Epistemic Entitlement 0198713525, 9780198713524

For most of the twentieth century, philosophers have explored the nature and extent of our knowledge-especially our know

167 74 3MB

English Pages 396 [407] Year 2020

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Epistemic Entitlement
 0198713525, 9780198713524

Table of contents :
Contents
Acknowledgements
Contributors
1. Introduction and Overview: Two Entitlement Projects • Peter J. Graham, Nikolaj J. L. L. Pedersen, Zachary Bachman, and Luis Rosa
Part I. Engaging Burge’s Project
2. Entitlement: The Basis for Empirical Epistemic Warrant • Tyler Burge
3. Perceptual Entitlement and Skepticism • Anthony Brueckner and Jon Alt
4. Epistemic Entitlement—Its Scope and Limits • Mikkel Gerken
5. Why Should Warrant Persist in Demon Worlds? • Peter J. Graham
Part II. Extending the Externalist Project
6. Knowledge, Default, and Skepticism • Ernest Sosa
7. Extended Entitlement • J. Adam Carter and Duncan Pritchard
8. Moorean Pragmatics, Social Comparisons, and Common Knowledge • Allan Hazlett
9. Internalism and Entitlement to Rules and Methods • Joshua Schechter
Part III. Engaging Wright’s Project
10. Full Blooded Entitlement • Martin Smith
11. Pluralist Consequentialist Anti-Scepticism • Nikolaj J. L. L. Pedersen
12. Against (Neo-Wittgensteinian) Entitlements • Annalisa Coliva
13. The Truth Fairy and the Indirect Epistemic Consequentialist • Daniel Y. Elstein and C. S. I. Jenkins
14. Knowledge for Nothing • Patrick Greenough
Index

Citation preview

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

Epistemic Entitlement

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

Epistemic Entitlement  

Peter J. Graham and Nikolaj J. L. L. Pedersen

1

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

3

Great Clarendon Street, Oxford, OX2 6DP, 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 2020 The moral rights of the authors have been asserted First Edition published in 2020 Impression: 1 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 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2019941428 ISBN 978–0–19–871352–4 Printed and bound in Great Britain by Clays Ltd, Elcograf S.p.A. 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.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

Contents Acknowledgements Contributors 1. Introduction and Overview: Two Entitlement Projects Peter J. Graham, Nikolaj J. L. L. Pedersen, Zachary Bachman, and Luis Rosa

vii ix 1

Part I. Engaging Burge’s Project 2. Entitlement: The Basis for Empirical EpistemicWarrant Tyler Burge

37

3. Perceptual Entitlement and Skepticism Anthony Brueckner and Jon Altschul

143

4. Epistemic Entitlement—Its Scope and Limits Mikkel Gerken

151

5. Why Should Warrant Persist in Demon Worlds? Peter J. Graham

179

Part II. Extending the Externalist Project 6. Knowledge, Default, and Skepticism Ernest Sosa

205

7. Extended Entitlement J. Adam Carter and Duncan Pritchard

223

8. Moorean Pragmatics, Social Comparisons, and Common Knowledge Allan Hazlett

240

9. Internalism and Entitlement to Rules and Methods Joshua Schechter

254

Part III. Engaging Wright’s Project 10. Full Blooded Entitlement Martin Smith

281

11. Pluralist Consequentialist Anti-Scepticism Nikolaj J. L. L. Pedersen

297

12. Against (Neo-Wittgensteinian) Entitlements Annalisa Coliva

327

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

vi



13. The Truth Fairy and the Indirect Epistemic Consequentialist Daniel Y. Elstein and C. S. I. Jenkins

344

14. Knowledge for Nothing Patrick Greenough

361

Index

389

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

Acknowledgements This edited collection is the result of the editors’ shared interest in different brands of epistemic entitlement. Peter first heard Tyler Burge use the phrase ‘entitlement’ as an undergraduate at UCLA. Peter then regularly attended Burge’s seminars at UCLA for nine years after returning to Southern California to start his appointment at UC Riverside. After completing a dissertation under the supervision of Crispin Wright at St. Andrews, Nikolaj moved to UCLA with a postdoc fellowship from the Carlsberg Foundation. Nikolaj also attended Burge’s seminars and started organizing regular one-day workshops to get epistemologists in Southern California together on a semiregular basis. The meetings were sponsored by Nikolaj’s Carlsberg funding and were usually held in Dodd 399 at UCLA. The workshops were run under the heading the “Southern California Epistemology Workshop” and later renamed the “Southern California Epistemology Network” (which is now run by Sven Bernecker, Karl Schafer, and their students at UCI and has also been established as an APA-affiliated group). The meetings brought together epistemologists from Southern California and occasionally from farther afield, including Yuval Avnur, Jason Baehr, Heather Battalay, Sven Bernecker, Tim Black, Paul Boghossian, Mikkel Gerken, Sandy Goldberg, Peter Kung, Luca Moretti, Michael Pace, Masahiro Yamada, Elia Zardini, and Aaron Zimmerman. At some point Nikolaj asked Peter if he would like to co-edit a volume on epistemic entitlement, loosely organized around themes from Tyler Burge’s use of the phrase on the one hand, and Crispin Wright’s on the other. What you see in this volume is the result. Tony Brueckner was another regular participant in Burge’s seminar. That’s when we first got to know Tony. We were blessed to get to know him much better over the years. Tony unfortunately died in April 2014. We got to know Jon Altschul largely through his participation in the Network. We both remember talks on Burge, entitlement, and brains-in-vats. Jon died suddenly in February 2016. We are moved by the opportunity to publish here a paper they wrote together soon after we invited them to contribute. We have chosen to dedicate this volume to their memory. We are deeply grateful to all of our contributors for all the time and effort they put into their outstanding essays, and for their generous patience. Peter is grateful for opportunities to spend time thinking about epistemic entitlement at the Northern Institute of Philosophy, Oxford University, the Cologne Center for Contemporary Epistemology in the Kantian Tradition, and sunny Southern California, at UCLA, UC Irvine, and UC Riverside. He is especially grateful for research support from the UC Riverside Academic Senate and from the three Deans he has had the pleasure to serve over the past seven years. Peter would also like to thank his partner James Jhun for his unbounded encouragement and continuous support.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

viii



Nikolaj is grateful to the Carlsberg Foundation for the funding that made it possible to start the Southern California Epistemology Workshop and Network. He is also grateful to both Crispin Wright and Tyler Burge for discussion of topics touched on in this volume. We are very grateful to Martin Noble who copy-edited the volume. And of course we couldn’t have pulled any of this off without the expert guidance and unwavering support of our editor, Peter Momtchiloff. Peter J. Graham Riverside, California Nikolaj J. L. L. Pedersen Seoul, South Korea

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

Contributors J A, Loyola University New Orleans Z B, Sam Houston State University A B, University of California, Santa Barbara T B, University of California, Los Angeles J. A C, University of Glasgow A C, University of California, Irvine D Y. E, University of Leeds M G, University of Southern Denmark P J. G, University of California, Riverside P G, University of St. Andrews A H, Washington University in St. Louis C S. I. J, University of British Columbia N J. L. L. P, Yonsei University D P, University of California, Irvine/University of Edinburgh L R, University of Cologne J S, Brown University M S, University of Edinburgh E S, Rutgers University, New Brunswick

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

1 Introduction and Overview Two Entitlement Projects Peter J. Graham, Nikolaj J. L. L. Pedersen, Zachary Bachman, and Luis Rosa

1.1 Putting Entitlement(s) in Context 1.1.1 Internalism, externalism, and the skeptical challenge For most of the twentieth century, in a tradition stretching back at least to Descartes, epistemology was dominated by internalist, typically hyper-intellectualist, antiskeptical approaches to knowledge and justification. The point of epistemology, as many practitioners saw it, was to rebut the skeptic in a non-question-begging manner by discovering justifications within our perspectives or point of view that established the existence and general features of the external world. Justifications would then be arguments, accessible on reflection to the justified individual, that cited other elements of the individual’s perspective (her sensations, perceptions, and beliefs, and patterns among these inner, psychological states—what “evidentialist” epistemologists call the individual’s evidence), that in turn combined with self-evident principles of reason to form persuasive rationalizing non-question-begging arguments for the conclusion that our beliefs about the external world, largely caused and sustained by sensory perception, are for the most part correct. Knowledge and justification were then entirely a function of the introspectively accessible elements of the knowing individual’s perspective, combined with consciously thought bridge premises. Hilary Kornblith helpfully called this the “arguments on paper” thesis. An individual’s knowledge and justification in effect turned on the arguments that the individual could write down on a piece of paper from non-question-begging premises— premises that did not presume knowledge of external reality—to substantive conclusions about the nature of external reality. An individual believer would then not only know that there’s a bottle there on the table, but also be in a position to know that she knows that there’s a bottle there on the table. The job of the epistemologist was then to discover and make explicit the form and content of these arguments. The epistemologist would reveal the structure and extent of our propositional justification for justified beliefs and knowledge about the external world and our knowledge of our knowledge. Descartes started this project in the modern period, and it dominated epistemology through most of the twentieth century. Burge’s chapter helpfully Peter J. Graham, Nikolaj J. L. L. Pedersen, Zachary Bachman, and Luis Rosa, Introduction and Overview In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0001

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



  

reviews some of this history. Some epistemologists today still largely see this task as the task for epistemology. By the turn of the twenty-first century, these epistemologists fell into a distinct minority. Two considerations stand out in our minds. The first is that this conception of knowledge objectionably hyper-intellectualizes ordinary empirical knowledge. If empirical knowledge requires non-question-begging justifications formulated, however poorly and sketchily, in the mind of the knowing individual, then very few individuals will know anything at all about the external world. For very few individuals can cite the relevant evidence, think and understand the bridge principles, and put two and two together. They frequently lack the required concepts, metarepresentational abilities, and argumentative skills. Surely non-philosophers know a great deal. Surely young children and animals capable of propositional thought know about their surroundings. Surely empirical knowledge is not the sole position of a sophisticated intellectual elite. Scientia is one thing; knowledge is another. A chorus of authors have made this point, starting in the late 1960s and early 1970s. Second, if empirical knowledge and justification really did require anti-skeptical, non-question-begging assurances drawn entirely from accessible elements in the individual’s perspective, then empirical knowledge is doomed, even for the sophisticated elite. There is just no rebutting the skeptic on his own terms. The elements of our perspective—our “internal” evidence, as it were—just isn’t enough on its own to provide non-question-begging assurances. This doesn’t mean we have to rebut the skeptic for empirical knowledge. Far from it. We can know a great deal—and so the skeptic’s denial of empirical knowledge is mistaken—even if we cannot prove, in a non-question-begging manner, that we have empirical knowledge. Though taking the skeptic on in his own terms is still a venerable pursuit among a handful of epistemologists, few believe it will ever succeed. It’s like the philosopher’s stone of epistemology. Thus, if we are forced to choose between skepticism about our knowledge of the external world on the one hand and a non-traditional, non-internalist theory of knowledge and justification on the other, we should choose the latter. Since the fall of traditional epistemology’s emphasis on refuting the skeptic, two traditions have emerged in analytic epistemology: externalism and (moderate) internalism. Both largely drop engaging with the skeptic as a pre-requisite. Both tend to drop hyper-intellectualist requirements as well, the externalist more so than the (moderate) internalist. The core externalist idea is that justification is not solely a function of what’s within the individual’s accessible perspective or point of view. Justification in the broad sense (in the sense required for all forms of knowledge), externalists argue, involves more than the internal, accessible, psychological elements in the individual’s point of view. Justification in the broad sense involves supplementation from elements “outside” of the individual’s perspective. What’s “in” the mind (what’s in the individual’s perspective) isn’t enough. That’s the truth, as it were, in skepticism: focusing exclusively on what’s in your mind, you won’t find a route back to truth and knowledge about the external world. Empirical warrant does not supervene only on the individual’s internal perspective or point of view, but requires supplementation from elements external to the individual’s point of view. The job of epistemology

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



from the externalist point of view is to identify the “external” supplements. What are they? Why do they underwrite empirical warrant? To what extent do they supplement internal elements that traditional internalists emphasize, and to what extent do they supplant those elements? The core (moderate) internalist idea is that justification (in the broad sense) is solely a function of what’s within or “internal” to the individual’s perspective or point of view. That’s the truth, as it were, in the traditional epistemological project: to find justification for beliefs about what is outside of the mind, look to what is inside the mind. But the contemporary internalist is much more relaxed or liberal (hence “moderate”) about empirical justification than the traditional internalist epistemologist. The contemporary internalist largely drops the requirement that empirical justification requires dialectical defeat of the skeptic, and largely drops the requirement that the individual be able to formulate arguments from premises about the individual’s perspective to conclusions about the external world. The contemporary internalist agrees with the traditional epistemologist in not seeking supplementation from factors outside of the individual’s perspective or point of view to understand empirical justification. Empirical justification, the contemporary moderate internalist asserts, supervenes on factors only within the individual’s perspective or point of view. Wittgenstein and Quine played important roles in setting the stage for the arrival of externalist thinking in epistemology. Early and influential externalists include David Armstrong, Fred Dretske, Alvin Goldman, Robert Nozick, Alvin Plantinga, and Ernest Sosa, among many others. Goldman and Sosa have independently carried the torch for over forty years. Externalism is now a powerful, mainstream approach throughout epistemology. A fundamental tenet is that justification requires our belief-forming competencies and processes to produce and sustain reliably true beliefs. Reliability supplements perspective. Internalist epistemologists who straddle the two centuries include Robert Audi, Laurence Bonjour, Earl Conee, Richard Feldman, Richard Fumerton, Susan Haack, Keith Lehrer, and John Pollock. Among these, Bonjour, Fumerton, and Lehrer stand out as holding on, where they can, to the traditional idea that empirical justification must survive an encounter with the skeptic.

