctivity depends on the vocabulary in terms of which we carry out our theorizing about perception. What the evidence in Section  suggests is that the integrativist’s description of perceptual strategies is not disjunctive relative to the kinds enshrined in our best theoretical descriptions of perception.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



isotropic—viz. that the range of input parameters to which its processing is sensitive is delimitable.16 Of course, in asking whether the range of input to which the processing of some mental process R is delimitable, we are always implicitly relying on some abstract (typically functional) characterization of R. Thus, Fodor’s claim that botanical beliefs potentially constrain the confirmation of astronomical theory only makes sense relative to the assumption that there is a (presumably functionally defined) mental process of theory confirmation, and that it draws on some range of input (e.g. when it is applied to an astronomical theory)—in this case beliefs. Thus, when we say that a process’s inputs are delimitable (hence, that the process is non-isotropic and modular), we mean that, relative to some typing of its proprietary states (such as beliefs in the confirmation case), there are no bounds on either (i) the subtypes of those states that can be relevant or (ii) the ways in which those subtypes can relate to each other, or both. We want to suggest that this idea—henceforth, ‘the isotropy criterion’—provides a potentially fruitful way of thinking about modularity (and the related notion of cognitive penetration). But before we go on to apply the criterion more widely and draw out its implications, we pause to note that it agrees with standard verdicts about the modularity or otherwise of paradigm cases.17 Thus, for example, the proponent of the isotropy criterion can agree with the standard verdict that belief fixation is nonmodular because any subtype of belief (astronomical, botanical, anatomical, etc.) can influence any other, and there is no a priori constraint on what form this influence takes. Likewise for analogical reasoning: in principle, considerations from literally anywhere in the conceptual system could be highly relevant to the reliability of analogically transferring conclusions from a first to a second domain, and they might be so in any number of ways. Assuming this is right, the isotropy criterion will classify the processes underlying analogical reasoning as isotropic, hence not modular. On the other hand, the isotropy criterion also sustains the standard verdict that the perceptual process leading to the Müller–Lyer illusion is modular: the illusion is the result of a process that integrates over lines (not e.g. color or motion), and only lines in certain arrangements (lines on opposite sides of the page, or disconnected, do not produce it). To the extent, then, that the isotropy criterion agrees with such standard verdicts about whether these paradigm processes are modular or not, it is to that extent not obviously on the wrong track. 16 As Fodor stresses, the fact that a non-delimitable output means that any set of information could in principle be relevant for a particular process does not show, in point of fact, that it is occurrently/actually relevant for that process. Otherwise astronomy would be even more complex than it is now. This will be important for our discussion of cognitive penetrability (Section ), where we will emphasize that showing that a system is isotropic—i.e. non-modular—by itself entails nothing about whether particular types of information actually play a role in its processing. 17 We don’t intend to commit to the correctness of these standard verdicts about these cases. Our purpose in appealing to the standard verdicts is only to suggest that the isotropy criterion is not extensionally inadequate in any obvious way.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

A more controversial example to which we can apply our proposed criterion comes from neuroscientific and psychological investigations into action control and prospective memory. These studies are uncovering a system, housed in dorsolateral regions of the prefrontal cortex, that appears to be in charge of representing higher-order relations (e.g. spatiotemporal relations, nesting relations) between different types of perceptual information in the context of action planning and control (Koechlin et al. ; Uithol et al. ). While much about the representational capacities of this system remains poorly understood/debatable, we want to use it to motivate a general possibility of which the system might turn out to be an instance. Namely, it is at least possible, given what is now known, that this system should turn out to be unrestricted in the range of perceptual parameters to which its processing is sensitive, and that these parameters can be organized in any sort of way relative to each other, dependent on the action-situation being represented. (Indeed, this possibility makes sense given the possibly infinite set of perceptual objects and features towards which we can direct and organize actions.) If so, then the isotropy criterion would classify this system as non-modular. We can contrast this case straightaway with the system discussed in Section ., which integrates contrast, luminance, and object information, underlying the perception of apparent motion. This earlier system seems to operate precisely over these specific inputs, in specific arrangements, and its functioning seems sensitive to these particular sorts of information (and not others). In this case, as in others we’ve listed, perceptual scientists have usefully attempted to delimit just the types of information that are integrated, and the computational principles via which the integration proceeds. That they are able to do so (when they are) enables us to describe the operation of the whole in terms of modular perceptual strategies, as per Section .. This contrasts markedly with the case of the action planning system: here, if we are right that there is no delimitation of the types of information relevant to the processing of the system, an analogous modular description is not on the cards. Whether this is so or not is, of course, an empirical issue—and one we take to be very much up in the air. What we want to emphasize, however, is that the possibility described by characterizing a perceptual system as non-modular is utterly coherent, so not to be dismissed by definition/stipulation.18

18 Given this way of thinking about modularity, it is natural to individuate modules by the (by definition, delimited) range of parameters to which their processing is sensitive. Moreover, if what we say about perceptual integration is true (Section ), modular systems will overlap—possibly significantly—in their associated ranges of parameters. It follows from these observations that there are likely to be a large number of modules in perceptual systems. For if module M interacts with or otherwise supplies input to modules M and M , then (some subset of) the parameters figuring in the range associated with M will also figure in the range associated with M and M ; hence it would appear that the range associated with M occurs as a part of not one but many module-individuative feature ranges. That said, we also think that there are explanatory contexts in which it is useful to partition these modules more coarsely, so it’s not the case that all of the many modules to which a particular set of parameters contribute are explanatorily relevant in every setting.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



The isotropy framework is meant to open conceptual space for a productive, empirically relevant account of modularity. Many questions, including how to individuate and count modules, and how to define relationships between them, remain to be addressed. The question we want to face next, however, is what, if any, implications this unconventional framework for thinking about modularity has for issues about cognitive penetration. This will be the topic of Section .

 Higher Perception, Modularity, and Cognitive Penetrability One important ramification of thinking about the modular/non-modular distinction in terms of isotropy, as we propose, is that that distinction comes apart from the cognitive/non-cognitive distinction. This is relevant because virtually all contemporary discussions of cognitive penetration (by both proponents and opponents) of which we’re aware have turned on supposing that these distinctions coincide, so that one can move freely between conclusions about modularity and conclusions about cognitive penetration. The literature about cognitive penetration has to date largely concerned the assessment of whether there is or is not crossing of a modular boundary (viz. a boundary demarcating perception) by ostensibly cognitive factors in a handful of cases. For example, Macpherson () appeals to cases of interactions between shape and color—specifically the result that heart shapes are perceived as redder than nonheart shapes of similar size—to argue for cognitive penetration of color experience. And Stokes () argues from the effect of increased size judgments for coins vs. similarly shaped disks that size perception is mediated by value judgments. In each of their cases, subjects are asked to match the stimulus to a comparator—the background in the color case, or another disk in the shape case—that can be adjusted manually by the subject until the test stimulus and comparator look the same along the relevant dimension. Crucially, these researchers take the effects in question to turn on categorical judgments that, according to them, can only be regarded as having been carried out by cognitive rather than perceptual processes: A more pressing concern is that, given the extent of perceptual integration, it could turn out that what are intuitively two different modules (say, one canonically associated with vision and another canonically associated with audition) end up being associated with the very same range of parameters; and if so, then it would appear that our view naturally leads to the conflation of intuitively distinct modules. We are not sure that there are genuine cases of this sort. But supposing there are, we suggest that our view can be saved by appealing to some further condition on the individuation of modules. That is, we can say that modules are individuated in terms of both the range of parameters to which their processing is sensitive and something else. And there are several different something elses available. For example, one might require that modules be associated with a canonical piece of (neuro)anatomy, a particular algorithm operating over the inputs, a characteristic form/degree of neural connectivity, etc. As far as we can tell, any of these further conditions would allow us to escape the conflation worry under consideration. We will ignore these niceties in what follows. (Thanks to Matthew Fulkerson for discussion of these matters.)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

While it may be plausible to think that the visual system is sensitive to heart shapes and autonomously responds to that shape making it seem more red than it is, it is far less plausible to think that the visual system can classify certain objects as being apples and autonomously alter the apparent colour of such objects. The reason is that classifying something as an apple doesn’t depend just on simple features, such as shape, that the visual system is responsive to, for many things have the shape of apples, such as cherries and other fruits. And while the visual system can respond to shape, it surely can’t classify an object as an apple, as it is not sensitive to the features that are required to do so. It is surely only the cognitive system that is capable of doing such classificatory work. (Macpherson : –).

On the other side, those who deny the existence of cognitive penetration typically attempt to undermine alleged cases of penetration by arguing that the latter do not involve the crossing of a modular boundary—e.g. by arguing that, in such cases of apparent unencapsulation, the relevant effects of cognition are due either to attentional processes that operate before perceptual modules are engaged or to post-perceptual judgments that operate after perceptual modules have finished (e.g. Pylyshyn ).19 Whatever one’s view about the success of these defenses, the point to notice is that these theorists are accepting the inference from a lack of modularity to cognitive penetration: they are agreeing with their opponents that if there were a crossing of a modular boundary, then that would be sufficient to demonstrate that perception is penetrated by cognition. Indeed, it is precisely because they accept this inferential pattern that their rejection of cognitive penetration depends on finding ways to deny that a modular boundary is crossed in the cases under discussion. Thus, while these theorists disagree about whether there are genuine cases of nonmodularity, all parties to the dispute accept the goodness of the inference from nonmodularity to cognitive penetration, and so confine their disagreement to the issue of whether certain cases qualify. In what follows we will suggest that the crucial inference underlying the dispute—and accepted by theorists on both sides—looks significantly less attractive once we think about modularity in the terms we have been urging. Before we come to this, however, we want to register a more internal concern. We think there are reasons to doubt that the classificatory status of the output of a representational process, invoked explicitly by Macpherson in the passage quoted above, is a good criterion for that state’s being cognitive (/cognitively penetrated). Though there are no uncontroversial notions for what counts as classificatory, Macpherson’s discussion seems to assume that for something to be classificatory, it must be categoryspecific, and must go beyond particular feature representations. But if that is right,

19 For what it is worth, Macpherson and Stokes address and argue against the possibility of defending impenetrability by these strategies in the cases they discuss. They argue against the attentional/preperceptual explanation on the grounds that, in the experiments cited, subjects have as much time as they need to adjust the comparator, hence have ample time for a wide variety of shifts in attention. They argue against the judgment/post-perceptual explanation on the grounds that this story would result in a deep inconsistency between the perceptual and judgmental interpretations if the affected vs. unaffected stimuli were perceived as the same. We take no official stand on the success of these arguments.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



then notice that, given the integrative view, a process could be classificatory even if it is entirely perceptual.20 To see why, consider investigations into the phenomenon of perceived ‘chasing’ (Gao et al. ; ; Gao and Scholl ) which have shown that certain patterns of movement between objects, even in the presence of many distracters, reliably elicit the representation that a second object is chasing a first. These investigations reveal that verdicts about chasing, no less than verdicts about applehood, seem to depend on more than just the simple features to which vision is responsive. Thus, for instance, if objects are not ‘facing’ their direction of movement (e.g. if a triangle’s lowest-degree angle is oriented away from the direction in which it moves), this inhibits the classification of the case as an instance of chasing. More generally speaking, it appears that the effect only occurs for figures whose shapes, as well as motions, stand in particular configural relations to one another—it is not just representations of features that are important, but instead features integrated in specific (but, given the integrative view, nonetheless perceptual) ways. Therefore, despite their being apparently classificatory in the intended sense, it is by no means obvious that representations of chasing (/the processes underlying that sort of representation) are cognitive. Indeed, while it is too early to endorse this conclusion with full confidence, initial research suggests strongly that the representation of chasing can be accounted for in terms of representations of shapes and motions (and configural relations among these) that are ordinarily counted as wholly perceptual, as opposed to beliefs or other canonically cognitive states. (For what it is worth, Gao and Scholl (: ) themselves advocate understanding the representation of chasing in perceptual rather than cognitive terms.) If we are right that representations of chasing might be examples of classificatory representations that are not themselves cognitive or cognitively mediated, then this would defeat the idea that any perceptual states sensitive to classificatory effects are ipso facto evidence of informational exchange across a modular boundary. In turn, this would mean that, even granting the boundary-crossing conception of modularity we rejected (Section ), cases of perceptual states sensitive to classificatory effects can’t be assumed without further argument to represent modularity violations or, therefore, grounds for concluding that cognition penetrates perception.

20 The version of the challenge to which we respond in the main text, taken from Macpherson’s description just quoted, revolves around the idea that the subject’s response is specifically classificatory. But there is a distinct, related formulation of the challenge grounded in the idea that the subject’s response is categorical. Besides wanting to follow Macpherson’s formulation of the worry, we prefer to carry out the discussion in terms of classification rather than categoricity because the latter tends to carry stronger conceptual/cognitive associations, and we are precisely attempting to leave it up for grabs whether and to what extent concepts/cognition play a role in generating the responses at issue. That said, note that our response to the case, which involves the example of chasing (see below), plausibly extends to the version of the challenge formulated in terms of categoricity as well. For, though one might initially have described subjects’ representations of whether one object is chasing another as both classificatory and categorical, our suggestion is that, on the evidence, these representations may be capable of being understood in noncognitive terms.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

Having registered this internal concern, we now want to raise a further, and less internal, point suggested by the case of representations of chasing: the processing underlying such classificatory/high-level representations might well turn out to be anisotropic. For, while, once again, it is still early days, the initial evidence suggests strongly that the inputs for the processes leading to classificatory verdicts about chasing are limited to specific sorts of perceptual information standing in specific types of relations. If so, this would mean that the processes leading to classificatory verdicts about chasing are delimited and specifiable in our sense. Therefore, on the isotropy construal urged in Section , these processes will plausibly turn out to be modular despite producing output representations that are no less classificatory than those considered by Macpherson and Stokes. Nor is chasing unique in being both plausibly anisotropic and ostensibly classificatory/high-level: investigators have proposed that the perception of certain social cues (Langton et al. ) and biological motion (Johansson ; Troje and Westhoff ) may admit of similarly modular interpretations (though, once again, we don’t mean to hold our conceptual point hostage to these verdicts about the particular phenomena). What this suggests, then, is that the classificatory nature of a representation is not a reliable guide to its being either cognitively penetrated or modular. Indeed, these considerations put us in a position to make the even stronger and more direct suggestion that (whatever one thinks about classificatory representations) at least some stereotypically cognitive processes might rely on a significantly delimitable range of input parameters—and hence, again, assuming our revisionary construal, might turn out to be modular. We suggest that simple arithmetic, for example, might be a case of this sort: it might well turn out that mental arithmetic draws on a delimited suite of psychological capacities, though these capacities and the associated contents on which they operate will plausibly be classified as conceptual rather than perceptual (for an overview of research in this area, see Ashcraft ). This gives reason for thinking that, on an isotropy-based reconstrual of modularity, the standard inference from modularity to cognitive impenetrability breaks down. Nor, given our preferred construal of modularity, is there reason to expect that non-modular (viz. isotropic) perceptual processing must be sensitive to the cognitive system. Thus, as we suggested in the case of prospective memory and action planning (Section ), it is least possible (and perhaps actual) that systems involved in the organization of perceptual information might draw on a non-delimitable range of inputs despite not drawing on inputs from cognition.21 This conclusion appears to comport well with recent work on perceptual category learning, which suggests that 21 One way in which one might wish to challenge our description of this system is by claiming that it is in fact cognitively penetrated, so not an instance of non-modularity without cognitive penetration after all (see e.g. Wu  for an analogous argument that the system for visual spatial constancy is cognitively penetrated by intentions). While we cannot hope to resolve the issue about prospective memory in particular (or about Wu’s case of visual spatial constancy) here, we stand by our claim in the main text: construed our way, the non-modularity of a system is not by itself a guarantee of its cognitive penetration.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



the processing underlying such learning becomes increasingly sensitive to integrative (/‘contextual’) cues—cues involving not just the presence of isolated parts of the stimulus, but the configural relations holding between parts—as subjects become more expert (Gauthier and Tarr ). Given the large (perhaps unlimited) number of objects about which we can form perceptual expertise, it is likely that this system won’t have delimitable inputs. However, as we saw with prospective memory, this point offers no reason to believe that the processing of configural cues underlying the resulting categorical representations must reflect input from cognition in particular. Hence there is no good inference here from the isotropy of the system to its having been penetrated by cognition. That is, there is no good reason to suppose, simply from the fact that such processes operate over a non-delimitable range of inputs, and that the outcome of learning over these inputs implements categorical effects, that they are sensitive to cognitive representations in particular.22 Now, the point we’re urging requires some care: for, while what we have said shows that there is one understanding of cognitive penetration on which the inference from isotropy to penetration fails, there is a neighboring reading of cognitive penetration on which that inference withstands the considerations we have brought to bear. Specifically, we have suggested that an isotropic process can fail to draw on cognitive input, so that such processes can fail to be cognitively penetrated as a matter of actual, historical fact. On the other hand, the isotropy of such processes means that we cannot foreclose, ex ante, the possibility that they might draw on—and that their operation would be affected by drawing on—any particular range of input, including, cognition. Even if our considerations suggest that isotropic processes might fail to be (actually) cognitively penetrated, they don’t show that such processes might fail to be cognitively penetrable. Even with this concession, however, we take our considerations to be relevant to the kind of reasoning on display in many recent discussions (e.g. Macpherson ; Stokes ), which attempts to infer from the absence of modularity to the conclusion that perception is actually cognitively penetrated (and thence, presumably, to the corollary conclusion that perception is cognitively penetrable, by application of a trivial modal axiom).23 22 Here we are assuming that there is a criterion of the cognitive—a standard that would settle whether a class of representations that affects some output is cognitive or not. For example, on the standard assumption that explicitly believed content is cognitively represented, one might test for the influence of specifically cognitive representations by determining whether perceptual processes are sensitive to (some degree of) relevant change in explicit belief. Thus, if the size or color judgments in the cases discussed could be mediated by previous belief change (e.g. being told that the coin was a counterfeit), this would support the view that the interactions in question are cognitive. These are only suggestions: we take no official stand here about the nature of the cognitive (or about the appropriate experimental tests for cognitive status). Rather, our central point is that the question of whether a class of representations is cognitive should proceed independently of the question of whether a particular process affected by members of that class is modular or not. 23 In fairness, Macpherson and Stokes explicitly argue for cognitive penetration from the premise that informational encapsulation fails, not from the premise that modularity fails. But since these authors interdefine the notions of modularity and informational encapsulation, this is a distinction without a difference.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

To be clear, our contention is not that there is no penetration by cognition going on in these cases; for we take the ultimate extent, type, and import of the relations between perception and cognition in such cases to be open questions. Rather, our point is that once modularity is construed in terms of isotropy, the proffered conclusion does not follow from the premises provided: the isotropy of a process, per se is no guarantee of its (actual) penetration by cognition.

 Conclusion We have argued that standard discussions of cognitive penetrability, perception, and modularity rest on an insufficient appreciation of the extent and role of integration within perception. But, unlike some who would take such considerations as reason for rejecting modularity altogether, we take them instead to motivate a more refined, isotropy-based conception of modularity, as opposed to the usual boundary-crossing criterion. Crucially, however, adopting the newer criterion for modularity means recognizing that the modularity/non-modularity distinction comes apart from the cognitive/noncognitive distinction, and that they together partition mental processes into four kinds rather than two. It allows for processes that are non-modular and cognitively penetrated (e.g. rational belief fixation); processes that are non-modular and not cognitively penetrated (e.g. prospective memory); processes that are modular and cognitively penetrated (e.g. mental arithmetic); and processes that are modular and not cognitively penetrated (the representation of chasing). We think modifying our understanding of modularity in this way, and accepting the resulting taxonomic revisions, carries significant benefits. It offers gains in clarity, allows us to recognize the importance of integration within perception without requiring us to give up useful notions of modularity, and leads to a more expansive view of the possible ways in which mental/perceptual states can interact. Our view leaves open important questions about how modular and non-modular systems interact within and across the perceptual/cognitive divide, just how that divide should be construed once the integrative picture is accepted, and what conditions distinguish modular cognitive systems from modular perceptual systems. But we believe that it represents a promising framework for pursuing these questions and, more generally, for thinking about modularity, cognitive penetration, and mental architecture.24

24 Thanks to Matthew Fulkerson, Brian Keane, Casey O’Callaghan, and the editors of this volume, for helpful discussions of this material that significantly improved the paper.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration 

References Anstis, S. M. (a). Levels of motion perception. In L. Harris and M. Jenkin (eds), Levels of Perception, –. New York: Springer. Anstis, S. M. (b). Moving objects appear to slow down at low contrasts. Neural Networks : –. Anstis, S. M. (). Factors affecting footsteps: contrast can change the apparent speed, amplitude and direction of motion. Vision Research : –. Ashcraft, M. H. (). Cognitive arithmetic: a review of data and theory. Cognition : –. Bach y Rita, P., Collins, C. C., Saunders, F., White, B., and Scadden, L. (). Vision substitution by tactile image projection. Nature : –. Barrett, H. C., and Kurzban, R. (). Modularity in cognition: framing the debate. Psychological Review (): –. Broackes, J. (). What do the color-blind see? In J. Cohen and M. Matthen (eds), Color Ontology and Color Science, –. Cambridge, Mass.: MIT Press. Buller, D. J. (). Adapting Minds: Evolutionary Psychology and the Persistent Quest for Human Nature. Cambridge, Mass.: MIT Press. Buller, D. J., and Hardcastle, V. G. (). Evolutionary psychology, meet developmental neurobiology: against promiscuous modularity. Brain and Mind (): –. Burnston, D. C. and Cohen, J. (). Perception of features and perception of objects. Croatian Journal of Philosophy : –. Carruthers, P. (). The Architecture of the Mind: Massive Modularity and the Flexibility of Thought. Oxford: Oxford University Press. Cohen, J. (forthcoming). Perceptual constancy. In M. Matthen (ed.), Oxford Handbook of Philosophy of Perception. Oxford: Oxford University Press. Coltheart, M. (). Modularity and cognition. Trends in Cognitive Sciences (): –. Field, G. D., Greschner, M., Gauthier, J. L., Rangel, C., Shlens, J., Sher, A., Marshak, D. W., Litke, A. M., and Chichilnisky, E. J. (). High-sensitivity rod photoreceptor input to the blueyellow color opponent pathway in macaque retina. Nature Neuroscience : –. Fodor, J. A. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, J. A. (). The Mind Doesn’t Work That Way: The Scope and Limits of Computational Psychology. Cambridge, Mass.: MIT Press. Gao, T, and Scholl, B. J. (). Chasing vs. stalking: interrupting the perception of animacy. Journal of Experimental Psychology: Human Perception and Performance : –. Gao, T., McCarthy, G., and Scholl, B. J. (). The wolfpack effect: perception of animacy irresistibly influences interactive behavior. Psychological Science : –. Gao, T., Newman, G. E., and Scholl, B. J. (). The psychophysics of chasing: a case study in the perception of animacy. Cognitive Psychology : –. Gauthier, I., and Tarr, M. J. (). Unraveling mechanisms for expert object recognition: bridging brain activity and behavior. Journal of Experimental Psychology: Human Perception and Performance (): –. Geldard, F. A., and Sherrick, C. E. (). The cutaneous ‘rabbit’: a perceptual illusion. Science (): –. James, T. W., Humphrey, G. K., Gati, J. S., Servos, P., Menon, R. S., and Goodale, M. A. (). Haptic study of three-dimensional objects activates extrastriate visual areas. Neuropsychologia (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



d. c. burnston and j. cohen

Johansson, G. (). Visual perception of biological motion and a model for its analysis. Perception & Psychophysics (): –. Kayser, C. and Logothetis, N. K. (). Do early sensory cortices integrate cross-modal information? Brain Structure and Function (): –. Koechlin, E., Ody, C., and Kouneiher, F. (). The architecture of cognitive control in the human prefrontal cortex. Science (): –. Kohly, R. P., and Regan, D. (). Fast long-range interactions in the early processing of luminance-defined form. Vision Research (): –. Langton, S. R. H., Watt, R. J., and Bruce, V. (). Do the eyes have it? Cues to the direction of social attention. Trends in Cognitive Sciences (): –. Lotto, R. B., and Purves, D. (). An empirical explanation of the Chubb illusion. Journal of Cognitive Neuroscience (): –. MacKay, D. (). Perceptual stability of a stroboscopically lit visual field containing selfluminous objects. Nature : –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. Mesulam, M. M. (). From sensation to cognition. Brain (): –. Meyer, K., and Damasio, A. (). Convergence and divergence in a neural architecture for recognition and memory. Trends in Neurosciences (): –. Nijhawan, R. (). Motion extrapolation in catching. Nature : –. O’Callaghan, C. (). Perception and multimodality. In E. Margolis, R. Samuels, and S. Stich (eds), The Oxford Handbook of Philosophy and Cognitive Science, ch. . Oxford: Oxford University Press. Pick, Jr, H. L., Warren, D. H., and Hay, J. C. (). Sensory conflict in judgements of spatial direction. Attention, Perception, & Psychophysics (): –. Prinz, J. J. (). Is the mind really modular? In R. Stainton (ed.), Contemporary Debates in Cognitive Science, –. Oxford: Blackwell. Purves, D., Shimpi, A., and Lotto, R. B. (). An empirical explanation of the Cornsweet effect. Journal of Neuroscience (): –. Pylyshyn, Z. W. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences (): –. Samuels, R. (). Is the human mind massively modular? In R. Stainton (ed.), Contemporary Debates in Cognitive Science, –. Oxford: Blackwell. Shams, L., Wozny, D. R., Kim, R., and Seitz, A. (). Influences of multisensory experience on subsequent unisensory processing. Frontiers in Psychology (). doi: ./fpsyg... Shepard, R. N. (). Perceptual-cognitive universals as reflections of the world. Behavioral and Brain Sciences : –. Sperber, D. (). Modularity and relevance. In P. Carruthers, S. Laurence, and S. Stich (eds), The Innate Mind: Structure and Contents, vol. , –. Oxford: Oxford University Press. Stein, B. E., and Stanford, T. R. (). Multisensory integration: current issues from the perspective of the single neuron. Nature Reviews Neuroscience : –. Stokes, D. (). Perceiving and desiring: a new look at the cognitive penetrability of experience. Philosophical Studies (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perceptual integration



Troje, N. F., and Westhoff, C. (). The inversion effect in biological motion perception: evidence for a ‘life detector’? Current Biology : –. Uithol, S., Burnston, D., and Haselager, P. (). Why we may not find intentions in the brain. Neuropsychologia : –. Ullman, S. (). The Interpretation of Visual Motion. Cambridge, Mass.: MIT Press. Vroomen, J., and de Gelder, B. (). Sound enhances visual perception: cross-modal effects of auditory organization on vision. Journal of Experimental Psychology: Human Perception and Performance (): –. Vroomen, J. and de Gelder, B. (). Ventriloquism and the freezing phenomenon. In G. A. Calvert, C. Spence, and B. E. Stein (eds), The Handbook of Multisensory Processes, –. Cambridge, Mass.: MIT Press. Wu, W. (). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies (): –. Young, M. P., Hilgetag, C. C., and Scannell, J. W. (). On imputing function to structure from the behavioural effects of brain lesions. Philosophical transactions of the Royal Society of London Series B, Biological Sciences (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Multisensory Perception and Cognitive Penetration The Unity Assumption, Thirty Years After Ophelia Deroy

 Introduction Most of our perception of the world does not come from a single sense: in most instances, our ears, nose, eyes, and skin are jointly stimulated, as are our proprioceptive, interoceptive, and vestibular systems. Importantly, these multiple streams of sensory information can be combined to lead to the perception of a single property, event, or object. As you grasp a golf ball from the ground, your perception of its shape uses both tactile and visual information. When you watch a tennis game, your perception of the ball bouncing on the ground combines both visual and auditory information. When you listen to a friend, your perception of her speech obviously relies on the auditory information, but also on the visual information you get from her lip movements. ‘Multisensory integration’ is the term used to cover these cases and many others where the joint stimulation of different sensory modalities contributes to the perception of the same entity.1 There is growing evidence that cases of multisensory integration are frequent in everyday life.2 As such, they should occupy a central place in our theories of perception. This is, however, rarely the case. Philosophical problems are often raised and discussed as if perception is distributed into distinct unisensory episodes, with vision as a paradigmatic case. This is still true of the question of cognitive penetration. Most of the examples discussed in the present volume ask whether the colours, distances, or 1 We certainly need to be cautious when using this term, as some cases of joint stimulation might deserve to be re-described not as cases of real integration, but as cases where sensory modalities lead to distinct perceptions of what only externally happens to be the same entity, or as cases where one modality merely influence the perception of an object in the other modality (see Deroy ; Deroy et al. ; O’Callaghan ). For the present purpose, it is sufficient to consider that at least some cases qualify as cases of multisensory integration. 2 See e.g. O’Callaghan () for recent illustrations.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  shapes we see are influenced by what we know or believe, and few wonder whether the shapes we touch and see, or the events we see and hear, are equally sensitive to these influences. The main goal of the present chapter is to correct for this absence—though before doing so, some preliminary justifications are needed. The first thing to recognize is that there might be reasons why multisensory cases are missing from debates about cognitive penetration, and why there is a focus on unisensory cases in philosophy. As the question of cognitive penetration is closely linked to modularist models of perception, and as multisensory integration is frequently seen as lying ‘beyond modularity’,3 it seems that there is no reason to apply the question to multisensory processing: why worry about a non-modular process satisfying a key criterion of modularity like cognitive impenetrability? Addressing this point is necessary before one can introduce multisensory cases in the debate. Arguably, and as discussed in several chapters in this volume, considering whether a certain perceptual process is cognitively penetrable can be somewhat detached from the question of the kind of modularity it presents. Several criteria of modularity coexist, and modularity can be seen as a question of degree, and not as all or nothing (see Deroy ). If there are reasons to consider that multisensory integration processes are modular to a certain extent, it is legitimate to ask whether they are influenced by cognitive states. Making a question legitimate is still not sufficient to make it important. The importance of the question of cognitive penetration in multisensory cases needs to be established from the presence of likely candidates for such top-down influences on multisensory integration. There are also good empirical reasons to wonder whether multisensory integration operates under certain higher cognitive assumptions. Isn’t the decision to integrate information from different senses into a single percept influenced by the perceiver’s representations of which feature goes with which? And aren’t such representations what we mean by concepts? Take the ventriloquist’s puppet and the fact that spatial information provided by vision tends to influence the information provided by audition: isn’t it because you know that speech sounds are coming from moving lips that you hear the sounds as coming from the animated dummy rather than from the still lips of the puppeteer? Consider also what happens at the movies. Whereas you perceive the sounds as coming from the lips of the characters you see on the screen, the loudspeakers are not located behind the screen but rather behind or around you (if you were introduced into a movie theatre with your eyes closed, this effect would certainly not occur and the sounds would be heard as coming from their real sources). In both cases, auditory and visual information combine to give rise to the integrated perception of an audiovisual (speech) event, occurring in a single place. The result of audiovisual integration observed in the ventriloquist case and at the movies is certainly in accordance with the belief that speech sounds are supposed to come from moving lips. But does it mean 3

See the title of a famous paper by Driver and Spence ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

that this belief plays a role in what you perceive? Could it be causally responsible for the resulting perception? Beliefs about sound sources and what they look like is certainly not necessary for the integration of sounds and visual objects or events. Newborns, cats, and adults seem to routinely integrate synchronous sounds and visual events even when they have no beliefs about one being the cause of the other, or them going together in any way. Beliefs are also not sufficient to explain why we are prone to perceive speech sounds as coming from moving lips: the auditory and visual stimuli must also fit certain temporal and/or spatial constraints for them to be integrated.4 So if beliefs are neither necessary nor sufficient to explain audiovisual integration, what role do they play? Because of the importance of bottom-up factors like time and/or space in multisensory interactions, we need to go to borderline cases to find instances where cognitive influences could play a role. To see this, one needs to turn to cases where neither perfect synchrony nor perfect spatial coincidence between stimuli occur (as should be the case in natural situations), but where auditory and visual stimuli are presented at slightly different times or locations in space than what is naturally the case. Despite a lack of true temporal or spatial coherence, experimental studies suggest that there is a ‘window’ during which these cues can be integrated and therefore perceived as synchronous and co-localized. Although spatial and/or temporal proximity are primary constraints and sufficient in many typical cases, there are circumstances where they do not determine in and of themselves whether two cues should be integrated or not. Since the s, several psychologists have suggested that the limits of the window of integration (i.e. the decision to integrate cues that are separated by a certain distance in time or space) are affected by pre-existing rules or representations about which sensory cues go or don’t go together, i.e. how congruent they are. Expressions such as ‘semantic effects’, ‘prior knowledge’, or ‘assumptions of unity’ are then used to refer to the idea that the degree of congruence between the presented cues can affect their probability of being integrated. To anticipate an example which will be discussed later, two random events, like a brief noise and a flash of light, are less likely to be integrated than congruent sounds and visual events—like speaking sounds and moving lips, or whistling sounds and steaming kettles. The fact that audiovisual integration can occur or not depending on the congruence between sounds and visual events seems difficult to explain, unless this congruence is somewhat represented by the system. There must be a representation or background assumption that speech sounds and visual lip movements, or whistling sounds and steaming kettles, go together or have the same source. Whether this assumption requires having a concept of speech and kettles, or beliefs about what a speaking human or a kettle looks and sounds like, is a way of asking whether audiovisual integration is cognitively penetrated. This, then, is 4 Just how necessary spatial coincidence is to integration is a matter of controversy, as discussed in Spence (). It is still an influential factor in many cases.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  what this chapter will ultimately address, once the relevance of framing the problem in terms of cognitive penetration has been established.

 Scaling Up the Problem of Cognitive Penetrability The issue of cognitive penetration is strongly connected to the architectural models introduced by Fodor () and Pylyshyn ().5 The introduction of these models was supposed to address a fundamental question: how do perceptual systems relate to the rest of our cognitive systems? Are they like a collection of autonomous subsystems, or are they dependent on so-called ‘higher’ cognitive states, notably on what we believe or expect to perceive?6 As the question of the dependence of perception on higher levels of cognition was framed as a question about modularity and cognitive penetrability, it has also encouraged a certain model of perception based on the two following assumptions: (i) Perceptual processing must be carried out by a plurality of non-overlapping sensory systems, which are functionally and neurologically distinct from one another. (ii) These various perceptual processes must take place at the same level in the mental architecture. Perceptual processing takes place after transducers have automatically converted incoming stimuli into signals, which can be used as inputs; its outputs are then fed to the central system, which compares the various representations delivered by the peripheral systems, and can form beliefs, desires, and other propositional states. The definition of modules, and these assumptions, have been mostly framed with the case of vision in mind, but they can be distinguished from this example, and more generally from the individuation of modules. Whether vision, for instance, constitutes a single module, as presented by Pylyshyn () and already proposed by Marr (), or whether the visual function can involve a collection of distinct modular systems (e.g. one for shape, one for colour, and one for motion), as argued by Zeki and Bartels (), is an open question, with several views lying between these two extreme options. The important part of the modularist claim is that sensory systems (whether they correspond to the common sense idea of a sensory modality or to more specific subsystems) are assumed to operate in parallel, at the same level in the hierarchy.

5 See the chapters by Lyons, Burnston and Cohen, and Briscoe (Chapters , , and ) in this volume; see also Deroy (in press) for a review of the initial and more recent debates surrounding modularity. 6 ‘Higher’ is a difficult metaphor, which is taken by some to mean ‘later in the chain of information processing’ and by others ‘higher in the cortical hierarchy’—with the two definitions not necessarily delivering the same verdict all the time. For the present purpose, it is sufficient to underscore that what philosophers count as propositional attitudes and conceptual representations are included in the category of ‘higher cognitive states’, but might not exhaust the category.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy M1

M5

M2

Central system

M3

M4

Figure . A shared assumption: modules correspond to non-overlapping encapsulated processing units, and their outputs feed directly into the central system

A good way to picture what these assumptions lead to is to represent perceptual processes like different branches of a star, all starting with transducers and ultimately leading to unified representations (see Figure .). With such a picture in mind, philosophers and psychologists interested in perception, and in cognitive penetration more specifically, tend to focus on one branch at a time, say on colour or visual shape perception. Whenever the question of cognitive penetration of perception is raised, it is raised for a specific output: modularist models will posit that this output be processed independently of anything else going on outside the modular branch, while their opponents suggest that higher cognitive states ‘feed into’ this branch. Take the case of colour perception, borrowed from Delk and Fillenbaum () and discussed by Deroy (), Macpherson (), and Siegel (). The modularist assumption which is challenged by proponents of cognitive penetration is that colour is processed independently of shape. If shape and colour were processed together, then the fact that different shapes of exactly the same reflectance property end up being perceived to have slightly different colours, in similar illumination conditions, would not be such a threat to modularity. A difference of shape input could explain the difference. Examples of cognitive penetration of size perception, discussed by Stokes (), suppose that size is processed independently of say, texture, colour, and weight. If size were processed alongside weight, colour, and texture, then the fact that coins and cardboard circles of exactly the same dimensions, but of different weight, colour, and textures, end up being perceived as slightly different in size would not be a threat to modularity. The difference in output could be explained by a difference in the set of inputs processed together within the module. In other words, the question of cognitive penetrability only makes sense when focusing on a single stream of sensory processing, supposed to operate in parallel to other sensory processing that could be

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  going on at the same time. To go back to the star model, the question of cognitive penetration arises mostly when one focuses on one branch at a time. This focus means that the debates about cognitive penetration remain restricted, and do not easily generate general conclusions about perception. From this way of dealing with one branch at a time, nothing follows (for instance) concerning what to think of other branches. It cannot be excluded (for instance) that the visual processing of shape is cognitively penetrable, while the tactile processing of shape is not. The idea that some processes could be modular and not others—what could be called ‘partial modularity’—contrasts with the tendency to consider that showing the modularity of some processes gives us good reason to accept that all the others are modular. More importantly, the investigation of cognitive penetrability on a branchby-branch basis simply leaves multisensory perception out of the picture. A large number of behavioural studies has revealed that sensory systems interact with one another at different levels. Perceptual processing is largely multisensory, which means that the kind of sensory information processed in a stimulated sensory stream is influenced by the sensory information processed in another (concurrently stimulated) stream7 . Noticeably, stimuli presented in different sensory modalities can provide complementary information about the same property or event and lead to elevated levels of performance which cannot be explained by simple probability summation. Both auditory and visual information are, for instance, used to lead to the perception of a certain object or event in space, in a way that is not just the sum of the visual information and the auditory information. Cases which result in the perception of a single multisensory property, event, or object count as multisensory integration. For instance, in the so-called ventriloquist effect, auditory spatial information is integrated with the spatial information conveyed synchronously by vision. The resulting experience can be described as hearing the sounds coming from a visual source (e.g. the dummy’s mouth) or hearing and seeing a unified speaking puppet.8 At first sight, any case in which one kind of sensory processing is influenced by another seems incompatible with the assumption that sensory systems do not overlap. Does the very existence of multisensory processing ask us to give up modularity? As discussed for instance by Burnston and Cohen (Chapter ), the assumption that sensory systems need not overlap or interact might not be essential to modularity, if modularity is defined first and foremost by cognitive impenetrability. It is possible to accept that sensory systems communicate with one another—what Burnston and Cohen call ‘horizontal penetration’—without seeing this as a threat for modularist models in general. Modularity with horizontal, crossmodal interactions still prevent ‘vertical penetration’—i.e. the influence of higher cognitive contents on interrelated perceptual processes. 7 8

In some cases, the influence can come from a non-stimulated stream (see Spence and Deroy ). See Spence and Bayne () and essays in Bennett and Hill () for abundant discussion.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

If the model of horizontal penetration can take into account crossmodal effects, it is still difficult to reconcile with multisensory integration. Crossmodal effects mean that the processing of one kind of sensory information is influenced by the processing of another sensory stream, and as such, they leave intact the idea that each one leads to its own output. The influence of audition on taste perception provides a good example. Tasting something sweet while hearing high-pitched sounds leads one to taste it as sweeter. This form of sweetness enhancement still leads to two distinct outputs, one auditory and one gustatory, with the latter being affected by audition. The processing of information in the auditory module modifies the output of the taste module, while still leading to a distinct auditory output. The idea of two horizontally penetrated modules still commits one to the existence of two modules with distinct outputs. It does not adequately account for processes of multisensory integration which are supposed to deliver a single output. Is it then impossible to fit multisensory integration within a modularist framework—even revised to authorize communications between sensory systems? One revision, independent from the previous proposal, is to modify the assumption of a strict parallelism between all kinds of sensory processing, therefore operating on a single level, and to authorize a hierarchy of sensory processing. Multisensory integration can take place after unisensory processing: the outputs of two or more unisensory streams can be used as inputs by another multisensory process later in the chain of information processing or in the cortical hierarchy.9 For instance, visual spatial information and auditory visual information can be processed separately and can then feed into a multisensory system which generates representations of audiovisual localization. More generally, it seems (at least theoretically) possible to systematically reinterpret the evidence of multisensory effects by attributing them to a second layer of processing where a first series of outputs are used as new inputs (see

(a)

(b)

Audiovisual module

Auditory module Auditory module Inputs from auditory transducer

Visual module Inputs from visual transducer

Inputs from auditory transducer

Inputs from visual transducer

Figure . Two ways in which multisensory audiovisual processing can be reconciled with modularity

9

This position is defended e.g. in Radeau (). See also Massaro ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  Figure .a). It then makes sense to wonder whether the new input–output system is modular, and therefore whether or not it is cognitively impenetrable. Another possibility to rescue the modular aspect of multisensory integration is to consider that one of the sensory systems at stake takes a wider range of inputs than we assumed in the first instance (see Figure .b). Instead of still thinking of visual processing and auditory processing as independent and interacting only if their outputs are used by a later system, it might also be the case that, under specific circumstances, some visual inputs are processed by the auditory system along with its primary auditory inputs. Again, it then makes sense to ask whether this input–output system is modular and cognitively impenetrable.

 Is Multisensory Integration Influenced by Higher Cognitive States? As seen above, recognizing that multisensory integration occurs does not in itself oblige one to move away from the traditional debates on modularity. If there is a hierarchy of uni- and then multisensory processing streams, or if certain processing might take a wider range of sensory inputs than initially thought, then it is legitimate to ask whether these newly defined multisensory processes are modular or not. It follows that it makes sense to ask whether these processes are cognitively penetrable. As mentioned in the introduction, the question is made all the more important by the evidence that multisensory integration is sensitive to ‘assumptions of unity’ or ‘prior knowledge’. While the majority of studies devoted to congruency effects have been published over the last couple of decades, some of the seminal experiments go as far back as the s. In a now classic early study conducted in the garage at the Oxford Psychology Department, Jackson () presented participants with a semicircular array of steaming kettles. These kettles did not emit any sound, but whistles were sounded by compressed air either at the same location as the steaming kettle or at another angle. Each time, participants were told that the whistling sounds would come from one of the steaming kettles, and were asked to indicate which one the sound was actually coming from. Participants’ responses showed a significant bias of sound localization towards the steaming kettle: in cases where the sound originated from a location other than the ‘silent’ steaming kettle but still close to it, there was a tendency to hear the sound as coming from this kettle. This, in a sense, was an expected ‘ventriloquist effect’. The interesting thing, though, was that the same task was conducted with bell sounds and shining lights—pairs of stimuli that the participants would not have usually encountered together—with less of a ventriloquist effect observed. Jackson concluded that audiovisual interactions were different depending on whether there were more or less grounds for supposing that the two stimuli originated from the same source in space.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

There are some problems with this early experimental setup, but its conclusions are supported by more recent studies. Using a variety of psychophysical and neurological techniques,10 several researchers have identified behavioural and neurological differences between cases where the pairs of audiovisual stimuli presented simultaneously correspond to either congruent or incongruent pairings. The selection of congruent pairs is most often done because the properties they code for are regularly associated in a single familiar object. The most natural, and indeed most used, description of the effect consists then in saying that pairs represented as, or known to belong to, a single object introduce a difference in multisensory perception. In their  review, Doerhmann and Naumer suggest that these effects of congruence deserve to be called ‘semantic effects’, as they must involve stored conceptual representations of objects. Earlier, Welch and Warren advanced the idea that multisensory integration (audiovisual again, in this case) involved an assumption of unity, which they defined as ‘the observers’ assumption that they are in the presence of a single distal object’: ‘The stronger the assumption of unity, the greater the tendency will be for perception to occur in a way consistent with a unitary event (i.e., with a normal stimulus situation)’ (Welch and Warren : ). The reasoning leading to this hypothesis, though, can be criticized. Welch and Warren start from the observation that ‘situations can vary from ones in which subjects hold a very strong assumption that what they see and what they feel, for example, are actually a single physical event (e.g. the finger) to ones in which this assumption is weak or even nonexistent’ (Welch and Warren : ). Based on the idea that two elements believed not to belong to the same object would not be integrated, or at least would not be likely to be integrated, Welch and Warren jump to the conclusion that believing that the two elements belong to the same object is a necessary condition for multisensory integration: If they (the subjects) strongly believe that the two sensory inputs are emanating from different physical events, no phenomenal intersensory discrepancy will prevail, thereby reducing or perhaps even eliminating the need for a perceptual resolution in the form of intersensory bias. In short, according to the current view, a necessary condition for the occurrence of intersensory bias is the registration, at some level, that the two discrepant sources of information are emanating from a single event. (Welch and Warren : ; emphasis added)

As integration is of course observed in the absence of belief, it does not follow that belief is necessary for integration. Welch and Warren were perhaps too quick to recognize the role of beliefs in multisensory processing, and might have reached for an extreme conclusion before considering alternative options.

10

See Doehrmann and Naumer () and Parise and Spence () for two recent examples.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception 

 The Problem of Parallel Explanations . Relying on contrast Many of the experiments which try to establish the presence of semantic effects on multisensory perception follow Jackson’s kettle study, and rely on a contrast between the presentation of non-congruent and congruent combination of features, in controlled spatial or temporal relations. The idea in these is to demonstrate that congruent cues—selected because they are characteristic of a familiar kind of object, like kettles—are bound differently than cues that are not congruent—for instance, random sounds and shapes. As detailed above, the assumption is that in the same conditions of spatial (or temporal) discrepancy, the visual shape of a silent kettle and a whistling sound are perceived as unified into a single object, whereas a mere visual flash and a noise are not. Most of the evidence in favour of cognitive penetration of perception relies on such contrasts, which are supposed to be only explicable through the influence of higher cognitive states on perceptual processing.11 The same sort of contrast is used in unisensory studies like Delk and Fillenbaum (), which compares the performance on a colour-matching task done on visual shapes which are conceptually associated to the colour red (like the shape of a heart) and on shapes which are not associated with red (like the shape of a mushroom). Susanna Siegel uses the presence of a contrast in her definition of cognitive penetrability: Cognitive Penetrability (second pass): If visual experience is cognitively penetrable, then it is nomologically possible for two subjects (or for one subject in different counterfactual circumstances, or at different times) to have visual experiences with different contents while seeing and attending to the same distal stimuli under the same external conditions, as a result of differences in other cognitive (including affective) states. (Siegel : )

Cognitive penetration is then used to explain a contrast between two perceptual contents, and does so by positing that these differences in perceptual content are caused by differences in cognitive states like beliefs or concepts. But how does a contrast establish that this causal relation holds? The empirical reasoning seems to go along the following lines: If identical stimuli, presented in identical conditions and attended in the same way give rise to different perceptual responses, and if the only relevant difference lies in higher-level cognitive contents, then the difference in responses can be attributed to cognitive penetration of perception. This reasoning underlies most of the empirical contrasts discussed in the literature.12 It crucially rests on two apparently opposing conditions. First, the stimulation must 11 12

See Macpherson () and Stokes (). See Deroy () for a discussion of unisensory cases.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

be identical in the two cases; second, there must be a relevant difference in higher cognitive contents between the two cases. How well are these conditions satisfied by the experimental protocols dealing with multisensory integration?

. Using differences in stimuli The identity condition for stimuli and conditions is often not strictly respected by contrast cases. In the kettle experiment, for instance, the auditory and visual stimuli are different: they consist here, of kettles and whistles, there, of flashes and brief sounds. Even if one controls for the duration of presentation, and abstract from the qualitative differences between stimuli, there is still a difference in the quantity of information delivered in the two series of cases. The difference of integration shown between congruent audiovisual stimuli and non-congruent ones could be due to differences in the richness of the stimuli presented—specifically, richness in temporal cues. It might be simpler to synchronize auditory and visual cues which present a large variation rather than more subtle ones of the same duration. As synchrony is not considered as a cognitive factor in multisensory integration,13 this explanation of the contrasted outputs suggests a purely bottom-up model, compatible with modularity. More generally, any method which contrasts congruent and incongruent stimuli introduces the possibility that the difference observed in the response comes from a difference at the level of inputs. It is actually difficult to imagine a case where nothing varies between the two contrasted series of cases, as differences in stimuli need to be introduced to trigger different kinds of higher cognitive states. Because of this limitation, the method of contrast makes it difficult to distinguish between two alternative explanations: E. Bottom-up explanation: The observed difference comes from a difference in the quantity of information provided in the two cases. E. Top-down explanation: The observed difference comes from stored information about congruent pairs of stimuli. Only the second option presents a threat of cognitive penetration of multisensory processing.

. Pushing higher cognitive contents too high Putting aside the first problem of inputs, a second problem arises from the method of contrast. Even if one grants that the contrast between outputs comes from a difference in stored information, this still does not show that this stored information is ‘cognitive’ in the sense that is needed to establish cognitive penetration. The method of contrast should not consider that all top-down factors accounting for the difference in perceptual content are necessarily cognitive in the sense that they can qualify for cognitive penetration. Even if the contrast between a pair of congruent stimuli and a 13

See Parise et al. (); Spence and Squire (); Vatakis and Spence ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  pair of incongruent stimuli is likely to make participants think differently about the two situations, this difference in higher cognitive states is not sufficient to implicate cognitive penetration. Most of the time, the method of contrasts can at best point to a correlation between the differences in higher cognitive states which are likely to be generated in the two situations, and the observed difference in the responses. This correlation is not sufficient to show that perception has been penetrated by higher cognitive states. For cognitive penetrability of perception to occur, the higher cognitive state must exert a causal influence on the processing of sensory information. Take the contrast used by Jackson in the kettle experiment. When an individual sees a kettle and hears a whistling sound, s/he probably happens to believe that kettles emit whistling sounds. There is a correlation between the increased tendency to perceive the whistling sounds and silent kettles as a single audiovisual entity, and the belief that kettles whistle. This correlation also does not yet show that audiovisual integration has been penetrated by this belief, or that this belief has in some sense causally contributed to the perception of a unified whistling kettle. At this stage, it is possible to recognize that experimental evidence often falls short of demonstrating more than a correlation. The inference from the fact that A and B are correlated to the fact that A caused B is often secured by the absence of another, better causal explanation. In the multisensory case, it is important to check whether there is no other systematic correlation, which would offer a competing explanation. This is certainly the case if differences in higher cognitive states generate changes in physiological arousal, which might in turn affect the processing of sensory information. Imagine that participants in Jackson’s experiment think about tea when they hear the whistling sound or/and see the kettle. As a result of this thought, they start thinking about scones, butter, jam, etc. and have a sudden secretion of insulin, which in turn influences audiovisual integration. Although the influence on audiovisual integration ultimately comes from the higher cognitive state triggered by the contrasted case, this is not a case where this higher state penetrates perception. Noticeably, the relation between the higher cognitive state and the result must be minimally sensitive to content, and not go through a deviant causal chain of either cognitive or noncognitive associations. As Pylyshyn states: If a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, that is, it can be altered in a way that bears some logical relation to what the person knows. (Pylyshyn : )

The positive clause which captures the exclusion of deviant chains relates to what Pylyshyn meant by ‘rational’ and ‘semantically coherent’. For cognitive penetration to occur, the influence on the processing of sensory information needs to come directly from a higher cognitive state whose content is relevant to the object of perception, and the processing needs to be sensitive to this content. Specifying what sensitivity means here would take us too far (see Deroy ). For the present purpose, the line of Potter Stewart on pornography might do: One can know that something satisfies

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

this semantic relevance and sensitivity when one sees it. In the present case, beliefs about sounds of kettles will do, but thoughts about scones and jam won’t. In most cases presented as cases of cognitive penetration, then, this kind of contentsensitivity is accepted, more than really demonstrated. What is evidenced is a mere compatibility between certain variations in perceptual responses (or experiences) and the reconstructed content of certain beliefs. The difference in audiovisual integration between flashes and beeps, and between kettle shapes and whistling sounds, is compatible with the fact that most people probably hold beliefs about the sounds of kettles while they hold no such beliefs for the sounds of flashing lights. It seems likely that, when presented with the two sets of stimuli, people will have the same kind of generic thoughts about kettles going with whistling sounds, and will not have thoughts about noises going with flashes. There is perhaps no way to control for differences in thoughts between participants. What needs to be shown is that there is no better theory than cognitive penetration by the same kind of thought granted to everyone to explain the statistically significant tendency to integrate whistling sounds and silent kettles more than noises and flashes. This is where attention must be given to alternative ways in which congruence can come to influence integration. Some non-conceptual representations of congruence, also known as ‘crossmodal correspondences’,14 can have effects on multisensory integration, and can be responsible for the contrast observed between certain combinations of stimuli. We know, for instance, that a crossmodal correspondence exists between higher-pitched sounds and brighter visual surfaces: when asked which sound goes best with a bright surface, most people consistently choose a higher pitch sound. What is important to notice, though, is that these correspondences are cognitively opaque, and are not taken to be justified or true. Before being asked, people would not volunteer to say that high-pitched sounds are brighter than low-pitched sounds, or go better with bright surfaces; once the forced response is given, they do not endorse it as being either justified or true. Evidence of the influence of these correspondences in infants and chimpanzees suggest that they do not rely on language.15 In this respect, crossmodal correspondences can underlie intuitive choices, such as in a forced-choice question, but subjects do not hold them as true. In this respect, they cannot count as contents of beliefs. Now, the correspondence between high pitch and bright is sufficient to explain effects of congruence similar to the kettle case. People have a greater tendency to integrate high-pitched sounds and bright visual objects than to integrate high-pitched sounds and darker objects. This effect of congruence corresponds to a top-down influence on multisensory integration, but it does not comes from the penetration of perceptual processing by the content of conceptual cognitive states, such as beliefs. Other kinds of non-conceptual correspondence could explain the privileged integration between congruent kettles and whistles, for instance the correspondence between rising pitch and rising visual movement. As shown in the musical case, and in the 14 15

See Deroy and Spence () and Spence (). See Ludwig et al. ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  linguistic expression itself, we tend to associate certain changes of pitch with directions of movement—with ‘rising’ pitch going up and ‘descending’ ones going down. These correspondences then explain why certain audiovisual combinations of sounds and movements can strike us as more congruent, when they relate in some way to this mapping between sounds and directions. The effect of congruence on multisensory integration observed in the case of the kettles might then also relate to some such non-conceptual, opaque, and non-rational representations of congruence, rather than deriving from our beliefs or knowledge about objects. Contrast cases such as the kettle one can therefore not distinguish between two alternative top-down influences for multisensory interactions: E(a). Top-down conceptual influence (cognitive penetration): The observed difference in perception between congruent and incongruent combinations of stimuli comes from conceptual representations about objects. E(b). Non-rational top-down influence: The observed difference in perception between congruent and incongruent combinations of stimuli comes from non-conceptual, non-rational representations of congruence between sensory features. These two interpretations coexist in the case of multisensory integration of whistles and kettles: Although we have beliefs about kettles and the sound they emit, it might be non-conceptual, non-rational associations which do the job. Again, the method of contrast is inconclusive.

Conclusions Cases of multisensory perception present an interesting challenge for claims of cognitive penetration of perception. Once the idea that they are not candidates for being modular is corrected, the question of their sensitivity to higher cognitive contents arise. Cases documented under the concept of ‘semantic effects’ are a good place to start, as they show that pairs of congruent stimuli (e.g. kettle shapes and whistling sounds) are not integrated in the same way as pairs of incongruent or non-congruent ones (e.g. random noises and flashes of light). Contrast cases, however, have a hard time distinguishing between three alternative explanations: E. Bottom-up explanation: The observed difference comes from a difference in the quantity of information provided in the two cases. E(a). Top-down conceptual influence (cognitive penetration): The observed difference in perception between congruent and incongruent combinations of stimuli comes from conceptual representations about objects. E(b). Non-rational top-down influence: The observed difference in perception between congruent and incongruent combinations of stimuli comes from non-conceptual, non-rational representations of congruence between sensory features.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

The main challenge, then, lies in finding a better way to establish that the influence on perception counts as cognitive penetration. I have argued here that one should distinguish between propositional or conceptual representations of congruence—which could count as cognitive penetrators in perception—and crossmodal correspondences—which do not threaten the cognitive impenetrability of perception and which represent non-rational and non-conceptual top-down influences on perception. This distinction raises an important challenge: what is the sign of a semantically coherent, rational influence which could distinguish between these two kinds of representations and ultimately between the explanations E(a) and E(b)? And by sign, I mean ‘empirical evidence’. After all, cognitive penetration must remain a testable claim about the mind, or it is only an idle question. A possible way to test the difference lies with the impact of brief training. A short change in exposure can temporarily affect our non-rational representations of congruence, but will not similarly affect our established beliefs about the world. To distinguish between the influence of our beliefs about audiovisual properties of kettles and the influence of a certain non-rational representation of audiovisual congruence, one could expose people to new combinations of sounds and shapes for a brief period of time. Beliefs that kettles whistle, for instance, should be maintained even if one is briefly presented with singing kettles in an experimental session. These beliefs are supported by other beliefs, or—granting that one has a certain degree of control over what one believes—the training in a laboratory setting will not be a reason to change beliefs about the world, where kettles whistle. By contrast, the presentation of singing kettles should temporarily affect non-rational representations of audiovisual congruency. A brief training will therefore provide the means to distinguish between beliefs and nonrational representations of congruence. If there is a difference in the audiovisual binding of singing sounds and steaming kettles similar to the one observed with whistling sounds and steaming kettles, then both effects could be attributed to non-rational representations of congruence, while if there is no difference in the audiovisual binding of singing sounds and steaming kettles, then the previous effect observed with whistling sounds and steaming kettles should be attributed to beliefs. I am not suggesting that this single experiment will crucially distinguish between the two options, and training experiments might present a blurred picture, where both effects occur. The point of suggesting an experimental design, however, was not to hope for a decision on alternative explanations, but to stress what I see as the real challenge, lying at the heart of the debate about cognitive penetration: not the distinction between perception and cognition, but differences between the effects of rational and non-rational representations.

References Bennett, D., and Hill, C. (eds) (in press). Sensory Integration and the Unity of Consciousness. Cambridge, Mass.: MIT Press. Delk, J. L., and Fillenbaum, S. (). Differences in perceived color as a function of characteristic color. American Journal of Psychology : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

multisensory perception  Deroy, O. (). Object-sensitivity versus cognitive penetrability of perception. Philosophical Studies : –. Deroy, O. (in press). Modularity of perception. In M. Matthen (ed.), Oxford Handbook of Philosophy of Perception. Oxford: Oxford University Press. Deroy, O. (). Multisensory integration and the many unities of consciousness. In D. Bennett and C. Hill (eds), Sensory Integration and the Unity of Consciousness, –. Cambridge, Mass.: MIT Press. Deroy, O., Chen, Y.-C., and Spence, C. (). Multisensory constraints on awareness. Philosophical Transactions of the Royal Society, Series B: Biological Sciences : . Deroy, O., and Spence, C. (). Why we are not all synesthetes (not even weakly so). Psychonomic Bulletin & Review : –. Doehrmann, O., and Naumer, M. J. (). Semantics and the multisensory brain: how meaning modulates processes of audio-visual integration. Brain Research : –. Driver, J., and Spence, C. (). Multisensory perception: beyond modularity and convergence. Current Biology : –. Fodor, J. (). Modularity of Mind: An Essay on Faculty Psychology. Cambridge, Mass.: MIT Press. Jackson, C. V. (). Visual factors in auditory localization. Quarterly Journal of Experimental Psychology : –. Ludwig, V. U., Adachi, I., and Matsuzawa, T. (). Visuoauditory mappings between high luminance and high pitch are shared by chimpanzees (Pan troglodytes) and humans. Proceedings of the National Academy of Sciences : –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research : –. Marr, D. (). Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. New York: Holt. Massaro, D. W. (). Perceiving Talking Faces: From Speech Perception to a Behavioral Principle. Cambridge, Mass: MIT Press. O’Callaghan, C. (). Perception and multimodality. In E. Margolis, R. Samuels, and S. Stich (eds), Oxford Handbook to Philosophy and Cognitive Science, ch. . Oxford: Oxford University Press. O’Callaghan, C. (). Intermodal binding awareness. In D. Bennett and C. Hill (eds), Multisensory Integration and the Unity of Consciousness, –. Cambridge, Mass.: MIT Press. Parise, C. V., and Spence, C. (). ‘When birds of a feather flock together’: synesthetic correspondences modulate audiovisual integration in non-synesthetes. PLoS One : e. Parise, C. V., Spence, C., and Ernst, M. (). When correlation implies causation in multisensory integration. Current Biology : –. Pylyshyn, Z. (). Computation and Cognition: Toward a Foundation for Cognitive Science. Cambridge, Mass.: MIT Press. Radeau. M. (). Auditory-visual spatial interaction and modularity. Current Psychology of Cognition : –. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Spence, C. (). Crossmodal correspondences: a tutorial review. Attention, Perception, & Psychophysics : –. Spence, C. (). Just how important is spatial coincidence to multisensory integration? Evaluating the spatial rule. Annals of the New York Academy of Sciences : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



o. deroy

Spence, C., and Bayne, T. (). Is consciousness multisensory? In D. Stokes, S. Biggs, and M. Matthen (eds), Perception and its Modalities, –. Oxford: Oxford University Press. Spence C., and Deroy, O. (). Crossmodal mental imagery. In S. Lacey and R. Lawson (eds), Multisensory Imagery: Theory and Applications, –. New York: Springer. Spence, C., and Squire, S. (). Multisensory integration: maintaining the perception of synchrony. Current Biology : –. Stokes, D. (). Perceiving and desiring: a new look at the cognitive penetrability of experience. Philosophical Studies : –. Vatakis, A., and Spence, C. (). Crossmodal binding: evaluating the ‘unity assumption’ using audiovisual speech stimuli. Perception & Psychophysics : –. Welch, R. B., and Warren, D. H. (). Immediate perceptual response to intersensory discrepancy. Psychological Bulletin : . Zeki, S., and Bartels, A. (). The autonomy of the visual systems and the modularity of conscious vision. Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

PA RT III

Substantive Impenetrability and Penetrability Claims

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Perception versus Conception The Goldilocks Test Fred Dretske

If one is going to quarrel about whether perception is modular, whether it is influenced by, say, cognitive factors, one needs to know what counts as perception. Since there are many ‘objects’ of perceptual verbs that are (often only implicitly) propositional in character, perception of such objects requires possession of the concepts deployed in that proposition and, hence, whatever knowledge it takes to have those concepts. One cannot see what is on the table—that it is an iPod, for instance—unless one knows what an iPod is. If one lacks this knowledge, one is ‘blind’ to what is on the table. One cannot see what it is. If seeing what is on the table in this sense counts as perception, then, clearly, cognition affects perception. Perception, some of it anyway, would be cognition. There would be nothing modular about it. Someone who lacks knowledge would have diminished perceptual powers. Despite good eyesight, they could not see what is directly in front of them.1 So the first step in trying to decide whether perception is modular, whether it is modifiable by cognitive affairs, is to carefully distinguish perception from cognition. This can be a tricky business. We can talk of an experience of an iPod in contrast to beliefs about or knowledge of the iPod. Or the way the iPod looks. Or we can invoke technical terms and speak of its ‘phenomenal appearance’. Such terms are generally intended to suggest a conscious stage in the processing of sensory information that is independent—logically independent—of those later cognitive or conceptual processes characteristic of knowledge, thought, judgment, and reasoning. But they only suggest it. They do not imply it. That is because these words themselves can be and often are used to describe mental states already contaminated by cognitive elements. If visual experience of an object is understood to comprise what the object looks like to you, what you see it as, or what it appears to you to be, then visual experience is infected with a cognitive virus from the get-go. With such an inclusive notion of experience, the 1 They could, of course, see it, the iPod, that which is in front of them. What they couldn’t see is what it is (that is in front of them).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. dretske

modularity of perception is no longer an open question. Perceptual experience is—by definition—not modular. To avoid this result one must focus, instead, on something we might call pure perception, what used to be called the sensation or (sometimes by psychologists) the percept. The target is the distinctively sensory aspect of a larger and more inclusive experience—something itself untainted by cognitive ingredients. We are looking, in short, for a stage2 in the processing of incoming information that, logically speaking, could occur in one whatever one’s thoughts, judgments, hopes, expectations, and values might be. We want perception of an iPod to be, in this respect, like stepping on an iPod, something a person can do whatever is going on in the rest of his mental life. But how do we isolate the sensory core, this cognitively unspoiled experience? Is there such a thing? Of course there is. It is something that sentient beings enjoy virtually every minute of their waking life. In what follows I offer a test, a criterion, a way of telling what should be included and—more importantly for present purposes— what should be excluded from this purified perceptual state. I call it the Goldilocks test. As a practical matter, Goldilocks is not always easy to apply. It is, for example, not clear how it is to be applied to gustatory and olfactory sensations. Even in the case of vision the test does not always yield reliable results. Not unless one makes unrealistic assumptions about the subjects being tested. I return to some of these shortcomings later. Nonetheless, even if never performed, Goldilocks is an effective analytical tool. It serves as a useful way of thinking about perception, an intuitive guide to the target—pure perception—whose causal isolation from cognitive affairs is at issue in debates about the modularity of perceptual processes. As a further benefit, Goldilocks also poses a challenge to those hoping to demonstrate the permeability of perception. To describe the test, and to illustrate its application, I will use an example of alleged cognitive penetration conveniently provided by Susanna Siegel in her recent book, The Contents of Visual Experience.3 Siegel argues that an expert’s perception can be affected by his or her knowledge in a dramatic way. A novice sees a pine tree and experiences merely a arrangement of variously colored shapes and textures. An expert forester, on the other hand, someone who effortlessly recognizes pine trees when he sees them, has quite a different experience. He experiences not simply the colors and shapes the novice sees, but also something the novice does not see: the property pine-treeness. The forester’s visual phenomenology is different. His beliefs and expectations, his reactions and dispositions are different, of course. He sees (the fact) that it is a pine tree. He sees, and thus knows, what kind of tree it is. He sees it as a pine tree. It looks

2 A conscious stage of course. We are not concerned, for instance, with peripheral activity in the receptors or the optic nerve. 3 I introduced Goldilocks in my commentary on Siegel’s book in an ‘Author Meets Critics’ session at the Pacific Division Meetings of the American Philosophical Association in Seattle, Wash., April .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perception versus conception



like a pine tree to him. He sees it to be a pine tree. None of this is true of the novice. But aside from these cognitive differences, there is also, according to Siegel, a difference in the forester’s visual experience of the tree, a difference in how the tree looks to him in a sense of ‘looks’ that indicates a genuine difference in visual phenomenology. This is an astonishing claim about cognitive penetration: someone coming to see a new property, a kind property, pine-tree-ness, as a result of acquired knowledge and experience of instances of that kind. I do not believe it. I think pine trees look much the same to experts as they do to novices. My skepticism is nourished by the failure of such alleged visual differences to pass the Goldilocks test. Before illustrating this, though, a word of caution. Quite obviously, what one knows, prefers, expects, and wants can causally affect one’s perceptual experiences. It can do so by affecting where one looks and therefore what one sees. That is to say, it affects experiences by changing the stimulus. No one defending the modularity of perception is denying these obvious facts. What is in question, instead, is whether such cognitive differences can alter the experience without changing the stimulus. Can they affect the processing of a stimulus—and thus the product of that process, the experience itself— without doing so merely by affecting the stimulus on which that process operates? That is the issue. I will therefore assume in what follows that a comparison of experiences that allegedly exhibit cognitive influence is a comparison of experiences that are responses to exactly the same physical stimulus. The expert forester and the novice are shown the same tree from the same angle in the same light for an interval so brief that no eye movements occur. Or, if we suppose the tree is seen long enough for eye movements to occur, I will assume that eye movements (including saccades) are exactly the same. Despite the implausibility of this ever occurring in real life, the assumption is required and completely harmless in the present context because the question being asked is not whether knowledge can affect which stimuli are processed (obviously it can), but whether it can affect how a stimulus is processed (controversial). So I assume that experiences allegedly exhibiting cognitive penetration, experiences that are allegedly different as a result of this influence, are responses to exactly the same stimulus. Using Siegel’s example, then, Goldilocks goes like this. E is an expert forester. N is a woodsy novice. If E’s visual experience is in some way enhanced, enriched, or merely altered by his special cognitive/conceptual condition with respect to pine trees, if the pine tree they both see looks different to E, this should become apparent in their pictorial representations of the tree. So we ask N to paint what he sees.4 E is asked to judge the painting. E can have one of three different reactions: Goldilocks #: too little; Goldilocks #: too much; and Goldilocks #: just right. Only Goldilocks #, I submit, is a credible response. 4 Assume for the sake of the thought experiment, the imagined test, that N has the artistic talents of a th-c. Dutch master.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. dretske

Goldilocks #: Too little. E complains that N has left something out of his painting: the pine-treeness of the tree.5 E experiences, he sees, pine-tree-ness when he looks at the tree, but he doesn’t see it when he looks at N’s painting. He doesn’t see it, of course, because N didn’t put it in his painting, and N didn’t put it there because he didn’t see it when he looked at the tree. According to Siegel, though, for E it is as though N left out the color of the needles or the shape of the cones—properties of the tree (and its parts) that both N and E see when they look at the tree.

Mystified by E’s complaint because he thinks his painting is virtually photographic in its realism, N hands E the paintbrush and invites him to fill in what is missing: ‘Show me, please, what you see that I don’t see.’ What is the forester going to add or change to make N’s painting a more accurate depiction of how the pine tree looks to him? What pigments will E select? Where will he apply the pigment? What details N so carefully recorded in his painting of the tree will E paint over? Since we are supposing N’s painting is virtually photographic in its realism, would E also find something missing in N’s photograph of the tree? Does E experience something cameras fail to capture? These, obviously, are meant to be rhetorical questions. If N was doing his job, he didn’t leave anything out of his painting. Neither did his camera. This is not to say that Goldilocks # is never the right reaction of an expert or trained observer to the efforts of a novice. It is not to deny the possibility of genuine perceptual learning.6 Yes, perhaps a person’s perceptual experiences can be enriched by exposure to and prolonged training with relevant stimuli. Perhaps after long experience tasting and comparing wines, the connoisseur actually begins to taste things (the hint of tannin) he didn’t taste before. Maybe musicians hear things (a change of key?) novices never hear. Maybe prolonged exposure or practice ‘tunes’ earlier processes in the sensory pathways to make them more sensitive to subtle differences important to specialists. If this is so—and there is certainly evidence that it is so—these improvements in perceptual acuity will result in greater discriminatory power. They will reveal themselves in a Goldilocks test. If E, as a result of long experience in the forest, actually perceives shades of green or subtle differences in bark texture that N does not see, this will be evident in E’s corrections to N’s painting. E will paint what he sees, the differences he sees, that N does not see. E will pass the test by responding with a plausible (to independent experts) Goldilocks #: 5 I ignore things that are necessarily left out in a two-dimensional representation of a three-dimensional scene—e.g. stereoscopic cues for depth. These will also be absent in E’s painting of the tree, so they are not something he would find ‘missing’ in N’s painting. 6 By ‘genuine perceptual learning’ I mean what Eleanor Gibson () meant when she defined perceptual learning as a ‘relatively permanent and consistent change in the perception [my emphasis] of a stimulus array following practice or experience with this array’. It is what Fahle (: ix) clearly intends by defining it as an improvement in discrimination as a result of learning to perceive something new that one could not perceive before. Psychologists (see Watanabe et al. ; Fahle et al. ) sometimes mean something more general by perceptual learning—e.g. an improvement in a human’s ability to perform perceptual tasks (e.g. distinguishing triangles from squares). This more general notion of perceptual learning clearly includes cases in which there may be no perceptual change at all and, therefore, nothing that is relevant to the topic of this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perception versus conception



N left something out, subtle differences in color or texture that E sees when he looks at a pine tree. But Goldilocks # is not a believable reaction with respect to pine-tree-ness. N didn’t leave this out of his painting. If he did, if Goldilocks # is E’s reaction to N’s painting, why can’t E restore the missing feature in his own depiction of the tree? His failure to do so provides us with convincing reasons to think there is nothing missing in N’s painting. In the relevant sense of ‘looks’ the pine tree looks the same to both of them. Goldilocks #: Too much. The forester complains that N has put too much in his painting. N depicted the tree as having five needles in each cluster with long slender cones, something not characteristic of pine trees in general. N did so, of course, because the tree he was painting turned out to be a White Pine tree and White Pine trees have five needles in each cluster and long slender cones. But in doing this N did not, the forester complains, depict the pine-tree-ness of the tree, something he sees when he looks at the tree. At best, N depicted white-pine-treeness or, maybe, an even more specific kind-property—eastern-white-pine-tree-ness. To paint pine-tree-ness N should have avoided details not characteristic of pine trees in general.7

Once again, I can’t imagine this to be a possible reaction, not one Siegel would put into the mouth of her expert. If one is going to see kind properties, one should be able to see them despite seeing details that are distinctive of more specific kinds. The forester, after all, saw these details when he looked at the tree itself. He noticed the five needles in each cluster and the long slender cones. That did not, or so we are told, prevent him from seeing the pine-tree-ness of the tree. So the details N included in his painting shouldn’t prevent E from seeing that property in N’s painting. This point should have been obvious from the beginning. N did not misrepresent what both he and E saw when he put White-Pine details in his painting. If these details did not prevent the forester from experiencing pine-tree-ness when he saw the tree, they shouldn’t prevent him from experiencing it when he looks at N’s painting. So ‘too much’ cannot be the right reaction. This leaves Goldilocks #, the only reaction I can imagine the forester having. Goldilocks #: Just right. N’s painting looks just right to the forester. It depicts the tree exactly as he sees it, exactly as he would have painted it. Nothing added, nothing subtracted. The forester sees pine-tree-ness in N’s painting in the same way he sees it when looking at the pine tree.8 So N’s painting is just right—just right to N who does not see pine-tree-ness in either the tree or the painting, and just right to the forester who sees pine-tree-ness in both. The forester’s experience is different from N’s, according to Siegel, when both see a pine tree, yes, but it is also different when both are looking at the painting. 7

Just as the concept of a pine tree leaves out properties not characteristic of pine trees in general. Siegel says that a non-pine tree (e.g. a hologram) could look exactly the same as a pine tree to an expert who knew it wasn’t a pine tree (Siegel : ). So the forester can see pine-tree-ness in something he knows is not a pine tree. I infer, therefore, that even if E knows N’s painting is not a pine tree, he could (according to Siegel) still see pine-tree-ness when he looks at it. If he cannot, we are back to Goldilocks #. N left something out of his painting—a property the forester sees when he looks at the pine tree but not when he looks at N’s painting. 8

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. dretske

If this is the option we are forced to accept, then since N’s painting of the pine tree is not a pine tree, E’s visual experience of N’s painting misrepresents the painting. It represents it as having a visible (to E) property—pine-tree-ness—it doesn’t have. N’s experience of the painting, on the other hand, is (at least in this respect) veridical. It represents what he sees (the canvas surface) as having a spatial arrangement of colors, shapes, and textures—colors, shapes, and textures the canvas actually has. What Goldilocks # means, of course, is that N always and unintentionally includes in his painting of a pine tree properties (pine-tree-ness) he does not see in the tree he is painting. This sounds suspicious. If N always and inevitably paints pine-tree-ness in his paintings of pine trees by arranging colors, shapes, orientations, and sizes (the properties he does see) the way he does, it begins to sound like pine-tree-ness is really just an arrangement, a configuration, of simple properties both E and N experience when looking at pine trees. If this is so, the property of pine-tree-ness that E sees is simply an arrangement, a spatially structured array of colored shapes that N also experiences when he sees pine trees. He just doesn’t know that that arrangement is characteristic of pine trees. It is like a child who does not know what a triangle is when seeing—and, yes, even painting—three lines arranged as in Figure .. Without realizing it, the child sees and paints triangularity because he sees and paints three lines arranged in one of the ways that constitute the normal look of a triangle. The child just doesn’t understand that that arrangement of lines is a triangle. There is, however, nothing more to seeing the property triangularity than seeing three lines that look the way Figure . looks to normal observers.9 To insist, on the contrary, that experts not only know (understand) more than novices but see more, that their visual experience is different, is to opt for Goldilocks #: N has left something out of his painting, something E sees when he looks at the

Figure . Triangulation 9 The existence of illusions and hallucinations should make it clear that one can see the property triangularity without seeing a triangle—i.e. when no object one sees actually has this property. If lines are arranged in three dimensions to look (from one’s perspective) like a closed figure and thus like a triangle, one sees (experiences) triangularity without actually seeing (experiencing) a triangle. If this sounds a little odd, think of seeing or experiencing the color red or movement without seeing or experiencing anything (any object) that is red or moving.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perception versus conception



Figure . Duck/rabbit

tree (or triangle). This leaves us with the question of why, despite seeing it, E cannot paint it. There are, of course, properties E knows the pine tree has when he sees that it is a pine tree that he cannot paint. Some properties are not paintable. But these properties contribute nothing to the phenomenology of seeing a pine tree. They are not represented in E’s experience of the tree. It may be objected, though, that Goldilocks yields counterintuitive results when applied to familiar perceptual phenomena and is for that reason unacceptable as a test for perceptual differences. I do not think so. Consider the familiar duck/rabbit (Figure .). When D sees the drawing as a duck and R (this could be D a moment later) sees it as a rabbit, are their perceptions different? Does it look different to them in that sense of the word ‘looks’ that captures the phenomenal character, the sensory aspect, of their perceptual experiences? Or is this instability in how one perceives the figure only a difference in how one interprets, or is disposed to interpret, the drawing (as of a duck rather than of a rabbit) or, perhaps, a difference in the pattern of eye fixations (on the bill of the duck rather than the nose of the rabbit)?10 Using the Goldilocks test, we ask D to paint the way the duck/rabbit looks to him. Will R regard this as an accurate painting of the way the duck/rabbit looks to him? Won’t they both draw the same thing—a duck/rabbit? If so, then Goldilocks decrees that R and D are having the same visual experience of the duck/rabbit. Since the difference in their experience is not paintable, it isn’t visual. This is not a visual phenomenon at all. One sees it as a duck and one sees it as a rabbit, yes; it looks like a duck to one, like a rabbit to the other, yes; but what they see—the duck/rabbit—is the same. And, as their drawings attest, it looks the same to both of them. They do not have different visual experiences of it. The same is true of other unstable perceptions. Consider figure and ground reversal (Figure .).

10 If the latter, the experiential differences are attributable to a difference in the stimulus, not to a difference in how the stimulus is processed. As such, the example would not illustrate the relevant cognitive influence on perception.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. dretske

Figure . Figure–ground ambiguity

There is clearly a difference between seeing a white figure—a vase, say—on a black background and seeing two faces in profile flanking a white background. This can change rapidly. Is it a visual difference? It is clearly a difference in how the figure looks—first this way, then that way—but, once again, this doesn’t settle matters. Is the difference in how the drawing ‘looks’ to be understood in a purely visual sense of this word or in some other way? Goldilocks makes it hard to see how this could be a difference in the experience. If asked to draw the way Figure . looks to them, won’t they both draw the same thing—viz. Figure .? How could they not? What else is left to draw? When we use words, verbal representations, we describe the figure differently, yes. One describes it as ‘a vase’ and the other as ‘two faces’, but their pictorial descriptions are the same—an ambiguous figure. Goldilocks tells us it is the latter, the pictorial descriptions, that are critical for identifying and distinguishing visual experiences. There are, nonetheless, practical limitations in applying Goldilocks. It is not at all clear, for instance, what the analogue of Goldilocks would be with taste, smell, or touch. I can ‘describe’ to you—using pictures—how something looks to me, but how do I describe in analogous sensory terms how things taste, smell, or feel to me? How do we use Goldilocks to decide whether a wine really tastes different, more complex, to the connoisseur, or whether he or she is simply able to identify something in the wine that the unwitting novice also tastes? To use Goldilocks to decide this, we need the analog of a picture, a ‘gustatory description’, a replication that a novice can produce that the expert can evaluate for accuracy and respond with Goldilocks #: ‘Yes, that is the way it tastes to me’ or Goldilocks #: ‘No, you left out the taste of apricots.’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perception versus conception



Aside from these difficulties, though, there are problems in using Goldilocks even in the case of vision.11 These problems relate to the fact that a painting of what one sees or of how things look is an expression of what one thinks (judges, believes, knows) one sees, not—not necessarily anyway—of what one actually sees. It may not, therefore, be a reliable guide to the character of the visual experience. This can be illustrated using an old example from epistemology: the speckled hen. Imagine a speckled hen with  clearly visible speckles on its facing side. N observes the hen and paints what he sees. E, an expert, observes the hen from the same angle for the same amount of time. He then examines N’s painting. He claims (Goldilocks #) N left something out. He painted too few speckles. E didn’t take the trouble to count the speckles so he can’t be sure exactly how many he saw, but he is quite sure there are more speckles on the hen than there are on N’s painting of the hen. As a result, E is convinced that his experience of the hen was more specklish than his experience of N’s painting. So if N’s painting is a true description of what N saw, there is more to what E sees when he looks at the hen than there is to what N sees when he looks at the hen. N left something out of his painting, the st speckle, something that E, but apparently not N, saw when he looked at the hen. Can we conclude from this that N and E had different visual experiences? That E saw things N failed to see? No. It shows, at best, that E thinks, or is inclined to think, that he saw more speckles on the hen than there were on N’s depiction of the hen— that, therefore, if the painting is an accurate representation of how the hen looked to N, E’s experience of the hen was richer, more specklish, than N’s. But why suppose that the painting is an accurate representation of how the hen looked to N? Maybe N failed to paint some of the speckles he saw. N (just like E) did not, after all, count the speckles. Maybe he saw more than he thinks he saw. Maybe, in fact, he saw all  speckles that E saw. N just misjudged, underestimated, how many that was. If this is what happened, then despite Goldilocks #, E’s experience was not more specklish than N’s. They had exactly the same visual experience. Both of them saw  speckles. It is only their judgments that are different—judgments that (in the case of N) resulted in his painting too few speckles. Pine trees, of course, are complex stimuli. So if N left out branches, cones, or needles in his painting of the tree, one cannot infer he didn’t see them. Likewise, if E regards N’s painting of the tree as ‘just right’ (Goldilocks #), we cannot infer that the painting is just right, that it faithfully represents what E experienced when he looked at the tree. Maybe E saw things that N did not see when they looked at the tree, but E doesn’t 11 I here ignore inverted spectrum possibilities—i.e. the possibility that red things look green to the novice and green things look red. Since this perceptual ‘inversion’ (since birth) also holds for the pigments the novice now uses to paint the things he sees, his painting of the pine tree will look ‘just right’ (i.e. green) to the expert (Goldilocks #), and just right to the novice (i.e. red), despite looking (in the relevant sense of ‘looking’) much different to the two of them (green to the expert, red to the novice). If such spectral inversion is deemed a real possibility, it is a possibility that challenges all efforts to compare intersubjective (but maybe not intrasubjective) experiences. Since it is a challenge to all conceivable tests, I do not take it as a serious objection to the Goldilocks test.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. dretske

notice (detect) their absence in N’s painting. This would then be a case of difference blindness, a failure to detect or identify visible differences. It is not genuine blindness, a failure to see visible differences. If this is possible—and I here assume it is12 —then Goldilocks is not an infallible guide to differences in experiences even in the case of vision where Goldilocks has its most compelling application. Not unless we assume—unrealistically—total visual recall on the part of the subjects being tested. This, though, only shows something that should have been evident from the beginning: Goldilocks is not so much a test, a practical tool for showing two experiences are the same (Goldilocks #) or different (Goldilocks #). I don’t expect anyone to actually use this test. Goldilocks provides, instead, a way of thinking about what would have to change if genuine penetration were to occur. A person’s experience would have to change in the sense of the word ‘experience’ in which a person could (in the case of vision) replicate the difference in pictorial terms. Inability to do so leaves one with reasons to think such differences do not exist. It is for this reason I conclude that one’s unstable experience of duck/rabbits and figure–ground is not—not in the strictest sense—a perceptual phenomenon. It is also why I remain skeptical of alleged accounts of cognitive penetration. Before I believe that experts see things I don’t see, I want a demonstration of the superior powers of discrimination this improved acuity confers on them. I want them to pass the Goldilocks test. Until they do, until I have a convincing Goldilocks # reaction, I will go on believing things look to them the way they look to me. Yes, they know a lot more than I do, and, yes, they come to know it by looking, but that doesn’t mean they see more.

Acknowledgments My thanks to Dennis Stampe and Judith Fortson for their help. I am grateful to Judith for suggesting ‘Goldilocks’ as an appropriate name for my test.

References Dretske, F. (). What change blindness teaches about consciousness. In J. Hawthorne (ed.), Philosophical Perspectives, vol. (): Philosophy of Mind, –. Oxford: Blackwell. Dretske, F. (). What we see: the texture of conscious experience. In B. Nanay (ed.), Perceiving the World, –. New York: Oxford University Press.