1.1.2 Two uses for ‘entitlement’: Crispin Wright and Tyler Burge The present volume was conceived as a venue to contribute to two major trends in epistemology, both associated with the phrase “epistemic entitlement.” The first is the externalist project, in particular Tyler Burge’s externalist project in epistemology, best known from his papers “Content Preservation” (1993) and “Perceptual Entitlement” (2003). Burge introduced ‘entitlement’ to name a kind of truth-conducive, knowledge supporting epistemic warrant, distinct from justification, the other kind of truth-conducive, knowledge supporting epistemic warrant.¹ The second is Crispin Wright’s engagement with different forms of skepticism, best known from his paper “On Epistemic Entitlement: Warrant for Nothing (and ¹ Pace Plantinga and his followers, for Burge (and many of our other authors) warrant does not guarantee truth. Though warrant contributes to and conduces knowledge, warrant is not “that, whatever it is, which makes the difference between knowledge and true belief.”

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



  

Foundations for Free)” (2004) and elaborated more recently in “On Epistemic Entitlement (II): Welfare State Epistemology” (2014). According to Wright, though we might not be able to refute the skeptic on his or her own terms, we might still have plenty to say. Wright introduced ‘entitlement’ to name a kind of non-evidential right to accept (to trust, presuppose, or take for granted) Wittgensteinian cornerstone propositions, a kind of right that makes it rational to claim knowledge, where this right is distinct from evidential, truth-supporting justifications. Even though the engagement with the skeptic no longer occupies center stage, engaging the skeptic is still a mainstay of contemporary epistemology, and Wright advances one of the more prominent positions in the contemporary discussion. This volume then discusses issues raised by, and surrounding, the work of two influential contemporary philosophers who think that epistemology should encompass two species of warrant, justification, and entitlement. For this reason, Burge and Wright share a commitment to a sort of epistemic pluralism. They both think that epistemology cannot be done adequately with one type of warrant, and hence introduce and draw a distinction between two species: justification and entitlement. Besides Burge’s chapter, the other chapters that directly engage Burge are from Gerken, Graham, and Brueckner and Altschul. Other chapters in the volume that take up and extend the broadly externalist project in epistemology are from Sosa, Carter and Pritchard, Hazlett, and Schechter. The chapters engaging Wright and Wright’s project are from Smith, Pedersen, Coliva, Elstein and Jenkins, and Greenough. In the rest of this section we sketch the differing approaches between Burge and Wright. We will then go on in the sections following to sketch and discuss the individual chapters that make up the volume. Because we’ll say a good deal about Burge’s approach when discussing his chapter in Section 1.2 of the introduction, we won’t go into the details just yet for Burge’s approach. ...  ’  :       To orient readers to Wright’s project, we’ll provide an overview. We’ll then provide a brief comparison with Burge’s project. We shall review Burge’s account in detail in Section 1.2. We hope you will go away with a good sense of the different senses Wright and Burge put to ‘warrant,’ ‘justification,’ and ‘entitlement.’ Wright is first and foremost engaging with the skeptic. Wright agrees with the skeptic that we possess no non-question-begging way of showing that we are not dreaming, that there is an external world, that our eyes are functioning normally and that we are in normal conditions, that the past conforms to our memories and evidence, that other minds really do exist, and that nature is largely uniform. What matters for Wright—and for understanding his project—is seeing what he thinks this shows. He doubts this shows we actually lack knowledge or justification to believe external world propositions (that’s a bottle on the table, for example), propositions about the past (I just walked back to the office), propositions about other minds (Jones is angry), or propositions about the future or the unobserved (your car is still in your garage). He grants that, perhaps, an externalist theory of knowledge and justification might be correct so that, in fact, if the right conditions obtain, we do know and justifiably believe these ordinary, everyday propositions (Wright 2004: 209–10).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



Rather, Wright thinks the skeptic’s challenge shows we do not know that we know these things. Wright’s concern is then the possibility that this inability to know (in a non-question begging way) that we know these everyday propositions threatens something important to us, namely our rational right to claim knowledge for these everyday propositions. Wright is out to show this isn’t so. Wright is out to show—through a priori reflection and self-knowledge (Wright 2004: 204, 209; Wright 2014: 220, 240–1)—that it can still be rational to claim that we know (many of) these everyday propositions, even if we do not (and possibly cannot) know through a priori and self-knowledge that we know these things. “What is put in doubt by skeptical argument is not our possession of any knowledge or justified belief . . . [but] rather our right to claim knowledge and justified belief ” (Wright 2004: 210). Wright’s project of “making out entitlements” is designed to address this doubt, to show that, in fact, despite what the skeptic has shown, we still enjoy a rational right to claim knowledge and justified belief. Wright’s “entitlement” project then sets out to show why it is rational for us to claim that we know these things. It is rational, he argues, provided we have an entitlement to accept (the relevant) cornerstone propositions for the everyday propositions. What is cornerstone proposition? What is it to accept such a proposition? And what is it to have an entitlement to accept such a proposition? “Call a proposition a cornerstone for a given region of thought just in case it would follow from a lack for warrant for it that one could not rationally claim warrant for any belief in that region” (Wright 2004: 168).

Here are some examples (Wright 2014: 216–17): • Nature is pretty uniform; there is a lot of natural law out there. • Others have mental states which are broadly manifest in what they say and do, and their circumstances. • There is an external material world, broadly manifest in normal sensory experience. • The world did not just come into being five minutes ago, replete with apparent traces of more ancient history, but genuinely has such a history, disclosed, in the round, in presently available data. If you didn’t have a rational right to claim that there is an external material world that we learn about through sense perception, then you could not rationally claim to know or justifiably believe that you have hands on the basis of your sensory experiences as of hands. So to have a right to claim we know everyday propositions, we need a right to claim cornerstone propositions. What is it to accept a cornerstone proposition? It is not to believe the proposition, but to trust that it is true, to act on the assumption that it is true, to presuppose it, to take it for granted (Wright 2004: 175–7, 194). What is an entitlement to accept a cornerstone proposition? Since the skeptic has shown that we lack an a priori, evidential right to claim that these propositions are true, our right to accept a cornerstone proposition must be non-evidential. Call

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



  

non-evidential rational rights to accept something (to act on the assumption that it is true or to take it for granted)—rights that do not show or establish that something is true through evidence of truth—entitlements. Entitlements are then rational rights (warrants) to accept (to trust) cornerstone propositions, and so rational rights (warrants) to claim to know cornerstone and everyday propositions. As Wright puts it, “an essential feature of the notion of entitlement [is] that it is a matter of rational trust” (Wright 2004: 205). Entitlement, for Wright, is not a matter of evidential warrant for belief. In other words, entitlements for Wright really are non-evidential in the sense that they are “warrants” to “accept” a proposition that “consist neither in the possession of evidence for [the proposition’s] truth, nor in the occurrence of any kind of cognitive achievement—for example, being in a perceptual state that represents it to one that P, or seeming to recollect that P—which would normally be regarded as apt to ground knowledge or justified belief that P” (Wright 2014: 214). Entitlements are rational warrants to accept cornerstone propositions, rights that form the basis of our rights to claim knowledge of such propositions. Beliefs importantly differ from acceptances. Beliefs are warranted evidentially (Wright 2004: 192–4). Acceptances are not beliefs; they are warranted nonevidentially. “ ‘Warrant for nothing’ [non-evidential warrant] is entitlement to trust. It is in the nature of trust that it gets by with little or no evidence. That is exactly how it contrasts with belief proper, and it is not per se irrational on account of the contrast. Entitlement is rational trust” (Wright 2004: 194). An entitlement to a cornerstone proposition shows why it is rational to “repose confidence” in a cornerstone proposition even when “one has no evidence or other form of cognitive warrant” for the proposition (Wright 2014: 221). Wright’s entitlement strategy is then a “skeptical solution” to the skeptical challenge in the following sense. Wright grants to the skeptic that we have no evidential right (no evidential warrant) to claim that we know everyday propositions and cornerstone propositions (we do not know that we know these things—the skeptic is right about that), but for all that we have a non-evidential right (an entitlement) to claim that we know these propositions (we have a right to accept these propositions). Though the skeptic is right (in one sense), there “is nevertheless no irrationality or capriciousness, in our proceeding in the ways we do—that we are warranted in so proceeding but warranted in a different way” (Wright 2004: 206). We are warranted (we have a rational right) non-evidentially to accept these propositions; we thereby have an entitlement to claim knowledge, rationally so. That’s Wright’s idea. Explaining what entitlement is, and what role it plays in the debate with the skeptic, however, is just the beginning of the battle. To succeed (on his own terms), Wright has to show through a priori reflection and self-knowledge that we have such entitlements. And trying to show that we do comprises the bulk of his two papers on the topic. He pursues four strategies for establishing entitlement (or four kinds or models of entitlement): strategic entitlement (modeled on Reichenbach’s pragmatic defense of induction), entitlement of cognitive project (modeled on the idea that some projects are so basic we cannot justify them by any other project), entitlement of rational deliberation (modeled on the idea that as agents we must deliberate on how to act, and in so doing we must rely on empirical subjunctive conditionals in our reasoning), and entitlements of substance (modeled on the

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



Kantian–Strawsonian idea that to have a concept of experience we must have a conception of bodies located objectively in space and time). A good deal of the subsequent literature, and the chapters in Part III, take up these attempts. We hope we’ve said enough to pique your interest in the chapters challenging Wright. It is an ambitious project. Can it be carried through? ...    ’   If that’s Wright’s project and his particular use of ‘entitlement,’ how does Wright’s project compare with Burge’s, and how does Burge’s use of ‘entitlement’ compare with Wright’s? Burge and Wright both distinguish between “entitlement” and “justification” and both say they are species of “warrant.” Even so, they mean very different things by these words, for they embed these terms in very different projects. We’re about to review Burge’s view on some of these issues in some detail, so here we will be brief. Once you’ve read through Section 1.2 on Burge, you may wish to return to the next few paragraphs again. If the differences are not clear on a first pass, they should be a second time through. Wright is engaging with the skeptic. Burge, for the most part, is not (though in his chapter he discusses a number of issues surrounding skepticism, including Moore’s argument, warrant transfer, closure, and so on). Burge’s project is very different from Wright’s. Burge is largely an externalist reliabilist, focusing on first-order knowledge and warrant, especially perceptual knowledge and warrant. For Burge, warrant (of any kind) is a constitutive contribution to knowledge, and so attaches in the first instance to belief. Wright grants the existence of such “externalist” knowledge, but notes its irrelevance to the skeptical challenge (as he sees it) and his skeptical solution. Wright’s target is the second-order claim to know. Here again is Wright: If knowledge, and justification, are essentially environmental—are constituted by (perhaps reflectively inscrutable) contingencies of our cognitive powers and the way that they enable us to interact with the external world—then no mere skeptical paradox, developed in the armchair, can show that we have no knowledgeable or justified beliefs. So why bother trying to make out entitlements? . . . what is put in doubt by skeptical argument is—of course—not our possession of any knowledge or justified belief—not if knowledgeability, or justification, are conceived as constituted in aspects of the external situation in which we come to a belief. (How indeed could armchair ruminations show anything about that?) What is put in doubt is rather our right to claim knowledge and justified belief. It is this which the project of making out entitlements tries to address and which, on which seems to me to be a correct assumption, externalism is impotent to address. (Wright 2004: 210, emphasis added)

Wright and Burge are then largely engaged in different projects, at different levels. Wright explicitly notes this difference (Wright 2014: 223). That’s two very different projects, with very different uses of the word ‘entitlement.’ Wright introduced the term ‘entitlement’ for his engagement with the skeptic, for a non-evidential right to accept cornerstone propositions, a right knowable a priori, that establishes a right to claim to know things, a right that may exist even if, as it turns out, the proposition is false. Wright contrasts entitlement (a nonevidential, non-truth-conducive right to accept) with justification and knowledge

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



  

(evidential, truth-conducive rights to believe). Wright introduced ‘entitlement’ for a very different purpose than Burge. As we’ll see below, Burge introduced ‘entitlement’ to draw a distinction between two kinds of (evidential, truth-supporting) warrant—entitlement and justification—where warrant is a constitutive contribution to knowledge (warrants are reliable routes to truth; warrants (in Burge’s sense) are ipso facto “evidential” (in Wright’s sense)). For Burge, justifications are warrants through reasons. Entitlements are warrants without reasons. We’ll say below what Burge means by reasons, but for now the idea that reasons are always propositional contributions to warrant shall suffice. Where Wright draws a distinction between justification (truth-conducive warrants—“Wright” justifications) and entitlements (non-evidential warrants—“Wright” entitlements), Burge draws a distinction between two kinds of truth-conducive warrants: warrants through reasons (“Burge” justifications) and warrants without reasons (“Burge” entitlements). Burge’s project (and so his distinction between “entitlement” and “justification”) is not about entitlements in Wright’s sense at all. Wright also notices this difference in terminology and finds nothing objectionable in granting Burge’s distinction between two kinds of evidential, truth-conducive warrants. In Wright-speak, evidential justifications can divide into propositional evidential warrants (reasons, in Burge’s sense) and non-propositional evidential warrants (entitlements, in Burge’s sense). Wright sees “no objection in constraining the notion of justification” so that justifications always involve propositional warrants (2014: 223). The different projects entail an interesting contrast that Wright also notes. Entitlements in Wright’s sense contrast with broadly externalist [warrants], [for Wright-entitlements] are essentially recognizable by means of traditionally internalist resources—a priori reflection and self-knowledge—and are generally independent of the character of our actual cognitisve situation in the wider world—indeed, are designed to be so. (Wright 2004: 209–10)

Wright-entitlements (if knowable) are knowable a priori. They are non-evidential rights to claim to know cornerstone propositions. Externalist warrants, on the other hand, are not in general knowable through a priori reflection and self-knowledge, but are generally only known empirically to obtain, contingently so. In sum, Burge and Wright are engaged in very different projects, and so give different meanings to the words ‘entitlement,’ ‘justification,’ and ‘warrant.’ In the rest of this introduction we provide our own interpretive summaries of the chapters.

1.2 Burge: “Entitlement: The Basis of Empirical Warrant” The 1970s and 1980s saw not only the emergence of externalist thinking in epistemology as we just saw, but also externalist thinking in the philosophy of language and mind. Influenced by Wittgenstein and Rogers Albritton, Hilary Putnam argued that what is in one’s descriptive perspective or point of view is not enough to determine the reference of natural kind terms. Saul Kripke and Keith Donnellan argued that what is in one’s descriptive perspective or point of view is not enough to determine the reference of proper names. David Kaplan and Howard Wettstein similarly made the

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



case for demonstratives and indexicals. Reference, these philosophers argued, requires supplementation from factors external to the individual’s descriptive point of view. Burge argued an even deeper point, viz. that an individual’s perspective or point of view—the individual’s representational states, percepts, concepts, and thoughts— themselves depend on factors outside of the individual’s brain and bodily history, construed narrowly. Burge’s point was not just that one’s descriptive thoughts were not enough to single out the references of one’s thoughts, but that thought itself—and representational mind in general, including perception—depends on relations outside of the individual’s neurophysiology and bodily behavior, narrowly individuated. To think about and perceptually represent bodies, colors, shapes, textures, substances, stuffs, and natural kinds, one needs to stand in causal-historical and explanatory patterns of interactions with instances of bodies, colors, shapes, natural kinds, and so on. Representation—perception and thought—requires supplementation from factors external to our bodies, narrowly construed. Along with more reductively inclined philosophers like Fred Dretske and Ruth Millikan, Tyler Burge ushered in the externalist (anti-individualist) revolution in the philosophy of mind. In the late 1980s and early 1990s Burge began to publish in epistemology, most importantly “Content Preservation” (1993) and “Perceptual Entitlement” (2003), among others. In these papers he advances a sophisticated reliabilist externalist account of epistemic warrant that incorporates his anti-individualist account of representational mind, while acknowledging the importance of the individual’s perspective or point of view.² Burge’s project in epistemology intertwines with his project in the philosophy of mind, for he sees epistemic norms as standards for the wellfunctioning of cognitive competencies. To understand epistemology, you must also know your psychology. And just as psychology requires supplementation from external relations, epistemology requires supplementation from external relations as well. Burge’s chapter “Entitlement: The Basis of Empirical Warrant” lays out his account of epistemic warrant within a normative framework. The framework connects functions, norms and goods: epistemic warrant is a good that consists in meeting norms a priori associated with belief ’s representational and epistemic functions. The functions include truth and being epistemically supported. The norms involve standards for reliably producing true beliefs through normal functioning. Applied to perception, Burge’s view of warrant is then a broadly reliabilist, teleological account where perceptual beliefs are prima facie warranted when formed and sustained through a well-functioning, reliable belief-forming competence. This view is moderate foundationalist: perceptual beliefs do not depend on inferential background support from other warranted beliefs for their prima facie warrant. Burge’s paper has six sections. The first lays out his positive account of warrant. The second illuminatingly covers some of the history of epistemology and argues

² So far, Burge has collected many of his papers in three volumes (Burge 2005, 2007, 2013). The first was on Frege, and includes papers on a priori warrant in particular. The second collects many of his papers in the philosophy of mind, and touches on epistemological issues. The third is epistemological through and through, collecting papers on the nature and epistemology of memory, self-knowledge, reason, understanding, reflection, and testimony. We believe another volume is in preparation, collecting papers on the nature and epistemology of perception.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    against the widely held (internalist) view that epistemic warrant hinges entirely on psychological states and events in the individual’s psychology, and their relations to one another. It also addresses the “clairvoyance” and “demon-world” cases against reliability theories of warrant. The third discusses Moore’s anti-skeptical argument, closure, and warrant transmission. The fourth discusses bootstrapping and “easyknowledge” objections as applied to his account of epistemic warrant. The fifth raises an objection from Bayesian confirmation theory to moderate foundationalism. Burge addresses the objection by examining in detail how Bayesian subjective probability theory fits into empirical visual psychology and his account of perceptual entitlement. The sixth concludes. If that is not enough, the footnotes address understanding, warrant for statistical inference, empirical versus a priori warrant, “consequentialism” in epistemology, de re belief, the generality problem, Sosa’s virtue epistemology, Goldman’s reliabilism(s), phenomenology and warrant, unconscious perception, modular reasons, the individual versus sub-individual distinction (the personal versus sub-personal distintion), blightsight, disjunctivism, epistemic dogmatism, contextualism about knowledge claims, closure, and more. Burge’s chapter contains something for nearly everyone in contemporary epistemology. Though it addresses warrant and cognition in general, Burge’s chapter focuses on perception. Alongside his other papers on the nature and epistemology of perception (especially “Perceptual Entitlement” (2003) and his book Origins of Objectivity (2010)), “Entitlement: The Basis for Empirical Warrant” constitutes a tour de force in the metaphysics and epistemology of perceptual belief. Burge’s chapter can be read as an extended defense of his teleological, reliabilist, moderate foundationalist view of perceptual warrant, developing the framework and responding to a host of standard and more recent objections to reliabilist and moderate foundationalist views of warrant, especially objections to the possibility of empirical entitlement. In the remainder of our introduction to his chapter, we will sketch our understanding of his account of warrant and his distinction between justification and entitlement. We realize we may make more than one mistake when interpreting Burge’s ideas, but we hope you find our interpretation useful. We would like to cover more, as there is so much of interest in Burge’s chapter worth introducing, especially his lengthy discussion of Bayesian conformation theory as applied to perceptual psychology and perceptual warrant. Because of the length of Burge’s chapter, we won’t provide a summary of these topics. On the other hand, because of their importance to the contemporary debate and to understanding Burge’s project, we will go into detail on two main topics—the nature of warrant and the distinction between entitlement and justification. This should also be especially helpful to those new to his work.

1.2.1 Epistemic warrant: a good, reliable route to truth Burge thinks knowledge results from the manifestation of a cognitive competence. A competence is a competence to do something, such as hit a home run, speak a language, or accurately represent objects and their properties in the physical environment. In general, we explain success that manifests competence through the normal functioning of the competence in normal conditions. To manifest a

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



competence is to succeed through its exercise non-luckily, non-accidentally so; manifestation entails the non-lucky, non-accidental successful exercise of the competence. Prima facie warrant takes a belief very close (and commonly all the way) to knowledge. Warrant makes the connection to truth “especially secure” by being a reliable route to truth. For Burge, roughly and approximately, knowledge results from the manifestation of a competence to form and sustain true beliefs, reliably so. Warrant is a constitutive contribution to knowledge, a constitutive contribution to manifesting a competence. Warrant is then the exercise of the competence, that in normal conditions manifests the competence, absent counter-warrants, internal incoherence, and Gettier failures. Warrant certifies or marks a belief as knowledge, absent counter-warrants, internal incoherence, and Gettier failures. Warrant puts belief on a good, reliable route to truth and knowledge. Put this abstractly, Burge’s view of warrant lies near “virtue reliabilist,” “competence,” and “proper functionalist” views.³ If knowledge entails success through normal functioning in normal conditions, what fixes normal functioning and normal conditions? Burge gives an etiological, explanatory answer: look to the explanatory connections between the subject matter of the competence (on the one hand) and the formation and nature of the competence (on the other) and you will find formative patterns between the subject-matter and the competence that fix normal functioning and normal conditions. This applies generally for epistemic competences. We will explain how it applies to perception (Burge 2003). Perception is a competence to accurately represent objects and attributes in the physical environment. The competence was formed and individuated through past patterns of interaction between the system and its subject-matter. This is the basis of Burge’s perceptual anti-individualism; perceptual states are what they are (they have their contents, and so their natures) because of past patterns of interaction between the system and the broader physical environment. The patterns begin with objects and their attributes in the environment; continue through a medium such as light; impact sensory receptors that in turn trigger processing within the perceptual system and then eventuate in perceptual representations that in turn guide behavior in the organism’s environment. These past patterns of interaction partly explain the formation and nature of the perceptual system, its perceptual state types, and so its ability to accurately represent objects and their attributes in the physical environment. Higher animals, especially humans, then form true perceptual beliefs though conceptualizations of accurate perceptual representations (reformatting perceptual representations into propositional form). The formative patterns of interaction specify what counts as normal functioning and normal conditions. Normal functioning is working or operating the way the system worked or operated when it was formed, when it came to be what it is; formative functioning (functioning along formative pathways forged by formative patterns within ³ “Virtue Reliabilists” include John Greco, Chris Kelp, Lisa Miracci, Duncan Pritchard, Ernest Sosa, and John Turri, among others. “Proper Functionalists” include Michael Bergmann, Kenneth Boyce, Peter Graham, Ruth Millikan, Andrew Moon, and Alvin Plantinga, among others. Burge notes some similarities and differences with Sosa and Plantinga in “Perceptual Entitlement” and in the present chapter.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    the psychology) determines normal functioning. Normal conditions are then those kinds of conditions where the system, in working or operating normally, came to be what it is, where it acquired the capacity to represent environmental objects and attributes; formative environments determine normal environments.⁴ The formative patterns ground a non-accidental explanatory path to success. Normal functioning in normal conditions leads to truth. When a perceptual belief is occasioned along this path, the belief is formed on a good route to truth. This path is non-accidentally and explanatorily associated with the nature of the competence; it’s the non-lucky, non-deviant route. That’s what makes it a good route. When perception manifests competence, it runs along this path. In general, warrant runs on such paths, on good routes to truth. Different competence, different formative patterns; different formative patterns, different pathways; different pathways, different good routes to truth. Hence Burge’s slogan: warrant is a good route to truth. Malfunction—failure to function normally or to operate well—then undermines warrant, for then the belief does not derive from the competence but the malfunction; the belief is then not on a good route to truth; it has strayed from the path, even if deviantly or luckily true. Other paths to truth are non-explanatory, accidental paths, at best. Success on any other route does not manifest competence. Only success on good routes manifests competence. Warrant, however, is more than a good route to truth. Warrant is also a reliable route to truth. Being a good route is not enough for being a reliable route; not all good routes are automatically reliable. A normally functioning exercise of a competence in normal conditions does not necessarily succeed. Competences are not necessarily reliable, let alone infallible. World-class major league batters strike out more often than they get a hit, let alone hit a home run. A cognitive competence may be unreliable even in normal conditions when functioning normally, like the professional baseball player at bat. Even perceptual competencies, Burge insists (see especially section II of Burge’s chapter and the appendix to Peter Graham’s chapter), need not reliably represent in normal conditions when functioning normally. The rabbit’s danger representation is inaccurate more often than right. Warrant is a good, reliable route to truth. The competence must also be reliably successful in normal conditions when functioning normally; exercise must manifest competence more often than not. The pathways must then be reliable routes to truth. Hence the patterns of interaction between the competence and its subject matter must not only explain the formation and nature of the competence, they must also underwrite and explain the reliability of the competence; they must ground general explanations of the reliability of the competence in normal conditions when functioning normally; the patterns must lay down pathways that comprise a reliable route to truth. Then, and only then, are beliefs appropriately derived from the competence certified or marked as knowledge, absent counter-warrants, internal incoherence, or Gettier failures. As we noted already, Burge places epistemic warrant within a normative framework. Warrant for Burge consists in fulfilling epistemic norms. These norms are