12 There is an enormous literature on change (or difference) blindness. I hope my example is sufficiently plausible not to need supporting argument. Despite differing judgments about what they saw, despite a Goldilocks # reaction on the part of E, my claim that N might have seen all the speckles E saw and that, therefore, they might have had exactly the same visual experience, is something that (if further support is needed) I defend in Dretske () and ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

perception versus conception



Fahle, M. (). Introduction. In M. Fahle and T. Poggio (eds), Perceptual Learning, ix–xx. Cambridge, Mass.: MIT Press. Fahle, M., Edelman, S., and Poggio, T. (). Fast perceptual learning in hyperacuity. Vision Research : –. Gibson, E. J. (). Perceptual learning. Annual Review of Psychology : –. Siegel, S. (). The Contents of Visual Experience. Oxford: Oxford University Press. Watanabe, T., Nañez, J. E., and Sasaki, Y. (). Perceptual learning without perception. Nature : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Cognitive Penetration and the Reach of Phenomenal Content Robert Briscoe

 Introduction The phenomenal character of a perceptual experience depends, in part, on its representational content. What it is like to enjoy the experience of seeing a bowl of fruit, for example, is determined by (among other things) the various shapes, sizes, colors, and textures that objects are represented as having in the experience. In keeping with recent usage (see e.g. Kriegel  and Bayne ), I shall employ the expression ‘phenomenal content’ to refer to that component of a perceptual state’s representational content that supervenes on its phenomenal character. Features that are represented in a perceptual experience, but that can vary without any change in the experience’s phenomenal character—supposing there be such—are outside the scope of phenomenal content as understood here.1 Are the phenomenal contents of visual experience informationally encapsulated? Can information originating from outside of the visual system influence the way an object appears to an observer, where ‘appears’ is interpreted in a phenomenal as opposed to epistemic sense and where the relevant influence on experience is direct, e.g. not mediated by shifts in selective attention? If so, then it is psychologically possible for two observers (or for one observer at different times) to have visual experiences with different phenomenal contents while seeing and attending to the same distal stimuli under the same external conditions as a result of differences in other representational mental states.2 This chapter critically examines two different ways of defending the claim that visual phenomenal content is not informationally encapsulated, that it can be ‘penetrated’ by 1 See Prinz () for an argument that conscious perceptual experiences can have contents that are not reflected in their phenomenal character. A visual experience, e.g., on Prinz’s account can represent an object as having the high-level property of being an apple even if that property makes no difference to the way the object (non-epistemically) visually appears. 2 This formulation is adapted from Siegel ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  nonvisual sources of information about the world or the perceiver’s body. I shall refer to this claim, in what follows, as the ‘penetrability thesis’ (or ‘Thesis P’ for short). Some philosophers have recently argued that acquiring the capacity to categorize an object as belonging to a certain high-level kind can alter the way the object subsequently appears. For example, your visual experience of seeing a tiger before acquiring the ability to distinguish tigers from non-tigers, it is claimed, can have a different phenomenal character than your visual experience after acquiring the ability. It is then argued that the best explanation of the putative phenomenal contrast between the two experiences is that the latter, unlike the former, represents the high-level kind property tiger. The conclusion that visual phenomenal content can come to include high-level properties—and thus is not restricted to low-level attributes such as shape, size, and color—has been referred to as ‘expansionism’ about phenomenal content, while its denial has been referred to as ‘restrictivism’ (Prinz ). I shall use this terminology here. Theorists who pursue one or another variation on this line of argument take what I shall call the ‘high road’ to Thesis P: they seek to show that visual experience can come to represent high-level properties via some sort of top-down influence on lower-level processing from cognitive systems involved in object recognition. Outputs emanating from systems dedicated to categorizing perceived objects on the basis of the highlevel kinds to which they belong can, under certain conditions, expand the stock of attributes that objects may be represented as having in conscious vision and hence, it is claimed, the way they thereafter look (see Siegel : ). A rather different approach to defending Thesis P eschews appeal to intuitions about how objects may phenomenally appear to an observer after acquiring certain recognitional dispositions. Further, it does not undertake to show that outputs from object-recognition systems can extend the reach of visual phenomenal content. Rather, it relies on psychophysical and neuroscientific evidence to motivate the claim that information originating outside the visual system can modulate the way an object’s low-level attributes visually appear. Accordingly, I shall refer to this second approach as the ‘low road’ to Thesis P. Unlike theorists who take the high road, theorists who take the low road attempt only to show that information present in areas outside the visual system can directly affect the value that some low-level, representational variable assumes, for example, by changing the weighting assigned to one or another source of optical stimulus information. This chapter is divided into three parts: Section  is dedicated to preliminaries and lays out the distinction between informational encapsulation and cognitive impenetrability. In Section , I critically examine three recent high-road arguments for Thesis P deployed respectively by Susanna Siegel (; ), Tim Bayne (), and William Fish (). I argue that none of them ultimately proves to be successful. In Section , I switch from offense to defense. In particular, I argue—in keeping with the low-road approach—that there is a substantial body of psychophysical and neuroscientific evidence that information originating outside the visual system can modulate

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

the way an object’s low-level attributes visually appear. Visual phenomenal content, I show, is significantly influenced not only by crossmodal interactions between vision and other exteroceptive senses such as touch and audition but also by interactions between vision and nonperceptual systems involved in motor planning and construction of the proprioceptive body image.

 Some Preliminary Definitions and Remarks Informational encapsulation is the most important of the nine definitive properties of a mental module enumerated by Jerry Fodor (; ). According to Fodor, if a perceptual input system such as vision or audition is informationally encapsulated, then the way it processes a set of inputs is computationally insensitive to whatever information might be present elsewhere in other modular or non-modular systems. Cognitive impenetrability, by contrast, only excludes high-level penetration, e.g. penetration by the subject’s beliefs, expectations, desire, and other centrally accessible mental states. As Pylyshyn puts it, ‘if a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, that is, it can be altered in a way that bears some logical relation to what the person knows (: ; emphasis added). That a perceptual input system is cognitively impenetrable, in this sense, is consistent with informational unencapsulation. Even if processing internal to the system is impervious to high-level influences of various kinds, it may be computationally sensitive to low-level information originating elsewhere in the brain. Fodor and Pylyshyn have focused primarily on the question of whether the nonconscious, ‘early’ visual system is informationally encapsulated and/or cognitively impenetrable. The notions of informational encapsulation and cognitive penetrability, however, can be usefully extended to discussions of conscious visual experience (e.g. Macpherson ; Siegel ; Wu ). Visual experience, to a first approximation, is informationally encapsulated if its phenomenal contents are not directly influenced by sources of information emanating from outside of the visual system. And visual experience can be characterized as cognitively impenetrable if its phenomenal contents are not directly influenced by high-level, centrally generated mental states, such as beliefs, desires, and intentions.3 Orthogonal to the distinction between informational encapsulation and cognitive impenetrability—a distinction with respect to putative sources or causes of penetration—is a distinction between different kinds of effects on phenomenal content. High-road arguments for Thesis P, as characterized above, maintain that information emanating from object recognitional systems can cause high-level properties 3 A direct causal link of some kind, however, isn’t enough. As Wayne Wu () has argued, to establish that visual phenomenal content is penetrated by information originating in some non-visual system Y, it is further necessary to provide a mechanism whereby the computations that support visual experience are able to exploit Y as a representational resource.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  to be represented in visual phenomenal content. For example, acquiring the ability to distinguish tigers from non-tigers might cause the high-level property tiger to be represented in one’s visual experience of a tiger. On the assumption that the source of penetrating influence here is a properly cognitive one, this would be a case in which a high-level penetrator has a high-level effect on phenomenal content. Alternatively, a high-level, cognitive penetrator might have a low-level effect on phenomenal content, e.g. by modifying the way an object’s shape, or orientation, or lightness appears. Correspondingly, a low-level penetrator could, in principle, have either low-level or high-level effects on phenomenal content.

 High-Road Approaches to Thesis P In this section, I critically assess three different arguments for the view that acquiring object-recognitional dispositions can cause high-level properties to be represented in the phenomenally conscious contents of visual experience. The first two arguments, developed respectively by Susanna Siegel () and Tim Bayne (), employ an essentially phenomenological methodology. The third argument, developed by William Fish (), by contrast, draws support from an array of recent psychophysical and neuroscientific findings.

. The method of phenomenal contrast In the method of phenomenal contrast, a putative difference in phenomenology between two overall experiences, OE and OE, is treated as an explanandum that different, rival hypotheses about the contents of visual experience compete to explain (Siegel ). Siegel brings the method to bear on two cases in which acquired recognitional dispositions seem to affect phenomenology. The first case contrasts one’s overall experience when looking at a page of Cyrillic text before learning Russian with one’s overall experience of looking at the same page after learning Russian. The second case contrasts one’s overall experience when looking for pine trees in a forest before and after learning to discriminate pines from other kinds of trees. After acquiring the relevant recognitional skills, ‘you can spot the pine trees immediately: they become visually salient to you’ (Siegel : ). The argument from these cases to what Siegel calls the ‘Rich Content View’ has the following structure (let ‘VE’ and ‘VE’ designate the visual experiences one has when enjoying overall experiences OE and OE, respectively): () The overall experience OE of which VE is a part differs phenomenologically from the overall experience OE of which VE is a part. () If OE differs phenomenologically from OE, then there is a phenomenological difference between visual experiences VE and VE. () If there is a phenomenological difference between VE and VE, then VE and VE differ in phenomenal content.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

() If there is a difference in phenomenal content between VE and VE, then it is a difference with respect to the high-level properties represented in VE and VE. For example, the change in how salient or attention-grabbing the pine trees are to you in the second case, Siegel suggests, is to be explained by the presence of the high-level property pine in the phenomenal content of VE and its absence in the phenomenal content of VE. Siegel defends premise () against re-descriptions according to which OE and OE differ in respect of either occurent non-sensory, cognitive phenomenology or ‘moodlike’ background phenomenology. There are other relevant alternatives in the Cyrillic text case, however. In particular, reading words in a language that one understands often elicits auditory (and sometimes visual) imagery that is absent when one scans a page of text written in an unfamiliar alphabet. Grasping the meaning of what we read can also, in many cases, affect our emotional responses. A sign on a forest trail that reads BHИMAHИE: MEДBEДИ ГPИЗЛИ ПOБЛИЗOCTИ! (BEWARE: GRIZZLY BEARS NEARBY!) will elicit quite different feelings from Russian speakers and nonRussian speakers, respectively. When it comes to the pine-tree example, the most promising skeptical tack, it seems to me, is rather to challenge premise (), i.e. to accept that there is some sort of difference in phenomenology between VE and VE, but to deny that it is specifically a difference with respect to the high-level kind properties that the two experiences respectively represent. It is plausible, as Siegel suggests, that an important respect in which the experience of seeing pine trees changes after one learns to recognize them has to do with their visual salience. Once you know what a pine tree looks like, the presence of a pine tree in your field of view (at least when you are looking for one) is apparent in a way that it wasn’t before. (And plausibly such a change in your experience is itself phenomenologically salient: you notice that it is now easier to find pine trees than it was before.) There are two different ways, however, in which visual salience can be conferred on a perceived object. On the one hand, objects sometimes have lowlevel features in virtue of which they ‘pop out’ from their background and capture the perceiver’s attention in an exogenous, bottom-up manner. A bright yellow triangle, for instance, will be especially noticeable when presented against a background filled with dark, purple discs. Evidently, acquiring the ability to recognize pines does not cause them to be visually salient in this sense. On the other hand, an object can be visually salient as the result of top-down factors that influence the allocation of overt or covert selective attention. The paradigm here is having a stored, inner representation of some kind—for example, a mental image or concept—that guides or otherwise influences you when searching for the object in a structured, visual scene. Presumably, it is this latter, recognition-based kind of visual salience that is at issue in Siegel’s second example. In what does recognition-based, visual salience consist? Recent models of top-down or ‘guided’ visual search suggest that high-level knowledge about a target object can

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  prime or configure the visual system to find it in a scene, in particular, by amplifying the conspicuity of the object’s location in what vision researchers refer to as the perceiver’s ‘saliency map’ (Koch and Ullman ; Wolfe ; Blaser et al. ; Wolfe et al. ; Underwood et al. ). A saliency map is an abstract, topographical representation that records the overall attentional strength of each location in visual space. The greater the attentional strength of a location on the map, the more likely it is to be selected. Evidence gathered in research on top-down search suggests that a location’s attentional strength is a function of both pre-attentive, bottom-up feature processing and top-down, cognitive influences of various kinds. For example, stored knowledge about an object’s visual appearance might increase the selectionrelevant weightings assigned to visual representations of certain low-level features, e.g. certain shapes or colors, if they are distinctive of the object (Wolfe et al. ). Alternatively, knowledge about the object’s most probable locations in the distal scene might increase the attentional strength of certain regions on the saliency map while decreasing the attentional strength of others. Subjects, for example, are much more likely to fixate locations on a sidewalk or road when looking for bicycles than when looking for chimneys or clouds (Henderson and Hollingworth ; Henderson et al. ). For present purposes, the main point is that the increased visual salience or attention-grabbingness of the pine trees in Siegel’s second case can be parsimoniously explained without recourse to the hypothesis that acquiring the ability to recognize pines causes a certain high-level kind property to be represented in visual phenomenal content. Rather, in the context of visual search, the visual salience of pines may be amplified in a top-down manner by knowledge about their distinctive, low-level features and/or their statistically most probable locations in the scene. Plausibly, both kinds of knowledge are often acquired in the course of learning to discriminate pine trees. It should be emphasized in this connection that implicitly categorized stimuli can capture visual attention even when subjects are not explicitly looking for them. Studies of inattentional blindness (IB) reported in Mack and Rock () provide evidence that nonconsciously perceived stimuli are often implicitly categorized at the highest levels of perceptual processing. According to Mack and Rock’s model of late attentional selection, objects that are implicitly categorized as highly meaningful or task-relevant have the ability to defeat IB and capture the perceiver’s attention (: chs  and ; also see Mack ). Another way, then, in which acquiring the ability to recognize an object can contribute to its visual salience is by making the object capable of defeating IB even when it isn’t the target of deliberate visual search. The object will seem especially salient, in other words, because implicit categorization has made it attention-grabbing. These proposals need to be fleshed out in more detail than is possible here. They hopefully serve to show, however, that the boost in a kind’s visual salience after subjects have learned to recognize its members can be parsimoniously explained without

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

recourse to the expansionist conclusion that the relevant high-level kind property has come to be represented in visual phenomenal content.

. Associative visual agnosia and the method of phenomenal contrast A contrast argument recently developed by Tim Bayne () appeals to changes in phenomenology that are consequent not on the acquisition of recognitional dispositions but rather on their loss in ‘pure’ associative agnosia. In relevant cases, visually acuity, form discrimination, D perceptual organization, and other low-level perceptual abilities remain intact, but patients are incapable of categorizing objects as belonging to certain high-level kinds, for example, as tigers, or teapots, or tomatoes. Bayne contends that associative agnosia is characterized by a loss of a ‘layer’ of high-level phenomenal content. The three brief arguments he presents in support of this claim are motivated by a description of deficits in a patient with visual agnosia provided by Rubens and Benson (): For the first three weeks in the hospital the patient could not identify common objects presented visually and did not know what was on his plate until he tasted it. He identified objects immediately on touching them. When shown a stethoscope, he described it as ‘a long cord with a round thing at the end’, and asked if it could be a watch. He identified a can opener as ‘could it be a key?’ Asked to name a cigarette lighter, he said, ‘I don’t know’, but named it after the examiner lit it. He said he was ‘not sure’ when shown a toothbrush. Asked to identify a comb, he said, ‘I don’t know’. When shown a large matchbook, he said, ‘It could be a container for keys’. He correctly identified glasses. For a pipe, he said, ‘Some type of utensil, I’m not sure’. Shown a key, he said, ‘I don’t know what that is; perhaps a file or a tool of some sort’. He was never able to describe or demonstrate the use of an object if he could not name it. If he misnamed an object his demonstration of its use would correspond to the mistaken identification. . . . Remarkably, he could make excellent copies of line drawings and still fail to name the subject. . . . He easily matched drawings of objects that he could not identify, and had no difficulty in discriminating between complex non-representational patterns differing from each other only subtly. He occasionally failed in discriminating because he included imperfections in the paper or in the printer’s ink. He could never group drawings by class unless he could first name the subject. (Rubens and Benson : –)

Bayne uses associative visual agnosia to develop a quick contrast argument for expansionism. Although the patient’s experience of low-level perceptual attributes is intact, it is nonetheless ‘extremely plausible’, he suggests, ‘to suppose that the phenomenal character of his visual experience has changed’ (Bayne : ). But since the patient, by hypothesis, has not lost low-level phenomenal content, it follows that he must have lost high-level or categorical phenomenal content. Restrictivists will regard this contrast argument as simply question-begging. There is nothing in the quoted passage from Rubens and Benson () that goes beyond the standard, textbook characterization of associative visual agnosia: viz. the presence of normal abilities to discriminate low-level perceptual attributes in the absence of abilities to recognize (properly categorize) the objects to which those attributes belong.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  There is nothing in this characterization, by itself, that underwrites the claim that associative agnosia leads to a ‘disruption to the phenomenal looks of objects’ (p. ). Bayne deploys two brief additional arguments in support of the proposal that ‘ “x looks F to S” can capture a genuine phenomenal-looks claim even when “F” expresses a property that is not sensory in any natural sense of that term’ (: ). Both proceed by defending view that the kind of object recognition that is impaired in visual agnosia falls on the perceptual side of the perception/cognition divide. The first argument is based on the claim that ‘object recognition of this kind resists doxastic penetration. It does not matter what one believes about an object; it still looks like a pipe, a stethoscope, or a cigarette lighter’ (Bayne : ; emphasis added). This argument, however, begs the question against restrictivism once again. The restrictivist about phenomenal content, as Bayne acknowledges, is quite happy to allow that a thing can look in a non-phenomenal, epistemic, or comparative sense like a pipe or some other high-level kind of object. What the restrictivist will not allow, without a convincing argument, is that a thing can also sometimes look in a phenomenal sense like a pipe. The second argument relies on the claim that object recognition cannot be restored in associative agnosia by the ‘insertion’ of a belief about the perceived object’s highlevel kind properties. The argument goes like this: () According to nonperceptual, doxastic models of object recognition, there are two components in visual recognition, a belief component and a looking component: seeing that such and such a type of object is present is a matter of forming an appropriate belief or judgment on the causal basis provided by one’s visual experience of the object. () But suppose that a patient with pure associative agnosia ‘suffers from a freak neurophysiological condition that causes him to believe that every object he is looking at is a pipe. This case satisfies . . . the causal condition on visual recognition, but it seems doubtful whether it suffices to reinstate the missing experiential content’ (Bayne : ). () Hence, doxastic models of object recognition are inadequate. ‘Perceptual recognition is not simply a matter of believing that such and such a type of object is present whilst enjoying low-level visual experience’ (Bayne : ). This line of argument is doubly problematic. First, premise () again simply assumes that a layer of experiential content is missing in visual agnosia. Second, it is doubtful that proponents of nonperceptual models of object recognition would accept the freak neurophysiological condition described in premise () as an adequate causal basis for recognizing something as a pipe. In order to recognize an object O as a pipe, it might be argued, it isn’t sufficient that seeing O just somehow causes you to form the belief that O is a pipe. Rather, most psychological theories of object recognition suggest that the adequate causal basis minimally would involve perceiving O to have a certain cluster of visual attributes, e.g. a certain shape and size, and, in addition, computing

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

the similarity between O and a stored, inner representation of the category pipe. From this empirically informed perspective, a ‘freak’ neurophysiological condition that somehow causes you to believe that every visually perceived object is a pipe is a condition in which no object is visually recognized as a pipe. Bayne, I conclude, fails to show that cases of pure visual associative agnosia provide any non-question-begging reason to think that the reach of perceptual phenomenality extends to high-level properties.

. Empirical arguments for high-level properties in visual experience Unlike Siegel and Bayne, William Fish () has evinced skepticism about the prospects of using an essentially phenomenological methodology to determine the scope of phenomenal content. Instead, he appeals to three different sources of empirical evidence to motivate the conclusion that high-level properties appear in the ‘sensory, presentational component’ of visual experience—in its phenomenal character— and not merely in our cognitive or ‘interpretive’ responses to that component.4 Although Fish is primarily concerned with the question of which properties appear in the sensory component of visual experience rather than with the question of whether or not visual experience is cognitively penetrable, the empirical evidence he reviews is certainly germane to the latter. As Fiona Macpherson () writes: People who subscribe to the existence of high-level content in visual experience are likely to reject cognitive impenetrability. This is because many of the arguments for high-level content proceed by arguing that learning can affect which visual experience one has. . . . For example, plausibly, the visual system does not come ready-made to represent pine trees or the specific individual that is my brother. This is a type of tree and a particular human that not all humans will encounter. If all such specific representational abilities had to be built-in to the visual system it would be enormous and unwieldy. But, perhaps on repeated exposure to pine trees or to my brother, a subject can come to notice features that all and only such trees have and that all and only that person has. And perhaps the subject’s knowledge of these features can feed into their visual system so that they come to have visual experiences that are sensitive to those features, at least sensitive in a way that they were not before. In this way, the subject would come to have visual experiences that they did not have before—visual experiences that represent pine trees or my brother. In other words, a very plausible mechanism for visual experiences coming to have high-level content is that the visual system is penetrated by the cognitive system. (: –)

Hence, if psychological findings suggest that high-level properties are represented in the phenomenal contents of visual experience—or, as Fish would rather put it, in its presentational component—then this would be seemingly good reason to suppose that visual phenomenal content has been penetrated by information originating outside of the visual system. 4 Although Fish’s naïve realist commitments lead him to speak of properties that appear in the presentational component of visual experience rather than properties that are represented in visual phenomenal content, this doesn’t affect my assessment of the evidence he adduces or the arguments he develops.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  Fish’s first source of evidence involves the finding that high-level object and scene properties can be perceived in the absence of focal attention. In particular, subjects engaged in a task that required them to attend to a -letter array presented in central vision were found to be able to determine with about % accuracy whether a scene presented in peripheral vision contained either animals or vehicles (Li et al. ). Here is the argument that Fish develops for the conclusion that high-level kind properties figure in the presentational component of visual experience: () ‘It would seem independently plausible . . . to suppose that, for processes of interpretation to take place, we would need to allocate additional cognitive resources to the task. . . . [I]f a property requires attention to be perceived, perhaps this is evidence that it should be located in the interpretative component of a visual experience; if it can be perceived preattentively, this is reason to think that it appears in the presentational component’ (Fish : ; emphasis added). () The best interpretation of findings reported by Li et al. () is that kind properties such as animal or vehicle can be perceived without attention. () So there is reason to think that kind properties such as animal or vehicle can appear in the non-interpretive, presentational component of visual experience. There is good evidence that an object’s properties can be perceived without attention. Indeed, psychophysical findings garnered in the previously mentioned studies of inattentional blindness suggest that a pre-attentively perceived object may be (implicitly) categorized as belonging to a certain high-level kind (Mack and Rock ; Mack ). The problem is that the same studies also provide compelling evidence that attention is necessary for conscious perception. In other words, a preattentively perceived object may be categorized by the visual system as an animal, or a vehicle, or some other high-level kind even though the object is absent from the sensory, presentational component of visual experience, contrary to premise ().5 In addition, the interpretation of the findings reported by Li et al. () in premise () is contestable. In particular, there is no reason to think that the scenes presented in subjects’ peripheral vision were completely unattended. Rather, as Jesse Prinz has pointed out, diffuse, nonfocal attention may have been allocated to them (Prinz : –). It is worth mentioning that the experimenters themselves conclude only that ‘rapid visual categorization of novel natural scenes requires very little or no focal attention’ (Li et al. : ; emphasis added). Fish’s second source of evidence comes from studies that suggest that high-level properties can be rapidly detected. For example, Fabre-Thorpe et al. report that human subjects can detect the presence of an animal with % accuracy and a median reaction time of  ms in natural images that are flashed for only  ms (: ).

5

See Cohen et al. (: ) for further evidence in support of this assessment.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

Fish assumes that, if properties can be detected quickly, then this is evidence that they appear in phenomenal character. Why? Quoting Prinz, he appeals to a dominant view of visual processing in cognitive science: [P]erception may work in the following way. After transduction, a signal is propagated through a hierarchically organized sequence of subsystems, which begin by producing representations of local features and move on to representations that are more global and invariant. (Prinz : )

The idea, I take it, is that if a property is detected quickly, then, given the above view of how visual perception works, its detection must occur at a relatively early stage in the visual processing hierarchy and, hence, prior to the involvement of cognitive, interpretative mechanisms. The problem is that the above account of perceptual processing actually implies that detection of viewpoint-invariant, high-level kind properties occurs at the uppermost stages of the processing hierarchy—for example, in the lateral occipital complex— not at earlier, feature-representing stages such as V or V (see e.g. Grill-Spector and Malach ). Categorization may be astonishingly rapid, but it is driven to a significant extent by even faster local feature processing. So, contrary to Fish, the rapidity with which objects and scenes are categorized as having certain high-level properties does not provide evidence that those properties are represented at a ‘preinterpretive’ level of perceptual processing. Fish’s final source of evidence comes from studies of visual adaptation and its aftereffects. Adaptation to an oblique grating, for example, can cause vertical lines subsequently to look tilted in the opposite direction. Similarly, adaptation to the downward flow of water in a waterfall can make stationary objects on the riverbank appear to flow upwards for a few seconds. While the underlying mechanism of adaptation isn’t fully understood, it is generally thought to involve reduction of activity and/or sensitivity in neural populations involved in processing the adapting stimulus attribute (Grill-Spector et al. ). Fish appeals to the existence of seemingly high-level adaptation aftereffects to motivate the claim that high-level properties appear in the presentational component of visual experience. In particular, he appeals to evidence provided by Burr and Ross () that numerosity is susceptible to adaptation. Adapting to a large number of dots on a circular patch, Burr and Ross report, decreases the apparent numerosity of a neutral (averagely populated) patch, while adapting to a small number of dots has the opposite effect.6 The conclusion that high-level properties can sometimes appear in the presentational component of visual experience is then inferred by means of the following argument:

6

A demonstration is available here:

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content 

Figure . Adapting texture density versus adapting numerosity. Reproduced, with permission, from Durgin ()

() ‘[A]ll agreed primary visual properties—the properties (such as size, orientation, shape, colour and motion) that everyone agrees appear in phenomenal character—are susceptible to adaptation’ (Fish : ). () ‘So if we can show that another property is also susceptible to adaptation, we have an argument that this property appears in phenomenal character too’ (Fish : ).7 () Burr and Ross () provide evidence that numerosity is susceptible to adaptation. () So there is evidence to think that numerosity, a high-level property, appears in phenomenal character. As Burr and Ross conclude: ‘just as we have a direct visual sense of the reddishness of half a dozen ripe cherries, so we do of their sixishness. In other words there are distinct qualia for numerosity, as there are for color, brightness, and contrast’ (: ). It is open to question, however, whether Burr and Ross () do in fact provide evidence for adaptation to numerosity. An empirically well-motivated, alternative explanation is that the numerosity judgments made in Burr and Ross’s experiment are based on perceived, relative texture density rather than perceived, relative numerosity (Durgin ). Consider Figure .. Following adaptation to the two circular texture patches in the top row, a neutral patch subsequently presented within the region adapted to the less numerous but more dense texture (bottom row, right) appears less dense, and hence less numerous, than a neutral patch presented within the region 7 The inference from () to () commits the converse error, but I shall put this objection aside to focus on premise ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

adapted to the more numerous but less dense texture (bottom row, left). Unlike numerosity, however, texture density is a low-level visual property. So evidence that texture density is susceptible to adaptation does not support the claim that high-level properties can appear in phenomenal character. Fish also appeals to evidence that adaptation to a human face can affect the visual appearance of subsequently seen faces. Adapting to a picture of a face that has been horizontally or vertically extended, for example, makes a picture of the original, nondistorted face appear distorted in the opposite direction (Webster and MacLin ). Adapting to an ‘anti-face’ (see Figure .)—a synthetic face created by locating the position of a target face relative to the average face in high-dimensional face space and then inverting the differences between them—makes the average face appear more like the target face and also makes it easier for subjects to identify the target face (Leopold et al. ). Adapting to a female face makes a subsequently presented androgynous face—formed by morphing between a female and male face—appear more masculine (Webster et al. ). Adapting to certain facial structures and textures and can lead to significant aftereffects in apparent age (O’Neil and Webster ).8 Corresponding to specific differences in configural and surface properties between faces, then, there are specific differences in adaptation and resultant visual aftereffects. Webster and MacLeod () argue that these aftereffects support a norm-based model of face coding and a normalization model of face adaptation. According to the norm-based coding model, individual faces are represented in the visual brain in terms of their direction and distance from a prototypical or average face in a highdimensional, perceptual face space (see Loffler et al.  and Leopold et al.  for consistent fMRI evidence). And according to the normalization model of face adaptation, adapting to an individual face ‘renormalizes’ perceptual face space so that the face that appears perceptually neutral or average is shifted in the direction of the adapting face.9 Studies and current theoretical models of face adaptation, while suggestive, don’t seem to provide compelling evidence that high-level face properties figure in what Fish calls the ‘presentational component’ of conscious visual experience. One source of difficulty is that aftereffects induced by complex, real-world objects are likely to reflect 8

For a detailed review of findings and theoretical assessment, see Webster and MacLeod (). Webster and MacLeod review several sources of evidence for a norm-based coding system. Here are two. First, adapting to a configurally distorted face significantly alters the appearance of a neutral, undistorted face, but adapting to the neutral face does not affect the appearance of the distorted face (Webster and MacLin ). The renormalization model predicts this asymmetry: ‘the neutral face simply reinforces the current norm or neutral point in face space and hence changes nothing, whereas the distorted adapting face induces a shift in the neutral point so that the previously neutral faces are no longer seen as such’ (Webster and MacLeod : ). Second, the appearance of the adapting face is itself altered by the adaptation process so that it looks less distinctive, i.e. less divergent from the norm, and, further, produces a global shifts in the same direction for all other faces in face space, e.g. adapting to a horizontally extended face makes every subsequently seen face look narrower. This again suggests the immediate result of adaptation is a ‘recentering’ of face space nearer to adapting stimulus so that the adapting face appears perceptually less distinctive and closer to the average. 9

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content 

Original face

Anti-face

Figure . Faces and their corresponding anti-faces. Reproduced, with permission, from Leopold et al. ()

changes in sensitivity at all levels of the visual processing hierarchy, starting as early as the retina (this is the problem of what Xu et al.  call ‘aftereffect inheritance’). In consequence, aftereffects that are suggestive of adaptation to some high-level attribute or category may in many cases be caused by adaptation of simple feature detectors or mechanisms that represent ‘holistic’ but low-level stimulus attributes. Thus Webster and MacLeod write: [T]he selectivity of the adaptation for different natural facial categories [e.g. race or gender] does not require that the mechanisms of face coding are directly tuned to these specific categories, for as long as the mechanisms can be differentially adapted by different categories they will result in selective aftereffects. (: )

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

[I]t remains unclear whether there are special axes in the representation of faces—perhaps corresponding to prominent natural categories—or if the selectivity of the aftereffects simply reflects the distances between the stimuli in face space . . . The point is that we cannot be confident that facial attributes that appear salient or ecologically relevant are the stimulus attributes that face processing is directly encoding. (: ; emphasis added)

In short, the available scientific evidence underdetermines the question of whether the aftereffects observed in face adaptation studies reflect changes in the sensitivity of neurons that explicitly represent high-level facial attributes or at the level of neurons that code for the lower-level ‘building blocks’ of possible faces. Hence, it is unclear at present how best to characterize the nature of face adaptation aftereffects. Preliminary grounds for skepticism about Fish’s assessment, however, comes from a recent study of cross-category adaptation. Javadi and Wee () adapted subjects to images of objects strongly associated with one gender, e.g. lipstick for females or motorcycles for males, and then asked them to judge the gender of an androgynous face. They found that subjects were more likely to characterize the appearance of the androgynous face as masculine when the images were of objects associated with women, and vice versa. It is highly implausible, however, that there is a high-level, visible property or ‘look’ that is common to feminine faces, tubes of lipstick, and tiaras (or to masculine faces, motorcycles, and electric razors). But, if not, then at least some gender-related visual aftereffects do not seem to arise in consequence of adaptation to the presence of high-level gender properties in visual phenomenal content.10

 Low-Level Effects on Visual Phenomenal Content According to Thesis P, visual phenomenal content is not informationally encapsulated, i.e. it can be penetrated by information from outside the visual system. In the last section, I criticized recent ‘high-road’ approaches according to which acquiring the ability to categorize an object as belonging to a certain high-level kind can cause the relevant kind property to be represented in visual experience (or to be present in what Fish calls its ‘presentational component’) and thereby alter the way the object subsequently appears. The approach to defending Thesis P, examined in this section, doesn’t undertake to show that inputs from cognitive systems can extend the reach of visual phenomenal content: it is neutral with respect to the question of whether such inputs can augment the range of attributes that objects may be represented as having in conscious seeing. 10 It should also be emphasized that there are numerous respects in which perceptual processing for faces is special, i.e. involves face-specific cognitive and neural processes (for discussion, see McKone et al. ). Hence, even if expansionist conclusions with respect to face perception were empirically warranted by face adaptation research, there would be no guarantee that they could be extended unproblematically to perception of other kinds of objects.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  Rather, it relies on psychophysical and neuroscientific evidence to motivate the claim that information originating outside the visual system can modulate the way an object’s low-level properties visually appear. There is a large body of evidence that when visual information about the environment conflicts with information received in another modality, vision typically dominates. In the ventriloquism effect, for example, the visually perceived location of a stimulus captures auditory localization of a sound source. Vision has also been found to exert a dominating influence on both touch and proprioceptive estimates of limb position and body posture (Stratton ; Harris ; Rock and Harris ). The effect of vision on other modalities is not limited to spatial localization. It can also have a striking influence on speech perception. In the McGurk effect, for example, which syllable a speaker is heard to pronounce strongly depends on how her lips visually appear to move. It would be a mistake, however, to think that when the senses conflict, vision is always the recalcitrant modality. While evidence that vision can affect phenomenal contents in other modalities has been available for over a century (Stratton ), recent findings from studies of multisensory integration suggest that the influence can also go in the other direction. In what follows, I present four cases in which sources of information originating outside of the visual system affect the way an object’s lowlevel properties appear in visual experience. The first three are cases of what I shall refer to as ‘synchronic, low-level informational penetration’ (or SLIP). A fourth case, by contrast, suggests that visual phenomenal content can sometimes be cognitively penetrated by the subject’s high-level intentions to engage in motor actions.

. Tactile-visual SLIP There are numerous, independently variable sources of information in the light sampled by the eye. The more important of these include binocular disparity, vergence, occlusion, motion parallax, texture gradients, shading, reflections, and relative size (for a comprehensive review, see Bruce et al. ). In many cases, however, visual estimation of an environmental property is optimized by recruiting sources of auxiliary information from outside the visual system. These may be directly integrated with available sources of optical stimulus information and/or used to change the weighting assigned to one or another source of such information. Receiving touch-based information about a surface’s D orientation, for example, can influence the way the surface subsequently visually appears. Ernst et al. () performed an experiment in which texture-based and disparity-based sources of optical information about a surface’s slant were set in conflict. They found that haptic feedback consistent with one or the other source of visual information changed the subsequent visual appearance of the surface’s slant for up to twenty-four hours. For example, when haptic signals were consistent with the texture-specified slant, the visual appearance of the surface’s D orientation was closer to the texture-specified slant than before training.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

This finding indicates that the weighting of a source of visual information in perceptual processing is not always determined by its overall ecological reliability, but can also be modulated contextually by feedback from other senses. In a recent study, van Beers et al. () undertook to estimate how rapidly the weightings assigned to visual cues change during training with haptic feedback. Subjects were instructed to place a cylinder flush onto a slanted, checkerboard surface. The apparent D orientation of the surface was perceptually determined on the basis of both monocular and binocular cues. In some trials, these cues could conflict; when this was the case, the haptic feedback that subjects received as they placed the cylinder on the surface was consistent with one but not the other cue. Van Beers et al. estimated the weights respectively assigned to monocular and binocular cues in each trial by monitoring the orientation of the cylinder just before it contacted the surface. They discovered that significant cue reweighting occurred after only about  conflict trials with small but non-negligible changes in weights after each trial (: ). This result constitutes empirical evidence that visual orientation perception is susceptible to synchronic, lowlevel informational penetration (SLIP) by touch.

. Audiovisual SLIP As mentioned above, in the most familiar examples of multisensory integration, vision exerts a dominating influence on other senses. The previous subsection, however, shows that the influence can flow in the other direction, with vision exhibiting susceptibility to change in the face of intersensory conflict with touch. In this subsection, I review evidence that what see can also be influenced by what we hear. In the ‘stream-bounce illusion’ (Sekuler et al. ), audition influences an object’s perceived trajectory in space. Subjects are shown an ambiguous display in which two objects can be perceived either as streaming through each other (without a change in their directions) or as bouncing off of each other (with a reversal in their directions), depending on whether or not a brief sound is heard when they coincide. The nature of the illusory effect, however, isn’t fully understood. One possibility is that the sound modulates motion perception simply because it causes attention to be momentarily withdrawn from the moving targets (Watanabe and Shimojo ). Consistent with this hypothesis it has been found that other transient attention-capturing stimuli, e.g. a visual flash or a brief tap on the subject’s finger, bias the visual system toward the bouncing interpretation of object motion. This speaks against the presence of audiovisual SLIP in the etiology of the illusion. Persuasive evidence for audio-visual SLIP, however, comes from the ‘sound-induced flash illusion’ (Shams et al. ; ). When a black disk is flashed once, but accompanied by two brief beeps, it is visually experienced as having been flashed twice. ‘This phenomenon’, Shams and co-authors suggest, ‘clearly demonstrates that sound can alter the visual percept qualitatively even when there is no ambiguity in the visual stimulus’ (: ). Information originating in the auditory system, it appears, can sometimes influence the phenomenal contents of visual experience.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  Several sources of evidence point to a perceptual, non-cognitive interpretation of this effect. First, the illusion is insensitive to the observer’s beliefs: knowing that the disc has only been flashed once does not eliminate the illusion. Such beliefindependence is often taken to be a hallmark of perception. Second, the effect does not appear to be due to either task difficulty or to a cognitive strategy that involves counting the number of beeps. Third, neuroimaging studies have shown that the illusion is correlated with changes in activity in low-level visual cortical areas, including V (Watkins et al. ; Mishra et al. ). This speaks strongly against a cognitive interpretation of the effect. Last, Rosenthal et al. () found that the illusion is significantly resistant to feedback training in which subjects are apprised about their errors. Together, these findings suggests that audition is genuinely affecting visual phenomenal content by means of SLIP.

. Proprioceptive-visual SLIP Size constancy is a basic feature of visual experience. We see a fire hydrant, for instance, as having roughly the same intrinsic or ‘absolute’ size despite variations in our perspective and other viewing conditions. What is it, however, to perceive an object’s intrinsic size and not merely its size relative to other objects in our field of view? Clearly, it is not enough that there be some proximal stimulus variable whose values correspond to variations in the extent of the object. Perceivers must also make use of a spatial scaling scheme of some kind that assigns intrinsic size meaning to those values.11 In the case of human perception, how is such scaling achieved? If the units in terms of which perceived spatial extents are scaled are not conventional units such as feet or inches (Peacocke ), then how are they scaled? One philosophically influential answer to this question, dating back to Malebranche in the th century, is that an object’s perceived size is scaled to one’s sense of the size of one’s own body. ‘Our sight’, Malebranche writes, ‘does not represent extension to us as it is in itself, but only as it is in relation to our body’ ([–]: I., §; quoted by Simmons ). The purpose of spatial vision isn’t to provide us with information about objects suitable for precise, mathematical calculation, but rather to adapt our actions to the structure of the D environment. In keeping with this purpose, visual experience represents properties such as size, shape, and distance using a bodyrelative scaling scheme that action-planning systems understand. As Simmons puts it, ‘Sensory perception . . . reflects one’s own body in the way it represents objects. This in turn helps to explain why sensory perception is especially suited to helping the mind to direct its body safely through the world’ (: ). The idea that visual perception makes use of an egocentric, body-relative scaling system finds supports in several decades of psychophysical work on size and distance perception. One reliable source of information about spatial layout, for example, is the ‘horizon ratio’ (Sedgwick ; ). Because the horizon line is always at eye level 11

An analogous problem arises with respect to distance perception. See Schwartz (: ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

in the optic array, the height of an object from its base to the point where it visually intersects the horizon is equal in height to the elevation of the perceiver’s eyes above the ground. Hence, the ratio of the object’s height (h) to the height of the perceiver’s eyes (e) is approximately equal to the ratio of the vertical visual angle subtended by the object (H) to the vertical visual angle subtended by the portion of the object below the horizon (E): h/e = H/E In other words, the height of an object, when estimated using the horizon ratio, is scaled in eye-height units. For example, if a tree’s horizon ratio is :, then the tree is approximately five eye-height units tall. There is psychophysical evidence that the visual system makes use of the horizon ratio, among other sources of information, when estimating an object’s height (Mark ; Warren and Whang ; Rogers ) as well as when estimating its egocentric distance (Ooi et al. ) and width (Wraga ).12 If human perceivers do make use of eye-height-scaled information, then it should, in principle, be possible to modify how far and/or how large objects appear in visual experience by altering a subject’s awareness of the size of her own body.13 This prediction is borne out by a recent study of the perceptual effects of the so-called ‘body swap’ illusion. Partial body-ownership illusions are well known. For example, in the rubber-hand illusion (Botvinick and Cohen ), subjects vividly experience tactile sensations as arising in a prosthetic rubber hand when it is stroked in synchrony with strokes applied by the experimenter to the subject’s own, unseen hand. Using a similar visuotactile stimulation technique, however, it is also possible to induce the illusion of owning an entire body distinct from one’s own (Lenggenhager et al. ; Petkova and Ehrsson ; Lenggenhager et al. ). In  study, van der Hoort and colleagues sought to determine whether it is possible (a) to induce illusory ownership of an abnormally large or small artificial body and, if so, (b) whether the induced illusory experience, in turn, influences visual perception of spatial layout. Subjects, lying on their back and looking toward their toes, were presented via a head-mounted display with real-time video of the abdomen and legs of an artificial body, varying between , , and  cm in height. To induce the body-swap illusion, the participant’s body and the artificial body were touched in synchrony for four minutes. This resulted in a powerful body-swap illusion, even when subjects were visually presented with the torso and legs of a small, -cm doll. ‘[M]ost participants’, van der Hoort et al. write, ‘were not aware of the extremely small size of the doll that

12

For a helpful overview, see Bennett (). Warren and Whang () showed that subjects overestimate height and width when effective eyeheight is underestimated. Their experimental paradigm, however, involved use of a false floor that surreptitiously increased the target’s H/E ratio rather manipulation of subjects’ internal representations of body size. 13

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  they felt ownership of. Instead, they experienced themselves to be located in a “giant world” ’ (: ). The experimenters then investigated whether the illusory change in body size also evoked changes in the perceived distances and sizes of external objects shown in the display. Subjects, when asked to walk with their eyes closed to the previously seen location of an object, it was found, walked a significantly longer distance during the small-body condition and a significantly shorter distance during the large-body condition. Verbal estimates of object size were found to be congruent with behavioral measures of distance perception. When subjects experienced themselves as owning a very large body, for example, they saw objects not only as closer but also as smaller than when they experienced themselves as owning either a very small or a normalsized body. In order to exclude the possibility that these effects were due to use of the body seen in the display as a visual, relative size cue, van der Hoort et al. manipulated the strength of the illusion by touching the prosthetic body in or out of synchrony with touches applied to the subject’s own body. Although the size of the body seen in display was the same in both conditions, only the synchronous touch condition lead to significant distortions in visual space perception. In other words, subjects only experienced distances as significantly shorter or longer when they had the illusory experience of owning the prosthetic body. Two points are important. First, perceivers’ representations of their own body size appear to be both highly labile and rapidly modifiable under conditions of intersensory conflict: when synchronous visual and tactile cues disagree, the brain swiftly alters its representation of the body’s metric properties in deference to vision. Second, the findings reviewed here suggest that perceivers use representations of their own body size to calibrate visual estimates of the distances and sizes of objects in the external word, much as Malebranche originally proposed. Taken together, they provide further support for proprioceptive-visual SLIP. In particular, they provide support for the claim that the computations underlying conscious visual space perception make use of nonvisual, proprioceptive information about body metrics as an informational resource.

. Intentions for action penetrate visual size perception A number of studies, beginning with Aglioti et al. (), have reported that the Ebbinghaus illusion has a significantly greater effect on visual awareness than on visually guided grasping. Although the disk surrounded by small circles in the illusion display typically looks about % larger than the disk surrounded by large circles, the increase in maximum grip aperture when reaching for the former disk exhibits a magnitude of only %. This finding in addition to various pieces of neuropsychological evidence has been used to motivate a dual-systems model of visual processing according to which ‘visionfor-perception’ and ‘vision-for-action’ are subserved by functionally and anatomically

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

distinct processing streams in the primate brain (Milner and Goodale /).14 According to proponents of the dual-systems model, the illusion has a different effect on visual awareness than on visually guided grasping because the former makes use of different sources of visual size information than the latter. On this model, how the size of an object appears in conscious vision should not influence grip aperture; conversely, how the size of the object is represented by motor systems that guide grasping should not influence representation of its size in conscious vision. At variance with this idea, however, Vishton et al. () found that the act of reaching for a disk in a three-dimensional version of the Ebbinghaus illusion significantly diminished the magnitude of the effect on subsequent perceptual estimation (.% for perceptual estimation vs. .% for grasping). Strikingly, they also found that when subjects were merely informed prior to engaging in the perceptual estimation task that they would subsequently be required to grasp the disk that appeared larger, the effect of the illusion on visual awareness was again significantly diminished (.% for perceptual estimation vs. .% for grasping). ‘Simply listening to a description of a reaching task’, the experimenters write, ‘seems to affect size perception’ (: ). These findings suggest that visual phenomenal content can be cognitively penetrated: high-level information from outside of the visual system seems to alter the way an object’s size appears. There are different possible mechanisms whereby such penetration might occur. Vishton et al. propose that ‘intending to reach for a target changes how the reacher perceives it’ and that ‘action choice changes the nature of visual size perception’ (: ). But how does action selection have this effect? One possibility (a) is that an abstract, high-level intention to act—either a ‘distal’ or ‘proximal’ intention in the sense of Pacherie —somehow exerts a direct influence on perceptual estimation, say, by changing the relative weightings assigned by vision to sources of depth information such as binocular disparity, vergence, and accommodation. Since size estimation depends, in part, on perceived distance, this could explain the influence of intention on perception. A second possibility (b) is that the effect is brought about via lower-level motor representations that implement the subject’s high-level intention. This would arguably still count as a case of cognitive penetration if the lower-level motor representations carried information from the subject’s high-level intention that influenced relative cue weighting or other visual computations (Wu ). A third possibility (c) looks to elicited motor imagery for the source of penetration. Possibility (c), however, is not entirely distinct from (a) and (b), since there is evidence that internally rehearsing the performance of an action activates representations at all levels in the motor processing hierarchy. A final possibility (d) is that the effect isn’t due to motor representations at all, but rather to the subject’s beliefs about the action she has been requested to perform. Future studies must investigate

14 For different philosophical assessments of the dual systems model, see Briscoe (), Clark (), and Briscoe and Schwenkler (forthcoming).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  which, if any, of these explanations best accounts for the effects that Vishton and his co-authors report.15

 Conclusion This chapter examined two approaches to the question of whether visual phenomenal content can be influenced by information originating outside the visual system. ‘Highroad’ arguments undertake to show that acquiring the capacity to categorize an object as belonging to a certain high-level kind can modify the way the object subsequently appears by causing the relevant kind property to be represented in perceptual experience. Both phenomenological and psychological considerations put forward in support of this conclusion were examined and found to be unconvincing. ‘Low-road’ arguments, by contrast, seek to show that information from outside the visual system can modulate the way an object’s low-level properties visually appear. In the previous section, I presented some compelling psychophysical and neuroscientific evidence for this view. The empirical findings scouted there suggest that visual phenomenal content is not only influenced by information imported from other peripheral input systems, such as audition and touch, but that it can also be penetrated by proprioceptive representations of the size of one’s own body as well as by computationally high-level motor intentions.16

References Aglioti, S., Goodale, M., and DeSouza, J. (). Size contrast illusions deceive the eye but not the hand. Current Biology : –. Balcetis, E., and Dunning, D. (). Wishful seeing: more desired objects are seen as closer. Psychological Science : –. Bayne, T. (). Perception and the reach of phenomenal content. Philosophical Quarterly : –. Bennett, D. (). How the world is measured up in size experience. Philosophy and Phenomenological Research : –. Blaser, E., Sperling, G., and Lu, Z. (). Measuring the amplification of attention. Proceedings of the National Academy of Sciences : –. Botvinick, M., and Cohen, J. (). Rubber hands ‘feel’ touch that eyes see. Nature : . Briscoe, R., and Schwenkler, J. (forthcoming). Conscious vision in action. Cognitive Science.

15 A number of psychologists in what might be called the ‘New New Look’ tradition have recently reported that anticipated motor effort and/or affective states can significantly influence subjects’ perceptions of spatial layout (Proffitt et al. ; Proffitt ; Balcetis and Dunning ). While many of these studies seem to provide further support for cognitive penetration of conscious visual space perception, recent counterstudies (Woods et al. ; Durgin et al. ; Durgin et al. ; Durgin et al. ) suggest that a number of the reported effects may be artifacts of experimental design. 16 I am grateful to Ruth Millikan for thought-provoking discussions of ideas in this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

Bruce, V., Greene, P., and Georgeson, M. (eds) (). Visual Perception: Physiology, Psychology, and Ecology, th edn. London: Psychology Press. Burr, D., and Ross, J. (). A visual sense of number. Current Biology : –. Clark, A. (). Perception, action, and experience: unraveling the golden braid. Neuropsychologia : –. Cohen, M., Cavanagh, P., Chun, M., and Nakayama, K. () The attentional requirements of consciousness. Trends in Cognitive Sciences : –. Durgin, F. (). Texture density adaptation and visual number revisited. Current Biology : R–. Durgin, F., Baird, J., Greenburg, M., Russell, R., Shaughnessy, K., and Waymouth, S. (). Who is being deceived? The experimental demands of wearing a backpack. Psychonomic Bulletin and Review : –. Durgin, F., DeWald, D., Lechich, S., Li, Z., and Ontiveros, Z. (). Action and motivation: measuring perception or strategies? Psychonomic Bulletin and Review : –. Durgin, F., Hajnal, A., Li, Z., Tonge, N., and Stigliani, A. (). Palm boards are not action measures: an alternative to the two-systems theory of geographical slant perception. Acta Psychologica : –. Ernst, M., Banks, M., and Bülthoff, H. (). Touch can change visual slant perception. Nature Neuroscience : –. Fabre-Thorpe, M., Delorme, A., Marlot, C., and Thorpe, S. (). A limit to the speed of processing in ultra-rapid visual categorization of novel natural scenes. Journal of Cognitive Neuroscience : –. Fish, W. (). High-level properties and visual experience. Philosophical Studies : –. Fodor, J. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, J. (). The Mind Doesn’t Work That Way. Cambridge, Mass.: MIT Press. Grill-Spector, K., Henson, R., and Martin, A (). Repetition and the brain: neural models of stimulus-specific effects. Trends in Cognitive Sciences : –. Grill-Spector, K., and Malach, R. (). The human visual cortex. Annual Review of Neuroscience : –. Harris, C. (). Insight or out of sight? Two examples of perceptual plasticity in the human adult. In Charles S. Harris (ed.), Visual Coding and Adaptability, –. Hillsdale, NJ: Erlbaum. Henderson, J., Brockmole, J., Castelhano, M., Mack, M., Fischer, M., Murray, W., and Hill, R. (). Visual saliency does not account for eye movements during visual search in realworld scenes. In R. Van Gompel, M. Fischer, W. Murray, and R. Hill (eds), Eye Movements: A Window on Mind and Brain, –. Amsterdam: Elsevier. Henderson, J., and Hollingworth, A. (). High-level scene perception. Annual Review of Psychology : –. Javadi, A., and Wee, N. (). Cross-category adaptation: objects produce gender adaptation in the perception of faces. PLOS One : e. Koch, C., and Ullman, S. (). Shifts in selective visual attention: towards the underlying neural circuitry. Human Neurobiology : –. Kriegel, U. (). Phenomenal content. Erkenntnis : –. Lenggenhager, B., Mouthon, M., and Blanke, O. (). Spatial aspects of bodily selfconsciousness. Consciousness and Cognition : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  Lenggenhager, B., Tadi, T., Metzinger, T., and Blanke, O. (). Video ergo sum: manipulating bodily self-consciousness. Science : –. Leopold, D. A., Bondar, I. V., and Giese, M. A. (). Norm-based face encoding by single neurons in the monkey inferotemporal cortex. Nature : –. Leopold, D. A., O’Toole, A. J., Vetter, T., and Blanz, V. (). Prototype-referenced shape encoding revealed by high-level aftereffects. Nature Neuroscience : –. Li, F. F., VanRullen, R., Koch, C., and Perona, P. (). Rapid natural scene categorization in the near absence of attention. Proceedings of the National Academy of Sciences of the United States of America : –. Loffler, G., Yourganov, G., Wilkinson, F., and Wilson, H. R. (). fMRI evidence for the neural representation of faces. Nature Neuroscience : –. Mack, A. (). Is the visual world a grand illusion? A response. Journal of Consciousness Studies : –. Mack, A., and Rock, I. (). Inattentional Blindness. Cambridge, Mass.: MIT Press. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research : –. Malebranche, N. ([–]). The Search After Truth, trans. T. Lennon and P. Olscamp. Cambridge: Cambridge University Press. Mark, L. (). Eyeheight-scaled information about affordances. Journal of Experimental Psychology: Human Perception and Performance : –. McKone, E., Kanwisher, N., and Duchaine, B. (). Can generic expertise explain special processing for faces? Trends in Cognitive Sciences : –. Milner, D., and Goodale, M. (/). The Visual Brain in Action, nd edn. Oxford: Oxford University Press. Mishra, J., Martinez, A., Sejnowski, T., and Hillyard, S. (). Early cross-modal interactions in auditory and visual cortex underlie a sound-induced visual illusion. Journal of Neuroscience : –. O’Neil, S., and Webster, M. (). Adaptation and the perception of facial age. Visual Cognition : –. Ooi, T., Wu, B., and He, Z. (). Distance determined by angular declination below the horizon. Nature : –. Pacherie, E. (). The phenomenology of action: a conceptual framework. Cognition : –. Peacocke, C. (). Analogue content. Proceedings of the Aristotelian Society, Supplementary Volume : –. Petkova, V., and Ehrsson, H. (). If I were you: perceptual illusion of body swapping. PLoS One : e. Prinz, J. (). Beyond appearances: the content of sensation and perception. In T. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Clarendon Press. Prinz, J. (). The Conscious Brain. Oxford: Oxford University Press. Proffitt, D. (). Embodied perception and the economy of action. Perspectives on Psychological Science : –. Proffitt, D., Stefanucci, J., Banton, T., and Epstein, W. (). The role of effort in perceived distance. Psychological Science : –. Pylyshyn, Z. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



r. briscoe

Rock, I., and Harris, C. (). Vision and touch. Scientific American : –. Rosenthal, O., Shimojo, S., and Shams, L. (). Sound-induced flash illusion is resistant to feedback training. Brain Topography : –. Rubens, A., and Benson, D. (). Associative visual agnosia. Archives of Neurology : –. Schwartz, R. (). Visual Versions. Cambridge, Mass.: MIT Press. Sedgwick, H. (). The visible horizon: a potential source of visual information for the perception of size and distance. Doctoral dissertation, Cornell University. Dissertation Abstracts International, , B–B. (University Microfilms no. –, ). Sedgwick, H. (). Space perception. In K. Boff, L. Kaufman, and J. Thomas (eds), Sensory Processes and Perception: Handbook of Perception and Human Performance, vol. , .–.. New York: Wiley. Sekuler, R., Sekuler, A., and Lau, R. (). Sound alters visual motion perception. Nature : . Shams, L., Kamitani, Y., and Shimojo, S. (). What you see is what you hear. Nature : . Shams, L., Kamitani, Y., and Shimojo, S. (). Visual illusion induced by sound. Cognitive Brain Research : –. Siegel, S. (). Which properties are represented in perception? In T. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Siegel, S. (). The Contents of Visual Experience. New York: Oxford University Press. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Simmons, A. (). Spatial perception from a Cartesian point of view. Philosophical Topics : –. Stratton, G. (). The spatial harmony of touch and sight. Mind : –. Underwood, G., Foulsham, T, van Loon, E., Humphreys, L., and Bloyce, J. (). Eye movements during scene inspection: a test of the saliency map hypothesis. European Journal of Cognitive Psychology : –. van Beers, R., van Mierlo, C., Smeets, J., and Brenner, E. (). Reweighting visual cues by touch. Journal of Vision (): –. van der Hoort, B., Guterstam, A., and Ehrsson, H. (). Being Barbie: the size of one’s own body determines the perceived size of the world. PLoS ONE : e (–). Vishton, P., Stephens, N., Nelson, L., et al. (). Planning to reach for an object changes how the reacher perceives it. Psychological Science : –. Watanabe, K., and Shimojo, S. (). Attentional modulation in perception of visual motion events. Perception : –. Watkins, S., Shams, L., Josephs, O., and Rees, G. (). Activity in human V follows multisensory perception. Neuroimage : –. Watkins, S., Shams, L., Tanaka, S., Haynes, J., and Rees, G. (). Sound alters activity in human V in association with illusory visual perception. Neuroimage : –. Webster M., Kaping, D., Mizokami, Y., and Duhamel, P. (). Adaptation to natural facial categories. Nature : –. Webster, M., and MacLeod, D. (). Visual adaptation and face perception. Philosophical Transactions of the Royal Society B : –. Webster, M., and MacLin, O. (). Figural aftereffects in the perception of faces. Psychonomic Bulletin and Review : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the reach of phenomenal content  Woods, A., Philbeck, J., and Danoff, J. (). The various perceptions of distance: an alternative view of how effort affects distance judgments. Journal of Experimental Psychology: Human Perception and Performance : –. Wolfe, J. (). Guided search .: a revised model of visual search. Psychonomic Bulletin and Review : –. Wolfe, J., Horowitz, T., Kenner, N., Hyle, M., and Vasan, N. (). How fast can you change your mind? The speed of top-down guidance in visual search. Vision research : –. Wraga, M. (). The role of eye-height in perceiving affordances and object dimensions. Perception and Psychophysics : –. Wu, W. (). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies : –. Xu, H., Dayan, P., Lipkin, R. M., and Qian, N. (). Adaptation across the cortical hierarchy: low-level curve adaptation affects high-level facial-expression judgments. Journal of Neuroscience : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Cognitive Penetration of the Dorsal Visual Stream? Brad Mahon and Wayne Wu

 Introduction The thesis of the cognitive penetration of vision asserts the existence of specific interactions between cognition and vision, namely a type of informational exchange between them. It remains unclear, however, whether cognition affects vision in this way, and even what would count as evidence for cognitive penetration. The issue engages both empirical and philosophical approaches, for on the one hand, the concept of cognitive penetration and the conditions that suffice for it must be clarified, and on the other hand, the claim of cognitive penetration is an empirical claim about interactions within the human mind and brain. As such, progress on the issue requires joint philosophical and empirical exploration in three steps: () philosophical clarification of the concepts needed to state the thesis of cognitive penetration; () articulation of a precise computational model of the visual system in question and of the informational exchange that satisfies the thesis of cognitive penetration; and () experimental work specifically to test such models. We believe that any convincing argument for cognitive penetration requires taking all three steps, a high hurdle. In this chapter, we focus on the first two steps. Is visually guided action cognitively penetrated? Specifically, we focus on the possible penetration of dorsal visual stream computations by semantic/conceptual representations of the function and purpose of the use of objects. Our goal will be to provide a clear definition of cognitive penetration that is conceptually fruitful and empirically tractable. This is necessary for the formulation of an adequate argument for cognitive penetration. We then provide a computational characterization of the dorsal stream’s role in guiding action. In Section , we explicate the concept of cognitive penetration. Section  then discusses the anatomical structure of the dorsal visual stream, while in Section  we examine its functional role in guiding motor action. In Section , we present a prima facie case that concepts, hence cognition, penetrate dorsal visual stream computations.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream 

 Conceptual Issues In this section we shall lay some conceptual groundwork for our discussion, focusing on the notions of modularity, informational encapsulation, semantic information, and concepts. These are theory-laden and controversial issues. Our aim is to be clear on how we shall understand the terms and to use them to raise specific questions about the visuomotor system. Nevertheless, we believe that our understanding of these notions is general enough to garner reasonably widespread acceptance. We can think of discussion of modules as highlighting certain features of processes or capacities, and in this regard, Jerry Fodor’s Modularity of Mind () provides a useful starting place for thinking about the properties of modules of interest to any cognitive psychologist (for recent empirical discussions of the notion, see Barrett and Kurzban ; Mahon and Cantlon ). In his list of such properties, Fodor included: information encapsulation, domain specificity, shallow outputs, dedicated neural realization, a characteristic developmental profile, and being fast, automatic, and innate. He never provided, however, a definition of modularity, but in later work he identified informational encapsulation as the ‘essence’ of modularity (Fodor ): Imagine a computational system with a proprietary . . . database. Imagine that this device operates to map its characteristic inputs onto its characteristic outputs . . . and that, in the course of doing so, its informational resources are restricted to what its proprietary database contains. That is, the system is ‘encapsulated’ with respect to information that is not in its database. . . . That’s what I mean by a module. In my view, it’s informational encapsulation, however achieved, that’s at the heart of modularity. (Fodor : )

We follow Fodor and focus on informational encapsulation, so the core issue concerns the exchange of informational content between processes and psychological domains. A specific issue that has been much discussed concerns whether the visual domain is informationally encapsulated from the cognitive domain, or whether encapsulation fails and vision is thereby cognitively penetrated by the cognitive domain. In general, the question is whether two putative modules, X and Y, exchange information in a particular way. More specifically, consider the computations carried out by X that draw on information from its proprietary database and its inputs. We can then say that X is informationally encapsulated from Y if and only if X cannot use information from Y in its computations. Let us think of Y as a cognitive capacity. Accordingly, the failure of informational encapsulation of X relative to Y entails the cognitive penetrability of X by Y, to use a phrase of Zenon Pylyshyn’s (). So, if the visual system is not informationally encapsulated from certain semantic systems, then the visual system is to an extent cognitively penetrated, or so we shall argue. When we speak of failure of informational encapsulation, we intend a version of cognitive penetration. Failure of informational encapsulation can come in two flavors. On the one hand, it might be the case that Y can informationally penetrate X but does not do so under normal circumstances. So, X is effectively informationally encapsulated relative to Y.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

This allows that Y could, in exceptional circumstances, informationally penetrate X’s computations. Accordingly, we can speak of potential informational penetration. This is not the case that interests us. Rather, we are interested in cases of actual informational penetration of X by Y, namely where the informational resources from Y are drawn on by X for use in X’s computation. Should this occur, then there is a further question as to the scope of this informational interaction: how pervasive is Y’s influence?1 The informational interaction is the crucial notion, and it is here that precise models of X’s computational properties are required. Let us say that X is involved in transformation of certain representations in its proprietary database, including its inputs, so as to compute a certain output type O (e.g. object categorization). Y then informationally penetrates X where the representations of Y are not part of X’s proprietary database and yet are used by X in computing O. One of the challenges in the debate about cognitive penetration is that more often than not, the computational role of X is underspecified; accordingly, concrete models of the penetration of X by Y remain obscure and hence hard to assess. Again, where Y is a cognitive capacity or process that informationally penetrates X, then Y cognitively (informationally) penetrates X. What makes Y relevantly cognitive? One response would be to say that where the information in Y is conceptual, then Y is cognitive, trading on the philosophical connection between concepts and thoughts: concepts are the building blocks of thought. Unfortunately, the notion of concepts is a matter of some heated controversy both in cognitive science and in philosophy, and so clarity is not achieved by merely invoking it. To make both conceptual and empirical progress on this issue, we propose to identify the conceptual with the semantic, as spoken of in some areas of psychology. Thus, we will have cognitive penetration of X by Y when Y encodes semantic representations or content over which X computes. The notion of semantic information is itself a matter of some controversy, but since our aim is to be as clear as possible about questions and proposals, the relevant sense of semantics at issue will always be task-relative. That is to say, we eschew general definitions of what semantics amounts to in respect of psychological processes, and focus instead on the information that is necessary for performing specific tasks. Specifically, in certain well-defined tasks, the relevant information needed for task performance will be operationally identified as semantic. In experimental cognitive science, ‘semantics’ is generally operationalized as the information that mediates 1 Informational encapsulation seems to be logically independent of the other conceptions of modularity that have been promoted by psychologists. For example, if we understand domain specificity to be the restriction of a capacity or process X to specific informational contents, then it seems that whether or not X and Y are informationally encapsulated relative to each other is largely independent of whether or not X and Y are domain-specific. There are, of course, special cases. For example, Y might involve computations over some informational content type A and also informationally penetrate the computations of X. Should X otherwise normally compute over informational content type B where B is in some relevant way distinct from information of type A, then informational penetration might undercut the domain specificity of X for B. Of course, should both X and Y operate only over A, then the failure of informational encapsulation does not undercut the domain specificity of X or Y. Spelling out the logical relations between the properties in Fodor’s list is a worthwhile exercise that we shall set aside.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  the mapping from input to output systems, where input and output systems are modality-specific in terms of both their content and their format (see Caramazza et al. ). The clearest demonstration of this approach to defining ‘semantic’level information comes from cognitive neuropsychological studies of patients with brain damage. The general criterion for positing a semantic-level impairment in such patients is a demonstration that the patient is impaired across multiple modalities of input and multiple modalities of output. So, for instance, while patients with visual agnosia would be impaired for picture naming (name a picture of a hammer), they would not be impaired for naming ‘hammer’ when hearing the sound of a hammer pounding, or touching a hammer, or when presented with a dictionary definition of a hammer. Thus, in the case of visual agnosia, the patient does not have a semanticlevel impairment (namely about hammers) but rather an impairment at the level of visual representations that interface with semantics. The evidence that semantics is not involved is the patient’s normal performance in naming within other modalities (e.g. sound or touch). A semantic-level impairment would manifest as impaired naming across all of the modalities through which the patient could be presented a stimulus of a hammer. Similarly, while patients with language impairments may be impaired for naming (potentially in writing or speaking or both) pictures or sounds or definitions, they would be able to (for instance) draw a reasonable hammer from memory (modulo drawing abilities), or match a hammer with a nail in a matching task. Patients with semantic-level impairments in respect of hammers would fail those tasks. An interesting task that remains contentious as to whether failure in it is indicative of a semantic-level impairment, and to which we will return below, is object use. Finally, a source of positive evidence that a patient may have a semantic-level impairment would be provided by impaired performance for making judgments about the typical attributes of an item (Is a hammer typically held by one or two hands? Are carpenters or mechanics more likely to use a hammer?). Again, it is important across all such tests that failures of a task are not due to uninteresting reasons. For instance, one might show that a patient who fails at naming across a range of inputs can repeat the word ‘hammer’ and read the word ‘hammer’, indicating that the problem is not with articulation. If a patient with a semantic-level impairment fails to be able to draw pictures from memory (draw a hammer), then it is important to show that the patient can copy pictures reasonably, ruling out low-level problems with drawing per se (and so on). The broader point is that by ‘semantic’ we mean to pick out an empirically defined class of processes or information that mediate the mapping of input to output systems. The nature of those processes and information will come into sharper focus as we frame the processes that are specifically relevant to exploring the hypothesis that dorsal stream object-directed actions are cognitively penetrated by semantic-level information. Specifically, the semantic information on which we shall focus concerns the function and the purpose of the use of objects: what they are for and what the goal is in using them.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

The psychological explication of ‘semantic information’ is not disconnected from the notion of concepts as discussed by philosophers. How to link concepts to specific information processes in the brain is a controversial topic, but there does seem to be a plausible link between concepts, characterized as the building blocks of thought, and semantic information in the task-relative sense. We assume that the possession of some concept X implies the presence of certain sorts of semantic information. In particular, concepts of functional objects imply the presence of semantic information about the function and purpose of the use of those objects. Such semantic information is degraded in patients with semantic dementia, and these semantic deficits are revealed in a variety of tasks that plausibly implicate defective cognition in respect of semantics, specifically the use and function of objects like hammers and spatulas. Patients can be unable to categorize these objects, describe their use, or understand sentences about them. Indeed, patients with semantic dementia are often unable to use objects appropriately once they have grasped them (Hodges et al. ). The action deficits seen in semantic dementia patients, who plausibly have degraded concepts concerning the use and function of certain objects, suggests that those concepts—and by implication associated semantic information—play a role in guiding action. An empirical prediction is that if such information is required, then semantic processing is needed to bring that information to bear in action such that disruption of such processing would disrupt action. While there is more empirical work to be done in this area, there is data suggesting this role for semantic information: drawing on semantic processing that is irrelevant to but concurrent with reaching for an object such as a spatula can disrupt appropriate reaching (Creem and Proffitt ). The hypothesis is that semantic processing used in recalling semantically associated words is also normally called upon by the action processing that guides functionally appropriate reaches to manipulable objects. When reaching for such objects, subjects must draw on semantic information about those objects such as representations of their function and the purposes for which they are to be used. Subjects who performed reach-irrelevant semantic processing tended to reach and grasp objects in functionally inappropriate ways (e.g. by the head of a spatula rather than its handle). The central hypothesis to be investigated, then, is that the involvement of concepts in visually guided action amounts to a form of cognitive penetration of visual processing necessary for functionally appropriate action. The empirical upshot is that the informational correlate of such concepts, semantic information, informs such processing. To return to conceptual issues, let the information in Y be necessary for performance of task T, or at least normally used to perform task T. Then if T is of a certain relevant kind, the information in Y will count as semantic. In this chapter, the relevant task will be functionally appropriate use of artifacts, so the relevant semantic information will concern the function and purpose of such objects. The central question can then be stated generally as follows: does X compute over the semantic information from Y, information identified as semantic relative to some task T?

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  In this chapter, the specific question is this: does the dorsal visual stream compute over semantic information from areas coding information about object use in tasks involving the appropriate action on those objects? If the answer is yes, then those areas cognitively penetrate the dorsal stream. Having philosophically explicated the hypothesis, the next two steps are to articulate a specific computational model of cognitive penetration and then to experimentally test it. In the remainder of this chapter, we focus on the computational articulation, necessarily leaving experiments for another time. We shall articulate hypotheses about the computations carried out by the dorsal stream and the sources of information that it draws on to perform those computations.

 What Is the Dorsal Stream? Anatomically, visual information is communicated at the neuronal level to the brain via ganglion cells in the retinas of the eyes. The overwhelming majority of those retinal ganglion cells send axons that synapse in the lateral geniculate nucleus (LGN) of the thalamus (a small-proportion synapse in the mid brain). From the LGN the great majority of forward connections then synapse in V, which is the earliest level of cortical visual processing, located at the pole of occipital cortex. The classic model of the dorsal vs. ventral stream is that visual information bifurcates at V, with a dorsal pathway projecting from V to dorsal occipital cortex, motion-sensitive area MT, and terminating in posterior parietal cortex, and a ventral pathway projecting from V to inferior and lateral temporal-occipital cortex (e.g. Goodale and Milner ; Merigan and Maunsell ; Ungerleider and Mishkin ). It is also the case that the small proportion of retinofugal fibers that synapse in the midbrain project to regions of the dorsal visual pathway (e.g. Lyon et al. ). Computationally, the dorsal pathway analyzes visual information in the service of physically interacting with the world, including getting an effector (e.g. hand) to the right location in space, and if relevant, shaping that effector to a target object (i.e. grasping). Of course, ‘visuomotor’ actions are not limited to the hands, although manual action has been an important testing ground for the dissociation of the ventral and dorsal streams. Any motor action that is visually guided is a visuomotor action, such as eye movements, walking around furniture, or jumping onto a ledge. The dorsal pathway supports spatial analysis of the location of objects in the service of this broad array of visuomotor actions. The ventral pathway subserves identification and semantic analysis, and is the major pathway through which semantic information is derived from visual input. Whereas the dorsal pathway is concerned with the true shape, size, and location of objects in the world, representations within the ventral pathway are largely invariant as regards orientation, size, and distance. For this reason, the ventral pathway is sensitive to contextual effects (e.g. that the moon looks large next to the horizon is a contextual effect that would be a quintessential byproduct of how the ventral stream

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

processes visual information). It is this difference between the ventral and dorsal streams that underlies the claim that the ventral stream, but not the dorsal stream, falls prey to some visual illusions, such as the Tichner Illusion. In this illusion, the context in which a central circle is shown affects its perceived size (a circle appears relatively larger if surrounded by smaller circles, and relatively smaller if surrounded by larger circles). Knowing this does not reduce the illusion—a case of the sort that Fodor and Pylyshyn have appealed to in order to demonstrate the cognitive impenetrability of visual experience. However, if the circles consist of poker chips that can be picked up, then the grip aperture of the thumb and finger can be measured, and it can then be tested whether a grasping action falls prey to the illusion. Experiments suggest that it does not (Aglioti et al. ; but see Carey ; Franz ; Smeets and Brenner ). This result is surprising if we think that action is guided by what we visually experience. Consequently, the grip should reflect the illusion. On its face, this result raises the possibility that the dorsal stream is immune to (at least this) illusion, and hence encapsulated from the information that gives rise to the illusion. Our phenomenological experience is generally thought to be aligned with the ventral pathway; it is an open issue as to whether the dorsal visual pathway is able to provide visual information that we can be aware of. However, we shall not be directly concerned with awareness as a way to divide the two streams. Rather, we emphasize the dorsal/ventral distinction at a computational or information-processing level. For example, it may be reasonable to suppose that while both the dorsal and ventral visual pathways are ‘driven’ by visual input, the dorsal pathway brings fewer assumptions about the world to bear on how it parses that input. Or, while it is not clear that the computations that form the dorsal visual pathway can be engaged in the absence of visual input, ventral visual pathway processes can be engaged by, for instance, visual imagery in the absence of visual input.2 The cleanest demonstrations of the functioning of the dorsal stream would consist of those situations in which visuomotor action is unequivocally ‘unguided’ by semantic information. A dramatic demonstration of what visuomotor behavior can look like in the absence of semantic input is provided by neuropsychological cases of blind sight. Blindsight patients have lesions to primary visual cortex—they are cortically blind; ‘blindsight’ refers to the phenomenon that in some patients, they can continue to make actions that avoid obstacles that ‘they cannot see’ (Goodale and Milner ). This is despite the fact that they lack a geniculostriate (i.e. LGN-V) pathway for the affected regions of the visual field. In one dramatic example of a patient with a confirmed lesion to all of early visual cortex, the patient could walk around obstacles placed in his path (de Gelder et al. ). The performance of such patients 2 The point being that the dorsal stream may be aligned with certain brain regions; but merely demonstrating the activation of such regions is not evidence of ‘dorsal stream’ computations. Such computations are defined in terms of the online computation performed over visual input.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  indicates that non-geniculostriate pathways (retina → midbrain → extrastriate cortex or retina → LGN → extrastriate cortex) are sufficient to support the online calculation of at least very coarse actions (e.g. Lyon et al. ; Schmid et al. ). Spared action in the context of impaired perception can also be observed in healthy participants. We noted above the empirical claim that manual grasping actions can show immunity, at least under some circumstances, to some visual illusions. In another demonstration of the independence of action from perception, Goodale and colleagues () had participants point to a target dot that was presented in the periphery. Participants were instructed that when the dot appeared, they should saccade to its location and then point to it. Unknown to the subjects, on a proportion of trials, the dot’s position would be slightly adjusted during the time when participants’ eyes were in flight from the fixation point to the dot’s (original) location. Participants did not notice these target jumps, and their eye movements automatically corrected via an intermediary landing spot. However, the finger-pointing movements were entirely accurate and updated to the new location of the target. In a subsequent study with a similar paradigm, Desmurget and colleagues () used transcranial magnetic stimulation (TMS) to disrupt neural processing in posterior parietal cortex during the saccade to the dot location. The authors found that TMS to posterior parietal cortex disrupted the in-flight correction of the pointing movement by the contralateral hand (contralateral to the TMS pulse), while ipsilateral pointing was unaffected by the TMS pulse. In some ways, therefore, certain dorsal stream computations are more starkly defined than ventral processes, in that the scope of their inputs is limited to online visual information and the scope of their computations is restricted to adjusting the trajectories of the body through space, in real time. In this way, the dorsal stream is one of the strongest candidates for informational encapsulation. The question is whether dorsal stream computations show cognitive penetration by semantic level information about objects. The semantic information that may be doing the penetration can be generally aligned with the ventral stream, but is not limited anatomically to the cortex that subserves the projection from V through to anterior temporal cortex. Relevant semantic, or semantically interpreted, information may also be represented by anatomical structures not generally considered part of the ventral pathway proper, such as prefrontal cortex. This does not change the conceptual issues at stake, since the relevant issue is whether dorsal stream processes are cognitively, i.e. semantically, penetrated.3 So what does the dorsal visual pathway know about the world? Take the situation in which a subject at a party is confronted with a beer mug and a wine glass on the 3 If prefrontal regions, which are known to be heavily interconnected with ventral temporal cortex as well as premotor cortex (Rizzolatti and Matelli ), are important for interfacing abstract representations of the visual environment with the goals of actions, then it matters whether (putative) penetration of dorsal stream processes is traceable to prefrontal cortex vs. the ventral visual pathway. This is a matter that must be explored further.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

table before her, filled with appropriate liquids. She also has a plate of food and is speaking to people around her. When she reaches for her wine glass, it is a dorsal stream process that calculates the aperture of her grip in flight as well as the velocity profile of that action. However, if the glass were to look slippery and thus require a different hold, that information may not be appreciated by the dorsal visual pathway— the dorsal visual pathway would have to be ‘told’ in one way or another that the glass was slippery. Similarly, if there was a smudge spot on part of the rim that the subject noticed while reaching toward the glass, she might rotate her grasp in flight so a clean part of the rim will come to her lips, without having to adjust her grip once the glass was prehended. Reorientation of the grip to accommodate such information is not a computation that the dorsal visual pathway, at least classically understood, would perform on its own (i.e. over its own propriety database of visual information). Rather, to perform that computation, the dorsal stream must be ‘told’ the relevant information. We shall return to this, as it may present an important test case of whether the dorsal stream is ‘told’ this information in a way that constitutes cognitive penetration. What becomes apparent at this point in the discussion is that we need a theoretically motivated definition of what should count as action vis-à-vis dorsal stream encapsulation. In other words, in the complex action of reaching out, picking up a glass, and taking a sip of wine, which components (or all?) of that complex action are attributed to dorsal stream computations? Identification of those aspects of the action that are attributable to the dorsal stream would then allow us to specifically frame the question of whether those actions are affected by semantic information.

 What Aspect of Complex Object-Directed Action Is Subserved by the Dorsal Stream? Consider again our subject at the party who is confronted with a beer mug and a wine glass on the table before her. Now, if there is to be any motor action in this context, she must select both a target and a response. The first challenge to action, then, is that there are two targets and indeed many possible responses. We can speak of the ‘behavioral space’ available to the subject at this time in terms of the set of input–output (target–response) mappings. Thus, we presume that the subject wants to drink, and so there are two potential targets (beer and wine). On the other hand, our subject could plan to throw one of the objects at a rival, and so it is also clear that for each object there are (at least) two responses: drinking and throwing. Thus far, we have described four available actions, defined by distinct input–output mappings. This yields what we can call a Many–Many Problem, for in normal action, a subject is presented with many inputs and many outputs that identify possible behaviors. The Many–Many Problem is resolved by action when a specific target is identified and acted on in a particular way.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  One version of the problem—what we can call the ‘deliberative version’—is solved when the subject decides what to do, say to drink some beer. So, her intention represents the action, a specific input–output mapping. Our subject must then produce the action, and it is here that the visual system must confront its version of the Many– Many Problem, what we can call the ‘non-deliberative’ Many–Many Problem. The challenge is that the target of action exemplifies many visual features, and to each ‘actionable’ feature, a set of motor responses can be made that would instantiate the intended action. There are, after all, many ways to hold a mug to drink from it. So, we can now imagine a more fine-grained behavioral space that maps different ways to implement the action type, drinking from a beer mug. While the deliberative Problem is solved by forming an intention, the non-deliberative Problem is solved by producing an actual action. Note that solving the non-deliberative Problem is a temporally extended process that extends to the completion of the action. In the action we are considering, ultimately drinking some beer, we can highlight two relevant phases once the action begins: (a) reaching for the target to manipulate it and (b) manipulating the target once it has been grasped. Thus, the non-deliberative Many–Many Problem can first be divided (at least empirically) into two separate actions. The empirical evidence motivating this distinction comes from patients with brain lesions that selectively compromise different aspects of complex actions, such as reaching out to take a drink from a glass. The reach-to-grasp action itself may be dissociable on empirical and conceptual grounds into grip scaling and targeting the object in visual space (a question we shall return to). In a series of studies, Goodale, Milner, and colleagues showed that human patients with selective lesions to bilateral lateral occipital regions can have a dense visual agnosia but intact reaching and grasping of objects. In other words, visual information could not be used to derive semantic information from visual input, but was available to guide action. In contrast, patients with lesions to posterior parietal regions can exhibit impairments for targeting objects with a reaching movement and/or scaling their grip aperture appropriately during object grasping—referred to as ‘optic ataxia.’ Patients with optic ataxia have normal perception and can extract semantic information from objects normally (i.e. their ventral stream is intact; see also Pisella et al. ). An interesting aspect of the behavior of patients with optic ataxia is that they can execute the complex object-associated actions associated with the function of an object, once the object is in hand. In other words, the impairment in optic ataxia concerns the reach-to-grasp component of the action. Stored knowledge—that for instance, hammers are used for pounding and pounding is accomplished with a swinging motion—can be intact in such patients. The other side of this potential dissociation is also observed: lesions to the left inferior lobule are classically associated with object apraxia. Patients with apraxia of object use may be able to reach out and grasp objects fine; the trajectory of their reaching movement and their grip scaling can

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

be normal, both processes supported by posterior/superior parietal regions, not the inferior parietal lobule. However, once they have the object in hand they are impaired at manipulating the object in the appropriate way to fulfill its function. Thus, they may reach out and grasp the hammer fluidly and without error, but then not be able to pantomime the swinging motion that is associated with using a hammer to pound nails. The failure in these patients is not because they do not appreciate what the object is. In fact, in many cases patients with apraxia of object use can name the same objects they fail to correctly use. Similarly, such patients may be able to retrieve function knowledge—such as the knowledge that hammers are used to pound nails—while still being unable to actually physically manipulate the hammer to accomplish pounding (for reviews of the clinical evidence, see Johnson-Frey ; Mahon and Caramazza ; Pisella et al. ; Rothi et al. ). The double dissociation between reaching and grasping on the one hand (impaired in optic ataxia) and object use (impaired in apraxia) on the other is empirical motivation for the fractionation of a complex action (such as reaching out to take a drink of beer) into component parts. It is clear that the computations underlying reach-tograsp actions are supported by the classically defined dorsal stream. However, it is not at all clear that complex object-associated manipulations could be supported by the dorsal stream. This is because such manipulations are not given by the visual input— they are stored knowledge that must be accessed. For instance, the knowledge of how pliers are used, or how a wrench is used, once the object is in hand is information that is stored. Rothi, Heilman, and colleagues in their influential model of apraxia have analogized those representations to lexical representations of words (e.g. Rothi et al. ). The implication then is that, on a conservative analysis, only reach-to-grasp components of actions are dorsal stream computations. The question that we address herein is whether such computations are cognitively penetrated.4

 Is There Cognitive Penetration of Dorsal Stream Computations? As we saw in the previous sections, dorsal stream computations are geared towards the generation of appropriate motor actions, and in the case of our party example, specifically the reach-to-grasp component of certain motor actions. The question we now raise is whether there is cognitive penetration of dorsal stream computations by semantic information coded outside the dorsal stream, specifically by semantic information concerning the function and use of objects. Our question of cognitive penetration concerns whether semantic information influences solving the non-deliberative Many–Many Problem focused on the reach4 An earlier argument in this direction, but phrased in terms of the conceptual content of vision, was given in Wu (), where it was argued that concepts are needed to organize attention in order to serve the appropriate use of artifacts.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  to-grasp component. Specifically, is semantic information needed to influence dorsal stream computations so as to adequately solve this Problem? Whether this is so is ultimately an empirical question, but let us first flesh out the space of possible ways in which dorsal stream computations could interface with semantic information in order to solve the non-deliberative Problem. We can think of three places that semantic information can influence dorsal stream computations: (a) at the level of input into the dorsal stream; (b) at the level of output from the dorsal stream; and (c) at the level of internal computations within the dorsal stream. It is only the third case where the influence would count as cognitive penetration. Let us consider the first possibility: that semantic information influences the input into the dorsal stream. One model that suggests this is Goodale and Milner’s proposal for how semantic information in the ventral stream affects dorsal stream computations. They write (: ): Although the dorsal and ventral streams have their own sophisticated and specialized languages, they both still retain contact with the more basic language of the retina. Rather like the human operator can instruct the robot via the two-dimensional optical array behind the lens of the robot’s camera, so the ventral stream can instruct the dorsal stream via the common retinotopic map in early visual areas.