⁴ Compare Millikan 1984; Graham 2012, 2014.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



standards a priori associated with belief ’s representational function. The representational function of belief is to be true. This representational function does not reduce to practical, biological functions. For every function, there are four norms a priori associated with the function: (i) function fulfillment, (ii) to operate well in serving the function on the occasion, (iii) to reliably fulfill the function, and (iv) to operate well on the occasion and to reliably fulfill the function. These norms are natural norms, in that they are not guiding norms. The individual need not be aware of, or even be able to be aware of, the norms. Fulfilling these norms are goods for the functional item. It is good for a belief to be true, for the belief-forming competence to operate well, and so on. These are goods for the functional item, not necessarily goods for the individual or goods all-things-considered. Epistemic warrant consists in fulfilling the fourth norm. Warrant then entails the normal functioning of the belief-forming competence, and the reliability of the competence in normal conditions. Meeting these norms is then a normative achievement, a good for the belief qua belief (Burge 2003, 2010, present chapter). Warrant is a standing in the space of natural norms, norms associated a priori with representational functions.

1.2.2 Two kinds of warrant: justification vs. entitlement In ‘Content Preservation’ and ‘Perceptual Entitlement’ Burge distinguished entitlement and justification in terms of access: justifications arose from reasons that were accessible to the subject; entitlements are contributions to warrant that need not be accessible to the subject. Burge has since abandoned this formulation (Burge 2013: 26–8). What then is Burge’s current basis for the distinction between entitlement and justification, two kinds of warrant? As we just saw, warrants in general are good, reliable routes to truth; they are exercises of cognitive competencies that reliably lead to truth and knowledge in normal conditions when functioning normally. The distinction between justification and entitlement would then be a distinction between two kinds of routes, two kinds of competencies. What’s the difference?⁵ Burge says justifications are warrants from, or involving, reasons. Justifications are routes to truth (exercises of cognitive competences) that involve reasons. When a theoretical reason is a good, reliable route to truth on which a belief is formed or sustained, the reason is a justification for the belief. Entitlements are then warrants without reasons. Entitlements are reliable routes to truth (exercises of cognitive competences) that do not involve reasons. When a belief is on a good route to truth that is not a reason, the route to truth is not a justification but an entitlement. What then does Burge mean by a reason?⁶ For Burge, reasons are actual and potential answers to ‘why’ questions, steps in explanations. As such they are necessarily propositional. Hence non-propositional perceptual representations, sensations ⁵ Burge now holds that justifications may be operative reasons in sub-personal, modular aspects of an individual’s psychology, not (fully) accessible to the whole individual (Burge 2013: 28, 489). Justifications are propositional structures within the individual’s psychology that provide rationalizing explanations for the belief-worthiness of the target belief for the individual, accessible or not. Justifications are elements of the individual’s perspective or point of view, not necessarily the individual’s accessible perceptive or point of view. For extensive discussion of Burge’s distinction and common mischaracterizations, see Graham forthcoming. ⁶ For Burge’s discussions of reasons, see Burge 2013: 192–3, 391–7, 490–5.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    and worldly states cannot be reasons. Instead reasons are propositional contents, combined with a mode (belief, judgment, intention). Having a reason involves thinking the reason, having a propositional attitude. Theoretical reasons are potential steps in explanation that show why the individual should believe something. They provide a rationalizing explanation for the individual for the credibility of the warranted belief. The argument or explanation need not be second-order or metarepresentational. “It is fundamentally at the same level as the belief: p because r, where p is the content of the belief and r is the reason” (Burge 2013: 490, n. 3). The argument or explanation can be in the individual’s psychology, even if the individual cannot give the explanation. Paradigmatically theoretical reasons are self-evident propositional attitudes (e.g. cogito thoughts or basic logical or mathematical thoughts), and warranted premises in deductive and inductive arguments. With this sketch of Burge’s view of reasons in place, let’s return to the distinction between justification and entitlement. What is justification? Warrant through deductive reasoning is the paradigm case. In deductive reasoning, the individual believes a conclusion on the basis of premises and competence with an inference rule. The premises are then operative reasons contributing to the overall warrant for the conclusion. For example, the warranted belief that the streets are wet inferred from the warranted beliefs that it is raining and that if it is raining then the streets are wet is warranted partly by the warranted premises. These premises are reasons—steps in the explanation of the beliefworthiness of the conclusion for the individual. The warrant for the conclusion is a justification. However, the justification versus entitlement distinction does not align with the distinction between inferential and non-inferential warrant. Some justifications are non-inferential in nature. Burge says cogito (and other self-verifying) thoughts, and our knowledge of basic logical and mathematical truths, are also warranted by reasons. Their warrants are justifications. Why does he say that? He sees these beliefs as warranted through (appropriate) understanding of the self-evident content of these propositions. Believed through understanding, these beliefs contain their own explanation for their belief-worthiness. These propositions contain reasons for themselves (Burge 2013: 18). Warranted by reasons, the warrants are justifications. Justifications are then inferential or non-inferential. Warrants from arguments are inferential justifications, and warrants from understanding self-evidence are noninferential. The justification versus entitlement distinction is not the inferential vs. non-inferential warrant distinction. What is entitlement? Perceptual warrant is the paradigm case. Warrant for a perceptual belief derives from the normal functioning of the reliable perceptual competence and the normal transition from a perceptual representation to a perceptual belief. Perceptual representations are not propositional and so are not reasons. Perceptual beliefs are not self-evident. The transition from perceptual representation to perceptual belief is not an instance of reasoning. Perceptual warrant arises from the perceptual competence. It does not require supplement from other warranted beliefs. Perceptual warrant does not involve or require reasons. Perceptual warrant is an entitlement. Entitlement is warrant without reasons.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



To understand Burge’s distinction, we find it helpful to distinguish between elements in a warrant and the total, overall warrant for a belief. Elements can be operative reasons or elements that are not operative reasons, and the total or overall warrant for a belief can be justification or entitlement. Burge calls operative reasons justifications. Elements that are not operative reasons are contributions to an entitlement. An element of a warrant is a justification when the element is an operative reason in the individual’s psychology. An element is not a justification when it is not a reason. The total warrant for a belief is justification when there is an element in the warrant that is a justification (an operative reason). The total warrant is an entitlement when no element in the warrant is a justification (an operative reason). Many overall warrants often involve operative reasons among the exercises of cognitive capacities responsible for warrant. Hence they are justifications (Burge 2013: 42, 196, 492). Deductive reasoning provides a good example of the mix. Deductive reasoning is a route to truth, provided the premise beliefs are warranted and the individual exercises a reliable competence with deductive reasoning, involving sensitivity to the inference rules. The inference rules, on pain of regress, are not themselves premises. They need not even be represented or representable within the individual’s psychology. The warrant to rely on inference rules is then an entitlement to rely on the rule, not a reason. When the individual infers that the streets are wet, the individual relies on the two warranted premises and her competence with modus ponens. Her warrant to rely on the inference rule is an entitlement, not a reason. The two premise beliefs are reasons. The overall warrant is then a mix of reasons (justifications) and an entitlement to rely on the rule, and so the overall warrant is justification. Furthermore, the warrants for the premise beliefs are often entitlements, not further reasons. Except in cases of self-evidence, warrants for premise beliefs are eventually entitlements, not reasons. Empirical warrants always derive, inevitably, from entitlements, not reasons. Hence Burge’s title: “Entitlement: The Basis of Empirical Warrant.” In a related paper, Burge imagines the possibility of an individual transitioning from a warranted belief to a conclusion, but the starting point does not function as a reason (Burge 2011/2013). He imagines a creature that moves from the belief that is movement in the bushes to the conclusion belief that danger is present, but with no learned evidence of the connection. He imagines the connection was forged entirely through evolution, an innate connection in the individual’s mind. In such a case, though the individual begins with a warranted perceptual belief, that belief does not function as a reason in the individual’s psychology. Nothing in the individual’s perspective rationalizes the transition or explains why the conclusion is beliefworthy. The conclusion belief is then warranted by entitlement, not justification. This means there are inferential entitlements, inferentially warranted beliefs where the warranted premises fail to combine with a competence with a deductive or inductive inferential rule so as to serve as a reason for the conclusion—warranted inference without reasons (Burge 2013). Why does Burge use ‘justification’ for warrants involving reasons? As Burge notes, the ordinary word ‘justification’ is connected to the activity of justifying. When

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    justifying, we argue for the conclusion. We thereby also explain the belief-worthiness of the conclusion. When justifying, we offer a justification. A justification would then be an argument or explanation in the individual’s psychology that explains or shows why the belief is credible for the individual, a rationalizing explanation for the beliefworthiness of the belief for the individual. Having a justification, however, does not require that the individual be able to provide the justification, conceive of the justification as a justification, or even access the justification. Burge does not require that the individual who possesses the justification be able to access the explanation, to evaluate the reasons as reasons, or to give the justification. For Burge, justifications are operative warranting reasons, accessible or not.⁷ Burge’s distinction reminds us of the broad scope of warrant, and the correlative error of focusing too narrowly on reasons, especially critical reasons, in one’s epistemology. Though reasons play an important role in epistemology, reasons are not the only route to truth. Entitlements are equally profound and extensive. Entitlements are involved in perception, memory, interlocution (testimony), self-knowledge, associative beliefformation, and deductive and inductive reasoning. Entitlement is a broad category. Philosophers who insist that warrant is a standing in the “space of reasons” tend to assimilate all warrant to justifications, ignoring or denying entitlements. The normal and reliable functioning of perception and competent transition to perceptual belief falls outside of the “space of reasons” in the “space of causes” and so cannot warrant perceptual beliefs. On these views, the warrant for a perceptual belief comes from other beliefs like “I saw that such and such” or “my visual system is working properly, and it appears to me as if such and such, so probably such and such is the case” or even “the best explanation of the overall patterns of experience and cognition in my mind involves the real-world hypothesis.” These views, Burge repeatedly remarks, hyper-intellectualize perceptual warrant, denying perceptual warrant and knowledge to higher non-human animals, human children, and even ordinary adults. They ignore or deny entitlement. But won’t theories that involve reliability, defenders of the “space of reasons” metaphor retort, end up assimilating knowledge—a normative achievement—to the reliable indications of thermometers and other “non-rational” devices? No. Warrant is not merely reliable correlation. Warrant involves reliable success through the normal functioning of a competence that fulfills norms associated with belief ’s representational function. Warrant, hence entitlement, is a normative achievement. We are not forced to choose between thermometers on the one hand and sophisticated critical reasoners on the other. The space of critical, reflective reasons is a subset of the space of critical reasons, itself a subset of the space of reasons. The space of reasons in turn is a subset of the space of norms. Justification is a standing in the space of reasons; warrant (justification and entitlement) is a standing in the space of norms. Burge’s chapter is an extended defense of this point of view with epistemology. As we noted already, he develops his normative framework in detail, provides historical

⁷ Burge once said justifications were necessarily accessible to the individual. He abandoned that view.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



context, argues against internalism, and replies to a score of objections, old and new, to reliabilist, moderate foundationalism, including the generality problem, the clairvoyance objection, the new evil demon problem, bootstrapping, easy knowledge, Moore’s anti-skeptical argument, and an objection from Bayesian confirmation theory. It’s a long and difficult read, but it more than repays the effort.

1.3 Brueckner and Altschul: “Perceptual Entitlement and Skepticism” Anthony Brueckner and Jon Altschul explore the implications, if any, of Burge’s writings on anti-individualism and perceptual entitlement on radical skepticism. In particular, they ask whether anti-individualism and its closely related veridicality thesis can be employed in an a priori anti-skeptical argument that defeats brain-invat (BIV) or evil demon scenarios, which hold that since “we cannot rule [BIV scenarios] out as non-actual,” we cannot know or are not even justified in believing that we are not BIVs. Anti-individualism is the thesis according to which an individual’s mental states get their content through “causal-formative interactions” and “representationally successful interactions” between mind and world in the phylogenetic history of the individual’s psychological system. The veridicality thesis holds that at some point in history an individual has had veridical representations. How might anti-individualism and perceptual entitlement be used to combat radical skepticism? One way to understand radical skepticism is as a claim that our beliefs in the external world lack any positive epistemic status. If we could provide an a priori argument that we are entitled to at least some of our beliefs about the external world, then the radical skeptic would be rebutted, at least somewhat. Here is how one such argument might go, from an earlier paper by Brueckner (2007: 160): (I) Anti-individualism holds for perceptual content. (Assumption) (II) An individual’s perceptual state types are reliably veridical in his perceptual system’s normal environment. (I) (III) This reliable veridicality is explained by the nature of the perceptual states. (I) (IV) If (i) an individual’s perceptual state types are reliably veridical in his perceptual system’s normal environment, and (ii) this reliable veridicality is explained by the nature of the perceptual states, then the individual, absent reasons for doubt, is entitled to hold beliefs appropriately based on the states. (Assumption) (V) Perceptual entitlement exists. [(II),(III),(IV)] Brueckner and Atshul point out that Burge himself would reject premise (II). Perceptual states may be highly unreliable; consider again a rabbit’s representation as of danger in the bush. Further, this argument gets the structure of Burge’s thought backwards. Rather than argue from the thesis of anti-individualism to the thesis that perceptual entitlement exists, Burge assumes that perceptual entitlement exists, and then wonders what could explain our so being entitled. Anti-individualism is supposed to provide us with such an explanation. This, of course, would be a questionbegging way to rebut the skeptic.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    What is the basis for our beliefs in anti-individualism and the veridicality thesis? Brueckner and Atchul point out that if it is Putnam and Burge style Twin Earth thought experiments—water/twatter, arthritis—then it would seem that such theses, insofar as they are true, are known a priori. If anti-individualism and the veridicality thesis are true and known a priori then, according to Brueckner and Atchul, it is true and can be known a priori that (K) “for there to be any perceptual representation, there must be some veridical perceptual representation (at some time or other).” Notice how weak K is; it does not entail that we are not brains in vats, but rather that at least one of our phylogenetic ancestors was not a BIV. Might we be able to get more, say that our perceptual beliefs are entitled? Recall that Burge thinks that anti-individualism can explain perceptual entitlement. But if we can know, a priori, that anti-individualism exists do we not get, by way of entailment, the conclusion that we can know, a priori, that perceptual entitlement exists? Burge thinks the answer to this is no. While anti-individualism can explain our perceptual entitlement, it does not entail that perceptual entitlement exists, for entitlement requires more than just the truth of anti-individualism, it requires the truth of the further thesis that our perceptual system is reliably veridical in normal conditions. This, Burge suggests, is only something we can know a posteriori. But even if we can’t know, a priori, that perceptual entitlement exists, we can know a priori that (#) there are some things that exist apart from [our] mind[s] (or were at some point things that existed apart from [our] minds). So anti-individualism is powerful enough to rebut very strong forms of radical skepticism, according to which I am the only sentient being ever to have existed or I am a disembodied spirit being tricked by an evil demon. According to anti-individualism, this is metaphysically impossible.

1.4 Gerken: “Epistemic Entitlement—Its Scope and Limits” Working within Burge’s vocabulary, Mikkel Gerken takes up the distinction between entitlement and justification. Gerken agrees that warrant generally arises from a nonlucky connection between belief and truth that arises from the exercise of the relevant cognitive competence. He disagrees over how to draw the distinction between entitlement and justification. He argues the difference has to do with reason considered as a faculty, not with reasons per se. He argues for the Reason Criterion. Reason Criterion (Justification) S’s warrant, W, for her belief that p is a justification if and only if W constitutively depends, for its warranting force, on the competent exercise of S’s faculty of reason. Reason Criterion (Entitlement) S’s warrant, W, for her belief that p is an entitlement if and only if W does not constitutively depend, for its warranting force, on the competent exercise of S’s faculty of reason. Justifications depend on the faculty of reason; entitlements do not. Simultaneously Gerken draws the internalism versus externalism distinction. Internalist warrants are justifications; externalist warrants are entitlements.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



Gerken’s idea seems to be this: the faculty of reason is a sophisticated high-level propositional competence to engage in deductive and inductive reasoning at the personal level. Reasoning is something that the individual does. Inputs to reason, such as perception, though it contributes to perceptual-belief formation, is not similarly high-level, and the computations in the perceptual system are not attributable to the individual. Perception inputs to reason without being an exercise of reason. Exercises of faculties of reason (on the other hand) are attributable to the individual; exercises of non-propositional competencies such as perception are not. Exercises of truth-conducive faculties of reason support justifications; exercises of truth-conducive competencies not involving faculties of reason support entitlements. Justifications are “internal” for exercises of reason are attributable to the individual; entitlements are “external” for exercises of the various sub-personal competencies are not. (The possibility of in principle, modular justifications would challenge this contrast.) Gerken uses the occasion to continue his case for pluralism about warrants: there are entitlements and justifications, with possible sub-species for each—in particular, there are sub-species of justification, including access justification and discursive justification—where a belief may enjoy more than one kind of warrant at the same time. One way of appreciating Gerken’s pluralism and his difference from Burge involves his treatment of cogito cases. Suppose I believe I am thinking or I am hereby thinking that water is wet. Burge argues the warrant comes from a reason (the proposition believed is self-evident, thereby containing its own explanation for the belief-worthiness of that very belief ), but its self-evidence must be understood. One’s warrant is then a justification, through understanding of a reason. Gerken sees things differently. He believes I would first enjoy entitlement for my belief, prior to understanding self-evidence, for cogito-type beliefs are well-functioning infallible. But once I understand the self-verifying nature of the belief, I gain a justification, through the exercise of reason. Gerken also argues against the idea that all justifications should be accessible. Consistent with his pluralism, however, Gerken concludes by canvassing cases where discursive justifications and accessible justifications play an important role in our social lives. A complete theory of warrant, Gerken holds, should recognize the plurality of warrants while recognizing the various epistemic roles they play.

1.5 Graham: “Why Should Warrant Persist in Demon Worlds?” Peter Graham’s chapter addresses a persistent challenge posed to reliability theories of warrant. If warrant entails the reliability of the belief-forming process, then perceptual warrant entails the reliability of perception. What then are we to say of the victim of demonic deception? Does that individual, who has no idea that she is being deceived, continue to enjoy warranted perceptual beliefs, or does it only seem to her as if she does, but in fact none are warranted at all? Graham (2012) distinguishes three types of demon-world cases. First, there is the case of a disembodied soul imagined by Descartes, fooled by the evil demon. Second, there is a molecule for molecule duplicate brain to an actual human brain,

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    somewhere in the universe that bears no causal, explanatory relationship to any other brain, nor any causal, explanatory relationship to any subject-matter. Maybe it is a swamp brain, created a moment ago. Third, there is the normal human being, unbeknownst to her, placed in misleading circumstances, maybe by a powerful scientist with “demonic” intent. Graham, like the vast majority of working philosophers of mind, denies the first and the second cases as impossible. The third, however, is a genuine possibility, and maybe even in our future, given the advance of virtual reality technology. What should we say, then, about the third possibility? Does warrant persist in such a “demon world” scenario? In “Perceptual Entitlement” (2003) and in his chapter included here, Burge says that it does. Graham’s chapter then interrogates Burge’s account of perceptual warrant and Burge’s “transcendental” explanation for the persistence of warrant in demon worlds (the notorious section VI of “Perceptual Entitlement”). Graham argues that Burge’s explanation falls short. Graham contrasts two “Burgean” theories of warrant, one where warrant is “bounded” to normal conditions (it only exists in normal conditions) and the other is “unbounded” (it may persist outside of normal conditions, and so in demon worlds). Graham argues that Burge’s explanation fails to explain why warrant should be unbounded (why it should persist in demon worlds). Readers interested in understanding Burge’s “transcendental” argument and why it might fall short should find this part especially worthwhile. However, Graham goes on to argue that the overall resources of Burge’s view provide material to distinguish two grades of warrant. The first grade consists in the normal functioning of the belief-forming competence, but only if the competence is reliable in normal conditions. The first grade would persist in demon worlds, for normal functioning can persist outside of normal conditions. The first grade is unbounded. That’s why warrant should persist in demon worlds. The second grade would consist in reliability through normal functioning in normal conditions. The second grade would not persist in demon worlds, for in demon worlds perception ceases to cause and sustain reliably true beliefs. The second grade is bounded. The first grade would depend upon, or be constructed out of, the second grade. The first grade is a constituent of the second. That’s why, at root, the first grade requires reliability. This view has the consequence that only competences that are reliable in normal conditions contribute to (first grade) warrant in any conditions. A perceptual competence that is not even reliable in normal conditions would not contribute to empirical warrant in any conditions, never mind demon conditions. The first grade of warrant is thus a reliabilist kind of warrant, despite being unbounded. That’s why warrant—the first grade—should persist in demon worlds. The chapters by Brueckner and Altschul, Gerken, and Graham explicitly take up themes from Burge. The following chapters in Part II of the volume from Sosa, Carter and Pritchard, Hazlett, and Schechter take up issues in broadly competence-based, modest foundationalist epistemology. Sosa takes up his distinction between animal and reflective knowledge. Carter and Pritchard raise the possibility of extended entitlement. Hazlett takes up the plausibility of Moorean knowledge claims. And Schechter makes a detailed case against method internalism.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