On this model, it seems that semantic information from the ventral stream is ‘fed back’ to early visual areas, presumably to identify certain regions in retinotopic maps for further processing by the dorsal stream. The idea would not be that semantic information is literally transposed or communicated to early visual regions, but that the semantic information would increase, for instance, the gain of neurons coding information in a particular region of the visual field that is processing the information about the selected object. Goodale and Milner invoke the metaphor of an operator of a robot pointing to a region in a map of the terrain around the robot to get the robot to head to that location. Analogously, when our subject decides to drink from the mug, their model holds that the ventral stream analyzes the properties of the mug as an artifact used to drink, and perhaps ‘highlights’ the relevant region in retinotopic maps corresponding to the retinal projection of the mug’s handle. In this way, the dorsal stream focuses on information from that region in space. We might think of this as a form of attention that helps to select relevant information for further processing. Whether this is how semantic information in the ventral stream influences dorsal stream computations—namely via the mediation of early visual areas affecting input into the dorsal stream—is an open empirical question. What we wish to highlight is that this does not seem to exhaust the role that the ventral stream may have in ‘telling’ the dorsal stream about semantically interpreted or behaviorally relevant aspects of objects. For example, even if part of the role of semantic information is to aid selection of locations as spatial targets for action by highlighting regions of retinotopic maps, this still leaves underspecified what action would be relevant within those regions. For example, as the Many–Many Problem demonstrates, for any given target there

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

are a variety of possible responses that are available. In the case of the handle on the mug, there are many possible ways to grip it, of which only a subset is appropriate to conventions for the object’s use. Put another way, given the two components of action in our example, reaching and grasping, we can see Goodale and Milner’s model as showing how semantics could aid reaching (i.e. targeting) by highlighting spatial locations. It does not, however, show how the specific grasp is selected by the dorsal stream, something which presumably must also be sensitive to the function of the targeted object and the intended purposes in using that object, and hence to semantic information regarding that object. Accordingly, even if Goodale and Milner’s model is correct, semantic information must play a role in selecting appropriate action in response to visual information. This leaves the other two points of influence: outputs and internal computations of the dorsal stream. Again, only the latter would count as cognitive penetration. Consider the possibility that semantics influences the output of the dorsal stream. A role for semantics in constraining the outputs of dorsal processing is perhaps easiest to conceptualize if we think of the dorsal stream as providing a set of action plans. Put in terms of the Many–Many Problem, we can think of the dorsal stream as constraining the behavioral options, but not to the extent that a single behavioral path is chosen. The constraints expressed over the multiple behavioral options may be biomechanical constraints of the body together with the physical (i.e. semantically uninterpreted) constraints about the shapes, sizes, and distances of objects. This yields a set of action plans that could be run given these biomechanical and physical constraints. On this model, further selection of the plans produced by the dorsal stream must be implemented to yield a specific input-output mapping. At this point, post-dorsal stream computation, appropriate action selection (input– output mapping) to implement a specific plan might require semantic information, say reaching for the handle of the mug in the right way. Still, even if the output of the dorsal stream is multiple in this way—a possibility that we leave open—one might think that it must still exhibit selectivity in that the initial behavioral space available to the agent is very large. It would seem imprudent for the dorsal stream to make its initial selections in the absence of reducing the Many–Many options without guidance by the relevant semantics. Thus, even if the dorsal stream outputs multiple behavioral options that are consistent with the biomechanical and physical constraints of the action context, if the dorsal stream is not outputting all possible actions consistent with these constraints, then dorsal stream computations are influenced by semantics. Otherwise, the number of possible actions would be overly large to allow for efficient action selection following dorsal stream computation. After all, qua a physical part, a handle can be acted on in many different ways, but there are fewer appropriate ways to act on a handle qua functional part. We saw above that the actual manipulation of the object is typically not something that is mediated by the fast visuomotor processes of the dorsal visual system. However, reach-to-grasp actions are classic dorsal stream computations. The potentially critical

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  test cases, then, are situations when an object is grasped in a way that anticipates its eventual use. Thus, if a hammer is on the table but oriented such that the handle is pointed away, then the appropriate reach-to-grasp movement would not be the easiest or shortest—the shortest would be to pick up the hammer by the head, or to pick it up with a power grasp with the head facing down as opposed to up. But if one were picking up the same hammer to hand to someone else, then it may be functionally appropriate to grasp the hammer by its head (consider the way one grasps a knife to cut vs. to hand to someone). The same situation is presented when picking up a wine glass that has a smudge on it that one wants to avoid with one’s lips. Compelling data relevant to this point are also provided by patients with object form agnosia. Recall that such patients are unimpaired for grasping objects. However, such patients show dramatic impairments for grasping objects in a way that anticipates the correct use of the object. In other words, the reach-to-grasp actions are well-formed actions, and the grip aperture and orientation are well calibrated to the part of the object that is grasped. But a patient such as the well-studied visual object agnosia patient, DF, would be as likely to pick up a hammer upside down or by the head as by the handle in the correct orientation. Similarly, in grasping a wine glass lying on its side, patients like DF would be as likely to pick it up initially with the cup facing up as with it facing down. Such patients would also be impaired for integrating information about the surface properties of an object (such as whether the handle was slippery) into their initial grasp. We have seen above that the dorsal stream has very definite limitations on the types of information that it can represent. The question is: where does the dorsal stream get the information that eventually constrains action output? Does information about the part of the rim of the glass that is smudged, or that the glass looks slippery, constrain action internal to the computations of the dorsal stream? We think that these types of considerations fit more naturally with the hypothesis that semantic information constrains the dorsal stream’s computations, rather than operating only over the inputs and outputs of the dorsal stream. If this hypothesis were true, it would open up the possibility for cognitive penetration of the dorsal stream, via semantic information influencing dorsal stream computations. We suggest the following sufficient condition for cognitive penetration: Y semantically penetrates X if semantic information encoded in Y is directly transmitted to X such that X computes over this information to generate its standard output. Thus, we can think of Y as any non-dorsal stream region that encodes action-relevant semantic information, and X as the dorsal stream. Specifically, if the function of the dorsal stream is to carry out computations that lead to reach-to-grasp actions of the right sort, then we can think of this as a computation that reduces the behavioral options presented in the Many–Many Problem. One way to think of this is that the dorsal stream selects a behavioral path through the available behavioral space, and the role of semantics is to aid in this selection, in addition to the standing biomechanical

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

and physical constraints. Dorsal stream computations would be, then, by hypothesis, sensitive to semantic information: they select an input–output map (or mappings) given relevant semantic information. Specifically, the reach and grasp identified by the dorsal stream is appropriate to the proper use of the object in question, say a beer mug. We can imagine the semantic influence on the dorsal stream in two ways, roughly via specification of the goal and via more direct interactions with the ventral stream. Let us begin with the influence of goals on action selection. Again, we emphasize that what counts as the correct model remains an open empirical question. Our goal is to delineate, at least partly, the relevant space of possibilities. In the case of the influence of goals on action selection, recall the deliberative and non-deliberative Many–Many Problems. We can think of the deliberative Problem, one that is solved by making a decision or forming an intention, as specifying the end point of action. At one level, this specification is abstract: for our subject at the party, it is to drink beer from a specific mug. We can imagine the influence of the representation of a mug invoked in the intention as a semantic informational constraint that affects action-relevant computations. It might delineate relevant action plans by the motor system such as specifying the endpoint of action, as it is the endpoint of the action that directly ‘interfaces’ with the intention. For instance, it is the final position that one wants the hammer to be in after grasping it that will depend on whether one intends to use it or hand it to someone else. The relevant aspects of the motor system are guided by the goals set by intention: if the goal is to drink from that mug, then the hand should be shaped on the handle in such-and-such way, namely in a drinking-appropriate grip. This can then provide a target for the dorsal stream: if the hand should land on this handle in such-and-such way, the dorsal stream must select a trajectory and grip that will achieve this endpoint, given the current layout of the world, the starting position of the hand, and relevant biomechanical constraints. In this way, the solution to the non-deliberative Many–Many Problem is constrained by semantics, and provides a way for semantics to constrain dorsal stream computations (for arguments that intentions do cognitively penetrate visual computations in the parietal cortex, specifically in the case of visual spatial constancy, see Wu ). A second way that semantic information could influence the dorsal stream is via the ventral stream. The relevance of the ventral stream to cognitive penetration is illustrated via the link between semantic dementia and semantic information encoded in the ventral stream. Early in semantic dementia, there are changes in the anterior medial temporal lobe (in addition to the frontal and parietal lobe). As we noted earlier, the symptoms observed in semantic dementia affect cognition of the function of artifacts, and it is the underlying semantic information that is relevant as a source of influence on the dorsal stream. For instance, consider the situation in which the mug must be grasped by its handle in a particular orientation in order to be able to drink from the mug. The identification of the handle (as such) is a process that would be mediated by ventral and not dorsal visual analyses. We have discussed how one

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  way in which the dorsal stream could be cued to the handle as the target of the action would be by ‘highlighting’ the region of space (represented in the early visual cortex) that corresponds to the handle. It is not clear, however, that the notion of a ‘handle’ as a behaviorally relevant affordance is something that can survive translation into the decidedly ‘asemantic’ language of the retina. If it cannot, then the expectation would be that the ventral representation of the handle (as such) is communicated directly to dorsal stream computations, such that the trajectory and grip scaling necessary to bring the hand to the handle in the right way can be computed given the physical and biomechanical information that the dorsal stream has available to it. It is important to note that the anatomical and functional connectivity that these hypotheses would suggest is at least plausibly demonstrated; or rather, there certainly are no data that rule out such connectivity, and there are data that could be taken to indicate the kinds of connectivity that would be expected according to those hypotheses. Specifically, it is known that the ventral temporal cortex is heavily interconnected with the lateral prefrontal cortex, which is in turn heavily interconnected with premotor cortex, which is interconnected with anterior intraparietal sulcus, aIPS (Rizzolatti and Matelli ). Thus, the pathway from semantics and intentions to motor planning stages, and the region of parietal cortex known to be critical for grip scaling (aIPS; e.g. Binkofski et al. ) is well established. There is also known to be connectivity, both functional (Mahon et al. ; ; Noppeney et al. ) as well as anatomical (e.g. Rushworth et al. ) between the relevant regions of temporal and parietal cortex, thus providing a reasonable pathway for the influence of ventral stream information on dorsal stream computations. Further work will be needed to delineate the relevant anatomical connectivity between temporal and parietal cortex as well as overlaying this on work demonstrating functional connectivity. Overall, we think there is a good prima facie case to be made for semantic influence on dorsal stream computations needed to specify reach-to-grasp movements. Such computations involve constraining behavioral possibilities in light of the nondeliberative Many–Many Problem. Should semantic information, whether mediated by the subject’s goals or via semantic information in the ventral stream, be brought to the dorsal stream to aid its solving the Many–Many Problem, this would count as an important and striking case of the failure of informational encapsulation of part of the visual system. Accordingly, there would be one sense in which the visual system, or at least the dorsal stream, is not encapsulated with respect to semantic information about object function and the purposes to which the object will be put to use. As we noted earlier, as such semantic information is plausibly deployed in judgments about object, it points to conceptual information and, given the tie between concepts and thought, to the failure of encapsulation of the dorsal stream from thought. That is, there is a prima facie case to be made that the dorsal stream is cognitively penetrated. Given our earlier point that the existence of cognitive penetration will likely be established with empirical inferences that result from careful experiments, there is no philosophical argument that establishes the existence of dorsal stream penetration

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



b. mahon and w. wu

by semantic information. Rather, the role of philosophical analysis has been to make clear the possibilities and to formulate the conditions for cognitive penetration. Computational models then provide a picture of what the mechanism of penetration might be, at least at an abstract level. To that extent, and at least in visually guided action, we are closer to assessing the actuality of the cognitive penetration of vision. The next, critical steps are to further elaborate the underlying models and to carry out experiments to test them. We believe that the fruitfulness of cognitive penetration as a matter for psychology and for much of philosophy of mind depends on researchers carrying out such experiments on the basis of the type of analysis that we have undertaken here. Cognition might very well penetrate vision. We hope to have taken the first two steps (of three) to establishing this.

References Aglioti, S., DeSouza, J. F. X., and Goodale, M. A. (). Size-contrast illusions deceive the eye but not the hand. Current Biology : –. Barrett, H. C., and Kurzban, R. (). Modularity in cognition: framing the debate. Psychological Review : –. Binkofski, F., Dohle, C., Posse, S., et al. (). Human anterior intraparietal area subserves prehension: a combined lesion and functional MRI activation study. Neurology : –. Caramazza, A., Hillis, A. E., Rapp, B. C., and Romani, C. (). The multiple semantics hypothesis: multiple confusions? Cognitive Neuropsychology : –. Carey, D. P. (). Do action systems resist visual illusions? Trends in Cognitive Sciences (): –. Creem, S. H., and Proffitt, D. R. (). Grasping objects by their handles: a necessary interaction between cognition and action. Journal of Experimental Psychology: Human Perception and Performance : –. De Gelder, B., Tamietto, M., van Boxtel, G., Goebel, R., Sahraie, A., van den Stock, J., Stienen, B. M. C., Weiskrantz, L., and Pegnas, A. (). Intact navigation skills after bilateral loss of striate cortex. Current Biology : R. Desmurget, M., Epstein, C. M., Turner, R. S., Prablanc, C., Alexander, G. E., and Grafton, S. T. (). Role of the posterior parietal cortex in updating reaching movements to a visual target. Nature Neuroscience : –. Fodor, Jerry A. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, Jerry (). The Mind Doesn’t Work That Way: The Scope and Limits of Computational Psychology. Cambridge, Mass.: MIT Press. Franz, V. H. (). Action does not resist visual illusions. Trends in Cognitive Sciences : –. Goodale, M. A., and Milner, A. D. (). Separate visual pathways for perception and action. Trends in Neurosciences : –. Goodale, M. A., and Milner, A. D. (). Sight Unseen: An Exploration of Conscious and Unconscious Vision. Oxford: Oxford University Press. Goodale, M. A., Pelisson, D., and Prablanc, C. (). Large adjustments in visually guided reaching do not depend on vision of the hand or perception of target displacement. Nature. : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

the dorsal visual stream  Hodges, J. R., Bozeat, S., Lambon Ralph, M. A., Patterson, K., and Spatt, J. (). The role of conceptual knowledge in object use: evidence from semantic dementia. Brain : –. Johnson-Frey, S. H. (). The neural bases of complex tool use in humans. Trends in Cognitive Sciences : –. Lyon, D. C., Nassi, J. J., and Callaway, E. M. (). A disynaptic relay from superior colliculus to dorsal stream visual cortex in macaque monkey. Neuron : –. Mahon, B. Z., and Cantlon, J. F. (). The specialization of function: cognitive and neural perspectives. Cognitive Neuropsychology (–): –. Mahon, B. Z., and Caramazza, A. (). A critical look at the embodied cognition hypothesis and a new proposal for grounding conceptual content. Journal of Physiology (Paris) : –. Mahon, B. Z., Kumar, N., and Almeida, J. (). Spatial frequency tuning reveals interactions between the dorsal and ventral visual systems. Journal of Cognitive Neuroscience : –. Mahon, B. Z., Milleville, S., Negri, G. A. L., Rumiati, R. I., Caramazza, A., and Martin, A. (). Action-related properties of objects shape object representations in the ventral stream. Neuron (): –. Martin, A., Ungerleider, L. G., and Haxby, J. V. (). Category specificity and the brain: the sensory/motor model of semantic representations of objects. In M. S. Gazzaniga (ed.), Higher Cognitive Functions: The New Cognitive Neurosciences, –. Cambridge, Mass.: MIT Press. Merigan, W. H., and Maunsell, J. H. (). How parallel are the primate visual pathways? Annual Review of Neuroscience : –. Noppeney, U., Price, C. J., Penny, W. D., and Friston, K. J. (). Two distinct neural mechanisms for category-selective responses. Cerebral Cortex : –. Pisella, L., Binkofski, F., Lasek, K., Toni, I., and Rossetti, Y. (). No double dissociation between optic ataxia and visual agnosia: multiple sub-streams for multiple visuo-manual integrations. Neuropsychologia : –. Pylyshyn, Z. W. (). Computation and Cognition. Cambridge, Mass.: MIT Press. Rizzolatti, G., and Matelli, M. (). Two different streams form the dorsal visual system: anatomy and functions. Experimental Brain Research : –. Rothi, L. J. G., Ochipa, C., and Heilman, K. M. (). A cognitive neuropsychological model of limb praxis. Cognitive Neuropsychology : –. Rushworth, M. F. S., Behrens, T. E. J., and Johansen-Berg, H. (). Connection patterns distinguish  regions of human parietal cortex. Cerebral Cortex : –. Schmid, M. C., Mrowka, S. W., Turchi, J., Saunders, R. C., Wilke, M., Peters, A. J., Ye, F. Q., and Leopold, D. A. (). Blindsight depends on the lateral geniculate nucleus. Nature : –. Smeets, J. B. J., and Brenner, E. ().  years of illusions. Journal of Experimental Psychology: Human Perception and Performance : –. Ungerleider, L. G., and Mishkin, M. (). Two cortical visual systems. In D. J. Ingle, M. A. Goodale, and R. J. W. Mansfield (eds), Analysis of Visual Behavior, –. Cambridge, Mass.: MIT Press. Wu, W. (). Visual attention, conceptual content and doing it right. Mind : –. Wu, W. (). Visual spatial constancy and modularity: does intention penetrate vision? Philosophical Studies : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Attention and Cognitive Penetration Christopher Mole

 Kuhnian Foundations In chapter  of The Structure of Scientific Revolutions (), Thomas Kuhn identified ‘the gradual and simultaneous emergence of both observational and conceptual recognition’ as being a characteristic feature of ‘all discoveries from which new phenomena emerge’ (p. ). He then noted, as a coda to the mainly historical line of argument that he had developed in support of this point, that ‘there is even evidence that these same characteristics are built into the nature of the perceptual process itself ’. Kuhn was not the first philosopher to suggest that cognition might exert an influence on perception, nor even the first philosopher of science to make that suggestion; but it was he who first put such claims on the empirical foundations provided by cognitive psychology, and who recognized that there was a distinct question to be asked here—a question that is neither entirely epistemological nor entirely phenomenological—about whether this influence of cognition is ‘built into the nature of the perceptual processing itself ’. By emphasizing the way in which the underlying psychological processes are built, he made an important contribution to one aspect of the philosophical debate about the cognitive penetrability of perception. The influence of that contribution continues to be felt. When claims about the cognitive influences on perception had figured previously in the philosophy of science—as they had in R. Norwood Hanson’s  book, Patterns of Discovery—they had not been related in the same way to empirical hypotheses about the structures of perception’s underlying processing. They had, instead, been made purely on the basis of phenomenological considerations, such as those pertaining to our experience of such ambiguous figures as the Necker cube and the duck/rabbit. Considerations of that sort were prominent in philosophy at the time when Hanson and Kuhn were writing, thanks in part to their appearance in Wittgenstein’s Brown Book (Wittgenstein ) (which had finally come to be published in the same year as Hanson’s Patterns). Kuhn had his own Wittgensteinian sympathies, and he

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  invokes the phenomenology of duck/rabbit cases himself (in chapter  of Structure) when the topic of cognitive penetrability returns to his discussion. He did see the importance of such cases, but did not take their phenomenology to be adequate for settling questions about cognition’s influence on perceptual processing. Nor did he take those questions about perceptual processing to be adequate for settling his own questions about scientific epistemology. He writes: Yet, though psychological experiments are suggestive, they cannot, in the nature of the case, be more than that. They do display characteristics of perceptions that could be central to scientific development, but they do not demonstrate that the careful and controlled observation exercised by the research scientist at all partakes of those characteristics. (Kuhn : )

By recognizing that certain questions about cognitive penetrability are distinct from his own epistemological questions, and from any questions that are purely phenomenological, and by placing those questions on an empirical footing (which we shall examine below), Kuhn identified cognitive penetrability as a topic that is proper to the science of cognitive psychology. This, at the time when he was writing, was a significant move. The science of cognitive psychology was then in the process of freeing itself from the constraints of behaviourism, and was establishing its credentials as a scientific discipline, distinct from phenomenology and from neurology, in which one could investigate the information-processing that underpins the mind. ∗∗∗

I do not think that it is of merely historical interest to note that our current approach to the cognitive penetrability of perception originated in this particular historical context. I shall be arguing below that our thinking about penetrability continues to be influenced by ideas that enjoyed currency in that context. I shall be suggesting that the influence of these ideas has, in some ways, been unhelpful. It has been particularly unhelpful in leading us to take an unwarrantedly dismissive attitude to cases in which the influence of cognition on perception is mediated by the mechanisms of attention. Such attention-mediated cases should not be dismissed. The usual reasons for dismissing them depend on psychological hypotheses that were once popular, but that have long been superceded. The cases of attention-mediated cognitive influence, as we understand them now, do demonstrate an epistemologically significant way in which the mechanisms that are responsible for our thinking and judging are entwined with the mechanisms that are responsible for our perception of the world. They do reveal the sort of influence that theorists of cognitive penetration should take seriously.

 Overt and Covert Attention The evidence that Kuhn cited, as providing ‘a wonderfully simple and cogent schema for the process of scientific discovery’ (: ), came from an experiment that

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

had been conducted in  by Jerome Bruner and Leo Postman. That experiment is now regarded as a classic of cognitive psychology, but at the time when Kuhn was writing he judged that it ‘deserves to be far better known outside the trade’ (p. ). Bruner and Postman presented their participants, via a tachistoscope, with a sequence of playing cards, some of which had been anomalously coloured, so that a six of spades might be coloured red and a five of diamonds black. At sufficiently short presentation times, the participants failed to report having had a clear experience of these colourings. In some cases they identified the cards as having their normal colours. In other cases they reported that the anomalous cards had their normal colours, but had them in some compromised way, reporting them as being ‘black and red mixed’, or ‘rusty black’, or ‘black with red edges’ (Bruner and Postman : ). Subsequent work has suggested that these ‘compromise responses’ may have been the result of some peculiarities in Bruner and Postman’s procedure. But even if we set the compromise responses to one side, it is a well-replicated result that, while normal cards are reported accurately, people do sometimes report having seen anomalously black cards as red, and anomalously red cards as black, when these are presented for a brief enough duration (see Kempen et al. ). This, although it is subject to divergent interpretations, can do most of the work that Kuhn wanted Bruner and Postman’s experiment to do. In introducing their experiment, Bruner and Postman do not themselves describe it as demonstrating a surprising result, with potential implications for the epistemology of science. They instead introduce it merely as an illustration of ‘a truism worth repeating’: The perceptual effect of a stimulus is necessarily dependent upon the set or expectancy of the organism. (Bruner and Postman : )

That claim may indeed be a truism when it is properly understood, but framing the point with reference to the ‘set or expectancy of the organism’ does little to advertise its commonsense appeal. Perceptual effects that are owing to what Bruner and Postman call ‘set or expectancy’ can more intuitively be described as effects of attention. The more recent empirical literature follows commonsense psychology in describing such effects in those terms (see e.g. Allport et al. ), but Bruner and Postman preferred to avoid the vocabulary of attention altogether. Kuhn himself says very little about attention in his treatment of their work. In avoiding the vocabulary of attention, Bruner and Postman were typical of psychologists writing in the s and s. It was only later, at around the time when Kuhn was writing Structure, that attention became properly established as a topic of psychological research. The first empirical evidence for cognitive effects on perception was therefore developed in the absence of any well-articulated theory of the role that attention might play in the production of those effects. And the first philosophical

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  discussions of cognitive penetrability were not informed by any psychological theory of attention’s nature or role. ∗∗∗

By the time that Kuhn reminded us of the Bruner and Postman experiments, in his  postscript to The Structure of Scientific Revolutions (reprinted in Kuhn ), attention had become established as a central topic of research in cognitive psychology. Although some cognitive psychologists still preferred to avoid the word ‘attention’ when framing their positions (see e.g. Broadbent : ), there were, by then, high-profile debates between rival accounts of the mechanisms by which attention operates. These debates were a major contribution to the fresh crop of psychological theory-building that had sprung up, like Bruner’s own ‘New Look’ approach to the psychology of perception, following the decline of behaviourism’s positivism-inspired conception of the way in which the science of psychology should operate. These two developments—the decline of behaviourism and the rehabilitation of attention as a topic of enquiry—were closely related (see Mole  for some discussion). In order for the study of attention to play its role in bringing about the overthrow of behaviourism, it was necessary to distinguish between attention’s overt and covert forms. The relationship between the mechanisms that are responsible for these two forms of attention continues to be a topic of investigation (see Armstrong  for a summary of some recent work), but the basic distinction between overt and covert attention has always been an easy one to draw, and to operationalize: Your attention to some item in your environment is overt if it is implemented by the bodily behaviour of pointing your eyes or ears in the direction of that item, or by some other bodily orientation of the organs by which you sense it (as when taking a lick, or reaching out to have a feel). Your attention is covert when a shift in the direction of that attention makes no such bodily difference, as when your attention is drawn by the salience of the stimuli in some psychological experiment, despite the fact that you succeeded in following the experimenter’s instruction to keep your eyes fixed on a central point; or as when your attention shifts from the person with whom you have been speaking to the gossip that is being exchanged on the other side of the room. Such shifts of attention can happen, as the early cognitive psychologists proved, without any reorientation of your eyes and ears (Cherry ). They can make a considerable difference to the way in which the scene is processed. When they do, attention is paid covertly. By establishing that covert attention really exists, and by providing a methodology by which the shifts of covert attention could be rigorously studied (at least for auditory stimuli), the early cognitive psychologists established that a theory of overt attention behaviours could not be adequate for explaining all attention-related phenomena. They thereby showed it to be essential for a theory of attention to break from the tradition of behaviourism. By making that break (and doing so at just the time when Chomsky was ejecting behaviourism from the study of language), they contributed to the foundations of what we now call the ‘cognitive revolution’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

Although it was crucial to these early cognitive psychologists to identify covert attention as a phenomenon distinct from overt orienting, it was also quite natural for them to develop hypotheses in which the covert forms of attention were understood to be implemented by internal orientation mechanisms, of a sort that are somewhat analogous to the bodily mechanisms that are responsible for attention’s overt instances. In overt attention a creature orients its sense organs to some subset of the several things that are present in its environment; in covert attention, it was natural for the early cognitive psychologists to suppose, the creature orients its internal processing resources to some subset of the information that its sensory organs have transduced. (See Broadbent  for an influential theory along these lines.)

 The Dismissive Attitude For as long as we think of covert attention in the way that these early psychologists found natural—for as long, that is, as we think of it as being implemented by a mechanism of internal resource orientation, analogous to the overt mechanism of pointing one’s sensory transducers in an appropriate direction—we will seem to be warranted in dismissing the attention-mediated influence of cognition on perception as being irrelevant to the question of whether the mechanisms of perception are cognitively penetrated. That dismissive attitude to attention-mediated cases is well represented in the current philosophical literature. Fiona Macpherson, in her recent work on the cognitive penetration of colour experience, gives a clear account of the role that the supposed analogy between the mechanisms of overt and covert attention plays in motivating this attitude. The attitude is not one that Macpherson herself adopts. Her own conception of the role that attention might play in effecting a cognitive influence on perception is a more nuanced one. But when characterizing the received view of attention-mediated influences on perception, she writes: We all know that which beliefs or desires one has can alter what one chooses to perceive because they can motivate the turning of one’s head, eyes and body—and the position of these will, of course, affect what perceptual experience one has. Similarly, . . . perhaps one can have different experiences just in virtue of attending to different parts of the scene. And, of course, the location one attends to in a scene can be influenced by one’s beliefs and desires. . . . Although both of these cases involve states of the cognitive system affecting which perceptual experiences one has, they are not typically taken to be cases of cognitive penetration. (Macpherson : )

As her argument progresses, Macpherson goes onto identify a pair of ideas that work together to provide the usual rationale for dismissing these attention-mediated influences, as if they were not genuine cases of cognitive penetration: The reasons that explain why people think that changes of spatial attention directed by one’s cognitive states should not count as cases of cognitive penetration are two-fold. The first is that

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  such cases are akin to selecting what spatial location to attend to by means of changing the position of one’s body or eyes. . . . The second [is the thought] that allocation of attention to a spatial location occurs before perceptual processing occurs. (: )

As the first part of Macpherson’s twofold reasoning makes clear, the supposed analogy between overt and covert attention plays a crucial role in the rationale for taking a dismissive attitude to the attention-mediated cases. Those cases of cognitive influence that are mediated by overt attention certainly do need to be dismissed, and for just the reason that Macpherson identifies: when the thoughts that we are cognizing play a role in determining how our sensory receptors are oriented, those thoughts are not thereby influencing the way in which our perceptual processes operate. Their influence is completed, as the second of Macpherson’s ideas suggests,‘before perceptual processing occurs’. It does not require the perceptual processing to be cognitively penetrated. When covert attention is thought of on the model of overt attention, the influence of covert attention will seem to be equally in need of being dismissed: Just as a cognitively motivated movement of the neck can make it the case that different stimuli are presented to the eyes, and so can enable cognition to influence the things that get perceptually processed without there being any penetration of perceptual processing by cognition, so, as Macpherson indicates, a cognitively motivated shift of one’s covert attention might make it the case that different stimuli are processed by the perceptual processing centres of our brains, and might thereby enable cognition to influence the things that our perceptual processes operate on, without the activity that takes place within those perceptual processing centres being subject to any cognitive influences. A mid-th-century conception of covert attention—as being the result of mechanisms of simple internal resource orientation, analogous to the mechanisms of bodily orientation—will therefore lead us to be dismissive of all cases in which attention mediates the influence of cognition on perception. Such cases will not seem to be genuine cases of cognitive penetration. It seems to be a two-step route of exactly this sort that leads Ophelia Deroy to argue, in a recent article, that certain cases which have been thought to show the cognitive penetrability of perception may in fact show no such thing. This, she thinks, is because there may be a difference of attention that could account for the way in which cognition exerts its influence in these cases. If attention does play such a role then the cases will fail to demonstrate cognitive penetrability because, Deroy claims: unless attention is fixed, there is room for the sameness of stimulation or external conditions not to defeat the claim that perception is cognitively impenetrable. The difference in experience can be explained right at the beginning by being introduced by attention. (Deroy : )

As the last sentence of this quotation suggests, the reason why Deroy takes it that an attention-mediated influence of cognition on perception will not qualify as an instance of cognitive penetration is because she takes it that the influence of attention is completed ‘right at the beginning’ of an item’s perceptual processing. Deroy must

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

here be understanding covert attention, in just the way Macpherson suggests, on the model of its overt counterpart. She, like the mid-th-century cognitive psychologists, must be taking covert attention to be a mere orienting of perceptual resources, which determines what it is that our perceptual processes take as their input but which, having pointed those processes in the right direction, leaves them to get on with their work. In what follows I shall suggest that the picture of attention that theorists like Deroy have assumed is mistaken. Just for the sake of a name, I propose to call it ‘the partition picture’. The idea behind it is that, when the influence of cognition on perception is mediated by attention, influence does not really qualify as a case in which the processes of perception are penetrated by processes of cognition. The mechanisms of cognitive processing exert only an orientation-determining influence on the mechanisms of perception, and they do this without thereby affecting the way in which the perceptual processes operate because the processes of attention stand as a partition between the processes of cognition and those of perception. When cognitive processes influence perception by influencing attention, they do so by acting through this partition, and so without being penetrative of it. It is this—this idea that the attention processes serve to partition off cognitive processes from their perceptual influences—that leads to attention-mediated cognitive influences being dismissed. It leads to the view that an attention-mediated influence of cognition on perception can never qualify as a genuine instance of cognitive penetration.

 Against the Partition Picture Although it was natural to adopt the partition picture in the context of early cognitive psychology, when behaviourism was first being overthrown and covert attention was being distinguished from its overt counterpart, that picture has long been superceded. We should no longer believe that attention is implemented in the sort of mechanisms that could stand as a partition between cognition and the perceptual processing that it influences. This is because we no longer believe that the mechanisms responsible for covert attention are disjoint from those that are responsible for cognition and perception. We now understand the selectivity of attention to be an immediate consequence of the competitive way in which the perceptual and cognitive processes themselves operate. Since it is implemented by the processes of perception and cognition themselves, attention cannot stand as a partition between them. Sections  and  unpack this line of thought. Section  examines some of the evidence for this conception of attention’s underlying mechanisms. Section  shows that, in the competition-based picture of covert attention, which has been erected in the partition picture’s place, the dismissive attitude to attention-mediated cognitive influences is no longer warranted.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  It should be admitted straightaway that the result of these arguments will not be to establish that all theorists who profess an interest in cognitive penetration should cease to be dismissive of attention-mediated cases. Different theorists mean different things by the phrase ‘cognitive penetration’, and different theorists are interested in the phenomena so named for different reasons. There may be some psychological purposes for which we will want to say that the attention-mediated influence of cognition on perception still fails to count as an instance of cognitive penetration, even once the updated conception of attention’s selectivity is in place. The point that I hope to establish in the present chapter is just that, for epistemological purposes, and for some of the other purposes that concern the philosopher of mind, the attentionmediated cases do need to be taken seriously. They serve to illustrate the sort of interpenetration of perception and cognition that underpins an important Kantian insight: that the world as we perceptually encounter it may be intelligible to us partly because the constraints of intelligibility play a role in determining the form of that very encounter. The attention-mediated cases will also be seen to illustrate a variety of cognitive penetration that theorists of ‘embodied cognition’ have wanted to emphasize. They show that the cognitive capacities of an organism are partly to be explained by reference to the way in which the cognitive conduct of that organism is implemented within an environment that the organism perceives, and in which it acts. On this picture—which was most clearly set out by Susan Hurley—the processes responsible for cognition are thoroughly interpenetrated by, and are not sandwiched between, the processes that are responsible for our perception of, and action in, the world (Hurley ). When this form of cognitive penetration is in question, the following arguments indicate that the attention-mediated cases should not be treated dismissively. ∗∗∗

The case for rejecting the partition picture is given by evidence that reveals the processes responsible for the allocation of attention to be inextricable from the processes that are responsible for the perception of the things to which we attend, and reveals them to be equally inextricable from the processes that are responsible for some conceptual classification of those things. Such evidence has been amassed over many years, in the course of an ongoing research programme in which a number of theories have been proposed, abandoned, reconceptualized, and refined. That programme employs a variety of methods, drawing on various branches of psychology and of cognitive neuroscience. The case for rejecting the partition picture is therefore not given by the results of any single experiment. There is, nonetheless, one particular series of experiments—those that have recently been conducted by Dwight Kravitz and Marlene Behrmann—that provides a focused source of evidence for the sort of phenomenon that the partition picture struggles to accommodate, but that can be accounted for easily by the competition-based picture which has replaced it (Kravitz and Behrmann ). I shall focus on these experiments in the discussion that follows,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

before turning in Section  to the details of the partition picture’s competition-based alternative. Like the great majority of the psychologists who examine visual attention, Kravitz and Behrmann measured the reaction times of undergraduates who were performing a simple visual task. They sat these undergraduates in front of a computer screen, on which a letter of the alphabet was briefly flashed up in some arbitrary location. That letter was either a T or an L. The students were asked to press one of two buttons, so as to indicate which. The experimenters measured how long it took for this button to be pressed. The allocation of the students’ attention was gauged by calculating the way in which these reaction times were influenced by the appearance, prior to the target letter, of an attention-directing cue. Kravitz and Behrmann presented this cue for a tenth of a second, and followed it by a tenth of a second break, before the appearance of the letter that was to be identified. The location of the cue gave an imperfect indication— accurate in only % of the cases—of the location in which the target letter would appear. It is a crucial feature of all such experiments that the attention-directing cue does not indicate the location of the letter to be identified with complete reliability, so that, when the participants direct their attention on the basis of that cue, they will not always be attending to the location where the target letter then appears. In the normal case, when the cue does direct attention to the correct location, the participants’ allocation of attention will be reflected by a reduction in the time that it takes for them to respond to the target that appears there. In the other cases, where the cue directs attention to a misleading location, the participants will show an increase in their reaction times, when responding to a target letter that is then presented elsewhere. Slowed reactions therefore indicate that a cue has drawn attention away from the location of the target; speeded reactions indicate that the cue has drawn attention towards the location of the target. By examining whether the occurrence of various cues cause reaction times to be slowed or speeded, relative to others, Kravitz and Behrmann were able to explore the way in which their cues had caused attention to be allocated. ∗∗∗

The screen on which Kravitz and Behrmann present their various cues and target stimuli was not an empty one. In each case it showed a couple of coloured shapes, one on either side of a small, central cross, on which the participants were told to fix their gaze (in an attempt to ensure that all the attention effects here are covert ones). In the simplest of their experiments, the shapes that Kravitz and Behrmann presented on either side of the central cross were coloured rectangles. In other experiments one or both of the shapes were rectangles with bulbous ends. In the most complex of the cases that Kravitz and Behrmann examined the ‘visual objects’ that were presented on either side of the screen were large H- (or h-) shapes, of the sort that are shown in Figure ..

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  Object Condition Identical Category

Non-Identical Category

Far-Object Between-Object

Cue 4.1°

Near-Object

10.4° 4.1° Within-Object 10.4°

Figure . Category task paradigm, illustration of the stimuli and task conditions. Originally published as fig.  in Kravitz and Behrmann (). Reproduced with kind permission from Springer Science and Business Media

It is a well-established fact that the presence of such shapes makes a difference to the way in which attention comes to be allocated, in response to whatever attentiondirecting cues are presented on or near these shapes. This difference manifests itself on those trials in which the appearance of a cue draws the participant’s attention to one location, but where the letter needing to be identified is then presented in another location, some distance away from the location that was cued. The effect of such misleading cues has long been known to depend, much as one would expect, on how close together the two locations are: a cue that draws one’s attention to a far-off location will be more detrimental to one’s reaction times than a cue that directs one’s attention to a location that is close to the place at which the target does then appear (Remington and Pierce ). What is more notable, for our present purposes, is that the effect of misleading cues also depends on whether the locations of the cue and of the target letter fall within the boundaries of a single object (Egly et al. ). A cue that occurs at one end of a rectangle will lead to a fast reaction if the target letter then appears at that same end of that same rectangle. It will lead to a somewhat slower reaction when the target appears at the other end of that rectangle. But the reaction time in this case, where the cue and the target occur at opposite ends of the same object, will be faster than the reaction time in a case where the cue appears at the end of an altogether different rectangle. And this is so even if the spatial separation of the cue and the target is held constant. There is therefore an advantage in the processing of target letters that fall within the same object as an attention-catching cue, in addition to any advantage that might be accounted for by their mere spatial proximity to that cue. ∗∗∗

This relatively old result—even without any of Kravitz and Behrmann’s more recent variations on it—is already enough to suggest that there must be some disanalogy between overt and covert attention, of a sort that would make the simplest versions of the partition picture untenable. When overtly attending to the things that are

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

on my right I can simply point my eyes and ears in a rightwards direction. No perceptual operation needs to have been completed in order to work out which direction that is; my sensory resources can simply be directed at an appropriate region of the egocentrically defined space, just by pointing them at it. This allows overt attention to be allocated before any perceptual processing has taken place. And it is this—this possibility of pre-perceptual allocation—that allows the immediate effects of overt attention allocation to be confined to a moment of orientation that occurs ‘right at the beginning’ of an item’s processing. This, as we saw in Section , is crucial to the rationale for dismissing the cases that involve such attention, as not being genuine instances of cognitive penetration. The same-object advantage effect that was described above—as demonstrated in Egly et al.’s () experiment—shows that the mechanisms of covert attention cannot be confined to an initial moment of orientation in the same way. It shows that, when a cue draws our covert attention to a part of the screen that has objects on it, our attention is not allocated merely on the basis of that cue’s egocentric spatial coordinates. Covert attention is instead allocated to the object within which the cue appears. The mechanisms by which this allocation is achieved cannot be mechanisms of simple spatial orientation, since the identification of the space that is bounded by an object requires some perceptual processing already to have been completed, in order for those object boundaries to be identified. Contrary to what Deroy assumes, in the passage quoted above, the allocation of covert attention cannot happen ‘right at the beginning’ of an item’s perceptual processing. Allocating covert attention is not simply a matter of pointing one’s resources in the right direction. Taken by itself, this disanalogy between overt and covert attention is enough to undermine the partition picture to only a small degree. It shows that at least some perceptual processing resources—those responsible for whatever processing identifies the target and cue locations as falling within the bounds of a single visual object—must be working together with the mechanisms of attention, to determine the way in which attention is allocated. But this is consistent with the idea that attention influences perceptual processing in a way that leaves most perceptual processing innocent of any cognitive penetration. Even if cognitive factors can cause attention to be allocated to visual objects, and not merely to egocentrically defined locations, the result of that attention being allocated might be simply to prioritize the processing of those objects, and not to change anything about the way in which their processing takes place. We might then be able to retain the spirit of the partition-picture by regarding attention— even object-based attention—in the way that Zenon Pylyshyn advocates: Attention is an extremely primitive mechanism that provides selection by only the most elementary properties, such as location and a few other primitively transduced properties. (Pylyshyn : )

Pylyshyn hopes to retain the spirit of the partition picture, in the face of the sameobject advantage effect, by treating object boundaries as ‘elementary properties’, and by treating the detection of those boundaries as an unusually primitive component of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  perceptual processing. It is only then that the lion’s share of perceptual processing can still be understood to happen subsequent to the allocation of attention, in systems that are thought of as being partitioned off from cognition’s influence. The import of experiments such as Kravitz and Behrmann’s is to make the case against Pylyshyn’s strategy by showing that it is not merely on the basis of ‘a few primitively transduced properties’ that attention is allocated. They do this by exploring a series of increasing complex variations on the same-object advantage paradigm. ∗∗∗

The first of Kravitz and Behrmann’s variations introduces the factor of colour. It shows that, just as there is a processing-time advantage for target letters that occur within the same object as the attention-grabbing cue, even when the cue appears in a different part of that object, so there is a processing-time advantage for target letters that occur within an object that is the same colour as the object within which the cue falls, even when that object is a different one. In order to show this, Kravitz and Behrmann presented their subjects with a display having two coloured rectangles on it. The subjects’ attention was directed by the appearance of a cue at one end of one of these rectangles. The letter needing to be recognized then appeared in another location, which might have been at the other end of the same rectangle, but which might also have been in the other rectangle (at a location equidistant from the location where the cue appeared). If the cue appears on a different object from the target letter then, as one would expect given the same-object advantage, the reaction time to that target is slower than it would be if the cue appears on the same object. But when the different object is also of a different colour from the object in which the cue appears then, Kravitz and Behrmann found, the reaction time is slower yet. This indicates that the result of an attention-grabbing cue’s occurrence is not simply to direct attention to a particular location, not even to a location that is defined by the boundaries of a visible object. Instead we have a case in which the appearance of a cue, at one end of a coloured object, is modulating the way in which attention is allocated to various parts of the visual scene, on the basis of their locations, on the basis of the ways in which they are grouped into objects, and on the basis of their colours. ∗∗∗

Kravitz and Behrmann also demonstrate an analogous effect for shape, using a similar experimental procedure. They show that, in responding to a target letter that appears within the boundaries of a bulbously ended rectangle, the cost of having one’s attention cued to the wrong object is greater if that object is not itself a bulbously ended rectangle. Attention, then, is not allocated to simple locations in egocentric space, nor is it allocated simply to the location of an object, nor even to an object, together with the things that share that object’s colour. Instead, Kravitz and Behrmann find that the distribution of colours, shapes, and objects in a scene must together interact to determine the way in which attention is allocated across that scene. They find this even in what might appear to be a maximally simple case of attention, in which attention is

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

allocated on the basis of a cue that is flashed up for  ms, and that is flashed up just  ms before the target stimulus appears. These first parts of Kravitz and Behrmann’s experiment sequence indicate that— contrary to the position that Macpherson articulates, that Deroy assumes, and that Pylyshyn states—attention cannot be allocated by a process of selection that takes place ‘right at the beginning’ of perceptual processing, before the processing in which colours, shapes, and object boundaries come to be represented. The process by which attention selects its targets must instead be understood to include the processes in which all of those features have already been detected. Contrary to Pylyshyn’s claim that ‘attention . . . provides selection by only the most elementary properties’ (: ), Kravitz and Behrmann have shown (together with the whole research programme for which they are here standing as representatives) that all sorts of features of a perceived scene can interact to determine the way in which attention is allocated to that scene, in response to a simple attention-grabbing cue. ∗∗∗

The results that we have been considering so far compromise one half of the picture according to which the processes responsible for attention stand as a partition between cognitive processing and perceptual processing. They show that those perceptual processes by which colours, shapes, and object boundaries are detected are not downstream from an attention process, which orients these perceptual processes, in response to a top-down, cognitive signal, but which then leaves them to get on with their work. The processes responsible for the allocation of attention must instead be dependent on the processes that are responsible for the detection of shapes, colours, and object boundaries. The attention-implementing processes and perception-implementing process are, to this extent, intertwined. Further experiments are needed in order to show that the other side of the partition picture is similarly compromised—i.e. to show that the processes responsible for the direction of attention are inextricable, not only from the perceptual processes, but also from at least some of the processes that are responsible for the cognitive business of applying learned concepts to things. If the attention processes overlap with the cognitive processes, as well as overlapping with the perceptual ones, then attention cannot stand as a partition between these processes, since it is not itself separable from either one of them. It is the last of Kravitz and Behrmann’s experiments that completes this argument. In this third part of Kravitz and Berhmann’s experiment, there were again a pair of objects present on either side of the screen, and again it was as parts of these objects that the attention-directing cues appeared. In some trials the participants were presented with a pair of objects such as those shown on the left hand side of Figure .. In other trials they were presented with the same objects but upside down, as shown on the right side of that figure. Since the capital H shape (shown on the far left of the figure) is symmetrical, there is always an H present, whichever way up the stimuli were presented. Notice, however, that on being turned upside down the other shape which

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  Kravitz and Behrmann use (the shape which is shown on the right of Figure .’s first part) changes from being an h shape to being something like a number . It no longer belongs in the same orthographic category as the shape that is presented alongside it, although its other perceptible properties are, to the greatest extent possible, held constant. In some cases, then, attention is cued to an H-shape and the target letter then appears in a different shape—an h—which belongs to the same orthographic category. In other cases attention is cued to an H-shape, and the target letter then appears in a different shape—a —which belongs to a different orthographic category. These categories are, of course, learnt ones, based on the arbitrary orthography of the Latin alphabet. If there is a perceptual difference between these cases, it must be a difference that is made by acquired concepts. Kravitz and Behrmann’s most impressive finding is that even this difference of arbitrary orthographic categorization makes a difference to the way in which a  ms cue causes attention to be allocated: They find that a cue that draws attention to one object has a greater cost in terms of the reaction time to a target that is presented within the bounds of another object when that other object belongs to a different orthographic category. The cue must therefore prompt a different allocation of attention, depending on whether the screen contains an H and an h, or an H and a . Since registering that fact requires learnt concepts to be brought to bear, attention and conceptual thought are shown to be at least somewhat intertwined. ∗∗∗

Taken together, Kravitz and Behrmann’s several experiments establish that it is the context of colour, of shape, and of semantics that influences the way in which a briefly presented cue modulates the allocation of attention. They show, in Kravitz and Behrmann’s own words, that: attentional selection, as reflected by pattern of median RTs [reaction times] to targets throughout a scene, is intimately tied to the perceptual representations mediating the entire scene. (Kravitz and Behrmann : )

If, as seems to be the case, it is the whole of the context in which an attention-catching cue appears that determines the way in which attention is allocated on the basis of that cue, it cannot then be that attention picks out a location, or an object, or a colour, the processing of which then goes ahead without being subject to any further attention-related influence. Instead, the whole perceptual processing of the attended and unattended items’ colours, locations, shapes, and even the conceptualization of their categories are all interacting to determine the way in which attention is allocated to those items. The partition picture, by which the dismissal of attention-mediated cases seemed to be warranted, therefore needs to be rejected. Attention cannot stand as a partition between cognition and perception because the processes responsible for cognition and perception are not separable from those responsible for the allocation of attention.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

We need now to see that no alternative warrant for the dismissive attitude is provided by the theory that is to be erected in the partition picture’s place, so that attention-mediated influences should be taken seriously, when the penetration of perceptual processes by cognition is at issue.

 Interpenetration and Integrated Competition The theory with which Kravitz and Behrmann suggest we understand their results is the biased-competition theory of attention. That theory, which is also known as the ‘integrated competition theory’, has been developed since the middle of the s. It originates in work by Robert Desimone, John Reynolds, and John Duncan, and has more recently been taken up by several others. Nothing in this area of study is uncontentious, but the basic tenets of the integrated completion theory are now endorsed very widely. Those tenets received one of their clearest statements in a  article by Desimone (published alongside a similarly definitive paper by Duncan, which set out the converging lines of evidence that support the theory). According to Desimone’s account, ‘any enhancing effect of attention on neuronal responses is best understood in the context of competition among all of the stimuli in the visual field for control over behavior’ (Desimone : ). This competition arises from some very basic facts about neuronal wiring. It is ubiquitous in the brain. At any one level of the brain’s processing architecture, whether the processing going on at that level is cognitive, perceptual, or even motor-related, there are several things that might be represented. At the earliest stages of visual processing these might simply be two points of light, both falling within the receptive field of a single cell. At later stages the representations generated will be larger and more abstract. It might be, for example, that there are some stages at which whole objects compete for representation, or where complete interpretations of a scene do so. Whenever there are several things that might be represented, those things will compete for the brain’s representational resources. To understand the importance of this ubiquitous competition for the explanation of attention, we need to add the hypothesis (for which Duncan assembles a range of supporting evidence) that the competition taking place within each level of the processing hierarchy biases the outcome of the competitions that are taking place at other levels. The result of this reciprocal biasing is that the ‘competition . . . is integrated between components of the sensorimotor network’, with the effect that ‘as an object gains dominance in any one system, responses to this same object are supported elsewhere’ (Duncan : ). When a clear winner emerges, in the large-scale competition that results from all of the brain’s local competitions being integrated, the winner of that integrated competition is, ipso facto, the thing to which attention is paid. In introducing this competition-based view of attention, Desimone explicitly contrasts it with the ‘classical view’, according to which covert attention is analogous to its overt counterpart, and so ‘essentially serves as an internal eye that can shift its focus

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  from one location to another’ (Desimone : ). The integrated competition theory tells us that, rather than being directed, like the eye, on the simple basis of spatial location: Processing can be biased in favour of stimuli possessing a specific behaviourally relevant colour, shape, texture, and so on, in parallel throughout the visual field, in addition to biases in favour of stimuli occupying a specific relevant spatial location. (Desimone : )

It is this prediction of the integrated competition theory that Kravitz and Behrmann’s  experiments confirm. It bears emphasis that, according to the integrated competition theory, the selectivity of attention is inherent to the basic structure of the neural architecture in which all of our perceptual, cognitive, and motor processes are implemented. There is no separate system that uses cognitive information to allocate attention to certain targets, and that operates before any perceptual processing has taken place. The system that allocates attention is, instead, the whole of the perceptual, cognitive, and motor system. Integrated competition takes place everywhere. The integrated competition theory therefore has no room for the idea that attention stands as a partition between cognition and perception. It entails that some cases of attention-mediated cognitive influence are cases in which the influence of ‘conceptual recognition’ is, as Kuhn said, ‘built into the nature of the perceptual process itself ’. ∗∗∗

One might object to the interpretation of the integrated competition theory that I have been advocating by maintaining that, when properly understood, the theory does not entail that the processes of perception are interpenetrated with the processes of cognition. Cognition, this objector might say, is still quite separate from the processes of perception. The integrated competition theory merely provides us with an account of the means by which one process, responsible for cognition, can influence a quite separate process, responsible for perception: it can do so by sending down a signal that biases the perceptual competition. But one might maintain that the sending of this signal does not require the cognitive and perceptual processes to be interpenetrating. This interpretation of the competition theory is not unprecedented, but it would, I think, be mistaken. The integrated competition theory should not be thought of as a theory that tells us what happens to the perceptual processing of objects as a consequence of attention having been directed at that object. It should, instead, be thought of as a theory of what it is for attention to be directed at an object. The allocation of attention is not a further explanandum, still needing to be accounted for after the integrated competition is done. Attention is allocated by the whole of the integrated competition. To be attended, according to the integrated competition theory, just is to be that integrated competition’s winner. Some theorists have tended to obscure this point. They write as if the competition that takes place in visual cortex were one thing and the processes responsible for

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

attention were another. Sabine Kastner, who has used the biased competition theory as a framework for a series of fMRI studies examining attention’s neural underpinnings, is an example. She argues as follows: In visual cortex (V), the magnitude of all attention effects increased from early to more advanced processing levels along both the ventral and dorsal pathways. This is consistent with the idea that attention operates through top-down signals that are transmitted via corticocortical feedback connections in a hierarchical fashion. Thereby, areas at advanced levels of visual cortical processing are more strongly controlled by attention mechanisms than are early processing levels. (Kastner : )

The weak top-down control of early visual processing that Kastner describes may not sound like a philosophically interesting instance of cognitive penetration. In taking attention to operate on V, by transmitting a signal to it, Kastner implies that the competitive processing in V is not itself a mechanism of attention. She implies that early perceptual processing is subject to control by an attention signal, which is transmitted to it from a separate set of ‘attention mechanisms’ (which she locates in a network of frontal brain areas). The operation of these mechanisms is not something that the integrated competition theory tells us about. It is not merely a verbal point to complain that Kastner’s reserving of the title ‘attention’ for those parts of the competition that take place in frontal cortex is unwarranted. The integrated competition theory was intended, in its original formulations, as an account in which all of the brain’s competitive processes can act as determinants of the way in which attention is allocated, because they are all integrated into a mutually biasing competition. So it is that Duncan writes: Competition, finally, is integrated between components of the sensorimotor network. . . . As an object gains dominance in any one system, responses to this same object are supported elsewhere. (Duncan : ; emphasis added)

And so it is that Desimone tells us that the ‘third tenet’ of the theory is that the integrated competition: can be biased in favour of one stimulus in a cluttered field by virtue of many different mechanisms, rather than by a single overall ‘attentional control’ system. Such mechanisms include both ‘bottom-up’, or stimulus driven influences (e.g. one stimulus has greater novelty or has a higher contrast than another) and ‘top-down’ feedback mechanisms (e.g. one stimulus has greater behavioural relevance than another). (: )

This is not to deny that the frontal areas that Kastner would have us identify as ‘the attention mechanisms’ play an important role in one part of the biased competition. It is ‘perhaps [the] most speculative tenet’ of the integrated competition theory that ‘a main source of the ‘top-down’ biasing inputs to ventral stream areas in extrastriate cortex derives from structures involved in working memory’ (Desimone : ). It is these structures involved in working memory that Kastner thinks of as implementing ‘attentional mechanisms’. But the idea that working memory is a

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  main source of the biasing signals that operate in the ventral stream certainly does not entail that our theory of attentional selection is not complete until it has been supplemented with a distinct theory of the way in which working memory decides which stimuli to bias. The integrated competition theory already gives us an account of the mechanism by which the attentional selection of working memory is achieved. It is achieved by a competition that is biased by, and integrated with, the other competitions that are taking place throughout the cortex. The competition that takes place in working memory is, no doubt, more interesting and complicated than the competition that takes place among the processes of early perceptual processing— it has certainly proven to be more difficult to study—but, contrary to what Kastner suggests, its role in constituting attention is not different in kind from the role that is played by competitions at other levels of the processing hierarchy. We should therefore interpret the integrated competition theory as telling us that attention emerges as the outcome of the total integrated competition, distributed across the full range of mutually biasing processing levels, including working memory, but also including those processes taking place in V. To say that one’s cognition can influence one’s perception by modulating one’s attention is, on this understanding, to postulate a direct connection between the processing that is responsible for cognition and the processing that is responsible for one’s perception. The processes of perception are so thoroughly interpenetrated by the processes of attentional selection as to be implemented by a subset of those very mechanisms. These mechanisms, when integrated into a cortex-wide competition, contribute to the cognitive work of keeping working memory focused on topics that are relevant to the demands of our current tasks. The processes that are responsible for perception are, according to the integrated competition theory as I have been interpreting it, a subset of those that are responsible for the cognitive work of ensuring that our attention is appropriately directed. ∗∗∗

When attention has been understood in this way there is no longer any rationale for dismissing cases in which cognitive influences on perception are mediated by attention. Cognition exerts an influence on the outcome of competitions taking place throughout the perceptual parts of the brain. Such an influence shows that our perceptual encounter with the world is constrained by the cognitive processes that operate on it. Perception is not a conceptually innocent input to the mechanisms of cognitive processing. Moreover, the perceptual processes are themselves a subset of the system that enables working memory to do its work. The competition that takes place within our perceptual processing centres contributes to the work of ensuring that working memory is appropriately directed. A theory that explains why the representations in working memory are handled intelligently should therefore make mention of the fact that our working memory is situated in a world that is perceived and acted in, just as proponents of embodied cognition have suggested. A complete account of working memory’s work should make mention of the sensorimotor systems with which the competition for representation in working memory is integrated. In these ways the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. mole

attention-mediated influences on perception illustrate a philosophically important variety of cognitive penetration.

References Allport, D. A., Styles, E. A., and Hsieh, S. (). Shifting intentional set: exploring the dynamic control of tasks. In C. Umiltà and M. Moscovitch (eds), Attention and Performance : Conscious and Nonconscious Information Processing, –. Cambridge, Mass.: MIT Press. Armstrong, K. M. (). Covert spatial attention and saccade planning. In C. Mole, D. Smithies, and W. Wu (eds), Attention: Philosophical and Psychological Essays, –. New York: Oxford University Press. Broadbent, D. E. (). Perception and Communication. New York: Pergamon Press. Broadbent, D. E. (). Task combination and selective intake of information. Acta Psychologica (): –. Bruner, J., and Postman, L. (). On the perception of incongruity: a paradigm. Journal of Personality : –. Cherry, E. C. (). Some experiments on the recognition of speech, with one and with two ears. Journal of the Acoustical Society of America (): –. Deroy, O. (). Object-sensitivity versus cognitive penetrability of perception. Philosophical Studies : –. Desimone, R. (). Visual attention mediated by biased competition in extrastriate visual cortex. Philosophical Transactions of the Royal Society B, : –. Duncan, J. (). Converging levels of analysis in the cognitive neuroscience of visual attention. Philosophical Transactions of the Royal Society B, : –. Egly, R., Driver, J., and Rafal, R. D. (). Shifting visual attention between objects and locations: evidence from normal and parietal lesion subjects. Journal of Experimental Psychology: General (): –. Hanson, R. N. (). Patterns of Discovery: An Enquiry into the Conceptual Foundations of Science. Cambridge: Cambridge University Press. Hurley, S. (). Consciousness in Action. Cambridge, Mass.: Harvard University Press. Kastner, S. (). Attention, neural basis of. In T. Bayne, A. Cleeremans, and P. Wilken (eds), The Oxford Companion to Consciousness, –. Oxford: Oxford University Press. Kempen, G., Hermans, B., Klinkum, A., Brand, M., and Verhaaren, F. (). The wordfrequency effect and incongruity perception: methodological artifacts? Perception and Psychophysics (): –. Kravitz, D. J., and Behrmann, M. (). Space-, object-, and feature-based attention interact to organize visual scenes. Attention, Perception, and Psychophysics (): –. Kuhn, T. (). The Structure of Scientific Revolutions. Chicago: University of Chicago Press. Kuhn, T. (). The Structure of Scientific Revolutions, nd edn. Chicago: University of Chicago Press. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. Mole, C. (). Attention. In E. Margolis, R. Samuels, and S. P. Stich (eds), The Oxford Handbook of Philosophy of Cognitive Science, –. Oxford: Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

attention and cognitive penetration  Pylyshyn, Z. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences : –. Remington, R., and Pierce, L. (). Moving attention: evidence for time-invariant shifts of visual selective attention. Attention, Perception, and Psychophysics (): –. Wittgenstein, L. (). Preliminary Studies for the ‘Philosophical Investigations’, Generally Known as The Blue and Brown Books. Oxford: Blackwell.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

PA RT IV

Cognitive Penetrability and the Phenomenology of Perception

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 ‘Looks the Same but Feels Different’ A Metacognitive Approach to Cognitive Penetrability Jérôme Dokic and Jean-Rémy Martin

 Introduction Cognitive penetrability (CP) is the general hypothesis that cognitive states (beliefs, desires, and preferences, perhaps imaginings) can ‘penetrate’, i.e. directly influence or modify, perceptual phenomenology and in particular perceptual content (what is perceived, or perceptually represented).1 In this chapter, we criticize what we see as a dubious premise implicitly or explicitly endorsed in many discussions about CP, either pro or contra. Roughly, this premise validates the transition from relevant differences at the level of perceptual phenomenology (i.e. those that are supposed to result from different cognitive states) to differences pertaining to perceptual content. Once this premise is exposed, the questions of whether cognitive states can influence perceptual phenomenology and, if there is such an influence, what its epistemological implications are, take a quite different form. The partisans of CP usually argue that the putative penetrating states (e.g. beliefs) directly modify or penetrate the contents of perceptual experiences. Nonetheless, we will show that the overall phenomenology of perception has another crucial, affective aspect, which is constituted by various kinds of conscious feelings (e.g. feelings of reality, familiarity, and confidence). In what follows, we will use the phrase ‘affective phenomenology’ to capture the feeling-based aspect of perceptual phenomenology. We shall claim that at least in central cases, the phenomenological changes that the

Both authors contributed equally to this work. 1 For a recent review, see Stokes (). This is a rough personal-level description of cognitive penetrability. CP is also defined in more empirical terms, as entailing the rejection of the encapsulation of perceptual processes (Fodor ; Pylyshyn ; for a thorough, empirically informed philosophical discussion, see Raftopoulos ). To be clear, we count as CP also cognitive modulation of perceptual feature processing.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin partisans of CP construe as being caused by penetrating states concern not the contents of perception proper but its affective side. The chapter is structured as follows. In Section , we introduce and give a standard definition of the notion of cognitive penetrability (CP). Arguments for the claim that perceptual content can be penetrated by cognitive states typically involve (whether explicitly or not) three distinct steps. First, contrast cases are presented in which the phenomenology of experience differs, while sensory stimulation and attention are fixed. Second, the relevant phenomenological differences are said to supervene on perceptual content: what is represented is different in each contrast case. Third, these differences are said to be the direct results of background cognitive states, such as different beliefs. Our main target in the bulk of the chapter (Sections –) is the second step of the argument for CP. We try to show that the relevant phenomenological differences between the contrast cases do not supervene on perceptual content. In Section , we argue that perceptual phenomenology is in fact dual: it has a sensory dimension but is also constituted by feelings which are in an important sense metaperceptual (which does not mean that they have metarepresentational contents). We then describe two pathological cases in which these feelings are obviously missing, in order to highlight their phenomenological contribution to normal perceptual experience. In Section , we show how the second step just described, according to which the phenomenological differences between relevant contrast cases affect perceptual content, works independently of the CP issue, within an influential conception of high-level recognition. Our conclusion is that the acquisition of visual recognitional capacities can change the affective dimension of our experience (some things come to feel familiar) while leaving perceptual content untouched (at least in principle). A fortiori perceptual content cannot be said to be penetrated by cognitive states, such as beliefs about the instantiation of high-level properties. In Section  we turn to an explicit argument for CP, which also hinges on the second step. According to this argument, our chromatic experiences can be penetrated by general beliefs about the characteristic colors of objects. We put forward a reinterpretation of the empirical facts referred to in the argument in terms of the role metaperceptual feelings, such as feelings of confidence, play in the causal modulation of our perceptual beliefs. Finally, in Section , we move to the third step described, according to which the relevant phenomenological differences result from background cognitive states. Since we have now identified the affective nature of this difference, the question becomes whether there is a sense in which metaperceptual feelings themselves can be said to be cognitively penetrable. Whatever answer we should eventually give to this question, we suspect that the epistemological consequences of the CP of feelings would be very different from, and much less disastrous than, the epistemological consequences of the CP of perceptual content itself.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’ 

 Cognitive Penetrability: Phenomenology and Content According to cognitive penetrability, the contents of perception can be modified or influenced by co-occurrent cognitive states, such as beliefs, desires, emotions, or imaginings. In other words, CP entails that what we perceive depends on our cognitive background, for instance what we independently believe to be the case. Here is a recent definition of CP, due to Susanna Siegel: If visual experience is cognitively penetrable, then it is nomologically possible for two subjects (or for one subject in different counterfactual circumstances, or at different times) to have visual experiences with different contents while seeing and attending to the same distal stimuli under the same external conditions, as a result of differences in other cognitive (including affective) states. (Siegel : –)

Following this definition, typical arguments for CP can be shown to proceed in three distinct steps, as follows: Step : Show that some aspect of perceptual phenomenology is modified in relevant contrast cases (intuitive claim).2 Step : Show that the relevant phenomenological differences are to be accounted for at the level of the experiences’ representational contents. Step : Show that these differences in the experiences’ representational contents are directly caused by differences in cognitive background states. As an illustration, take a recent example from Siegel (). Jill, for independent reasons, believes that Jack is angry. The phenomenology of Jill’s visual experience of Jack’s face before and after she acquired the belief that Jack is angry is different (Step ). The difference in phenomenology concerns the content of Jill’s visual experience, which now represents Jack’s face as angry (Step ). Finally, the phenomenology of Jill’s visual experience of Jack’s face is modified as the result of Jill’s belief that Jack is angry at her (Step ). Thus, Jill’s visual experience has been cognitively penetrated. Let us dwell a bit on the second step of the argument. Some philosophers would see it as being rather straightforward, on the grounds that perceptual phenomenology supervenes on perceptual content (see e.g. Tye ). Independently of the issue of whether there are non-representational sensory qualia (such as color sensations), the transition from perceptual phenomenology to perceptual content is correct in many cases. For instance, what it is like to see a red cube is different from what it is like to see a blue sphere. We can safely predict that this phenomenological difference concerns the contents of the visual experiences: the former visually represents a red cube, whereas the latter visually represents a blue sphere. 2 This is the starting point of what Siegel () calls ‘the method of phenomenal contrast’, which we are now going to present in a simplified form. This method can be part of an argument for CP, but it is itself independent of CP, as we shall see in Section .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin However, there are cases in which Step  is harder to justify. For the sake of illustration, consider the feeling of familiarity. Certainly, there is a phenomenological difference between a situation s in which we see someone for the first time and a situation s in which we see the same person after she has become familiar to us. What is the nature of this phenomenological difference? Is it reflected at the level of the contents of the visual experiences? There is certainly a sense in which the person ‘looks the same’ in both situations, in some suitable, experiential sense of ‘looking the same’ (see Lyons ). In other words, our visual experiences in both situations seem to have the same sensory contents. The same visible state of affairs is presented in our experience. As we shall argue, perceptual phenomenology is dual: it has both a sensory and an affective dimension. While the sensory phenomenology of perception supervenes on perceptual content, differences pertaining to the affective phenomenology of perception do not always entail differences at the level of what is perceptually represented. In the case in point, perceptual content is just sensory content.3 Not only does the person look the same but feel different, but this is all that can be perceived: a person with a specific visual appearance. The feeling of familiarity does not affect perceptual content, because (as will become clear later) it is metaperceptual rather than perceptual. In what follows, we shall suggest that many candidate examples of CP discussed in the literature are at best cases in which the affective dimension of the phenomenology of perception has been modified, while its sensory aspect remains untouched. In other words, Step  might fail quite independently of whether the relevant phenomenological difference is caused by cognitive states, i.e. whether Step  itself can succeed. In the meanwhile, though, we are going to present the dual view of perceptual phenomenology in more detail.

 The Dual Phenomenology of Perception The sensory aspect of perceptual experience constitutes an essential and pervasive part of its phenomenology. However, it does not represent the whole of perceptual phenomenology. When you are observing the pigeon just there perched on the branch, you have some experiences of grayish shades, of shapes, of movements, and so on. Now, together with these sensory experiences you also feel that you are looking at a real pigeon (and not, for instance, at a mere projection of your imagination). Let us call this ‘the feeling of reality’ (which some authors also call ‘the feeling of presence’; see e.g. Matthen, ). 3 Here we use the notion of perceptual content in such a way that perceptual content is apparent (manifest or experiential) content, i.e. content consciously available to the subject in her perceptual experience. Some authors use the notion of perceptual content in a different way, e.g. according to Lyons (), perceptual content can be richer than apparent content. On this view, perceptual content shows up at the level of the beliefs delivered by the visual system. However, even Lyons could accept our claim that metaperceptual feelings do not affect the apparent content of experience.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  In Dokic and Martin (), we argued that the feeling of reality constitutes the relational phenomenology of perception, i.e. the experience of being related to actual things, rather than mere representations of them. If the feeling of reality is lost, as in specific pathologies (which we shall discuss), the relational phenomenology is no longer instantiated and subjects are not able, on the basis of their perceptual experiences, to judge whether what they are perceiving (e.g. that there is a pigeon perched on the branch) is actual or not (even if the sensory contents are identical before and after the advent of the pathology). Therefore, along with other types of feelings which can be bound to sensory contents, such as the feeling of familiarity and the feeling of confidence, the feeling of reality contributes to the overall phenomenology of perception. In order to understand how the relevant feelings work, it is necessary to make a detour via the psychological theory of metacognition.4 One can distinguish between two sources of metacognitive judgments: theory-based and experience-based judgments (Koriat ; ). Theory-based judgments are the result of explicit inferences from our encyclopedic background knowledge or from knowledge that we have about our own cognitive skills. In contrast, experience-based judgments result from an affective experience—what is referred to as metacognitive feelings. As an illustration, the belief that we know who was the forty-third President of the USA can either be based on the reasoning that we have learned the names of all the USA Presidents at school and/or that we have an excellent memory or it can be based on the gut feeling that we know the answer. Feelings of knowing are examples of metacognitive feelings among others, including feelings of forgetting, feelings of confidence/unconfidence, feelings of certainty/uncertainty. These specific affective experiences constitute the output of monitoring processes, which involve implicit inferences from a set of internal cues, such as availability of partial information or fluency. In this respect, the level of availability of partial information or the level of fluency will modulate the quality of the corresponding feeling (Koriat ; Dunlosky and Bjork ). In previous work, we hypothesized that the distinction between theory-based and experience-based judgments can also be applied to metaperceptual judgments (Dokic and Martin ). Feelings present in metamemory, such as the feelings of certainty/uncertainty and confidence/unconfidence, but also feelings of familiarity, are present in metaperception too. The judgment that one is unsure if the particular bird one is looking at is a blackbird or a raven can be based on either theory (e.g. one explicitly tries to categorize the bird on the basis of its features, and eventually fails to determine whether it is a blackbird or a raven) or experience (one has a feeling of uncertainty about the kind of bird perceived, perhaps because it is too dark or there 4 The literature on metacognition (which is not the same as metarepresentation) is now considerable. See e.g. Nelson (), Reder (), Levin (), Proust (; ), Dunlosky and Bjork (), Beran et al. ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin is a mist). Similarly, the judgment that one has already seen this place in the past can be based on either theory (e.g. one explicitly remembers that it is not the first time that one comes in this town) or experience (the place just feels familiar). Finally, as described above, our ordinary perceptual experiences are usually accompanied by feelings of reality. The judgment that there is a real as opposed to an imagined pigeon there on the branch can be based on either theory (e.g. one realizes that there is no reason to imagine a pigeon and that one never imagined anything like that in the past) or experience (the pigeon feels present). Normally, of course, when we look at things around us we have the feeling of being related to actual things rather than mere representations, imaginings, or memories. Feelings of reality could be the result of a reality-monitoring process, which assesses if an informational state has been internally or externally generated.5 Studies in metamemory show that experimental manipulation of processing fluency with respect to a target will influence the strength of the subject’s feeling of knowing, independently of the specific content at stake (e.g. Koriat ; ). In other words, the feeling of knowing is not sensitive to memory contents as such, but only to the quality of retrieval or collateral cognitive processes. In perception, too, the feelings of confidence, familiarity, and reality seem to be sensitive to the quality of perceptual or collateral cognitive processes rather than perceptual contents as such. The feeling of confidence is the conscious reflection of the strength or optimality of the perceptual response, while the feeling of familiarity seems to be a function of both observed and expected processing fluency (Whittlesea and Williams ). It could be thought that the feeling of reality strongly depends on features of the perceptual content the subject is currently enjoying. For example, the subject who seems to see a pink elephant while heading home might not feel this to be related with an actual thing but think that she is just hallucinating (all other things being equal). If it appears that the elephant is not realistic enough she won’t undergo a feeling of reality of what she seems to see. However, in some cases of hallucination subjects undergo a feeling of reality although the content is a horrible yet two-dimensional creature. (Note that the subjects can be perfectly aware that they are hallucinating; see Shanon .) Similarly, several experimental studies of the feeling of reality in the context of virtual reality found that the degree of realism is not a determining factor 5 It has been proposed that a self- or reality-monitoring mechanism is necessary to sort out what comes from the outside world and what is due to the subject herself (e.g. Helmholtz ; Frith ; ). For instance, the question arises as to how the brain can distinguish between inner and outer speech events given that both can have the same contents. The proposal is that a subpersonal mechanism monitors the source of the event (i.e. either internal or external) and gives rise at the conscious level to a feeling of agency attached to internal events (imaginings) and, we argue, to a feeling of reality attached to external events (percepts) (Frith ). In this regard, Frith (e.g. ; ) and many others suggested that auditory hallucinations in schizophrenia would result from a breakdown of the self-monitoring mechanism, so that patients attribute to external sources active inner speech (in other words, this breakdown leads patients to lose the feeling of agency that is normally attached to their inner speech or, in our view, to have a feeling of reality for their own inner speech) (see e.g. Blakemore et al. ; Ditman and Kuperberg ; Farrer and Franck ; Fletcher and Frith ; Ford and Mathalon ; Ford et al. ; Johnson et al. ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  for the feeling of reality (i.e. the feeling of reality remains identical between conditions of high- or low-level realism; Sanchez-Vives and Slater ). Interestingly, it appears that ‘the graphics frame-rate is positively correlated with reported [reality]’ (SanchezVives and Slater : , who actually use the term ‘presence’, but in the sense of reality at stake here), in that purportedly a well-suited frame rate improves the fluency of processing. As other types of metaperceptual feelings, feelings of reality are essential to the overall phenomenology of ordinary perception. In particular, these feelings, and not perceived objects themselves (sensory contents), constitute the phenomenology of perception as a relation to the world. When we are looking at some object in the world we have both a sensory experience and a relational experience, viz. the fact that we feel en rapport with an actual object (rather than with a mere representation of this object). In contrast, objects and their properties constitute the sensory side of our perceptual experiences. In this view, the phenomenology of perception is Janusfaced. To the extent that perceptual phenomenology is dual, we can predict that the sensory and the relational/affective dimensions of perceptual experiences can vary independently of each other. Such variations show up more saliently in pathological cases such as derealization disorder (DD) and Capgras syndrome (CS), which we now briefly present.

. Derealization disorder Patients presenting DD experience a kind of affective detachment with the external world, which then appears strange, inconsistent, and unreal (Sacco ): ‘[their perception is not] lived but is more like a mechanical, purely receptive sensory process, unaccompanied by its affective-tone’ (Parnas and Sass : ). Patients report that it is as if they were watching a movie or a picture of the world, instead of the real things around them: ‘In fact, the people and things around you seem as unreal to you as if you were only dreaming about them’ (Shorvon et al. : ). Shorvon et al. report another patient’s own description: ‘Through the eyes I look out at a world that might be a picture of the world’ (p. ). It is as if patients were confronted with a mere representation of the world. The causes of DD are multiple, including cognitive, neurological, and psychiatric disturbances (Sacco ). Whatever etiology DD has, the crucial thing we would like to highlight is that the content of the subjects’ experience should motivate the presence of an affective phenomenology that is in fact absent. They do not have the feeling of being directly presented with the actual sensory contents that they are nevertheless entertaining. In other words, the patients’ perceptual states are no longer accompanied by feelings of reality, as Shorvon et al. noted: ‘[their] perceptions do not awaken a feeling of reality’ (p. ). To our mind, the experience of derealized patients illustrates the relative independence between sensory phenomenology and affective/relational phenomenology, at least with respect to the feeling of reality. In addition, subjective reports of patients support the thesis that the affective phenomenology is a necessary condition for

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin the relational phenomenology to be instantiated, as it is deeply affected in patients. At the level of perceptual judgments, to the extent that the sensory aspect of the patients’ perceptual experiences is not affected they can produce judgments about the properties of objects with a high degree of confidence (e.g. ‘This pigeon is gray’). On the contrary, to the extent that the affective aspect of the patients’ perceptual experiences is deeply affected they cannot produce judgments about the actuality of things (e.g. ‘There is a pigeon’) with a high degree of confidence.

. Capgras syndrome Patients with CS hold the belief that one of their relatives (e.g. their spouse) has been replaced by an impostor. The formation of this delusional belief would result from the fact that the affective component that normally accompanies our perceptual experiences of familiar persons is disturbed (i.e. absent) in patients. The affective component refers to the feeling of familiarity. In fact, some models propose the existence of two information-processing pathways of face recognition. A first pathway (the ventral visuo-semantic one) encodes semantic information about facial features to construct a visual image of face and constitutes the medium of overt recognition. A second pathway (the dorsal visuo-affective one) processes the affective response to familiar faces and is responsible for both the feeling of familiarity and the covert automatic recognition.6 In CS, as in DD, the sensory phenomenology of subjects should motivate the presence of an affective phenomenology that is in fact absent. The person the subject is perceiving looks (perhaps exactly) like one of her relatives, but the subject lacks the feeling of familiarity associated with the ordinary perception of a familiar person. As a consequence, she does not feel en rapport with the relative. The person ‘looks right’ but ‘feels wrong’ to the subject, who subsequently confabulates about the explanation of her weird experience and forms delusional beliefs. We may surmise that the experience of Capgras patients illustrates the relative independence between sensory phenomenology and affective/relational phenomenology with respect to the feeling of familiarity. Moreover, the feeling of familiarity is, at least in some conditions, necessary for the phenomenology of high-level recognition. In contrast, the sensory phenomenology is only sufficient for the phenomenology of similarity in that the patient refrains from judging that her relative is actually present but admits that the person she is looking at is visually strictly similar to the relative. Having now argued that the overall phenomenology of perception is dual, we review two important families of cases which have been identified in the literature as potential targets of what we have called ‘Step ’, namely the attempt to show that relevant 6 See e.g. Bauer (). The neuroanatomical validity of the original two-route model has been questioned and refined (e.g. Ellis and Lewis ). Nonetheless, the idea of a dissociation between overt and covert recognition is much less controversial. In the same way, the precise nature of the experiential content of the Capgras delusion, or what exactly this content involves, is still under discussion (e.g. Bayne and Pacherie ; Dokic ; Pacherie ; Reimer ; Young ; ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  phenomenological differences supervene on perceptual content. We shall reinterpret these cases in terms of the dual view of perceptual phenomenology, and suggest that these differences belong to the affective aspect of perceptual experience, and may not concern perceptual content itself.

 High-Level Recognition: Pine Trees, Cardinals, and Other Familiar Things An important issue in philosophy of perception concerns the admissible contents of perceptual experience (Hawley and Macpherson ). Austere conceptions of perceptual content allow only low-level sensory qualities, such as color, shape, size, texture, and movement, to be represented in perception, whereas liberal conceptions of perceptual content are happy to let high-level properties, including kinds and artifacts such as being a pine tree or a table, enter the representational contents of experience. In principle, CP arguments are independent of the foregoing issue, as they can focus on either low-level or high-level properties (see Macpherson ). Indeed, in the next section we are going to discuss one of these arguments as applied to color, one of the paradigms of a low-level property. Our aim in this section is to criticize an influential strategy to the effect that high-level properties can figure in the contents of perception. If this strategy is not committed to CP, it can be exploited as a central stage in an argument for the latter. In this respect, it is worth discussing, since it illustrates a common mistake, or at least a blind spot, in many discussions about the relationship between the phenomenology of perception and its contents. Consider two relevant contrast cases, involving visual experiences E and E, respectively before and after the subject has learned to visually recognize cardinals as such. In our view, the ability to recognize a kind of things on the basis of perception is first and foremost a judgmental skill. The subject who can recognize a cardinal when she sees one is able to form spontaneous demonstrative judgments of the form ‘This is a cardinal’, where the relevant token of ‘this’ refers to the perceived bird. Spontaneous judgments are non-inferential, in the sense that they do not present themselves as conclusions of explicit pieces of reasoning or trains of thoughts. They are made ‘just like that’, on the basis of what the subject perceives.7 There is arguably a phenomenological difference between E and E (Step ). What it is like to see what is in fact a cardinal when one lacks the relevant recognitional ability is different from what it is like to see it when one possesses and exercises this ability. Siegel () (see also Bayne ) argues for the validity of the transition, in such a case, from differences in phenomenology to differences in contents (Step ).

7 See McDowell (). Of course, spontaneous judgments can be inferential in another sense, if they result from implicit (unconscious) inferences.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin For instance, one might suggest that the concept cardinal must be used in a correct specification of the content of E, whereas it is not required to capture the content of E (whether or not the contents of these experiences are themselves conceptual). On this suggestion, the content of E is in a sense richer than the content of E, since it involves a high-level property (being a cardinal) over and above low-level properties already present in E. Given the dual phenomenology of perception, an obvious alternative interpretation is that the acquisition of the relevant recognitional ability has changed the affective dimension of the relevant visual experiences. More precisely, the subject has become familiar with cardinals, so that the perception of what is in fact a cardinal generates a feeling of familiarity. This feeling is integral to the phenomenology of E but is absent from the phenomenology of E. In our view, the alternative interpretation is compatible with the fact that high-level properties such as being a cardinal do not literally enter the contents of our perceptual experiences. Siegel () envisages and rejects the possibility that the relevant phenomenological difference between experiences such as E and E is in fact non perceptual. She mentions ‘cognitive’ phenomenology and ‘background’ phenomenology as possible candidates. On the one hand, perhaps the familiarity gained through the acquisition of a recognitional ability, i.e. what we are going to call the phenomenology of familiarity, is determined either by a ‘commitment-involving’ cognitive attitude (like judgment or belief) or by a cognitive attitude which does not involve such commitment (like merely entertaining a proposition). Siegel raises two objections to this claim. First, the phenomenology of familiarity is belief-independent, in the sense that it can be experienced even if the subject does not believe that she is looking at a familiar bird or tree (for instance, she falsely believes that she is facing a hologram). Second, it seems to be thought-independent as well, in the sense that the subject does not even have to think or explicitly entertain the proposition that what she perceives is a familiar bird or tree. On the other hand, perhaps the phenomenology of familiarity can be explained by the instantiation of standing, background states, like moods. Here, Siegel makes two observations. First, background states, such as drunkenness and depression, usually color the phenomenology of perception by affecting perceptual content: ‘depression can cause things to look grey; drunkenness can cause them to look blurry’ (: ). Second, background states have ‘non-local’ effects on phenomenology; they affect the whole visual scene and not merely a particular perceived entity. Siegel briefly discusses another possible suggestion, according to which the phenomenology of familiarity is generated within perceptual experience itself, but by a ‘non-representational feeling of familiarity’ (p. ). So the phenomenological difference between E and E would belong to the phenomenology of perception, but without affecting the contents of experience. Siegel objects to this suggestion on the grounds that ‘familiarity is not the sort of thing that could be felt without any representation of something as familiar’ (p. ). Familiarity, considered as what is

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  gained through the acquisition of a recognitional ability, cannot be a ‘raw feel’, but must be attributed to a particular perceived entity: a particular bird or tree, which must then be represented as familiar. Interestingly, none of these objections applies to our view. To begin with, affective phenomenology is indeed not cognitive in Siegel’s sense. We can have a feeling of familiarity with respect to a perceived entity concurrent with a belief that this entity is not familiar at all (see Section ).8 Remember that metaperceptual feelings, such as the feeling of familiarity, do not result from beliefs or thoughts, but are grounded on the subpersonal monitoring of the quality of perceptual and collateral cognitive processes. Indeed, to the extent that holograms of pine trees or cardinals yield perceptual inputs strongly similar to actual visual inputs of such natural kinds (after the subject has learned to recognize them), then the feeling of familiarity will be de facto identical in both cases. Moreover, metaperceptual feelings are often local in the sense that they are bound to specific sensory contents. The feeling of familiarity can be free-floating (perhaps in the case of the déjà vu experience), but it is often bound to a particular entity, like a bird, tree or person (or at least the corresponding bundle of low-level sensory properties). Does it follow that we are committed to the claim that feelings of familiarity are ‘raw feels’? Siegel seems to reason as follows: since feelings of familiarity cannot be raw feels, they must contribute to the representational contents of experience. What does it mean to say that feelings of familiarity are representational? The contents of feeling of familiarity might be either objectual or propositional. On the one hand, a given feeling of familiarity can be bound to a perceived person, and its content might be exhausted by the person itself. However, if the content of the feeling is purely objectual in this sense, it does not add anything to the content of experience: the entity to which it is bound is already a constituent of sensory content. On the other hand, even if we assume that the content of the feeling is propositional, so that the perceived person is explicitly represented as familiar, it does not follow that it must be conceived as contributing to what is perceived rather than felt by the subject. As we have noted above, the feeling of familiarity is, at least in part, a function of the fluency of perceptual and/or cognitive processes. Plausibly, learning to recognize kinds of birds on the basis of her perception has heightened the fluency of whatever categorization processes will lead to her spontaneous judgments of the form ‘This is a bird of kind K’. The fluency of these processes, which are partly perceptual and partly cognitive, is what generates a feeling of familiarity bound to the sensory presentation of a particular bird. Of course, there are cases in which we can experience a feeling of familiarity without actually categorizing the perceived object as a cardinal or, in the case of persons, our 8 Similarly, we can have a feeling of reality bound to the sensory experience of a particular thing without believing that that thing is real (perhaps because we know that we are in a virtual reality setting). In this sense, both feelings of reality and feelings of familiarity can be belief-independent.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin good friend Pierre. For instance, we can have a ‘tip-of-the-tongue’ experience as to the identity or kind of what we see (Schwartz ). In this case, there is no explicit categorization process which could ground our judgment of identity (‘This is Pierre’) or kind (‘This is a cardinal’). Still, just as the feeling of knowing can be sensitive to the fluency of aborted or truncated retrieval processes (Koriat ), the feeling of familiarity with respect to a perceived object might be sensitive to the fluency of aborted categorization processes. That is, the subject feels as if she could easily identify or categorize what she is seeing, while actually failing to do so. From the brain’s point of view, the initial steps of the categorization process are ‘promising’, so to speak, which is enough to generate a conscious feeling of familiarity, even if the subject is eventually unable to form a specific identity or recognition judgment. What we have suggested so far is that the acquisition of recognitional abilities can change the affective phenomenology without the relevant high-level property entering perceptual content. One might go further and claim that the contents of the experiences before and after this acquisition can in principle be the same. I have learned to recognize cardinals on the basis of their visual appearance, but cardinals look the same now as before; the way they look has not changed just because I am now able to form spontaneous judgments about the visible presence of cardinals. Similarly, the way Pierre looks to me now can be the same as the way he looked to me when I first met him. What has changed concerns only the affective dimension of perceptual phenomenology; now both cardinals and Pierre feel like familiar entities to me. Of course, it is probable that in fact the contents of the relevant experiences, for instance E and E, will not be quite the same because the ability to recognize cardinals on the basis of their visual appearance can make me attend to different and perhaps more subtle features of the perceived birds. In other words, attentional differences can be responsible for changes at the level of perceptual content, but these changes will arguably concern only low-level properties represented in my experience. In a nutshell, the acquisition of a visual recognitional ability can affect the phenomenology of perception in only two ways. On the one hand, it changes the affective dimension of our perceptual experience, since we will experience new feelings of familiarity bound to specific sensory contents. Such a change can be explained without positing an enrichment of perceptual content by high-level properties, such as being a cardinal, being a pine tree, or being identical to Pierre. On the other hand, it brings about new attentional styles, which can change the sensory dimension of our perceptual experience. Neither way of affecting the phenomenology of perception entails CP. The generation of feelings of familiarity by itself does not add anything to perceptual content, whereas the influence of attention on perceptual content is usually considered to be ‘pre-perceptual’, and thus insufficient to justify CP (see e.g. Pylyshyn ; Raftopoulos ). In the next section, we examine a putative case of CP of color experience by general beliefs or memories. Again, our strategy will consist in showing that the inference

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  from perceptual phenomenology to perceptual content (Step ) should not be made too hastily.