1.6 Sosa: “Knowledge, Default, and Skepticism” In his contribution, Ernest Sosa further develops his distinctive approach in virtue epistemology, focusing on skepticism. When and why, he asks, are we default “justified” or “entitled” to presume, or take it for granted, that our belief-forming competencies are in the right situation for their successful performance? Given the possibility—especially the close possibility—that we are not in the appropriate situation for the successful performance of our cognitive capacities, the skeptic claims, we do not possess knowledge, even when we succeed and form a true belief through the successful exercise of a cognitive competence. Sosa disagrees. We can rightly ignore such possibilities—even close possibilities. He begins his chapter with an analogy. Consider a baseball outfielder playing in a night game. The outfielder can properly ignore the possibility that the lighting system might suddenly fail, even if they might due to poor maintenance. For monitoring for good lighting conditions is not a part of the outfielder’s competence. Catching fly balls, throwing to the right base, communicating with teammates, and so on, make up the fielder’s competence. The fielder can catch a fly ball due to competence, without ruling out the possibility that the lights might fail, even when the possibility is modally close. Monitoring for good lighting conditions is not a part of the fielder’s competence. That the lights are on is a background condition, properly taken for granted. But what about the fielder’s belief that he will catch the ball if he tries? There’s a good chance that belief will be false, for the lights might go out. Assuming they do not, does he know he will catch the ball if he tries? Though his fielding competence rightly ignores that the lights might go out, can his cognitive competence rightly ignore that possibility? A safety theorist says no. Though previously a safety theorist himself, Sosa now says yes. How does he make his case for an affirmative answer? Here’s an abbreviated version of a thought experiment he develops throughout the chapter to support his case. Suppose aliens flew close to the Earth. When they did, they considered putting all of us in the Matrix. They then ran a lottery. If we win the lottery—a one in a million chance—they leave us alone. So easily we’re in the Matrix. But we win the lottery, and they leave us alone. Suppose NASA learns of the flyby, and the lottery we very luckily won. Did we lack knowledge before NASA’s discovery? Do we lack knowledge after NASA’s discovery? Surely, Sosa thinks, the quality of our cognitive performances before and after NASA’s discovery is the same. Surely, he thinks, we know even though we only luckily won the lottery, just as the fielder can catch a fly ball due to competence only if only luckily the lights stay on. Our cognitive competences do not require monitoring for the possibility that such abnormal spoilers intervene. That we are not Matrixed is a background condition, properly taken for granted, absent considerations to the contrary. Through further analogies with pilots, surgeons, and those walking home at night, Sosa explores the category of default assumptions in domains of performance. Different domains of performance involve different competencies, with different default assumptions. Some competencies involve monitoring a wide range of circumstances, some do not. The fielder can rightly presume good lighting conditions, even if they might easily fail. We too can rightly presume normal conditions for our cognitive competencies, even if they too might easily fail.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    What then is a default assumption? In general, he argues, a default assumption is an assumption that a certain background condition obtains, one that (a) is required for the competence to be in place, but (b) can be in place unsafely, and unbeknownst to the agent. And how does this help with the skeptic? For many of our cognitive competences, given their proper domain, we need not monitor for the possibility that background conditions fail to obtain, that abnormal spoilers intervene, even if those possibilities are close possibilities.

1.7 Carter and Pritchard: “Extended Entitlement” Lisa uses the notebook app on her smartphone to keep track of what she needs to get when she goes grocery shopping. When she runs out of broccoli, she adds it to the list. When she purchases it, she crosses it off and it disappears from the list. Unlike Lisa, Zac doesn’t use a smartphone to keep track of what he needs in the grocery store. Instead, Zac relies purely on his innate memorial capacities. One might think that Lisa and Zac differ dramatically in the cognitive processes that they use when they go grocery shopping. Zac relies purely on his memory, and so his beliefs about what he needs to buy are basic, non-inferential beliefs; Lisa, on the other hand, relies on perception and inferential capacities (she infers that if ‘broccoli’ is displaying in her notebook app, this means that she ought to buy it), so her beliefs are non-basic, inferential beliefs. Zac’s memory beliefs paradigmatically enjoy entitlement, whereas Lisa needs a reason to rely on her smartphone. However, many philosophers in recent years have suggested that this common intuition is false. Lisa and Zac are both relying on memory, it’s just that Lisa is relying on an extended cognitive competence while Zac is not. This idea has become known as the hypothesis of extended cognition. Put more precisely in terms of a principle of parity, we can say: Parity Principle: If, as we confront some task, a part of the world functions as a process which, were it to go on in the head, we would have no hesitation in accepting as part of the cognitive process, then that part of the world is part of the cognitive process. (Clark and Chalmers 1998: 8)

Were Lisa’s process of recalling her shopping list occurring as mental projection within her mind, we would not hesitate to say that Lisa is remembering what she needs to buy. That this process is occurring outside of her mind should make no difference. J. Adam Carter and Duncan Pritchard, in their contribution, aim to provide further support to the hypothesis of extended cognition by arguing that, all else being equal, if the only difference between Lisa’s and Zac’s cognitive processes is that Lisa’s is extended while Zac’s is not, then Lisa and Zac should enjoy the same kind of positive epistemic status. Not only should Lisa’s and Zac’s beliefs about what to buy be equally warranted, but the source of their warrant should be identical. That is, their warrants should equally be a basic, non-inferential defeasible warrant. Why might we think that Lisa’s epistemic source is basic? Carter and Pritchard point out that if we take hypothesis of extended cognition seriously, and if we assume, per hypothesis, that Lisa and Zac’s respective cognitive processes are “psychologically on a par,” then we need to stipulate that Lisa’s cognitive process, much like Zac’s, is

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



reliably available, easy to access, automatically endorsed, and was previously endorsed by Lisa. Insofar as Lisa always has her phone on her, is adept at using her phone, does not question whether the things on her list are genuinely things she needs, and adds things to her list because she wants to be able to recall them later, then the way that Lisa uses her phone is the same as the way that Zac uses his biologically based capacity for recall. One still might think that there is an important disanalogy between Lisa and Zac’s respective cognitive processes. Because Zac’s belief that he needs broccoli is sourced in biological memory, he is likely to remember why he needs broccoli (he will, for example, remember running out of it). Lisa, on the other hand, is unlikely to remember why she needs broccoli (is it because she ran out, or is because she is making something that requires more broccoli?). Biological memory, one might think, provides reasons for the beliefs that it supports. Against this, Carter and Pritchard point out that, following Burge (1993), one can be warranted in one’s memory-based belief even if one does not recall why one remembers the proposition in question, even if one does not have the meta-belief that one’s memory is reliable. For example, Zac might remember that he needs chickpeas, but can’t remember why he needs chickpeas, only that he does. (When he gets home he sees his cookbook open to the hummus recipe and says to himself, “That’s why I needed chickpeas!”) The experience might be even more familiar if we consider many of the things that we learned in school. We might remember many of the theorems we proved in geometry class in high school, but not remember any of the actual proofs we did for them. Memory, Burge says, provides us with a pro tanto entitlement to form a belief based on that memory, regardless if we remember the basis of the original belief (did I prove the theorem, or did I read it in a textbook?). Lisa not being able to recall why she put broccoli on the “to buy” list does not diminish the epistemic status of the belief she formed. Just like Zac, Lisa is entitled to believe that she needs to purchase broccoli.

1.8 Allan Hazlett: “Moorean Pragmatics, Social Comparisons and Common Knowledge” Consider the following common argument for radical skepticism: 1. You know that you have hands only if you know that you’re not deceived by a malicious demon. 2. You don’t know that you’re not deceived by a malicious demon. 3. Therefore, you don’t know that you have hands. According to commonsense, premise two of this argument is false. I know that I’m not deceived by a malicious demon; I know this even though I can’t provide any evidence for it. I can’t prove that I’m not being deceived by a malicious demon. But just as I don’t need to prove that my perceptual capacities are reliable in order to be entitled to rely on them, I don’t need to be able to prove that I am not being deceived by a malicious demon in order to know that I am not.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    But to many, premise two doesn’t seem to be false. In fact, to many it seems to be true. This is, after all, what makes it a powerful argument. But if it is false, why does it seem to be true? In this chapter, Allan Hazlett provides an error theory of sorts for why it might seem that we don’t know that we’re not deceived by a brain in a vat, even though we do know that we are not. The reason for this, firstly, is that it is common knowledge that we are not being massively deceived, and secondly, in some contexts, knowledge attributions (i.e., to assert, “S knows that p”) of common knowledge, even when true, will seem false. Hazlett will then supplement the case for moderate foundationalism by removing an intuitive obstacle to its acceptance. There are three steps in Hazlett’s argument. In the first step, Hazlett argues that “knowledge attributions sometimes seem false when they are heard as having misleading implications.” In certain contexts, when an individual asserts that S knows that p, the individual is doing more than merely asserting that S has a belief which would not easily have been false. In certain contexts, for you to say that S knows that p will carry with it the implication that S can provide grounds for her belief that p. If it turns out that S cannot provide grounds then, given the context, those you told this to will believe what you said to be false, even if true. The second step in Hazlett’s argument concerns the function of knowledge attributions. Hazlett notes that knowledge attributions can serve many functions. They can, for example, serve to flag reliable informants or signal the end of inquiry. Hazlett does not wish to deny that knowledge attributions may serve these functions. However, knowledge attributions can also function as a social comparison. Sometimes, when you claim that S knows that p, you are claiming that S stands out among her epistemic peers; S unlike R knows that p. If knowledge attributions are sometimes (heard as) false when they are heard as having misleading implications, and if one implication of a knowledge attribution can be that the knower stands out epistemically among her peers, then knowledge attributions will sometimes be (heard as) false when it is false that the knower stands out epistemically among her peers (even if it is true that S knows that p). The third step in Hazlett’s argument concerns the idea of common knowledge. According to Hazlett, a proposition p is common knowledge among a group if every member of that group knows that p and no member of that group has a better epistemic standing regarding p than any other member of that group. When a knowledge attribution concerns a piece of common knowledge, then in contexts where knowledge attributions imply that the knower stands out epistemically among her peers, such attributions will seem false since a knower can’t stand out epistemically among her peers vis-à-vis a piece of common knowledge (since her peers will be in equal epistemic standing to her). Hazlett suggests that the reason it seems true that we don’t know that we’re not being massively deceived is that an assertion to the contrary would be an attribution of knowledge regarding a piece of common knowledge with the implication that we, the knower, stand in greater epistemic standing with respect to this proposition than others. But of course, this is false. We all know that we are not being massively deceived.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



1.9 Joshua Schechter: “Internalism and Entitlement to Rules and Methods” Joshua Schechter explores different explanations for why we are pro-tanto justified in forming beliefs in accordance with basic belief-forming methods (BFMs), such as modus ponens, inference to the best explanation (IBE), and transformation rules of our perceptual system. Schechter’s main target in his essay is so-called method internalist accounts of justification to rely on BFMs. Schechter considers three ways of defending method internalism. The first, extreme method internalism, appeals to analogies concerning the use of instruments. Intuitively, when we read a thermometer and form a belief about the current weather, we are pro-tanto justified in this belief because we have a justified belief that the thermometer is reliable. Perhaps the source of our justification in relying on BFMs is similar to the source of our justification in relying on such instruments, namely a justified belief that our BFMs are reliable. The problem with extreme method internalism is that it gives rise to an explanatory regress. If our justification in relying on BFMs is explained by a justified belief in their reliability, the internalist must explain why this further belief is justified. The internalist has two options: she can appeal to further justified beliefs, in which case she will be stuck in an “infinite explanatory chain,” or she can appeal to some further belief-forming method, in which case she is stuck in a viciously circular explanatory chain. The second way of defending method internalism appeals to what would intuitively make an individual epistemically responsible, namely possessing the capacity to defend one’s belief. Insofar as proper defense is a justified one, on what Schechter calls Defense Internalism, one is justified in relying on one’s BFMs only if one “has an independently justified defense of the method.” Schechter identifies four problems with defense internalism. First, it is unclear how relying on a method is akin to “staking a claim,” which in general is what one would need a defense for. Second, defense internalism faces the same regress problems that extreme method internalism faces. Third, it seems that no defense could be provided for many basic BFMs, such as modus ponens and IBE, for any such defense would itself appeal to the very rules that are being defended. Fourth, and finally, defense internalism would require that an individual possess an infinite number of justified defenses. The third and final rationale for defending method internalism concern our reactions to skeptical scenarios. Insofar as our phenomenal experience is indiscernible from that of a brain in a vat, it is intuitive to think that we are no more justified in relying on our BFMs than is a brain in a vat. The mental internalist argues that the explanation for this symmetry is the fact that justification supervenes on our mental states. Insofar as my envatted counterpart has the same mental states that I do, we have the same justification for our respective beliefs. While Schechter admits he doesn’t have a compelling objection to mental internalism (but compare Burge’s chapter), he does think we have reason to resist it, first, because our intuitions about skeptical scenarios are sometimes opaque and

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    equivocal, and second, because there are better accounts of why we are justified in relying on BFMs. Schechter argues that the best strategy for explaining why we are justified in relying on BFMs is to appeal to process accounts of justification, such as proper function reliabilism (e.g. Burge, Graham), competence reliabilism (e.g. Sosa), and Schechter’s preferred approach, the rational obligations account, according to which our pro-tanto justification for reliance on BFMs is explained by the fact that relying in such BFMs is constitutive of the rational activities we are rationally obligated to engage in (e.g., Enoch and Schechter). What these accounts have in common is that they explain basic epistemic norms in terms of the role that following these norms play in certain normative processes. But in explaining this, all of these accounts make use of, in different ways, features of the world that are external to an epistemic agent’s point of view. Schechter’s chapter completes Part II of the volume. The five chapters in Part III all directly engage Crispin Wright’s “entitlement” project, where Wright sets out to show, a priori, that we have a rational but non-evidential right to accept certain cornerstone propositions (a “Wright-entitlement”), and so a right to claim to know these propositions, against the skeptic. In different ways, all five directly challenge the viability of Wright’s project.

1.10 Martin Smith: “Full Blooded Entitlement” ‘Entitlement,’ as Wright uses the term, picks out an unearned epistemic good, a right to assume, trust, or take for granted. But what kind of unearned epistemic good is it? Martin Smith, in his chapter, argues that if entitlements are to do their anti-skeptical work, they must do the same kind of epistemic work as justifications. This means that an entitlement is an unearned justification. Smith calls such an entitlement a fullblooded entitlement. He contrasts this with anaemic entitlement and moderate entitlement, both of which Smith argues are unstable. Common to all three notions of entitlements is the idea of an unearned epistemic good. The distinguishing mark of anaemic entitlement is that the epistemic good is one that does not give rise to a right to believe. Rather, anaemic entitlement underwrites something less than belief, perhaps trust towards the truth of a proposition, or some other such attitude of lesser conviction. Smith argues that entitlement, so understood, would violate closure. Here is Smith’s idea. The purpose behind appealing to entitlement is to defeat the skeptic. So, assuming the skeptic is defeated, we can assume that we are justified in believing, say, that I have two hands. But the proposition I have two hands, entails the proposition I am not a handless brain in a vat. If justification is closed under known entailment, if I am justified in believing the former then I am justified in believing the later. But by hypothesis I only had anaemic entitlement toward the latter proposition, not a justification to believe. Moderate entitlement, unlike its anaemic cousin, does give rise to a right to believe. However, unlike justification, moderate entitlement cannot “supply the needs of knowledge.” That’s what makes it “moderate.”

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



Smith challenges the proponent of moderate entitlement by asking the following question: why is it that moderate entitlement (plus an anti-Gettier condition) cannot get us to knowledge? Answering this question presents a dilemma to the proponent of moderate entitlement. Perhaps the reason that moderate entitlement cannot get us to knowledge is that knowledge is simply beyond our reach. But if this is the case, then entitlement would be parallel to justification, albeit an unearned justification. If, however, knowledge is not beyond reach, then moderate entitlement, like anaemic entitlement, will violate closure, for if knowledge is not beyond reach, then presumably, if I know anything, I know that I have hands (if I don’t know this, then we clearly haven’t, contra hypothesis, defeated the skeptic). But the proposition I have hands entails the proposition I am not a handless brain in a vat. But, per hypothesis, that I am not a handless brain in a vat is something I am only entitled to believe, and so not something I can know. If we want to be an entitlement theorist, it would appear that the only option we have left is full blooded entitlement. But then, it seems, Wright’s distinction between entitlement and justification threatens to collapse.

1.11 Nikolaj Jang Lee Linding Pedersen: “Pluralist Consequentialist Anti-Skepticism” A cornerstone proposition c for a given region of thought is a proposition such that warranted belief in c is necessary for warranted belief in any proposition concerning that region of thought. The proposition I am not a brain in a vat is a cornerstone proposition for warranted beliefs about the empirical world. A response to the skeptic is concessive if it grants to the skeptic that one cannot have evidence for the truth of cornerstone propositions. In this chapter, Pedersen, like Wright, asks how one could be warranted and rational in accepting cornerstone propositions in the absence of evidence supporting such propositions. How, in other words, is Wright-style entitlement possible? Critics (Pritchard (2005) and Jenkins (2007)) have argued that Wright has no good response to this question, that such entitlement is impossible. Pedersen develops an alternative to Wright-style entitlement, approaching the question from an epistemic consequentialist perspective. Warrant and rationality attach to epistemic strategies that maximize expected epistemic value. If truth is the only epistemic good, then in order for there to be a consequentialist warrant for accepting cornerstone propositions, it had better be the case that accepting the cornerstone propositions maximizes expected true beliefs. However, assuming that we can grant no greater probability to, say, not being a BIV than we can to being a BIV, no value-maximizing strategy emerges for accepting cornerstone propositions (for, as it turns out, both accepting and not accepting cornerstone propositions generate zero epistemic value on this monistic conception of epistemic value). And because no value-maximizing strategy emerges, there can be no consequentialist account of cornerstone warrant and rationality. Pedersen argues, however, that if we abandon monism about epistemic value and instead embrace epistemic pluralism—where error avoidance, coherence, and

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    meta-cognitive truth-link are also epistemic values we should aim to maximize—we will find that there is a consequentialist warrant that attaches to belief in cornerstone propositions. Consider, as an example, the epistemic good of coherence. Accepting the cornerstone proposition that I am not a brain in a vat coheres better with the rest of my beliefs than does accepting that I am a brain in a vat, regardless of the truth of these propositions. So accepting this cornerstone proposition would maximize the epistemic value of coherence. Insofar as this strategy delivers a genuine warrant and rationality for acceptance of cornerstone propositions, it would seem as though there is simply no need for Wright’s controversial notion of non-evidential entitlement to provide a concessive response to the skeptic. Pedersen argues that, while the conditions on Wright-style entitlement may be satisfied, this does not suffice to account for the positive epistemic standing of acceptance of cornerstones. Instead appeal is made to a pluralist consequentialist—story. This shows that the real epistemological work is done by this alternative story, not by entitlement. Pedersen concludes that there is no need for Wright-style entitlement, conceived as a species of warrant.

1.12 Annalisa Coliva: “Against (Neo-Wittgensteinian) Entitlements” Coliva’s chapter takes a critical stance toward Wright’s view that appealing to default or unearned warrant can provide us with a response to the skeptic. Coliva argues that (1) Wright’s view is not genuinely Wittgensteinian, and (2) regardless of its genealogical lineage, Wright’s articulation of the idea of an unearned warrant is problematic on its face. Coliva argues that Wittgenstein offers two arguments against skepticism. According to the first argument against Descartes’ “dream skeptic” from Mediation, the skeptical judgment “I am now dreaming” cannot be intelligibly maintained. Either the judgment occurs in a dream or it does not. If it occurs in a dream then such a judgment would be meaningless, for such judgments would occur outside the context of a language game. But a meaningless judgment could not be the basis of a rational epistemic argument for rejecting belief in the external world. If, however, the judgment occurs in waking life, then the coherence of our waking life would be sufficient to refute the judgment. Dream skepticism, suggests Coliva’s Wittgenstein, falls flat on its own two feet. Wittgenstein’s second argument attacks the “Humean skeptic” who holds that we cannot know that we have hands because we cannot know that we are not brains in a vat. Much like with dream skepticism, Wittgenstein argues that Humean skepticism is self-undermining. According to Coliva’s Wittgenstein, skeptical doubts can rationally be raised only when there exist reasons to raise them. But such reasons necessarily can only exist in the context of a language game. So skeptical doubts can rationally be raised only when it constitutes an appropriate move in the language game. Now a language game, much like the game of baseball, has certain rules that constitute or designate correct moves in the game. Just as it is nonsensical or meaningless to ask why three strikes rather than four constitute an out, it is similarly

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



nonsensical to question the fundamental rules or norms of a language game. Coliva suggests that the fundamental rules of the language game that we play include the rule that we take for granted that there exists an external world. While Wittgenstein doesn’t hold that we must play this language game, there can be no rational reason for transitioning to a different game since what counts as rational is internal to the game we are playing. The only way we can ask skeptical questions is by playing a language game, but the question is nonsensical by the very rules of the game we are playing. Coliva argues that Wright’s notion of epistemic entitlement as unearned warrant is not Wittgensteinian, most perspicuously because, unlike entitlements, the fundamental norms of our language games are not epistemic. Rather, these constitutive norms delineate the epistemic; they are the conditions that make epistemic inquiry possible. Whereas Wright characterizes entitlements as rational, Wittgenstein characterizes the fundamental norms of our language game as non-rational. Whatever their relation to Wittgenstein, Coliva argues that the notion of entitlement as unearned warrant ought to be rejected for the notion is an “utterly mysterious” one. Unlike the more familiar idea of propositional justifications, which are tied to epistemic achievements (such as the proper exercise of our perceptual systems), entitlements exist in the so-called “space of reason,” untethered to our epistemic practices, but somehow rationally supporting them. How, Coliva asks, can trusting such a proposition be rational? And if not, how can they rationalize empirical beliefs?

1.13 Daniel Elstein and Carrie Jenkins: “The Truth Fairy and the Indirect Consequentialist” Jenkins (2007) argues that the prospects of accounting for the epistemic rationality of cornerstone acceptance look rather bleak for proponents of Wright-style entitlement. Jenkins dismisses a number of strategies, including the strategy of counting X as epistemically warranted and rational if X maximizes expected epistemic value, pace Pedersen. This is a form of direct epistemic consequentialism because X’s epistemic standing is determined directly by the epistemic consequences of X itself. She dismisses direct consequentialism on the grounds that it falls prey to the “Truth Fairy problem.” The problem is this: suppose that S has evidence neither for nor against p. Furthermore, suppose that S encounters the Truth Fairy, a powerful being, who promises S many good epistemic consequences and very few—if any at all—bad epistemic consequences if she accepts p. In this case direct consequentialism issues the verdict that S is epistemically warranted and rational in accepting p. This seems like a hard pill to swallow. In their chapter, Elstein and Jenkins explore whether rule consequentialism might give the proponent of entitlement the resources to account for the epistemic rationality of cornerstone acceptance and respond to the Truth Fairy problem. Rule consequentialism can be stated as follows: (RC) X is epistemically warranted and rational if and only if X is allowed (or recommended) by the rule set R the internalization of which maximizes epistemic value.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    Rule consequentialism is a form of indirect epistemic consequentialism because the epistemic standing of X is not tied directly to the epistemic consequences of X itself but rather to a set of rules. Elstein and Jenkins consider three versions of rule consequentialism: (RC1)

(RC2)

(RC3)

Strict evidentialism: Adopt, without exception, the rule: accept p if and only if you have sufficient evidence that p. Flexible evidentialism: Adopt evidentialism (accept p if and only you possess sufficient evidence that p) as a rule but allow exceptions to the rule, sustained by other specific rules. Rule-based direct consequentialism: Adopt, without exception, the rule of direct consequentialism: accept p if and only if doing so maximizes expected utility.

Elstein and Jenkins observe that (RC1) will not do from the point of view of proponents of entitlement. It would lead to skepticism because the entitlement theorist agrees with the skeptic that there can be no such thing as evidential warrant to accept cornerstone propositions. (RC3) will not do either because it falls prey to the Truth Fairy problem. This leaves (RC2). What specific rules might sustain exceptions to the evidentialist rule? Elstein and Jenkins suggest adding specific rules that recommend accepting individual cornerstone propositions (the reliability of induction, the general trustworthiness of the senses, the existence of the external world, not being a BIV, etc.). They take seriously the prospects of this version of epistemic consequentialism being of help to the entitlement theorist. The shape of (RC2) aligns with Wright’s framework. The evidentialist rule can be taken to capture instances of justification, i.e. evidential warrant. The specific rules that create exceptions to the evidentialist rule can be taken to capture entitlements, i.e. non-evidential warrant. While there is no evidence supporting entitled trust, such trust nonetheless gets classified as epistemically warranted and rational. The specific rules that recommend trust in cornerstones are members—together with the evidentialist rule—of the set of rules whose internalization maximizes expected utility. At the same time (RC2) steers clear of the Truth Fairy problem: neither the evidentialist rule nor the specific “cornerstone rules” recommend accepting the propositions offered by the Truth Fairy.