 The Color Case, or the Importance of Being Confident . Color constancy, the memory color effect, and cognitive penetrability When you are strolling in the streets of Paris, you see plenty of colors, shades, and so on. You can easily remark that the illumination of building surfaces is different according to their orientation relative to the sun. Then, while you are watching the Pantheon you see some parts as being lighter than other (shaded) parts. Nonetheless, you still see the Pantheon as a uniformly colored surface (and not as an alternation of yellowish and grayish surfaces). This phenomenon reflects a perceptual mechanism known as ‘color constancy’, which ensures that our color experience of a particular uniformly colored object remains relatively unchanged (i.e. constant) under varying illumination conditions. Color constancy suggests that the visual system not only takes into account the lightness/brightness differences of surfaces but also the illumination conditions (Rock ). Some authors have wondered if our knowledge or memories about the color of familiar objects could help us to maintain color constancy for these objects, especially when the illumination conditions are unknown or neutral (Granzier and Gegenfurtner ; Hansen et al. ; Olkkonen et al. ). In the studies of Hansen et al. () and Olkkonen et al. (), subjects had to adjust the color of depicted familiar fruit objects (orange, banana, and so on) until these objects subjectively appeared gray or, in other words, achromatic (see also e.g. Witzel et al.  and Delk and Fillenbaum ). As an illustration, participants were shown a photograph of a real banana upon a gray background on the monitor screen. Their task was to adjust the color of the banana (which was globally yellow) until it appeared gray. Results showed that ‘the settings for the banana deviated from the neutral gray adaptation point [as it was measured for each subject]’ (Hansen et al. ) when compared to objects with nonfamiliar and non-characteristic colors (i.e. random noise patches and uniform points of light). More precisely, participants adjusted the color of the banana to a slightly bluish hue, which is the opponent color of yellow. It was as if participants continued to see the banana yellowish while the neutral achromatic point was already reached. The same kind of effect was found for some of the other fruits and vegetables photographs tested (lemon, orange, carrot, and so on). This effect is thought to be evidence of what is referred to as a ‘memory color effect’, or MCE (e.g. Hansen et al. ; Olkkonen et al. ; Witzel et al. ). A memory color is the typical color of a specific object that we have learned and memorized through repeated exposures with this object. The MCE ‘refers to the idea that memory colors modulate the color appearance of the respective objects’ actual

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin colors’(Witzel et al. : ) (e.g. the achromatic banana purportedly appears yellowish to the subject). Gegenfurtner and colleagues (Hansen et al. ; Olkkonen et al. ; Witzel et al. ) claim that our prior knowledge about the characteristic colors of objects can influence our color perception of such objects. In this respect, the knowledge that a banana (for instance) is yellow supposedly influenced the way subjects perceived its actual achromatic color appearance. We thus have a potential case of cognitive penetrability here (Macpherson ). A participant of CP could indeed argue that prior knowledge directly influenced (Step ) perceptual phenomenology (Step ) at the level of the representational chromatic contents of the experiences (Step ). In the following two subsections we assess the viability of the CP interpretation of the experiences described above. Then we discuss the decisional anti-CP hypothesis recently put forward by Zeimbekis () and, in the last subsection, we formulate our own alternative account.

. Against Step : a short methodological critique One could claim that prior knowledge directly influenced the representational contents (Step ) only if the MCE found in the study of Hansen et al. () reveals an actual perceptual effect. Nonetheless, the specific methodology used in this study raises some doubts about the perceptual interpretation of the MCE. In particular, the adjustment method used in Hansel et al.’s () experiment (see e.g. Kingdom and Prins ) does not disclose the nature of the observed MCE. The signal detection theory (SDT) teaches us that when a perceptual task involves conditions for which the evaluated performances of subjects are different, this change of performance can either reflect a change in the perceptual experience of subjects (i.e. sensitivity) or a change in the way they decided to respond (i.e. decision criteria). The experimental paradigms designed from the SDT precisely can discriminate between these two alternatives. In contrast, the protocol of Hansen et al. () leaves us uncertain as to whether the MCE is a perceptual or a decisional effect. For instance, knowing that a banana is yellow, subjects might have raised their decision criteria so that they stopped adjusting the color when they felt quite certain that the banana is now achromatic, i.e. when they were really confident in their response (in the framework of the SDT, subjects are said to be ‘conservative’). In contrast, in the case of a stimulus without a characteristic color, like a random noise patch, subjects lowered their decision criteria so as to stop adjusting the color; they did not have to be as confident as in the case of a stimulus with a characteristic color (subjects are said to be ‘liberal’). Note that a change in the decision criteria of subjects might have produced exactly the same MCE as the one found in the study of Hansen et al. (). However, on the assumption that these authors have uncovered a truly perceptual phenomenon, the question arises as to whether the interpretation of this effect in terms of CP is well suited.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’ 

. Against Step : the CP hypothesis and outline shapes In their  study, Olkkonen et al. (using the same protocol as the one described above) found that the MCE of non-neutral color stimuli, fruits and vegetables, is absent when pictures are not realistic photographs but only outline shapes of these same fruits and vegetables stimuli. The outline shapes lack many of the perceptual cues present in the photographs, namely complex color distribution, three-dimensionality, and texture patterns. Assuming that this effect does not reflect a decisional bias but a truly perceptual effect, how can we explain the differences between realistic pictures and outline shapes? The first thing to note is that the CP hypothesis would have probably entailed the inverse prediction, viz. that the MCE should have been stronger for nonrealistic pictures than for realistic pictures. Arguably, the more perception is degraded or underspecified, the more the penetrating state would have room to modify (i.e. penetrate) the current perceptual content (Hohwy ). Conversely, the more the perceptual state is optimal and stable, the less the penetrating state should have the opportunity or room to modify it.9 In this regard, the putative cases of CP usually concern degraded perceptual conditions, as precisely in the experiments of Hansen et al. () and Olkkonen et al. (): the penetrating state (whatever it is) operates only when perception becomes strongly suboptimal, i.e. when the color parameters reach the near absolute threshold of subjects. Gegenfurtner and colleagues propose to explain the differences between realistic and non-realistic pictures in the following way: ‘to activate the visual representation [of fruits and vegetables] strongly enough to induce an illusory color percept, the object has to have all relevant visual features—shape, shading, and texture—present’ (Olkkonen et al. : ). Recall that these authors argue that the MCE results from the influence upon perceptual states of prior knowledge about the characteristic colors of familiar objects (Hansen et al. ). If this interpretation is taken at face value, it does not fit with the explanation in terms of the strength of visual representations, and it cannot explain the differences between realistic and unrealistic pictures in that the latter are the most likely to be sufficient to activate prior knowledge relative to the characteristic colors of familiar objects they represent. As the authors themselves acknowledge, ‘If the memory color effect is solely due to the learned object–color association, it should appear independently of the particularities of their perceptual features’ (Witzel et al. : ). In this respect, Witzel et al. () found a MCE for stimuli lacking threedimensionality, texture, and complex color distribution (e.g. picture of a Smurf or of the Pink Panther)—in other words, the features that are precisely lacking in the 9 The remark applies to all kinds of putative penetrating states. Macpherson (), for instance, argues that in studies like that of Olkkonen et al. (), beliefs penetrate only indirectly color perceptual contents through imaginative states. Very likely, it is easier for imagination to impregnate a perceptual state when the latter is non-optimal (e.g. underspecified). The arguments that will follow against the CP interpretation apply to the imaginative interpretation too.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin outline shapes of Olkkonen et al. (). Witzel et al. () therefore reinterpret the lack of a MCE for these specific outline shapes by saying that the objects they represent are less recognizable in comparison to the photographs of these same objects. More precisely, they designed a preliminary reaction time experiment in which subjects should indicate as fast as possible in a forced-choice task the color of achromatized manmade and natural objects displayed on a monitor screen.10 Then experimenters recorded the accuracy rate and the reaction time of subjects. With regard to the stimuli of Olkkonen et al. (), authors found the following pattern of results: ‘the medians of the accuracy rates and average reaction times for the eight photos . . . were % and  ms. For the corresponding eight silhouettes [i.e. outline shapes] these were % and  ms, respectively. For the seven photos of the white-painted fruits these medians were % and  ms’ (p. ) (note that for all the outline shapes present in this preliminary experiment, the accuracy rate reaches %). The accuracy rate is very high even for the outline shapes, which means that they were easily recognized (i.e. in order to attribute the accurate color to an object, we need to recognize what object it is). In this context, why do Witzel et al. () argue that the difference of MCE between outline shapes and photographs is explained by the lesser recognizability of the former in comparison to the latter? They based their claim upon the fact that subjects are slower to respond for outline shapes than for photographs. However, we can express doubts about Witzel et al.’s () interpretation of the different results obtained by Olkkonen et al. () between outline shapes and photographs in terms of recognizibility. First, after all outline shapes are well recognized as indicated by the high levelof-accuracy rate of subjects when matching the right color to these objects. Second, recognition of outline shapes is very likely improved by the fact that they are presented along with their realistic versions, as Witzel et al. () themselves admit: ‘the outline shapes in general did result in a quite high accuracy rate (%) in comparison with the photos (%). This probably had to do with the fact that the silhouettes of objects corresponded in size and shape to the photos of the same objects’ (p. ). Third, the claim that outline shapes are less recognizable than photographs seems to mean only that subjects need slightly more time to recognize them in comparison to photographs (as you need more time to recognize the face of a familiar person in the twilight because some perceptual cues usually exploited by the perceptual system are lowered because of poor light). Now, to the extent that in the adjustment task of Olkkonen et al. () subjects had all the time they wanted to adjust the color of pictures, the fact that subjects need slightly more time to recognize the outline shapes should 10 In this study Witzel et al. () wondered whether the MCE works not only for natural objects but also for manmade objects. To this aim, they used the same method as the one elaborated by Hansen et al. () and Olkkonen et al. (). The preliminary experiment was designed to select the objects with the highest diagnosticity (i.e. the objects that strongly refer to a typical color). Note that Witzel et al. () included the fruit and vegetable stimuli of Olkkonen et al. () in the preliminary experiment but not in the main experiment, precisely because they wanted to test the MCE for artificial objects.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  not raise any problem. Fourth, even if outline shapes are less recognizable, the fact is that they are recognized. Therefore, they should activate the prior knowledge about the typical colors of objects they represent and generate a MCE, which rests upon the activation of such prior knowledge (according to Gegenfurtner and colleagues). In other words, the memory colors are all-or-nothing states. Once the recognitional concept of the object represented by a realistic or unrealistic picture is triggered, the memory color is activated in exactly the same way. The memory colors constitute a kind of (not necessarily explicit) declarative knowledge encoded in long-term memory, and therefore do not co-vary with the low-level properties of objects. Witzel et al. ()’s interpretation of the difference between outline and realistic shapes in terms of the differential level of recognizibility between these two categories is not conclusive. To sum up, the CP interpretation of the results of Hansen et al. () and Olkkonen et al. () is not convincing. At least, it is not well suited to account for the various findings: if the prior knowledge that bananas are characteristically yellow is constant over all the realistic and non-realistic conditions, it cannot be the determinant factor that accounts for the differences between realistic pictures and outline shapes.

. The memory color effect: an influence of memory on categorization judgments? For similar reasons, the recent anti-CP interpretation of these experiments by Zeimbekis () is not entirely satisfactory. Zeimbekis argues that the differences between shapes with characteristic colors and color-neutral shapes can be explained by a specific decisional bias owing to experimental uncertainty conditions. In particular, in the experiments described so far, ‘the colors from which the subject has to choose extend from the least-yellow grays, across her mean settings for the concept gray, and into the least-blue grays. There are no perceived cutoff points between these regions of the quality space, and thus, no perceived last yellow-gray’ (p. ). In such uncertainty situations, Tversky and Kahneman () have shown the existence of a specific, anchoring bias. That is to say, the activation of a color concept can anchor the subject in a specific decision and change the way she categorizes or judges her color experiences. Imagine that you have to categorize a particular shade that looks neither clearly red nor clearly orange. In this case, your judgments of categorization will depend on some decisional processes. If there are no influencing factors, your judgment that the shade is orange or red will vary randomly over time (i.e. all other things being equal, your decisional processes will depend on the shade itself, and to the extent that it is intrinsically ambiguous, you will never be inclined to a specific category). Nonetheless, the decisional processes, and then the judgments of categorization, can be biased toward one of the categories. For instance, if all the other priming shades presented before the ambiguous target shade were clearly red, you will have a tendency to judge more often the ambiguous shade as red (by resemblance if the shade is more red than orange) or more often as orange (by contrast if the shade is

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin more orange than red). In other words, the priming effect anchored you in a specific judgment. With respect to the experiments of Hansen et al. () and Olkkonen et al. (), the fruit pictures may play the role of anchoring the judgment of subjects in a particular way. The concept banana (for example) may activate the color memory associated with bananas (i.e. yellowness) and change the way subjects draw the concept gray (compared to color-neutral shapes). That is to say, the subjects’ judgments ‘are anchored in the concept yellow, which is triggered each time she looks at the [e.g. banana] shape’ (Zeimbekis : –). Although Zeimbekis does not discuss the absence of a MCE for outline shapes, we can make some predictions from his hypothesis. In particular, we predict that if his hypothesis is right, there should not be any difference between outline shapes and photographs, although evidently there is. More precisely, to the extent that outline shapes are able to trigger the recognitional concepts of objects they represent, the corresponding memory colors should be activated and the anchoring process engaged. As a consequence, we should see similar decisional biases between outline shapes and photographs. One might reply that when subjects reach their mean for grayness in the case of photographs, that is the point where the pictures’ colors are indiscriminable from the background (hereafter, I-point). Some cues persist that sustain online recognition of the represented object, namely three-dimensionality cues and texture patterns (i.e. from these cues the subject is still able to mentally construct a representation of the represented object). To the extent that these cues are absent in the case of outline shapes, when subjects reach the I-point there is nothing left that could sustain online recognition (i.e. the subject is not able to form a mental representation of the represented object anymore). Therefore, the corresponding memory colors are deactivated and the anchoring bias cannot be engaged.11 A direct consequence of this suggestion is that there could not be a MCE for any outline shapes because they lack the relevant cues sustaining recognition at the I-point. However, we have seen that Witzel et al. () found a MCE even for outline shapes lacking three-dimensionality, texture, and complex color distribution. This result suggests that for memory colors to have an influence on subjects’ color categorization processes at the I-point, memory colors need not to be sustained by the online recognizing of objects, which would be enabled by the presence of relevant cues in the case of photographs. This is true whether the influence of memory colors is perceptual, as argued by Gegenfurtner and colleagues, or decisional, as suggested by Zeimbekis (). In other words, a memory color representation can continue to be active in the absence of a stimulus-driven representation of the object. A possibility is that at the I-point a mental representation of the object persists for some time, which therefore continues to activate the corresponding memory color, which in turn triggers 11

We thank an anonymous reviewer for this suggestion.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  the MCE. The presence of a MCE for the outline shapes of Witzel et al. () could also suggest that some cues still persist after all, enabling the online maintaining of a mental representation of the objects at the I-point. Independently of the empirical results of Witzel et al. (), we can indeed doubt that subjects cease to have a mental representation of the represented object just at the I-point. That does not seem very plausible. It is sufficient that such representation persists for a while to continue to trigger the relevant memory color and produce a MCE. If this is true, however, the outline shapes of Olkkonen et al. () should have exhibited a MCE. Although we agree with Zeimbekis that the different effects found in the studies of Gegenfurtner and colleagues do not reflect a perceptual effect, we have raised some doubts as to whether the decisional hypothesis he proposes can explain the lack of a MCE for the outline shapes in Olkkonen et al. () (as well as the presence of a MCE for the outline shapes in Witzel et al. , if we take into consideration the suggestion that the absence of a MCE for the outline shapes in Olkkonen et al.  results from the absence of the relevant cues at the I-point).

. Cognitive penetrability and metacognition At this stage, we would like to sketch an alternative hypothesis that stresses the metacognitive component at stake in the experiments described so far. Indeed, the subject is very likely to reflect on her visual experience in order to achieve an adjustment task. Such reflection can potentially be influenced either at the level of judgments, as proposed by Zeimbekis, or at the level of metacognitive feelings, as we now suggest. In Section , we indicated that a metacognitive judgment can be either theory-based or experience-based. The second kind of judgment is made possible by the presence of feelings that are sensitive to the quality with which the information is processed (e.g. whether it is fluently encoded or not). We surmise that the subjects’ judgments of subjective certainty about the colors of the presented objects are experience-based, and more precisely based on feelings of confidence bound to color-categorization processes. We hypothesize, first, that color-categorization processes are more fluent in the case of realistic images than in the case of neutral shapes and, second, that feelings of confidence about what colors are perceived are accordingly stronger for the former case.12 Variations at the level of the strength of these feelings may lead to inaccurate metacognitive judgments, which in turn impact the correctness of the subjects’ categorical judgments. On this hypothesis, fruit concepts and the associated memory colors do not directly influence the subjects’ categorical judgments or the experience’s representational contents (Step ). 12 One could also hypothesize that realistic pictures enhance the fluency of perceptual processes as a whole, including the fluency related to shape or texture processing. In this context, participants’ feelings of confidence (about the object) will also be enhanced. Stronger feelings of confidence could subsequently bias color-categorization judgments because participants would misinterpret these feelings as being related to the current task.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin Our hypothesis should be seen as an inference to the best explanation. In particular, it best explains the different results obtained by Hansen et al. () and Olkkonen et al. (). If realistic fruit pictures enhance the feeling of confidence about perceived color, the experience-based judgment of certainty that the banana (for instance) is now achromatic will be delayed. In other words, the subject will persist in being subjectively certain that the current shape is yellow longer in the case of a banana or a lemon than in the case of color-neutral shapes. As a consequence, subjects will adjust the neutral gray adaptation point farther away for the former than for the latter. Another way to describe the situation is to say that the period of uncertainty about the color of the image (yellow-gray or neutral-gray) is delayed for realistic fruit images compared to color-neutral shapes. In this context, there are at least two options as to why or how this delay occurs. A first option is that the feeling of confidence about perceived color is stronger for realistic pictures than for neutral shapes from the start, i.e. when participants begin to adjust the color and the level of saturation is high for both kinds of stimuli. In this view, because the level of confidence is initially higher for realistic pictures, the achromaticity judgment for these pictures in comparison to neutral shapes will be delayed. Although possible, this option is unlikely. It would indeed be surprising that the level of confidence about the perceived color is higher for realistic pictures than for neutral shapes when the level of saturation is nonetheless quite high for both kinds of stimuli. A second option is that the decreasing rate of confidence level is slower for realistic pictures than for neutral shapes. The high fluency of colorcategorization processes in the case of realistic shapes slows down the decreasing of confidence as the color becomes more and more ambiguous. Subjects will be more confident that a barely yellow realistic picture of a banana is yellow than in the case of an equally barely yellow neutral shape. Our hypothesis can equally account for the differences between realistic pictures and less realistic ones. On this hypothesis, the more realistic the picture is, the stronger the feeling of confidence will be. Therefore feelings of confidence will be stronger for photographs than for outline shapes. Contrary to the prior belief that bananas are yellow, metacognitive feelings can come by degree. It is an essential trait of feelings that their strength can be modulated and constitutes a kind of gradient (e.g. Koriat ; ).13 This explanation fits well with another result of Olkkonen et al. (). They designed a third kind of stimuli, which consist in fruit pictures with reduced texture surface. These stimuli are less realistic than fruit photographs but more realistic than outline shapes. Results show that the MCE for stimuli with reduced texture surface is in the midst between fruit photographs and outline shapes. In this regard, the strength of feelings of confidence for stimuli with reduced texture surface will be in the midst between fruit photographs and outline shapes. 13 Note that in metacognitive tasks subjects are asked to evaluate the strength of their feelings by using a scale, and they perform this exercise easily.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  It remains to be explained why Witzel et al. () found a MCE for their outline shapes. Here is a suggestion: the strength of metacognitive feelings is further modulated by the presence of a contrast between events, stimuli, situations, and so on (e.g. Whittlesea and Williams ; ). For instance, the familiar aspect of your friend will be stronger if you meet her in an unexpected place than in her home. Another intuitive example is given by quiz games: we are usually more confident in the choice of a response if we selected it among alternative options than if we had been asked to give the response without options. Experimentally, sentences presented with a high visual contrast are felt to be more credible than sentences presented with a low visual contrast (because a high visual contrast raises the fluency of perceptual processing). This is especially true when the visual contrast of the current sentence deviates from the visual contrast of the previous one. In other words, high-contrast sentences preceded by a low-contrast sentence will improve more drastically the subjective confidence in the truth of the sentence than a high-contrast sentence preceded by a high-contrast sentence (Hansen et al. ). In the experience of Olkkonen et al. (), outline shapes of familiar objects are presented not only with neutral shapes but equally with realistic photographs of the same objects. In contrast, in the experience of Witzel et al. () the outline shapes are only presented along with neutral shapes (i.e. they are not presented along with their realistic versions). As a consequence, the outline shapes in this study appear much more familiar than neutral shapes, but the outline shapes in Olkkonen et al. () appear much less familiar than photographs. Accordingly, the feeling of confidence is raised for the outline shapes of Witzel et al. () while it is lowered in the study of Olkkonen et al. () and the process of adjustment is modulated correspondingly. We predict that if the outline shapes of the latter study were presented only along with neutral shapes, the former would exhibit a MCE. To sum up, experiments which can seem to show that our color experience is penetrated by prior beliefs about the characteristic color of things are more plausibly interpreted as showing that the affective dimension of our perceptual phenomenology is modified, while the sensory contents of the experience remain untouched. Underor overconfidence can lead us to behave as if we saw the colors of things differently from how we actually see them, and thus to slightly misinterpret the contents of our own experience. Our main argument against CP has consisted in attacking the claim according to which, in central putative cases of cognitive penetrability, differences in phenomenology (Step ) would concern differences in experiences’ perceptual contents (Step ). In the last section, we briefly address the question of whether there is a sense in which metacognitive feelings themselves can be said to be cognitively penetrable (Step ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin

 Conclusion: Can Feelings Be Cognitively Penetrated? Our aim so far has been relatively modest, since we wanted to show that reference to manifest changes at the level of perceptual phenomenology, even if they would result from cognitive states, is not enough to show that there is CP of perceptual content. Now assume that we are right and that perceptual content itself, as it is determined by the sensory dimension of perceptual phenomenology, is cognitively impenetrable. The question arises as to whether metaperceptual feelings, like the feelings of reality, familiarity, or confidence, can be directly influenced by cognitive states (ArangoMuñoz ). This is a difficult question, which obviously requires more conceptual analysis and empirical background than we can provide here. In this last section, we offer only preliminary remarks. It is certainly true that many metaperceptual feelings are not caused by cognitive states, but generated in a bottom-up way thanks to low-level mechanisms. Feelings of reality and feelings of familiarity might be caused, at least in some cases, by independent stimulus-driven processes in the dorsal stream, belonging to either the visuo-motor system (Matthen ) or the visuo-affective system (Young ). Feelings of confidence can surely operate in the absence of any relevant penetrating state, by reflecting, at the least in part, the fluency of perceptual and/or cognitive processes (Jacoby and Dallas ; Koriat ; ; Oppenheimer ; Wurtz et al. ). Concerning the MCE, we reported that subjects take more time to match the accurate colors to fruit images when the latter are outline shapes than when they are photographs (Witzel et al. ). This could be taken as evidence that outline shapes are processed less fluently than photographs, at the perceptual and/or post-perceptual level. In particular, familiar objects might improve the fluency of color processing itself in a purely bottom-up way.14 Of course, even if it happened that color concepts or prior beliefs could directly influence the fluency of low-level perceptual processes, this would not amount to a CP effect of perception. First, it is not the processing of perceptual features itself that is affected but only the quality with which such features are processed. Second, fluency cues have no contents; they are just signals for the relevant monitoring processes which interpret these signals as indicating whether the current processing is going right or not. Fluency cues thus can only influence subjects’ decisions (e.g. in affecting the quality of some feelings), not perceptual contents. The architecture of metaperceptual feelings allows for some modularity effects. For instance, something can feel present, or familiar, even if we independently believe or know that it is not present (perhaps because we suffer from a hallucination) or familiar (perhaps because we are really facing a hologram). It is probable that these feelings are not as persistent as optical illusions, and will eventually disappear if our background 14 The enhanced fluency of the color processing might result from the fact that a familiar object to which a specific color is attached may prime or pre-activate this specific color (e.g. Ganis and Kosslyn ; Gegenfurtner ). But priming is a pre-perceptual notion in the sense that it happens before the actual stimulus is processed.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  beliefs say otherwise, but the fact that there is some affective resilience shows that feelings cannot be simply identified with beliefs or judgments. In the case of high-level recognition (Section ), some authors would be inclined to think that the acquisition of general beliefs about, say, cardinals is what generates the feeling of familiarity toward perceived cardinals. Arguably, when the subject has acquired the ability to recognize cardinals on the basis of visual perception, she has also acquired general beliefs about how cardinals usually look. However, it is not obvious that these beliefs cause or are responsible for the feeling of familiarity. Another obvious interpretation is that the acquisition of beliefs about cardinals and corresponding feelings of familiarity are joint effects of a common cause, namely the acquisition of the recognitional ability itself, considered as a practical skill. On this interpretation, there is no direct or indirect causal relation from beliefs to metaperceptual feelings. Both feelings of familiarity and feelings of confidence are sensitive at least to the fluency of processes within the brain. It might be argued that they are not sensitive to the fluency of perceptual processes as such, but also to the fluency of post-perceptual, cognitive processes.15 For instance, in the case of the MCE, familiar objects might improve color categorization processes rather than perceptual fluency itself. Our feelings may or may not be bound to sensory contents, but even when they are, they might reflect features of both perceptual and cognitive processes. Thus, what we conceive as belonging to perceptual phenomenology, namely feelings bound to sensory contents, can reflect post-perceptual processes as well. To the extent that background states such as beliefs can influence such processes, they might also be able to influence the corresponding feelings, generating a form of CP of the affective dimension of perceptual phenomenology. Even if metaperceptual feelings can be penetrated by cognitive states, or more precisely cognitive processes, this form of CP is very unlike the CP of perceptual content. To begin with, metaperceptual feelings may not have intrinsic contents, so cognitive states could at best influence their non-semantic properties, such as their strength. Independently of the issue of content, the epistemological consequences of each form of CP are quite different. It has been argued that the CP of perceptual content threatens certain epistemological views, such as dogmatism (see Siegel ; but see also Lyons  for a criticism of Siegel’s argument). If the belief that Jack is angry makes Jill see Jack angry, there might be a kind of epistemic circle: the experience of seeing Jack angry both justifies and is penetrated by the belief that Jack is angry. Now if CP only reaches the affective dimension of perceptual phenomenology, no such circle arises. Even if Jill feels certain that Jack is looking angrily at her, she is not thereby justified in believing so. On the contrary, at least on the normative level, her visual experience would justify the belief that Jack is looking at her, say, in a neutral way. 15 Psychologists acknowledge that metacognitive feelings can reflect either perceptual or conceptual fluency. For instance, the latter distinction looms large in Bullot and Reber’s () metacognitive theory of aesthetic feelings.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin Metaperceptual feelings play a causal role in the formation of perceptual beliefs, but may not affect the justificatory relation between the contents of perceptual experiences and the contents of our perceptual beliefs. However, it is safe to say that we currently lack a good epistemological model of metacognitive feelings (for preliminary remarks, see Dokic ), so we leave the question of their epistemic role or power to another occasion.

References Arango-Muñoz, S. (). The nature of epistemic feelings. Philosophical Psychology (): –. Bauer, R. M. (). The cognitive neuropsychology of prosopagnosia. In H. D. Ellis, M. A. Jeeves, F. Newcombe, and A. Young (eds), Aspects of Face Processing, –. Dordrecht: Nijhoff. Bayne, T. (). Perception and the reach of phenomenal content. Philosophical Quarterly : –. Bayne, T., and Pacherie, É. (). Bottom-up or top-down? Campbell’s rationalist account of monothematic delusions. Philosophy, Psychiatry, and Psychology (): –. Beran, M., Brandl, J. L., Perner, J., and Proust, J. (eds) (). Foundations of Metacognition. Oxford: Oxford University Press. Blakemore, S. J., Smith, J., Steel, R., Johnstone, E. C., and Frith, C. D. (). The perception of self-produced sensory stimuli in patients with auditory hallucinations and passivity experiences: evidence for a breakdown in self-monitoring. Psychological Medicine (): –. Bullot, N. J., and Reber, R. (). The artful mind meets art history: toward a psycho-historical framework for the science of art appreciation. Behavioral and Brain Sciences (): –. Delk, J. L., and Fillenbaum, S. (). Differences in perceived colour as a function of characteristic color. American Journal of Psychology : –. Ditman, T., and Kuperberg, G. R. (). A source-monitoring account of auditory verbal hallucinations in patients with schizophrenia. Harvard Review of Psychiatry (): –. Dokic, J. (). Person recognition and the feeling of presence. In B. Nanay (ed.), Philosophy of Perception: New Essays, –. Oxford: Oxford University Press. Dokic, J. (). Seeds of knowledge: noetic feelings and metacognition. In M. Beran, J. L. Brandl, J. Perner, and J. Proust (eds), Foundations of Metacognition, –. Oxford: Oxford University Press. Dokic, J., and Martin, J.-R. (). Disjunctivism, hallucinations, and metacognition. WIREs Cognitive Science : –. doi: ./wcs.. Dunlosky, J., and Bjork, R. A. (eds) (). Handbook of Metamemory and Memory. London: Taylor and Francis. Ellis, H. D., and Lewis, M. B. (). Capgras delusion: a window on face recognition. Trends in Cognitive Sciences (): –. Farrer, C., and Franck, N. (). Self-monitoring in schizophrenia. Current Psychiatry Reviews : –. Fletcher, P. C., and Frith, C. D. (). Perceiving is believing: a Bayesian approach to explaining the positive symptoms of schizophrenia. Nature Reviews Neuroscience (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  Fodor, J. (). The Modularity of Mind: An Essay on Faculty Psychology. Cambridge, Mass.: MIT Press. Ford, J. M., and Mathalon, D. H. (). Electrophysiological evidence of corollary discharge dysfunction in schizophrenia during talking and thinking. Journal of Psychiatric Research : –. Ford, J. M., Roach, B. J., Faustman, W. O., and Mathalon, D. H. (). Synch before you speak: auditory hallucinations in schizophrenia. American Journal of Psychiatry : –. Frith, C. D. (). The Cognitive Neuropsychology of Schizophrenia. Hillsdale, NJ: Erlbaum. Frith, C. D. (). The self in action: lessons from delusions of control. Consciousness and Cognition : –. Ganis, G., and Kosslyn, S. M. (). Multiple mechanisms of top-down processing in vision. In S. Funahashi (ed.), Representation and Brain, –. Tokyo: Springer. Gegenfurtner, K. R. (). Cortical mechanisms of colour vision. Nature Reviews Neuroscience : –. Granzier, J. J. M., and Gegenfurtner, K. R. (). Effects of memory colour on colour constancy for unknown coloured objects. i-Perception : –. Hansen, J., Dechêne, A., and Wänke, M. (). Discrepant fluency increases subjective truth. Journal of Experimental Social Psychology : –. Hansen, T., Olkkonen, M., Walter, S., and Gegenfurtner, K. R. (). Memory modulates color appearance. Nature Neuroscience : –. Hawley, K., and Macpherson, F. (eds) (). The Admissible Contents of Experience. Oxford: Wiley-Blackwell. Hohwy, J. (). The Predictive Mind. Oxford: Oxford University Press. Jacoby, L. L., and Dallas, M. (). On the relationship between autobiographical memory and perceptual learning. Journal of Experimental Psychology: General : –. Johnson, M. K., Hashtroudi, S., and Lindsay, D. S. (). Source monitoring. Psychological Bulletin  (): –. Kingdom, F. A. A., and Prins, N. (). Psychophysics: A Practical Introduction. New York: Academic Press. Koriat, A. (). The feeling of knowing: some metatheoretical implications for consciousness and control. Consciousness and Cognition : –. Koriat, A. (). Metacognition and consciousness. In P. D. Zelazo, M. Moscovitch, and E. Thompson (eds), Cambridge Handbook of Consciousness, –. New York: Cambridge University Press. Levin, D. T. (ed.) (). Thinking and Seeing: Visual Metacognition in Adults and Children. Cambridge, Mass.: MIT Press. Lyons, J. C. (). Perception and Basic Beliefs: Zombies, Modules, and the Problem of the External World. Oxford: Oxford University Press. Lyons, J. C. (). Circularity, reliability, and the cognitive penetrability of perception. Philosophical Issues : –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. Matthen, M. (). Seeing, Doing, and Knowing. Oxford: Oxford Unversity Press. McDowell, J. (). Avoiding the myth of the given. In J. Lindgaard (ed.), John McDowell: Experience, Norm, and Nature, –. Oxford: Blackwell. Repr. in J. McDowell, Having the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 j. dokic and j.-r. martin World in View: Essays on Kant, Hegel, and Sellars, –. Cambridge, Mass.: Harvard University Press, . Nelson, T. O. (). Consciousness and metacognition. American Psychologist : –. Olkkonen, M., Hansen, T., and Gegenfurtner, K. R. (). Colour appearance of familiar objects: effects of object shape, texture and illumination changes. Journal of Vision : –. Oppenheimer, D. M. (). The secret life of fluency. Trends in Cognitive Sciences (): –. Pacherie, E. (). Perception, emotions and delusions: revisiting the Capgras delusion. In T. Bayne and J. Fernandez (eds), Delusions and Self-Deception, –. Hove: Psychology Press. Parnas, J., and Sass, L. A. (). Self, solipsism, and schizophrenic delusions. Philosophy, Psychiatry, and Psychology (): –. doi:./ppp... Proust, J. (). Metacognition and metarepresentation: is a self-directed theory of mind a precondition for metacognition? Synthese : –. Proust, J. (). The Philosophy of Metacognition. Oxford:Oxford University Press. Pylyshyn, Z. W. (). Seeing and Visualizing: It’s Not What You Think. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press. Reder, L. M. (ed.) (). Implicit Memory and Metacognition. Hove: Psychology Press. Reimer, M. (). Is the impostor hypothesis really so preposterous? Understanding the Capgras experience. Philosophical Psychology (): –. Rock, I. (). Perception. New York: Scientific American Books. Sacco, R. G. (). The circumplex structure of depersonalization/derealization. International Journal of Psychological Studies : –. Sanchez-Vives, M. V., and Slater, M. (). From presence to consciousness through virtual reality. Nature Reviews Neuroscience : –. Schwartz, B. L. (). Tip-of-the-Tongue States: Phenomenology, Mechanism, and Lexical Retrieval. Mahwah, NJ: Erlbaum. Shanon, B. (). The Antipodes of the Mind: Charting the Phenomenology of the Ayahuasca Experience. Oxford: Oxford University Press. Shorvon, J. H., Hill, J. D. N., Burkitt, E. and Halstead H. (). The depersonalization syndrome. Proceedings of the Royal Society of Medicine : –. Siegel, S. (). Which properties are represented in perception? In S. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs (): –. Stokes, D. (). Cognitive penetrability. Philosophy Compass : –. doi: ./ phc.. Tversky, A., and Kahneman, D. (). Judgment under uncertainty: heuristics and biases. Science : –. Tye, M. (). Ten Problems of Consciousness. Cambridge, Mass.: MIT Press. von Helmholtz, H. (). Handbuch der Physiologischen Optik. Leipzig: Voss. Whittlesea, B. W. A., and Williams, L. D. (). Why do strangers feel familiar, but friends don’t? A discrepancy-attribution account of feelings of familiarity. Memory and Cognition (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

‘looks the same but feels different’  Whittlesea, B. W. A., and Williams, L. D. (). The source of feelings of familiarity: the Discrepancy-Attribution Hypothesis. Journal of Experimental Psychology: Learning, Memory, and Cognition (): –. Witzel, C., Valkova, H., Hansen, T., and Gegenfurtner, K. R. (). Object knowledge modulates colour appearance. i-Perception : –. Wurtz, P., Reber, R., and Zimmermann, T. D. (). The feeling of fluent perception: a single experience from multiple asynchronous sources. Consciousness and Cognition : –. Young, G. (). In what sense ‘familiar’? Examining experiential differences within pathologies of facial recognition. Consciousness and Cognition (): –. Young, G. (). Beliefs, experiences and misplaced being: an interactionist account of delusional misidentification. Phenomenology and the Cognitive Sciences (): –. doi: ./s---. Zeimbekis, J. (). Color and cognitive penetrability. Philosophical Studies. doi: ./ s---.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Cognitive Penetrability and Consciousness Athanassios Raftopoulos

 Introduction: Phenomenal Consciousness and Cognitive Access Consciousness Block (d: ) distinguishes between two concepts of consciousness, phenomenal and access consciousness. ‘Phenomenal consciousness is just experience; access consciousness is a kind of direct control . . . a representation is access conscious if it is actively poised for direct control of reasoning, reporting and action.’ Access consciousness is the consciousness of content that is available to the cognitive centers of the brain; it is information that is broadcast in the global neuronal workspace (Block a: –; b: ). The qualification ‘actively poised’ is meant to exclude from being characterized as access conscious those beliefs that are not occurrent, as the pieces of knowledge acquired in the past that rest inactivated in long-term memory. On the other hand, the definition of access consciousness cannot demand of actual use Block (e: ). Block (a: –; e) thinks that access consciousness or reflexivity is phenomenal consciousness (PC) plus reflection. Reflexivity is phenomenality plus reflection on the phenomenality that is, ‘phenomenality plus another state, one that is about the phenomenal state’ (Block a: , ). This is the reflexive sense of consciousness that contrasts with phenomenality. In access consciousness, subjects have a state, e.g. a belief or thought, which is about their own experience. Thus, they have thoughts or beliefs about their perceptual contents; ‘A-conscious states are “propositional attitude” states like thoughts, beliefs and desires’ (Block e: ). Block is clear that access conscious states are first-order beliefs and not second-order beliefs that one is having such beliefs. Forming a thought about perceptual contents presupposes that cognition has access to these contents. A representation ‘is access-consciousness if it is made available to cognitive processing’ (Block : ). Block (a; b; c: ) claims that ‘there can be phenomenally conscious states that are not cognitively accessible’,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  and goes on to maintain ‘I will be talking about cognitive accessibility instead of reportability’ (Block c: ). For these reasons, I will be referring to cognitive access consciousness (CAC) instead of access consciousness, where the consciousness implicated concerns perceptual experience. Block seems to construe CAC as enabling subjects to form either beliefs whose contents are the contents of perception, or beliefs whose contents are the thoughts with this perceptual content; this is the distinction between a person having a thought with a content and the same person having a thought that they are having a thought with that content. When the representational content of a perceptual state is accessed for cognitive reasons, this content becomes the content of a thought that the viewer can entertain simply by being in that state. Thus, CAC contents are the contents of perceptual thoughts/beliefs; the contents of CAC states are necessarily conceptual being the contents of thoughts; ‘the beliefs have their semantic contents conceptualized by the persons whose beliefs they are’ (Davies : ). One might wish to add a second component to the definition of CAC, to wit, that a perceptual state is CAC not only if the content of the perceptual state is available as the content of a thought but also if the viewer is able to think that she is in that state. This is a stronger sense of CAC. Thus, not only should the viewer be able to entertain a thought with the perceptual content as content, but also she should be able to entertain the higher-order thought that she has that thought. Davies (: ) argues that even the stronger notion of CAC is not sufficient for consciousness as it applies to beliefs because ‘there are imaginable cases in which a subject has a belief, and is thereby in position to judge that she has that belief, and even does judge that she has that belief, but where both the first-order belief and the secondorder belief would be intuitively reckoned as unconscious beliefs’. It follows that a state can be cognitive access-conscious in the stronger sense without being conscious to the subject, i.e. without the subject realizing or noticing that she in such a state. Davies’ conclusion strikes me as paradoxical, for how can it be that a state is cognitive access-conscious and yet also an unconscious state? The paradox can be removed if one construes the two occurrences of ‘conscious’ in ‘cognitive access consciousness’ and in ‘unconscious state’ in a different way. In CAC the term ‘consciousness’ designates the fact that cognition has access to some perceptual content, while ‘consciousness’ in ‘unconscious state’ means that the subject does not notice or realize that she is having such a state. I call the consciousness of X that is accompanied by the subject’s realization that she is a state with X as content, ‘awareness in the strong sense’. Thus, the second meaning of conscious presupposes that the subject be aware in the strong sense that she is having such a state, and this is the sense in which ‘conscious’ being is used in psychology. It is important, thus, to clarify the conditions under which beliefs are conscious or not. An intuitive answer is that one may entertain beliefs and use them for various purposes even though they are not conscious that they entertain these beliefs (as in the case of using implicit premises); the beliefs that a subject has without noticing

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

or realizing that they have them are called ‘implicit’. One could draw from this the conclusion that an implicit belief is an unconscious belief. Underneath this view lies the assumption that a state is conscious if the persons who have it are aware that they are in this state, in the sense that they realize that they have such a belief. Either that person has a second-order thought that they are entertaining such a belief, or they have a second-order experience that they are in such a state. If one subscribes to this view, what makes a mental state of a person conscious is the person’s awareness (strong sense) that they are in this state. Dretske (: –), however, argues that what renders a person’s state conscious is not some sort of second-order awareness that they are in such and such state, or that they are having that state. A state is conscious because being a certain sort of representation makes one aware of the properties of an object and of the object itself of which it is a sensory representation. That is, a state is conscious if it plays a cognitive role in making a person aware of some facts or things (a first-order consciousness). ‘Beliefs are conscious, not because you are conscious of them, but because, so to speak, you are conscious with them.’ Beliefs that are thought of as implicit but play a cognitive role are deemed as conscious. Dretske does not claim that everything that happens to someone when they become aware of some object or event is conscious. However, a belief has to be conscious in order for a person to be made aware of things and events through this belief. It follows that an implicit belief—i.e. a belief that can be had even though subjects do not realize or notice that they have it—can be a conscious belief. It is easy to see that Block’s CAC fits this construal of consciousness. A CAC state, a belief, is conscious since it makes one aware of, say, a drilling noise even one does not realize that one has this belief—i.e. even though the belief is implicit. In view of these, I employ the term ‘implicit’ to designate beliefs of which subjects have CAC but they are not aware in the strong sense that they have them, and the term ‘explicit’ to denote states of which subjects are aware in the strong sense. If the conscious/unconscious distinction is not equivalent to the explicit/ implicit distinction, since both explicit and implicit beliefs are conscious, what is an unconscious state? An intuitive answer is that a state, or a stimulus, or properties of stimuli are unconscious if we cannot report about them even when they are attended. This definition is in line with the practice in psychology of talking about implicit processing in cases such as binocular rivalry or spatial neglect; in these cases, even though some stimuli are in the visual fields and fully attended, they never reach the level of awareness that would allow reporting them. An implicitly represented stimulus, in contradistinction, can become explicit and be reported when attended as the literature in change blindness, for example, suggests. According to Block’s definition, CAC with respect to perceptual states is defined as the active poised-ness of some perceptual content for cognition, which entails either that the content is the content of some belief (the weak notion of CAC) or that there is also a higher-order belief whose content is that one is having the thought with that perceptual content (the strong notion of CAC). As we saw, Block (e: ) opts

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  for the first option. In either case, nothing in the definition entails that these thoughts should be such that subjects should be aware that they are having them. Being information that is broadcast in the global neuronal workspace is a necessary condition for the subject to become aware in the strong sense of this thought, but is not a sufficient condition. Let us turn now to PC. Block’s definition of phenomenal consciousness (PC) leaves it open whether Block intends it to be intransitive (just a raw feel) or transitive (consciousness of some content). If PC is the qualitative aspect of experience—the ‘what it is like’ to have it—it may be intransitive. Block’s work, however, suggests otherwise. Discussing Spelling’s experiments to argue that phenomenality overflows cognitive access, Block (a: ) writes: ‘I also allow phenomenal consciousness of the shapes themselves without any reflexive consciousness of them. I say that the subjects have phenomenal experience of the shapes.’ Thus, he talks about consciousness of objects and properties. This relates the phenomenal content of an experience to the PC of having the experience, and hence, PC is transitive. In an earlier paper, Block (e: ) opens the way to a transitive notion of consciousness by admitting that many phenomenal contents are also representational, and therefore that the notion of PC has a transitive component as well. At the same time, there is an intransitive ingredient in Block’s PC: We may suppose that it is platitudinous that when one has a phenomenally conscious experience, one is in some way aware of having it. Let us call the fact stated by this claim—without committing ourselves on what exactly that fact is—the fact that phenomenal consciousness requires Awareness. (This is awareness in a special sense, so in this section I am capitalizing the term.) Sometimes people say Awareness is a matter of having a state whose content is in some sense ‘presented’ to the self or having a state that is ‘for me’ or that comes with a sense of ownership or that has ‘me-ishness’. (Block c: )

Block’s own view confirms this analysis: ‘I argue that “phenomenal consciousness” in my sense of the term can be either transitive (take an object of which the subject is conscious) or intransitive’ (b: ) Therefore, to say that a state is phenomenally conscious is not just to say that there is something it is like to be in it, but also to say something about differences in what it is like: ‘phenomenally conscious content is what differs between experiences as of red and green’ (p. ; emphasis added). Burge (; ) distinguishes between the phenomenality associated with a perceptual experience and the PC of the same experience. Burge (: , ) defines PC as ‘part of how the perceptual capacity presents its objects—entities in the physical environment. The qualitative elements in consciousness . . . are aspects of ways of referring; they are part of the perspectival framework of perceptual reference.’ Block (b: ) maintains that his intransitive sense of consciousness (his ‘Awareness’ in the special sense) corresponds to Burge’s phenomenality, whereas his transitive sense of consciousness corresponds to Burge’s phenomenal consciousness. Burge also distinguishes between phenomenal consciousness and access consciousness, since one

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

can be phenomenally conscious of some content without having access consciousness to it: ‘Certain complex, pre-attentive, pre-conscious categorizations occur but are not available to conscious report . . . At the level of conscious access, individuals are oblivious to what they perceive’ (Burge : ). Burge thinks that one can have both phenomenal consciousness and access consciousness of the conceptual content of empirical thoughts, since empirical thoughts take up the mode of presentation of the perceptual experience that gave rise to them: Correct conceptualization of a perceptual attributive involves taking over the perceptual attributive’s range of applicability and making use of its (perceptual) mode of presentation. (Burge : )

It should be noted that Burge’s access consciousness includes awareness on the part of the viewer that she is having an experience with such and such content, since without cognitive access individuals are oblivious to what they perceive. Block (a: ) thinks that the contents of phenomenal states of which one has phenomenal consciousness can be nonconceptual. At the same time, since CAC is PC plus reflection, one can be phenomenally conscious of contents of which one has CAC. This may be read in two ways. Either Block thinks that a viewer has CAC of the nonconceptual content of the experience as soon as this is conceptualized without this content being affected otherwise by conceptualization; the conceptualized content is the nonconceptual content plus concepts. In this case, the PC in the equation ‘access consciousness is phenomenal consciousness plus reflection’ is consciousness of nonconceptual content. Or, Block thinks that a viewer has access consciousness to conceptualized perceptual content that is, or may be, different from the nonconceptual content, in which case ‘phenomenal consciousness’ in the equation is used to signify visual consciousness in general, i.e. the consciousness involved in the visual modality irrespective of the nature of the conscious content. Burge (: –, n. ) holds that the perceptual representational content of which one has PC is not conceptual (nonpropositional), and that PC is the sort of consciousness one has of the nonconceptual content of experience. At the same time, he thinks that one can have PC of conceptual content. These entail that he uses PC to mean the visual consciousness associated with both a visual scene and the perceptual beliefs one forms about this scene; for Burge, one has access consciousness of the nonconceptual content to which concepts have been added. Adding to this his view that perceptual beliefs take over the mode of presentation of the phenomenal content, one draws the conclusion that, for Burge, the conceptualization of the phenomenal content does not change its phenomenology: the conceptualized content looks the same as the nonconceptual content. The relation between PC and nonconceptual content paves the way to relating Block’s and Burge’s notions of phenomenal and access consciousness to Dretske’s work. A caveat is needed first. Block employs PC to designate the phenomenology of perceptual experiential content (in addition to determining the phenomenal,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  non-representational character of experience); ‘the phenomenal neural correlate of consciousness for visual motion determines the experiential content of visual motion . . . That content itself is a kind of phenomenology, a kind of consciousness’ (Block b: ). Block claims that if there could be a phenomenal content without ‘anything that could be called awareness of it, one might not want to apply the term “consciousness” to it’—a view with which Block disagrees. Block uses here ‘awareness of content’ to mean content that one notices or realizes that one has it. Thus, awareness is stronger than consciousness, since one can be phenomenally conscious of some content without being aware of it. Viewers that possess linguistic capabilities can in principle report the contents of which they are aware in the strong sense, but they cannot report contents of which they are only phenomenally conscious. Note that the strong usage of awareness that goes beyond consciousness in the way explained above differs from Block’s (e: –) earlier usage of awareness as more or less equivalent to consciousness. There, Block had called PC ‘awareness’, and the fact that one may be both phenomenally conscious and access conscious of an event or object ‘conscious awareness’: Suppose that you are engaged in intense conversation when suddenly at noon you realize that right outside your window, there is—and has been for some time—a pneumatic drill digging up the street. You were aware of the noise all along, one might say, but only at noon are you consciously aware of it. That is, you were P-conscious of the noise all along, but at noon you are both P-conscious and A-conscious of it. Of course, there is a very similar string of events in which the crucial event at noon is a bit more intellectual. In this alternative scenario, at noon you realize not just that there is and has been a noise, but also that you are now and have been hearing the noise. In this alternative scenario, you get ‘higher-order thought’ as well as A-consciousness at noon. So on the first scenario, the belief that is acquired at noon is that there is and has been a noise, and on the second scenario, the beliefs that are acquired at noon are the first one plus the belief that you are now and have been hearing the noise. (Block e: )

Block says that when you become access-conscious of the noise, you realize that you are hearing a noise and a few lines below the cited passage he writes that at noon you are consciously aware of the noise. Evidently, he uses ‘conscious awareness’ to signify what I have called ‘awareness in the strong sense’. So, despite the difference in terminology, the motive underlying the distinctions is the same: the need to distinguish a form of consciousness, the PC, which does not require that a subject be aware (in the strong sense) that they have phenomenal consciousness or awareness. So, we can have PC of an event or object without being conscious aware of it. When we acquire access consciousness of the event by forming an implicit belief about it, we acquire conscious awareness (i.e. we acquire phenomenal awareness (PA), where ‘awareness’ is understood in the strong sense) of the event or object—i.e. we realize that we are phenomenally conscious of that event or object. It follows that a state that is CAC allows us to realize or notice that we are phenomenally conscious of the event or object. At the same time, this perceptual belief is a first-order belief that is not necessarily accompanied by a second-order belief to the effect that one has this

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

perceptual belief. The lack of the second-order belief means that, although we have come to realize that we were and are phenomenally conscious of the noise, we do not realize that we have the thought that we are now and have been hearing the noise. It follows that a perceptual thought that is CAC allows us to realize or become conscious aware of our phenomenal content, but we do not realize or become consciously aware that we are having a perceptual thought; we simply have it. Dretske (; ) uses ‘awareness’ more liberally, since he does not demand that in order for a person to be aware of X, they should be able to notice or realize that they see X. This is the weak sense of awareness. Dretske distinguishes ‘thing awareness’ from ‘fact awareness’. The former is perceptual awareness of an object or event, and is awareness of nonconceptual content; it is the awareness to which the experience owes its phenomenology. The latter is awareness that something is the case, and is awareness of conceptual content, since for Dretske (: ) the awareness of a fact is a perceptual belief. Dretske (p. ) relates thing awareness with phenomenal, nonconceptual seeing, and fact awareness with doxastic seeing. ‘Unless one understands the difference between a consciousness of things . . . and a consciousness of facts . . . and the way this difference depends, in turn, on a difference between a concept-free mental state (e.g. an experience) and a concept-charged mental state (e.g. a belief), one will fail to understand how one can have conscious experiences without being aware that one is having them.’ One can have thing awareness of X without believing or realizing that one sees X. Perceivers have fact awareness when they reflect on their perceptual content and they are aware of the fact that they have such content; they realize or notice that they have it. ‘Awareness’ in thing awareness is awareness in the weak sense, while ‘awareness’ in fact awareness is awareness in the strong sense. Dretske’s ‘thing awareness’, being the awareness responsible for visual phenomenology and the sort of awareness one can have without realizing or believing that one sees something—is similar to Block’s CAC.1 In this chapter, I use the term PC to signify the consciousness of the phenomenal, nonconceptual content of perceptual experience. Thus, PC is equivalent to phenomenal awareness as used by Hatfield (), Raftopoulos (), and Tye (; ), and to Dretske’s (; ) thing awareness. Since I hold that nonconceptual content is pre-attentive perceptual content, my PC is also equivalent to Lamme’s (; ) phenomenal awareness. Thus, I do not use PC as a generic term for visual phenomenology. I use the term ‘visual consciousness’ to signify the generic consciousness associated with the visual modality. Whenever I use the term PC as a generic term for visual consciousness, I will mark it with an asterisk (PC∗ ). Following Block, I use CAC to signify the consciousness of conceptual perceptual content that 1 It is debatable whether Dretske’s fact awareness is similar to Block’s CAC, although they both hold that fact awareness and CAC respectively are about conceptual contents. I will not discuss this problem here however.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  is available to the cognitive centers of the brain. As I have said, CAC content may be content that one does not notice or realize that one has; it is content of which one is not aware in the strong sense of the term. To put it differently, and in view of the fact that CAC states are implicit beliefs, a viewer may be unaware that they are having this belief or even that they are believing that they are having this belief. As I argue in Section , after  ms of visual processing, one becomes aware in the strong sense of the term of some of the conceptual contents that were thus far implicit, i.e. of which they had CAC only. In other words, some of the thoughts or beliefs that have thus far been implicit become explicit. Thus, this person acquires awareness of some of the implicit thoughts, which thus become explicit. This person notices or realizes that she is having these thoughts with such and such content. I call this sort of cognitive access ‘cognitive access awareness’ (CAA). To simplify things, in discussing the PC of some content, I restrict the scope of the terms to perceptual experience; I do not mean it to apply either to all phenomenally conscious states (such as feelings, and emotions) or to hallucinatory states.2 Let me also clarify that by attention I mean cognitively driven, endogenous attention, which is opposed to stimulus-driven, exogenous attention. Block (a; b: –), Burge (: ), Lamme (), and Raftopoulos () hold that attention determines the transition from PC to CAC, in the sense that attention is indispensable for phenomenal content to become cognitively accessible. When attention intervenes, the objects in phenomenal consciousness that were not selected leave their trace in consciousness, and one keeps being phenomenally conscious of them for some time. For Block, CAC is PC (or PC∗ , depending on which one of the two ways discussed above one reads Block’s claim) + reflection. Consequently, once attention intervenes, the content of which one has PC or PC∗ becomes content that can be accessed by cognition; it becomes cognitive content. If one reads Block as holding the view that one can have CAC of nonconceptual content of which one has PC, and that what happens is that attention and conceptualization just add concepts to

2 I do not think, however, that the definition of PA as awareness of some object or event raises any real difficulties with respect to the case of hallucinations. I do not discuss the reasons for this here except to say two things. First, I think that hallucinations may involve perceptual awareness of something, specifically awareness of proto-object bound together inappropriately in such a way that the subject may experience an object that is phenomenologically of exactly the same kind as an object of genuine perception, or she may experience an object that does not exist in the environment (say, a dragon or an alien creature). Second, if hallucinations do not involve awareness of a real object, I would agree with Tye’s (; ) and Smith’s () representationalist or intentionalist respectively accounts of hallucination, in which the objects of hallucinations are intentional objects. ‘Although Macbeth did not see a dagger, by contrast it is true to say that what he saw was shaped thus and so, had such and such color or luster, and so on for all sensible qualities perceptible by sight. If, therefore, we re-phrase Macbeth’s question as ‘is this dagger-shaped thing I see before me?’ intentionalism can give an affirmative answer.’ (Smith , ) Note that although Block is not a representationalist in the sense that he thinks that qualia as properties of neural states have a role to play in accounting for visual phenomenology, Block also accepts that perceptual experiences have a representational character as well.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

the nonconceptual content rendering it suitable for CAC, Block is committed to the thesis that attention does not affect nonconceptual content and does not change phenomenology; pre-attentional and attentional perceptual contents look the same. If one reads Block as espousing the view that attention changes the nonconceptual content of the experience as it transforms it to conceptual content and affects its phenomenology, then, for the equation to hold, Block’s phenomenal consciousness should be PC∗ , since PC∗ designates visual consciousness in general. In this case, pre-attentional content and its phenomenology differ from attentional content and its phenomenology. Several commentators think that Block takes the first option; ‘Block’s argument presupposes that visual consciousness itself remains static during episodes of cognitive access’ (Stazicker : ). Things are clearer with Burge. His statement ‘the correct conceptualization of a perceptual attributive involves taking over the perceptual attributive’s range of applicability and making use of its (perceptual) mode of presentation’ entails that conceptualization does not affect phenomenology. Nonconceptualist philosophers (Dretske ; ; Hatfield ; Jackendoff ; Tye ; ) who hold that PC is awareness, either in the weak or the strong sense, of nonconceptual content, and that this awareness exhausts the visual phenomenology of a scene, have two options available to them concerning phenomenology when attention starts modulating perceptual processing. They could claim that attention and the concomitant conceptualization do not change phenomenology, because if they did, PC would not exhaust visual phenomenology and something more would be needed to account for the visual phenomenology associated with conceptual visual states. Or, they could claim that once attention intervenes and conceptualization occurs, the resultant states are not visual but an amalgam whose phenomenology is not visual but something else (Tye : ), or go further and, following Jackendoff (), argue that perceptual beliefs are not characterized by visual phenomenology but by visual understanding. In this chapter, I argue that attention affects visual phenomenology. Since attention’s role signifies the onset of the cognitive penetration (where concepts enter the contents of perception) of the hitherto cognitively impenetrable perception, cognitive penetration changes the phenomenology of a visual scene; the scene before cognitive penetration looks different from the scene after cognitive penetration. Consequently, using a single term to signify the visual consciousness of a visual scene is not sufficient for the following reasons. Philosophers who think that PC is a generic term designating visual consciousness (PC∗ ) are divided into two categories: those who hold that attention does not change PC∗ , and those who think that although attention affects phenomenology, the fact that there is phenomenology associated with the experience warrants the usage of PC∗ for the phenomenology of a visual scene. The first category faces the problem that evidence shows that attention changes phenomenology. The second faces the problem that it uses a single term to designate two different sorts of phenomenology, which may create confusion.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  Philosophers who associate PC with nonconceptual content face a problem with the phenomenology of the contents of perception that are conceptually modulated/cognitively penetrated on the assumption that contents that are conceptually modulated are perceptual contents. If PC concerns nonconceptual content, apparently a new term is needed to designate the visual awareness of conceptually affected perceptual content. I introduce ‘conceptually modulated visual consciousness’ (CMVC) to denote the visual consciousness of the conceptual, perceptual content of late vision. In this introduction, which serves as the first section of the chapter, I have presented and discussed the terms PC, PA, CMVC, CAC, and CAA. In the second section, I briefly present early vision, a non-attentional, cognitively impenetrable visual stage. I then turn to late vision, and argue that in late vision there is conceptual modulation of visual processing through attentional effects. To counteract the view that the conceptually modulated part of vision is not related to visual awareness but to visual understanding, I claim that although late vision culminates in a perceptual belief, this belief is not the result of a discursive inference,3 that it is inextricably linked to a visual scene, and that phenomenality is constitutively a part of the perceptual belief that late vision outputs. In the third section, I argue for the view that in late vision attention, and thus cognitive penetration, affect phenomenology, and that the conceptual content formed in late vision differs from phenomenal, nonconceptual content since it presents the world differently. Finally, in Section , I elaborate on PC, CMVC, CAC, and CAA, and the contents associated with them. I argue that CAC and CAA emerge in late vision and that there are contents of which one can have (a) CMVC and CAC; (b) CMVC and CAA; and (c) only CAC or CAA without CMVC.

 Early Vision and Late Vision . Early vision Early vision includes both a feed-forward sweep of signal transmission in which signals are transmitted bottom-up and which lasts, in visual areas (i.e. up to inferior temporal cortex (IT)) for about  ms, and a stage at which lateral and recurrent connections between neurons allow local recurrent processing (Lamme ). This sort of recurrent processing starts at – ms, is restricted within visual areas, does not involve signals from cognitive centers, and culminates at about  ms. The unconscious feed-forward sweep extracts high-level information that could lead to categorization, determines the classical receptive field of neurons and their basic tuning properties, and results in some initial feature detection. When local recurrent processing starts, further binding and segregation occur. The representations formed

3 I use the term ‘discursive’ to distinguish ‘inferences’ as understood by philosophers from ‘inferences’ as understood by vision scientists who think of any transformation of signals carrying information according to some rule as a form of inference.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

at this stage are restricted to including information about the observable features of objects, i.e. information about spatiotemporal properties, surface properties, viewercentered shape, color, texture, orientation, motion, and affordances, in addition to the representations of objects as bounded, solid entities that persist in space and time. I call the representations of the objects in early vision ‘proto-objects’. When local recurrent processing starts, PC arises (Block a; b; d; Lamme ; ; Raftopoulos ). By being restricted within the visual system and by not involving signals from the cognitive areas of the brain, the feed-forward sweep and local recurrent processing are cognitively impenetrable or conceptually encapsulated. This leads to the thesis that early vision contents are nonconceptual provided that concepts4 do not figure inherently in the perceptual system—a provision that I have rejected elsewhere (Raftopoulos ; ). During early vision, multiple stimuli are all represented in parallel and there are no attentional effects. Neurophysiological evidence for this thesis comes from various findings (discussed in Raftopoulos : ch. ) that suggest that the first signs of cognitively driven attentional effects on visual areas up to V occur after  ms. Early vision, therefore, is a pre-attentional stage that is cognitively impenetrable in that its processes are not directly affected by cognitive signals. A ‘direct’ explication of ‘pre-attentional’ is needed. First, my claim does not entail that there is no selection during early vision, as it is well known that there are non-attentional selection mechanisms (Lamme ) that filter information before it reaches awareness. These mechanisms are not considered to be attentional, because they occur very early and do not involve higher brain areas associated with the attentional mechanisms (prefrontal cortex, parietal cortex, etc.) Second, ‘pre-attentional’ should be construed in relation to cognitively driven attention that affects perceptual processing directly, the claim being that early vision involves processes that are not affected directly by attention. Cognitively driven attention is opposed both to stimulus-driven attention and to the effects of either spatial or feature/object cueing before stimulus onset. This phenomenon is referred to as ‘the attentional modulation of spontaneous activity’, or as ‘the preparatory effect’. I have discussed this extensively elsewhere (Raftopoulos ; ; forthcoming) and I will only repeat here the main conclusion. The cueing effects are restricted to affecting the tunings of some neurons and facilitate perceptual processing either from selected locations or for selected stimuli before stimulus presentation. Once the stimulus is presented, there are no top-down cognitive effects registered in visual areas for the duration of early vision. The effects of pre-cueing reflect a change in background neural activity that biases visual processing; that is, they set the initial values of some parameters figuring in the equations 4 Concepts are constant, context independent, and freely repeatable elements that figure constitutively in propositional contents; they correspond to lexical items. Concepts are items that have meanings and are not themselves meanings as in the Fregean tradition.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  guiding state transformations. These biases, which are top-down driven, rig-up visual processing since they are anticipatory and occur before stimulus presentation but they do not emerge as part of perceptual competition and in this sense they are not intrinsic to perceptual processing (Nobre et al. : ) which is otherwise unaffected by topdown effects. Thus, the pre-cueing attentional effects constitute a case of rigging-up the early visual processing; they are not cases of top-down cognitive effects on it. This sort of modulation of early vision is an offline cognitive effect in that it affects early visual processes before these start to take place. Once the processes of early vision start, there are no top-down cognitive effects that transfer conceptual information that is used by the processes of early vision. It should be stressed that ‘early vision’ as used here does not refer to neuroanatomical areas of the brain, but is a computational, temporal, representational, functional definition of a stage of visual processing. It cannot refer to neuroanatomical areas of the brain because all areas of the visual brain from LGN onwards are eventually penetrated in a top-down manner from higher-level cognitive signals. It is temporal because it holds that early vision lasts for a certain amount of time during which visual processing is not cognitively penetrated. It is functional/computational in that it relates visual processing with two sorts of processes in the brain, namely the feed-forward sweep and the local recurrent processing. Finally, it is representational because it relates the states formed during it with a certain type of content, nonconceptual content. This delineation of early vision permits the claim that early vision is cognitively impenetrable despite the fact that the neural sites at which its processes take place are at a later time imbued with top-down, cognitive information. One could object at this juncture that in normal circumstances, when visual processing is being continually updated by new sensory input, it does not make sense to claim that some of this processing is not being cognitively penetrated, when the areas doing that processing are receiving input from cognitive areas of the brain on account of the processing of previously received information. The answer in a nutshell is that during late vision the role of top-down cognitive signals, mediated by attentional top-down control, is to test the various hypotheses formed concerning object identity. They do that by taking advantage of the rich iconic information about the visible properties of the objects stored in early visual areas (see Raftopoulos  for a detailed account). As a result, the activations of the neurons in the early visual areas that store this iconic information are enhanced owing to the top-down modulation. It is important to note that the cognitive modulation concerns only those neuronal assemblies that store information concerning the particular visual scene. When new information enters the retina, as in realistic conditions is always the case, the new information is stored in different neuronal assemblies, since the previously engaged assemblies continue to store the initial information for up to  ms (iconic memory). The new neuronal assemblies, however, have not been affected by the topdown cognitive signals that modulate only the assemblies that store information about the initial visual scene.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

. Late vision Starting at – ms, signals from higher executive centers including mnemonic circuits intervene and modulate perceptual processing in the visual cortex, and this signals the onset of global recurrent processing. Within  ms, for example, poststimulus parietal areas in the dorsal system but also areas in the ventral pathway (IT cortex) semantically process the low spatial frequency information and determine the gist of the scene based on stored knowledge that generates predictions about the most likely interpretation of the input even in the absence of focal attention (Barr ). This information re-enters the extrastriate visual areas in the IT and modulates (at about  ms) perceptual processing by facilitating the analysis of high spatial frequency information—by specifying, for example, certain cues in the image that facilitate target identification. At about  ms, hypotheses concerning the identity of objects in the scene are formed, and are tested through the detailed iconic information stored in early visual circuits, mainly in V (Peyrin et al. ). This leads to the recognition of the object(s) in the scene. This is signaled by the P ERP waveform5 at about  ms in the IT cortex, whose neurons contribute to the integration of information of low and high spatial frequencies. Thus, cognitive effects on visual areas manifest themselves at  ms. Recurrent interactions with areas outside the visual areas allow storage in visual working memory and give rise to global recurrent processing, which allows standing knowledge (information stored in long-term memory in the synaptic weights of the neurons) to modulate, when activated, visual processing which up to that point was nonconceptual. During global recurrent processing, the conceptualization of perceptual content starts and the contents of the states formed in this stage are conceptual. Global recurrent processing allows the emergence of CAC and CAA by rendering the content of the perceptual states of this stage available to the cognitive states. What determines the transition from PC to CAC is attention when it affects visual processing (Block a; b; Burge : ; Lamme ; Raftopoulos ). The work of Lamme and colleagues shows that CAA (awareness in the strong sense) of contents arises at  ms after stimulus onset, a result that is supported by studies of attentional blink in which the P is elicited at latencies of about – ms (Hopfinger et al. ) and by studies correlating behavioral visibility ratings and recordings of ERPs (Sergent et al. ). The P component indexes explicit detection, and thus conscious awareness (Evans et al. ; Niedeggen et al. ). Let me draw attention to one of the repercussions of the discussion of the processes of late vision to counteract one possible objection to the view that nonconceptual contents and cognitive-access conscious, conceptual contents are different. One might 5 The P waveform is elicited at – ms and is generated in frontal/central, central/parietal, parietal/occipital areas, the temporal lobe, the temporal/parietal junction, and neighboring neocortical regions. The generating sites and timing onset show that P is associated with semantic processing and with the subjects’ reports. P is thought to signify the consolidation of a representation in working memory.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  claim that cognition grasps nonconceptual content and conceptualizes it by bringing it under the auspices of cognition, outputting eventually some perceptual beliefs— a claim that is compatible with Block’s and Burge’s views on the relation between CAC and PC. CAC refers to episodes of thinking or reflection on the contents of one’s perceptual experience. That is, perceptual nonconceptual contents are brought before the mind that reflects on them. One could argue that the contents of PC and CAC are the same; the difference is at the level of access/cognitive uptake, not at the level of what is represented. It is plausible that the cognitive-access conscious content is just nonconceptual content accessible for purposes of conceptualization. The thesis underpinning this objection is erroneous for at least two reasons. First, since conceptual engagement requires attentional modulation of visual processing, the thesis presupposes that attention does not change the representations and the phenomenology of perceptual content; it just renders the phenomenal content available to cognition. As we shall see, this is wrong. The contents of PC, as determinates and not determinables, and the contents of CAC cannot be the same. Second, CAC is made possible by global recurrent processing, which involves among other things working memory. Since concepts are constituents of thoughts that are occurrent states and are formed in working memory, concepts are representations formed in working memory. When working memory is at work, concepts are activated and affect the perceptual processes top-down. Thus, the ensuing perceptual content is conceptualized content. The content that is accessible to the cognitive uptake is content that has been conceptually modulated. As such, CAC content includes semantic, non-visible information (class membership, for example). Nonconceptual content, however, is restricted to the visible properties of the objects in a visual scene. It follows that the two contents are different; nonconceptual contents are not simply brought before the mind that by reflecting on them renders them CAC contents. Let us return to the contents formed in late vision. Global recurrent processing marks the beginning of the conceptualization processes of the nonconceptual content that early vision outputs. Perceptual contents, by being conceptualized, become available to, and thus are accessed by, cognition; one has a CAC state, a perceptual belief. CAA, which is the awareness (in the strong sense) of conceptual contents, arises at a later stage in late vision, specifically with the onset of P; a belief is a CAA state if viewers notice or realize that they have it. For example, a viewer has CAA that a red cat is in front of her, which means that she forms the explicit perceptual belief that there is a cat in front of her. The belief being explicit is reportable. The formation of this state presupposes that the animal in the visual scene has been definitely recognized as a cat and the color has been identified as some shade of red. For this to happen, as we saw, various implicit hypotheses/beliefs are formed and tested before a definitive recognition of the object in the visual scene and of its properties occurs once one of the tentative hypotheses is accepted. These implicit hypotheses are constructed when perceptual information becomes available to cognition, which uses it to form the tentative hypotheses regarding the identity of the visual objects. Thus, the contents of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

the relevant late vision states are CAC, and a viewer forms implicit beliefs concerning the possible objects in the visual scene. There is thus a significant amount of time during which implicit beliefs, i.e. CAC states, are formed in late vision without viewers being aware that they have these beliefs, i.e. without the viewer having CAA of them. This last remark brings us to another issue worth discussing. If late vision is conceptually modulated and some of its contents have propositional forms, and if the role of concepts and stored knowledge consists in providing some initial interpretation of the visual scene through the determination of its gist, one is tempted to say that this stage relies heavily on inferences (this is what hypothesis testing amounts to) and thus differs radically from the processes in early vision. In this sense, it may be better to construe late vision not as a visual stage but as a discursive stage involving thoughts. Jackendoff () distinguishes visual awareness from visual understanding. Visual awareness concerns representational contents associated with Marr’s /D sketch, which is a viewer-centered representation of the visible surfaces arrayed in depth. The D sketch, which includes the unseen surfaces not represented in the /D sketch, is a result of an inference and is not visually experienced. One is aware of the D object or of category-based representations but this is not visual awareness; it is visual understanding. Similarly, Dretske (: ) distinguishes between a concept-free mental state, an experience, and a concept-charged mental state, a perceptual belief. On this view, perception properly speaking is restricted to that stage of visual processing that is not conceptually modulated. There is no visual awareness of conceptual content, only visual understanding. Late vision corresponds to doxastic seeing and involves thoughts and inferences, and is thus a discursive stage resulting in visual understanding. Against this thesis, I have argued (Raftopoulos ) that the sorts of processes and contents of which late vision is constituted suggest that late vision is not a discursive stage characterized by visual understanding and the drawing of discursive inferences, but a visual stage that usually outputs a perceptual belief and that has its own phenomenology, i.e. it involves visual awareness. Cognitive transformations are not necessarily inferences rules, although they can be described in terms of inference rules; therefore, even though visual processes can be described in inferential terms this does not mean that they involve inference making. The states of late vision that have propositional contents, despite their conceptual nature, are not pure perceptual beliefs but experiences that have visual awareness. Perceptual beliefs formed in late vision during the time of the perceptual encounter are not a pure, context-free thought but are inextricably connected to the object of perception and are tied to the idiosyncratic viewpoint of viewers by making use of their physical presence and occupation of a certain location in space-time; this makes these thoughts essential indexicals (Raftopoulos ; Raftopoulos and Muller ). The inextricable link between thought and perception in late vision explains the essentially contextual character of perceptual beliefs, in Perry’s () and Stalnaker’s (: –) sense. The proposition expressed by the belief cannot be detached from the perceptual context in which it is held, and cannot be reduced to another belief in

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  which some third-person content is substituted for the indexicals in the thought. The inextricable link with a perceptual context means that late vision involves essentially visual circuits, and thus an inherent visual phenomenology.

 Attention and Phenomenology Let me now turn to the differences in the phenomenology of a visual scene that result from the attentional modulation of perceptual processing. An example of the difference between PC on the one hand and CMVC and CAA on the other of the same scene comes from the experiment of Carrasco et al. () that demonstrates how attention can alter the mode of presentation of a visual scene to a viewer. Carrasco et al. presented their participants with two oriented gratings and asked them to report the orientation of the lines of the highest-contrast grating. There were two conditions in the experiment. In the first, attention was not focused. In the second, a dot preceded one of the two gratings, automatically attracting attention to it. Since participants were asked to report on the orientation of the grating, their decision concerning which grating was higher in contrast was implicit rather than explicit. When the difference in contrast between the two gratings was large, the presence of the dot had no effect. When the two grating had similar contrasts, in the non-attentional condition the participants did not make any choice, which means that they visually experienced both gratings as similar in contrast. However, in the attentional condition the participants tended to report the orientation of the grating that was preceded by the dot and was thus attended. This means that the participants visually experienced the attended grating as having higher contrast than the other grating even though the two gratings had similar contrasts. Attention changed the appearance of the attended grating by increasing its apparent contrast; to put it differently, attention changes the phenomenological experience of the contrast, making bright lines appear brighter and dark lines appear darker. Extensive literature on the effects of feature-based attention on the processing in early visual areas suggests that attention affects the phenomenology of colors at later stages of visual processing owing to the top-down effects of attention (Muller et al. ). In general, feature-based attention can change the phenomenology of a visual scene in two ways. One is through top-down modulation of visual processing during late vision—an effect that is delayed in time since, as we have seen, attention starts modulating visual processing at about  ms at the extrastriate cortex, predominantly in feature-sensitive V in the ventral pathway and MT in the dorsal pathway, and at about  ms at V (Lamme and Roelfsema ). Thus, the effects of feature attention within the relevant feature-sensitive area start several hundred ms after the subjects perceive in early vision a particular feature. The other way the phenomenology of a visual change can change is through the effects of cues that are presented in advance of the stimulus. As we have seen, pre-cueing facilitates the feed-forward sweep of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

the pre-cued feature and, hence, one should expect re-entrant signals concerning the attention on this feature to enhance activity in the areas of V sensitive to this feature earlier than the re-entrant signals concerning the attention to other features. In this case, the phenomenology of the scene is different from the phenomenology of the same scene without the cue. Another example of the effects of attention on visual phenomenology concerns perceptual constancies, a term used to describe the fact that in perception the properties of objects that are perceived tend to remain constant in consciousness although the viewing conditions may change. Even though the retinal image of the objects in the environment changes as we move around them, or as they move with respect to us, our perception of some of their properties remains relatively constant. If the perceptual shape or size constancy is due to cues that are available in the retinal image, shape or size constancy can be achieved by purely visual information and occurs automatically. The viewer has PC of the same shape or size despite differences in viewing conditions. If the perceptual shape or size constancy cannot be effectuated through visual information and conceptual sources are needed, the shape or size constancy is achieved in late vision. In this case, the viewer has CAA and sees the same size or shape in the doxastic sense of seeing (i.e. the viewer ‘judges’ that the shape or size is the constant despite changes in viewing condition). This entails that if purely visual information is insufficient for perceptual constancies and nonvisual information is needed to that effect, if the nonvisual information cannot be brought to bear (as where attention is diverted elsewhere), the viewer should be aware of changing shapes or sizes. The role of knowledge in perceptual constancies is evidenced in Granrud’s () research, which shows that younger and older children and adults perceive, for distances longer than  m, distant objects as smaller than their actual size. Thus, with respect to the way things look in perception, viewers do not achieve size constancy at these distances. Older children and adults, however, know that distant objects are misperceived and they supplement perception with judgments that allow them to estimate accurately the physical size of the objects. They achieve perceptual constancy in the object mode, i.e. they can report the actual physical size of the distant object despite the fact that the object looks smaller, and show that they can distinguish between how things are in the world as opposed to how things seem in their sensation. This suggests that developmental changes in the accuracy of size estimation of fardistant objects result from the development of reasoning. For distances up to  m, size constancy is achieved automatically without any cognitive involvement on the basis of visual cues only. Granrud (: ) thinks that at short distances, depth cues such as ocular convergence, binocular disparity, and motion parallax provide precise information that can be used for accurate size constancy. For longer distances, these same cues do not provide accurate distance information to support size constancy, which breaks up, or is achieved by cognitive means that dictate effective strategies; for long distances, size constancy is the result of a judgment. This research suggests that for longer distances size constancy is not really

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  a phenomenon that concerns visual phenomenology but a phenomenon that is within the realm of perceptual understanding, what Hatfield (: –) calls awareness of the cognition, because the viewers do not have a visual experience as of a same size but instead judge that the objects have the same size despite the fact that the objects look different in size. In such cases, size constancy is not accompanied by the visual phenomenology of two equally sized objects but is a purely cognitive effect that assumes the form of doxastic seeing. Note, however, that the viewer is made cognitively aware of, can think of, and can report the fact that the two objects look different. That is, the viewer notices their differing phenomenal sizes since the conceptualization process has not changed the visual representation of the sizes of the objects, although attention has increased their contrasts. Therefore, the pre-attentional nonconceptual content of which one has PC, and the attentional conceptually modulated content of which one has CMVC, coincide with respect to the values of the property ‘size’. The two objects look in perception to have different sizes both in early vision and in late vision because the way the sizes of the objects look is not affected by attention. The overall phenomenology accompanying the two experiences, however, is not the same because contrasts differ and because, as we shall see, the conceptual perceptual content has a stability and coherence that the pre-attentional content lack. In such cases, the output of late vision is the perceptual belief that the two objects have different sizes, and this belief is accompanied by the phenomenology of two differently sized objects. The viewer, however, because of her knowledge, abstains from endorsing this perceptual belief and forms the perceptual judgment that the two objects have the same size. The role of cognitive effects in determining constancies has also been shown by Epstein and Broota () in a size constancy study. This time, unlike what transpires in Granrud’s experiments, knowledge has a phenomenological impact on the perception of size in that it changes through attention the perceptual experience of size. While Granrud’s experiments show that size constancy is not a visual phenomenon since it breaks down in the participants’ visual experience and surfaces in their perceptual judgments, in Epstein and Broota’s experiment the phenomenology of the participants’ experience is affected by their knowledge, and size constancy is achieved at the level of visual awareness as this is affected by attention; when attention intervenes, subjects visually experience size constancy. Epstein and Broota asked the participants to hold their right hand in front of them, to look directly at a finger, and to start moving the hand towards and away from their face. Although the retinal image of the finger changes as the hand moves, the finger appears to have the same size because of the operation of size constancy. Thus, the participants are visually aware of the size constancy. Since there are no visual cues that could explain the constancy, size constancy is the effect of the familiarity with the size of the hand. This being a cognitive effect, size constancy takes place in late vision. It follows that the visual awareness involved is awareness of conceptual content or CMVC, because it is the knowledge of the fact that the fingers do not change size

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

that modulates perceptual processing and enables size constancy. The participants also form the belief that their finger has a constant size and report it. Now, when the participants hold a finger in front of their face and look steadily at the finger while they are moving their hand, if they try to look at the moving hand without moving their eyes from the finger, size constancy breaks down and the farther the hand is from their face, the smaller the hand looks, although the experience is described as fleeting and unstable. Thus, when attention is directed elsewhere and the size constancy operations fail, one is made aware of a percept that corresponds directly to the size of the object in the retinal image. Therefore, one can be aware of the pre-constancy size when the constancy is achieved by cognitive mechanisms that intervene late in visual processing (i.e. when the visual system does not automatically achieve constancy through purely visual cues) and the conditions are such that the mechanisms fail in their task, as when attention is diverted elsewhere. Since the preattentional stage of visual processing is early vision, the pre-constancy sizes are the contents of the stages of early vision of which one has PC. This is one of the rare occasions in which careful experimentation allows us to have cognitive access to the nonconceptual, pre-attentional contents of our perceptual experience—i.e. to the contents of which we have PC—and to be able to report them. Notice that in these cases, the PC contents differ from the contents of which we have CMVC. A final example of the effects of attention/cognition on visual phenomenology concerns color perception. The bipolar cells in the retinal ganglion layer and the lateral geniculate nucleus that receive input from the cone cells measure differences between red (L) and green (M), blue and yellow (differences between L plus M and S), and black and white (the sum of L and M). These processes explain the visual system’s ability to detect wavelengths, but they do not explain conscious representation of colors. People with achromatopsia, a condition resulting from a defect to extrastriate areas in the neighborhood of V that inhibits chromatic color vision, have no conscious hue experience (Heywood and Kentridge ) but they detect wavelengths. Whether they are also conscious of this information is another problem that I will pursue here. There is evidence suggesting that the neural correlates of conscious hue experience may be further upstream in the brain, i.e. regions in the inferior convexity of the temporal lobe are critical for color processing. In this vein, studies show that neurons in the posterior part of the superior temporal sulcus are responsive to specific hues (Komatsu et al. ) as well as after-image activity (Conway and Tsao ). These results show that attention modulates neural activity when color information is behaviorally relevant, leading to a recruitment of higher neural areas. The results are consistent with the view that in passive and relatively inattentive viewing only general features of the color stimulus are determined, such as approximate wavelength and relative local contrast. The determination of a specific hue, however, occurs when cognitively driven attention modulates perceptual processing and requires processing in higher neural regions. Thus, it is the synergy of top-down modulation and activity

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  in the visual areas that results in the conscious perception of hues. As Brogaard and Gatzia (forthcoming: ) remark: ‘The experiences of hues, unlike the experience of an approximate wavelength and a relative contrast, are high-level and can be compared with experiences of faces the content of which is computed by ventral areas in the close vicinity.’ Thus, it seems that attentional control changes the content of the perception of colors because only when attention intervenes do viewers have conscious perception of hues, as opposed to perception of approximate wavelengths and local contrasts; attention affects the phenomenology of colors. The role of attention in visual processing is not restricted to changing the ‘values’ of the determinable properties of object in a visual scene (e.g. increasing contrasts, or achieving size constancy). It also concerns the way some of the generic properties of the objects of perception appear in consciousness, because attentional selection brings with it a profound difference in the way things are experienced. The ‘phenomenal’ pre-attentive percept consists of only tentatively but uniquely bound features that are segmented from visual information and are candidate objects for further processing (Driver et al. ; Lamme ; ; Wolfe et al. ). Similarly, in Rensink’s () theory, although proto-objects that are formed in the pre-attentive stage can be complex structures, they are coherent only over a small region and over a limited amount of time; they have limited spatial and temporal coherence (where ‘coherence’ means that the structures refer to parts of the same system in space and time— in other words, that their representations in different locations and over different times refer to the same object). Proto-objects are very volatile, in that they are either overwritten by subsequent stimuli or else fade away within a few hundred ms; the volatile representations last for about half a second. Each time the eyes move and new light enters the eye, the proto-objects fade away and new ones are generated. This information is a short-lived, vulnerable, and not easily reportable form of visual experience (Lamme : ). In the absence of attention, only some fleeting unstable representations of objects can be constructed that contain sparse information. This explains the unstable and fleeting character of the experience of the pre-attentional object corresponding to the image in the retina described in the Epstein and Broota experiment. In this section, I presented evidence that supports the following conclusions. First, attention changes not only the values of the properties of objects or surfaces such as contrast or size as they appear in the phenomenology of the experience, but also the coherence and stability of the visual representation, which affects the way contents are experienced. Second, cognitive factors may cause a viewer to form a visual judgment that the world is different from how it looks in perceptual experience—a judgment that negates the perceptual belief that late vision outputs, i.e. the belief formed when one takes perception at face value (Granrud ). Third, occasionally it is possible to have a fleeting, unstable experience of the pre-attentional content of perception, i.e. of the phenomenal content of perception before it is affected by the viewer’s knowledge.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

 Forms of Consciousness and Perceptual Contents As I said in the introduction, some philosophers construe PC as the consciousness related to the phenomenology of a visual scene in general, and think that it can be consciousness of both nonconceptual and conceptual perceptual content. Specifically, philosophers who hold that perception has, in addition to nonconceptual content, a conceptually structured content think that one has CAC of this content, or, equivalently, that one has implicit beliefs with this content (Block a; b; c; Raftopoulos ; Burge : –). Block (a; b; c), Burge (: – ; ), and Horgan and Tienson () claim that one can have PC of conceptual content too, and thus use PC as a generic term for visual consciousness (which I have called PC∗ ). Burge also thinks that when the content of perception is conceptualized, its phenomenology does not change, and Block may hold this view too (see the discussion in the introduction). Burge (: ), in particular, argues that perceptual thoughts constitute cases in which access is overdetermined because ‘thoughts are both phenomenally and access conscious’—a view shared by Block (e: ): ‘you were aware of the noise all along, one might say, but only at noon are you consciously aware of it. That is, you were P-conscious of the noise all along, but at noon you are both P-conscious and A-conscious of it.’ All these philosophers distinguish PC (or PC∗ ) from CAC on the ground that one can have PC of some content without having CAC of it. Attention is thought to determine the transition from PC to CAC, and to mark the transition from a stage of visual processing that is restricted within visual areas to a stage that is cognitively modulated. Thus, it is commonly held that attention is required so that the contents of which a viewer has PC become accessible to her cognition; attention marks the onset of the conceptualization of the hitherto nonconceptual content (Burge : ; ; Block b: , ; Dretske ; ; Lamme ; ; Raftopoulos ). Block (a: ; e) and Burge (; ) think that reflexive consciousness is phenomenal consciousness plus reflection. The content of a phenomenal state is reflexively conscious just in case it is also the content of one of the subject’s thoughts, which entails that reflexive consciousness is synonymous to CAC. As I explained in the introduction, this is usually taken to entail that CAC makes available to cognition the perceptual content of which one has PC—which means, in turn, that when attention intervenes and renders CAC possible the phenomenology of the experience remains unaltered. In view of the evidence that attention alters phenomenology, one could come to Block’s rescue and argue that he uses phenomenal consciousness as a generic term for the phenomenology of a visual scene irrespective of the nature of the perceptual content of the relevant perceptual states. In this case, Block’s equation CAC = PC∗ + reflection does not entail that attention does not alter phenomenology. It simply states that one keeps having some sort of visual consciousness when attention intervenes. Underneath this construal of the relation between PC∗ and CAC is the view that as

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  ‘experience’ is used, all experiences are phenomenally conscious, and so they all are cases of PC∗ ; some of them are cognitive access conscious and some are not. The distinction between PC∗ and CAC is a distinction between the ways a visual scene looks to a viewer and the ways the viewer thinks of the content of this same scene. This view is compatible with the empirical finding that attention alters phenomenal content, insofar as the new content is still content with some phenomenology. However, given that attention affects phenomenology, there remains the problem that a single term, PC∗ , is used to designate two different forms of phenomenology. Here is another way to explain the problem faced by this group. One could ask, ‘Why could not we remain phenomenally conscious of some contents that then become cognitively accessible? If this is the case, why do we need to introduce a third form of consciousness?’ Let us suppose that by ‘phenomenal consciousness’, one means visual consciousness in general, i.e. PC∗ . There are two ways to read this question. According to the first, which reflects Burge’s and perhaps Block’s views, this question presupposes, as the term ‘remain’ shows, that phenomenal contents do not change when they become available to cognition. This is wrong: attention changes perceptual content and its phenomenology. The content of which one has PC cannot be the content available to their cognition; it is a different content in many respects. The point now is the following: if the phenomenology of attended content is different from the phenomenology of non-attended content, we need to distinguish between the two sorts of phenomenology. According to the second way of reading the question, it presupposes that when some content becomes cognitively accessible it is still content of which one is phenomenally conscious without any commitment that the phenomenology before and after the availability to cognition should remain the same. It can change, since all that matters is that the cognitive access content is content of which one has visual consciousness. In this case, the problem is that if, as the evidence shows, nonconceptual and conceptual contents are different and their phenomenology is different, it would be good to distinguish between the two sorts of phenomenology. Other philosophers relate PC with the phenomenology of the nonconceptual content of perception. In this case, PC is described as the awareness of just having some perceptual content, as when one is aware of an object in one’s visual field without being aware in the strong sense of the term of the object (Dretske ; ; Hatfield ; Raftopoulos ; Tye ; ). These nonconceptualists tie PC to the nonconceptual content of perception, and think that one cannot have PC of content of which one has CAC because the latter is conceptualized content. This view relates PC with nonconceptual content and CAC with episodes of thinking on the contents of perceptual experience, and does not address the issue of the visual phenomenology of conceptual content. The reason is that these philosophers think that once concepts start modulating perceptual processing, either the ensuing states are mixed conceptual and nonconceptual states that do not belong to perception properly speaking and have a mixed phenomenology which is different from PC

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

(Tye ), or the ensuing states are not perceptual at all but rather episodes of thoughts (Dretske ; ; Jackendoff ), and therefore that what characterizes them is visual understanding rather than visual consciousness. Thus, PC and CAC characterize nonconceptual and conceptual content respectively. When attention intervenes, the phenomenology of the visual scene is not affected but the visual content is transformed to some other sort of mixed or to a non-perceptual content with no inherent phenomenology. Against this view, I claimed in Section  that late vision—the visual stage that is modulated by attention and concepts—should not be construed as thought-like but as a perceptual stage with its own visual phenomenology, that of conceptual content. If we use PC exclusively for nonconceptual content, we have to introduce a new term to cover the phenomenology of conceptual content. To this end, I use the term ‘conceptually modulated visual consciousness’ (CMVC). Since attention changes visual phenomenology by changing perceptual contents, the contents of which one has CMVC differ from the nonconceptual contents of which one has PC. PC, as I use the term, emerges during early vision and makes us visually conscious of the nonconceptual content of perception, and therefore is neatly mapped to early vision contents. Things are more complicated with late vision, and an examination of the relation between the three forms of consciousness that emerge in late vision, namely CMVC, CAC, and CAA, is needed. Recall first that CAC is consciousness of conceptual perceptual content (a CAC state is a belief), that is accessed by cognition without the viewer being aware of it in the strong sense of the term, i.e. without the viewer noticing or realizing that they have this belief (this belief is implicit). CAA concerns beliefs formed at about  ms post-stimulus and of which the viewer is aware in the strong sense and thus can report them; they are explicit beliefs. Late vision starts at about  ms before objects in the scene are recognized and consolidated in memory. Since object recognition takes about  ms and since CAA of an object, i.e. the formation of an explicit belief concerning it, presupposes a representation that has been consolidated in working memory, one has CAA of that object long after conceptual processing started. Since, when conceptually modulated processing starts, implicit beliefs about the content of perception are formed, viewers have CAC of the contents. CAC precedes CAA, which becomes available later when objects are identified and categorized. Recall that CMVC is the visual consciousness of conceptual contents formed in late vision. Contents in late vision of which viewers have CMVC are also contents that are parts of implicit perceptual beliefs; they are also CAC contents. Here is why. The contents of which viewers have CMVC are both conceptual and perceptual contents. Being perceptual contents, they present the world as being such and such and, thus, viewers have CMVC of them. Being conceptual contents, they are parts of perceptual beliefs and are thus CAC contents. It follows that contents of which viewers have CMVC are also CAC contents. The reverse relation, however, does not hold. States that are CAC are not necessarily CMVC because viewers have CAC of semantic properties

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  of object that (as I will argue) are not contents of which viewers could have any form of visual consciousness. Now, since late vision starts at about  ms and CAA at about  ms, the contents of the states processed between  ms and  ms are not available yet to CAA since they are the contents of implicit beliefs. This means that the fact that some contents are CMVC does not entail they are also CAA contents. In other words, since CAC differs from CAA, viewers can have CMVC of content that are CAC; but since these contents are not as yet (or may never become) available to CAA, the viewers do not have CAA of them. Thus, viewers can have CMVC of X and form implicit beliefs concerning X without realizing or noticing that they have these beliefs when X is an object or feature that is not available to CAA; when it does become available, viewers have both CMVC and CAA of this content; that is, they form explicit beliefs about X. Allow me to elaborate the discussion in Section .. When late vision starts, the brain formulates tentative hypotheses concerning the identity of the attended objects in the visual scene. Sometimes the first hypothesis formed passes the test, and the object is recognized in about  ms. At other times multiple hypotheses must be tested until one is selected, in which case object recognition takes longer (recall that P is registered – ms after stimulus onset). The tentative hypotheses formed are implicit beliefs, whereas the hypothesis that is selected can become an explicit belief about the object but may remain as an implicit belief. In other words, the tentative hypotheses that are eventually rejected are CAC states that never reach the level of CAA. The selected hypothesis, on the other hand, is initially CAC but may become CAA. When the object is recognized, viewers realize that they see it, i.e. they become aware of it in the strong sense, or (to use Block’s term) they become conscious aware of it. It is then that they can report it. Note that for this to happen, viewers need not realize or notice that they have the belief corresponding to the selected hypothesis; the belief may be implicit, i.e. CAC. Thus, Block is right that to become conscious aware or aware in the strong sense of an object, the confluence of PC and CAC is needed. However, not all beliefs that are formed in late vision and which are CAC make us realize that we see an object; only the belief that passes the tests can make us aware in the strong sense of the object. This is why we do not experience, in the sense of realizing that we see, all the possible alternative objects that match the information in the image, but only the object that corresponds to the hypothesis that passes the test. (Occasionally we may realize that we see an object as X or be aware of X and then change our mind and realize that the object is actually Y. In this case, we initially choose one hypothesis that, erroneously, passed the tests, and then for some reason or other reject it and select another hypothesis concerning the identity of the object.) One might object at this juncture that the various alternative hypotheses formed in late vision are not all CAC states; only the belief that wins is CAC. To answer, let us apply the criteria for CAC. According to Block, a state is CAC if it is actively poised for control of action or reasoning, which, as we saw, ensures that CAC states

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

are occurrent beliefs. The various alternative beliefs concerning object identity are occurrent states. However, although they pass the ‘occurrence’ test, are they actively poised for control? Recall that Block insists that the designation ‘actively poised’ does not require actual use, only the potentiality for such use. The alternative hypotheses certainly have this potentiality because, were they selected, they would have influenced action or reasoning. There is another criterion for CAC, though, which is that CAC is phenomenality plus reflection. The alternative hypotheses are the result of the brain ‘reflecting’ on the information in the image to form the hypothesis; they are beliefs concerning the identity of the object in the visual scene. Is phenomenality involved, however? One might argue that since only the selected hypothesis leads to conscious awareness of an object, only this belief has phenomenality associated with it, and thus it is the only one that should be deemed CAC. This objection overlooks the fact that the tentative hypotheses are formed on the basis of information in the image retrieved from the scene in early vision. Since this is iconic information of which we have PC, the tentative hypotheses are associated with PC. This last remark neutralizes the objection that subjects do not notice or realize that they see the objects corresponding to the tentative hypotheses, and thus that there is no phenomenality involved, since phenomenal consciousness is not awareness in the strong sense. In late vision, one has visual consciousness of the visible features of objects as they are modulated by attention. One is not visually conscious, however, of the D representation of an object, or of its semantic non-observable properties. This follows from the fact that the visual consciousness of both early-vision and late-vision content is subserved by the recurrent loops within the visual areas only, and these areas encode only visible information. Specifically, we know that the visual brain can represent textures, colors, shapes, motions, orientations, and the rest of the visible features of objects. We also have some solid evidence concerning the areas at which these features are represented and the relevant processes. None of the visual areas (including IT), however, is known to represent category membership, and this means that there is no form of visual consciousness associated with category membership per se. (However, once an object is categorized some of its visible features may be highlighted and attended, and this changes the way these features look. Since attention changes the phenomenology of a visual scene, and since the attentional modulation of visual processing goes in hand with the conceptualization of the perceptual contents, this may be a sense in which higher-level concepts make their presence phenomenologically immediate.) The CMVC one has during the stages of late vision is subserved by recurrent processing in the visual brain as it is modulated by attention. Although viewers are not visually conscious of the D representation of an object, they can form thoughts about it. They can have CAC or CAA of the D representation of an object. Hence, given that CMVC is necessarily accompanied by CAC, CMVC of some content entails the CAC of this content, but that some content is CAC does not entail that this content is also available to CMVC.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  Viewers, therefore, can have CAC of the color of an object of which they have CMVC—i.e. they can form the implicit thought that they see an object with that color. Viewers can also have CAA of that color, (i.e. an explicit thought) and report it. They cannot, however, have CMVC of the D representation even though they can have CAC or CAA of it. Let me explain this a bit further. The conceptual contents of late vision are about the conceptualized visible properties of objects as they have been modulated by attention, and about the semantic properties of objects as they have been retrieved from memory. A viewer has CMVC only of the visible properties of the objects and not of their semantic properties because only the former have a phenomenological presence. A viewer also has CAC of both visible and semantic properties, but she has CAA of only some of them. The reason is that during the testing of the various tentative hypotheses concerning the identity of an object in the visual scene, some of the iconic information from the visual scene stored in visual circuits is revisited so that the hypotheses be tested. This information is accessed by cognition, and is therefore cognitive access conscious. Of all this visible information, only the information related to the hypothesis concerning the identity of the object that is selected is consolidated in working memory and thus becomes available to CAA. Similarly, even though various pieces of semantic information may have been retrieved from memory and used to facilitate object recognition and identification, the viewer will have CAA only of the semantic properties that belong to the identified object. Let us return to the relations between CMVC and CAA when both forms of awareness are available (after  ms of visual processing). When one views a scene, after  ms they have CMVC and CAA of contents that result from the synergy of bottom-up information retrieved from the scene and top-down information from cognitive centers. That is, they are aware in the strong sense both of the perceptual content and of the corresponding perceptual belief. The CMVC of the scene concerns conceptual content of which one is visually conscious. The CAA concerns conceptual content of which the viewer is cognitively aware (of the color of an object or of its class membership, for example). As already stated, a content that is CMVC may become CAA content, but not all CAA contents are CMVC, because the contents of the states formed in late vision of which the viewer has CMVC consist in information represented in the visual areas of the brain as they are modulated by top-down signals, whereas CAA contents may include semantic information as well. These considerations have some important implications concerning the relation of CAA, CAC, CMVC, and PC with perceptual contents. First, PC starts at about  ms after stimulus onset; one has PC of the viewer-centered three-dimensional surfaces in depth, for example, of all the objects in the visual scene in – ms. CMVC starts about  ms post-stimulus; one starts having CMVC of the visible properties of attended objects at about  ms. CAC also starts at  ms; one has CAC of these visible properties at the same time, as well as of some of the semantic properties of the object in the scene. Finally, CAA starts at about  ms; one has CAA of some of the visible properties—namely those that are associated with the hypothesis

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

about object identity that is selected among the various hypotheses about identity and is consolidated in working memory—but also of D representation, category membership, and other semantic properties, at  ms. Second, one can have CMVC of the observable properties of a visual scene but one cannot have CMVC of semantic properties of the objects in the scene. Moreover, viewers can have CAC and CAA of both visible and semantic properties. The conceptual contents of thoughts in late vision include the information that the object in the scene is, say, red and oval, but they also include the information that it is a teapot, and that it is heavy; viewers have CAA of a heavy teapot, but they do not have CMVC of them. Third, although CAA emerges during late vision, this does not mean that one has CAA of the contents of late vision in their entirety, as there is a significant time interval during which late vision processes occur but there is no CAA. Only contents that are needed for the purposes of cognitive or motor responses are eventually selected and become CAA contents. There are thus conceptual perceptual contents of which viewers have CAC but not CAA.

 Concluding Remarks Philosophers who think that PC is a generic term designating the visual consciousness associated with a visual scene (PC∗ ) are divided into two categories with respect to the problem of the effects of attention and cognitive penetration on perceptual processing: those who hold that PC∗ does not change when attention intervenes and visual processing becomes cognitively penetrated, and those who think that, even though cognitive penetration affects phenomenology, the fact that there is still phenomenology associated with the experience warrants the usage of the term PC∗ . The first category faces the problem that empirical evidence shows that attention/cognitive penetration changes the phenomenology of a visual scene. Thus, pre-attentional phenomenology should be distinguished from post-attentional phenomenology. The second category faces the problem that it uses a single term, namely PC∗ , to designate two different forms of phenomenology, and this may create confusion. Other philosophers who associate PC with nonconceptual content face a problem with the phenomenology of the conceptual contents of perception, i.e. the contents of late vision, on the assumption that contents that are conceptually modulated are perceptual contents. If PC concerns nonconceptual content, apparently a new term is needed to designate the visual awareness of conceptually affected perceptual content. I introduced the term CMVC to denote this sort of awareness. A viewer has CMVC of the conceptual experiential content related to visual features and of the general organization of a visual scene, as these have been modulated by cognitive influences. CAC covers contents of which viewers have CMVC; a viewer can implicitly think that ‘[that] in front of me is red and circular’. It also signifies the availability to cognition of semantic contents that viewers employ in their implicit beliefs, which being semantic and not visual contents are not accompanied by CMVC. CAA, finally,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  denotes viewers’ explicit beliefs about visual objects and their properties. CAA entails that subjects notice the visible and semantic properties of visual objects. CAA also covers cases of seeing in the doxastic sense, that is, the formation of explicit beliefs about the D model of an object, class membership, and other semantic properties. Thus, one has CAA of fact that ‘the can in front of me is red, circular, and heavy’.

References Barr, M. (). The proactive brain: memory for predictions. Philosophical Transactions of the Royal Society, Biology : –. Block, N. (a). Paradox and cross-purposes in recent work on consciousness. In Collected Papers, vol. , –. Cambridge, Mass.: MIT Press. Block. N. (b). Two neural correlates of consciousness. In Collected Papers, vol. , –. Cambridge, Mass.: MIT Press. Block, N. (c). Consciousness, accessibility, and the mesh between psychology and neuroscience. Brain and Behavioral Sciences : –. Block, N. (d). How not to find the neural correlate of consciousness. In Collected Papers, vol. , –. Cambridge, Mass.: MIT Press. Block, N. (e). On a confusion about a function of consciousness. In Collected Papers, vol. , –. Cambridge, Mass.: MIT Press. Block, N. (). Perceptual consciousness overflows cognitive access. Trends in Cognitive Sciences (): –. Brogaard, B., and Gatzia, D. (forthcoming). Color and cognitive penetrability. Topics in Cognitive Science, special issue on cortical color. Burge, T. (). Reflections on two kinds of consciousness. In Foundations of Mind, vol. , –. Oxford: Clarendon Press. Burge, T. (). Origins of Objectivity. Oxford: Clarendon Press. Carrasco, M., Ling, S., and Read, S. (). Attention alters appearance. Nature Neuroscience : –. Conway, B. R., and Tsao, D. Y. (). Color architecture in alert macaque cortex revealed by fMRI. Cerebral Cortex (): –. Davies, M. (). Tacit knowledge and subdoxastic states. In C. MacDonald and G. MacDonald (eds), Philosophy of Psychology: Debates on Psychological Explanation, –. Oxford: Blackwell. Dretske, F. (). Conscious experience. Mind (): –. Repr. in A. Noë and E. T. Thompson (eds), Vision and Mind (Cambridge, Mass.: MIT Press, ). Dretske, F. (). Naturalizing the Mind. Cambridge, Mass.: MIT University Press. Dretske, F. (). Perception without awareness. In T. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Driver, J., David, G., Russell, C., Turatto, M., and Freeman, E. (). Segmentation, attention and phenomenal visual objects. Cognition : –. Epstein, W., and Broota, K. D. (). Automatic and attentional components in perception of size-at-a-distance. Perception and Psychophysics : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



a. raftopoulos

Evans, M. A., Shedden, J. M., Hevenor, S. J., and Hahn, M. C. (). The effect of variability of unattended information on global and local processing: evidence from lateralization at early stages of processing. Neurophysiologia : –. Granrud, C. E. (). Judging the size of a distant object: strategy use by children and adults. In G. Hatfield and S. Allred (eds), Visual Experience: Sensation, Cognition, and Constancy, –. Oxford: Oxford University Press. Hatfield, G. (). Perception and Cognition: Essays in the Philosophy of Psychology. Oxford: Clarendon Press. Hatfield, G. (). Phenomenal and cognitive factors in spatial representation. In G. Hatfield and S. Allred (eds), Visual Experience: Sensation, Cognition, and Constancy, –. Oxford: Oxford University Press. Heywood, C. A., and Kentridge, R. W. (). Achromatopsia, colour vision, and cortex. Neurological Clinics of North America : –. Hopfinger, J. B., Luck, S. J., and Hillyard, S. A. (). Selective attention. In M. S. Gazzaniga (ed.), Cognitive Neuroscience, rd edn, –. Cambridge, Mass.: MIT Press. Horgan, T., and Tienson, J. (). The intentionality of phenomenology and the phenomenology of intentionality. In D. Chalmers (ed.), Philosophy of Mind, –. New York: Oxford University Press. Jackendoff, R. (). Consciousness and the Computational Mind. Cambridge, Mass.: MIT University Press. Komatsu, H., Ideura, Y., Kaji, S., and Yamane, S. (). Color selectivity of neurons in the inferior temporal cortex of the awake macaque monkey. Journal of Neuroscience : –. Lamme, V. A. F. (). Why visual attention and awareness are different. Trends in Cognitive Sciences (): –. Lamme, V. A. F. (). Independent neural definitions of visual awareness and attention. In A. Raftopoulos (ed.), The Cognitive Penetrability of Perception: An Interdisciplinary Approach, –. Hauppauge, NY: NovaScience. Lamme, V. A. F., and Roelfsema, P. R. (). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neuroscience : –. Muller, M. M., Andersen, S. K., Trujillo, N. J., Valdes-Sosa, P., Malinowski, P., and Hillyard, S. A. (). Feature-selective attention enhances color signals in early visual areas of the human brain. Proceedings of the National Academy of Science USA : –. Niedeggen, M., Wichmann, P., and Stoerig, P. (). Change blindness and time to consciousness. European Journal of Neuroscience : –. Nobre, A. C., Rohenkhol, G., and Stokes, M. G. (). Nervous anticipation: top-down biasing across space and time. In M. Posner (ed.), Cognitive Neuroscience of Attention, nd edn, –. New York: Guilford Press. Perry, J. (). Knowledge, Possibility, and Consciousness. Cambridge, Mass.: MIT Press. Peyrin, C., Michel, C. M., Schwartz, S., Thut, G., Seghier, M., Landis, T., Marendaz, C., and Vuilleumier, P. (). The neural processes and timing of top-down processes during coarseto-fine categorization of visual scenes: a combined fMRI and ERP study. Journal of Cognitive Neuroscience (): –. Raftopoulos, A. (). Defending realism on the proper ground. Philosophical Psychology (): –. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and the Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

penetrability and consciousness  Raftopoulos, A. (). Late vision: its processes and epistemic status. Frontiers in Psychology (). Doi: ./fpsyg... Raftopoulos, A. (). The cognitive impenetrability of the content of early vision is a necessary and sufficient condition for purely nonconceptual content. Philosophical Psychology. Doi: ./.., –. Raftopoulos, A. (forthcoming). The cognitive impenetrability of perception and theoryladenness. Journal for General Philosophy of Science. Raftopoulos, A., and Muller, V. (). Nonconceptual demonstrative reference. Philosophy and Phenomenological Research (): –. Rensink, R. A. (). The dynamic representation of scenes. Visual Cognition : –. Sergent, C., Baillet, S., and Dehaene, S. (). Timing of the brain events underlying access to consciousness during the attentional blink. Nature: Neuroscience, online version, –. Smith, A. D. (). The Problem of Perception. Cambridge, Mass.: Harvard University Press. Stalnaker, R. C. (). Our Knowledge of the Internal World. Oxford: Clarendon Press. Stazicker, J. (). Attention, visual consciousness and indeterminacy. Mind and Language (): –. Tye, M. (). Visual qualia and visual content revisited. In D. Chalmers (ed.), Philosophy of Mind, –. Oxford: Oxford University Press. Tye, M. (). Nonconceptual content, richness, and fineness of grain. In T. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press. Tye, M. (). Consciousness Revisited: Materialism without Phenomenal Concepts. Cambridge, Mass.: MIT Press. Wolfe, J. M., Oliva, C., Horowitz, T. S., Butcher, S. J., and Bompas, A. (). Segmentation of objects from background in visual search tasks. Vision Research : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Seeing, Visualizing, and Believing Pictures and Cognitive Penetration John Zeimbekis

Visualizing and mental imagery are thought to be cognitive states by all sides of the imagery debate (Tye, Pylyshyn, and Kosslyn). Yet the phenomenology of those states has distinctly visual ingredients. This has potential consequences for the hypothesis that vision is cognitively impenetrable, the ability of visual processes to ground perceptual warrant and justification, and the distinction between cognitive and perceptual phenomenology. I explore those consequences by describing two forms of visual ambiguity that involve visualizing: the ability to visually experience a picture surface as flat after it has caused volumetric nonconceptual contents (Sections  and ), and the ability to use a surface initially perceived as flat to visualize three-dimensional scenes (Section ). In both cases, the visual processes which extract viewer-centered volumetric shapes (equivalent to Marr’s /D sketch) have to rely solely on monocular depth cues in the absence of parallax and stereopsis. Those processes can be cognitively penetrated by acts of visualizing, including ones that draw on conceptual information about kinds. However, the penetrability of the visual processes does not weaken their ability to provide perceptual warrant and justification for beliefs (Section ). The reason is that picture perceptions—whether they are stimulus-driven or based on acts of visualizing—are different to object perceptions both phenomenologically and in terms of their functional roles as states. Thus, although the penetrability of the visual processes does mean that subjects can have visual experiences with contradictory (D and /D) contents, perceptual belief is adopted at most towards one set of contents, and questions of warrant and justification are raised only for those contents. A ruleproving exception is provided by trompe-l’oeils (Section ). I use the expressions ‘volumetric content’, ‘D contents’, and so forth to designate the representational content of visual experiences, whether the experiences are caused by object perceptions or by picture perceptions, and whether they are stimulusdriven or driven jointly by acts of visualizing. I take such visual experiences to be representational, contentful states which divulge accuracy or correctness conditions. The accuracy conditions can be thought of in terms of Peacocke’s () scenario

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  content. For example, to have a visual experience with volumetric contents is to represent volumetric shapes at certain egocentric locations. What it is to have visual experiences with volumetric representational contents when perceiving flat surfaces, and what kind of epistemic predicament this is, will be explained at length. On the other hand, why visual experiences have content and accuracy conditions in the first place, and what it is to have content and nonconceptual content in particular, are not topics dealt with here. For concise replies to the last two questions, see Macpherson, Chapter , section .

 Pictures and Visual Ambiguity Among theories of depiction, there is considerable consensus that pictures can cause not only visual experiences with volumetric contents but also veridical visual experiences of picture surfaces as flat. The assumption is so widespread that it splits into several varieties, depending on whether viewers can have both visual experiences at once. One variety holds that it is possible to have both the experience of volumetric shapes and the experience of flatness at once (Peacocke : , ; Schier : ; Lopes : ; Walton : –). Another says that we always have both experiences at once (Wollheim : ; Gibson : ; Rock ). A third holds that we cannot have both experiences at once, but can have each one separately (Gombrich : , and perhaps Cutting and Massironi ). But there is one point on which all of these authors agree: that we are capable of having both kinds of visual experience, both of flatness and of depth, of picture surfaces. If this consensus on the phenomenology of picture perception is anything to go by, pictures qualify as visually ambiguous figures. The consensus is worth noting because, as Macpherson has said in a related context,1 ‘the evidence that one must appeal to here is introspective and the more people that agree that a change takes place in their experience the better’ (Macpherson : ). Yet, if we admit the phenomenological claims as a form of evidence, then pictures should support not one but several kinds of visual ambiguity. In one kind, visually experiencing a picture (or even parts of it at a time) as a flat surface (one perpendicular to the line of vision) requires great mental effort. Figure . provides an illustration. If we can visually experience such pictures or parts of them at a time as flat, it will only be subsequent to the pictures’ having caused visual experiences with volumetric content; the mental effort could involve consciously shifting attention across the surface until we find a way to blind ourselves to regions that work as depth cues and prevent the cues from triggering depth representations. Even Peacocke’s relatively modest claim,

1 The context is Macpherson’s discussion of the ability to perform Gestalt switches with ambiguous figures; but in the cases Macpherson has in mind the switch is not from D (/D, i.e. egocentrically D) to D visual contents, as will be the case with the examples introduced here.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

Figure . Visually driven picture perception

that we experience the outlines of object-representations in pictures as flat,2 seems wrong where this picture is concerned: the left outline edges of the pyramids seem further away than the right edges. The problem is that vision refuses to process just the outlines; it processes the interior lines as edges and yields a mental representation of a regular volume with a certain orientation, which settles the egocentric distances of the object’s parts, including parts of the outlines. Whether or not Figure . is in fact visually ambiguous (as one would expect given the claims that pictures can also cause visual experiences of their surfaces as flat), it is clear that when we perceive it we first become aware of volumetric visual contents and then have trouble suppressing this visual experience by making an agentive mental effort. How do the volumetric visual contents of such ‘natural’ picture perceptions emerge in the first place? We could sketch the following hypothesis, postponing until the next section its details and justification. The picture’s surface includes monocular depth cues which can trigger brain representations of volume even without calculations from parallax and stereopsis; those cues exploit early visual processes dedicated to constructing volumetric contents; the outputs of those early processes are brain representations corresponding to Marr’s /D sketch; those representations provide the contents of phenomenal awareness; and those phenomenal contents constitute the visual experience of the picture’s content. If the hypothesis is plausible, as I will try to show in Section , then, in this form of visual ambiguity, volumetric contents are experienced without any help from visualizing or visually imagining, which can be driven consciously and are usually thought to be higher-level processes. (See Tye : , ; Pylyshyn : ; Pylyshyn ; Kosslyn .) On the other hand, 2 ‘The silhouette both is and is experienced as a flat surface, or at least as occupying a plane: any description of the experience omitting this point is incomplete’ (Peacocke : ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  D contents emerge—if they emerge at all, or when they do3 —only subsequently, as a result of some conscious activity. Today we frequently have such picture perceptions, which feel effortless and automatic, because many of the pictures we perceive (including non-agentive pictures like photographs) contain cues which successfully exploit object-perception processes like segmentation and the construction of egocentric depth representations out of twodimensional distributions of light. But while some depth cues (line junctions) were mastered in prehistoric times (Cavanagh ; Biederman and Kim ), it took a long time for drawing and painting to develop techniques which could cause visual and perceptual contents comparable to those caused by objects about shading, textures, colours, lines, orientations, volumes, and the depth relations between objects (see e.g. Kubovy  on depth relations; Gombrich : ch. , on colour). Moreover, once mastered, the techniques were often subsequently avoided by painters trying to develop new pictorial styles—i.e. new ways of causing a visual impression of depth and volume without relying on existing techniques. When a picture does not include adequate cues to evoke some of those features (shading, textures, lines, orientations, etc.), the brain cannot form representations of the intended picture contents solely on the strength of stimulus-driven or hard-wired processes. On the strength of those processes, the brain may represent the picture as a flat surface with uneven stains, or else it may capture only some local depth features (like apparent occlusions) but not global ones or depth relations between objects. In those cases, we have to perform conscious acts of visualizing by using the picture surface as a prop. This gives rise to a second kind of visual ambiguity supported by pictures. This time, the surface naturally yields a visual experience of a flat surface; but subsequently, by performing conscious acts of visualizing, we can look at the surface and generate experiences of depth and volume. Such cases of depiction are often provided by what Cutting and Massironi () call fortuitous pictures: natural or other objects whose colour and line patterns can, but do not always, support visualizing activities which allow ‘seeing-in’. The passage below, taken from Leonardo’s advice to apprentice painters, seems to describe this kind of ambiguity: If you look at any walls soiled with a variety of stains, or stones with variegated patterns, when you have to invent some location, you will therein be able to see a resemblance to various landscapes graced with mountains, rivers, rocks, trees, plains, great valleys and hills in any combinations. Or again you will be able to see various battles and figures darting about, strangelooking faces and costumes, and an endless number of things which you can distill into finelyrendered forms. And what happens with regard to such walls and variegated stones is just as with the sound of bells, in whose peal you will find any name or word you care to imagine. (Leonardo : )

For instance, it is possible to visually experience Figure . as the stained flat surface that it is. But it is also possible to visually experience the figure depthwise as figure 3

I think they do emerge in some cases; see Section .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

and ground; and in fact concepts can help us to settle the figure–ground relation, for instance if we’re told to see the lower part as a field of wheat and the upper part as a grey sky. A question that emerges about such cognitively driven acts of visualizing is whether they just add a cognitive ingredient to mental contents without altering the outputs of the visual processes leading up to Marr’s /D sketch, or instead tamper with the way vision itself generates figure–ground segregations; I postpone this discussion until Section . A third form of picture-related visual ambiguity occurs when neither D nor /D visual contents are the more natural output of vision, so that neither prevails in perception. Details of pictures taken out of context can be ambiguous in this way because their depth cues can easily be overridden in isolation. Many Rorschach blots are also in this category: they cause representations of volume only indirectly (through recognition, by exploiting templates for outlines); they do not include shading and enclosed lines that could signal convexities, concavities, or what Marr calls surface orientation discontinuities (Marr : –). The weakness of Rorschach blots as pictures is described by Gibson when he writes that their ‘invariants are all mixed up together and are mutually discrepant instead of being mutually consistent or redundant’ (Gibson : ). A fourth kind of ambiguity is bi-stability, in which both visual interpretations are volumetric and both are generated bottom-up, as apparently occurs when we switch among different perceptions of the Necker cube. It seems plausible to hold, as Pylyshyn does (: ch. ), that both visual experiences of the Necker cube are performed purely by bottom-up processes jointly with hard-wired processes. Finally, a fifth kind of ambiguity is that found in perceptions of Jastrow’s duck/rabbit drawing. Both experiences caused by the drawing are experiences of the picture’s content, namely, an object belonging to a three-dimensional kind. However, if those experiences involve any representation of volume, that representation depends almost entirely on prior knowledge encoded in sortal concepts, not stimulus-driven depth cues. In fact, it is not immediately clear that there is any change of visual contents in the Jastrow’s ambiguity, although this has been claimed recently (see Macpherson : ; Siegel  makes a related point, discussed here in Section ). In the Jastrow figure, representations of volume and depth are not caused by bottom-up or hardwired visual processes, as they are by the Necker cube, because the Jastrow contains no internal lines suggestive of surface discontinuities—only a couple of weak volumetric

Figure . J. M. W. Turner, Approach to Venice, oil on canvas, ; detail. Courtesy National Gallery of Art, Washington, DC

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  cues from shading (the shading that suggests a concavity between the head and neck). So any change of visual experience as we switch from one experience to the other is unlikely to be a change in content construed as what would have to be the case in the world for the experience to be true. The change seems to be more superficial; it has been described by Lyons as ‘manipulating the representation produced’ previously by the bottom-up processes without altering it, or as ‘facilitating pop-out of certain patterns’ and yielding ‘a late experiential effect, leaving the nonconceptual early perceptual states unaffected but influencing the nondoxastic seemings’ (Lyons ). While cases belonging to all five kinds of ambiguity have been described and discussed in the literature on depiction, the kinds of visual ambiguity supported by pictures have not yet been distinguished, nor the differences between them explained. Making the distinctions is important for two sorts of reasons. First, ambiguous figures have always been used to test claims about the cognitive impenetrability of certain visual processes and certain ways of drawing the perception/cognition distinction (Churchland ; Macpherson ). Deniers of cognitive penetrability (Fodor ; Pylyshyn ; Raftopoulos ; ) have in turn responded to each kind of ambiguity-based penetrability claim. Some kinds of ambiguity are easy enough for modularists to deal with, others less so. But pictures are an extremely rich resource of different kinds of visual ambiguity; the first two kinds just isolated have not been described in the penetrability literature but pose a significant threat to the impenetrability hypothesis. Secondly, understanding the different forms of visual ambiguity is essential to understanding what pictures are. For example, they show that picture perception is not a single kind of mental state, be it a higher-level state of visualizing or a lower-level state which is the output of earlier visual processes;4 they allow us to describe the mental states and contents caused by different kinds of pictures, and thus to relate those states and contents to other states such as perceptual belief and memory; they allow us to account for what is called naturalism in depiction; and so on. Thus, issues about the perception/cognition distinction turn out to be essential for an account of depiction; though in this chapter I will focus on the first set of issues. The first two kinds of visual ambiguity described correspond to two kinds of picture perception. The first, illustrated by Figure ., could be called bottom-up or ‘natural picture perception’. The ambiguity it gives rise to is one in which an object naturally causes /D visual contents, but with some mental effort we can also see the object as flat. The second, illustrated by Figure ., could be called ‘cognitively driven picture perception’; the ambiguity it gives rise to is one in which an object naturally causes D visual contents, but with some mental effort we can see the object as /D. How exactly do these two kinds of visual ambiguity relate to the cognition/ perception distinction and to cognitive penetrability? If all picture perceptions were bottom-up and hard-wired, or if all picture perceptions were cognitively driven acts 4 e.g. Abell and Currie (: ), Wollheim (; ) and Levinson (: ) seem committed to an account of picture perception as the output of pre-doxastic processes; Walton () to a doxastic-level account (though see Walton : , for a denial of this). Most authors on depiction do not address the issue at all; an exception is Levinson ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

of imagining, or if some were bottom-up and others cognitively driven, this would not necessarily entail any form of cognitive penetrability. The bottom-up picture perceptions could be compatible with impenetrability of the visual processes up to Marr’s /D sketch, while the cognitively driven ones could conceivably be such that they added a cognitive phenomenology to mental contents without altering the outputs of the visual processes leading up to Marr’s /D sketch. It is along similar lines that Raftopoulos (; ), Fodor (), and Pylyshyn () have dealt with the kinds of ambiguity supported by the Jastrow figure and the Necker cube. But there is a difference between those forms of visual ambiguity and the forms described above which involve switching from D to /D visual contents and vice versa: only the latter two have the potential to cause trouble for the hypothesis that visual processes required to construct /D nonconceptual content out of the D sketch are cognitively impenetrable. My strategy here is twofold. I will concede that there are good counterexamples to the claim that visual processes leading up to Marr’s /D sketch are cognitively impenetrable. Sections  and  describe two forms of visual ambiguity, and argue that they imply the cognitively penetrability of visual processes required to construct /D content out of the D sketch. I believe that trying to exclude all such cases of penetrability would be barking up the wrong tree. Instead—and this is the second part of the strategy—I will claim that even if it turns out that the functioning of some key visual processes can sometimes be tampered with by processes that count as cognitive, this does not lead to the expected pernicious epistemic and epistemological consequences (Sections  and ). The reason is that visual experiences caused by pictures fail to cause perceptual beliefs. This also amounts to something of an enabling condition for depiction as a form of representation, as opposed to a source of illusions.

 Bottom-Up or Natural Picture Perceptions The hypothesis that some pictures cause visual experiences with /D contents without any conscious mental contribution on behalf of the viewer is plausible, because otherwise picture perception would always be an effortful, slow performance for subjects, involving acts of visualizing for each feature of the picture that can be used to imagine a depth relation or a volumetric shape. Instead, most picture perception is caused in an effortless, quick, and automatic way, suggesting that it is subserved by hard-wired and bottom-up processes, not deliberate acts of imagining. So the hypothesis is compatible with the phenomenology of many picture perceptions; and it also allows us to explain the other cases: in the absence of cues that trigger automatic, dedicated processes, we have to perform slow, attention-consuming acts of visualizing to grasp the picture contents. A sketch follows of how bottom-up

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  picture perceptions could occur. It focuses mainly on the construction of volumetric shapes from monocular cues, and for reasons of space omits the segregation of objects during picture perception and the ways in which different object-recognition processes are coopted by different kinds of pictures (issues which are discussed in Zeimbekis ). The natural place to look for support for the claim that many pictures cause egocentric, conscious representations of D shapes is Marr’s theory of vision. Marr’s working hypothesis is that initial, two-dimensional retinal inputs are built up into adequate representations of the three-dimensional objects that cause them, and this makes its details particularly interesting for a theory of depiction. An important part of Marr’s proposal cannot be applied to pictures. It is the part of his theory (Marr : ch. ) according to which binocular disparity—the slight difference in perspective from which each eye views the object—is exploited by the visual system to compute the relative depths of surfaces in a scene. In pictures, all points of the surface from which the eyes receive stimuli are at roughly the same location on the back-to-front axis (the axis perpendicular to the picture’s surface and passing through the centre of the viewer’s body, which is identical to the z axis of Peacocke’s nonconceptual scenario content; see Peacocke : ). Therefore, in picture perception, the brain cannot compute depth relations from binocular disparity and has to rely on monocular depth cues. The same applies to parallax from movement relative to the object. When we move relative to the picture, points on the surface are not seen as moving at different speeds and parallax cannot be exploited to compute depth. Picture perception is even insensitive to changes in the angle from which the picture is viewed; the picture can be rotated up to  degrees on its vertical axis without this deforming the threedimensional representations it causes (Cutting ), an effect Kubovy () calls ‘robustness of perspective’. However, Marr also relies heavily on hypotheses about how we extract volumetric and depth information from monocular cues which exploit neither binocular disparity nor parallax. Here, his proposal is that parts of the visual system take two-dimensional representations of objects or scenes as inputs, apply hard-wired processes to them, and yield three-dimensional shape representations. The inputs of the processes are edges (lines) and their orientations and junctions, i.e. the outputs of earlier visual processes which generate a brain representation of differences in light intensities. These are treated by vision as discontinuities in the distance of surfaces from the viewer (‘occluding contours’), convexities or concavities of surfaces (‘surface orientation discontinuities’), and depthwise curvature (‘surface contours’) (: –). Texture patterns are also interpreted to yield representations of the depthwise slant of a surface (pp. –). According to Marr, the brain applies a set of simplifying constraints when interpreting occluding contours; for example, points close on the D contour are assumed to be close in D space (p. ), an assumption which is nicely illustrated by Richard Gregory’s impossible triangle. When those assumptions are applied to

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

two-dimensional views of certain regular D shapes,5 they yield accurate threedimensional representations of those shapes. (Note that monocular cues are likely to be an essential resource for vision when objects are too distant for binocularity or parallax to calculate the relative distances of parts of an object.) On Marr’s model, the output of these processes—the /D sketch, an egocentric volumetric representation—subserves kind recognition and therefore precedes the application of concepts. Volume is assigned as a condition for recognition and classification. The opposite—classification as a condition for assigning shape—would amount to a form of sortalism like the one criticized by Campbell (see his account of the ‘delineation thesis’; Campbell : ). On the sortalist scenario, we would carve up the visual scene into volumetric objects by using sortal concepts, not on the basis of bottom-up stimuli jointly with hard-wired visual processes like those that Marr describes; those processes would underdetermine the assignment of shape, allowing different shapes to be assigned depending on which concepts we applied. A sceptic might argue that in picture perception, due to the absence of any differential stimuli from stereopsis and without performing the processes that compute them, the brain would not be in the kinds of states that qualify as causal antecedents for having three-dimensional object representations. While we do have D object representations at some point in picture perception, the sceptic might argue that they are not caused by lower-level visual processes but are instead the effects of some kind of higher-order imagining or visualizing. My response to this scepticism is twofold. First, I do not deny that there are cognitively driven picture perceptions which involve conscious acts of imagining. Such picture perceptions are different to the ones I call ‘bottom-up’ or ‘natural’, and the sceptic would have to explain the differences. Admitting bottom-up picture perceptions can explain the difference between perceiving pictures like Figure . and perceiving pictures in non-naturalistic styles like cubism. For example, Picasso’s The Guitar Player () can support a visual representation of a guitar player—at least if the concept of visual representation is construed to include the contents of visualizing. But to succeed in having the visual representation of a human body in the position of a guitar player, the subject first has to make conscious hypotheses about depth relations and volumetric shape. Before we read the picture’s title and make use of concepts, all we can pick up in terms of volume or depth are, at best, some local occlusion effects which do not allow us to reconstruct the volumes corresponding to the object described by the concepts. Figure ., on the other hand, yields that kind of information naturally and effortlessly as soon as we glance at the picture, suggesting that Marr is right to hold that the brain can reach

5 The regular shapes are ‘generalized cones’, shapes whose cross-sections have the same shape but can vary in size (e.g. a sphere, pyramid, or cylinder). Although the objects we have to recognize in the visual environment are not usually generalized cones, they can be analysed coarsely into regular component shapes for the purposes of kind recognition (Marr ; Biederman ; ; see also the brief discussion which follows).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  volumetric object representations without either the benefit of binocular disparity or input from recognitional concepts. Object-recognition theories after Marr seem to confirm this hypothesis. According to one such theory, Biederman’s theory of recognition by components (‘RBC’; Biederman ; ), the features of objects on the basis of which we analyse them into components, such as vertices, which are interpreted as concavities or convexities, are ‘generally invariant over viewing position’ (: ). As a result, the information required to analyse an object into primitive volumetric components can be extracted from a single two-dimensional representation of the object (: ; : –). A picture of an object causes a two-dimensional view of that object from a single viewpoint, so on this theory, picture perceptions can suffice to cause volumetric object representations. Note that on RBC theory, these structural representations are objectcentred, not viewpoint-relative; they correspond to the objects in Marr’s D sketch of the visual scene and are thought to be the immediate causal antecedents of object recognition on this theory as well as Marr’s. Competing views of object recognition have shown that RBC is inadequate for explaining certain recognition tasks and fails to match a quantity of experimental data. The competing accounts are called ‘image-based’, ‘viewpoint-relative’, or ‘multipleviews’ theories. Image-based or viewpoint-relative models of object recognition emerged from evidence that recognition also relies on viewpoint-relative information in ways that RBC does not account for (Tarr and Pinker ; Bulthoff and Edelman ; Edelman and Bulthoff ; Tarr ; Ullman ). In a series of experiments, subjects who were acquainted with new kinds of objects could still only recognize them from viewpoints similar to those under which they were presented to them, contradicting the predictions of RBC theory. (This behavioural evidence has also been backed by neurological findings; see Logothetis et al. .) Moreover, RBC’s structural representations are too coarse-grained to capture the recognition of individuals, or even to subserve many fine-grained generic classifications. These theories could be seen to present a challenge to the idea that a volumetric (egocentric, /D) representation is constructed prior to recognition, since they could be compatible with the claim that volumetric representation is added cognitively post-recognition, not visually prior to recognition. However, viewpoint-dependent theories do not make the claim that the brain representations that trigger recognition are representations of objects in two dimensions. The relevant distinction is that between structural, object-centred representations, and representations that are relativized to viewpoints. The representations in Marr’s /D sketch are also relativized to viewpoints, yet they are volumetric and represent depth; they contrast with subsequent brain representations which permit mental rotation, belong to the D sketch, and qualify as structural or viewpoint-independent. The conclusion that both viewpoint-relative and viewpoint-independent mechanisms are involved in recognition is supported by neurological findings on an area of the human brain called the lateral occipital

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

complex (LOC). According to Kourtzi et al. (), this area represents perceived object shape, a relatively high-level cognitive representation, and plays a role in object recognition. Kourtzi et al. show that while one subregion of the LOC represents two-dimensional shape, another subregion encodes the represented threedimensional shape of objects and appears to ‘mediate object and scene recognition based on rather abstract three-dimensional representations’ (Kourtzi et al. : ; see  for other studies which show that the LOC represents shape threedimensionally). Volumetric representations on Marr’s hypothesis and on the RBC hypothesis, and viewpoint-relative representations on the hypotheses of Tarr, Bulthoff, and Ulmann, are considered causal antecedents for object recognition and classification, and therefore must be nonconceptual states. Putting aside for a moment the non-naturalistic and fortuitous pictures that require agentively driven attention and conceptual input to contribute to depth interpretation, and focusing on the important category of naturalistic pictures, this yields the following result. Descriptions of the nonconceptual contents of perception by Evans (), Peacocke (), and Raftopoulos () are consistent with the descriptions Marr gives of the /D sketch. Both are perceptiondependent, unlike conceptually encoded representations; both are spatially egocentric; and in both cases, shape, orientation, texture, and colour (including shading) information is more fine-grained than prior conceptually encoded information about such properties. (For examples, see Marr’s illustrations of how we visually experience flat surfaces that contain monocular depth cues; : –.) On several current theories of the neural correlates of consciousness, the kinds of shape representations which according to Marr constitute the /D sketch require local recurrent processing—a kind of brain state that occurs at approximately – ms after stimulus onset and precedes personal-level awareness—which is thought to be the neural correlate of phenomenal awareness in visual perception (Block ; ; ; Lamme ; ; Raftopoulos : –). If that hypothesis is true, then the egocentric volumetric shape representations of Marr’s /D sketch constitute the nonconceptual contents of visual perception. Thus, the sketch of bottom-up picture perceptions that I have drawn here would explain an important part of the phenomenology of picture perception: by the time we reach personal-level awareness and the ability to think about the picture, the picture has already caused a visual experience of a three-dimensional scene. This conclusion seems accurate phenomenologically; for a great number of pictures, picture perception is not the result of any mental effort but occurs naturally. However, the bottom-up account of many picture perceptions also raises new questions: if there are bottomup picture perceptions, then what differentiates those experiences from perceptual illusions, and especially from the visual experiences caused by trompe l’oeils? These questions are dealt with in Section .

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing 

 Picture Perception and Cognitive Penetrability As stated in Section , there is considerable consensus among depiction theorists that picture perceptions can support both /D and D visual contents. An easy way to account for that insistence would be to hold that such claims of visual ambiguity are limited to cognitively driven picture perceptions and do not apply to bottom-up picture perceptions. This response would sidestep the underlying issue of how viewers could avoid having contents caused by bottom-up processes. The problem with the response is that the theories tend to claim that pictures generally—not just pictures that require cognitively driven forms of picture perception—can support such visual ambiguity, and rely on naturalistic pictures such as Constable’s paintings. Consider Figure ., which is similar to one of the drawings Marr (: ) uses to illustrate the interpretation of surface orientations as volumetric shapes by early, hard-wired visual processes. The natural visual interpretation of the figure yields a nonconceptual representation of a cube seen from above, with point A in front of point B. Marr’s explanation of the experience of depth is this: ‘If the occluding contour shown with thick lines is present on its own, one perceives a hexagon. The interior lines change it into a cube, since they suggest that the occluding contour is not planar’ (: ). But although it is initially difficult to perceive this figure and visually represent it as flat, it is possible. It suffices to be told to look for three rhomboids, or for a regular hexagon, or to see the enclosed lines as radii. If you see the hexagon, attention is distracted from point A to the perimeter and you experience a short-lived change of visual experience. Note that Figure . is not bistable: we return naturally to the cube perception once we relax the effort of attention, so there is one visual interpretation we could call natural. Natural picture perceptions share this characteristic

B

A

Figure . Opaque cube

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

with what Kanizsa and Gerbino call ‘amodal completions’, as distinct from what they call ‘represented’ or ‘merely thought’ completions, which work like cognitively driven picture perceptions. (For examples of such revisable amodal completions see Kanizsa and Gerbino : figs .c and .a.) There appear to be two different ways to go about seeing the drawing as flat. The first is to consciously direct attention away from the drawing’s enclosed lines and on to the perimeter, optionally by using the concept rhomboid or hexagon to focus attention differently. That way, we can avoid performing the hard-wired interpretation of the intersection of the three enclosed lines as three-dimensional. Conscious manipulations of attention are a personal-level activity even when they are not concept-driven; when they are concept-driven (e.g. because we are given hexagon or rhomboid), the concept determines where attention is directed. But there is a second way to get the same effect, which this time does not require us to attentionally ignore the enclosed lines in Marr’s cube: we can just see them differently, as radiating from the centre of the hexagon. In that case, we focus on the very part of the picture that contains its depth cue and still get a D representation. If that is so, and if we appeal to the attention-shift argument to explain the change in visual experience and content, we are exposed to the rejoinder that viewers directly interfere with the processes that construct volumetric representations. The rejoinder may or may not be correct, but as it stands it pits one introspective claim against another. If the dispute between deniers and defenders of penetrability hinges on such a conflict, then it seems to reach a stalemate. This is important because the key argument by deniers of penetrability where ambiguous figures are concerned is that shifts in where spatial attention is focused on the scene change which data is processed, without changing the way such data is processed (Fodor ; Pylyshyn ; Raftopoulos ; ). If Marr’s opaque cube is visually ambiguous in the way described, then it is a clear counterexample to the attention-shift argument for a key part of perception: the construction of /D visual contents from lines of the optic array that signal surface orientation discontinuities and convexities. Which other strategies are available to block the conclusion that the switch from /D to D visual contents implies the penetrability of visual processes by agentively driven processes? For one kind of ambiguity seen in Section , that found in Jastrow’s rabbit/duck figure, it can be argued that what changes each time is only the conceptual representation caused by the nonconceptual visual contents, and that the visual experience and contents remain the same throughout. That argument has been used to counter Churchland’s use of perceptual ambiguity to challenge Fodor’s () claims about the impenetrability of visual processing modules. Churchland writes that we can make figures like Jastrow’s ‘flip back and forth at will between the two or more alternatives, by changing one’s assumptions about the nature of the object or about the conditions of viewing’, concluding that ‘some aspects of visual processing, evidently, are quite easily controlled by the higher cognitive centers’ (Churchland : ). Discussion of the Jastrow figure received a new twist when Macpherson

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  () claimed that in Jastrow-type ambiguity, the nonconceptual content of visual experience also changes. (A distinct but substantially similar position is Siegel’s  generalized claim that conceptual content influences perceptual content.) Irrespective of whether those claims are true and what kind of penetrability they amount to, the claim that in Figure . the switch is merely conceptual cannot even get off the ground. It is the nonconceptual representation of shape, in particular the /D sketch itself, that changes. For example, points A and B go from being at different distances on the z axis of Peacocke’s nonconceptual scenario content to being equidistant on that axis. There is no comparable modification of visual nonconceptual content in the Jastrow figure’s ambiguity. A related strategy for countering cognitive penetrability claims is to use low-level perceptual illusions like the Müller–Lyer illusion (Fodor ; Pylyshyn ; ) to argue that the modules that yield visual experiences as outputs cannot be influenced by belief. But the Müller-Lyer illusion, like the chessboard illusion, is not visually revisable. In the chessboard illusion, the visual processes that interpret changes in light intensity in order to preserve light constancy turn out to be impenetrable. We cannot (as it were) ‘reach down’ from personal-level awareness into the brain processes that interpret light intensity data, tamper with them, and produce an experience of a different colour. However, we do seem to be able to do something like this with certain processes that yield volumetric representations, since, for example, when we switch from /D to D contents in Figure ., the orientation of line AB changes. When we do this we bring our perceptions into line with our beliefs—exactly what Fodor and Pylyshyn seek to deny by their use of lower-level illusions. Thus, to the extent that bottom-up picture perceptions are visually revisable at the personal level of awareness—with the exception of trompe l’oeils, which I will argue constitute ruleproving exceptions (Section )—such picture perceptions do not constitute lower-level illusions. Perhaps certain depiction theorists can come to the rescue of cognitive impenetrability. Some accounts of depiction hold that we have experiences of pictures as D and /D simultaneously (Wollheim : ; Gibson : ; Rock ), which contradicts the account given of Figure .. In its extreme form—the claim that we always visually experience pictures as D and /D simultaneously—this proposal (which consists essentially of phenomenological claims) is in fact contradicted by the phenomenology: in bottom-up picture perceptions, which are the most frequent kind, the two experiences are separated by the considerable mental effort required to switch from one of the visual experiences to the other. But in fact, even to say that we sometimes have a visual experience both of the surface and of the picture content seems wrong. In order to see a cube drawing as both flat and cubical at the same time, we would have to represent the intersection of the enclosed lines (point A) as occupying two locations at once. There are several ways of generating visual contents from pictures, but neither the natural, hard-wired ways nor the attention-driven visual interpretations make us represent a point at two different locations simultaneously.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

Although the visual system can yield separate and conflicting depth interpretations of scenes, each interpretation always seems to be consistent (Waltz ; Pylyshyn : –; Cutting and Massironi  also give several illustrations of this visual assumption in their discussion of line interpretation). When vision cannot solve the problem of depth relations in a scene consistently, the result is an incomplete content and visual experience of only part of the picture at a time, as when we see the devil’s pitchfork or certain drawings by Escher. Moreover, it is not part of the phenomenology of the picture perception that we see the point in two places at once; on the contrary, it is very much part of the phenomenology that we see the point at one location or the other, and that the two experiences are separated temporally by an ‘attentional switch’ (which itself seems to be positively experienced). In fact, in the case of the simple outline drawing of a cube, there is little else to the phenomenology of the perception than the locating of these points in space. In that case, what could explain the insistence of Wollheim, in particular, on the twofoldness thesis? Wollheim holds that the twofold experience of pictures is a single experience with two ‘aspects’. Perhaps we could give the following construal of that claim. Matthen (: –) has argued that pictures do not engage the motion-guiding visual system, a visual system thought to be implemented in the dorsal pathway and to be distinct from the system dedicated to recognition and to representing object features (Ungerleider and Mishkin ; Milner and Goodale ). According to Matthen, the impact of this physical difference on the phenomenal character of picture perceptions is that they do not cause a ‘feeling of presence’ (: ). So perhaps the dorsal system dedicated to navigation ‘knows’ that the picture is a more or less flat object, while at the same time the ventral system picks up the volumetric contents and depth relations from the picture’s surface. This would solve the apparently paradoxical nature of phenomenological accounts of twofoldness found not only in Wollheim (: ) but also in Peacocke (: , ) and in Gibson (: ). Wollheim insists that the viewer ‘remains visually aware not only of what is represented but also of the surface qualities of the representation’ (: ; emphasis added), even describing the experience as ‘twofold attention’. Perhaps this can be squared with the kind of twofoldness patched together in the last paragraph. Even so, a problem would remain: there is no room in this account of twofoldness for the fact that we consciously perform mental actions in order to switch from one set of visual contents to the other. If we really are visually aware of two contents at once, as Wollheim holds, they cannot be the two sets of contents that we have trouble switching to and from when we perceive Figure . (or, for that matter, Figure .). Another attempt to explain Wollheim’s convictions about twofoldness could be that they result from conflating two senses of ‘seeing’: a contentful and a purely causal sense. The surface causes the picture content and in this sense we see it, but when it does this, we don’t see the surface where it is—since visually experiencing the content means experiencing different points of the surface at different distances. Once again, this does not account for Wollheim’s particular position, since he insists that we do not see the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  surface in the bare sense that the surface is a causal stimulus, but visually experience it. Yet another explanation could be this, which concerns colour in particular. The colour that we place as a feature at a location, usually behind the picture surface, is the colour caused by the surface. So one may hold that we have an accurate perception of the surface in respect of colour and at the same moment an inaccurate perception in respect of location. The location component of the content is inaccurate because there is no blue there, while the feature component is accurate because there is blue (but not where it is located). In that case, we do not see the colour in two places at once: we see it behind the picture surface. This account would be compatible with typical phenomenological descriptions of perceiving pictures in Wollheim, as when he says that we ‘marvel endlessly at the way in which line or brush stroke or expanse of colour is exploited to render effects or establish analogies that can only be identified representationally’ (Wollheim : ). However, the solution cannot be extended to shapes. Colours (including non-chromatic ones in etchings and drawings) yield shape information, but adopting the solution for shapes would again lead to the problem encountered earlier, of seeing the same point in two places at once. It’s very tempting to conclude that the position defended by Gombrich and Cutting is right: we cannot perceive both the surface and the content of pictures at the same time. The upshot of this discussion of simultaneity and twofoldness (theses to the effect that we have visual experiences of pictures as D and /D simultaneously) is that they do not offer a plausible alternative to the account given of visual ambiguity in bottom-up, naturalistic picture perceptions. Such pictures can admit a form of visual ambiguity which implies that certain visual processes which lead from the D sketch to the /D sketch are penetrable by consciously driven processes. This modulation of visual processes does not just change which data is processed (as occurs in cases of attention-shifting) but the way visual data is processed. I think it unlikely that all bottom-up picture perceptions are subject to this form of ambiguity, but it suffices to establish the penetrability of the relevant visual processes that many of them are. The robustness of the picture contents caused by naturalistic pictures—the difficulty we have in generating visual experiences with D contents while looking at naturalistic pictures—favours depiction and is an enabling condition for pictorial representation. For picture perception as we know it to occur, we have to be able to continue to exploit the depth representations formed by earlier cognitive stages even after we have reached the stage at which we can form doxastic states. Otherwise, the mental representation of what the picture represents would evaporate each time we stopped making an effort to direct spatial attention in the right way. Imagine the opposite scenario. If the /D visual interpretation did not prevail naturally each time we looked at the picture, perceiving pictures would be a painfully slow and attention-consuming activity; we would only be able to understand one part of a picture at a time with considerable effort. In that case, seeing the opaque cube-drawing as a cube would be as hard to do as it is now to see the cube-drawing as a flat surface. Picture perception would require great mental concentration on one region of the picture at a time. Higher-level

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

attention is a valuable and energy-consuming resource for the brain, and thus it is unlikely that depiction would be as widespread an activity as it actually is. Instead, the /D visual interpretations prevail, and depiction as we know it—especially the profusion of different individual styles in drawing and painting—has emerged by exploiting that robustness. At the same time, admitting cognitively driven picture perceptions alongside bottom-up ones (see Section ) can explain both why we can use found objects as fortuitous pictures and how we understand picture styles that exploit or frustrate visual procedures in different ways, either for aesthetic effect or due to technical limitations.

 Cognitively Driven Picture Perceptions In the second form of visual ambiguity supported by pictures, the picture surface causes a veridical visual experience with D content in a bottom-up way by the time we reach phenomenal awareness; but it can subsequently support a visual experience of a /D scene if we agentively control spatial attention, optionally with backup from conceptually encoded information. The less ambiguously and more fluently a surface causes a D visual representation, the greater the conscious effort required to use the surface to have an experience with volumetric and depth content. For example, in Figure .a, the visual system detects no occlusion. It is unlikely to detect a surface discontinuity because the change in colour is not sharp or regular enough. In the absence of other depth cues, the object is likely to suggest to the visual system a roughly planar, irregularly stained surface. (With some priming, however, we can see the figure as a convexity lit from the left.) Figure .b supports an experience of the lower and upper halves as foreground and background (see Section ). If we have trouble performing this segregation, it helps to think of the figure as a field with stalks against a grey sky. If the same object is viewed upside-down (.c), the figure–ground effect (a)

(b)

(c)

Figure . Ambiguous experience of volume

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  goes away, possibly because of an implicit assumption that light comes from above (see Ramachandran and Rogers-Ramachandran ). But in that case, we can reinstate the figure–ground segregation and the nonconceptual depth representation if we are given conceptual cues; e.g. if we’re told to see the figure as a sandy stretch in the foreground with reeds or shrubs in the background. Both the phenomenal character and the nonconceptual content of visual perception change when the figures are seen as either segregated figure and ground or as a convexity, instead of as a stained flat surface. According to theories that defend perceptual impenetrability (Pylyshyn ; : ; Raftopoulos : ), the processes that construct the nonconceptual volumetric representations in Marr’s /D sketch are part of early vision and thus claimed to be impenetrable. Perceptual representations of depth are part of Marr’s /D sketch. (The /D sketch is no less three-dimensional than the D sketch; the difference is that the /D sketch is ‘a viewer-centered representation of the depth and orientation of the visible surfaces’, the D sketch a representation whose ‘coordinate system is object centered’; Marr : .) But because conceptual information can determine which nonconceptual depth representations we have of the figures, this activity of visualizing or seeing-in is not just a case of perceptual ambiguity which can be contained within lower-level processes. Instead, it fits Pylyshyn’s definition of cognitive penetrability: the visual experience of depth is ‘altered in a way that bears some logical relation to what the person knows’ (Pylyshyn : )—namely, to conceptual information or memories of the textures of specific kinds of scenes. Thus, the processes that yield volumetric representations as outputs turn out to be cognitively penetrable. Note that for Figure .b, while we can have conceptual representations (a field of wheat against a grey sky), perhaps we can also have a nonconceptual depth representation without having to use any concepts, solely on the basis of the figure– ground segregation. However, in that case, the segregation would not have to be performed bottom-up; it is possible to perform it consciously by imagining that there is a distance depth-wise between the top and bottom parts of the figure.6 This would count as a cognitively driven (personal-level, though not conceptual) form of amodal completion. In Kanizsa and Gerbino’s terminology, the depth perception in Figure .—and perhaps .c, depending on the subject’s perceptual set—could be described as a case of represented completion as opposed to perceptual completion. As Kanizsa and Gerbino put it, such a construal of the figure does not give us the impression of being faced with something ‘objective,’ independent of us, not influenced by our will or our cognitive set. Indeed, properties such as phenomenal givenness and independence from the observer characterize a perceptual datum and distinguish it from a datum that is merely thought. (Kanizsa and Gerbino : ) 6 If we perform the figure–ground segregation, do we have the same visual experience of Fig..b with and without the concept? Siegel () thinks we do not; Lyons (: ) thinks that the conceptual phenomenology may be ‘a late experiential effect, leaving the nonconceptual early perceptual states unaffected but influencing the nondoxastic seemings’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

Briscoe () also calls similar completions ‘cognitive’; and elsewhere he uses the term ‘make-perceive’ to describe comparable top-down cognitive activities: ‘one engages in make-perceive when one projects or “superimposes” a mental image on a certain region of the visually perceived world’ (Briscoe : ). Like the claim about penetrability made in the last section, this one too is of a different kind to the claims made in Macpherson (). Macpherson discusses cases of perceptual ambiguity in which she argues that the phenomenal character of perception changes but the nonconceptual content can remain the same, whereas the cases of seeing-in described here bring about changes in content along with phenomenal character. This is because the cases discussed in Macpherson () essentially concern mental rotations of represented shapes without altering those shapes, whereas the cases described here concern alteration of the shapes themselves (from D to /D or vice versa). On the other hand, the form of penetrability described here may be compatible with part of Macpherson’s () account of the effect of mental imagery on vision.7 The need for top-down cognitive influence in picture perception is not limited to fortuitous pictures. Paintings and drawings in many styles require subjects to make an active personal-level effort to visually experience their contents. This can be as a result of technical limitations like absent or imperfect perspective (in Roman wall paintings, for instance), or deliberate choice in caricature and in partly abstract styles (expressionism, cubism, and so on). Many of Turner’s paintings, or parts of the paintings (of which Figures . and . are photographs), are in this category. While it is possible to perceive depth in those paintings, Ruskin held that Turner’s paintings teach the eye to be ‘innocent’ about content and focus on traits of the picture surface; for Ruskin, as Gombrich put it, ‘we do not even see the third dimension, only patches of colour and textures’ (: ).

 Picture Perceptions and Perceptual Beliefs To claim that cognitive and agentive states can modulate the outputs of visual processes is potentially to admit circularity into the justificatory relation between perceptual experience and belief. As Siegel () puts it, if Jill visually experiences Jack’s face as angry because she believes that Jack is angry, then visually experiencing Jack’s face as angry can no longer justify her belief that Jack is angry. Siegel’s question applies in the following way to the forms of visual ambiguity outlined. On one hand, the experiences generated when we see Figure . as flat, and when we see Figures .a–c as /D, are in some sense visual experiences. At the same time, those experiences are caused jointly by us—by agentively driven activities jointly with bottom-up processes. The 7 That is, I think Macpherson’s () thesis that colour perception is cognitively penetrable is contestable (see Zeimbekis ), but I agree with the general point that mental imagery can penetrate vision (in shape perception, at least).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  suggestion is that we can contribute to deciding which visual experiences we have; and this would vitiate the ability of visual experience to justify belief. Note that, in one sense, it is relatively trivial that we can decide which visual experiences to have. If we accept that we can never experience simultaneously all of the nonconceptual visual contents that a visual scene could cause, because we cannot focus attention on all parts of a visual scene at once,8 then, given that we are capable of agentively directing the focus of visual attention, it follows trivially that we choose what to visually experience. For example, if applying the concepts pine and tree to the same object required having different visual experiences of it, that would not require having contradictory contents and would not vitiate the justificatory relation: the visual experience which supports the concept pine would also support the concept tree; the concepts name kinds standing in a determination relation. But that is not what happens in cases of visual ambiguity—at least not in the ones described here.9 The ambiguity of Figures . and . allows us to have contradictory visual contents, so the threat is that those visual contents can support contradictory attributions or beliefs. There are two ways to block the inference from cognitive penetrability to the vitiation of perceptual justification. The forms of penetrability described afflict picture perception, so one response would be to claim that even if processes which generate volume perception in picture perception are cognitively penetrable, volume perception processes in object perception are not penetrable. Another response would be to show that of the two states caused in each case of visual ambiguity, at most one state has the roles of causing and justifying beliefs—in other words, that only one state counts as a perception functionally and epistemically. In fact, these two questions turn out to be very closely related. I’ll start from the second question, and this will give us the means to settle the first question. If the capacity of visual perception to justify beliefs is vitiated by either of the forms of visual ambiguity outlined, it is unlikely to be the second form, in which visual processes naturally yield a veridical perception of the picture’s surface as flat and we subsequently use the surface as a visual prop to visualize a D scene. It immediately seems suspect to suggest that a cognitively driven act of visualizing could compete as a state for perception’s causal and justificatory roles. When a fortuitous picture, like the wall stains described by Leonardo, naturally causes a visual experience of a flat surface, that experience causes a perceptual belief with the same content. Visual experiences of depth are produced after the formation of that perceptual belief by agentively controlling the focus of spatial attention. They are something we do akin to a mental action, not something that happens to us as the outcome of stimulusdriven and hard-wired processes functioning autonomously. Those states do not feel like perceptions, and there is no reason to think that they have the causal role of causing 8 9

For a description and plausible defence of this position see Naccache and Dehaene (). It does happen in Siegel’s example of perceptions of pine trees.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

perceptual beliefs which could compete with the belief caused by the initial, veridical visual experience. Matters are more complicated when it comes to penetration of visual processes which support naturalistic picture perceptions. Remember that on the present hypothesis, in naturalistic picture perceptions—much as in object perception—by the time we reach phenomenal awareness, we have already formed a mental representation of a /D scene. The processes which cause the /D representation turn out to be cognitively penetrable: with mental effort, we can succeed in having a veridical experience of the picture, or parts of it at a time, as a flat surface. Thus, it seems that a subject could choose to have a veridical or a non-veridical visual experience at will, and the experiences would have contradictory contents. But this time, we cannot explain away the falsidical experience as a consciously driven act of visualizing; because the experience caused by an act of visualizing is veridical this time. Therefore, there seems to be only one way to avoid the conclusion that subjects can have experiences with contradictory contents, which can justify contradictory beliefs. It is to argue that the /D visual experiences do not cause beliefs; in fact, to argue that, as states, they do not fulfil the functional role of perceptions and in that sense at least do not constitute perceptions. I will sketch an explanation of why I think this may be so. If I am right, then the fact that vision is cognitively penetrable does not—at least in such cases— imply any vitiation of perception’s role in justifying and grounding beliefs. Stimulus-driven picture perceptions take a free ride on the processes that subserve object perception, so they have much in common with object perceptions; but that does not mean that the two are identical. The key difference is that in picture perception the brain does not calculate volumes and depth relations by using either binocular disparity from stereopsis or parallax from movement. According to Marr (: ch. ), stereopsis is largely responsible for generating the ‘viewercentered representation of the depth and orientation of the visible surfaces’ (p. ) in object perception—the /D sketch which, as we saw, is likely to provide the nonconceptual content of visual experience. Picture perception, on the other hand, can exploit neither stereopsis nor parallax, and relies exclusively on the kinds of monocular depth cues described in Section . Therefore, binocular picture perception generates /D brain representations, and the nonconceptual contents of phenomenal awareness, not just by using monoptic perception, but by using a subset of the processes of monoptic object perception: not only is stereopsis absent, as it is in monoptic vision, but parallax from movement, which is available to monoptic object perception, is also absent from picture perception. As brain states, picture perceptions are very different to object perceptions, if only for these reasons (additional ones will be given shortly). These differences between the two kinds of visual states seem to be reflected by differences in their phenomenology. As Briscoe () points out, the case of Susan Barry (reported by Sacks ), who had monoptic vision and only experienced stereoptic vision for the first time at an advanced age, suggests that the visual experiences

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  caused by stereoptic and monoptic vision can be distinguished on phenomenological grounds: I noticed the edge of the open door to my office seemed to stick out toward me. Now, I always knew that the door was sticking out toward me when it was open because of the shape of the door, perspective and other monocular cues, but I had never seen it in depth. It made me do a double take and look at it with one eye and then the other in order to convince myself that it looked different. It was definitely out there. [. . .] While I was running this morning with the dog, I noticed that the bushes looked different. Every leaf seemed to stand out in its own little -D space. The leaves didn’t just overlap with each other as I used to see them. I could see the SPACE between the leaves. The same is true for twigs on trees, pebbles on the road, stones in a stone wall. Everything has more texture. (Sacks : ; quoted by Briscoe : )

Susan Barry’s reports of the phenomenology of her visual experiences suggest that depth is represented much less vividly10 in monoptic object perceptions than in stereoptic ones—despite the fact that here, parallax is available to the subject, unlike in picture perceptions. She also reported that before her vision was corrected, she predicted that she could imagine what stereoptic vision was like, but retracted the claim after she experienced stereopsis (Sacks : ), which suggests that stereopsis acquainted her with a new kind of experience. Briscoe describes the difference between seeing monoptically and seeing stereoptically as a ‘dramatic influence of binocular depth information on the spatial phenomenal character of our visual experience’ (: ). His explanation of the phenomenological difference is that after the brain calculates three-dimensional shape on the basis of binocular disparities in the light information reaching the retina, the information and processes are ‘not lost in our conscious visual experience of the object. Indeed . . . we can literally see the difference made by their presence (and absence) in the light available to the eyes’ (Briscoe : ). How does this case throw light on picture perceptions? The difference between object perceptions and natural, stimulus-driven picture perceptions is that between (a) constructing the depth representations available to phenomenal awareness by means of stereopsis, parallax, and hard-wired processes that interpret monocular cues, and (b) constructing those representations only by means of monocular cues. Susan Barrytype experiences are subserved by an intermediate state, (c): parallax plus monocular cues, but no stereopsis. Since there is a phenomenological difference between (a) and (c) that a subject whose physiology is working in the relevant respect (unlike Susan Barry before her vision was corrected) is able to distinguish, there is all the more reason to think that a subject should be able to tell apart (a) normal object perception from (b) picture perception. Apart from the suppression of stereopsis and parallax, there are other grounds for distinguishing (natural, stimulus-driven) picture perceptions from object perceptions, 10

Whether or not this implies a change in representational content; a point I will discuss.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

both as states and phenomenologically. As we saw in Section , Matthen (: –) has argued that pictures do not engage the motion-guiding visual system, but only the visual pathway dedicated to recognition and to representing object features (the distinction is made by Ungerleider and Mishkin , and Milner and Goodale ). Matthen claims that this physical difference is reflected phenomenologically by the fact that picture perceptions do not cause a ‘feeling of presence’ (: ). Lack of engagement of the motion-guiding system is a functional effect. The absence of the brain states and processes that calculate disparity and parallax, the lack of engagement of the dorsal visual system, and increased reliance on monocular cues may not affect only the phenomenology of picture and object perceptions: it may also mean that the causal antecedents for perception to play its functional role of causing perceptual beliefs are not satisfied, something which could directly prevent even natural, stimulus-driven picture perceptions from causing perceptual beliefs. Another feature of picture perception which is compatible both with the lack of stereopsis and parallax and with the lack of engagement of the motion-guiding system, could be the lack of subject independence. Subject independence is claimed by Siegel () to be part of the phenomenal character of object perceptions. This feature is absent from picture perceptions not only because the locations of objects represented in picture perception are insensitive to parallax, but also because, as Cutting () has shown, shapes represented in picture contents withstand considerable changes in the viewing angle. I have given a number of reasons why the visual states caused by natural, stimulusdriven picture perceptions should differ from object perceptions, and some evidence and reasons why the phenomenology of the states should also differ. Now, is this difference between object perceptions and natural picture perceptions a difference of sensory or cognitive phenomenology? According to Sacks (: ch. ), the suppression of stereoptic representations of depth can have an impact on representational content itself; for example, it can lead to miscalculations about the shape and location of objects. And Briscoe, as we saw, describes the phenomenological difference between monoptic and stereoptic vision as being due to the ‘influence of binocular depth information on the spatial phenomenal character of our visual experience’, and writes that ‘we can literally see the difference’ made by calculation of shape from binocular disparities (: ; emphasis added). This suggests that the phenomenological contrast between stereoptic and monoptic vision is due to differences of representational content. Nevertheless, differences of representational content may not exhaust the phenomenological differences between stereoptic and monoptic vision, nor those between object perceptions and natural picture perceptions. One reason for this is provided by Matthen’s proposal. Lack of engagement of the motion-guiding system by pictures would have no effects on the object features represented visually, but it would have the effect of suppressing the epistemic feeling of presence, according to Matthen. In that case, the contrast between the phenomenologies would not co-vary with changes to representational

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  content. That does not necessarily mean that the missing ingredient has to be a form of cognitive phenomenology.11 Perhaps the epistemic feeling of presence supervenes on early, unconscious visual processes; that is certainly suggested by Matthen’s () account (see Section ).12 A similar point applies to subject-independence, which Siegel considers part of perceptual experience: if it is absent from picture perceptions, that does not mean that picture perceptions and object perceptions differ in terms of cognitive phenomenology; yet nor is the difference one of visual contents. When reporting her experiences, Susan Barry contrasts representing the orientation of a door ‘because of the shape [and] perspective’ to ‘seeing it in depth’, and representing the ‘overlap’ of leaves to ‘seeing the space between the leaves’. There are different ways to understand these passages. It is true that the closer objects are to the viewer, the more binocular disparity can reveal of the space that would be occluded by either one of the monocular views. On one reading, this is what Susan Barry is reporting, and it is a change in phenomenology that co-varies with a change in visual representational content. However, such changes to representational content can also be obtained through parallax, so their phenomenology should have been familiar to Susan Barry before her vision was corrected—they would not constitute a novel form of phenomenological vividness or salience. So it is possible that the passages report a phenomenological change that does not co-vary with representational content. Briefly, Susan Barry’s reports may not be about seeing more occluded space than before, but about seeing space differently. To conclude, perhaps stereoptic and monoptic vision differ phenomenologically not only because they have different visual contents, but also in ways that overflow any differences between their visual contents.

 Illusory Picture Perceptions This account of the causal and epistemic role of picture perceptions is supported by a consideration of pictures that do cause illusions: trompe l’oeils. ‘Trompe l’oeil’ is an ambiguous term which sometimes designates a picture and sometimes the effect of a picture. Here, I use the term for pictures in contexts in which they succeed in causing higher-level illusions, typically signalled by a sense of surprise when the illusion is dispelled. (I do not use the term for paintings that are extremely naturalistic but appear in contexts or conditions in which they do not produce the illusion, like William Harnett’s Old Models; see Goldstein : .) Figure ., while it is only a photograph of a picture (a mural), provides a good illustration.13

11

Thanks to an anonymous OUP reviewer for asking me to pursue this question. See also Dokic and Martin, this volume, Ch. , who describe epistemic feelings as ‘the output of a monitoring process which involves implicit inferences from a set of internal cues, such as availability of partial information or fluency’. 13 Thanks to John Pugh for generously granting me permission to reproduce this work. 12

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

Trompe l’oeils, for as long as they work, make us ascribe properties represented by the picture to locations in the space represented by the picture. The picture contents— nonconceptually represented shapes, colours and textures, concepts for properties and kinds, and object files for objects to bind the features and instantiate the properties— reach doxastic-level awareness wholesale, and form the structured contents of a perceptual belief. We could form such a belief if we turned a corner and came upon the building in Figure .. That in turn implies that the phenomenal character of the experience, and the causal role of the state, are indiscernible from those of an object perception. Trompe l’oeils do not get that far under any contextual conditions. A precondition for their success is that the picture should not be presented as a picture, something which would affect ‘perceptual set’: the preparedness for certain categories of object, scene, or action which is capable of biasing the visual processing of depth relations such as figure–ground segregation and could therefore defeat the trompe l’oeil’s ability to cause an illusory belief (see Vecera : –, for an overview of the concept of perceptual set, and Peterson and Hochberg  for the claim that perceptual set can affect figure–ground segregation). This is why successful trompe l’oeils usually appear in particular visual contexts, which could be described as unexpected in one sense, but also—in another sense—as consistent with expectations. On one hand, they benefit from being embedded in contexts where we do not expect to see a picture, which is why they are often painted on architectural features like external or internal walls, ceilings, and columns. On the other hand, the content of the trompe l’oeil should be consistent with the context in which it is embedded, which is why contents frequently consist of fluting, panels, mouldings, and other features consistent with the architectural setting. When such contextual conditions are met, even coarsely made pictures which are not naturalistic can momentarily fool the mind into confusing the picture perception with an object perception.

Figure . John Pugh, Siete Punto Uno. Reproduced by kind permission of the artist: artofjohnpugh.com

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  But the condition which is of direct interest here is that trompe l’oeils work as long as the absence of binocular disparity can go undetected and as long as parallax is neutralized, if we are not stationary. Thus, the ability of the picture to cause the illusionistic effect increases with the viewer’s distance from it, and the distance required is itself proportionate to the depth represented in the picture contents: the shallower the scene represented, the smaller the binocular disparities it would cause and the harder it is to detect their absence. A frequently used theme in trompe l’oeils are false window casings painted high on a building, which meet all of these conditions, as well as the earlier conditions connected with perceptual set. When trompe l’oeil illusions are dispelled, it is usually because the absence of parallax is detected as we move relative to the wall, column, or ceiling in which the picture is embedded. Once the illusion is dispelled, the experience of the picture seems to change. For example, when we realize that the lines suggesting the presence of a window casing are planar and that there are no window casings where we represent them, we can still mentally represent the casings—in other words, the content of the picture—as we look at the building. Support for this claim comes from the phenomenon that Kubovy () calls ‘robustness of perspective’: picture perception is relatively insensitive to changes in the angle from which the picture is viewed; according to Cutting (), pictures can be rotated up to  degrees (on their vertical axis) without this deforming the three-dimensional representations they cause. This is a very substantial degree of rotation, so we should be able to represent picture contents throughout the movements that make us detect the lack of parallax. What changes throughout the movements is the illusory nature of the picture contents, not the contents themselves. The fact that picture perceptions and trompe l’oeil illusions represent the same contents under different psychological modes or attitudes corroborates the claim that picture perceptions are distinct from perceptual illusions. Therefore, trompe l’oeils are rule-proving exceptions to the theory that picture perceptions and object perceptions have different phenomenal characters. They are exceptions because they produce picture perceptions which are momentarily indiscernible from object perceptions. But they are rule-proving exceptions because what makes them indiscernible from object perceptions is the fact that we do not detect the absence of binocular disparity and parallax. As such, they confirm the thesis that lack of stereopsis and parallax underlie the phenomenal difference between the remaining cases of bottom-up picture perception and object perceptions.

 Conclusion A key part of vision—the processes that generate volumetric shapes and depth relations from monocular cues—is cognitively penetrable, with the result that what subjects visually experience can be a function of conscious acts of visualizing and semantic information. One would expect the cognitive penetrability of those processes

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

to damage their capacity to warrant and justify beliefs. However, the epistemic situations in which subjects find themselves provide no scope for cognitive penetrability to cause such damage. Consider a simple illustration of these epistemic situations. When we view Rubin’s vase/face figure, the ambiguity is restricted to visual experiences; it does not translate into any ambiguity at the level of perceptual beliefs, because neither of the visual experiences causes a perceptual belief that there is either a vase or a pair of faces located before us. Issues of warrant and justification simply do not emerge. They would emerge only for visual experiences of the figure as a flat, stained surface—and in that case, in a way that does not damage vision’s capacity to provide perceptual warrant. The same applies to all the ambiguous figures classically used in the penetrability literature. To the extent that they are pictures, they can support forms of visual ambiguity that imply cognitive penetrability; but not in ways that entail the epistemological consequences usually expected of cognitive penetrability. Could we conclude that the processes which generate volume perception are cognitively penetrable only during picture perception, not in the wider context of the processes that take place in object perceptions? Think of object perceptions on a range. At one end are cognitively driven picture perceptions; next are pictures with weak depth cues; after that, naturalistic pictures with good depth cues, followed by trompe l’oeils; and finally, object perceptions in good viewing conditions. Object perceptions would be even harder to visually experience as D than trompe l’oeils, which are already very hard to visually experience as D. Stereopsis, visual exploration using parallax, and possibly the engagement of the dorsal stream’s motion-guiding systems would make it so difficult to have such visual experiences that we would have to actively visualize the scene before us as being D with a great mental effort. If it was possible to have such a state, it would have to be by keeping still to prevent the effect of parallax, and viewing the object through one eye to stop the brain calculating binocular disparity; even then, we would still have monocular cues to contend with, so such a state could still not emerge solely on the basis of stimulus-driven and hard-wired processes. Now, if we could not get to the point of having such a state, the processes which generate volume perception would not be cognitively penetrable in visual object perception taken as a wider set of visual processes than picture perception; they would only be penetrable in picture perception. This would be an interesting outcome, because challenges to impenetrability and modularity from visual ambiguity always use pictures as examples: pictures are used to show that there is penetration of some process, and then the conclusion is that that process is generally penetrable. The conclusion would be false if in object perception we simply could not get ourselves into visual states whose contents were two-dimensional. But suppose that we could have such states. It is plausible that we can, because viewing conditions can be far from optimal in object perception (especially for distant objects, for the reasons seen in the discussion of trompe l’oeils), and the visual system is made to be able to deal with such conditions (just as it seems to do when we can make

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  sense of non-naturalistic pictures). Then—for the same reason that cognitively driven picture perceptions do not cause perceptual beliefs—these consciously sustained mental representations of real scenes as flat would not cause perceptual beliefs either. They would be visual experiences that we ourselves actively sustain as agents by performing acts of visualizing, not states that happen to us like purely stimulus-driven and hardwired visual contents. Yet, it would remain that the processes which generate volume perception in object perception are cognitively penetrable; it is just that the states resulting from penetration would not cause beliefs, and could therefore not vitiate the justification of beliefs by visual perception. In either case—whether the visual processes that construct volume and depth are penetrable only in picture perception, or also in object perception—it transpires that we can admit the cognitive penetrability of fundamental visual processes without threatening the epistemic relation between visual perception and belief. So we do not have to deny cognitive penetrability to uphold the perceptual justification of belief.

References Abell, C., and Currie, G. (). Internal and external pictures. Philosophical Psychology (): –. Biederman, I. (). Recognition by components: a theory of human image understanding. Psychological Review : –. Biederman, I. (). Visual object recognition. In S. Kosslynand and D. N. Oshershon (eds), An Invitation to Cognitive Science, –. Cambridge, Mass.: MIT Press. Biederman, I., and Kim, J. (). , years of depicting the junction of two smooth shapes. Perception : –. Block, N. (). Consciousness and accessibility. Behavioral and Brain Sciences : –. Block, N. (). On a confusion about a function of consciousness. Behavioral and Brain Sciences : –. Repr. in N. Block, O. Flanagan, and G. Güzeldere (eds), The Nature of Consciousness: Philosophical Debates (Cambridge, Mass.: MIT Press, ). Block, N. (). Two neural correlates of consciousness. Trends in Cognitive Sciences (): –. Briscoe, R. (). Vision, action and make-perceive. Mind & Language : –. Briscoe, R. (). Mental imagery and the varieties of amodal perception. Pacific Philosophical Quarterly (): –. Bulthoff, H., and Edelman, S. (). Psychophysical support for a two-dimensional view interpolation theory of object recognition. Proceedings of the National Academy of Sciences of the United States of America : –. Campbell, J. (). Reference and Consciousness. Oxford: Oxford University Press. Cavanagh, P. (). The artist as neuroscientist. Nature (): –. Churchland, P. M. (). Perceptual plasticity and theoretical neutrality: a reply to Jerry Fodor. Philosophy of Science : –. Cutting, J. (). Rigidity in cinema seen from the front row, side aisle. Journal of Experimental Psychology (): –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. zeimbekis

Cutting, J., and Massironi, M. (). Pictures and their special status in cognitive inquiry. In J. Hochberg (ed.), Perception and Cognition at Century’s End, –. Orlando, Fla.: Academic Press. Edelman, S., and Bulthoff, H. (). Orientation dependence in the recognition of familiar and novel views of three-dimensional objects. Vision Research (): –. Evans, G. (). The Varieties of Reference. Oxford: Oxford University Press. Fodor, J. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Fodor, J. (). A reply to Churchland’s ‘Perceptual plasticity and theoretical neutrality’. Philosophy of Science (): –. Gibson, J. (). The Ecological Approach to Visual Perception. New York: Taylor and Francis. Goldstein, B. (). Pictorial perception and art. In B. Goldstein (ed.), Blackwell Handbook of Sensation and Perception, –. Oxford: Wiley-Blackwell. Gombrich, E. ([]). Art and Illusion: A Study in the Psychology of Pictorial Representation. London: Phaidon. Kanizsa, G., and Gerbino, W. (). Amodal completion: seeing or thinking? In B. Beck (ed.), Organization and Representation in Perception, –. Hillsdale, NJ: Erlbaum. Kosslyn, S. (). Image and Brain: The Resolution of the Imagery Debate. Cambridge, Mass.: MIT Press. Kourtzi, Z., Grodd, W., and Bulthoff, H. (). Representation of the perceived -D object shape in the human lateral occipital complex. Cerebral Cortex : –. Kubovy, M. (). The Psychology of Perspective and Renaissance Art. Cambridge: Cambridge University Press. Lamme, V. (). Neural mechanisms of visual awareness: a linking proposition. Brain and Mind : –. Lamme, V. (). Why visual attention and awareness are different. Trends in Cognitive Sciences (): –. Leonardo daVinci (). On Painting. New Haven, Conn.: Yale University Press. Levinson, J. (). Wollheim on pictorial representation. Journal of Aesthetics and Art Criticism (): –. Logothetis, N., Pauls, J., and Poggio, T. (). Shape representation in the inferior temporal cortex of monkeys. Current Biology (): –. Lopes, D. (). Understanding Pictures. Oxford: Oxford University Press. Lyons, J. (). Circularity, reliability, and the cognitive penetrability of perception. Philosophical Issues : –. Macpherson, F. (). Ambiguous figures and the content of experience. Noûs (): –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. Marr, D. ([]). Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. Cambridge, Mass.: MIT Press. Matthen, M. (). Seeing, Doing, and Knowing: A Philosophical Theory of Sense Perception. Oxford: Clarendon Press. Milner, A., and Goodale, M. (). The Visual Brain in Action. Oxford: Oxford University Press. Naccache, L., and Dehaene, S. (). Reportability and illusions of phenomenality in the light of the global neuronal workspace model. Behavioral and Brain Sciences : –. Peacocke, C. (). Depiction. Philosophical Review : –. Peacocke, C. (). A Study of Concepts. Cambridge, Mass.: MIT Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

seeing, visualizing, and believing  Peterson, M., and Hochberg, J. (). Opposed-set measurement procedure: a quantitative analysis of the role of local cues and intention in form perception. Journal of Experimental Psychology: Human Perception and Performance : –. Pylyshyn, Z. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences : –. Pylyshyn, Z. (). Seeing and Visualizing: It’s Not What You Think. Cambridge, Mass.: MIT Press. Pylyshyn, Z. (). Things and Places: How the Mind Connects with the World. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Ambiguous figures and representationalism. Synthese (): –. Ramachandran, V., and Rogers-Ramachandran, D. (). Seeing is believing. Scientific American—Mind (Jan.): –. Rock, I. (). Perception. New York: Scientific American Library. Sacks, O. (). The Mind’s Eye. New York: Knopf. Schier, F. (). Deeper into Pictures: An Essay on Pictorial Representation. Cambridge: Cambridge University Press. Siegel, S. (). Subject and object in the contents of visual experience. Philosophical Review (): –. Siegel, S. (). The Contents of Visual Experience. New York: Oxford University Press. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs (): –. Tarr, M. (). Rotating objects to recognize them: a case study of the role of viewpoint dependency in the recognition of three-dimensional objects. Psychonomic Bulletin and Review (): –. Tarr, M. J., and Pinker, S. (). Mental rotation and orientation-dependence in shape recognition. Cognitive Psychology (): –. Tye, M. (). The Imagery Debate. Cambridge, Mass.: MIT Press. Ullman, S. (). Three-dimensional object recognition based on the combination of views. Cognition : –. Ungerleider, L., and Mishkin, M. (). Two cortical visual systems. In D. Ingle, M. Goodale, and R. Mansfield (eds), The Analysis of Visual Behavior, –. Cambridge, Mass.: MIT Press. Vecera, S. (). Toward a biased competition account of object-based segregation and attention. Brain and Mind : –. Walton, K. (). Depiction, perception, and imagination: responses to Richard Wollheim. Journal of Aesthetics and Art Criticism (): –. Walton, K. (). Mimesis as Make-Believe: On the Foundations of the Representational Arts. Cambridge, Mass.: Harvard University Press. Waltz, D. (). Understanding line drawings of scenes with shadows. In P. H. Winston (ed.), The Psychology of Computer Vision, –. New York: McGraw-Hill. Wollheim, R. (). Art and its Objects. Cambridge: Cambridge University Press. Wollheim, R. (). Painting as an Art. Princeton, NJ: Princeton University Press. Wollheim, R. (). On pictorial representation. Journal of Aesthetics and Art Criticism (): –. Zeimbekis, J. (). Color and cognitive penetrability. Philosophical Studies (): –. Zeimbekis, J. (). Pictures, perception and meaning. MS.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

PA RT V

Cognitive Penetrability and Nonconceptual Content

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Cognitive Penetration and Nonconceptual Content Fiona Macpherson

In recent work, I have argued that perceptual experience can be penetrated by cognitive states, such as beliefs and desires (Macpherson ). I have done so in two ways. One was by claiming that the best explanation of some colour experiments was that cognitive penetration was occurring. The second was by proposing a mechanism that could explain how at least some instances of cognitive penetration might occur: beliefs and desires generate perceptual imagery which adds to, contributes to, or alters perceptual experience. The mechanism allows one to explain how cognition could affect perception whilst making reference only to psychological phenomena, the existence of which is supported by good independent evidence. In addition to providing a further reason to believe in the existence of cognitive penetration, a second motivation for proposing the mechanism, which unlike the first I did not articulate in that paper, is that it allows one to endorse the existence of cognitive penetration while retaining a view of perceptual experience that has many attractive features. One of these is that experience has nonconceptual content; and this feature is the focus of this chapter. While some have argued that perceptual experience can be nonconceptual only if it is not cognitively penetrated (Raftopoulos and Müller ; Raftopoulos ), I disagree. This chapter consists in making good the claim that the occurrence of cognitive penetration by means of the interaction of perceptual imagery and perceptual experience is compatible with experience having nonconceptual content. In so doing, I will explain that mechanism and also examine more generally the relationship between cognitive penetration and nonconceptual content. I distinguish between ‘classic’ cognitive penetration and cognitive penetration ‘lite’. I argue that all forms of cognitive penetration lite are compatible with all reasonable specifications of nonconceptual content, while classic cognitive penetration is compatible with only one specification of nonconceptual content. In Section , I introduce the notion of the content of experience and cite the reasons that have motivated people to think that a nonconceptual view of the content of experience is plausible. I clarify and refine the claim that cognitive penetration and

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

nonconceptual content are compatible in order to set aside a trivial sense in which they are compatible. In Section , I outline the cognitive penetration claim that I am discussing, which is one about perceptual experience, and distinguish it from one that can be made about brain mechanisms, in particular the early visual system. In Section , I explain one model of cognitive penetration—the classic model— and consider whether it is compatible with four different notions of nonconceptual content. I argue that classic cognitive penetration is compatible with one, but only one, of these notions of nonconceptual content. Thus, if one thought that the only kind of cognitive penetration was classic cognitive penetration, then it would be tempting to think that, at least in the core senses of nonconceptual content, cognitive penetration and nonconceptual content were incompatible. In Section , however, I outline the other model of cognitive penetration—cognitive penetration lite. I explain the reasons for thinking that there could be such a form. One reason derives from consideration of particular examples of cognitive penetration. Another reason stems from consideration of one mechanism that would explain some cases of cognitive penetration. I consider, in Section , whether cognitive penetration lite is compatible with the four accounts of nonconceptual content elucidated in Section . I show that it is straightforwardly incompatible with the first account of nonconceptual content, but compatible with the third and fourth accounts. The third account is, I believe, the standard and most influential account of nonconceptual content. I argue further that there is a crucial ambiguity in the second account of nonconceptual content on which the question of the compatibility of cognitive penetration lite with that account of nonconceptual content turns. I argue that disambiguated one way there is compatibility. Disambiguated another way, that account becomes almost equivalent to the first account of nonconceptual content. That is the account of nonconceptual content that is not compatible with cognitive penetration lite. However, I provide reasons to think that that specification of nonconceptual content is spurious and ought to be abandoned. Thus, I argue that cognitive penetration lite is compatible with all three reasonable specifications of nonconceptual content.

 Setting the Scene: Conceptual and Nonconceptual Content Propositional attitudes, such as beliefs, desires, occurrent thoughts, and judgments, represent, or are about, the world. Another way of saying this is that they are states that have content, and the content is what is represented by the state. For example, if Morag believes that Partick Thistle are the best team in Glasgow then her belief represents that Partick Thistle are the best team in Glasgow, and one can say that the content of her belief is that Partick Thistle are the best team in Glasgow. Propositional attitudes are attitudes, such as holding true (in the case of belief) or wanting to be true (in the case of desire), that we take towards propositions. Different views hold that

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  the proposition either is, or expresses, the content of the propositional attitude. There are many options as to what one should take the content of a propositional attitude to be: a proposition or proposition-like entity composed of concepts; a proposition or proposition-like entity composed of objects, properties, and relations; the objects, properties, and relations, or the state of affairs specified by the relevant proposition; or the set of possible worlds in which the relevant proposition is true; or something else. Are perceptual experiences representational states that have content? This is a difficult question to answer. Many people think that they are (although some think that they involve a more direct relation to the world—one of acquaintance or presentation). One oft-cited reason for thinking that they are representational states is that when having a perceptual experience the world seems a certain way to one.1 For example, it may seem to one as if a black pentagon is in front of one. And the way it seems would be the way that the world is represented to one as being. Another reason often given for thinking that perceptual experiences have content is that they have accuracy conditions.2 The idea is that one can always assess a perceptual experience for accuracy with respect to the actual world. That one can, suggests that one can identify the circumstances in which the experience is, or would be, accurate (there is a black pentagon in front of you), and that those circumstances are to be identified with what the experience represents. There is debate about whether the reasons set out above for thinking that experiences have content are sufficient. Some people consider they are not, for they think that the fact that a state can be associated non-arbitrarily with a proposition does not show that the state has content. Such people wish to impose stricter conditions on what it is for a state to have content. What further conditions should be imposed, and whether experience meets them, is a matter for debate, discussion of which would take me too far from my present purposes. I will simply assume that perceptual experiences represent and have content. Doing so will have no effect on the debate that follows. That debate is whether, on the assumptions that experiences have content and that some experiences have some nonconceptual content, nonconceptual content is compatible with experience being cognitively penetrated. As in the case of propositional attitude representation, there are different options as to what one can think the content of a perceptual experience as of a black pentagon is: a proposition or proposition-like entity that represents a black pentagon (that may or may not be composed of concepts); a black pentagon located in space in front of one; or the set of all worlds in which there is a black pentagon in front of one; or something else. What is it for content to be conceptual and what is it for it to be nonconceptual? It is difficult to give a pithy answer to this question as there are different notions of what is required—notions that I will articulate at length in Section . However, there are two claims that are common to all notions. The first is, as Bermúdez and Cahen 1

See e.g. Byrne ().

2

See e.g. Siegel ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

() state, that the notion of nonconceptual content is contrastive—it is elucidated by distinguishing it from some particular conception of conceptual content. The second is that some important aspect of explaining how states with nonconceptual content represent the world can be done without postulating a crucial role for concepts. A common conception of conceptual content is that it is the kind of content that the propositional attitudes have. One reason for thinking this is that a crucial role is postulated for concepts in the having of propositional attitudes: in order for a subject to believe or desire something, he or she needs to possess the concepts that are required to specify what it is that is believed or desired. For example, in order to believe or desire that Partick Thistle will be promoted this year, Morag has to possess the concepts of Partick Thistle, of promotion, of this year, and of being. What exactly it is to possess a concept is one of the most difficult issues to address in philosophy. Many different answers have been given and contested. A review of these would take me too far from my present purposes, but roughly speaking, she has to be able to think of these things, know what these things are in some sense, or have an idea of them. (Those people who hold that the content of propositional attitudes is a proposition composed of concepts will have a further reason to think that there is a crucial role for concepts in the having of a propositional attitude.) Because it is fundamentally contrastive, the notion of nonconceptual content, which is often attributed to perceptual experiences but also to subpersonal states, is often introduced by contrasting it with the content of the propositional attitudes. Nonconceptual theorists hold that experience has either just nonconceptual content or both conceptual and nonconceptual content. If experience has both conceptual and nonconceptual content, and only the conceptual content is affected by cognition, then, in a trivial sense, cognitive penetration is compatible with nonconceptual content— but that is an uninteresting sense only. What I am interested in, by contrast, is best expressed by the following two questions: (i) On the assumption that experiences only have nonconceptual content, can experiences be penetrated by cognitive states, such as beliefs and desires? (ii) On the assumption that experiences have both conceptual and nonconceptual content, can experiences be penetrated by cognitive states, such as beliefs and desires, so that they have different nonconceptual content from that which they would have had if unpenetrated? In the rest of this chapter, when I speak of ‘the question of whether cognitive penetration is compatible with nonconceptual content’ I will mean to address jointly questions (i) and (ii). Why is this question interesting? The view that experience has nonconceptual content is an attractive view of the nature of perceptual experience. This is because there is some reason to think that it provides good explanations of many features that it is plausible to believe perceptual experiences have. If one accepts that cognitive penetration occurs and one then finds out that it is not compatible with nonconceptual

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  content, then one would have to reject the idea of nonconceptual content and find a better explanation of the features of experience that motivate its postulation. Thus it is an empirical adequacy constraint on a view of the content of experience that it is compatible with cognitive penetration. Showing that the views are compatible allows one to endorse the attractive nonconceptual view of the nature of perceptual content. One might worry that this motivation does not carry much weight, for, as I have shown, there is a trivial sense in which cognitive penetration and nonconceptual content are compatible: nonconceptual content exists but it is unaffected by cognitive penetration. However, the examples of cognitive penetration that I argued existed in Macpherson () involved contents pertaining to fine-grained shades of colour. This is one of the central examples of contents that proponents of nonconceptual content wish to hold is nonconceptual (as I will explain). Therefore the motivation for thinking that this question is interesting should not be dismissed. I list below the features that it has been said that experience has that have motivated some to postulate nonconceptual content, and briefly indicate why one might think that they call for an explanation in terms of nonconceptual content. Such a list does not consist in a defence of the claim that experience has these features; and the reasons I cite for thinking that nonconceptual content provides a good explanation of them does not consist in a defence of the claim that nonconceptual content is the best explanation of them. There has been detailed argument about these points in the extant literature that I will not rehearse or adjudicate here. In this chapter I am simply assuming that experience has nonconceptual content, not arguing for it. (I am also assuming that there is cognitive penetration—not arguing for it.) In addition to bringing to the reader’s attention the reasons that one might have for holding that experience has nonconceptual content, another reason for listing them is to be in a better position, later in the chapter, to address the question of whether cognitive penetration is consistent with experience having nonconceptual content. To this end, here are six alleged features of experience that have motivated people to hold that experiences have nonconceptual content: I. Experiences can be, and often are, very fine-grained. Some people hold that such experiences are more fine-grained than our conceptual capacities. For example, we can see and experience millions of different shades of colours. We do not typically possess names for each of these shades. But even if we assigned names to each of the shades, we would be unable to attach the labels to the shades at a later time by means of just looking at them using only our memory—at least if more than a few seconds have passed. Thus, after seeing or experiencing specific shades of colour, we cannot keep them in mind for more than a few seconds. So one might hold that we do not have concepts of each of the specific shades of colour that we can see and experience, except when seeing or experiencing them, or just seconds afterwards. I may of course keep in mind the names of some of these shades—such as ‘burnt umber’—or

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

descriptions of these shades—such as ‘the shade of colour of your sofa’. But as we cannot pick out the shades of colour that these refer to at a later time, there is reason to think that we do not fully have in mind concepts of these colours in the way that we do when we are experiencing the shade. Thus, one might think that experiences can represent in more detail than my concepts—or at least the concepts that I have when not seeing or perceiving. Perhaps when seeing and experiencing, I can come to have concepts as fine-grained as my experience, but these inherit their content from those experiences, and they disappear seconds after the experiences disappear. (These considerations do not apply to categories of shades of colour. We do possess words for and concepts of red, blue, green, pink, orange, and so on, and can attach these labels appropriately to shades that we see or experience.) This fineness of grain of experience with respect to shades of colour applies to many other properties, such as pitches and volumes of sound and the size and shape of things.3 II. Experiences represent in a unit-free manner. Peacocke () brings attention to the fact that when our experience represents the size of objects, say the length of the table in front of me, it does not represent its length in feet or centimetres, or any other unit. Likewise, when my experience represents how loud a sound is, or how cold the room is, the quantity is not given in any particular units of measurement. He argues that conceptual representation cannot represent in this fashion. III. Experiential representation is analogue nature. Dretske () claims that conceptual representation is digital and chunks together information, in the way that a digital watch chunks together units of time into seconds or minutes and represents which second or minute it is but not when within that second or minute time-frame the present lies. In contrast, an analogue watch, with its constantly moving hands, represents not only seconds and minutes but when within those units the present lies. The position of the hand varies as the time does. Unlike discrete, digital, conceptual representation, Dretske claims that our experiences of properties like colour, length, temperature, pitch, and volume are all continuous and analogue in nature, and hence (he claims) nonconceptual. The claim that experience is analogue is not the same as the claim that it is finegrained. Experience could be fine-grained while still being digital, by having lots and lots of digital categories. For example, a digital watch that included a digital count of the seconds passing would be fine-grained compared to one that only included a digital count of the minutes, but neither would be the same as an analogue clock. IV. Perceptual experiences that represent the world can be had by young children and animals. It is common to think that young children and animals do not 3 This consideration has been discussed by e.g. Evans (), Raffman (), Peacocke (), Tye (), and McDowell ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  possess concepts. If that is true, then if they can have experiences that represent the world, those experiences cannot have conceptual content.4 V. Plausibly, it is in virtue of having perceptual experiences that we acquire perceptual concepts, such as those of shape, size, colour, pitch, heat, sour, and malodourous. If we acquire such concepts through experience, then one might hold that experience must represent these features prior to our possessing the concepts required to canonically describe the content of the experience, and hence that there must be nonconceptual content. VI. The content of experience can represent contradictions. Plausibly, we can experience contradictions when having visual experiences as of impossible tri-bars and other impossible figures made and drawn by artists.5 One might hold that propositional attitudes with conceptual content cannot have contradictory content, and thus that the content of perceptual experience must be nonconceptual. However, one might think that this idea needs refining, for it is not obvious that states with conceptual content cannot be contradictory. For example, one can entertain the thought with conceptual content that P and not P, which represents a contradiction. One therefore might instead claim that one cannot be in a state with contradictory conceptual content when the attitude of that state asserts that the world is a certain way, as belief does, but which the mere entertaining of a thought, and other propositional attitudes, do not. Then one could claim that perceptual experiences assert that the world is a certain way, and that unlike belief they can represent the world to be contradictory. Thus one could argue that the content of perceptual experiences must be nonconceptual. Further, in light of dialetheists’ apparent beliefs that there can be true contradictions, one might wish to further refine the thesis to be that one cannot rationally be in a conceptual state that asserts that the world is a certain way, and represent it as containing true contradictions, but one can rationally have a perceptual experience that asserts that the world is a certain way, and represents it as containing true contradictions.6 Thus far, I have introduced the notion of content and of conceptual and nonconceptual content. I have listed six alleged features of perceptual experience that have motivated people to hold that perceptual experience has nonconceptual content. I have explained that the key questions that will be addressed here is whether cognitive penetration is compatible with nonconceptual content and why one should be interested in this question.

4

For further discussion see Bermúdez () and Peacocke (). See Crane () and Macpherson (). Crane also claims that one has such a contradictory experience when undergoing the waterfall illusion, also known as the motion after effect. It is far from clear, however, that one does undergo such an experience in that case. 6 Dialetheism was developed by Priest (), who would of course resist the thought that the view is irrational. 5

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

 Cognitive Penetration In this section, I address the question of what cognitive penetration is. I outline two cognitive penetration claims—one about brain mechanisms and one about perceptual experience. It is the claim about perceptual experience that is the focus of this chapter. There have been two forms of the claim that perception is cognitively penetrated. One form of the claim is about brain mechanisms, and has been discussed almost exclusively with respect to vision, to which I will also limit my discussion here. The claim is that early vision can be cognitively penetrated for ‘the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, that is, it can be altered in a way that bears some logical relation to what the person knows’ (Pylyshyn : ). The early visual system is defined functionally, as a system that takes attentionally modulated signals from the eyes (and perhaps some information from other sensory modalities) as inputs, and produces shape, size, and colour representations—representations of visual properties—as output. These basic representations are then processed further and at some point may be categorized and identified, drawing on a person’s memory and knowledge, as being objects and properties of particular kinds. Higher-level cognitive states can affect the position of one’s eyes and what one attends to, and that will of course usually have an effect on what processing takes place in early vision. For example, my desire to look to the left, caused by my belief that there is an interesting insect there, may cause me to move my eyes and focus my attention in a way that alters which input the early visual system receives and hence which processing takes place in it. But according to many that discuss this thesis, such as Pylyshyn (), those instances do not count as instances of penetration. This is because attention and eye movements are said to affect that which feeds into the early visual system, not the early visual system itself. Whether one should rule out these effects as instances of cognitive penetration is a matter of debate. Likewise, whether the constraints of semantic coherence and logical relations between that which does the penetrating and that which is penetrated should be a requirement for the existence of cognitive penetration is also a matter that is disputed. The second form of the claim about cognitive penetration concerns perceptual experience, not early visual processing. This is the claim that philosophers have tended to be interested in. (Psychologists have been mostly interested in the claim about early visual processing, although this is starting to change and they are becoming increasingly interested in this second claim about experience.) Broadly, the claim is that, with certain conditions fixed—what is perceived, the perceiving conditions, and the state of the sensory organ—it is possible for two subjects (or one subject at different times) to have different perceptual experiences on account of the differing content of the states of their cognitive systems. States of the cognitive system will include the having of propositional attitudes, such as beliefs and desires, the having of occurrent thoughts and judgments, the having of certain concepts, and the state of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  having certain concepts primed so that they are more likely to be used in the formation of propositional attitudes, occurrent thoughts, and judgments. Two experiences will count as different in this context if they have different representational contents— which, one might plausibly hold, in all or many cases will go hand in hand with having different phenomenal characters, as I will assume here.7 I said that this was ‘broadly’ the claim because one can debate (as one can in the case of the cognitive penetration claim regarding brain mechanisms) whether one should exclude the effects of attention as counting as instances of cognitive penetration. Likewise, one can discuss whether one ought to insist on a semantic or logical connection between the content of the state that penetrates and the perceptual experience—or indeed some stronger connection between the content of these states.8 Discussion of these two points, interesting though they are, lies outside the scope of this chapter. Deciding them one way or another is not relevant to my concerns. The claim that these two cognitive penetration claims—one about brain mechanisms and one about perceptual experience—should be distinguished is sometimes given lip service, but often not adhered to. For example, when discussing the claim about brain mechanisms, Pylyshyn () explicitly claims that he is not discussing a claim about perceptual experience and that one should not think of the early visual system as determining the nature of visual experience. Yet, he adduces evidence about whether experience is penetrated in defence of his position. The claim that I am considering here, and will refer to as ‘the cognitive penetration claim’, is the claim that (at least some) perceptual experiences are cognitively penetrated. Drawing on what has been said in this section and the previous one, the claim is that with certain conditions fixed—namely, what is perceived, the perceiving 7 The claim that there can be differences in the content of experience without a difference in the phenomenal character of experience of vice versa is a claim that strong representationalists would deny. Whether they are right, and on what grounds, has been discussed at length by many. I will assume it for convenience here, although I have argued elsewhere that the claim, in some forms, is not true. See Macpherson (; ; ). 8 For example, some people might think that there are cases that do not count as cases of cognitive penetration, even though the cases are ones in which a cognitive state causes a particular perceptual experience to occur and there is a semantic or logical connection between the content of a cognitive state and the content of the perceptual experience. They might deny that they are cases of cognitive penetration because they think that the semantic or logical connection between the states exists merely by accident. For example, suppose that Murdo believes that aliens are attacking Earth. This belief causes stress, which induces a migraine. Suppose that whenever Murdo has a migraine, he experiences flashing lights in the top half of his visual field. So suppose that Murdo visually experiences flashing lights in the sky on account of having his migraine. The content of this experience bears a semantic relation to the belief that caused it—the belief about the alien attack—but one might want to deny that this is a case of cognitive penetration. This is because it seems to be by chance that the nature of Murdo’s migraines are such that they produce experiences of flashing lights, and that that content bears a semantic relation to the belief that induced the migraine. In light of this kind of case, someone might want to claim that, in addition to a semantic relation between the content of the cognitive state and the perception, the intermediate state that does the penetrating—in the example just given, the migraine state—has to have the content that it does because the cognitive state that causes it has the content that it has. Or they might insist that there has to be a direct transfer of content from the cognitive state without going through any intermediate state into the perceptual state. Working out clear versions of these stronger claims that one might wish to stipulate lies beyond the scope of this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

conditions, the state of the sensory organ, (and perhaps the attentional focus of the subject)—it is possible for two subjects (or one subject at different times) to have perceptual experiences with different contents, on account of the differing content of the states of their cognitive systems. (Whether one should add, ‘in virtue of the content of the cognitive system having semantic, rational, or logical connections to the content of the perceptual experience’, I will leave for the reader’s decision.) I am not going to argue here that the cognitive penetration claim is true. I am only going to consider the question of whether cognitive penetration is compatible with nonconceptual content.

 Is Classic Cognitive Penetration Compatible with Nonconceptual Content? In this section, I outline one model of cognitive penetration that I call ‘classic’ cognitive penetration.9 On this model, the following takes place in a case of cognitive penetration: (a) You have a certain propositional attitude that p (e.g. you believe that computers look like grey boxes). (b) p involves a concept C (for example, the concept of being a computer), in the sense that C is required to specify the content that p and possession of C is required in order to have the propositional attitude that p. (c) The content of this propositional attitude causally affects your perceptual processes yielding an experience with the content that q (e.g. an experience with the content that there is a computer present) that involves C (in the sense that C is required to specify the content that q). (d) The content that q is a content that an experience could not have unless it was affected by propositional attitudes in this way. Classic penetration occurs when each of the conditions (a)–(d) occurs. It follows that, when classic penetration occurs, you could not have an experience with the content that q, unless you possessed the concept C. The question I will now address is whether classic cognitive penetration is compatible with nonconceptual content in the sense that I outlined in Section . There are different accounts of what nonconceptual content is, and hence of the difference between nonconceptual and conceptual content. I identify four different accounts, which I believe are exhaustive of the accounts that are found in the literature. I specify each account and, in turn, discuss whether that notion of nonconceptual content is compatible with classic cognitive penetration. 9 The classic model of cognitive penetration should not be confused with the classic view of perception. The classic view of perception is that cognitive penetration does not occur. The classic model of cognitive penetration is one model of what occurs when cognitive penetration occurs.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  The accounts of nonconceptual content detailed below specify what it is for a state to have nonconceptual content, on the assumption that it has content. In other words, these accounts do not specify what it is for a state to have content in the first place. They only specify what it takes for a state that has content to count as having nonconceptual content. When considering accounts of conceptual and nonconceptual content, I will not specify each time the caveat ‘on the assumption that the state has content’, but the reader should take it to be in place each time. The first account of nonconceptual content is that ‘the existence of cognitively impenetrable mechanisms is a necessary and sufficient condition for nonconceptual content’ (Raftopoulos and Müller : ).10 The idea is that if a state or an experience that has content is produced by a system that cannot be cognitively penetrated, then that state or experience has nonconceptual content. If a state or an experience that has content is produced by a system that can be cognitively penetrated, then that state or experience has conceptual content. Raftopoulos and Müller () and Raftopoulos () hold that early vision is a cognitively impenetrable system or mechanism and that some states of early vision can be identified with (at least some) visual experiences. They therefore believe that (at least some) perceptual experiences are not cognitively penetrated. Thus they maintain that those perceptual experiences have nonconceptual content. Raftopoulos and Müller are clear examples of people who believe, tout court, that if perceptual experience is cognitively penetrated then its content cannot be nonconceptual and must be conceptual. Classic cognitive penetration is clearly not compatible with this notion of nonconceptual content. Nonconceptual content is simply defined as that which is produced by a cognitively impenetrable mechanism. It should be noted that this conception of nonconceptual content is nonstandard. When the other notions of nonconceptual content are outlined, this conception of nonconceptual content will clearly stand out as being rather different from the rest, which are more similar to each other. One might wonder why this definition of nonconceptual content has been adopted by Raftopoulos and Müller. I believe that I have an answer to this question, which can be gleaned by considering in detail the other accounts of nonconceptual content. Once I have done so, I will return to answer this question in Section . The second account of nonconceptual content is one articulated by Bermúdez and Cahen (). They begin by claiming that the view that the contents of propositional attitudes and the contents of perceptual experiences are conceptual is motivated by the conjunction of two thoughts: () In specifying what a thinker believes, what a perceiver perceives, or what a speaker is saying by uttering a certain sentence in a particular context, one has to be as faithful as possible to how that thinker, perceiver, or speaker apprehends the world. 10

See also Raftopoulos () for affirmation of the same account.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

() How a thinker, perceiver or speaker apprehends the world in having beliefs about it, perceiving it, or speaking about it is a function of the concepts he possesses. They claim that nonconceptual theorists reject the second thought. Thus, they define nonconceptual content as being content that is not a function of the concepts possessed by the subject of the state that has that content. Much the same idea was mooted in earlier work by Bermúdez (: ) who claims: A creature has perceptions with conceptual contents . . . to the extent that its perceptual representations of the environment are determined by its classificatory and recognitional abilities.

This yields another way of saying that nonconceptual content is content that is independent of the concepts that the subject has: the content is not determined by those concepts. I will amalgamate these accounts and speak of them, for ease of reference, as the Bermúdez and Cahen account. Recall that on the classic model of cognitive penetration, you could not have an experience with a certain representational content unless you possessed a certain concept. Hence, when classic cognitive penetration occurs, the content of experience clearly is a function of, and is determined by, the concepts that the subject of the state with that content possesses. Therefore the content of a state that is classically penetrated must be conceptual according to this second account of the distinction between conceptual and nonconceptual content. Classic cognitive penetration is therefore clearly also incompatible with this second account of nonconceptual content. The third account of nonconceptual content is that an experience has nonconceptual content if and only if a subject of an experience with that content need not possess the concepts required to specify that content. (The experience will have conceptual content otherwise.) This account of nonconceptual content is the orthodox account of nonconceptual content. It is often called the ‘state account’ of nonconceptual content, and is frequently contrasted with the ‘content account’ that we will consider next. (I will also sometimes use the terms the ‘state sense’ of conceptual and nonconceptual content and the ‘content sense’ of conceptual and nonconceptual content to refer to these two accounts of the difference between the types of contents.) The state account is discussed by, among others, Evans (), Cussins (), Crane (), and Peacocke (; ). This account of nonconceptual content specifies a necessary and sufficient condition that the subject of an experience must meet in order for his or her experience to have nonconceptual content. It is important to note that this account of nonconceptual content is modal. Crucially, it does not say that an experience has nonconceptual content if and only if a subject does not possess the concepts required to specify the content of the state. Rather, it says that an experience has nonconceptual content if and

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  only if the subject need not possess those concepts in order to have the experience. So the subject may, in fact, possess the concepts required to specify the content of the experience. What is vital is that it is possible for the subject to have the experience and not possess those concepts. Again, recall that on the classic model of cognitive penetration, what your experience represents is such that you could not have that experience unless you possessed a certain concept. That concept is one that is required to specify the content of the experience. Thus, on this third account of nonconceptual content, when classic cognitive penetration occurs, the content of experience will count as being conceptual. Thus, classic cognitive penetration and nonconceptual content, conceived of as per the state account, are incompatible. Lastly, there is the fourth account of nonconceptual content. While the state account laid down conditions that the subject of the state must meet in order for the state to be nonconceptual, this ‘content account’ of nonconceptual content lays down conditions solely that the content itself must meet. According to the content account, an experience has nonconceptual content if and only if the content of the state does not have concepts as constituents. (The experience will have conceptual content otherwise.) This account was articulated and differentiated from the state account by Heck (). In Section , I mentioned what the contents of a state might be taken to be. I explained that they could be taken to be various things. Extant accounts have, variously, taken them to be () propositions composed of concepts, or () objects and properties, or () sets of possible worlds. According to the content account of nonconceptual content, if the content of experience is what the first of these three suggestions says it is, then it is conceptual content. The latter two conceptions of content are conceptions of nonconceptual content. That is because only in the second and third cases the content is not composed of concepts. The state account of nonconceptual content and the content account of nonconceptual content are completely independent of each other. The view that experience is conceptual in the state sense and the view that experience is nonconceptual in the state sense are each compatible both with the view that experience is conceptual in the content sense and with the view that experience is nonconceptual in the content sense. If an experience is cognitively penetrated on the classic model, that tells us nothing about what the constituents of the content are. Therefore, it is perfectly compatible with experience being classically penetrated that its content not be concepts but objects or properties, or sets of possible worlds, instead, and hence be nonconceptual content as stipulated by the content account. Thus, this fourth conception of nonconceptual content is compatible with classic cognitive penetration. Despite this independence of the state account of conceptual and nonconceptual content and the content account of conceptual and nonconceptual content, Bermúdez () argues that a good explanation of why experiences have conceptual content in the state sense (i.e. a good explanation of why there are some experiences that one

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

cannot have unless one possesses a certain concept) is that that concept is a constituent of the content—in other words that they have conceptual content in the content sense. On the basis of this, one might try to argue that, as classic cognitive penetration is incompatible with the state account of nonconceptual content, this provides a good reason to think that it is incompatible with the content account of nonconceptual content. One might try to claim that a good explanation of what goes on in classic cognitive penetration is that the content of a propositional attitude that p, which involves the concept C (in the manner specified at the start of this section), interacts with your perceptual processes by making C a constituent of the content of the experience—and thus giving it conceptual content according to the content account. However, cognitive penetration gives us a reason to think that there is another good explanation of why an experience that is classically cognitively penetrated cannot have nonconceptual content according to the state account: the propositional attitude that p, which involves the concept C, and hence requires possession of C, is simply a necessary cause of your having the experience. The concept C does not become a constituent of the content, but possession of it is necessary in order to have the experience. Thus, we need not think that if an experience has conceptual content according to the state account, it has to have conceptual content according to the content account. To summarize, we have seen that three out of four accounts of the difference between conceptual and nonconceptual content will count the content of an experience that is the result of classic cognitive penetration as having conceptual content and not nonconceptual content. The exception was the fourth account of nonconceptual content, according to which an experience has nonconceptual content if and only if the content of the state does not have concepts as constituents. Classic cognitive penetration could be (but clearly need not be) nonconceptual in this sense. One can appreciate now that if one thought that the only form of cognitive penetration was classic cognitive penetration, then one might be tempted to think that cognitive penetration, simpliciter, and nonconceptual content were incompatible. One might set aside the content account—the fourth account—as not as important as the traditional state account of nonconceptual content—the third account. (Or one might, as we saw that someone akin to Bermúdez might do, argue that a good explanation of the incompatibility of classic cognitive penetration with the state account—the third account—is that it is incompatible with the content account.) However, as we will see, once we have the model of cognitive penetration lite in our purview we should no longer be tempted to hold that cognitive penetration and nonconceptual content are incompatible.

 Cognitive Penetration Lite In this section, I articulate another model of cognitive penetration and give reasons to think that such a model is plausible. In the next section, I go on to consider to

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  what extent experience being penetrated in this manner is compatible with its having nonconceptual content. Here is a second model of cognitive penetration: (a) You have a certain propositional attitude that p (e.g. you believe that something red is likely to be found at your present location). (b) p involves a concept R (e.g. the concept of being red), in the sense that possession of R is required to specify the content that p and possession of R is required in order to have the propositional attitude that p. (c) The content of this propositional attitude causally affects your perceptual processes yielding an experience with the content that q (e.g. that there is something red present) that involves R (in the sense that R is required to specify the content that q). (d) The content that q is a content that an experience could have without being affected by propositional attitudes in this way. Call this cognitive penetration ‘lite’. The idea is that there is a form of cognitive penetration which is such that, although it causes an experience with content q to come about and, on that occasion, the experience with content q probably would not have come about unless cognitive penetration had occurred, on other occasions it would be possible to have an experience that represented that q without cognitive penetration occurring.11 Why should we think that such a form of cognitive penetration is possible? One reason comes from thinking about particular examples of cognitive penetration, two of which I will discuss below. A second reason comes from thinking about the mechanism that might explain some cases of cognitive penetration. I will explain each of these in turn. It should be remembered that in the discussion of these two reasons, I am not trying to establish that cognitive penetration exists. I am assuming cognitive penetration exists and considering what forms of it there could be and whether they are compatible with nonconceptual content. One example of an alleged case of cognitive penetration, discussed at length in Macpherson (), occurs in an experiment by Delk and Fillenbaum (). To summarize the experiment briefly: different shapes were cut out of a uniformly coloured orange paper. Some of these shapes were shapes of characteristically red objects, such as hearts and lips, and some were not. These shapes were placed against a uniformly coloured background that could be adjusted in colour from yellow, through 11 I say that the experience with content q probably would not have come about unless cognitive penetration had occurred, not simply that it would not have come about, because there could be cases of pre-emption where, although cognitive penetration is the cause of an experience with a certain content coming about, the cognitive penetration could have pre-empted another event that would have caused the same effect if the cognitive penetration did not occur. I take it that such cases of pre-emption would be rare. In the absence of pre-empted causes, if the cognitive penetration had not occurred then the experience would not have had the content that it did.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

the shade of the orange of the paper, and into red. Subjects were instructed to ask for changes in the colour of the background to be made until it matched the colour of the shape placed in front of it. Subjects chose a background that was redder when the characteristically red objects were placed in front of it than they did when the other objects were placed in front of it. The latter were matched to a more orange colour. (The more orange colour was, objectively, the more accurate match.) Suppose that these are cases of cognitive penetration in which subjects’ beliefs about the characteristic colour of some of the shapes, or some other relevant belief or beliefs, penetrate their experience of those shapes so that they see some of them as redder than others. (See Macpherson  for a detailed discussion of whether this case or similar cases are cases of cognitive penetration.) Suppose then that when looking at the orange heart, subjects’ beliefs that hearts are red penetrate their respective experiences and cause them to have experiences as of a red heart. It seems clearly possible that that type of experience could have been produced by perception alone, such as might occur in a simple, non-cognitively penetrated veridical experience of a heart that is cut out of red paper—an experience as of a red heart.12 In other words, there is reason to think that there could be cognitive penetration lite. A second example is discussed by Tye (), and concerns experiences of the face/vase ambiguous drawing. When one looks at that drawing, one can experience it either as two faces looking at each other or as a vase. One account of what occurs when looking at the drawing is that one either has an experience that represents faces or one has an experience that represents a vase. However, some people think that faces and vases cannot be represented by visual experience. They think that only low-level properties can be represented in visual experience, such as shape, colour, position, object-hood, figure, and ground.13 Such people would be inclined to think that when looking at the drawing, one can have two different experiences, one of which would prompt one to judge that there were faces present, if one possessed the concept of faces, and the other of which would prompt one to judge that there was a vase present, if one possessed the concept of a vase. But they would hold that the difference in what the experiences represented would be in the low-level properties. For example, they might claim that, in one experience, the portion of the picture corresponding to the faces would be represented as figure and the portion corresponding to the vase as ground, and in the other experience the reverse would be true. Tye holds this second view. According to him, what determines which experience you have could involve top-down processing, such as the priming or employment of 12 Someone might be tempted to think that such experiences couldn’t be had, for they might think that a subject’s belief that hearts are red might affect their experience when looking at a red heart yielding a non-veridical experience of the colour of the heart—perhaps an experience of a deeper shade of red than is present. While that could (but need not) be the case for a subject who had the belief that hearts are red, it would not be the case for a subject who lacked the belief that hearts were red. And it is definitely possible that there could be such a subject. 13 A summary of the debate about whether high-level properties can be found in visual experience or whether only low-level properties can be is found in Macpherson ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  either the concept faces or the concept vase. However, he thinks that you could also have each experience purely in virtue of bottom-up processing. In other words, the figure and ground status of parts of the drawing could be assigned purely by the visual system with no top-down processing occurring. Moreover, this could, but need not, happen in the absence of the possession of the concept faces and the concept vase in virtue of different assignments of figure and ground. If you lacked the concepts you would not go on to judge that what you experienced was faces or a vase, but you would be having the same experience that would allow you to so judge were you to possess those concepts. Tye’s account of this case invokes the idea of cognitive penetration lite. A cognitive state involving the employment of certain concepts can penetrate perception to produce a certain experience, but it would be possible to have that experience without cognitive penetration occurring, by purely bottom-up processing. Thus, if there can be cognitive penetration, it seems that there is no reason to think that there could not be cognitive penetration lite. I have not argued here that there are any such cases, but simply that one should think that such cases are possible if you think that cognitive penetration is possible. Besides consideration of particular cases such as these, another reason to think that there can be cognitive penetration lite comes from reflection on a mechanism that might explain some cases of cognitive penetration. As I stated in the introduction, I have postulated such a mechanism in Macpherson (). We have reason to believe that each stage of the mechanism could occur, independently of any considerations of cognitive penetration. The evidence suggesting the existence of the phenomena that constitute each stage of the mechanism, plus the overall ability of the mechanism to implement cognitive penetration, forms a reason to think that cognitive penetration lite could occur. Let me explain. We know that cognitive states, such as beliefs, thoughts, and desires, can cause some states with visual phenomenal character that are not genuine perceptions of the world to come into existence. Such states are commonplace: • perceptual imagery (both intentional and unbidden); • dreams; • hallucinations. Often, but not always, the content of such states reflect one’s beliefs or desires. For example, one might desire to live in a large castle with turrets and crow steps and therefore frequently perceptually imagine such a castle. Alternatively, one might have visited the Alhambra and so have beliefs about what it looks like. In consequence, one might dream about the Alhambra and one’s dream may reflect the beliefs that one holds about how it actually looks. Similarly, people suffering from hallucinations brought on by Parkinson’s disease can hallucinate deceased partners or pets.14 Such 14

See e.g. Barnes and David ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

hallucinations draw on a subject’s knowledge and beliefs about those people and animals and how they looked, and perhaps his or her desire for them to be present. We also know that the phenomenal character of those experiential states can interact with and affect the phenomenal character and content of an experiential state with perceptual phenomenal character that is involved in actually perceiving the world. In fact, in such cases, subjects do not seem to be aware of two states—an imaginative, dream, or hallucinatory one and a perceptual one. So, a more accurate description of what is taking place is that the processes that typically do create perceptual imagery, dreams, or hallucinations interact with the perceptual process to yield one state with phenomenal character. The phenomenal character of the state is determined by the contribution of both processes: imagery, dreaming, or hallucination, on the one hand, and perception on the other. We have reasons to think that such interactions do occur. Consider, for example, the Perky effect (Perky ). Subjects are placed facing a white wall and asked to visually imagine a certain object; unbeknown to them, an image that is above conscious visible threshold for normal subjects is shone onto the wall. In these conditions, subjects subsequently report that they had visual imagery but they deny that they saw anything. However, what they report imagining is influenced by the image that is shone onto the wall. A standard interpretation of what is going on is that subjects mistook their perceptual experience for visual imagery (Thomas ). Another interpretation is that subjects did not consciously perceive the image shone onto the wall because their carrying out the imagery task resulted in the image being below conscious visible threshold for the subject (Segal  and Segal and Fusella , as reported in Thomas ). That is a plausible thought, because some instances of high cognitive load are known to affect which stimuli reach consciousness (see e.g. Carmel et al. ). On this alternative interpretation, it is claimed the subjects unconsciously perceived the image, and that influenced what they went on to visually imagine. It does not matter for my purposes which of these interpretations is right; what is clear is that the resulting experience has aspects that are contributed by perceptual processing and aspects that are contributed by the imagination. While the image projected on the screen affected what was reported, indicating that perceptual processing must have played a part in generating the experience, what was imagined affected what was reported too. Subjects often reported elements to their experience in addition to those present in the image that was projected onto the screen. For example, when only the outline shape of a book was projected onto the screen, a subject reported imagining a book with writing on it; when only the outline shape of a leaf was projected, a subject reported imagining a leaf with veins on it. And in a twist on the experiment (Segal  and Segal and Fusella , cited in Thomas ), an object unrelated to the object that the subject was instructed to visually imagine was projected onto the wall. A tomato was projected onto the wall while subjects were instructed to imagine the New York skyline. Subjects reported imaging the New York skyline at sunset. Therefore, we know that processes that generate visual imagery under voluntary control, hence processes

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  affected by thoughts and beliefs and desires, together with perceptual processing, can yield one phenomenal experience whose content and phenomenal character are affected by and determined by both. Similar interactions are known between processes that underlie dreaming and perception. For example, some experiences have elements that are contributed by both dreaming and perceptual processes, as when real-world stimuli affect what one dreams about. For example, the sound of a loud bang might cause one to dream that an earthquake is occurring. As in the Perky case, it does not interest me here whether it is right to say that the resultant experience, or elements of it, are really dreamed or perceived or both. What is crucial is that what appears to the subject to be one experience has elements that are clearly caused by both dreaming and perceptual processes. Likewise, some experiences are known to be created by both hallucinatory and perceptual processes. For example, many actual hallucinations are partial hallucinations—and thus unlike the philosophers’ favoured example of total hallucinations where everything experienced is hallucinatory. In partial hallucinations, objects are visually hallucinated to be in a scene that is actually perceived by the subject. For example, during Lilliputian hallucinations, very small people are hallucinated and seem to be seen in the environment that the subjects otherwise accurately perceive. For example, small people might be experienced as on the carpet, or peeking out from behind the curtains, or on the subject’s food (Chand and Murthy ). These three examples—the Perky effect, the incorporation of perceptual elements into dreams, and partial hallucinations—show that states with visual phenomenology can be produced that have elements contributed by imagination, dreams, or hallucinations on the one hand and perception on the other. Thus, they show how states with visual phenomenology can be influenced by cognitive states, for the processes that lead to imagination, dreams, and hallucinations can be so influenced. The idea is that cognitive states cause some imaginative, dream, or hallucinatory processes to come into existence. These processes interact with the perceptual process to yield one state with visual phenomenal character, the content and character of which is therefore partly caused by cognitive states. It is easy to see why this mechanism could produce cognitive penetration lite if we think that hallucinations, dreams, and visual imagery can, at least on some occasions, have the same phenomenology as visual perceptual experiences. For in those instances, the addition of such phenomenology to perceptual experiences would produce an experience that could have been had by perception alone. So the resultant experience would be an example of an experience produced by cognitive penetration lite. It seems highly plausible that some hallucinations and dreams have the same phenomenology as some visual perceptual experiences. In the case of visual imagery, the Perky experiments provide some support for the traditional idea that visual imagery is like a form of perception: that had when looking at faint, de-saturated,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

indistinct, or blurry things, or when perceptual conditions are such as to produce faint, de-saturated, indistinct, or blurry experiences of things. At least they do on the first interpretation of the experiments discussed about according to which the faint perceptual experiences were mistaken for visual imagery. To give an example, consider a subject in an experiment like that of Perky’s who is asked to imagine a black pentagon. Unbeknown to the subject, an image of a white hexagon with a black perimeter is projected onto a white wall that the subject is looking at. Plausibly the subject might report having visual imagery as of a grey hexagon (perhaps with a black border). Such an experience could plausibly have been had in pure perception without any cognitive penetration occurring: the straightforward veridical perception of a grey hexagon with a black border. Therefore, it seems highly plausible that some hallucinations and dreams have the same phenomenology as some visual perceptual experiences. In this section, I have outlined the idea of cognitive penetration lite and provided three reasons to think that, on the assumption that there can be cognitive penetration, there is reason to think that cognitive penetration lite could exist. I have not tried to establish that cognitive penetration exists. In the next section, I examine whether cognitive penetration lite is compatible with nonconceptual content.

 Is Cognitive Penetration Lite Compatible with Nonconceptual Content? I will consider, in turn, each of the four accounts of nonconceptual content that I considered in Section , and whether cognitive penetration lite is compatible with each. The first account of nonconceptual content was that of Raftopoulos and Müller, according to which it is a necessary and sufficient condition for a state to have nonconceptual content that is produced by cognitively impenetrable mechanisms. Thus a state produced by cognitively penetrable mechanisms has conceptual content. Clearly, a state produced by cognitive penetration lite cannot have nonconceptual content on this notion of nonconceptual content, and so cognitive penetration lite is not compatible with this notion of nonconceptual content. Note, as I did previously, that this not the standard account of nonconceptual content. I will comment further on their motivation for holding such an account later in this section. Recall the second account of nonconceptual content posited by Bermúdez and Cahen. They hold that the content of experience is nonconceptual to the extent that it is not a function of, or not determined by, the concepts of the subject. It is conceptual otherwise. Consider an instance of cognitive penetration lite that produces a cognitively penetrated experience. On that occasion, the experience is produced by cognitive penetration lite and so the experience produced will be a function of, and determined by, the subject’s concepts. Those concepts will have had a causal role in bringing about

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  the experience with that content and, bar pre-emptive causation, a different experience would have occurred. However, in general, the existence of that type of experience— the type that has the same content and phenomenal character—need not be a function of, or need not be determined by, one’s concepts, for that kind of experience could have been produced without cognitive penetration coming about. Another way of putting the same point is that a token experience may be produced by cognitive penetration lite and so the token may be a function of, or determined by, the subject’s concepts, but this does not entail that the existence of the type of experience—and hence other tokens of that type—are a function of, or determined by, a subject’s concepts. (Again, when I talk of a type of experience here, and subsequently, I mean to refer to types individuated by the fact that they have the same content and phenomenal character.) This raises the question of which reading Bermúdez and Cahen had in mind when stating their definition of conceptual content as occurring when the content is a function of, or determined by, the concepts of the subject. There is an ambiguity lurking in their specification of conceptual content. Did they intend the ‘on that occasion’ reading or did they intend the ‘in general’ reading? I will return to answer this question, but now I will consider whether cognitive penetration is compatible with each of the two disambiguations. On the ‘on that occasion’ reading of their account, the content of an experience produced by cognitive penetration lite will be conceptual. The existence of the token experience is a function of, or is determined by, the subject’s concepts. On the ‘in general’ reading, the content of an experience produced by cognitive penetration lite could be nonconceptual. If cognitive penetration produces a token experience of a type, other tokens of which can occur by pure perception alone, then while possession of concepts played a causal role in the subject coming to have this token experience, there seems no reason to think that, in general, coming to have that type of experience requires possession of concepts. Indeed, on the assumption that pure perceptual experiences do not require concepts in order to have them, then the type of experience produced by cognitive penetration lite will not require possession of concepts in order to have it either. Thus, on the second disambiguation of this view of the distinction between conceptual and nonconceptual content, cognitive penetration lite is compatible with nonconceptual content. The third account of nonconceptual content is the state account, according to which an experience has nonconceptual content if and only if in order to have an experience with a certain content a subject need not possess the concepts required to specify that content. Clearly cognitive penetration lite is compatible with this notion of nonconceptual content. As the type of experience produced by cognitive penetration lite could occur without cognitive penetration occurring, there is no good reason to think that a subject need possess certain cognitive states or possess certain concepts in order to have an experience with that content. On the assumption that there can be pure perceptual experiences that have nonconceptual content, there is no reason to suppose that cognitive penetration lite could not produce tokens of that type of experience.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

The ‘in general’ reading of the account of nonconceptual content given by Bermúdez and Cahen is very similar in spirit to the state account of nonconceptual content. The state account is clearly making reference to types of experience, and the view is that a type is nonconceptual if tokens of that type can exist in a subject without the subject possessing certain concepts, just as the ‘in general’ reading of the Bermúdez and Cahen view is the view that a type of experience is nonconceptual if the content of some tokens of the type of experience need not be a function of, or not determined by, the concepts of a subject. Thus we have so far found two, rather similar, views of nonconceptual content that are compatible with cognitive penetration lite. The fourth account of nonconceptual content is that an experience has nonconceptual content if and only if the content of the state does not have concepts as constituents. Again, as in the case of classic cognitive penetration, the issue of whether experiences have nonconceptual content in this sense is orthogonal to whether they are produced by cognitive penetration lite. There is no reason why experiences produced by cognitive penetration lite should not be nonconceptual in this sense. One could hold that such experiences were not composed of concepts but of objects and properties, or sets of possible worlds. Thus cognitive penetration lite is compatible with the content account of nonconceptual content. To summarize the discussion so far in this section, cognitive penetration lite has been found to be incompatible with nonconceptual content, understood in the manner of the first definition put forward by Raftopoulos and Müller. It is incompatible with it on the ‘on the occasion’ disambiguation of the second account of nonconceptual content put forward by Bermúdez and Cahen. It is compatible with it on the ‘in general’ reading of that account of nonconceptual content. It is also compatible with the third (state) account and fourth (content) account of nonconceptual content. Thus, cognitive penetration lite is compatible with more accounts of nonconceptual content than classic cognitive penetration is. Classic cognitive penetration is only compatible with the fourth (content) account. I now want to examine in more detail the two accounts of nonconceptual content with which cognitive penetration lite was found to be incompatible. As I have twice noted, the first account of nonconceptual content—that proposed by Raftopoulos and Müller—namely, that it is a necessary and sufficient condition for a state to have nonconceptual content that it is produced by cognitively impenetrable mechanisms, is rather different from the other definitions. As I indicated earlier, one might wonder why they propose it. My suggestion is that their account of nonconceptual content is derived from reading the second—Bermúdez and Cahen—definition of nonconceptual content on its ‘on that occasion’ disambiguation. Recall that on that disambiguation a state has nonconceptual content if and only if the content of the state on that occasion—that token state—is a not a function of, or is not determined by, the concepts of the subject. What would it be for a token state to be that way? It would be for the concepts of the creature not to have a causal influence on the bringing about of that state. In other

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  words, it would be for the propositional attitudes, occurrent thoughts and judgments, the possessing of certain concepts, or the priming of certain concepts not to have an influence on the fact that that state came to exist. It would be for the state not to be cognitively penetrated. Thus, this disambiguation of Bermúdez and Cahen’s definition is tantamount to saying that a state has nonconceptual content if and only if it has not been cognitively penetrated, and it has conceptual content if and only if it has been cognitively penetrated. It is not a big step to propose, as Raftopoulos and Müller do, that a state has nonconceptual content if and only if it has been produced by cognitively impenetrable mechanisms, and that a state has conceptual content if and only if it has been produced by cognitively penetrable mechanisms. The precise way that the relevant mechanisms are subsequently individuated will determine whether there are any states that are not in fact penetrated but which count as being produced by cognitively penetrable mechanisms, and hence count as having conceptual content. What is clear, however, is that on this account any state that is cognitively penetrated will have conceptual content. Should we accept either the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account or the Raftopoulos and Müller account of nonconceptual content? Bermúdez and Cahen themselves do not intend the ‘on this occasion’ disambiguation of what they say. They explicitly cite Tye’s example of the face/vase ambiguous figure, in which he claims top-down processing may be, but need not be, involved in the production of one or other experiences of the face/vase ambiguous figure, as being a reason to think that Raftopoulos and Müller’s account of nonconceptual content should be rejected: It is consistent with the notion of nonconceptual content that the representation in question is in fact produced in what Raftopoulos and Müller would call a conceptually mediated way, whereby concepts in the subject’s possession influence the content of a personal level representation . . . That is, it does not seem to be a necessary condition on the nonconceptual content of representations in general that they be insulated from the propositional attitudes. (Bermúdez and Cahen : sect. .)

This reasoning would also apply to rejection of the ‘on this occasion’ reading of their account, thus it would be highly unlikely for them to endorse it. But why should we reject the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account and the Raftopoulos and Müller account of nonconceptual content, rather than be pluralist, thinking that there are simply different accounts of nonconceptual content? If we adopted a pluralist account, then we could simply say that cognitive penetration lite is incompatible with nonconceptual content as it is conceived on the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account and the Raftopoulos and Müller account, but is compatible with the other accounts of nonconceptual content. Only Raftopoulos and Müller hold their account, and I know no one who holds the ‘on this occasion’ disambiguation of Bermúdez and Cahen’s account (or have indeed

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

pointed out this ambiguity in it). Raftopoulos and Müller, then, are alone in recommending this type of account. It is rather different from all the other accounts because, as we have seen, their account of the conceptual/nonconceptual distinction allies it very closely with the issue of whether cognitive penetration occurs. As such, it bears little relation to the concerns that originally motivated the introduction of the notion of nonconceptual content into the literature, namely the alleged properties of experiences I–VI, mentioned above, which the other accounts are motivated to explain. That Raftopoulos and Müller’s account is not concerned with whether experience has these properties is brought out by considering cognitive penetration lite, which provides us with an example of cognitively penetrated states, which according to Raftopoulos and Müller’s account must be conceptual states, but which it is plausible to think still have the properties I–VI, if any perceptual experiences have them. (Similar reasoning applies to the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account.) Why is it plausible to think that states produced by cognitive penetration lite have the properties I-VI, if any perceptual experiences have them? Consider them again: I. II. III. IV. V. VI.

Fineness of grain. Unit-free representation. Analogue representation. Can be had by young children and animals. Explains the acquisition of concepts. Can represent contradictions.

Cognitive penetration lite produces experiences of a type that could have been had by pure perception alone, without cognitive penetration occurring. If one thinks that pure perceptual experiences possess the above properties (or any subset of them), then so too will the other tokens of the same type produced by cognitive penetration lite. Even if one thought that only some pure perceptual experiences possess the above properties, I can see no good reason to think that cognitive penetration lite will produce only experiences that lack those properties. On the contrary, there is every reason to suspect that it will produce experiences with those properties. Another way to show that it is plausible that experiences produced by cognitive penetration lite could have the above properties starts from supposing that some cognitive penetration lite occurs by means of the mechanism that I outlined in Section —namely, by the processes underlying visual imagination, dreaming, or hallucination interacting with the processes of perception to produce one experience with a phenomenal character and content determined by both processes. If experiences are produced in such a fashion, then there is good reason to think that some of them are likely to have properties I–VI. This is because experiences had in visual imagination, dreaming, and hallucination could have the above properties if pure perceptual experiences can have them. As I have argued, there is reason to think that the phenomenology of some of these experiences is the same as some experiences had during pure perception. It is reasonable to think that the output of the result of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  two processes, each of which can alone produce experiences with properties I–VI, could be experiences with those properties also. And even if no experience had in visual imagination, dreaming, or hallucination has the same phenomenology as an experience produced by pure perception, the phenomenology of some of these experiences seems similar enough to that of experiences produced by pure perception that it might be reasonable to attribute enough of properties I–VI to them to warrant thinking that their content is nonconceptual. If so, it is reasonable to think that the output of the result of two processes, each of which can alone produce experiences with nonconceptual content, could be experiences that also have nonconceptual content. So Raftopoulos and Müller’s account and the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account will count experiences produced by cognitive penetration lite as having conceptual content, but it is plausible to think that at least some of these experiences will have the properties I–VI already outlined, if any perceptual experiences do. So, the two accounts will classify as having conceptual content some experiences that have the properties that have motivated people to believe that experiences have nonconceptual content. This gives one some reason to reject these accounts as accounts of nonconceptual content. They don’t track the properties of experience that have motivated people to hold that experiences have nonconceptual content. They just look like accounts of something else: accounts of states that are not cognitively penetrated or not produced by mechanisms that can be cognitively penetrated. If we reject the idea that Raftopoulos and Müller’s account and the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account should be accepted as accounts of nonconceptual content, then cognitive penetration lite is compatible with all accounts of nonconceptual content. If we are pluralists and accept Raftopoulos and Müller’s account and the ‘on that occasion’ disambiguation of the Bermúdez and Cahen account as special types of accounts of nonconceptual content, then, while cognitive penetration lite is not compatible with them, it is with all other accounts. Either way, cognitive penetration lite is compatible with the main and traditional accounts of nonconceptual content.

Conclusion I have distinguished two forms of cognitive penetration: the classic model and cognitive penetration lite. I have not argued that cognitive penetration occurs, but that if it does occur, there is reason to think that cognitive penetration lite could occur. My argument was by means of consideration of two examples—the colour cut-out case and Tye’s face/vase ambiguous figure case—and on consideration of the indirect mechanism involving visual imagery, dreams, or hallucinations that I outlined and showed would explain the existence of some cases of cognitive penetration. I outlined four different accounts of nonconceptual content and showed that classic cognitive penetration was only compatible with one: the content account of noncon-

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

ceptual content. I showed that cognitive penetration lite was straightforwardly compatible with the two traditional accounts of nonconceptual content: the state account and the content account. I then showed that there was an ambiguity in Bermúdez and Cahen’s account. One disambiguation—the ‘in general’ disambiguation—yielded an account very similar in spirit to the state account and an account that is compatible with cognitive penetration lite. The other disambiguation—the ‘on that occasion’ disambiguation—was very similar to Raftopoulos and Müller’s account. These accounts were not compatible with cognitive penetration lite, but I provided some reasons to resist accepting these as accounts of nonconceptual content. In summary, whether cognitive penetration and nonconceptual content are compatible depends on the accounts of each that one is considering. However, I have demonstrated that a good number of forms of cognitive penetration and nonconceptual content are compatible. Thus, if one is motivated to hold that experience is nonconceptual, and one is motivated for the reasons that experience has at least some of the six properties that have traditionally motivated the positing of nonconceptual content, then this should not stop one from holding that cognitive penetration, in particular cognitive penetration lite, can occur. I outlined an indirect mechanism involving the processes underlying visual imagery, dreams, or hallucinations—processes that uncontroversially can be driven by cognitive states—to explain how there could be cases of cognitive penetration lite. This indirect mechanism not only explains how cognitive penetration lite could come about, but also explains its compatibility with the traditional forms of nonconceptual content.15

Acknowledgments This work was supported by two grants from the Arts and Humanities Research Council (grant numbers AH/I/ and AH/L/).

References Barnes, J., and David, A. S. (). Visual hallucinations in Parkinson’s disease: a review and phenomenological survey. Journal of Neurology, Neurosurgery & Psychiatry : –. Bermúdez, J. L. (). The Paradox of Self-Consciousness. Cambridge, Mass.: MIT Press. Bermúdez, J. L. (). What is at stake in the debate about nonconceptual content? Philosophical Perspectives (): –. Bermúdez, J. L., and Cahen, A. (). Nonconceptual mental content. In E. N. Zalta (ed.) The Stanford Encyclopedia of Philosophy: Byrne, A. (). Intentionalism defended. Philosophical Review : –. 15 Enormous thanks are due to Clare Batty and Craig French for reading and providing very helpful comments on earlier drafts of this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

nonconceptual content  Carmel, D., Saker, P., Rees, G., and Lavie, N. (). Perceptual load modulates conscious flicker perception. Journal of Vision (): article no. ; doi: ./.. Chand, P. K., and Murthy, P. (). Understanding a strange phenomenon: Lilliputian hallucinations. German Journal of Psychiatry : –. Crane, T. (). The nonconceptual content of experience. In T. Crane (ed.), The Contents of Experience, –. Cambridge: Cambridge University Press. Cussins, A. (). The connectionist construction of concepts. In M. Boden (ed.), The Philosophy of Artificial Intelligence, –. Oxford: Oxford University Press. Delk, J. L., and Fillenbaum, S. (). Differences in perceived color as a function of characteristic color. American Journal of Psychology (): –. Dretske, F. (). Knowledge and the Flow of Information. Cambridge, Mass.: MIT Press. Evans, G. (). The Varieties of Reference. Oxford: Oxford University Press. Heck, R. G. (). Nonconceptual content and the space of reasons. Philosophical Review : –. Macpherson, F. (). Novel colours and the content of experience. Pacific Philosophical Quarterly (): –. Macpherson, F. (). Colour inversion problems for representationalism. Philosophy and Phenomenological Research (): –. Macpherson, F. (). Ambiguous figures and the content of experience. Noûs (): –. Macpherson, F. (). Impossible figures. In E. B. Goldstein (ed.), The SAGE Encyclopedia of Perception, –. Thousand Oaks, Calif.: Sage. Macpherson, F. (). The admissible contents of experience. In K. Hawley and F. Macpherson (eds), The Admissible Contents of Experience, –. Chichester: Wiley-Blackwell. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. McDowell, J. (). Mind and World. Cambridge, Mass.: Harvard University Press. Peacocke, C. (). Analogue content. Proceedings of the Aristotelian Society : –. Peacocke, C. (). A Study of Concepts. Cambridge, Mass.: MIT Press. Peacocke, C. (). Does perception have a nonconceptual content? Journal of Philosophy : –. Perky, C.W. (). An experimental study of imagination. American Journal of Psychology : –. Priest, G. (). In Contradiction. Dordrecht: Nijhoff. Pylyshyn, Z. W. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences (): –. Raffman, D. (). On the persistence of phenomenology. In T. Metzinger (ed.), Conscious Experience, –. Munich: Academic. Raftopoulos, A. (). Perception and Cognition: How Do Psychology and the Cognitive Sciences Inform Philosophy? Cambridge, Mass.: MIT Press. Raftopoulos, A., and Müller, V. C. (). The phenomenal content of experience. Mind and Language (): –. Segal, S. J. (). Processing of the stimulus in imagery and perception. In S. J. Segal (ed.), Imagery: Current Cognitive Approaches, –. New York: Academic Press. Segal, S. J., and Fusella, V. (). Effects of images in six sense modalities on detection (d’) of visual signal from noise. Psychonomic Science : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



f. macpherson

Siegel, S. (). Do visual experiences have contents? In B. Nanay (ed.), Perceiving the World, –. Oxford: Oxford University Press. Thomas, N. J. T. (). Mental imagery. In E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy: Tye, M. (). Ten Problems of Consciousness: A Representational Theory of the Phenomenal Mind. Cambridge, Mass.: MIT Press. Tye, M. (). Nonconceptual content, richness, and fineness of grain. In T. Gendler and J. Hawthorne (eds), Perceptual Experience, –. Oxford: Oxford University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

PA RT VI

Cognitive Penetrability and Realism

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Perceptual Content, Cognitive Penetrability, and Realism Jonathan Lowe

‘[C]ognitive penetrability of perception’ signifies the effects of cognitive states on the content of perceptual states. . . . Now, conceptual frameworks constitute theories (in a broad sense of the term, not necessarily in a strict scientific sense); thus, ‘cognitive penetrability of perception’, by implicating the conceptual apparatus in perceptual modulation, also signifies ‘theory-ladenness of perception’. (Raftopoulos : xxi)

Ever since Zenon Pylyshyn first coined the somewhat slippery term ‘cognitive penetrability’ over thirty years ago (see Pylyshyn :  and Fodor : ), it has been commonplace for philosophers to argue vigorously either for or against the thesis that human perceptual systems, including the human visual system, are cognitively penetrable. Often, it seems, they do so because they have other and (in their eyes) more important axes to grind—notably, the defence of realism against various forms of antirealism—rather than purely out of a scientific interest in the topic. In line with the remarks by Athanassios Raftopoulos just quoted, I shall take it that, if human perceptual systems are ‘cognitively penetrable’, the implication is that human perception is inextricably ‘theory-laden’. And it is this—at least in Raftopoulos’s view— that means that the cognitive penetrability of perception poses a serious threat to realism. In what follows, I shall examine Raftopoulos’s position on this issue in some detail, focusing first of all on the question of what role, if any, sortal concepts play in the content of visual perception. Later in the chapter, I shall criticize some aspects of Raftopoulos’s own defence of realism and propose an alternative approach.

 The Question of Realism What reason have we, if any, to be realists about the ‘external world’ of commonsense ontology? It is this perennial philosophical question that lies at the heart of Athanassios

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

Raftopoulos’s excellent and thought-provoking book, Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? (Raftopoulos ). Raftopoulos maintains that realism—as opposed to various types of constructivism, conventionalism, and relativism—can be defended only if it can be argued that perception includes some species of nonconceptual content which serves to ‘individuate’ objects in the perceiver’s environment and thereby fix the reference of perceptual demonstratives in a manner which does not rely on the perceiver’s repertoire of concepts (including, above all, her sortal concepts). And he argues at length that just such a species of nonconceptual content can indeed be attributed to perception and exploited in an argument for realism. In this chapter, I shall explain why I think that it may be questioned both whether realism can be defended only in this way and also, more fundamentally, whether it really can be defended in this way. To this end, I shall raise some doubts concerning the adequacy of Raftopoulos’s conception of ‘object individuation’, as this notion is applied to perceptual processes. In particular, there seems to me to be some confusion—or at least a questionable degree of idiosyncrasy—in Raftopoulos’s use of the terms ‘object’ and ‘individuation’, and likewise in his related use of the terms ‘sortal’ and ‘identification’. I shall argue that the perceptual mechanisms appealed to by Raftopoulos cannot in fact serve to single out environmental objects uniquely, and that our perceptual systems must be able to exploit sortal distinctions of a very broad kind in order to achieve this task. But I shall also argue that, even if the perceptual individuation of objects relies in this way upon the deployment of at least some sortal ‘concepts’, this need pose no serious threat to realism, since it is perfectly possible and indeed very plausible to contend that the relevant sortals are ones that are, for evolutionary reasons, both ‘innate’ and well-attuned to real distinctions in nature. Consequently, they need not and should not be seen as ‘floating free’ of such distinctions or as being empirically and metaphysically ‘ungrounded’, in any sense that could give comfort to an antirealist. In addition to all this, I shall argue that Raftopoulos’s response to antirealism is ultimately question-begging and circular, failing to tackle antirealism at its real philosophical roots. This is because Raftopoulos formulates and defends his account of the nonconceptual content of perception in terms which just presume the truth of current empirical scientific theories in neurophysiology and neuropsychology and thus in terms which are already presumptively realist. If one is to argue non-questionbeggingly from nonconceptual perceptual content to realism, then, I believe, one needs to be able to find in perception a species of nonconceptual content which is both introspectively accessible by the perceiver and suited by its intrinsic character to provide good inductive or abductive evidence for the existence of an ‘external world’ conforming in its broad nature and structure to the realist’s conception of such a world. Towards the end of the chapter, I shall attempt to show that such a species of nonconceptual perceptual content can indeed be identified. It will become evident that nonconceptual content of this kind differs markedly from the type of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  nonconceptual content appealed to by Raftopoulos, notably in being nonrepresentational in character. In effect, then, my claim will be that Raftopoulos has not delved deep enough into the structure of nonconceptual content to find a level of such content that is suited to the task of combating the antirealist in terms which do not implicitly beg the question against such an opponent.

 An Outline of Raftopoulos’s Argument for Realism Raftopoulos, as I have just indicated, has as his principal aim the defence of a form of realism, although he makes it clear that it is only a relatively modest or moderate form of realism that he has in mind. It is, nonetheless, one that is clearly opposed to the antirealism implied by the many versions of constructivism, conventionalism, and relativism that have pervaded much thinking in the philosophy of science and metaphysics during the last fifty years or so. Raftopoulos seems to think, perhaps rightly, that one important common root in all of these antirealist positions is the conviction that all perception—and, more specifically, all perceptual content—is thoroughly and inextricably infused with concepts or interpretations or descriptions. One finds this idea, for instance, in Wilfrid Sellars’ () rejection of what he called ‘the myth of the given’, in N. R. Hanson’s () insistence that all observation is ‘theory-laden’, and in Thomas Kuhn’s () talk about scientific ‘paradigms’ and ‘normal science’. Raftopoulos accordingly considers that realism can be defended only if it can be shown that perception includes some kind of nonconceptual content, which serves to ‘individuate’ objects in the perceiver’s environment in advance of the perceiver’s deployment of concepts or descriptions of any type to characterize or classify those objects. This would be a kind of content which suffices to fix the reference of perceptual demonstratives—‘this’ and ‘that’, as used by the perceiver to pick out such objects, whether in speech or merely in thought—in a way which does not rely on the perceiver’s idiosyncratic stock or repertoire of classificatory concepts. For the latter, by implication, are all regarded as being suspect, from an antirealist point of view, on account of their supposed cultural or historical relativity and consequent lack of permanence and universality—a suspicion that Raftopoulos himself seems to share at least in some measure. In this fashion, then, Raftopoulos hopes to marshal evidence against the antirealist in favour of at least a modest species of realism, according to which we human subjects are not radically mistaken about the real nature of the ‘external world’—to the extent, at any rate, that our basic referential apparatus in everyday talk and thought succeeds in latching on to particular things which do genuinely exist as distinct and intersubjectively accessible constituents of mind-independent reality. Before beginning my detailed critique of Raftopoulos’s position, I should like to quote two important passages from his book which encapsulate key features of his approach and main line of argument. The first passage expresses a very

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

general claim—with which I have considerable sympathy—concerning the relationship between perception and our knowledge of the external world: [T]o make any claims whatsoever about the externalia, one must argue from perceptual states to the external world. (Raftopoulos : )

The second passage provides Raftopoulos’s own very helpful and concise summary of his core argument for realism: Here is my argument for a restricted form of realism: (i) To get in contact with the world through vision, one must break the circle of representations. (ii) To break that circle . . . , at least some visual states must have nonconceptual representational content. (iii) There is a part of visual processing . . . that induces in us states with nonconceptual representational content. (iv) The nonconceptual content has features that we consciously perceive as of properties of objects . . . (v) Our interactions with the world on the basis of this content are largely successful. (vi) In order for [this to be the case], a subset of the [nonconceptual content] must present correctly some of the real properties of the external objects. (Raftopoulos : )

Now, I must emphasize that I agree with Raftopoulos that visual perception includes a distinctive kind of nonconceptual content. I disagree with him only about certain aspects of this content and about how, precisely, considerations concerning such content can be marshalled in favour of realism. Here I should reveal, without reservation, that all of my own philosophical sympathies lie entirely on the side of realism. However, with regard to the question of realism, I want to stress my two key points of disagreement with Raftopoulos. The first is that I am not convinced that realism can be defended, at least in some measure, only by appealing to some notion of nonconceptual content. The second is that I do not consider that Raftopoulos focuses on the right kind of nonconceptual content for the purposes of an argument from perceptual content to realism, nor indeed that the kind of argument from such content that Raftopoulos deploys is one that will or should make any impression upon, much less satisfy, his antirealist opponents—who are, I consider, entitled to regard his style of argument as being fundamentally question-begging and circular.

 Object Individuation and Sortal Concepts Before we look more closely at the question of realism, I want to register some fundamental doubts that I have concerning Raftopoulos’s account of the processes of ‘object individuation’ involved in perception. To be blunt, I think that he is somewhat confused—or, at least, confusing—both about ‘objects’ and about ‘individuation’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  Indeed, I think that he is also confused or confusing about what he calls ‘identification’ and ‘sortals’. Most importantly, I do not believe that the perceptual mechanisms that he appeals to in his account of the nonconceptual content of perception can in fact serve to single out environmental objects uniquely and in that sense ‘individuate’ them. However, it seems that by ‘individuation’ Raftopoulos himself doesn’t really mean individuation in this sense: rather, he means something much more like discrimination, as I shall soon try to explain. And this potential source of confusion may serve to obscure some key aspects of his account. Moreover, by ‘identification’ Raftopoulos evidently means something like classification and, more specifically, classification under a sortal concept or description—and this too is a potential source of confusion. For, as we shall shortly see, these uses of the terms ‘individuation’ and ‘identification’ are rather non-standard ones, at least within the philosophical community. The overall picture that Raftopoulos seems to be offering us is something like the following one. Even before our sortal concepts are brought to bear upon the contents of perception, our perceptual mechanisms serve to ‘parse’ the environment around us into discriminably distinct ‘objects’, thus ‘individuating’ those objects for us and enabling us, for instance, to track them as they move. Only when our higher-level cognitive capacities are brought to bear on these perceptual contents do we ‘identify’ any of those objects, that is, classify them, whether explicitly or only implicitly, as being objects of this or that sort or kind: for instance, as being a table, a cat, or a tree. And then Raftopoulos’s point is that, while the latter process of ‘identification’ is conceptually infused or theory-laden and hence subject to antirealist anxieties about the intersubjective variability of different ‘conceptual schemes’, the initial process of object ‘individuation’, being nonconceptual, is not vulnerable to such anxieties and consequently serves to provide an intersubjectively invariant domain of reference for perceptual demonstratives, thereby securing the truth of at least a moderate form of realism concerning the ‘external world’. All human subjects, it may thus be supposed, at least ‘parse’ their perceived environment into the same basic ‘objects’, ensuring a limit to the amount of possible disagreement between them regarding the constituents and structure of ‘external reality’. It won’t be a case, then, of ‘anything goes’, depending on one’s conceptual scheme, regarding the nature of the ‘external world’, rendering that world effectively noumenal, in a Kantian sense. Let me focus now on some key matters of terminology. I prefer to use the term ‘individuation’, at least as it applies to an activity of perceivers or of their cognitive mechanisms, to mean—and in this regard I believe that I am just following current mainstream philosophical practice—the singling out of an object as the particular individual thing that it is: not just a discrimination of its boundaries (see Lowe ). A simple example will show what I mean by this distinction. If we have before us, say, a certain cat, Tibbles, then Tibbles will at this moment be composed of a certain discrete hunk of feline tissue. And, of course, Tibbles and that hunk of feline tissue

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

will share their spatial boundaries. If, say, Tibbles is black and is sitting in front of a white screen, then a normally sighted person will certainly be able to discriminate Tibbles from his surroundings, because Tibbles’s spatial boundary will be sharply delineated by the colour contrast between his blackness and the whiteness of the screen. But to discriminate Tibbles in this sense is not at all the same as individuating him in the standard sense of the term ‘individuate’, since Tibbles and the hunk of feline tissue composing him have the same boundaries and yet are different individual objects. ‘Discrimination’, then, does not mean the same as ‘individuation’ in the standard philosophical sense of the latter term. However, the perceptual mechanisms to which Raftopoulos appeals in his account of nonconceptual content can only secure the discrimination of environmental objects, not their individuation in the standard sense—a fact which is unfortunately obscured by Raftopoulos’s non-standard use of the term ‘individuation’ to mean, in effect, mere discrimination. According to Raftopoulos, ‘When one perceives a scene, the first thing one usually does is parse it in[to] discrete objects, thereby individuating them’ (: , emphasis added). But Raftopoulos’s usage will naturally suggest to many of his philosophical readers that the perceptual mechanisms to which he appeals in his account of nonconceptual content really can serve genuinely to individuate environmental objects, when in fact they can do no such thing. Next, there is the question of the appropriate use of the term ‘object’. Here, I suggest, we should insist that something counts as an ‘object’ only if it has determinate identity—i.e. only if there is a fact of the matter as to which object it is (see Lowe : – and Lowe ). Again, in this I am just following standard philosophical practice, I believe. Both Tibbles and the hunk of feline tissue composing him—call it H—are objects in this sense. But, most importantly, they are clearly different objects, since Tibbles can, in time, come to be composed of a different hunk of feline tissue, owing to metabolic processes that Tibbles necessarily undergoes in virtue of being a living organism and, more specifically, an animal. Now, a subject of perception can individuate—single out uniquely—a particular object only if she can identify it, in the sense of tell which object it is and thereby tell it apart from other objects. This, of course, is not Raftopoulos’s use of the term ‘identify’ since, for him, to ‘identify’ an object is just to classify it as being of this or that sort or kind and hence to apply a certain sortal concept to the object. According to Raftopoulos, the term ‘object identification’, as he uses it, ‘convey[s] the notion of an object . . . as falling under a certain [sortal] description . . . [or being] a member of a category’ (: ). But, again, I find his terminology here confusing and certainly out of line with common practice in philosophical debate, however things may be in other disciplines such as (perhaps) cognitive science or sociology. The upshot, however, is this. Because Raftopoulos means ‘classify sortally’ by ‘identify’ and thinks that such classification operates only subsequently to the ‘individuation’ of objects—by which he really means only their discrimination—he never really gets to grips with the question of how a subject can acquire and exercise an ability to tell which object an object is: that is, to

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  identify objects, in my philosophically more familiar sense of ‘identify’, which links ‘identification’ to an object’s individual identity, rather than to its general sort or kind. Discrimination of an object clearly does not serve to identify it in this sense, since, for example, Tibbles and the hunk of feline tissue H are discriminated together by our nonconceptual perceptual mechanisms, as the latter are characterized by Raftopoulos, because Tibbles and H have the same spatial boundaries. But, given that these mechanisms cannot identify objects, in my sense of ‘identify’, they don’t provide our higher-level classificatory concepts with items that are even suitable to be assigned to this or that sort or kind. They don’t serve to pick out objects with determinate identities, prior to any sortal characterization of them, to which sortal classifications can then be applied. And yet this is what Raftopoulos’s readers, owing to his misleading choice of terminology, may well be tempted to suppose that those mechanisms can do and, indeed, what he himself may well suppose them to be capable of. Moreover, it seems that an account like Raftopoulos’s can never explain what we now need to explain—namely, how a subject can acquire and exercise an ability to tell which object an object is and tell it apart from others—for the following simple reason. Identifying an object—in my sense of being able to tell which object it is— only makes sense relative to an appropriate classification of that object as belonging to a certain broad sort or kind, because the identity criteria of individual objects are clearly sortal-relative. For instance, the criterion of identity governing Tibbles is that of a living organism and differs fundamentally from that governing H, which is that of a hunk of matter. A living organism can, for instance, survive—i.e. preserve its identity through—a change of its material parts, whereas a hunk of matter cannot. Hence, only a subject possessing at least an implicit grasp of this criterial difference can be in a position to identify Tibbles, in my sense of ‘identify’, i.e. tell which object Tibbles is: for this requires the subject at least to be able to tell, for instance, that Tibbles is distinct from H, in the sense of know that Tibbles and H are not one and the same object—even though Tibbles and H presently coincide spatially and so share the same spatial boundaries, rendering them perceptually indiscriminable at the present moment.

 A Problem for Realism? Raftopoulos might well urge at this point that what I have just said concerning object identification renders me vulnerable to the attacks of antirealists, because I am committing myself to the claim that a subject cannot ‘individuate’ perceptible objects in the environment independently of deploying her repertoire of sortal concepts, thus making all perceptual content—or, at least, all perceptual content of a relevant kind— irredeemably ‘theory-laden’ and hence open to cultural and historical variation and instability. But, in the first place, I don’t see why the concession that certain sortal concepts play the role in question, even if it has to be made, renders anyone who makes it automatically vulnerable to antirealist attacks of this sort. Even granted

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

that perceptual demonstratives, on the view I am now proposing, would rely on the deployment of sortal concepts of certain very broad kinds—such as the concepts of ‘living organism’ and ‘hunk of matter’—I see no reason why a realist cannot plausibly claim that the sortal concepts in question, at least in a primitive form, are innate components of human cognitive and perceptual systems, being built into those systems in the course of human cognitive evolution and shaped by evolutionary processes in such a way that they are attuned to real distinctions in nature. Just because they are ‘conceptual’ resources, we needn’t conclude that they are therefore completely arbitrary or mere products of cultural or linguistic history. Indeed, when Raftopoulos himself talks of the ‘objects’ which, according to him, the human perceptual system succeeds in ‘individuating’ nonconceptually, it turns out that what he has in mind are something like so-called ‘Spelke-objects’, which are topologically connected, possess spatial boundaries, are ‘rigid’, and can move continuously as cohesive wholes (see Spelke  and Raftopoulos : –). But these seem to be, to all intents and purposes, very close to what I meant earlier by ‘hunks of matter’—certainly, it is the latter that most nearly satisfy the characterization just given in terms of connectedness, boundedness, rigidity, and the ability to undergo cohesive continuous movement. However, Raftopoulos seems to be labouring under the misconception—which, to be fair, is quite common, especially amongst cognitive scientists and psychologists—that when we use the term ‘object’ to mean something like this we are not deploying a sortal concept, and hence that an ‘object’ in this sense is something that can be singled out prior to any classification of it as belonging to a certain sort or kind. That is simply not so, however, and this is a point whose importance is such that it deserves further elaboration. It would be quite wrong to suppose that we can single out a particular object in the foregoing sense and then only later assign it to any one of indefinitely many different sorts or kinds, such as ‘cat’, or ‘toy’, or ‘ornament’. For an ‘object’ in this sense—a ‘Spelke-object’ or ‘hunk of matter’—is already conceived as something of a certain sort or kind, albeit a very broad kind, governed by a criterion of identity which is in fact incompatible with (for example) the criterion of identity governing any cat. The very fact that ‘objects’ in this sense are said to move implies that there must be some principle constraining their identity both at a time and over time, since movement requires the same object to be in different places at different times—a crucial point to which I shall return. A hunk of matter is not something that can be classified as being a cat, for no hunk of matter can be identical with cat. The reason is simple and we have already pointed it out: a cat, being a living organism, can preserve its identity through changes which cannot be survived by a hunk of matter, and indeed vice versa. And the same reasoning applies to a Spelke-object, if indeed that differs significantly from a hunk of matter: a Spelke-object cannot be—i.e. be identical with—a cat or other living organism. It can at most coincide for a period of time with a cat, sharing its spatial boundaries for a while with the latter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  So ‘object’, in the sense implicitly understood by Raftopoulos, is itself a sortal term, not a term which can find application prior to any sortal classification whatever of the items encountered in perception. Of course, I do not personally recommend this use of the term ‘object’, precisely because it is so misleading. In my own usage, the term ‘object’ is what David Wiggins (; ) calls a ‘dummy’ sortal, carrying with it no commitment to any particular criterion of identity for the items to which it applies. But, for reasons already made plain, one cannot identify or individuate (in my senses of these terms) anything as an ‘object’ in this sense, precisely because it is so indiscriminate in its application. The term ‘object’ as used by Raftopoulos, however, is far from being so indiscriminate. As I say, it seems to be pretty much what is meant by ‘Spelke-object’ and very close to what I mean by ‘hunk of matter’. And neither of the latter terms is unconstrained in its application by considerations to do with identity at and over time—the sort of constraints that are standardly captured by the criterion of identity governing the application of a sortal term. Indeed, a sortal term, in the broadest sense, precisely is a general classificatory term whose use is constrained by a specific criterion of identity governing the items to which it applies. That is how the expression ‘sortal term’, originally coined by John Locke, was recruited into the current philosophical literature by philosophers such as P. F. Strawson () and David Wiggins (), so that to use it in any other way would be seriously misleading. It turns out, then, that Raftopoulos himself is, very probably, actually committed to something like the position that I am proposing, whereby our pre-reflective perceptual mechanisms not only discriminate objects in our environment but also (in my sense) individuate at least some of them, as individual instances or exemplars of a certain broadly defined sort or kind, namely, as something very like individual ‘hunks of matter’ or (if this is anyhow significantly different) ‘Spelke-objects’. It is then a moot point whether these perceptual mechanisms exhibit an application only of this one general sortal classification, with its associated criterion of identity, in their prereflective ‘parsing’ of the perceived environment into discrete objects moving through space and time, or whether, for instance, they also apply a sortal classification roughly corresponding to the concept of a living organism, with its very different associated criterion of identity. Certainly, there is strong empirical evidence that even very young infants are sensitive, in their instinctive perceptual responses, to the distinction between things that are ‘alive’ and things that are not, but that is not enough to show that they have a sensitivity to differences between the identity conditions of living and non-living things—i.e. between the criterion of identity governing living organisms and, for example, the criterion of identity governing hunks of matter, or the criterion of identity governing material artifacts. But let me emphasize once again the crucial point that any theorist who maintains that our pre-reflective perceptual mechanisms can somehow pick out items in our environment and track them as they move must be committed to acknowledging that those mechanisms somehow embody a sensitivity to the synchronic and diachronic identity conditions of the items in question and must thereby be, effectively, classifying those items sortally. For precisely what it means for

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

an item to fall under a certain sortal characterization, however broad, is for it to comply with a certain criterion of identity both at a time and over time. Theorists, including Raftopoulos, who think that items can be ‘picked out’ and ‘tracked’ independently of any sortal characterization of them whatever betray, I believe, a seriously distorted understanding of what ‘sortal’ classification really involves (see further Lowe ). It involves first and foremost a proper sensitivity to an item’s distinctive identity conditions, as captured by the criterion of identity governing it. It is quite false, however, to think that a ‘sortal’ term in this sense has to be something as nearly as specific as, for example, ‘cat’, or ‘table’, or ‘tree’. It can be quite as general as ‘object’, in the somewhat misleading sense in which this term is used by Raftopoulos, for even in this sense it is necessarily associated with the satisfaction of certain identity conditions both at a time and over time, given that it is being used to denote items capable of moving through space and time.

 How May We Argue for Realism? I said earlier that, in my view, Raftopoulos’s strategy for defending realism isn’t the only available one, nor even an entirely satisfactory one. As I have just indicated, I believe that one can defend realism by appealing to the innateness of certain ‘sortal’ concepts, at least as successfully as one can by deploying Raftopoulos’s strategy—which isn’t, of course, to imply that either strategy is fully satisfactory. At the same time, I do think that Raftopoulos is right to contend that the only way to defend realism compellingly against the sorts of antirealists that he has in mind is to argue from perceptual content to realism, not least because their scepticism arises from doubts about the reliability of perception as a guide to mind-independent truth. I am also happy to acknowledge that it is only the nonconceptual content of perception that should ultimately be appealed to for this purpose—although not because I share the antirealists’ scepticism about the universality and stability of human concepts quite generally. However, my immediate objection to Raftopoulos’s own attempt to implement an argument for realism of this type is—as mentioned earlier—that his version is at bottom just question-begging and circular. This is because he openly and deliberately appeals to scientific evidence drawn from empirical work by neurophysiologists and neuropsychologists in support of his claims that perception has the type of nonconceptual content that he attributes to it. However, this is just to assume the truth of the scientific theories that such evidence relies on: it is to assume, for example, that brains exist, that they are composed largely of neurons, and that our perception is a product of brain processes involving the interaction of neurons stimulated by physical events—events such as the emission of photons—occurring in our external environment and affecting our sensory organs, such as our retinal surfaces. But these are all blatantly realist assumptions of a quite specific kind, all of which are thoroughly ‘theory-laden’ and so subject to the antirealists’ sceptical concerns.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  To argue at all compellingly against the antirealist, we need to be able to appeal to features of the nonconceptual content of perception which can be identified wholly introspectively, without any appeal to antecedently assumed empirical scientific theories. Now, is the kind of nonconceptual content described and appealed to by Raftopoulos of a suitable type for this dialectical purpose? I think not. For even if it could be contended—as Raftopoulos himself does not contend—that it is purely introspectively ascertainable that perception has nonconceptual content of this kind, this is supposed to be content which is still representational, even though not conceptual, in character. It is supposed, more specifically, to be content which represents the physical environment of the subject as being a certain way: namely, as being articulated into three-dimensional objects situated outside the subject and capable of undergoing movement with respect to the subject. But any thoroughgoing antirealist—such as (to take a classic historical example) the th-century idealist philosopher George Berkeley ()—will very likely urge that the mere fact that our perceptual states represent to us the existence of such an external, three-dimensional environment in which we ourselves are located gives us no good reason to suppose that such an environment really does exist. In short, we have yet to be persuaded of the veridicality of any such representational content, notwithstanding its supposedly ‘nonconceptual’ character. Why—the antirealist may ask—couldn’t extramental reality (if indeed there is any) be utterly unlike the ‘external world’ as the nonconceptual content of our perception represents it to be? Why shouldn’t it be the case that extramental reality—God, perhaps—equips us with this kind of representation not because it is in any way veridical, but just because it is somehow conducive to our wellbeing or our survival for us to take extramental reality to be like this? In this context, then, the crucial step (v) in Raftopoulos’s own summary of his argument, cited earlier in Section , has very little dialectical force against his antirealist opponents. Step (v), recall, was this: ‘Our interactions with the world on the basis of this [nonconceptual representational] content are largely successful’ (Raftopoulos : ). But, as I have just indicated, a sufficiently sophisticated antirealist, such as Berkeley, can readily contend that ‘success’ in our projects is perfectly consistent with the radical nonveridicality of the representational contents of our perceptions because, say, the real source of our perceptions is a benevolent deity who has our wellbeing in mind in conferring upon our perceptions contents of precisely this ‘illusory’ kind.

 An Alternative Approach Fortunately for realism, there is, I believe, a solution to the foregoing difficulty, but it is one which involves delving much deeper into the introspectively accessible nonconceptual content of perception, to a level not recognized by Raftopoulos—nor, to be fair, by most present-day philosophers of perception, given their hostility to anything that is suggestive of ‘sense data’. I can only very roughly sketch this solution

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

here, given the space available. And, for ease of exposition, I shall focus exclusively on the case of visual perception, as indeed Raftopoulos himself largely does. There is, I believe, a type of nonconceptual visual content that is, mathematically speaking, topologically two-dimensional. That is to say, the visual field of a normally sighted human subject is, at any given moment, typically segmented into regions, often made salient by phenomenal colour differences, which individually and collectively exhibit a two-dimensional geometry—a claim which, ironically enough, Berkeley himself vigorously defended. For simplicity, I shall concentrate on the case of monocular vision, since the stereoscopic effects of binocular vision, while not affecting the topological structure of the visual field, in the sense of rendering it anything other than two-dimensional, are of course significant for depth perception and certainly give the perceiver an impression of three-dimensionality, even though the visual field itself is not extended in three dimensions. I shall deal fairly briefly with this topological point here, since I have discussed it much more fully before (see Lowe ). The following is a simple theorem in the topology of two-dimensional surfaces. Given any two separate points, A and B, in such a surface, it is always possible to draw a closed line or loop, L, around A—this loop being confined to the surface in question—such that any line (likewise confined to the surface) connecting A to B must intersect (i.e. cross) the loop L. Figure . illustrates this fact. Now, to see how this theorem is exemplified by the visual field of a normally sighted human subject, consider any two ‘points’ in that field: for instance, suppose that one is looking at a distant hillside on which two houses appear as two small coloured ‘dots’, A and B, separated from one another. Then it will always be possible to take a thin piece of coloured wire bent into the shape of a loop, L, and position it so that it appears to encircle A, in such a fashion that any other thin piece of coloured wire, positioned so as to appear to connect A to B, must appear to intersect or cross the loop L. Since the word ‘appear’ is multiply ambiguous, it is vitally important, of course, that we should understand its intended sense of ‘appear’, as I use it in the foregoing example. This is sometimes called the ‘phenomenal’ sense of ‘appear’. The sense in question is, I

B

A

Figure . Topology of two-dimensional surfaces

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  think, best brought out by means of further examples. A good example for this purpose is the following one, which draws on the phenomenon of ‘size constancy’ in visual perception. If one holds one’s two thumbs together in front of one, and then gradually moves one of them further away, all the time fixing one’s gaze on the two thumbs, it is manifest that there is a perfectly good sense in which the more distant thumb does not appear to get any ‘smaller’ as one moves it progressively further away. This is the sense of ‘appear’—sometimes called the epistemic sense—which is intimately related to the fact that one will not be at all inclined to judge, on the basis of what one sees, that the more distant thumb is in fact any smaller than the nearer thumb. However, if one repeats this exercise while directing one’s gaze not at the two thumbs but rather at a point a few degrees of arc above or below them, it is manifest that there is another perfectly good sense in which the more distant thumb does indeed appear to get ‘smaller’ as one moves it further away. The ‘size constancy’ phenomenon that manifests itself in the first version of the procedure entirely vanishes in the second version, and indeed in a very striking fashion. Evidently, the more distant thumb ‘appears smaller’, in this second sense, than the nearer thumb precisely in virtue of the fact that its visual extension occupies fewer degrees of arc in one’s visual field. This fact obtains, of course, in both versions of the task, but it is manifestly obvious to the subject only in the second version and takes considerable pains to detect in the first, although this can be achieved with practice and is something that a trained artist is more likely to succeed in doing than someone not used to judging differences in visual extension. (Actually, artists often estimate such differences precisely by holding their thumb up against distant objects to compare their relative extensions.) Now, the sense in which the more distant thumb appears smaller than the nearer thumb in the second version of the procedure is the same sense of ‘appear’ that I am applying in my description of the experiment with the two distant houses and the two pieces of wire. In the latter experiment, there is also, of course, a perfectly good sense in which the two wires may not ‘appear’ to intersect or cross one another, since one of them may ‘appear’, in this sense, to be in front of the other. Similarly, in this sense of ‘appear’, the second piece of wire may not ‘appear’ to connect the two houses, since it may not ‘appear’, in this sense, to touch either of them—nor, in this sense of ‘appear’, may the two houses ‘appear’ to be small coloured ‘dots’. This is, once again, the so-called epistemic sense of ‘appear’. Setting terminology aside, however, I simply want to point out that we are dealing, in such examples, with two distinct senses of ‘appear’—the sense in which, for instance, the two thumbs do ‘appear’ to remain the same in size and the sense in which they do not—and that it is solely the second sense that I am appealing to in my argument concerning the topology of the visual field. This is a sense of ‘appear’ which I am, somewhat reluctantly, prepared to go on calling the ‘phenomenal’ sense, now that we have managed to pin it down. That epithet is, however, potentially very misleading. After all, the size-constancy effect, which is present in the first version of the exercise with the thumbs, is certainly a visual ‘phenomenon’, in the sense that a subject may be perceptually conscious or aware of

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

it, and so it may sound odd to say that in this version the two thumbs appear only ‘epistemically’ and not also ‘phenomenally’ to have the same size. Anyway, terminology aside, the thesis is that visual appearances, in the relevant sense of ‘appearance’—whatever we choose to call it—are two-dimensional. And the hillside example with the two houses seems to confirm this thesis quite compellingly. In particular, I would expect any philosophically untutored (and so philosophically unbiased) subject to be perfectly willing to say, when experiencing that example, that the two wires ‘appear’ to cross one another: and that is the crucial fact about the example for my purposes. Plainly, in that example, there does not ‘appear’ to be a visual gap between the two wires, in the way in which there does ‘appear’ to be such a gap between, say, the opposite sides of the wire loop or indeed between the two houses. The fact that there may be a real gap between the two wires, in the direction of the subject’s line of sight, and that the subject may know this to be the case, in no way undermines the fact that there does not ‘appear’ to be a visual gap between them. Such a real gap in the line of sight cannot be experienced as having any visual extension in the visual field, and that, fundamentally, is why the visual field is topologically only two-dimensional. This is something that Berkeley was perhaps the first philosopher to recognize fully, for which he deserves much credit rather than the ridicule that is often heaped upon him by many modern philosophers of perception, especially those with a mission to defend some kind of naïve realism. And here I should perhaps emphasize that, in strict topological terms, a spatial manifold can only possess an integral number of dimensions, so that it would be wrong to suppose that the visual field could somehow be more than merely two-dimensional without being fully three-dimensional. In that connection, we should not be misled by David Marr’s well-known notion of the ‘/D sketch’ (Marr ), whatever its merits in his theory of visual information-processing. No ‘sketch’ could literally have a dimensionality of /, in the strict topological sense. What I have been seeking to establish is that the visual field of a normally sighted human subject is, at any given moment, typically segmented into colourdifferentiated regions, with these regions individually and collectively exhibiting a two-dimensional geometry, and that this constitutes a basic nonconceptual—and indeed nonrepresentational—feature of the content of human visual experience. However, for the purposes of an argument for realism, it is important to consider the visual field not merely as it is at a moment, but rather as it evolves over continuous periods of time. For what we then find is that the colour-differentiated two-dimensional visual shapes into which the visual field is segmented typically change in such a way that they may normally be described mathematically as the projections, upon a twodimensional surface, of rigid three-dimensional objects moving relative to that surface in a continuous fashion. Another simple visual experiment illustrates this important fact about the typical evolution over time of the human visual field, as follows. Take a thin piece of coloured wire bent into the shape of an equilateral Euclidean triangle and view this, to begin with, face-on in a small central region of one’s visual field, by holding it at a suitable distance away from one’s eye. Next, gradually bring the

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  wire triangle progressively closer to the centre of one’s eye. It will then be observed that the phenomenal character of one’s visual field changes in the following way: a progressively larger triangular region of that field, outlined by the phenomenal colour of the wire, will be presented to view, in such a fashion that the sum of the internal angles of this triangular region progressively increases beyond  degrees to a theoretical maximum of almost  degrees, attainable only when one’s eye is actually inside the wire triangle. Moreover, if the wire triangle is gradually rotated about one of its sides at any time during the foregoing process, so that it is no longer viewed face-on, the triangular region of one’s visual field that one experiences during this process will exhibit a progressively larger disparity between one of its internal angles and the other two: the angle opposite the side in question will increase in size, eventually becoming obtuse, while the other two angles will become increasingly acute. There is, of course, a very obvious physiological explanation for the visual phenomena described in this experiment. This is that each region of the human visual field corresponds to a geometrically similar region of the subject’s retina, which is an approximately hemispherical surface and thus one that possesses an approximately spherical two-dimensional geometry. Thus, in the case of the wire triangle, as it is brought progressively closer to the eye, a progressively larger spherically triangular region of the retina is stimulated by light rays from the wire incident upon the retinal surface. Now, obviously, this physiological explanation is not something that we can appeal to in a debate with a thoroughgoing antirealist, for whom the truth of the physiological claims involved in that explanation are still entirely open to question. Nonetheless, the phenomenological fact itself, for which this physiological explanation has been offered, is unassailable, since it is ascertainable by introspection alone in the manner just described. And bear in mind here that while the instructions for performing the experiment are couched in physical terms, the phenomenal character of the visual experience thus generated does not call upon any such terms for its proper description. And this is why phenomenological facts of this kind are suitable for recruitment in an argument in defence of realism, since appealing to them begs no relevant question against the antirealist. We are now in position to propose, as a case of abductive argument or ‘inference to the best explanation’, that it is reasonable to infer, as a cause of these pervasive geometrical characteristics of the human visual field as it typically evolves over time, the actual existence of real three-dimensional rigid objects occupying a real Euclidean space and indeed a real ‘projective’ mechanism, in the form of something akin to light rays, incident upon a real hemispherical sensory surface, akin to the retina— in short, an ‘external’ world possessing the basic structure and form that realism traditionally assigns to it and in which we as perceiving subjects are located and embodied. Of course, this is not at all to imply that human subjects can only acquire a belief in such an external world by inference from the geometrical characteristics of their visual fields as they evolve over time—whether by some conscious process of inference or by some unconscious or ‘sub-personal’ one executed by the visual

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



j. lowe

systems in their brains. We are not concerned here with the acquisition of belief in an external world, only with its justification. Obviously, too, there are in principle alternative possible ‘explanations’ for the relevant geometrical characteristics of the visual field other than the ‘realist’ one which posits a real three-dimensional Euclidean space occupied by objects ‘external’ to the subject. For instance—taking a lead from Descartes or Berkeley—it is possible to suppose that God or an ‘evil demon’ has just decided, for good or ill, to give us visual experiences which present us with a visual field possessing precisely the projective properties that have just been described, thus making it seem to us as if we inhabit a real three-dimensional Euclidean space. But it seems evident that any such alternative ‘explanation’, while conceivable, is really just parasitic upon the preferred realist explanation of the relevant phenomenological facts. The alternative hypotheses are all ones which claim that the ‘true’ explanation mimics the effects of the realist explanation, without providing any detailed account of how or why these effects are supposed to be produced. And that is why the realist explanation is indeed the best—because the simplest—explanation, even if it is not the only possible explanation. The phenomenological facts in question appear, then, to provide us with good but not conclusive reasons to be realists—and this, really, is the best that one could ever hope to achieve in a debate with any confirmed antirealist. Notice, crucially, that this is not a response to the antirealist that could compellingly be advanced if one were restricted to appealing, as Raftopoulos is, to the representational nonconceptual content of perception. For attributing that kind of content to visual experience merely registers the fact, already acknowledged by antirealists such as Berkeley, that it seems to us that we perceive an ‘external’ three-dimensional world. The line of argument that I am proposing explains that seeming, in a detailed fashion, by appeal to specific mathematical relationships holding between the nonrepresentational nonconceptual content of visual perception and the geometrical structure of the ‘external’ world as hypothesized by the realist. This means that the realist’s hypothesis has a genuine and substantive explanatory role to play in accounting for the structure of our perceptual experience, and can be shown to fulfil this role without begging any question against the antirealist.

 Concluding Remarks It turns out that we can hope to formulate an argument from perceptual content to realism which does not beg any question against the antirealist by assuming the truth of any theory couched in terms of objects and properties supposedly belonging to an extramental ‘external world’. But in order to construct such an argument, we need to be able to identify introspectively certain aspects of the nonconceptual content of perception which are not only nonconceptual but also nonrepresentational in character, unlike the kind of nonconceptual content appealed to by Raftopoulos, which is not only representational but also, supposedly, not attributable to perception solely on the basis of introspection and without the support of empirical scientific

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

content, penetrability, and realism  theory. I have tried to show that introspectively accessible nonconceptual content of the required sort can in principle be found, in the form of the characteristic projective geometrical features of the human visual field as it evolves over time. These features appear to provide good evidence for the truth of a moderate form of realism, according to which we really do inhabit an ‘extramental’ three-dimensional space occupied by persisting objects possessing relatively stable spatial configurations—in effect, something like ‘hunks of matter’ or ‘Spelke-objects’: a space that is Euclidean in form, at least at the scale at which our sensory engagement with it takes place. As it stands, my proposed argument for realism is, of course, just a rough preliminary sketch, which needs to be filled out much more fully to be made entirely convincing. But it at least indicates the kind of task that needs to be undertaken, I believe, if one is to engage the antirealist on terms which are not implicitly question-begging and circular, in any attempt to argue from perceptual content to realism.

References Berkeley, G. (). Philosophical Works, ed. M. R. Ayers. London: Dent. Fodor, J. A. (). The Modularity of Mind. Cambridge, Mass.: MIT Press. Hanson, N. R. (). Patterns of Discovery. Cambridge: Cambridge University Press. Kuhn, T. S. (). The Structure of Scientific Revolutions. Chicago: University of Chicago Press. Lowe, E. J. (). The topology of visual appearance. Erkenntnis : –. Lowe, E. J. (). The Possibility of Metaphysics: Substance, Identity, and Time. Oxford: Clarendon Press. Lowe, E. J. (). Sortals and the individuation of objects. Mind and Language : –. Lowe, E. J. (). Individuation, reference, and sortal terms. In A. Raftopoulos and P. Machamer (eds), Perception, Realism, and the Problem of Reference, –. Cambridge: Cambridge University Press. Marr, D. (). Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. New York: Freeman. Pylyshyn, Z. W. (). Computation and Cognition: Toward a Foundation for Cognitive Science. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press. Sellars, W. (). Science, Perception and Reality. London: Routledge and Kegan Paul. Spelke, E. S. (). Principles of object perception. Cognitive Science : –. Strawson, P. F. (). Individuals: An Essay in Descriptive Metaphysics. London: Methuen. Wiggins, D. (). Sameness and Substance. Oxford: Blackwell. Wiggins, D. (). Sameness and Substance Renewed. Cambridge: Cambridge University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

 Cognitive (Im)Penetrability of Vision Restricting Vision versus Restricting Cognition Costas Pagondiotis

Introduction The debate on the cognitive penetrability of vision concerns the possible dependence of vision on cognition. A proper evaluation of this debate requires a specification of the two relata involved in the dependence relation and of the kind of relation itself, if there is one. As we will see, there are different suggestions as to which the relata are and what exactly the dependence of the one on the other is. I argue that the terms in which the debate was framed lead to a problematic dilemma that may be roughly formulated as follows: either vision is not dependent on cognition, namely, it is cognitively impenetrable, and consequently vision is alienated and disconnected from thought, or vision is dependent on cognition, i.e. it is cognitively penetrated, and as a result the distinction between vision and thought is blurred. In this chapter I examine this dilemma and suggest a possible way out. The chapter consists of two main parts, Sections  and  and Section . The first part examines Pylyshyn’s approach to cognitive penetrability and his suggestion that only the initial stage of visual processes, called ‘early vision’, is cognitively impenetrable, whereas ‘late vision’ is cognitively penetrable. This suggestion focuses mainly at the sub-individual level and restricts the dependence of vision on cognition, by restricting the scope of the first relatum to late vision. Moreover, it takes the second relatum, cognition, for granted and makes no distinction between different kinds of penetrating cognition. The suggestion implies that late vision is penetrable by the same kind of cognition by which states of the central systems are penetrable. I will argue that this approach disconnects early vision content from the world and from the cognitively penetrated visual content available to the perceiver, on the one hand, and, on the other, it blurs the distinction between the cognitively penetrated visual content and the content of thought.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



For this reason, in the second part of the chapter I suggest an alternative approach based on the opposite move. I formulate this suggestion at the level of visual experience. The first relatum, visual experience, is taken for granted, while a restriction is introduced on the kind of cognition that can directly penetrate visual experience. In particular, the suggestion is that visual experience, unlike thought, is directly penetrated only by practical non-propositional knowledge. This implies that the cognitive penetrability of vision should be distinguished from the cognitive penetrability of thought, since vision is cognitively penetrated only by practical nonpropositional knowledge. This suggestion, if tenable, opens up space for a middle position between the two alternatives that create the dilemma since (a) differentiating the cognitive penetrability of visual experience from the cognitive penetrability of the content of thought preserves the distinction between vision and thought, and (b) taking visual content as penetrated by practical nonpropositional knowledge makes it an objectdependent content available to the embodied perceiver. In order for this position to be appropriate for overcoming the dilemma, the way vision is distinguished from thought must account for the distinctive content of visual experience. I will argue that this position succeeds in accounting for the distinctive content of visual experience by analysing the kind of dependence visual content has on practical non-propositional knowledge. To this effect, I initially examine the kind of dependence between the two relata that Pylyshyn’s approach introduces. Pylyshyn conceives the dependence as causal and, more importantly, as semantic-rational. He further specifies this semantic-rational relation in terms of a particular kind of inference, one which is related to the inferential promiscuity of beliefs. This amounts to a strong notion of inference, which presupposes that the inferentially connected representations have contents that (i) are propositional (given that they can be contents of beliefs), and (ii) are not domain-specific. If one adheres to this strong version of cognitive penetrability, then a system with representations that depend rationally on certain representations of another system without satisfying (i) and/or (ii) has to be taken as cognitively impenetrable, even though it is not informationally encapsulated. Thus, between the informationally encapsulated systems and the systems that exhibit strong cognitive penetrability, a variety of systems can be specified that are neither informationally encapsulated nor strongly cognitively penetrable. I call this kind of cognitive dependence ‘weak cognitive penetrability’, and argue that it should be further specified in anti-representational terms, as the dependence of visual content on practical nonpropositional knowledge. This is a more radical option on cognitive penetrability found in the work of a number of philosophers and scientists who reject the framework of the Representational and Computational Theory of Mind (RCTM) and support anti-representationalist approaches to perception and cognition. In these approaches, the processes subserving perceptual and cognitive capacities are not taken as consisting in computational transformations of symbolic representations. This precludes an understanding of cognitive penetrability in terms of a (promiscuous or not) inferential dependence of symbolic

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

representations of the perceptual systems on symbolic representations of the cognitive systems. Thus, Pylyshyn’s semantic criterion of cognitive penetrability is not applicable to the anti-representational approaches. For this reason, I attempt to develop an account of weak cognitive penetrability that is not based on the inferential dependence of symbolic representations of the perceptual systems on symbolic representations of higher-level systems. In particular, I examine different kinds of constitutive dependence of perceptual content on practical nonpropositional knowledge, and suggest that this dependence should be conceived as a rational non-inferential relation, for it is in this way that we can account for the distinctive content of visual experience. I focus mainly on a particular kind of practical nonpropositional knowledge, sensorimotor knowledge, and I explain further the rational non-inferential dependence of visual content on sensorimotor knowledge in terms of the structure of the act of perceiving. I suggest that, according to the weak cognitive penetrability I am proposing, what makes available to the perceiver the sensorimotor knowledge on which visual content depends is the implicit and non-receptive bodily self-consciousness constitutively involved in the act of perceiving.

 The Traditional Framework for the Cognitive Impenetrability of Perception Pylyshyn () coined the term ‘cognitive penetrability’ in the context of the traditional RCTM. He conceived cognitive penetrability as a semantic-logical dependence of the content of symbolic representations processed in a system on the content of representations, such as goals and beliefs, processed in cognitive systems. Pylyshyn (a: ), for instance, explains the notion as follows: If a system is cognitively penetrable then the function it computes is sensitive, in a semantically coherent way, to the organism’s goals and beliefs, that is, it can be altered in a way that bears some logical relation to what the person knows.

This is supposed to be a rational and not just a causal dependence of a system on the cognitive systems. In such a case, representations outputted by the cognitive systems serve as internal input (internal feedback) to the system along with possible exteroceptive input. Pylyshyn specified further this semantic-rational relation in terms of inference: if a system is cognitively penetrable by the cognitive systems, the content of representations processed in this system depends inferentially on the content of representations outputted by the cognitive systems. The term ‘inference’ is used here in a specific way, as Pylyshyn (: ) acknowledges: I prefer to reserve the term ‘inference’ for a process that Stich () calls ‘inferentially promiscuous’—i.e., that is not restricted in a principled way to what information it can use.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



Pylyshyn makes this qualification because, according to the RCTM, the processes that take place within non-cognitive systems, like the various perceptual systems, are explained in computational terms, namely as syntactic transformations of symbolic representations. Thus, many researchers consider these processes as inferential when they describe them at the semantic-intentional level. For example, Marr (: ) states: ‘the true heart of visual perception is the inference from the structure of an image about the structure of the real world outside.’ Although both the states of the visual system and the states of the cognitive systems are involved in relations that can be characterized as inferential in the broad sense of the term, only the latter are inferentially promiscuous. The paradigmatic states of the cognitive systems are beliefs. Specifying inferential promiscuity, Stich (: ) notes: ‘Provided with a suitable set of supplementary beliefs, almost any belief can play a role in the inference to any other.’ This inferential integration of beliefs is an expression of their holism; and since any belief is possible to play a role in the inference of any other, the inferential relations among them are not domain-specific. Thus, although the visual system has states that are inferentially related to one another, it can be cognitively impenetrable if it is inferentially insulated from the states of the cognitive systems. At this level of generality, the cognitive impenetrability of a computational system does not seem to imply that the content of its states and their inferential relations should be of a different type from the content of beliefs and their inferential relations. The only restriction that cognitive impenetrability seems to set concerns the source of the contents used. As Pylyshyn (b: ) notes: Describing the computations carried out by early vision in Bayesian terms is perfectly reasonable and has indeed led to some useful insights. The crucial point that is relevant to the present thesis is that the Bayesian formulations, though they can be viewed as probabilistic inferences, are cast over optico-spatial properties and do not take into account such things as the probability that what one sees is an X, given that one knows from general considerations that Xs are likely to be at this place at this time.

But, perhaps, the way Pylyshyn defines the cognitive impenetrability of a state sets an additional restriction to its content, namely that it must be unconscious. Fodor () examines the relation between cognitive impenetrability (or, in his terminology, informational encapsulation) and unconsciousness but he finds it extrinsic, since there are unconscious states that are not encapsulated. He gives as an example of unencapsulated unconscious state one’s subdoxastic acquiescence in the rule of modus ponens: Sub-doxastic knowledge of such principles must be accessible to practically all mental processes, since practically all inferential processes exploit them in one way or another. One’s subdoxastic beliefs about validity and confirmation are thus quite unlike one’s subdoxastic beliefs about the rules of grammar; though both are unconscious, the former are paradigms of promiscuous and unencapsulated mental states. So the connection between unconsciousness and encapsulation cannot be intrinsic.(Fodor : )

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

Pylyshyn’s and Fodor’s approach to cognitive impenetrability concerns primarily the sub-individual level. For example, their discussion of the cognitive impenetrability of visual system deals with the processes and states in this system from the moment input affects the retina up to the point of the appearance of visual experience. Only the final output of these processes, visual experience, is conscious and has a phenomenal character. Thus, the problem of cognitive impenetrability, as was initially formulated by Pylyshyn and Fodor, concerns primarily sub-individual processes and states. Current discussion focuses also on the individual level and examines the cognitive impenetrability of perceptual experience.1 But if one accepts the framework of RCTM, then the cognitive penetrability of the sub-individual states processed in the perceptual systems implies the cognitive penetrability of, at least, a part of perceptual experience. In the first part of the chapter (up to the end of Section ), I will focus mainly on the discussion about the cognitive impenetrability of sub-individual processes and states, and I will presuppose Pylyshyn’s semantic criterion of cognitive penetrability. To further clarify this semantic criterion, it is helpful to contrast the possible semantic-rational dependence of perception on cognition with an ‘external’ way cognition may influence perception, namely through an action affecting the input to the perceptual system. For example, when I intend to see what is on my left, I turn my eyes to the left and, as a result, my perceptual experience changes. But this change in the perceptual content depends directly only on external input, namely the part of reality that happens to be in view. The movement of the body and of the eyes serve just as a means to change external input and, through it, perceptual content. This is what Raftopoulos (: ), interpreting Pylyshyn’s approach, calls ‘indirect cognitive penetrability’.2 Indirect cognitive penetrability is not cognitive penetrability proper, since the content of thought and of the ensuing action does not bear any semantic-rational relation to the content of perception. This is illustrated by the fact that if, in the absence of the relevant motivating thought and of the consequent action, the perceiver had been moved by someone else so that his eyes focused on the same location in the same lighting conditions, then his perceptual experience would not have been different. Thus, the influence of thought on perception must be internal to count as a case of direct cognitive penetrability. Still, not every internal influence can be a semantic-rational influence. For example, it might be the case that a top-down signal internally modulates the functioning of a perceptual system without affecting the content of the representations processed. This would still be a case of indirect cognitive penetrability, which could be called ‘internal’ 1

See e.g. Macpherson () and Siegel (). In one respect, the notion of instrumental dependence of perception on the output of action that Susan Hurley (: ) introduces is similar to the notion of indirect cognitive penetrability: ‘[t]he dependence of the contents of perception on output is instrumental when it operates via input rather than directly. Output plays the role of means to input, on which perception depends directly.’ It is important to note, however, that Hurley works in a radically anti-representational framework. 2

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



to distinguish it from the former kind, the ‘external’ indirect cognitive penetrability. In the next section we will see a particular example of internal indirect cognitive penetrability.3

 Restricting Cognitively Impenetrable Perception to Early Vision Unlike Fodor’s initial position presented in his seminal book on the modularity of mind (Fodor ), the main contemporary proponents of the cognitive impenetrability of visual processes restrict their claim to early vision.4 Only early vision is considered to be cognitively impenetrable, whereas late vision processes, unlike what Fodor used to hold, are conceded to be cognitively penetrable. But what is the output of early vision? This is an open question, as Pylyshyn (: ) acknowledges: There is a great deal that is unknown about the output, for example, whether it consists of a combinatorial structure that distinguishes individual objects and object parts, and whether it encodes nonvisual properties, such as causal relations, or primitive affective properties, like ‘dangerous’, or even some of the functional properties that Gibson referred to as ‘affordances’. In principle, the early-vision system could encode any property whose identification does not require accessing general memory, and in particular that does not require inference from general knowledge.5

Pylyshyn takes cognitive influences to be effected mainly by different kinds of attention, and he attempts to show that attention does not affect directly early vision, but only processes that take place either before the onset of early vision or after its completion: Our hypothesis is that cognition intervenes in determining the nature of perception at only two loci. In other words, the influence of cognition upon vision is constrained in how and where it can operate. These two loci are: (a) in the allocation of attention to certain locations or certain properties prior to the operation of early vision . . . (b) in the decisions involved in recognizing and identifying patterns after the operation of early vision. (Pylyshyn a: )6

3 Fodor (: ) speculates on the possibility of what we called ‘internal indirect cognitive penetrability’: ‘Heaven knows what psychological function “descending pathways” subserve. . . . One thing is clear: if there is no cognitive penetration of perception, then at least “descending pathways” aren’t for that.’ 4 See Pylyshyn (a) and Raftopoulos (). 5 In a more positive vein, Raftopoulos (: ) states that the ‘information delivered by that process is spatio-temporal (that is, information about location, spatial relations, orientation, and motion) and information about size, color, orientation, and viewer-centered shape, and is used for object individuation and tracking’. 6 See also Pylyshyn (: – and ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

Pylyshyn’s strategy allows him to reject the suggested cases of cognitive penetrability by placing them on a time-scale that either precedes or follows early vision. However, the way he formulates the first locus is incomplete and imprecise. It refers only to cases of external indirect cognitive penetrability, namely cases in which attention causes some kind of bodily movement that results to a change of the exteroceptive input. But there might be also cases where attention influences internally the perceptual system, and in these cases what is of importance is not whether attention takes place prior or even during early vision but whether it does (or not) affect inferentially the content of the representations processed during early vision. For example, an attentional process prior to early vision could cognitively penetrate it in a direct way by sending a signal that affects inferentially the content of the representations processed during early vision. On the other hand, an attentional process during early vision might not affect early vision inferentially but in some other way that does not constitute a case of direct cognitive penetrability. For example, it has been argued that the enhancement of the activity of all the neurons that correspond to the attended location does not modulate cognitively in a direct way the processing in early vision because it does not introduce a bias between targets and non targets.7 Thus, exempting cases of cognitive processes that take place after the process of early vision (see locus (b) in the quotation above), what is crucial in deciding whether a cognitive process influences cognitively early vision is not whether it is prior or concurrent with early vision, but whether or not it affects early vision inferentially.

. Problems with the connection between early vision content and the world It has been argued that the cognitive impenetrability of early vision entails that its output is nonconceptual and affords a direct connection with the world.8 Pylyshyn calls this direct connection ‘visual indexing’. A visual index, which Pylyshyn calls ‘FINST’ (for FINger of INSTantiation), performs ‘a demonstrative or preconceptual reference function’ (Pylyshyn : ), which simply picks out the referent without describing it: ‘Like natural language demonstratives (such as “this” or “that”) this direct connection allows entities to be referred to without being categorized or conceptualized’ (p. ). Thus, since the content of the states processed by early vision is causally constituted and cognitively impenetrable, it is nonconceptual and provides a direct connection with the perceived objects. Here it needs to be argued independently that the system of early vision has states with some kind of referential function, since a system can be cognitively impenetrable without having any states with referential function. So, one has to argue on independent grounds that early vision states have a referential function. Pylyshyn, following Fodor and Dretske, opts for a kind of causal theory of mental content. Whether or not a causal theory can account for the representational content of perceptual states is a debatable subject itself. For present purposes, I will simply argue that the cognitive 7

See Raftopoulos (: ).

8

See Raftopoulos ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



impenetrability of early vision, as defined by Pylyshyn, does not necessarily entail that its output depends exclusively on the world. As we have seen, Pylyshyn’s semantic criterion of cognitive penetrability is based on a strong notion of inference which is exemplified in the inferential promiscuity of beliefs. However, this leads to a very restricted notion of cognitive penetrability, since the kind of inference it introduces presupposes, among other things, that the inferentially connected representations must have contents that are (a) propositionaldiscursive (given that they can be contents of beliefs) and (b) domain-general, in the sense that they are not necessarily about a specific domain. If one accepts this strong version of cognitive penetrability, then a system A with representations that depend inferentially on certain representations of a system B without satisfying (a) and/or (b) has to be considered as cognitively impenetrable. This is the case if, for example, the representations of system A can depend inferentially only on domain specific representations9 or/and on non-propositional representations of system B. In this case, system A must be considered as cognitively impenetrable although it is not informationally encapsulated, since information from another system directly influences rationally the content processed in system A. In other words, the conception of cognitive penetrability based on the strong notion of inference applies only to systems that process domain-general representations with propositional-discursive contents. Thus, if one defines the notion of cognitive penetrability in terms of the strong notion of inference, the scope of the concept of cognitive impenetrability becomes too broad: even a system the representations of which depend in a non-promiscuously inferential way on representations of other systems has to be considered as cognitively impenetrable. Thus, the claim that early vision is not cognitively penetrable, in the sense that it does not depend in an inferentially promiscuous way on the representations of other systems, does not necessarily entail that early vision is exclusively data-driven, since it may also inferentially depend on propositional domain-specific and/or non-propositional representations of other systems. Moreover, as we will see in Section , there might be some kind of rational dependence which is not inferential. For example, the task dependence of vision10 could be conceived as a non-inferential dependence on some kind of practical nonpropositional knowledge. To recapitulate, between the informationally encapsulated systems and the systems that exhibit strong cognitive penetrability, a variety of systems can be specified that are neither informationally encapsulated nor strongly cognitively penetrable. I will call this kind of dependence of a system on other systems ‘weak cognitive penetrability’. Weak cognitive penetrability, if tenable, threatens the direct connection with the world that early vision supposedly affords, since, in this case, early vision content 9 This could be the dependence of motor systems on central systems, if motor systems are to be taken as not cognitively penetrable in the strong way. 10 See Raftopoulos (: ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

does not depend only on the stimulus. But, even if one grants that the content of the representations processed during early vision is not cognitively penetrable either in the strong or in the weak sense, this claim does not suffice to establish the direct connection with the external world that Pylyshyn assumes. This is because early vision outputs depend not only on the stimulus but also on the kind of processing that takes place during early vision. As Pylyshyn (: ) holds, the processing during early vision is modulated by a number of intrinsic principles that are hard-wired in the visual system: The visual system follows a set of intrinsic principles independent of general knowledge, expectations, or needs. The principles express the built-in constraints on how proximal information may be used in recovering a representation of the distal scene.

These intrinsic principles constitute a kind of implicit knowledge encapsulated in the visual modules. But still, the penetration of the content of early vision by implicit knowledge impugns the purported direct connection with the external world.11 It has been argued that the implicit principles reflect the geometry of our world and higher-order physical regularities that govern the behaviour of objects, since only in this way the evolutionary success of vision could be explained.12 However, even if one adopts this ‘reflection’ metaphor, it is more plausible, I think, to suppose that the intrinsic principles ‘reflect’ geometrical and physical regularities of the world in relation to the particularities of the embodied perceiver. After all, a visual system is not like a camera that just registers the incoming light; it processes and transforms the information from the incoming light in a way that is also sensitive to the geometry and certain physical properties and capacities of the eyes and body of the perceiver.13 In other words, even if one grants that early vision processes representations with content that is not cognitively penetrable either in the strong or in the weak sense, the intrinsic principles do not simply ‘reflect’ the unmediated effect of the world on the organism. Moreover, even if one grants that early vision is not cognitively penetrable either in the strong or in the weak sense and its intrinsic principles do not change, differences at the level of the output of early vision depend exclusively on the proximal stimuli. Thus, if two proximal stimuli are the same, the early vision system will produce the same output, even if the etiology of the two proximal stimuli is different—say, they are produced by different distal stimuli. So, the cognitive impenetrability of early vision does not secure in any way the perceiver’s direct access to the world, since early vision processing is indifferent to differences in the etiology of proximal stimuli. For what matters, the processing could be the same even if all proximal stimuli were produced not by the physical world but by a computer.

11 13

12 See Raftopoulos (:  and ). See Churchland (: ). This is one way to understand the claim that perception is essentially embodied.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



. Problems with the connection between early vision and visual experience But, even if one ignores these objections and grants that early vision outputs representations with content directly connected with the world, the question that remains is how this connection, which at this stage is encapsulated in early vision, allows the perceiver direct access to the world. This problem concerns the transition from the sub-individual to the individual-level of description; in particular, it concerns the way the encapsulated content of early vision affects the cognitively penetrated content of late vision processes and finally the content of visual experience that is available to the perceiver. Formulating the problem at the level of visual experience, the main question is: how does the content of early vision influence the distinctive content of visual experience? If the content of visual experience does not depend appropriately on the content of early vision, the latter proves to be disconnected from experience and even more so from thought. Of course, much hangs on how exactly this dependence should be understood. Given that we are dealing with a dependence between representational states, it seems that it should be some kind of semantic-rational dependence; if the content of the representations outputted by early vision did not have any semantic-rational relation with the content of the representations processed in late vision and finally with the content of visual experience, then the dependence of visual experience on early vision could not be described at the intentional level. In such a case, early vision should not be appropriately characterized as a kind of vision, but just as an enabling condition of visual experience.14 The most crucial problem at this juncture is to specify what is required for a subindividual content to become content appropriately accessible by an individual. This intimate relation between the individual-level representations and their possessor is notoriously elusive and difficult to account for. Individual-level representations, unlike sub-individual representations, are not just representations of something but also representations possessed by and attributed to an individual.15 But an individual A can possess a mental representation, attributed to her by individual B on the grounds of publicly available evidence, without A being able to have a more immediate access to her mental representation than B. This more immediate access to one’s own mental representations—not available to an external observer—is a first-personal access. Formulated in these terms, the crucial problem is what is required for a sub-individual content to become first-personal content. Pylyshyn does not raise this problem; but he briefly discusses a related problem, namely how to interpret the notion of conscious experience. His aim is not to advocate a particular way of interpreting this notion but just to show that ‘whether something is or is not part of the content of our experience is not self-evident’ (Pylyshyn : ).

14

See McDowell ().

15

Cf. Kriegel ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

To this effect he refers to Dretske’s, Block’s, and Rosenthal’s approaches on this issue. In what follows, I will focus on Block’s approach.16 The output of early vision has a peculiar status. On the one hand, if we grant that it is the product of a cognitively impenetrable process, it is not cognitively penetrated. On the other hand, if it is to play a role in the conceptual processes of late vision, it must be appropriately poised for use by this system, i.e. it must be cognitively penetrable by late vision. But in order for the output of early vision to be cognitively penetrable by late vision in Pylyshyn’s strong sense, its content must be inferentially promiscuous. Block (: ) introduces the notion of access consciousness to generalize this idea: A state is access conscious (A-conscious) if, in virtue of one’s having the state, a representation of its content is () inferentially promiscuous (Stich ), that is, poised for use as a premise in reasoning, () poised for rational control of action, and () poised for rational control of speech.

If the output of early vision is access-conscious, then it is cognitively accessible by late vision and other cognitively penetrated systems. Access consciousness is a functional notion. The content of a representation in a system is access-conscious relative to a different system that uses the content of the representation in a semantic-rational relation. As Block (: ) notes: What makes content A-conscious is not something that could go on inside a module, but rather informational relations among modules. Content is A-conscious in virtue of (a representation with that content) reaching the Executive System, the system in charge of rational control of action and speech.

But does access consciousness suffice to make the content of the output of early vision individual-level content, let alone first-personal content? This does not seem to be the case. After all, there is not yet any individual in the picture, just one part of the brain receiving input from another part. Unless the individual is identified with a certain part of the brain, it cannot be said to have access to the output of early vision. Block (a: ) adopts Dehaene’s model of broadcasting in a global workspace to make more explicit the idea of cognitive accessibility involved in the notion of access consciousness: Think of perceptual mechanisms as suppliers of representations to consuming mechanisms which include mechanisms of reporting, reasoning, evaluating, deciding, and remembering. There is empirical evidence that it is reasonable to think of perceptual systems as sending representations to a global active storage system, which is closely connected to the consuming systems. Those representations are available to all cognitive mechanisms without further processing.

This model, though it emphasizes the central role that global workspace plays, remains a description of the functional organization of the mind at the sub-individual level: the

16

See also Raftopoulos (: ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



global workspace is just a subsystem of the brain. So there is no individual at this level of description that could access cognitively the output of early vision. But even if we were to identify the individual with a subsystem of his brain that can access cognitively the output of early vision, nothing in this picture could account for the subjective character of experience when accessed first-personally, since the individual could have had exactly the same cognitive access without being conscious of it. One might suggest that the difference between sub-individual and individual-level content is that the former lacks phenomenal character. When I experience the blue of the ocean, there is something it is like for me to have this experience. Indeed, Block () introduces phenomenal consciousness as distinct from access consciousness. He conceives phenomenal consciousness as an intrinsic property of certain mental states that is independent of whether they are access conscious: ‘P-consciousness is not a functional notion. . . . [C]ontent gets to be P-conscious because of what happens inside the P-consciousness module’ (Block : ). Block () holds that the content of a state can be phenomenally conscious independently of whether the subject is aware of this content. Though he acknowledges that phenomenal states often involve a ‘me-ishness’ about them,17 he does not consider it as a constitutive element of phenomenal states, but just as an extra quale that may characterize some of them. However, a phenomenal state that has not a subjective character or—in Block’s terms—a ‘me-ishness’ is a free-floating state, not yet related first-personally to any individual. There is nothing it is like for me to have this supposed phenomenal state. As Levine (: ) notes: ‘Phenomenal states/properties are not merely instantiated in the subject, but are experienced by the subject. Experience is more than mere instantiation, and part of what that “more” involves is some kind of access.’ Block (a: ) concedes Levine’s point and introduces a constitutive noncognitive access relation of phenomenal states to the subject that he calls ‘awareness access’. This kind of awareness is not a higher-order relation to the phenomenally conscious state, since that would make it a kind of cognitive access consciousness— extrinsic to the state. But still it is a kind of access: ‘Something worth calling “accessibility” may be intrinsic to any phenomenally conscious state, but it is not the cognitive accessibility that underlies reporting’ (p. ). He further specifies this non-cognitive access in terms of a reflexive relation: ‘A conscious experience is reflexive in that it consists in part in an awareness of itself ’. Block (b: ) holds that a same-order or a deflationary view is adequate to account for what in his previous work called ‘me-ishness’ and now conceives as an element that characterizes any phenomenally conscious state:

17 See Block (: ).‘P-conscious states often seem to have a “me-ishness” about them; the phenomenal content often represents the state as a “state of me”.’

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

In , the only option I saw for explaining awareness-access in noncognitive terms was as a kind of phenomenal property I called ‘me-ishness.’ But now I see that awareness-access can be adequately understood in terms of ‘same-order’ and deflationary theories, so there is no need for cognitive or other ‘higher-order’ accounts.’

Thus, if we follow Block’s suggestion, we can attribute the first-personal character of phenomenal content to the noncognitive awareness access involved in it, which can be further specified as a same-order reflexive relation. This same-order noncognitive reflexive relation is what accounts for the subjective character of phenomenal states. I think that the general idea of this approach is on the right track, but it certainly needs further clarification. However, I would like first to raise some objections about the use of this general idea in the context of early vision. If we take for granted that the phenomenal content (and the awareness-access intrinsically involved in it) is constituted in the encapsulated process of early vision, it is legitimate to hypothesize that this content could have been produced even in the absence of any cognitive system. This means that even an early vision modulein-the-vat would suffice to produce individual-level contents. But this is certainly a counterintuitive result because it allows for individual-level content in the absence of any individual. Moreover, conscious occurrent thoughts and other purely cognitive episodes are also given first-personally and, thus, have subjective character. This is what differentiates them from unconscious thoughts and other unconscious purely cognitive states.18 Thus subjective character is not only an intrinsic element of perceptual episodes but also of conscious occurrent thoughts. But conscious occurrent thoughts do not have phenomenal content.19 It follows that subjective character is not exclusively an intrinsic characteristic of phenomenal states. If we insist on accounting for the subjective character of conscious occurrent thoughts and other cognitive episodes in terms of a corresponding same-order noncognitive reflexive relation, then we should clarify how is it possible for this reflexive relation not to be ‘contaminated’ by the cognitive relations that conscious occurrent thoughts bear to other cognitive states. This calls for a deeper understanding of the kind of noncognitive reflexive access that is involved in the first- personally given conscious episodes. What kind of access is this and how can we avoid the unwelcome consequence of attributing individual-level contents to the outputs of encapsulated modules? In section . I will make a specific suggestion on this issue to the effect that the proper kind of access is a nonreceptive practical access. There is, however, another general issue with the connection between early vision and visual experience. This issue concerns the very content of the output of early vision and whether it (or a part of it) contributes to the phenomenology of visual experience. We may distinguish two ways the output of early vision could be related to the content 18 19

See, also, Kriegel (: –). At least not the phenomenal content of perceptual states.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



of visual experience: the output (or a part of it) could be available to visual experience either with or without further processing. If it is available through further processing and given that this processing would take place in the late vision system which is cognitively penetrated, the content of the output of early vision would first be transformed into a kind of cognitively penetrated content before it becomes part of the content of visual experience. But then, this transformed content is, by Pylyshyn’s strong notion of cognitive penetrability, inferentially promiscuous and propositional-discursive. However, a propositionaldiscursive content cannot account for visual phenomenology, since a thought could have the same content without exhibiting any visual phenomenology. If the output of early vision (or a part of it) is available to visual experience without further processing, then the content of visual experience would involve a cognitively impenetrable part. The problem with this option is that the phenomenology of visual experience is quite unlike the presumed phenomenology of the content of the output of early vision. The stability and the seamlessness of our visual phenomenology cannot be accounted for in terms of the evanescent and volatile content of early vision. After all, we experience the blue of the ocean and not the blue of the fleeting output of early vision. One could object that, as Pylyshyn claims, early vision indexes to worldly objects, and hence that the blue we experience is the blue of the indexed object.20 So there is no intermediate object introduced by early vision. This is certainly right, if we take for granted Pylyshyn’s use of visual indexes. However, as Pylyshyn (: ) acknowledges: ‘The individual items that are picked out by the visual system and tracked primitively are something less than full-blooded individual objects.’ This is because the visual index allocated to the picked out object does not yet represent its properties or its location.21 It is for this reason that Pylyshyn calls the particular objects ‘visual objects’ or ‘proto-objects’. Thus, initially, the visual index does not encode the colour of the picked-out object. But early vision does not just allocate visual indexes to visual objects, it gradually builds representations of certain properties of the indexed object. As Pylyshyn (: ; emphasis added) holds: In principle, the early-vision system could encode any property whose identification does not require accessing general memory, and in particular that does not require inference from general knowledge.

Thus the real issue is the following: if the representations of the properties of an indexed object, such as colour, outputted by early vision are available to visual experience without further processing, can the content of these representations account for visual phenomenology?

20

Thanks to an anonymous referee for raising this objection. See e.g. Pylyshyn (: ): ‘What we need is a way to refer to individual things in a scene independent of their properties or their locations. This is precisely what FINSTs provide.’ 21

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

The stability and the seamlessness of our visual phenomenology is characterized by the fact that what we visually experience is not exhausted by what is sensorily present. This is a widespread characteristic of perceptual content, namely that it presents various perceptual constancies. Experiencing such constancies is the ability to perceive physical properties of objects as unchanging, although their sensory presence changes due to the changing physical relation one has to the object. In these cases, the unchanging properties are perceptually present even though they are not sensorily present. But the representations of properties of an indexed object outputted by early vision are restricted to what is sensorily present. Thus, early vision representations cannot account for this aspect of the phenomenology of visual experience.22 Another distinctive feature of perceptual phenomenology is the presentational character of perceptual content. To take the case of vision, the presentational character is what differentiates the phenomenology of visual experience from the phenomenology of quasi-visual states such as visual imagination or visual recall: when I visually imagine or visually recall something, it does not appear to me as bodily present in the external environment. The presentational character has to do with the distinctive directness of visual phenomenology compared to the directness of the phenomenology of quasi-visual states such as visual imagination or visual recall: vision provides us access to bodily present objects. However, the presentational character of visual experience cannot be accounted for just by what is sensorily present to early vision, since the same sensory qualities could be represented in visual imagination or visual recall without having presentational character. But, perhaps, one could object that the mental representations involved in visual imagination and visual recall are less vivid. This objection, however, is inadequate, since presentational character is not a matter of degree of vividness: a visual experience could also be less vivid, without this affecting its presentational character. For example, even in dense fog the objects seen display presentational character. The same holds for the visual experience of a myopic person who is not wearing her glasses: though her visual field is blurry, the perceived objects continue to have presentational character. Thus, the presentational character of visual experience cannot be accounted for simply in terms of the sensory character of the cognitively impenetrable representations available to early vision.23 But if the cognitively impenetrable content of early vision cannot account for the phenomenology of visual experience, and late vision is cognitively penetrable—which, according to the strong interpretation, means that its content is inferentially promiscuous and propositional-discursive—how could one account for visual phenomenology? 22 It can further be argued, following Smith (: ch. ), that perceptual constancies cannot be attributed to the presence of some descriptive content. But if there are aspects of perceptual content that are neither sensory nor descriptive, then we should look for a third source that might determine perceptual content besides the external world and, possibly, the higher cognitive systems. In Section , I will suggest practical non-propositional knowledge as a factor that determines perceptual content. 23 For a more detailed argumentation see Pagondiotis ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



 Restricting Penetrating Cognition to Practical Nonpropositional Knowledge Thus far, I have examined Pylyshyn’s approach to cognitive penetrability of vision— an approach that restricts the dependence of vision on cognition—by restricting the scope of the first relatum to late vision. Moreover, this move takes for granted the second relatum, cognition, since it does not make any distinction between different kinds of penetrating cognition. This implies that late vision is considered as penetrable by the same kind of cognition by which states of the central systems are penetrable. I have called this kind of cognitive penetrability ‘strong cognitive penetrability’. I argued that this approach disconnects early vision content from the world and from the cognitively penetrated visual content first-personally available to the perceiver, and cannot account for visual phenomenology. Moreover, the approach blurs the distinction between the cognitively penetrated visual content and the content of thought, since no principled distinction between them is introduced. For these reasons, in the remainder of the chapter I explore an alternative suggestion based on the opposite move. I formulate this suggestion at the level of visual experience. In particular, I take the first relatum, visual experience, for granted, and introduce a restriction on the kind of cognition that can directly penetrate visual experience. My suggestion is that visual experience, unlike thought, is directly penetrated only by practical nonpropositional knowledge. This implies that the cognitive penetrability of vision should be distinguished from the cognitive penetrability of thought, since vision is cognitively penetrated only by practical nonpropositional knowledge. This is one possible way to specify what I called ‘weak cognitive penetrability’. As we have seen in Section ., if one remains in the context of the representational and computational theory of mind, weak cognitive penetrability can be further specified as a non-promiscuous inferential dependence of the content of visual representations on the content of representations of some cognitive system. A non-promiscuous inferential dependence could be a dependence on propositional domain-specific and/or nonpropositional representations of other systems. However, in the remainder of the chapter I will focus on weak cognitive penetrability of vision, understood as the dependence of vision on practical nonpropositional knowledge. This is a more radical option on cognitive penetrability found in the work of a number of philosophers and scientists24 who reject the framework of RCTM and support anti-representationalist approaches to perception and cognition. In these approaches, the processes subserving perceptual and cognitive capacities are not taken as consisting in computational transformations of symbolic representations. This precludes an understanding of cognitive penetrability in terms of a (promiscuous or not) inferential dependence of symbolic representations of the perceptual systems on 24

See e.g. Dreyfus (), Haugeland (), Hurley (), Noë (), O’Regan ().

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

symbolic representations of the cognitive systems. Thus, Pylyshyn’s semantic criterion of cognitive penetrability is not applicable to the anti-representational approaches. For this reason, I have to develop an account of weak cognitive penetrability that is not based on the inferential dependence of symbolic representations of the perceptual systems on symbolic representations of higher-level systems. Before examining this problem, let us see why perception has to depend on practical nonpropositional knowledge. Churchland () pointed to this direction in his exchange with Fodor. Churchland accepts that the way we perceive the world does not change immediately after we change our beliefs about it. It takes training and practice to change the way we perceive the world: Who ever claimed that the character of a scientist’s perception is changed simply and directly by his embracing a novel belief? None of the theorists cited in Fodor’s paper have defended such an unrealistic view. And all of us have, at some point or other, emphasized the importance of long familiarity with the novel idiom, of repeated practical applications of its principles, and of socialization within a like-minded group of researchers. (Churchland : –)

It is for this reason that Churchland opts for a diachronic cognitive penetrability of perception by cognition. This amounts to holding that perceptual processing is not innately specified but exhibits a plasticity over time due to the acquisition of new skills. Thus, Churchland (: ), can be seen as suggesting a restriction of the scope of cognition that can directly penetrate perception to some kind of practical nonpropositional knowledge involved in skills of applying the newly acquired propositional knowledge in a wide variety of circumstances. But what exactly is the kind of dependence perception has on practical nonpropositional knowledge? Alva Noë () has recently explored in detail this dependence, so his work is a good entry point to the problem of specifying weak cognitive penetrability in anti-representational terms.25

. The proper dependence of visual experience on sensorimotor knowledge Noë (: ) holds that ‘perception depends on the possession and exercise of a certain kind of practical knowledge’. He understands this dependence as constitutive. More specifically, starting from a direct realist position—namely that perception presents us directly with the external objects and some of their intrinsic properties26 — he attempts to show that perceptual content rests on practical knowledge of the way perspectival properties of objects would change as we move relative to them. Thus, we manage to see a coin as circular, even when, seen from an angle, its perspectival shape 25 See also Hurley (). Hurley has devoted a significant part of her book on the intimate interrelation between perception and action. Her basic idea is that the intimate interdependence between perception and action at the personal level rests on the fact that they both emerge from the same complex dynamic sensorimotor system as described at the subpersonal level. Thus, according to Hurley, perception is not subserved by some kind of subpersonal input system that functions as an interface between world and mind, nor action emerges, correspondingly, from a subpersonal output system. 26 See Noë (: ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



is not circular but elliptical, because we possess practical knowledge of the way the perspectival shape of the coin would change as our point of view changed in relation to the coin. Noë (: ) holds that perceptual content presents both the intrinsic and the perspectival properties of objects, and for this reason he considers it as having a dual aspect: ‘The plate looks to be circular (it really does) and it looks elliptical from here (it really does).’ Noë’s main idea is that perceptual content depends constitutively on the possession of this kind of practical knowledge which he calls ‘sensorimotor knowledge’. But how should we understand the particular constitutive relation? One way is to take perceptual content as consisting, even partly, of the content of this sensorimotor knowledge. Thus, to visually experience a coin as circular is to know how its current perspectival shape would change relative to my visual exploration of the coin. This approach seems to reduce the perceptual presentation of an intrinsic property to practical knowledge of how the presentation of a corresponding perspectival property changes relative to the perceiver’s exploratory movements. It is for this reason that Noë considers the circularity of the coin as virtually present in perceptual content, but he insists that this ‘virtual presence is a kind of presence, not a kind of non-presence or illusory presence’ (Noë : ). In other words, the perceived intrinsic properties of objects are ‘present as available, rather than as represented’. However, one could note that, even after the reduction suggested by Noë, the perspectival properties of the object continue to figure in the content we get after the reduction. More specifically, the reductive approach we are examining is in fact an attempt to reduce one aspect of perceptual content to the other plus the relevant sensorimotor knowledge: the perceptual presentation of intrinsic properties is taken to consist in knowing how the presentation of perspectival properties would change relative to one’s exploratory movements. Noë (: ) is keen to emphasize that the presentation of perspectival properties is not a mere sensation, some kind of sense-datum; it is about ‘genuine, relational properties of things’. However, his account creates new problems about how to deal with the perceptual presentation of perspectival properties. Noë (p. ) admits that there is a threat of regress if the experience of perspectival properties is taken as primitive: ‘Do we see them by seeing how they look? This would threaten to lead to infinite regress (after all, one would need to experience the looks of the looks in order to see how things are, and so on, ad infinitum).’ For this reason, he opts for an approach that reduces the presentation of perspectival properties to the practical knowledge involved in a set of more basic sensorimotor skills: ‘The plate looks elliptical to me, because, to indicate its shape, I can (and indeed, in some sense, must) move my hand in a characteristic manner. That is, to experience a thing as elliptical is precisely to experience it as occupying a particular kind of region in one’s egocentric, sensorimotor space’ (Noë : ). It is this reduction that allows (p. ) to hold that the content of experience is virtual all the way in, namely, that it is all present as available, as accessible. However, this result does not fit well with the phenomenology of perceptual experience and with direct realism with respect to perception, because, if things are

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

as Noë describes, then vision, for example, would not allow us to distinguish what is occurrent, bodily present in front of us, from what is not. Moreover, we do not experience the mere availability of the intrinsic properties of objects relative to our exploratory movements; we experience the very properties themselves. In other words, the perceived intrinsic properties of objects are given categorically and not dispositionally to the perceiver. If perceptual content is reduced to ‘a set of counterfactual implications for sensorimotor activity’, then, as Campbell notes, ‘[t]he ordinary world, there independent of us, there for us to explore, has simply disappeared’ (Campbell : ). Unlike what Noë claims, perceptual content is not virtual all the way in. Noë seems to presuppose here that, if perceptual content is not present as available, the only alternative is to be present as represented, a claim that he rejects since he takes it to necessarily entail the acceptance of mental representations. However, these are not the only available options. Alternatively, the perceptual relation could be taken as an acquaintance relation with the objects and properties,27 or as an intentional relation where perceptual content consists of object-dependent senses.28 Both these alternatives are versions of direct realism. In what follows, I will adopt the latter.29 The endorsement of this alternative entails the rejection of Noë’s claim that sensorimotor knowledge is a constituent of perceptual content. But, then, in what way is perceptual content dependent on sensorimotor knowledge? I will employ McDowell’s use of the Sellarsian idea that, besides the logical dependence of our beliefs on observation reports, there is a logical dependence in the opposite direction, namely a dependence of observation reports on our world-view that is rational but non-inferential. McDowell, in using this idea, puts perceptual experiences in the epistemological position in which Sellars puts reports of observation: Unlike the dependence that traditional empiricism focuses on, this dependence in the opposite direction is not inferential. Suppose one knows by looking that some object is green. That the lighting conditions are appropriate for telling what colours things have is not a premise in an inferential justification one could appropriately give for one’s claim that the thing is green. On the contrary, one’s justification for the claim is simply that one sees that the thing is green. But it is a way of putting what Sellars urges in introducing the second dimension of dependence, as exemplified in the case of colour experience, to say that the very possibility of one’s having that justification—a justification consisting in the fact that one sees that the thing is green— depends on one’s having suitable knowledge about the effects of lighting conditions on colour appearances. And though it is not inferential, this dependence is rational. (McDowell a: )

McDowell concedes that the background knowledge on which certain fundamental kinds of experience depend may be rather practical, not theoretical-propositional. He uses the example of colour experience: 27 28 29

For a recent use of this Russellian idea in favour of a direct realist approach, see Campbell (). See McDowell (). See Pagondiotis () for a defence of a version of this alternative.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



It would be infelicitous to describe the bit of our world view that is embodied in colour experience . . . as a theory. The general knowledge (so called) that Sellars invokes need not be acceptance of a body of propositions at all, inferentially articulated or not. It might be simply a responsiveness in practice to differences in lighting conditions, a practical rather than theoretical grasp of their significance for the possibility of telling what colours things have by looking. So the dependence of colour experience on background knowledge need not be a case of experience being theory-laden. (McDowell a: –)

This is a weaker notion of constitutive dependence which does not impugn the presentational character of fundamental kinds of perceptual experience; rather, it concerns the conditions that justify the authority of the perceiver. This is reflected in the way a viewer responds in case his observational judgment about the colour of an object is challenged. The viewer can initially respond that the colour is, say, red, because he sees it. But if pressed to provide further justification, he cannot give any premises that lead to the conclusion that what he sees is red. The viewer does not offer ‘premises for an inference to the truth of . . . [his] claim’ (McDowell : ). Instead, he justifies his own authority to make judgments of this kind in general. Thus, he may respond that he knows how to recognize red or that he knows how to distinguish whether the illumination conditions are appropriate for telling what the colour is. In this line, McDowell (b: ) suggests that the background knowledge on which perceptual content depends ‘may be a purely practical ability to discriminate appropriate from inappropriate conditions’. To summarize my suggestion about how to specify weak cognitive penetrability in anti-representational terms, I proposed that we should understand the constitutive dependence of perceptual content on sensorimotor knowledge in a weak way: sensorimotor knowledge is not a constituent of perceptual content, but something on which perceptual content depends in a rational, non-inferential way. This solution avoids the danger of reducing perceptual knowledge to some kind of practical knowledge. Moreover, it is compatible, as we saw, with a direct realist approach to perception, since perceptual content consists of object-dependent senses. More specifically, the weak constitutive dependence of perceptual content on sensorimotor knowledge allows the perceptual content of the embodied perceiver to be anchored to the world. This is because the acquisition of sensorimotor knowledge requires an intimate cooperation between the embodied perceiver and the world. Just as one cannot become an expert tango dancer if he does not learn to be in tune with his partner, so one cannot become an expert perceiver if one does not learn to be in tune with the way the perceived scene changes as a function of, among other things, the change of distance, angle, and movement between the perceiver and the perceived scene. This attunement between perceptual content and world does not suffice to make perceptual content a ‘reflection’ of the world. After all, perception is not infallible and often, as in the case of illusions, it systematically presents the world in a nonveridical way. However, even though the attunement between perceptual content and world does not suffice to make the former a reflection of the latter, it establishes a systematic

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

dependence of the modifications of the perceptual content on the changes of the world. What is crucial for perception is to present the changes of the world in a stable and systematic way even if not always veridically. This is a sufficient ground for scientists and ordinary observers to communicate and to find out what is or is not veridical.30 More specifically, the fact that perceptual experience is cognitively penetrable in a different way from thought allows the constitution of two different modes of presentation of the world, the perceptual and the discursive. The relative autonomy of each mode makes possible the mutual testing of the coherence between the two modes as well as within each one of them. My suggestion on the constitutive dependence of perceptual content on sensorimotor knowledge is a way to specify weak cognitive penetrability in anti-representational terms. This dependence is not a dependence on a distinct cause on which perceptual content depends inferentially. The sensorimotor knowledge involved in perceptual content does not necessarily entail that perception depends on some actual bodily movement, since there is perception even in the absence of any bodily movement. However, any perception depends on sensorimotor knowledge in the sense that it involves anticipations. Anticipations concern the future and, in that sense, cannot be efficient causes of perceptual content. Thus, the dependence of perceptual content on sensorimotor knowledge is not (efficiently) causal. If we have to use the idea of causation to characterize the dependence of perceptual content on sensorimotor knowledge, then we could hold that the latter constitutes part of the formal cause of perceptual content. I have already characterized this dependence, involved in weak cognitive penetrability, as rational non-inferential. In the next section, I will further specify the particular dependence as nonreceptive and implicit.

. The presentational character of visual experience and its dependence on sensorimotor knowledge Let me finish with a suggestion about how we should understand more specifically weak cognitive penetrability in anti-representational terms. I will focus on the presentational character of visual content and I will suggest an account of it that will make clearer the way visual content depends on sensorimotor knowledge. As we have seen in Section ., the presentational character has to do with the distinctive directness of visual phenomenology compared to the directness of the phenomenology of quasivisual states such as visual imagination or visual recall: vision provides us access to bodily present objects. Visual content is not just object-dependent, but it is dependent on bodily present objects. To account for the presentational character of visual experience, we have to examine the interdependency between the perceived object and the perceiver. My suggestion is that the presentational character of the perceived object is due to the intimate relation between the perceiver’s awareness of the object and the implicit bodily self-knowledge 30

Cf. Fodor (: –).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



of the perceiver. To take a simple example, when we look at a static object and move our eyes, we do not experience the object as moving; we experience it as remaining still due to our implicit bodily self-knowledge. Thus, there is a close coordination, or rather coupling, between the awareness of the object and the implicit bodily selfknowledge of the perceiver. When this coordination breaks down, what we are aware of loses its presentational character. This is why the appearance of an after-image in our visual field is not experienced as perceiving. The after-image does not afford us any appropriate way to explore it, since it is completely dependent on the movement of our eyes. My suggestion is that the presentational character of an experienced object does not rest simply on our awareness of the way things visually-sensorily look or of how the way they visually-sensorily look changes. Rather, it rests on our awareness of how the way things visually-sensorily look changes relative to our bodily movement or rest. In other words, the presentational character involves amodal elements and does not amount just to a kind of sensory-like presence. Thus, even if it were possible to have a visual-like experience where the changes in how things sensorily look were indistinguishable from such changes in a genuine visual experience, the subject could still be aware of a difference that would be revealed as a sense of passivity that would characterize the former case of awareness. This can be exemplified in the case of hallucination. Hallucinatory content is not dependent on any objects in the subject’s visual field, so it is not dependent on any retinal image, and this makes it also independent from the subject’s bodily movements—a fact that explains the sense of passivity that may accompany a hallucinatory experience. This approach to the presentational character of a perceived object requires analyzing the act of perceptual awareness. This act seems to involve a double consciousness: besides the consciousness of the perceived object, visual perception also involves a consciousness of perceiving, namely a self-knowledge that concerns the very act of perceiving. The consciousness of perceiving is integral to the act of perceptual awareness. As we have seen, Block (a) considers this as a noncognitive access that is not a higher-order awareness, but it is involved constitutively in perceiving as a same-order reflexive relation. We may now further specify the consciousness of perceiving as implicit and nonreceptive.31 The implicitness and nonreceptiveness of the consciousness of perceiving account for the transparency of perceptual experience, since the only object we are receptively aware of is the perceived object. One aspect of the consciousness of perceiving is bodily self-knowledge. Bodily self-knowledge is not a kind of receptive knowledge: it does not inform us of our experiencing body as something that involves qualities for which the distinction 31 Cf. Rödl (). A related notion has been elaborated in the phenomenological tradition. For a recent discussion of it, see Gallagher (), Zahavi (), and Thompson (). See also Longuenesse (). There is, though, an important differentiation between Rödl’s approach and the phenomenological approach: only the former approach considers the self-conscious relation as intentional. However, the examination of this issue is beyond the scope of this chapter.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

sensorily given vs. nonsensorily given applies. My suggestion is that this is so because bodily self-knowledge does not rest on an organ that is to some extent under our voluntary control.32 Our voluntary control on the organ of vision makes possible for us to change what we see by changing our view on what we see, and for the same reason, it makes possible for us to interrupt our seeing by closing our eyes. Bodily self-knowledge does not afford us any corresponding degrees of freedom. We cannot interrupt it, neither can we change our way of access to our body. It is for this reason that bodily self-knowledge does not afford us ‘views’ of our body. This inability to make a distinction between different ‘views’ of our body does not allow us to distinguish based on bodily self-knowledge between the way our body appears and the way it is. It is for this reason that bodily self-knowledge does not inform us of our experiencing body receptively. The fact that perceptual awareness involves constitutively the consciousness of perceiving—an aspect of which is bodily self-knowledge—accounts for the egocentric elements of perceptual experience. Thus, when we see an object, we are not only conscious of the object, we are also implicitly and nonreceptively conscious of our orientation, our distance and our movement in relation to the object. This is a necessary condition for seeing the world as independent of us. More particularly, our ability to move our eyes and our body makes possible for us to change views on the world. Thus the same object can be viewed from different angles, and this allows us to distinguish the way the object appears from the way it is. In this sense, vision presents its objects as independent of the awareness of them. Thus, vision presupposes that the perceiver has bodily self-knowledge of the motion or rest of his visual organ (and, more generally, of his body to the effect that his body influences the motion or rest of his visual organ). The mere movement or rest of the visual organ without an accompanying bodily self-knowledge of these changes would make the relation of vision to its subject matter analogous to the relation of bodily consciousness to the body. That is, every change, that such kind of ‘vision’ would make us aware of, would appear as a change of its subject matter. And it would be impossible to distinguish the way things visually appear from the way they are. That would not allow us to experience the objects of vision as independent of our visual awareness of them. If what I have said is tenable, then bodily self-knowledge of the motion or rest of the visual organ must be a constituent part of the consciousness of perceiving and necessary for experiencing the perceived objects as independent of the visual awareness of them. Thus, my awareness of the sensible qualities of an object and of the way they change relative to my bodily movement or rest involves also a nonreceptive self-knowledge of the kind of experience I entertain—namely, of whether, for example, I see, hear, or just visually imagine the object. Of course, I can be wrong about the kind of conscious act I undergo, but the point is that there must always be a 32

See Armstrong (: –). Cf. Shoemaker (: –).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



nonreceptive consciousness of this sort when I am perceptually directed towards something; perceiving something, as opposed to, say, imagining it, has a different cognitive significance for the subject: when I visually imagine a lion, as opposed to seeing it, I do not have, ordinarily, the tendency to run away. The consciousness of perceiving is not inferred at a later stage from the content of perception, but it is an integral part of the act of conscious perception. This precludes that there are free-floating perceptual experiences not yet related first-personally to any individual. If my suggestion about the double consciousness involved in perception is tenable, it provides a more specific way to understand weak cognitive penetrability in antirepresentational terms. In particular, it gives an account of how sensorimotor knowledge penetrates perception: what makes available to the perceiver the sensorimotor knowledge on which perceptual content depends is the implicit nonreceptive bodily self-consciousness constitutively involved in the act of perceiving. I suggested that specifying weak cognitive penetrability in anti-representational terms allows us to account for the distinctive content of visual experience in a way that preserves the distinction between vision and thought without disconnecting visual content from the world and the embodied perceiver. This suggestion raises further questions for future investigation, the most pressing of which is, I think, how thought depends on the visual content available to the embodied perceiver. But I hope to have shown that the suggestion is worth pursuing.33

References Armstrong, D. (). A Materialist Theory of the Mind. London: Routledge and Kegan Paul. Block, N. (). On a confusion about the function of consciousness. Behavioral and Brain Sciences : –. Block, N. (a). Consciousness, accessibility, and the mesh between psychology and neuroscience. Behavioral and Brain Sciences : –. Block, N. (b). Author’s response. Behavioral and Brain Sciences : –. Campbell, J . (). Sensorimotor knowledge and naïve realism. Philosophy and Phenomenological Research (): –. Churchland, P. M. (). Perceptual plasticity and theoretical neutrality: a reply to Jerry Fodor. In P. M. Churchland, A Neurocomputational Perspective: The Nature of Mind and the Structure of Science, –. Cambridge, Mass.: MIT Press. Dreyfus, H. (). What Computers Still Can’ t Do: A Critique of Artificial Reason. Cambridge, Mass.: MIT Press. Fodor, J. (). The Modularity of Mind: An Essay on Faculty Psychology. Cambridge, Mass.: MIT Press. 33 I am grateful to Vaios Papatheocharis for his critical comments on various drafts of this paper that helped me clarify my arguments. I also thank Athanassios Raftopoulos, John Zeimbekis, and the two anonymous referees for their comments and suggestions.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



c. pagondiotis

Fodor, J. (). A Theory of Content and Other Essays. Cambridge, Mass.: MIT Press. Gallagher, S. (). Bodily self-awareness and object perception. Theoria et Historia Scientarum (): –. Haugeland, J. (). Mind embodied and embedded. In Having Thought: Essays in the Metaphysics of Mind, –. Cambridge, Mass.: Harvard University Press. Hurley, S. (). Consciousness in Action. Cambridge, Mass.: Harvard University Press. Kriegel, U. (). Subjective Consciousness: A Self Representational Approach. Oxford: Oxford University Press. Kriegel, U. (). Personal level representation. Protosociology : – Levine, J. (). Two kinds of access. Behavioral and Brain Sciences : –. Longuenesse, B. (). Self-consciousness and self-reference: Sartre and Wittgenstein. European Journal of Philosophy (): –. Macpherson, F. (). Cognitive penetration of colour experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research : –. Marr, D. (). Vision: A Computational Investigation into Human Representation and Processing of Visual Information. New York: Freeman. McDowell, J. (). De Re senses. Philosophical Quarterly : –. McDowell, J. (). The content of perceptual experience. Philosophical Quarterly : –. McDowell, J. (a). Response to Costas Pagondiotis. Teorema : –. McDowell, J. (b). Response to Stella Gonzalez Arnal. Teorema : –. McDowell, J. (). Brandom on observation. In B. Weiss and J. Wanderer (eds), Reading Brandom: On Making It Explicit, –. New York: Routledge. Noë, A. (). Action in Perception. Cambridge, Mass.: MIT Press. O’Regan, K. (). Why Red Doesn’t Sound Like a Bell. Oxford: Oxford University Press. Pagondiotis, C. (). Hallucination, mental representation, and the presentational character. In F. Macpherson and D. Platchias (eds), Hallucination, –. Cambridge, Mass.: MIT Press. Pylyshyn, Z. (). Computation and cognition: issues in the foundations of cognitive science. Behavioral and Brain Sciences : –. Pylyshyn, Z. (a). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences (): –. Pylyshyn, Z. (b). Author’s response. Behavioral and Brain Sciences (): –. Pylyshyn, Z. (). Visual indexes, preconceptual objects, and situated vision. Cognition : –. Pylyshyn, Z. (). Seeing and Visualizing: It’s Not What You Think. Cambridge, Mass.: MIT Press. Pylyshyn, Z. (). Things and Places: How the Mind Connects with the World. Cambridge, Mass.: MIT Press. Raftopoulos, A. (). Cognition and Perception: How Do Psychology and Neural Science Inform Philosophy? Cambridge, Mass.: MIT Press. Rödl, S. (). Self-Consciousness. Cambridge, Mass.: Harvard University Press. Shoemaker, S. (). Self-knowledge and ‘inner sense’, Lecture I: The object perception model. Philosophy and Phenomenological Research (): –. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Smith, A. D. (). The Problem of Perception. Cambridge, Mass.: Harvard University Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

cognitive (im)penetrability of vision



Stich, S. (). Beliefs and subdoxastic states. Philosophy of Science : –. Thompson, E. (). Representationalism and the phenomenology of mental imagery. Synthese : –. Zahavi, D. (). Subjectivity and Selfhood: Investigating the First-Person Perspective. Cambridge, Mass.: MIT Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

Afterword Epistemic Evaluability and Perceptual Farce Susanna Siegel

Several contributors in this volume observe that each part of the notion ‘cognitive penetration of perception’ can be specified in multiple ways.1 The first parameter concerns the kind of perception at issue: is it early vision, or only those parts of perceptual experience produced by early vision, or all perceptual experience, or perceptual judgment downstream of experience, or any perceptual state or event, or the entire perceptual system? The second parameter concerns what counts as a cognitive influence: in addition to belief, suspicions, desires, and fears, do such influencers include affect, imagination, attention, sensorimotor dispositions, any information learned through patterns of exposure, or stored assumptions used in perceptual inference? Is ‘cognitive’ a placeholder for ‘psychological’, or does it mark a distinction in the mind between cognition and perception? The third parameter is the relationship that characterizes the influence: is it merely causal? Is it semantically relevant? Does it include associations between properties that are represented by the influencer and by the perceptual state? Taken together, the chapters in this volume make the case for a family of phenomena that differ depending on how each of these parameters is fixed. Most contributors focus on whether the architecture of the mind allows influences on perception that are defined by setting the parameters listed above.2 Many offer arguments with a deflationary flavor: a phenomenon that is only superficially similar to ‘properly’ cognitive penetration has been mistaken for it,3 or a phenomenon that is superficially similar to properly cognitive penetration doesn’t share the epistemically interesting 1 See the contributions by Deroy (Ch. ), Machery (Ch. ), Macpherson (Ch. ), Mole (Ch. ), and Stokes (Ch. ). For comments and discussion, thanks to Ned Block, Alex Byrne, Jeremy Dolan, Eric Mandelbaum, Farid Masrour, and especially Zoe Jenkin. 2 See the contributions by Briscoe (Ch. ), Burnston and Cohen (Ch. ), Deroy (Ch. ), Dokic and Martin (Ch. ), Dretske (Ch. ), Machery (Ch. ), Macpherson (Ch. ), Mole (Ch. ), Wu and Mahon (Ch. ), Raftopoulos (Ch. ), and Pagondiotis (Ch. ), as well as Macpherson (). 3 Deroy (Ch. ), Dokic and Martin (Ch. ), Machery (Ch. ), and Raftopoulos (Ch. ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

upshots that properly cognitive penetration is supposed to have.4 Nearly all of these contributors assume an architecturally significant distinction between perception and cognition, relative to which the kind of influence on the perception that they call attention to do not count as cognitive. Given these pluralities, we can see that there is no single phenomenon of cognitive penetration. There is therefore no single version of the existence question: ‘Is there cognitive penetration?’, or the evidential question: ‘Is this experimental result evidence of cognitive penetration?’, or the epistemic question: ‘Does cognitive penetration have philosophically important epistemic consequences?’ In response to the plurality of phenomena, it is useful to postpone the existence and evidential questions until one or another theoretical purpose for the notion of cognitive penetration has been specified. If the theoretical purpose concerns the architecture of the mind, we can ask: could we learn that the human mind has or lacks architecture X, if we learned that phenomenon Y did or didn’t occur in our minds? This question could guide us in selecting phenomenon Y to use in formulating corresponding existence, evidential, and epistemic questions. Stokes advocates this general ‘consequentialist’ approach, with an eye toward singling out a unique phenomenon for us to attach to the label ‘cognitive penetration’.5 But nothing precludes us from using this approach multiple times, to yield multiple existence, evidential, and epistemic questions. A different response to the plurality starts directly with epistemically significant or epistemically interesting phenomena, and considers which psychological structures would give rise to them, without trying to decide which of those structures, if any, exemplify cognitive penetration. This approach skips the existence, evidential, and epistemic questions, and avoids the need to distinguish between perception and cognition. There need not be any notion of cognitive penetration that is defined in such a way that some instances of it would give rise to the epistemic implications of interest. I’ll apply this second approach to two sets of epistemic phenomena. First, I consider which psychological precursors of perceptual experience impact its power to provide rational support for certain other propositions. Many contributors mention a negative epistemic upshot that they assume cognitive penetration could generate, and take this negative upshot to be the main thing at stake in whether the architecture of the mind allows it.6 I make the case that this attitude may tie 4 See the contributions by Briscoe (Ch. ), Lowe (Ch. ), Pagondiotis (Ch. ), and Zeimbekis (Ch. ). From the opposite end, Mole’s conclusion (Ch. ) is inflationary: attention-mediated influences on perception have been excluded from definitions of cognitive penetration, when in fact they should be included. And Lyons (Ch. ) argues from a reliabilist perspective that the same negative epistemic upshot can apply to both cognitive penetration and mere influence on judgment. 5 Stokes (Ch. ). 6 Zeimbekis (Ch. ) mentions ‘epistemically pernicious consequences usually expected of cognitive penetration’ to which the effects he focuses on are immune. Dokic and Martin (Ch. ) likewise suspect that the epistemological consequences of the cognitive penetrability of feelings would be very different from,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



the architectural and the epistemic questions together too tightly, and argue that the negative upshot is only one interesting upshot among many that psychological influences might have on perceptual experience. Second, I identify an even wider family of phenomena that I call ‘perceptual farce’ to help us consider how psychological precursors could impact the role of perceptual experience in reflecting and sometimes masking social forces. I discuss each of these epistemic phenomena below.

 The Epistemic Evaluability of Perceptual Experience Suppose you see Jack walking toward you. Seeing his face can give you reason to think that he is angry. What is the role of your visual experience in giving you reason to form this belief about Jack? Many philosophers hold that your visual experience purports to represent features of Jack and his face, in the sense that your experience would be inaccurate if things in the external world weren’t the way your experience presented them as being. In presenting you with Jack and his facial expression, your experience identifies something that might be reasonable for you to believe: for instance, ‘He is angry’. A natural next idea is that under certain conditions, you can have reason from your experience to believe what your experience suggests to you. In the history of analytic philosophy, both this construal of perceptual experience and its power to give you reason to believe what it suggests are relatively new. Davidson famously held that only a belief can justify another belief. Quine didn’t talk about perceptual experience at all in discussing belief formation—only about ‘sensory stimulation’.7 Classical foundationalists shared Quine’s (and perhaps Davidson’s) impoverished construal of perceptual experience on which they did not purport to represent conditions in the external world, and took the main challenge for epistemology to be to explain how the transition from introspective beliefs selfascribing such impoverished ‘sensory inputs’ to beliefs about the external world could be rational. Despite their differences, both older and newer construals of perceptual experience and its epistemic role agree that experiences (variously called ‘sense-data’, ‘havings of sense-data’, ‘conscious sensory inputs’, ‘percepts’, or ‘perceptual experiences’) are the kinds of states that cannot be formed rationally or irrationally, relative to the and much less disastrous than, the epistemological consequences of the cognitive penetration of perceptual content itself ’. Machery (Ch. ) writes: ‘philosophers are largely concerned with the cognitive penetrability hypothesis because it seems to deprive perceptual experience of its distinctive role in the justification of beliefs. But, since it is dubious that degraded and ambiguous perceptual experience has any such role, the influence of beliefs, desires, emotions, etc., on this kind of experience is of little philosophical relevance.’ Lowe (Ch. ) and Pagondiotis (Ch. ) focus on putative threats to realism from cognitive penetrability. 7 Davidson (). Quine () also worked with notions of experience, observation, and evidence that were much less impoverished, but never assimilated sensory stimulation to any of these.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

same epistemic norms that we apply to beliefs when we talk about whether they are epistemically well formed or epistemically badly formed. Davidson, Quine, and the classical foundationalists would all agree with their contemporary detractors that we can’t reason our way to experiences, the way we can reason our way to belief. A fortiori, we can’t reason our way well to experiences, nor can we reason our way to them epistemically badly. The standard examples of cognitive penetration have been taken to call this assumption into question, even when those examples are fictional. Several contributors discuss the case of Jack and Jill: Jack believes Jill is angry at him, and his belief helps explain why, when he sees Jill, her face looks angry to him. If you saw Jill’s face, you’d see her neutral expression for what it is. Jack is having an illusion brought on by his belief. Now consider whether his visual experience of Jill’s face as angry provides him with additional rational support for believing that she is angry. Normally, when you look at someone, you can gain evidence from how they look about their mood. If Jack can’t do that in this case, due to the influence of his belief on his experience, then it seems that the epistemic power of his perceptual experience is reduced by one of its causes. A second fictional example is an oversimplified form of influence on color experience by ‘memory color’. A gray banana looks yellow, due to your belief that bananas are yellow.8 That belief prevents you from seeing the grayness of the banana for what it is, just as Jack’s belief prevents him from seeing Jill’s neutral expression for what it is. We can then ask whether your yellow-banana experience provides as much rational support as it might otherwise do for believing that (i)

the banana you see is yellow,

or for increasing your confidence in the generalization that (ii) bananas are yellow. Here too, many contributors assume that if memory color operated through the influence of a belief that bananas are yellow on color experience, the epistemic power of color experience would be reduced with respect to one or both of these propositions. The assumptions listed so far are directed at the epistemic power of perceptual experiences to support beliefs that helped produce the experiences. But parallel questions arise for fears and desires. Would Jack’s fear that Jill is angry be confirmed by experiencing her as angry, if his fear helped produce her experience? If you hoped that bananas are yellow and this hope generated a yellow experience upon seeing a 8 The central experimental papers are Hansen et al. (), Olkkonen et al. (), and Witzel and Gegenfurtner (). It is in dispute whether the influencing state is a belief, and if so, what its content is (‘Bananas are yellow’ or ‘Banana-shaped and textured things are yellow’.) The example in the text is fictional because it stipulates both that the influencing state is a belief and that the subject experiences the banana as yellow, whereas in the real case it is experienced as yellowish (though see Zeimbekis  for an argument that it is simply experienced as gray, with no effect on color experience at all. The phenomenon is also discussed by Brogaard and Gatzia (forthcoming) and Macpherson ().)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



gray banana, would that influence detract from the experience’s epistemic power to support the belief that the banana is yellow, or that bananas are yellow? These questions probe whether experiences can lose any of their epistemic power by virtue of influences from doxastic states, conative states, or fear. If they can, then the perceptual experience is itself epistemically evaluable, in the following sense: it depends for its formation on other psychological states, and that dependence impacts its power to provide rational support for believing other propositions. (I return shortly to different ways in which the epistemic power of experience may be impacted.) None of the contributors contest the substantive epistemological thesis that the experiences in at least some of these scenarios would lose some (or all) of their rational support they could otherwise offer for certain propositions.9 But many have epistemologically deflationary aims of a different sort. They assume that cognitive penetration has epistemic effects like these, and then they argue that the experimental data most often put forward as examples of cognitive penetrability are at best evidence of ‘lateral’ or ‘intraperceptual’ effects, effects mediated by attention or affect, or simply effects on perceptual judgment, where this is downstream from perceptual experience.10 Memory color, multisensory integration, the Levin–Banaji faces, and various results from ‘New New Look’ experiments about wishful seeing are redescribed by several contributors as falling squarely on the perceptual side of the distinction between perception and cognition. They conclude that the effects could not have any special impact on the epistemic power of experiences, on the grounds that entirely intraperceptual processes are a-rational. Even if preconscious processing involves unconscious inference, such as Bayesian reasoning, it is at best as-if rational or as-if irrational, due to the fact that it operates entirely at the level of perception. Such inferences are not subject to the epistemic norms we apply to persons, when we say that the person is rational or irrational, by virtue of their psychological 9 The contributors don’t contest this, but it is contested by e.g. Huemer (), Tucker (), Fumerton (), and Pryor (), and defended by Jenkin (forthcoming), Lyons (), McGrath (; ), Siegel (; a), Teng (forthcoming), and Vance (). Even those who deny that cognitive penetration by itself could reduce the epistemic power of experience agree that if the subject learned that her experience had been produced in those ways, she would have a defeater that undercut the experience as a source of rational support. The issue is whether any psychological influences on perception by themselves, absent awareness of them, could have this effect. 10 See the contributions by Briscoe (Ch. ), Machery (Ch. ), Deroy (Ch. ), Dokic and Martin (Ch. ), and Zeimbekis (Ch. ), as well as Zeimbekis (), and Brogaard and Gatzia (forthcoming). Following Deroy (), many of these writers classify the stored generalization in the case of memory color as perceptual on the grounds that the measurable effect is weaker for line drawings and stronger for realistic pictures. A banana will continue to look yellowish even after its hue is adjusted far past gray, but a line drawing of a banana will not be adjusted as far past gray as a realistic picture of a banana. Categorizing the item as a banana thus seems not to determine how strong the effect is, since the perceiver recognizes the item as a banana both times. The difference in strength of effect seems to come from factors that differ between the line drawing and the realistic picture, such as texture and shape information. But which properties a state represents does not settle whether the state itself belongs to perception or to belief (cognition). We know and hence believe that banana-shaped and textured things tend to be yellow, alongside our knowledge that bananas are yellow. (A different and more powerful ground for this classification is the relative strength of the effect with ‘daylight’ colors yellow and blue, and relative weakness for red and green. Why the effects are uneven in this way remains to be explained.)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

processes being epistemically well-formed and maintained, or epistemically ill-formed and maintained. (For example, a person is irrational to the extent that they hold an epistemically ill-formed belief.)11 According to the deflationary idea that several contributors express, no inferences leading up the putative cases of cognitive penetration they discuss are subject to these epistemic norms, however redolent those inferences may be to ones that are.12 When is a route to perceptual experiences epistemically evaluable? We are used to asking how reliable or truth-conducive our perceptual experiences may be. Reliabilists about epistemic justification take the answer to bear on the epistemic power of those experiences, whereas non-reliabilists don’t.13 Even fictional cases of cognitive penetration like the two we started with (Jack and Jill, fictionalized memory color) open the possibility that some routes to perceptual experiences might be epistemically evaluable. The kind of epistemic evaluability I introduced earlier is independent of reliabilism. If a route to experience is epistemically evaluable, then due to the relationship between the experience and psychological states that help generate it, its epistemic power is impacted. How could it be impacted? (Here I pick up the question left behind earlier.) In three ways: it either has less power to support certain beliefs than they would absent those precursors;14 or it has more power to do so; or it has the usual amount, but has it in part by virtue of its relationship with those psychological precursors. Most discussions of putative cases of cognitive penetration focus on the reduction of epistemic power. The fact (if it is a fact) that there could in principle be any impact on the epistemic powers of experience due to influences by what you believe, suspect, want, know, or fear suggests that there may be such a thing as a epistemically evaluable route to perceptual experience. The philosophical problem is then to identify which routes these would be. If Jack’s beliefs or fears could influence his anger-experience of Jill in a way that reduces its epistemic power, why couldn’t his stored representations that bananas are yellow reduce the epistemic power of the yellow-banana experience?15 Which features of a route to experience could make it epistemically evaluable?

11 Note that it would not be a foregone conclusion from the fact that a process is person-level (a process the person undergoes, as opposed to part of the person) that it is epistemically evaluable. If I blink my eyelids rapidly, knowing that the blinking will make me dizzy, this route to my dizziness is not thereby subject to person-level epistemic norms. For more on the distinction between personal and sub-personal processes and explanations, see Drayson (). 12 Deroy (Ch. ) may have this idea in mind when she describes perceptual representations as ‘nonrational’: ‘The effect of congruence on multisensory integration observed in the case of the kettles might then also relate to some such non-conceptual, opaque, and non-rational representations of congruence, rather than deriving from our beliefs or knowledge about objects.’ Similar ideas are expressed by Brogaard and Gatzia (forthcoming). 13 On reliabilism, see Goldman and McGrath (forthcoming) and Goldman (). 14 We can think of reduction as reduction below a baseline, relative to which we normally have pretty good reason to believe our eyes. 15 I thank Zoe Jenkin for pressing this question, and for many illuminating discussions about how to answer it. She addresses the question in her paper ‘Perceptual expectation and epistemic downgrade’.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



Some contributors rely on a distinction between perception and cognition to draw the line dividing the epistemically evaluable routes to experience from the a-rational ones. But many distinctions do business under the label ‘perception versus cognition’. These include: the distinction between iconic and propositional format; early vision versus visual cognition; conceptual versus non-conceptual content; representations that depend on current stimuli versus those that don’t; representations internal to sensory modalities versus those external to them.16 The elements of all these distinctions are themselves underspecified. Even if we could regiment the perception/cognition distinction by sharpening and selecting one of the many distinctions that go under that label, further principles would still be needed to explain why the processes that are purely perceptual in the selected sense fall outside the domain the epistemic norms. A more direct approach would probe directly which routes to perceptual experience are epistemically evaluable. How can we probe this question? I outline three approaches. I illustrate each one using the case of memory color, and discuss whether the effect on color experience from memory color would be epistemically evaluable, even it results from processes that many contributors classify as perceptual rather than cognitive. My discussion is not meant to establish that influences by memory color on color experience are epistemically evaluable.17 But it highlights the kinds of considerations that help answer the larger question of which routes to perceptual experience are epistemically evaluable, without relying on an independent classification of an influencing state as either cognitive or perceptual.

Approach : Find the limits from below Let us take it as a fixed point that there limits from below on which routes to perceptual experience are epistemically evaluable: not every route to perceptual experience is epistemically evaluable. When a route to perceptual experience is not epistemically evaluable, we can say that it is a-rational.18 To find the limits from below on which routes to experience are epistemically evaluable would be to identify in general terms the a-rational routes to experience. Routes to experience that fall outside those limits would then be decent candidates for being epistemically evaluable. To execute this approach, one can consider examples of a-rational psychological processes in 16 See Fodor () and Carey () on iconic vs. propositional format; Evans () on conceptual vs. non-conceptual content; Pylyshyn (), Raftopoulos (), Brogaard and Gatzia (forthcoming) on early vision; Beck (forthcoming) on stimulus-dependent vs. stimulus-independent representations. 17 If memory color had epistemically evaluable effects on color experience, there are multiple effects these could be. The option most often discussed is reduction of epistemic power of experience, relative either to the content of the color experience (‘x is yellow’) or the content of influencing state (‘banana-shaped and textured things are yellow’). Other options echo the ones listed earlier: increase the power of experience to epistemically support its contents or other propositions, relative to a baseline; or simply change the factors by virtue of which the experience provides baseline amount of justification. 18 What is it for a state S or process P to be a-rational? At a minimum, there is no respect in which her being in S or undergoing P constitutes her being rational, and no respect in which it constitutes her being irrational.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

which psychological states influence perceptual experience, focus on representations involved in those routes to experience, and try to identify which features of those representations, or of the process linking them to experience, makes that route to experience a-rational.19 Some contributors suggest that if putative cases of cognitive penetration are due chiefly to the influence of a representation that is ‘implicit’,20 then it has no epistemically significant effects on perceptual experience. To make contact with this suggestion, we can start with some examples of uncontroversially a-rational processes containing uncontroversially implicit representations that encode transitions from one representation to another. So let us take it as another fixed point that a-rational processes include: the inference from contrasts to edges, the inference from certain edges to depth representations, and the inference from certain spatiotemporal cues to representations of causation.21 How can we generalize beyond these examples to assess whether other representations are epistemically evaluable, such as the ones involved in memory color that may go on to influence perceptual experiences? I’ll consider several features of them and argue that either they don’t apply to memory color representations, or they are poor candidates for illuminating what makes a process a-rational, or both. A first suggestion is that the paradigmatically a-rational inferences encode information in a different format than the belief that bananas are yellow. This feature is a poor candidate for making a process a-rational. If perceptual experiences or imaginative states of yellow oblong shapes had an iconic format, then this format would presumably not preclude the experience from providing reason to believe that something yellow and oblong was nearby, or the imagination from providing reason to believe that yellow things can be oblong. A second suggestion is that the implicit representations cannot be formulated by the subject at all, or else cannot be formulated without significant investigation. This feature applies as well to the heuristics uncovered by Tversky and Kahneman, and to implicit assumptions that guide our reasoning but that take a lot of reflection or

19 A different approach would aim to identify in general terms what can make a state a-rational, and then see if those general terms would count any representations involved in memory color as a-rational. For instance, a potentially sufficient condition for being an a-rational representational state is that its formation is not due to any of the properties it represents, and it cannot develop or be adjusted in any way that would make it sensitive to those properties. This proposal would not count any memory color representations as a-rational, on the assumption that those representations are learned from patterns of exposure to the properties they represent. Thanks to Eric Mandelbaum for discussion. 20 The term ‘implicit representation’ can be confusing if one thinks of implicit states as rules or transitions from one representation to another, such as the modus ponens inference rule, or the rule for computing edges from contrasts. For purposes of discussion, we can think of these states as implicitly representing modus ponens, or the conditional that if there are contrasts meeting certain specifications, then there are edges that meet certain other specifications. To say that they are implicit marks their functional difference from the representations that they relate. Shea () offers a useful discussion of implicit representation. 21 Another example of an implicit representation: the syntactic rules that allow us to discern whether a string of words is an English sentence.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



experimental investigation to unearth, such as the kinds we often say were ‘not on the radar’ in studying scientific theories from the past.22 Since both implicit assumptions and formation of beliefs using heuristics are epistemically evaluable, it is not in general true that an inability to formulate the content of an implicit representation (or unawareness that one has a representation) make the reliance on an implicit representation a-rational. A third suggestion is that implicit representations that encode transitions from contrasts to edges are representations we can have without having any concepts of the properties contrast or edge. (A parallel observation holds for the other examples of implicit representations we started with.) If lacking such concepts made this process arational, then some other explanation would be needed of why they remain a-rational once we gain concepts of those properties. And in the case of memory color, most of us have concepts of the color yellow, and of the shapes and textures characteristic of bananas (smooth in parts, stubby at the stem, etc). A fourth suggestion is that the transitions in our paradigms of a-rational processes are immalleable by any other psychological process. Such immalleability arguably could help make a psychological state a-rational. This strategy may be a promising way to find the limit from below, but it does not apply to memory color, so long as the links between shapes, textures, and colors could change by some of the same processes that established them in the first place, such as patterns of exposure. A last suggestion is that the implicit representations have no function in the mind other than to execute the transitions that they encode. A consequence of this limited functional role is that the implicitly represented information is unavailable for use in a wide range of inferences. If its limited inferential availability is part of what makes reliance on implicit representations a-rational, then our question should be: how much inferential availability, and what kind, would the key influencing state need to have, for its role in producing perceptual experience to be rationally evaluable? It is implausible to require global inferential availability in order for it to be an ingredient in a epistemically evaluable process. Think of compartmentalized beliefs, which are not available for a full range of processing. Suppose you forget that you have an appointment with X at noon when you make an appointment with Y at noon. The partial inaccessibility of your belief that you will meet X at noon does not stop it from participating in epistemically evaluable inferences. For instance, upon finding X’s umbrella you might plan on giving it back to her when you see her at noon, without yet noticing your conflicting appointments. The epistemic status of the resulting belief that X will get her umbrella back seems influenced by its reliance on the belief that you are

22 Consider some medieval alchemists concept of weight, which would allow a few pounds of lead to be turned into hundreds of pounds of gold. Presumably they could not easily formulate the assumption that weight was not an extensive magnitude. For discussion of contrasting conceptual schemes related to weight, see Carey (: Ch. ).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

meeting X at noon.23 Partial inaccessibility is no bar to participating in epistemically evaluable routes to belief. So it is not true in general that inaccessibility of a psychological state prevents it from impacting the epistemic status of other states that depend on it. If limited inferential availability makes reliance on implicit representation arational, then in probing whether any effect of memory color on experience is epistemically evaluable, we should consider how wide the range of circumstances is in which the generalization about color is activated. Call shapes and texture characteristic of bananas ‘B-shapes’ for short, and suppose that memory color effects results from an intra-perceptual generalization that B-shapes tend to be yellow. What activates this generalization? Can it be activated by an imagination or hallucination of B-shapes, as well as by perception of them? As the range widens, it looks more like a case of compartmentalized belief than the highly circumscribed syntactic information, or a mere transition from one state to another. These suggestions do not yet illuminate why there are limits from below on the epistemic evaluability of perceptual experience. But they put pressure on excluding the influence of memory color representations on perceptual experience from the domain of epistemic norms.

Approach : The explanatory approach How else might we probe which routes to perceptual experience are epistemically evaluable? A second approach starts by asking whether there is anything that could be explained by the hypothesis that some routes to perceptual experience are epistemically evaluable. An explanatory role for this hypothesis would give us some reason to believe it. Any cases in which the hypothesis plays an explanatory role will contain factors that make the route to experience epistemically evaluable. We can then try to figure out which factors these are, drawing on the specific character of the case. A straightforward application is the case of Jack and Jill. The main ingredients of the argument are these: Bootstrapping intuition: Jack’s experience that Jill is angry does not give him additional rational support for believing that Jill is angry. The Inferential Hypothesis: The Bootstrapping intuition is best explained by the inference-like relation of rational dependence that the experience stands in to the antecedent beliefs.

23 What is the epistemic status of your belief that X will get her umbrella back (the ‘umbrella-belief ’)? Its epistemic status seems obviously influenced by your belief that you’ll meet X at noon (the ‘meeting-X belief ’). On one view, the umbrella-belief is well-founded by the meeting-X belief, together with your belief that you’ll return X’s umbrella then. A different option is that the umbrella-belief is ill-founded, because your belief that you’ll meet X is noon is made irrational by your belief that you’ll meet Y at noon, and that you can’t meet them both at once. On either option, a belief with limited accessibility (the meeting-X-belief) influences the epistemic status of another belief. The options differ with respect to whether a prohibition against inconsistency governs relationships between beliefs in different compartments.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



The Inferential Hypothesis entails the Epistemic Evaluability hypothesis: Epistemic Evaluability Hypothesis: Some routes to perceptual experience are epistemically evaluable.

An explanatory approach makes the case that the Bootstrapping intuition is best explained by the Inferential Hypothesis. How could the Epistemic Evaluability Hypothesis explain the Bootstrapping intuition? A natural idea is that the intuition is explained by illicitly circular structure that would emerge if Jack strengthened his belief in response to his experience. More argument is needed to show that this is the best explanation. But it is a natural candidate. If it is a strong candidate, then we have likely identified a structural relationship between experience and its psychological precursors that makes the experience epistemically evaluable. We can then ask in what other domains the same structure might be found. In particular, we can ask what it would take for such a structure to be found in the case of memory color. Could there be analogs in that case of the Bootstrapping intuition? More generally, is there anything for the Epistemic Evaluability Hypothesis to explain about memory color? To address these questions, let us examine more closely the putatitve illicitly circular structure that many find in the Jack and Jill case. When one and the same state both generates an experience and is psychologically strengthened by it, this structure is a good candidate for being irrational. In the case of Jack and Jill, Jack’s belief that Jill is angry plays two roles: it helps generate an experience in which Jill looks angry to Jack, and it is strengthened by that very experience. It is strengthened in the sense that Jack increases his confidence that Jill is angry on the basis of the experience. What about memory color? First, suppose that one strengthens one’s confidence (a form of belief) that B-shaped things tend to be yellow, on the basis of an experience that X is B-shaped and yellow, and suppose that a generalization that B-shapes are yellow helped generate that very experience. Finally, suppose that the generalization takes the form of something other than a belief. If the state that causes the experience is not the state that is strengthened by it, then the strengthening is a poor candidate for irrational circularity.24 This situation is the one many contributors think we are in. According to them, we have a ‘merely perceptual’ representation that B-shapes are yellow (or a ‘merely perceptual’ association between B-shapes and yellow), that representation helps produce a yellowish experiences, and those yellowish experiences are perfectly poised to rationally strengthen beliefs about the color of the banana. By contrast, in the Jack and Jill case, the threat of illicit circularity arises, in part because one and the same state generates the experience and is strengthened by it, if 24 Why is it a poor candidate for irrational circularity, if the state that generates the experience and the state it strengthens have overlapping accuracy conditions (e.g. both have accuracy condition ‘B-shapes are yellow’)? It is a poor candidate because sharing accuracy conditions is not enough to make a transition irrational, e.g. when one endorses one’s perceptual experience, one forms a belief with a content that is also a content of one’s experience, but endorsement is not thereby irrational.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

Jack strengthens his belief on the basis of his experience. For there to be forms of illicit intraperceptual circularity in which the generalization helps produce the experience, there would have to be an analogous two-part structure. First, there would have to be a way for the experience to psychologically strengthen the generalization, consistent with its belonging to a perceptual system. The strengthening might consist in a stronger ‘yellowification’ signal in response to B-shapes, or a wider range of circumstances in which B-shapes produce a yellowification signal, or a wider range of circumstances in which both B-shapes and yellow signals are sent.25 Second, the strengthening by experience would have to be epistemically evaluable. Here we are asking whether an intraperceptual generalization can be rationally strengthened by experience, as a way to assess whether a route to experience from the generalization is epistemically evaluable. Supposing that the generalization in memory color is intraperceptual, can it be either rationally or irrationally strengthened by experience? Here there is a major disanalogy with the case of Jack and Jill, where the circular structure includes Jack’s belief. Jack’s belief is the kind of state that can be strengthened rationally or irrationally by experience. In contrast, whether generalizations that aren’t beliefs can be strengthened rationally by experience is up for grabs. So long as this is the case, we don’t have an analogy to the Bootstrapping intuition about Jack’s belief. Since the Bootstrapping intuition was the thing to be explained in the explanatory strategy outlined earlier, that strategy has no straightforward extension to intraperceptual cases. The explanatory approach could be extended in this way only if there were grounds for thinking that the generalization can be strengthened rationally or irrationally by experience. Perhaps such grounds could be leveraged into an argument for the potential epistemic evaluability of circular intraperceptual structures.26

Approach : The Defeat Model The circular structure invoked by the Explanatory approach is not the only structure that is a candidate for an epistemically evaluable intraperceptual process. A different structure involves overriding unconscious perceptions that offer rational support for believing their contents. To fix ideas, consider a structurally analogous route to belief. You know that X is behind a curtain and you can’t see X, but you know that X is B-shaped. You’ve got some reason to think X is yellow, but it doesn’t come from perceiving X—it comes 25 If in addition there was an unconscious representation of a gray banana, it might weaken the association, in a way worked against the strengthening by the yellowish experience. If the unconscious perception is a distribution of probabilities over colors, then the increment of strengthening might depend on the relative strength of the ‘yellow’ possibilities. 26 In ‘Perceptual expectations and epistemic downgrade’, Jenkin develops in detail the idea that there can be epistemically illicit intraperceptual circularity.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



Table A Inference overrides unconscious perception Inference Stored representation Current representation

B-shapes tend to be yellow X is a B-shape

Experience

X is yellow

Unconscious perception

X is gray

from your knowledge that B-shaped things tend to be yellow. Then the curtain is lifted and you see X. X is gray, and X looks gray. You’ve got no reason to believe anything is abnormal, and your experience is caused primarily by X in the usual manner. There will be some cases with this structure in which it will be rational for you to update your antecedent belief and form the belief that X is gray. Normally, when it is rational to update one’s belief in this way, the information provided by one’s perception of X outweighs the information one has by virtue of one’s relationship to the premises of the inference. And normally, in this situation it is irrational to be guided by the premises of the inference instead of by the perception. Could any memory color effect on perceptual experience fit this model? An analogous route to experience might take a form illustrated by this dialogue between the visual system and a banana: VS: What color are you? X: I’m gray. VS: But you’re B-shaped. You must be yellow. Here, the stored information linking B-shaped things to yellow intervenes before the experience, but after the representation of gray that is triggered by the gray banana. An inference from the generalization generates an experience overriding an unconscious perception that x is gray (see Table A). Continuing the analogy with the curtain case, psychologically, the intraperceptual generalization that B-shaped things tend to be yellow would take the place of knowledge of that generalization, and unconscious perceptions that X is B-shaped and that X is gray would take the place of knowing that X is B-shaped and seeing X once the curtain is lifted. Could this process be epistemically evaluable? Here is an argument that it could.

Argument from defeat of unconscious perception P. An unconscious perception that represents that x is F can have epistemic power to support believing that x is F. P. If an unconscious perception that represents that x is F can have epistemic power to support believing that x is F, then it can be irrational for it to be overridden in a route to a perceptual experience with the content: x is F.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

P. If a process irrationally overrides an unconscious perception, then the process is epistemically evaluable. Conclusion: Some routes to perceptual experience are epistemically evaluable. Premise  is a natural elaboration of the idea of a epistemically evaluable process. The key premises are P and P. Defenses of them might start from the following considerations. On premise P: The case of conflicting appointments discussed earlier illustrates a kind of state that is not entirely accessible to subsequent reasoning and processing, but that has rational power to support subsequent beliefs all the same. We can distinguish a psychological aspect of these beliefs from an epistemic aspect. Psychologically, their availability for inference is limited. Epistemically, they are epistemic resources that subject fails to make full use of. Even when they aren’t accessed for inference (as they would be if you realized that you had conflicting appointments), they retain the features that ground their epistemic force once they are accessed. Are compartmentalized beliefs the only states with this pair of related psychological and epistemic features? Arguably, they belong to a wider category that includes perceptual experiences with a low degree of attentiveness. Just as each compartmentalized belief about your appointment can be made available for reasoning through recall, so too inattentive experiences can be made available through shifts of attention. Just as the compartmentalized belief retains its epistemic force when unaccessed, arguably the inattentive experiences do too.27 Unconscious perceptions seem to fit the same pattern, when the transition to phenomenal consciousness leaves all their other features intact. Once it is phenomenally conscious, it will belong to the subject’s epistemic resources (assuming it is not defeated or otherwise downgraded with respect to its epistemic power). Just as the inattentive experience could easily become attentive, so too the unconscious perception could easily become phenomenally conscious. In both cases, the result of the transition clearly belongs to the subject’s epistemic resources. Given that only a minimal transition is needed to cross the threshold into the realm of epistemic resources, such states may be epistemic resources all along, and compartmentalization, inattentiveness, or unconsciousness are merely ways of limiting our access to them. In the face of these similarities, principled grounds are needed for excluding unconscious perceptions from the domain of unused epistemic resources. These considerations are far from decisive, but they suggest a case that might be made for P. Premise P purports to take us from the epistemic power of certain unconscious perceptions to the epistemic evaluability of processes that override them. Suppose it is granted that an unconscious perception can provide rational support for believing its contents. Then consider a process that could take account of that support, but doesn’t.

27 For discussion of inattentive experiences and their epistemic status, see Siegel and Silins (). I thank Nico Silins for discussions of unused epistemic resources.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



Such a process seems analogous to a process that ignores evidence that one has. If ignoring evidence is irrational, then the process that leads to this result is irrationally outweighing the evidence, and hence must be the kind of process that is epistemically evaluable. I’ve considered three approaches to discovering what belongs in the domain of epistemic norms. Each approach suggests that some routes to perceptual experience are epistemically evaluable, in the sense discussed earlier: the perceptual experience depends for its formation on other psychological states, and that dependence impacts its power to provide rational support for believing other propositions. One might try to develop any of the approaches to argue that the epistemic evaluability of perceptual experience extends beyond canonical cases of cognitive penetration. All three approaches give us tools to analyze how far into the mind epistemic norms extend. I’ve argued so far that the epistemic significance of psychological influences on perception is not necessarily limited to canonical cases of cognitive penetration. Next I turn to the epistemological significance of an even broader family of effects to which canonical cases of cognitive penetration belong.

 Varieties of Top-Down Effects on Perception The family of phenomena falling under the general rubric of cognitive penetration belongs to an even wider class of potential influences on perceptual experience and judgment. We can distinguish myriad ways in which both perceptual experience and perceptual judgment could be influenced by other psychological states we are in. Studying the broader family of effects may give us tools to analyze how social forces may operate through perception. To see how the broader family of effects applies to understanding of social phenomena, consider Keith Payne’s disturbing experiment in which participants more often misclassify a tool (pliers, wrench, or a drill) as a gun when primed with pictures of Black men, compared with subjects who have been primed with pictures of White men.28 From this result, we know that whatever psychological state the prime puts the subjects in, it influences their answers on the classification task. How is the visual experience of subjects in Payne’s experiment affected? How do the pliers look to the subjects, when they see them? The experiment does not speak to this question, and the results could be explained in a range of ways: • Disbelief: The pliers look to the subject exactly like pliers. Classification errors are driven by something that makes the subjects disbelieve their perceptual experience. 28 Payne (). The subjects are non-Black American college students, though similar effects are found for Black American college students.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

• Bypass: The pliers look to the subject exactly like pliers. Classification errors do not result from responses of any kind to the experience—not even disbelief. • Haste: The pliers do not look like pliers or like a gun, but subjects jump to the conclusion that it’s a gun, before perceiving enough detail to decide the matter on the basis of what they see. • Introspective error: The pliers look to the subject exactly like pliers. But they make an introspective error in which they take themselves to experience a gun. • Automatic, disowned behavior: The pliers looked to the subject exactly like pliers, but the state guides behavior that subjects immediately afterward will on reflection regard as mistaken. • Cognitive penetration: The pliers look to the subject like a gun, due to the state activated by the Black prime (where by stipulation the state is activated by the Black prime counts as cognitive). The first three options (Disbelief, Bypass, Haste) are effects on perceptual judgment. The first two (Disbelief, Bypass) along with Introspective Error impact the role of experience in forming judgment, rather than the contents of experience. The third (Haste) leaves open whether the prime impacts the extent of perceptual processing as well as the response to it in judgment. There could be hasty judgment, hasty perceptual experience, or both. The last option (cognitive penetration) relies on a distinction between extraperceptual or intraperceptual influencers, which, as we saw earlier, remains fluid absent further regimentation. The rest of the options are insensitive to how the influencing state stands relative to the domain of perception.29 Most discussion of epistemological impact of psychological influences on perception has focused on canonical (even if fictional) cases of cognitive penetration. But if we start from the assumption that social forces influence perception without our awareness, and our question is how they do so, drawing these distinctions gives us a place to start. In all of these cases, a perceiver ends up either perceptually experiencing what she already suspects or fears to be the case, or forming beliefs on the basis of perception that confirm her suspicions or fear. We might say that they are all cases of perceptual farce. The farce is that perception seems to open our minds to the things around us, but doesn’t. It purports to tell us what the world is like, so that if need be, we can check our beliefs, fears, and suspicions against reality and can use it to guide our actions—but it doesn’t.30 As the distinctions drawn above show, perceptual farce is not specific to even the most canonical cases of cognitive penetrability. It can operate through influences by one’s own outlook on perceptual judgment, or by neutralizing

29 For discussion of the cognitive underpinnings of implicit bias, and of the type of state that the Black prime puts participants in, see Levy (forthcoming), who argues that it isn’t a belief, Mandelbaum (forthcoming), who argues that it is a belief, and Brownstein (forthcoming), who reviews a wide range of options. 30 By this characterization of perceptual farce, any case of illusion, no matter how it was generated, would be a case of perceptual farce. So some cases of it will be more interesting than others from the point of view of analyzing the impact of social configurations on perception.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



the role of experience in guiding those judgments, or by selecting which stimuli will be experienced in the first place—leaving perceptual experience itself perfectly faithful to the external things that it helps us perceive. Perceptual farce also encompasses systematic effects on attention. Any pattern of attention will include some stimuli and exclude others. Not all such patterns will give a misleading impression of openness to the world. But some seem to do so clearly. Consider outgroup hiring. Here is a domain in which the main task of inquiry is to respond to new information one gets from the applicants’ dossiers. Information in the dossier, we can suppose, comes with a level of detail that provides the kind of evidence that can outweigh antecedent generalizations. By comparison, you might have excellent evidence that on the whole, -year-olds are poor drivers. Upon getting to know a specific -year-old, you will be better placed to assess whether this generalization is true of her. Similarly, in the case of outgroup hiring, readers bring to the process their generalizations about outgroup applications, poised to use them as they would use any background information in assessing new information. Just as you might come across a -year-old who is a good driver, contrary to your antecedent assumption about -year-olds, so too you might come across an outgroup applicant who is stronger than antecedent assumptions would predict. If social forces, or the affective profile that goes with them, influenced the process of information uptake by putting a halt to processing of information that is at odds with the generalizations, then specific information that should modulate generalizations in the face of new information would not get a chance to play that role. When the generalizations (unlike the one about young drivers) are unjustified, as it is in the many real-world cases of bias in outgroup hiring, a poor epistemic situation will be perpetuated. This scenario gives us a model for formulating hypotheses about perception of other people. In the domain of social perception, we can ask whether features that are congruent with antecedent social assumptions or with relative social positions are systematically selected for experience, whether incongruent features are systematically anti-selected, or whether patterns of attention are entirely independent of relative social positions of the perceivers. If we suspect that social positions may be reflected in perceptual situations, this is a useful hypothesis to try and test. An application of these ideas may be found in the domain of gaze-following. Ingroup participants more readily follow gaze of ingroup members than of outgroup members, whereas outgroup participants follow gaze of both ingroup and outgroup members.31 To the extent that gaze-following indicates confidence that the followed person’s object of attention or experience of it is epistemically valuable, it is reasonable to hypothesize that this result reflects an underlying pattern of social valuation. This kind of selection effect shapes our epistemic situation. In general, selection effects that embody confirmation bias at the level of perception will be cases of perceptual farce.32 31 32

Adams and Kveraga (forthcoming). I discuss this kind of confirmation bias and the rational role of selection effects in Siegel (b).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

The many modes of perceptual farce are useful for analyzing any domain that is ridden with resistance to taking in evidence that is available from perception. Payne’s experiment and others like it show that contemporary forms of racism are fueled by this sort of resistance.33 And in philosophy, a systematic account of why we should expect blindness to perceptually available facts is developed by Iris Murdoch.34 Why, if people are presented with such specific information, does it not influence their decisions and their behavior? What role is the information playing, if it isn’t guiding belief and behavior? It is no surprise that people sometimes respond poorly to evidence when they deliberate. But when the new specific information comes from perception, some explanation is needed of why the information in these situations has so little impact on what subjects go on to believe or do, when in other cases it has so much impact. The distinction between cognitive penetration and other forms of perceptual farce is important for the psychology of perception, and for understanding the architecture of the mind. But whether cognitive penetrability occurs or not is less important for understanding the role of perceptual experience in shaping and sustaining our outlook on the world, especially in the domain of social perception. Perceptual farce in the form of standard visual illusions has long fueled discussions in perceptual epistemology. When perceptual farce reflects social forces that would be better eradicated, or when it reflects moral limitations in other ways, it poses a host of normative questions about the epistemic role of judgments and perceptual behavior that they belong to. The epistemic questions surrounding putative cases of cognitive penetrability are just the tip of the iceberg.

References Adams, R., and Kveraga, K. (forthcoming). Social vision: functional forecasting and the integration of compound social cues. Review of Philosophy and Psychology, special issue on cognitive penetrability, ed. Z. Jenkin and S. Siegel. Beck, J. (forthcoming). Marking the perception/cognition boundary. MS. Brogaard, B., and Gatzia, D. (forthcoming). Is color experience cognitively penetrable? Topics in Cognitive Science. Brownstein, M. (). Implicit bias. In Stanford Encyclopedia of Philosophy, Spring Edition, ed. E. N. Zalta. Online . Carey, S. (). The Origin of Concepts. Oxford: Oxford University Press. Correll, J., Park, B., Judd, C. M., Wittenbrink, B., Sadler, M. S., and Keesee, T. (). Across the thin blue line: police officers and racial bias in the decision to shoot. Journal of Personality and Social Psychology : –. Davidson, D. (). A coherence theory of truth and knowledge. In E. LePore (ed.), Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson, –. Oxford: Blackwell. 33 34

See Dovidio and Gaertner (), Correll et al. (), also Goldin and Rouse (). For discussion, see Katsafanas (forthcoming) and Manne (forthcoming).

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

afterword



Deroy, O. (). Object sensitivity versus cognitive penetrability of perception. Philosophical Studies. Online: doi: ./s---. Dovidio, J., and Gaertner, S. L. (). Aversive racism and selection decisions:  and . Psychological Science (): –. Drayson, Z. (). The uses and abuses of the personal/subpersonal distinction. Philosophical Perspectives (): –. Evans, G. (). The Varieties of Reference. Oxford: Oxford University Press. Fodor, J. (). Modularity of Mind. Cambridge, Mass.: MIT Press. Fumerton, R. (). Siegel on the epistemic impact of ‘checkered’ experience. Philosophical Studies : –. Goldin, C., and Rouse, C. (). Orchestrating impartiality: the impact of ‘blind’ auditions on female musicians. Working Paper  (Jan.), National Bureau of Economic Research. Goldman, A. (). Reliabilism. Stanford Encyclopedia of Philosophy (Spring ), ed. E. N. Zalta: Goldman, A., and McGrath, M. (forthcoming). Epistemology: A Contemporary Introduction. New York: Oxford University Press. Hansen, T., Olkkonen, M., Walter, S., and Gegenfurtner, K. R. (). Memory modulates color appearance. Nature Neuroscience (): –. Huemer, M. (). Epistemological asymmetries between belief and experience. Philosophical Studies : –. Jenkin, Z. (forthcoming). Perceptual expectations and epistemic downgrade. MS. Katsafanas, P. (forthcoming). The moral significance of perceptual experience. MS. Levy, N. (forthcoming). Neither fish nor fowl: implicit attitudes as patchy endorsements. Noûs. Lyons, J. (). Circularity, reliability, and the cognitive penetration of perception. Philosphical Issues (): –. Macpherson, F. (). Cognitive penetrability of color experience: rethinking the issue in light of an indirect mechanism. Philosophy and Phenomenological Research (): –. Mandelbaum, E. (forthcoming). Attitude, association, inference: on the propositional structure of implicit attitudes. Noûs. Manne, K. (forthcoming). Let me look again. MS. McGrath, M. (). Siegel and the impact for epistemological internalism. Philosophical Studies : –. McGrath, M. (). Phenomenal conservatism and cognitive penetration: the bad basis counterexamples. In C. Tucker (ed.), Seemings and Justification, –. Oxford: Oxford University Press. Olkkonen, M., Hansen, T., and Gegenfurtner, K. R. (). Color appearance of familiar objects: effects of object shape, texture, and illumination changes. Journal of Vision (): –. Payne, K. (). Prejudice and perception: the role of automatic and controlled processes in misperceiving a weapon. Journal of Personality and Social Psychology (): –. Pryor, J. (). The skeptic and the dogmatist. Noûs : –. Pylyshyn, Z. (). Is vision continuous with cognition? The case for cognitive impenetrability of visual perception. Behavioral and Brain Sciences (): –. Quine, W. V. O. (). Word and Object. Cambridge, Mass.: MIT Press. Raftopolous, A. (). Cognition and Perception: How Do Psychology and Neuroscience Inform Perception? Cambridge, Mass.: MIT Press.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



s. siegel

Shea, N. (). Distinguishing top-down from bottom-up effects. In S. Biggs, M. Matthen, and D. Stokes (eds), Perception and its Modalities, –. Oxford: Oxford University Press. Siegel, S. (). Cognitive penetrability and perceptual justification. Noûs : –. Siegel, S. (a). The epistemic impact of the etiology of experience. Philosophical Studies (): –. Siegel, S. (b). Can selection effects on experience influence its rational role? In T. Gendler and J. Hawthorne (eds), Oxford Studies in Epistemology, iv, –. Oxford: Oxford University Press. Siegel, S., and Silins, N. (). Consciousness, attention, and justification. In D. Dodd and E. Zardini (eds), Skepticism and Perceptual Justification, –. Oxford: Oxford University Press. Teng, L. (forthcoming). Cognitive penetrability and justification: the imagination argument. MS. Tucker, C. (). If dogmatists have a problem with cognitive penetration, you do too. Dialectica (): –. Vance, J. (). Emotion and the new epistemic challenge from cognitive penetrability. Philosophical Studies (): –. Witzel, C. et al. (). Object knowledge modulates color experience. i-Perception : –. Witzel, C., and Gegenfurtner, K. R. (). Categorical sensitivity to color differences. Journal of Vision (): –. Zeimbekis, J. (). Color and cognitive penetrability. Philosophical Studies : –.

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

Author Index Aarts, H.  Abell, C.  Adams, R.  Adolphs, R.  Aglioti, S. ,  Allport, D. A. ,  Alter, A. L. ,  Anstis, S. M. – Arango-Muñoz, S.  Armstrong, D.  Armstrong, K.  Ashcraft, M. H.  Bach y Rita, P.  Balcetis, E. , , , –, , , –,  Banaji, M. , –,  Barr, M.  Barrett, H. C. ,  Barry, S. ,  Bartels, A.  Baye, T. , –, , , –, , –, – Beck, J.  Bayesian ,  Bayne, T. ,  Beets, I. A. M.  Behrmann, M. , , , , , , , ,  Bennett, D. ,  Benson, D. ,  Beran, M.  Bergeron, V. , ,  Berkeley, G. , , ,  Bermudez, J. L. , –, , , –, – Bhalla, M. , ,  Biederman, I. , ,  Binkofski, F.  Bitter, D.  Bjork, R.  Blakemore, S.  Blaser, E.  Block, N. , –, –, , –, –, –, , –, ,  Bogen, J. ,  Bolles, R. C. –,  Botvinick, M.  Brenner, E.  Brewer, W. F. ,  Briscoe, R. , , , –, , , , , –, –,  Broackes, J. , 

Broadbent, D. E. ,  Brogaard, B. , – Broota, K. D. , ,  Brown, H.  Brownstein, M.  Bruce, V.  Bruner, J. S. –, , , , , , , –, , – Buller, D. J. – Bullot, N.  Bulthoff, H, – Burge, T. , , –, –, –, – Burnston, D. , , , , , , , , ,  Burr, D. – Byrne, A. ,  Cahen, A. , –, , –, – Campbell, J. , ,  Cantlon, J.  Caramazza, A. , ,  Carey, D. P.  Carey, S. , ,  Carmel, D.  Carrasco, M. , ,  Carruthers, P. –,  Carter, L. F.  Cavanagh, P. –,  Chand, P. K. ,  Changeux, S.  Chaumon, M.  Cherry, E.  Chomsky  Churchland, P. M. , –, –, , , , , , , , , , , , , ,  Clark, A. ,  Cohen, J. , , , , –, , , , ,  Cohen, M.  Coltheart, M. ,  Conway, B.  Correll, J.  Crane, T. ,  Creem S.  Crouzet, S.  Currie, G. ,  Cussins, A.  Cutting, J. , , , , , ,  Dallas, M.  Damasio, A. R. 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



author index

Davidson, D. – Davies, M. ,  De Gelder, B.  Dehaene, S. , –, ,  De-Medonsa, M.  Delk, J. L. , , , , , , –, –, –, , , ,  Delorme, A. – Dennett, D.  Deroy, O. , –, , , , , , –, , –, –, , , , – Descartes  Desimone, R. , , , , ,  Desmurget, M.  DeWald, D.  Dienes, Z.  Ditman, T.  Doehrmann, O.  Dokic, J. , –, , , , , , –,  Dolan, J.  Dovidio, J.  Downes, S.  Drayson, Z.  Dretske, F. , , –, , –, , , , , , , , , –, , , ,  Dreyfus, H. ,  Driver, J. ,  Duhem, P.  Duncan, J. , , , ,  Dunlosky, J.  Dunning, D. , , , –, , , ,  Durgin, F. H. –, , , –, ,  Edelman, S.  Egeth, H.  Egly, R. ,  Ehrsson, H.  Ellis, H. D.  Enns, J.  Epstein, W. , ,  Erdelyi, M. H. –,  Eriksen, C. W.  Ernst, M. ,  Evans, G. , , , , ,  Evans, J.  Evans, M. A.  Fabre-Thorpe, M.  Fahle, M.  Farrer, C.  Feyerabend, P. ,  Field, G. D.  Fillenbaum, S. , –, , , , –, –, –, , , ,  Firestone, C. 

Fish, W. , , , –,  Fisher, S. C.  Fletcher, P. C.  Fodor, J. A. –, –, , –, , , , , , , –, –, –, , , –, , –, , –, , , –, , , –, –, , –, , ,  Ford, J. M.  Fortson, J.  Francis, G.  Franck, N.  Franz, V.  Frederick, S.  Freuder, E.  Frith, C.  Fumerton, R.  Furnham, A.  Fusella, V. , ,  Gaertner, S. L.  Gallagher, S.  Gallese, V.  Ganis, G.  Gao, T. , ,  Garner, W. R. – Gatzia, D. , – Gauthier, I.  Gegenfurtner, K. R. –, –, ,  Geldard, F. A.  Gerbino, W. , ,  Gibson, E. J. , , , , –,  Goldin, C.  Goldman, A. I. ,  Goldstein, B.  Goldstone, R. L. ,  Gombrich, E. , , ,  Goodale, M. A. , , , –, , –, ,  Goodman, C. C. , , , , , , –,  Granrud, C. , ,  Granzier, J.  Gregory, R.  Grill-Spector, K. –,  Haber, M.  Haber, R. N. , –,  Hansen, J.  Hansen, T. , , –, , , , , , –, ,  Hanson, N. R. , , , ,  Hardcastle, V. G.  Harnett, W.  Harper, R. S.  Harris, C.  Hatfield, G. , , ,  Haugeland, J. , 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

author index Hawley, K. ,  Haxby, J.  Heck, R. G. ,  Heilman K.  Helmholtz, H.  Henderson, J.  Henrich, J.  Heywood, C.  Hill, C.  Hochberg, J.  Hodges, J.  Hohwy, J.  Hollingworth, A.  Holmes, G.  Hommel, B.  Hopfinger, J.  Horgan, T.  Huemer, M.  Humphreys, G. , –, , ,  Hurley, S. , , , – Hutchison, J. J.  Hutton, E.  Israel, D.  Jackendoff, R. , , , ,  Jackson, C. V. , , ,  Jacoby, L.  James, G.  James, T. W.  Jastrow , , ,  Javadi, A.  Jenkin, Z. , –,  Johansson, G.  Johnson-Frey, S.  Johnson, M.  Johnston, K.  Kahneman, D. , , , , ,  Kanizsa, G. , ,  Kastner, S. ,  Katsafanas, P.  Kayser, C.  Kempen, G.  Kentridge, R.  Kim, J. ,  Kingdom, F.  Kirchner, H.  Kitcher, P.  Klatzky, R. L.  Koch, C. ,  Koechlin, E. , ,  Kohly, R. P.  Komatsu, H.  Koriat, A. –, –, , ,  Kornmueller, A. E.  Kosslyn, S. M. , , , 



Kottenhoff, H.  Kourtzi, Z.  Kravitz, D. , , , , , , , , ,  Kriegel, U. , , , , ,  Kubovy, M. , ,  Kuhn, T. S. , , , , , –, ,  Kukla, A.  Kuperberg, G.  Kurzban, R. ,  Kveraga, K.  Lakoff, G.  Lambert, B. L. ,  Lamme, V. A. F. –, –, , –, –, –, , , –,  Landis, D.  Langton, S. R. H. , ,  Lashley, K. ,  Lechich, S.  Lenggenhager, B.  Leonardo, d. ,  Leopold, D. A. – Levin, D. T. , –, ,  Levine, J.  Levinson, J.  Levy, N.  Lewis, M. B.  Li, F. F.  Li, Z.  Liu, H.  Loffler, G.  Logothetis, N. K. ,  Longuenesse, B.  Loomis, J. M.  Lopes, D.  Lotto, R. B.  Lowe, E. J. , –, , –, , , – Ludwig, V. U.  Lund, O.  Lyon D. ,  Lyons, J. C. –, , –, , , , , , –, , , –, –, –, , , , , , , ,  Macdonald, J. ,  Machery, E. , –, , , , ,  Mack, A. ,  MacKay, D.  MacLeod, D. – MacLin, O.  Macpherson, F. , –, –, , –, , , , –, , –, , , , , –, , , –, , , , , , , , , –, , ,  Mahon, B. Z. , , , , –, , , , , 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



author index

Malach, R.  Malebranche, N. , ,  Mandelbaum, E. , ,  Manne, K.  Mark, L.  Marr, D. , , –, , –, , , –, , , , , –, –, –, , ,  Martin, J. R. , , , , , , –,  Masrour, F. ,  Massaro, D. W.  Massironi, M. , ,  Matelli, M. ,  Mathalon, D.  Matthen, M. , , , , – Matthews, L.  Maunsell, J.  McCauley, R. N.  McClelland, J.  McCurdy, H. G.  McDowell, J. , , , ,  McGrath, M. , – McGurk, H. , , ,  McKone, E.  Melchner, M.  Merigan, W.  Mesulam, M. M.  Meyer, K.  Michotte, A.  Milner, A. D. , –, , – Milner, D. , ,  Mishkin, M. , , ,  Mishra, J.  Miyamoto, Y.  Mole, C. –, –, , , , – Muller, M.  Muller, V. C. , , , , , , –, , , , , – Murdoch, I.  Murthy, P. , 

O’Reilly, R. C. ,  Olkkonen, M. , –, –,  Ontiveros, Z.  Ooi, T.  Oppenheimer, D. 

Naccache, L.  Nanay, B.  Naumer, M. J.  Nelson, T.  Newman, L.  Nichols, S. ,  Niedeggen, M.  Nijhawan, R.  Nisbett, R. E.  Nobre, A. ,  Noe, A. –, – Noppeney, U. 

Quine, W. V. O. , –

O’Callaghan, C. , ,  O’Neil, S.  O’Regan, K. , 

Pacherie, E. , ,  Pagondiotis, C. , , , , , – Parise, C. V. ,  Parnas, J.  Payne, K. ,  Peacocke, C. , , , –, , , –, –,  Perky, C. W. , , , – Perner, J.  Perry, J.  Peterson, M. A. , –, , ,  Petkova, V.  Peyrin, C.  Pick, Jr, H. L.  Pierce, L.  Pinker, S.  Pisella, L. ,  Plutynski, A.  Posner, M.  Postman, L. –, , , – Priest, G.  Prins, N.  Prinz, J. J. , , , , , , –, – Proffitt, D. R. , , , , , –, , ,  Proust, J.  Pryor, J.  Purves, D.  Pylyshyn, Z. W. –, , –, , , , –, –, , , –, , –, –, , –, , –, –, , , , , , , , , –, , , , , –, –, –, –, , –, , –, 

Radeau, M.  Raffman, D.  Raftopoulos, A. –, , , –, –, , –, –, , , , , –, –, , , , , , –, –, , –, –, , , , , , , –, –, , –, , , ,  Ramachandran, V.  Reber, R.  Reder, L.  Regan, D.  Reimer, M.  Remington, R.  Rensink, R. , 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

author index Reynolds, J.  Riddoch, J. , –, , ,  Rizzolatti, G. ,  Rock, I. , , , , , ,  Rödl, S. ,  Roelfsema, P. R. –, –, ,  Rogers-Ramachandran, D.  Rosenthal, O. ,  Ross, J. – Rothi, L. J. ,  Rouse, C.  Rubens, A. ,  Rumelhart, D.  Rushworth, M.  Ruskin, J.  Russell, R. ,  Sacco, R.  Sacks, O. , – Samuels, R. – Sanchez-Vives, M.  Sass, L.  Schier, F.  Schmid, M.  Scholl, B. J. , , ,  Schooler, K.  Schupbach, J.  Schwartz, B.  Schwartz, R.  Schwenkler, J.  Schwitzgebel, E.  Sedgwick, H.  Segal, S. J. , ,  Sekuler, R.  Sellars, W. , ,  Sergent, C. ,  Shams, L. , ,  Shanon, B.  Shea, N. ,  Shepard, R. N.  Sherrick, C. E.  Shimojo, S.  Shoemaker, S.  Shorvon, J.  Siegel, S. , , –, –, , , , , , –, , , , –, –, , , –, , , , –, –, , , , , ,  Siewert, C.  Silins, N.  Simmons, A.  Slater, M.  Smeets, J.  Smith, A. ,  Spekreijse, H. ,  Spelke, E. S. –, , –,  Spence, C. , –, , , , 



Spencer-Bowdage, S.  Sperber, D.  Sperling, G.  Squire, S.  Stalnaker, R.  Stanford, T. R.  Stark, C.  Stazicker, J.  Stefanucci, J. K.  Stein, B. E.  Sterelny, K.  Stich, S. , , –,  Stokes, D. , –, –, –, –, –, –, , , , , , –, –, , , , – Stokes, M.  Storm, R. W. ,  Strack, F.  Stratton, G. M. ,  Strawson, P.  Super, H.  Swets, J.  Tabery, J.  Tajfel, H. ,  Tanner, W.  Tarr, M. J. , – Taylor, A.  Taylor, J.  Teng, L.  Thomas, N. J. T.  Thompson, E.  Thorpe, S. J. – Tienson, J.  Treisman, A.  Troje, N. F.  Tsao, D.  Tsuchiya, N.  Tucker, C.  Turner, W. M. ,  Tversky, A. , , ,  Tye, M. –, , –, , , –, , –, , , , , , –, ,  Uithol, S.  Ullman, S. , , ,  Underwood, G.  Ungerleider, L. G. , , , ,  van Beers, R.  vanBoxtel, J. J. A.  van der Hoort, B. – Van Essen, D.  Van Gulick, R.  vanUlzen, N. R.  Vance, J.  VanRullen, R. 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



author index

Vatakis, A.  Vecera, S. –, , ,  Vernon, B.  Vishton, P. , ,  Vroomen, J.  Walton, K. , ,  Waltz, D.  Warren, D. H. ,  Warrington, E. K. ,  Watanabe, K.  Watanabe, T.  Watkins, S.  Webster, M. – Wee, N.  Weiskrantz, L.  Welch, R. B.  Westhoff, C.  Whitham, E. M.  Whittlesea, B. ,  Wiggins, D.  Williams, L. , 

Witt, J. K. , , ,  Wittgenstein, L. , ,  Witzel, C. –, –,  Wolfe, J. , ,  Wollheim, R. , , , ,  Wolter, J.  Woods, A. J. , , ,  Wraga, M.  Wu, W. , , , –, , –, , , , , , , , ,  Wurtz, P.  Xu, H.  Yeh, Y.  Young, G. ,  Young, M. P. , ,  Zahavi, D.  Zeimbekis, J. , , , , , , , , , –, , , , , ,  Zeki, S. –, 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

Subject Index abilities  conceptual  modulo drawing  normal  perceptual  recognitional , ,  representational  access consciousness , , , , , ,  achromaticityjudgement  achromatopsia ,  action(s) , , , , , , , , , , , , , , ,  appropriate  control/control of ,  deficits  grasping ,  manual  object-directed/associated , , ,  overt attentional  philosophy  plan(s)/-ning , , , , , ,  situation  stream vision-form  vision  components of ,  goal of  mental ,  motor , , , ,  (visually) guided/guiding , , , , , ,  visuomotor ,  wellformed  actionable  adaptation , , ,  cross-category  diachronic  face ,  high-level  mechanism of  visual , ,  affective/relational phenomenology , , , , ,  dimension , , , , ,  phenomenology , , , , ,  affordances ,  agnosia  associative , , , , , ,  object form ,  visual , , , , –, , , 

visual associative  visual object patient  agnosticism  ambiguity/ambiguous , , , ,  figure(s) , , , , , , , , , , , , , ,  Jastrow’s/Jastrow type/figure ,  perceptual , ,  visual , , , , , –, , , , , ,  anterior intraparietal sulcus  antirealism/st , , , , , , , ,  anxieties  opponents ,  skepticism  approach alternative , ,  anti-intellectualist  architectural  direct realist ,  internalist  low-road  (es) highroad ,  apraxia , ,  influential model  object/object-use , ,  ataxiaoptic , ,  attention/nal , , , , , , , , , , , , , , , –, , , –, , , –, –, , , , , –, , , , , , ,  -based interpretation (see interpretation attention-based) bias (see bias attention) cognitively driven – direct , ,  effects , , ,  endogenous , ,  exogenous ,  feature , , , ,  higher-level  -directing , ,  inattentiveness mechanisms (of) , , , , , , ,  -mediated influence , , , , ,  modulation , , , , ,  overt , , , , 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



subject index

attention/nal (cont.) overt action (see action overt attentional) perceptual  performance  pre-attentional , , , , , , , ,  pre-attentive , ,  process(es) , ,  selection , , , , , ,  -shift interpretation (see interpretation attention-shift) shifting/-shift argument , , , , ,  spatial –, , , , , –,  attitude(s) doxastic  evaluative ,  psychological  dismissive attitude , , , ,  attributions aesthetic property  concept-level  contradictory  mental state  awareness , , , , , , , , , , , , , , , , , , , , , , , , , , ,  access , , ,  cognitive access (CAA) , , , , , , , , , , , ,  conceptually modulated visual  conscious (access) , , , ,  doxastic-level  PC  perceptual , , , ,  personal-level ,  phenomenal (PA) , , , , , , , , ,  visual , , , , , , ,  background phenomenology ,  Bayesian formulations  Bayesian priors , ,  behavior/al , , , ,  path ,  space , , ,  behaviourism , ,  belief , , , , , , , , , , , , , , , , , , , ,  bias (see bias belief) botanical  compartmentalized ,  conscious  fixation , ,  illusory  implicit , , , , , ,  introspective 

explicit , ,  perceptual , , , , , , , , , ,  unconscious  bias anchoring ,  attentional  belief ,  implicit measures  implicit  perceptual , ,  problematic  publication ,  response , , ,  binocular binocularity  disparity , , , , , , , ,  object (see object binocular) vision (see vision binocular) blind blindnessgenuine  blindness inattentional (IB)  bottom-up , –, , , , , –, , , , , –,  explanation ,  processing  brain , , , , –, , , , , , , , , , , , , , , , , , –, , , –, , , –, , , , ,  areas ,  circuits  damage  lesions  mechanisms , , ,  plasticity  processes , ,  regions  representations (see representations) state(s) , ,  competitive processes  plasticity ,  point of view  capacities cognitive , , , ,  perceptual  recognitional ,  Capgras syndrome (CS) , ,  circularity , , ,  epistemic , ,  illicit  intraperceptual  cognition/cognitive(ly) , –, , , , , , , , , , , –, , –, , –, , , , ,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

subject index , , , , , , , , , , , , ,  approach, (meta) cognitive ,  architecture –, , , ,  assumption(s), higher ,  capacities , , , ,  defective  embodied ,  metacognitive judgement (see judgement metacognitive) access awareness (CAA; see awareness cognitive access) access consciousness (CAC; see consciousness cognitive access) accessible ,  attitude (see attitude cognitive) background  capacities (see capacities cognitive) centres , , , , ,  content (see content cognitive) distinction , , , , , , , ,  driven attention (see attention cognitively driven) effects top-down ,  encapsulation (see encapsulation cognitive) episodes  experimentalscience  factors , , , ,  impenetrability/impenetrable (see impenetrability cognitive) influence , , , , , , , , , , , , –, , , , , , , ,  mechanisms , , ,  modulation ,  noncognitive –,  noncognitive penetration penetrability/penetrated/penetration (see penetrability cognitive) phenomenology/phenomenon (see phenomenon/phenomenology cognitive) phenomenon/phenomenology (see phenomenon/ phenomenology cognitive) processes/processing , , , ,  processors (see processors cognitive) psychology/psychologist , –,  states , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  system (see system cognitive) transformations  categorization judgements (see judgement(s) categorization)



tasks  process , ,  colour (colour)less  -categorization judgements (see judgements colour-categorization) constancy  extractor ,  memory (see memory colour) perception (see perception colour) phenomenology of (see phenomenology of colours) qualia  computations/computational(ly) , , , , , , , , , , , , , , , , , , , , , ,  articulation  capacities of a substrate  characterization  computationalism/ computationalist , , , ,  condition ,  device  explanations  framework/theory of mind/doctrine –, , ,  function(s) , ,  mechanisms  model , ,  principles  processors  psychology  resources  transformations  visual (see vision) work  dorsal stream , , – concept/conceptual(ly) , , , , , ,  abilities (see abilities conceptual) acquisition  conceptualization  content (see content conceptual) encoded information  information ,  issues , ,  mind  nonconceptual system (see content nonconceptual) nonconceptual states  nonconceptualists  nonconceptualist philosophers  possession  preconceptual  recognition (see recognition conceptual) recognitional 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



subject index

concept/conceptual(ly) (cont.) representation (see representation conceptual) representationalism (see representationalism nonconceptualist) schemes  sortal , , , , ,  conscious/consciousness , , , , , , , –, , –, , , , , , ,  access awareness (see awareness conscious access) acts , , ,  awareness (see awareness conscious) belief (see belief conscious; belief unconscious) consciously , , , , , , , , , , , ,  consciousness (CMVC) , , , , , , , ,  cognitive access (CAC) –, –, –, – perception (see perception conscious; perception unconscious) perceptual content (perceptual content conscious) perceptually  phenomenal (PC) , , , , , , , , , , , , , , , , , , , ,  phenomenally , , , , , ,  pre-conscious  processes visual unconscious  reflection  representations (representations conscious) self-consciousness ,  sensory inputs (see sensory inputs conscious) thoughts (see thoughts unconscious) visual , , , , , , , , , ,  visual processes (see processes visual unconscious) un(consciousness)  visual contents (see contents conscious visual) consequences debate-neutral ,  mental architecture ,  negative epistemic  consequentialism conjunctive  disjunctive , , ,  consequentialist conception (see conception consequentialist) constructivism , , , 

constructivists  content(s)  / D  awareness (see awareness content) -based condition ,  cognitive  conceptual , , , , , , , , , , , , , , , , , , , , , , , , ,  conceptual/nonconceptual , , , , , , , , , , ,  hallucinatory  informational – mental , ,  nonconceptual , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  of thought , , ,  perceptual , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  phenomenal , , , , , , , , –, , , , , –, , , , , , , , , , –, , , , ,  representational , , , , , , –, , , ,  semantic , ,  visual , , , , , , , , , , ,  volumetric , , ,  conventionalism ,  cortex dorsal occipital  early visual ,  extrastriate , ,  inposterior parietal  posterior parietal  prefrontal , ,  premotor  subcortical temporal(-occipital)  ventral temporal  visual , , , , ,  crossmodal correspondences , ,  Interaction , , cues binocular  mnemonic  monocular , , , , ,  perceptual ,  social , , , 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

subject index spatiotemporal  visual – depression  derealization disorder ,  distal intentions ,  distal layouts  dorsal (stream’s visual) computations ,  occipital cortex (cortex dorsal occipital) pathway , , , , ,  stream , , , , , , –, –,  stream computations (see computations dorsal stream) system(see system dorsal) visual pathway , , ,  visual system (see system visual dorsal) post-stream  doxastic  irrational etiology (see etiology irrational doxastic) sub(doxastic)  acquiescence  beliefs(see beliefs subdoxastic) egocentric egocentrically  space  visual experience (see visual experience egocentric) volumetric representations (see representations egocentric volumetric) empiricism  encapsulated/encapsulation , , , , , , , , , , , ,  informational , , , , , , –, , , , , , , , , ,  informationally encapsulated , , , , , , , , , , , , , , –, , , , , , , ,  modules , , ,  monadic conception of  perceptual  un(encapsulated) modules  un(encapsulated) systems (see systems unencapsulated) un(encapsulation) informational  un(encapsulation) of perception  endogenous attention (see attention endogenous) effects  spatial attention (see attention spatial) epistemic/epistemology /epistemological(ly) , , , , , , ,  circularity , , 



downgrade   feelings , , , , , , power reduction of ,  epistemological(ly)  consequences ,  illicit feedback  implications  questions  reliabilist (see reliabilist epistemology) etiology circular  irrational doxastic  exogenous effects  attention (see attention exogenous) experience(s)hallucinatory  non-veridical  perceptual , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  phenomenological  visual , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  feed forward sweep , , , , , ,  fixation belief (see belief fixation) folk psychology  theory  Goldilocks , , – hallucinations/hallucinatory , , , , , , ,  content (see content hallucinatory) experience (see experience hallucinatory) states  illusion , ,  body swap  crossover  Ebbinghaus , , ,  flash-lag  rubber-hand  sound-induced flash , ,  stream-bounce  illusion ventriloquist  illusory  belief (see belief illusory) picture perceptions (see perceptions illusory picture)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



subject index

information(al) /computational dependence  content (see content informational) encapsulation (see encapsulation informational) exchange ,  interaction , , ,  -ly penetrate (see penetrate informationally) proximal  semantic , , , , , , , –, , , , , , , , , , ,  im(penetrability)/impenetrable cognitive –, , , –, –, , , , –, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , –, –, , ,  internalist principle of justification  internalized regularities  internally , , ,  interpretation(s) , , , ,  attention-shift , ,  intraperceptual , ,  judgement (see judgement intepretation) memory (see memory interpretation) visual , , ,  judgement(s) (-al) skill  (-al) uncertainty  interpretation ,  metacognitive ,  metaperceptual  perceptual ,  knowledge nonpropositional , , , – object , ,  prior , , , , ,  self-knowledge , , ,  bodily ,  implicit bodily ,  sensorimotor , , , , -,  lateral geniculate nucleus (LGN) , , , ,  low-level ‘building blocks’  perceptual (see perceptual low-level) problems  processes (see low-level processes) property ,  visual processes (see visual processes lower-level)

memory , , , , ,  colour –, , , –, – colour effect (MCE) – long-term , , , , ,  memory prospective – metamemory , ,  visual –,  working , , , , , ,  mental architecture , , –, , , , , , ,  arithmetic ,  high-level states imagery , , ,  phenomenon (see phenomenon mental) processes (see processes mental) processing (see processing mental) representation (see representation mental) state , , , , , , , , , ,  metaphysically ,  metaphysics ,  mind , , , , , ,  and brain ,  concept (see concept mind) modular/modularity , , , , , , , , , , ,  (-lar) account of perception (see perception modular account) /non-modular distinction  ‘boundary-crossing’ criterion  a doctrine of weak  anti(modularists) ,  beyond ,  boundary , , , ,  isotropy-based/criterion ,  modularist perception (see perception modularist) modularists , ,  nonmodular  of mind , , , , , , , , ,  partial  perceptual , , ,  processing  pro-modularists  theory of ,  monocular cues (see cue monocular) depths , , , , ,  object (see object monocular) vision  motion biological , ,  parallax  -sensitive 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

subject index motor action (see action motor) control  planning ,  pre cortex (see cortex pre-motor) responses ,  Necker cube , , ,  New Look , , , , , , –, –, , , , , , , , , ,  New New Look , , , , , , ,  nomological  object(s) , , , , , , , , , , , , , , , , , , , , ,  apraxia (see apraxia object) boundaries ,  classification  -directed actions (see actions object-directed) identification , ,  individuation , , , ,  intentional  knowledge (see knowledge object) of perception/perceived , , , , , , , , , ,  proto-(objects) , , , ,  recognition , , , , , , , , , , , , , , , , , , , , – representation ,  visual , , , , , , ,  overt  attention (see attention overt) mechanism  recognition (see recognition overt) Parkinson’s disease ,  penetrability/penetrate(d)/penetration/ penetrable cognitive , , , –, –, –, , –, –, –, , , –, –, –, –, –, –, –, , , –, – , , , , –, –, , , , –, –, –, , , , , , , –, –, , , –, –, –, –, , –, –, , , , , , –, , , , , , , , , , , –, –, –, , , , , , , , , , , – , , , , , , , ,  debate , ,  informational , , ,  lateral , –,  non-cognitive , 



perceptual/postperceptual ,  synchronic, low-level informational (or SLIP) , , ,  perception , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  /cognition distinction ,  abilities (see abilities perceptual) ambiguity (see ambiguity perceptual) architecture(s) of meta(perception)  meta(perceptual) , , , , ,  meta(perceptual) feelings  meta(perceptual) judgements (see judgements metaperceptual) attention (see attention perceptual) awareness (see awareness perceptual) belief (see belief perceptual) intraperceptual circularity  bias (see bias perceptual) bodily  capacities (see capacities perceptual) colour(see colour perception) conscious (see conscious-perception) competition (see completion perceptual) computation (see computation perceptual) conscious (see conscious perceptually) content high-level , , , ,  content (see content perceptual) content conscious (see content perceptual conscious) cues (see cues perceptual) discernment  early states ,  experience (see experience perceptual) illusions (see illusions perceptual) effect , ,  encapsulation (see encapsulation perceptual) experience (see experience perceptual) illusions (see illusions perceptual) imagery , ,  impenetrability (see impenetrability perception) information , ,  infra(perceptual)  inputs  interaction  intra(perceptual) circularity (see circularity intraperceptual) interpretation (see interpretation intraperceptual)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



subject index

perception (cont.) judgements (see judgements perceptual) learning , , , , , ,  mechanism , , , , , ,  metafeelings ,  modularity (see modularity perceptual) modulation  monoptic  multisensory  object of (see object perception) ordinary ,  penetration/penetrability (see penetrability perception) phenomenology (see phenomenology perceptual) picture , –, , , –, –, – plasticity (see plasticity perceptual) post(perceptual) , , ,  process(-ing) , , , , , , , , , , , , , , , , , , , –,  representation (see representation perceptual) state(s) , , , , , , , , ,  systems (see systems perceptual) visual , , , , , , , , , , ,  warrant , ,  phenomena/phenomenal(ly) , , , ,  awareness (see awareness phenomenal) consciousness (PC) (see consciousness phenomenal) content (see content phenomenal) contrast (see contrast phenomenal) dual ,  sensory , ,  -lity , , ,  phenomenological  approach  difference , , ,  experience (see experience phenomenological) phenomenology , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  of colours ,  perceptual , , , , , , , , , , , ,  sensory , , , , , , ,  visual , , , ,  , -, , , , , 

phenomenon cognitive , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  mental ,  target ,  visual , , , , , , , , , , , , , ,  picture partition , , , , , , ,  perception , , , , , , , , , , , , , , , , , , , , ,  realistic , , , ,  plasticity , , , , , , , , , , , , , ,  processes/processing cognitive (see cognitive process) low-level ,  mental (see processing mental) object-recognition  personal-level  post-perceptual cognitive  recognitional , ,  retrieval  semantic  top-down  top-down visual  computations visual  visual early  visual , , , –, –, , , , , , , , , , , –, , , , , , –, , , , , , , , , , , ,  proposition/propositional ,  -discursive , ,  attitudes (see attitudes propositional) knowledge , , , ,  nonpropositional  knowledge (see knowledge nonpropositional) representations (see representations non-propositional) psychiatric disturbances  proximal information (see information proximal) projections  stimuli ,  rational a-(rational) , – realism/realist , , , , –, , , , , , , , , , , –, , –, –, , , , 

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

subject index scientific , ,  recognition Biederman’s theory of (RBC) ,  conceptual ,  face , , ,  high-level , , ,  object (see object recognition) observational  online  overt  visual , , ,  recognitional abilities (see abilities recognitional) capacities (see capacities recognitional) concept (see concept recognitional) processes (see processes recognitional) referential function  relativism ,  conceptual  reliabilism , ,  reliabilist/reliabilists epistemology (see reliabilist epistemology) framework ,  representation(s) , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,  anti-terms , , ,  anti-representationalist ,  approach(es) anti-representational/-ist , , ,  brain , , ,  , , , , ,  coherence , ,  conscious  content , , , , , , –, , , ,  experiential  mental representation , , , , , , ,  presentational component –, ,  volumetric , , , , , ,  representation(al) , , , , , , , , , ,  abilities (see abilities representational) and Computational Theory of Mind (RCTM) , , , ,  coherence , ,  content(s) (see content representational) states , ,  non-representational , , , , , , ,  visual , , , , ,  representationalism ,  nonconceptualist 



representationalist  theory  -computationalist theory of mind  restrictivism ,  realism/-ists , , , , , , , –, , , , , , –, , , ,  anti(realists) , , ,  direct , , ,  pictures (see picture realistic) recognition , , , , ,  object, (see object recognition) recognitional ability(ies) (see abilities recognitional) concept (see concept recognitional) relation causal , , , , , ,  salience/-ncy  salient  schizophrenia  scotopic  segmentation , , ,  segregation figure–ground  object (see object segregation) semantic , , , , , , , , , , , , ,  analysis  a(semantic)  congruence  content (see content semantic) criterion , , ,  dementia ,  information (see information semantic) information processing systems (see systems semantic information) input ,  interpretation  -level impairment  coherence/coherent , ,  processing (see processing semantic) properties , , ,  -rational , , , , ,  representations (see representations semantic) systems (see systems semantic) sensory content (see content sensory) experience , , ,  input  modality ,  multi(sensory) interaction , ,  modules  perception (see perception multisensory)

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i



subject index

sensory (cont.) phenomenology (see phenomenology sensory) post(sensory)  qualia  receptors  representation (see representation sensory) sensorimotor knowledge (see knowledge sensorimotor) stimulation ,  uni(sensory) processing(see unisensory processing) sensation(s) non-sensorily  signal detection theory (SDT)  size constancy , –,  relative  skepticism , , ,  sketch(es) ,   / D (Marr’s) , , , , , , , , , , , , , , –, ,  D (Marr’s) ,  D (Marr’s) , , , ,  spatial , , , , , ,  attention (see attention spatial) boundaries , , ,  configurations  coordinates  location , ,  orientation  representation (see representation spatial) separation  system(s) anisotropic  boundaries  cognitive , , , , , , , , , , , , , , , , , ,  computational , , , ,  dorsal , ,  higher-level ,  information-processing , , , ,  perceptual , , , , , , , , , , , , , , , , , , , , , , , , , , , , , –, , ,  semantic  information processing  storage  visual , , , , , , , , , , , , , , , , , , ,  dorsal  early , ,  visuomotor , 

thalamus  theory of recognition Biederman’s (see recognition Biederman’s theory) -laden/theory-ladenness , –, , , –, , ,  thoughts , , , , , , , , ,  perceptual ,  unconscious  top-down , , ,  cognitive effects (see cognitive effects top-down) cognitive signals (see cognitive signals top-down) conceptual influence (see conceptual influence top-down) effects , , ,  processing (see processing top-down) signal , ,  visual processing (processing top-down visual) valence(d)  ventral  distinction  pathway , , , , ,  stream , , , , , ,  system (see system ventral) temporal cortex (see cortex ventral temporal) visuo-semantic (see visuo-semantic ventral) vertical axis ,  virtual reality ,  vision/visual (ly) action  D representations (see representations D visual) adaptation (see adaptation visual) agnosia (see agnosia visual) ambiguity (see ambiguity visual) ambiguous/ambiguity (see ambiguity visual) areas , , , , , , , , , , , , , ,  attention (see attention visual) awareness (see awareness visual) binocular  brain  computations , , ,  content (see content visual) consciousness (see consciousness visual) conceptually modulated , ,  subcontent visual , , , , , , , , , , , , , , ,  early vision , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

i

i i

i

i

i

OUP CORRECTED PROOF – FINAL, //, SPi i

i

subject index , , , , , , , , , , , , ,  early visual areas , , , ,  cortex ,  processes , ,  visual processing , –, , ,  system , ,  experience , , , , , , , , , , , , , , , , , , , , , , –, –, , –, , , , , , , , , , , , –, , , –, , , –, , , , , –, , , , , ,  information , , , , , , , , , , –, , , ,  interactions audiovisual  early cortex (see cortex early visual) early system (see system visual early) experience (see experience visual) experience egocentric (see experience visual egocentric) extension ,  feature ,  field , , , , , , , , ,  –, ,  gap  guided , , , ,  illusions (see illusions visual) image/imagery , , , , , , ,  imagine/imagination , , , , , ,  impression 



index(-ing) ,  information , , , , , , , , , , , , –, , , , ,  input , , , , ,  interpretation (see interpretation visual) low-level  memory working (see memory working visual) monoptic , , ,  non(visual)  objects (see object visual) output  perception (see perception visual) phenomenon/phenomenology (see phenomenon/phenomenology visual) processes/processing (see processes/processing visual) recognition (see recognition visual) recognitional ability (see ability visual recognitional) recognitional capacities (see capacities recognitional visual) representation (see representation visual) scene , , , , , , , , , , , , , , , , , , , , , , , ,  visual colour (see colour visual) visualizing , , , , , , , , , , , , –, ,  visuo-affective system  visuomotor , , , ,  action (see action visuomotor) system (see system visuomotor) visuo-semantic ventral 

i

i i

i