1.14 Patrick Greenough “Knowledge for Nothing” Greenough’s chapter offers an extensive discussion of Wright-style entitlement. Greenough distinguishes between four kinds of entitlement epistemology. The various types differ with respect to their commitments vis-à-vis the internalism/externalism divide, the place accorded to knowledge in epistemology, and their ability to address different forms of skepticism. Type i entitlement epistemology—a type identifiable in certain parts of Wright’s work—involves a concession to the Cartesian skeptic: skeptical arguments of

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



the Cartesian sort show that empirical knowledge is impossible. Thus, the Type I entitlement epistemologist embraces a knowledge nowhere epistemology: the focus of epistemology should not be knowledge or other external states, but rationality, justification, entitlement, or warrant construed along internalist lines. However, while empirical knowledge is impossible, our empirical beliefs—ordinary as well as cornerstone ones—are nonetheless warranted. This is what Wright has labeled the “Russellian Retreat” (Wright (1991: 88)). Still, a certain kind of endogenous skepticism remains. There can be no such thing as an evidential warrant to believe cornerstone propositions because the best attempts to acquire such warrant are bound to fail for principled reasons, due to epistemic circularity. (Wright takes the appropriate attitude for cornerstones to be trust, a species of acceptance different from belief, as we saw above. Greenough operates in terms of belief and gives a reason for doing so. In introducing his chapter we follow his path and speak in terms of belief.) The best attempts to acquire an evidential warrant to accept cornerstone propositions take the form of a so-called I-II-III argument, i.e. arguments of the following form: I My experience is in all respects such that it appears (to me) that I have hands. II I have hands. III There is an external world. The transition from II to III is sustained by the suppressed premise that, if I have hands, then there is an external world. This is taken for granted: it ranks as a conceptual truth, given an ordinary conception of hands as being the kind of thing located in the external world. Wright is a conservative regarding the transition from I to II. He thinks that there can be evidential warrant to believe ordinary empirical propositions such as II, but that this warrant cannot be acquired purely on the basis of I. (This commitment contrasts with the dogmatism of Jim Pryor, e.g. Pryor (2000).) The warrantability of the transition from I to II is held hostage to an antecedent, independent warrant to believe the type III proposition, the cornerstone. Cornerstone propositions constitute an authenticity-condition: they say that the attendant circumstances are authentic or conducive to enquiry. Thus, absent a warrant to believe a cornerstone proposition, there is no warrant to think that experiences represent—rather than misrepresent—the world. I-II-III arguments cannot be a means of acquiring an evidential warrant to believe the type III proposition, by way of I and II serving as evidence for III. For, the transition from I to II depends on there being an antecedent warrant to believe III, and so, due to epistemic circularity, the I-II-III argument misfires. However, since I-II-III arguments are supposed to constitute the best attempts to acquire an evidential warrant to believe cornerstone propositions, this means that no attempt can succeed. Hence, there can be no such thing as an evidential warrant to believe cornerstone propositions. On the basis of this line of argument I-II-III skeptics conclude that there can be no non-circular evidential warrant to believe cornerstone propositions. Type I entitlement epistemologist endorses the Russellian Retreat and, so, thinks that both ordinary empirical beliefs and empirical cornerstone belief are warranted. Ordinary empirical beliefs are evidentially warranted (construed along conservative

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    lines). However, since it is granted that the I-II-III skeptic is right that there can be no evidential warrant to believe cornerstone propositions, the only strategy for accommodating cornerstone warrant is to buy into a non-evidential type of warrant. Greenough argues against the tenability of Type I entitlement epistemology. Consider I know that there is an external world—what might aptly be called a “higher-order cornerstone proposition.” It is a cornerstone because doubting commits one to doubting a wide range of other propositions (e.g., I know that I have hands), and it is higher-order because it embeds an occurrence of “know.” Now, according to Greenough, since it is a cornerstone, it must qualify as an entitlement. In turn, since entitlement is non-evidential warrant, it is warranted. However, recall that Type I entitlement epistemology is concessive towards Cartesian skepticism: there can be no empirical knowledge. So, I don’t know that there is an external world. But the Cartesian skeptic’s line of argument justifies the belief I don’t know that there is an external world. Justification is evidential warrant, and so, I don’t know that there is an external world is warranted. Putting the two together: both I know that there is an external world and I don’t know that there is an external world are warranted. However, no tenable epistemology allows both a proposition and its negation to be warranted. What gets Type I entitlement epistemology in trouble? The Russellian Retreat, it would seem. It is the concession to the Cartesian skeptic—empirical knowledge is impossible—that lays the foundation for the warrant for I don’t know that there is an external world. In light of this one might think that it would be advisable for entitlement epistemologists to abandon the Russellian Retreat. Greenough identifies two strands in Wright’s work on entitlement that suggest precisely this kind of rejection. He uses these two strands to formulate what he calls respectively “Type III Entitlement Epistemology” and “Type IV Entitlement Epistemology.” There are significant differences in detail between these two epistemologies. However, for present purposes it is crucial to note that they are unified by the combination of a rejection of the first-order Russellian Retreat and an endorsement of a meta-Russellian Retreat. It is possible to know that there is an external world. This is implied by the rejection of the first-order Russellian retreat. Additionally, while it is impossible for me to know that I know that there is an external world, I am nonetheless warranted in believing that I know that there is an external world. This is the meta-Russellian Retreat. Greenough maintains that both Type III and Type IV entitlement epistemology are unsustainable, pressing a worry structurally similar to the one pressed against Type I entitlement epistemology. According to Greenough the most promising form of entitlement epistemology is Type II entitlement epistemology. Like Type III and IV entitlement epistemology (and unlike Type I entitlement epistemology) it involves a rejection of the first-order Russellian Retreat, and so, a rejection of first-order Cartesian skepticism. However, unlike Type III and Type IV entitlement epistemology Type II involves no Russellian Retreat at any level. This means that there is no concession to the Cartesian skeptic (or Cartesian meta-skeptic) at any level. Empirical knowledge is not impossible at any level, first-order or higher. Greenough’s proposal is that this anti-skeptical line is to be sustained by minimalism about knowledge. Minimalism about knowledge is a minimal form of externalism. It says that (empirical) knowledge does not merely

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

, , ,  



consist in the internal states of the subject and the truth of the proposition believed by the subject. There is likewise an external component: the world must cooperate more than making the relevant proposition true. However, minimalism remains silent or neutral on the exact nature of this external component—therein lies the minimalism (and the potential to circumvent challenges leveled against more substantive, robust forms of externalism—reliabilism, say). Minimalism about knowledge is intended to be internalism friendly. It is compatible with internalism about warrant, i.e. with the idea that the internal states of the subject determine warrant. This is indeed the package that Type II entitlement epistemology involves: minimal about knowledge (i.e. a lightweight form of externalism) and internalism about warrant. Minimalism about knowledge is meant to address first-order Cartesian skepticism and its higher-order counterparts. Entitlement, construed along internalist lines, is meant to deliver a response to I-II-III skepticism. According to Wright entitlement is warrant for nothing. Type II entitlement epistemology makes available a parallel story about cornerstone knowledge. Cornerstone knowledge involves three things: entitlement, truth, and minimal externalism. Entitlement is for nothing. If the world cooperates, truth comes for nothing and so, too, does the minimal externalist condition. Hence, knowledge is for nothing. One of the main messages of Greenough’s contribution is that it is a mistake to try to externalize entitlement, i.e. to make the unearned warrant applicable to cornerstones subject to external conditions. Entitlement is a purely internal matter. However, as seen, Greenough still takes knowledge to be subject to a minimal externalist condition. Type II entitlement epistemology is thus pluralist twice over. Like Type I, Type III, and Type IV entitlement epistemology it accommodates two species of warrant: evidential warrant and non-evidential warrant, i.e. justification and entitlement. Additionally, Type II entitlement epistemology incorporates a lightweight form of externalism in the form of minimalism alongside internalism about entitlement. Altogether the five chapters in Part III offer sustained critical engagement with Wright’s entitlement project in epistemology.

References Brueckner, A. 2007. Content Externalism, Entitlement, and Reasons. In S. Goldberg (ed.), Internalism and Externalism in Semantics and Epistemology. Oxford: Oxford University Press. Burge, T. 1993. Content Preservation. The Philosophical Review 104(4): 457–88. Burge, T. 2003. Perceptual Entitlement. Philosophy and Phenomenological Research 67: 503–48. Burge, T. 2005. Truth, Thought, Reason: Philosophical Essays, Volume 1. Oxford: Oxford University Press. Burge, T. 2007. Foundations of Mind: Philosophical Essays, Volume 2. Oxford: Oxford University Press. Burge, T. 2010. Origins of Objectivity. Oxford: Oxford University Press. Burge, T. 2011/2013. Epistemic Warrant: Humans and Computers. In T. Burge, Cognition through Understanding: Self-Knowledge, Interlocution, Reasoning, Reflection: Philosophical Essays, Volume 3. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    Burge, T. 2013. Cognition through Understanding: Self-Knowledge, Interlocution, Reasoning, Reflection: Philosophical Essays, Volume 3. Oxford: Oxford University Press. Clark, A. and Chalmers, D. 1998. The Extended Mind. Analysis 58: 10–23. Graham, P. J. 2012. Epistemic Entitlement. Noûs 46: 449–82. Graham, P. J. 2014. Functions, Warrant, History. In A. Fairweather and O. Flanagan (eds.), Naturalizing Epistemic Virtue. Cambridge: Cambridge University Press. Graham, P. J. forthcoming. What is Epistemic Entitlement? In J. Greco and C. Kelp (eds.), Virtue-Theoretic Epistemology. Cambridge: Cambridge University Press. Jenkins, C. S. I. 2007. Entitlement and Rationality. Synthese, 157: 25–45. Millikan, R. 1984. Language, Thought, and Other Biological Categories. Cambridge: MIT Press. Pritchard, D. 2005. Wittgenstein’s On Certainty and Contemporary Anti-scepticism. In D. Moyal-Sharrock and W. H. Brenner (eds.), Readings of Wittgenstein’s On Certainty. London: Palgrave Macmillan. Pryor, J. 2000. The Skeptic and the Dogmatist. Noûs 34: 517–49. Wright, C. 1991. Scepticism and Dreaming: Imploding the Demon. Mind 100: 87–116. Wright, C. 2004. Warrant for Nothing (and Foundations for Free)? Proceedings of the Aristotelian Society 78 (suppl.): 167–212. Wright, C. 2014. On Epistemic Entitlement (II): Welfare State Epistemology. In D. Dodd and E. Zardini (eds.), Scepticism and Perceptual Justification. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

PART I

Engaging Burge’s Project

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

2 Entitlement The Basis for Empirical Epistemic Warrant* Tyler Burge

Most of what we know is supported by beliefs formed fairly directly from sense perception. Many such beliefs constitute knowledge. Over its history, philosophy has not done well in characterizing these beliefs or in saying wherein they are knowledgeable. One reason is that philosophy has done poorly in characterizing perception. A science of perception was not available until the late nineteenth century, and did not bloom into a rich enterprise until the 1970s. Some of the early mischaracterizations are perhaps to be expected. However, even since the 1970s, philosophy has mostly done poorly in describing perception and in providing an epistemology of empirical belief, because philosophy has not used science well. Another reason is that philosophy did not focus squarely on ordinary perception-based beliefs until the second half of the twentieth century. The primary topics of epistemological reflection, from Plato through Kant, were more impressive sorts of knowledge. Roughly, the topics were scientific knowledge, including mathematical knowledge, and knowledge gained through reflective understanding. Ordinary sorts of knowledge have not, until the last half-century, been as squarely targeted. The traditional focus on knowledge deluxe is defensible. But it has led to hyper-intellectualized accounts of the first knowledge obtained through sense perception. That knowledge is not knowledge deluxe. A satisfying history of why these distortions occurred has yet to be written. Sometimes it was just assumed that if knowledge closely connected to sense perception is to support rich kinds of empirical knowledge, it must itself be rich, or in some way golden. That assumption is mistaken. Richness accrues from cross-checking, testing, ruling out alternatives, systematization, mathematicization, reflection. Science expands from its humbler beginnings. Concern with scepticism also fed the distortion. Scepticism is a rich tool for provoking search for philosophical understanding. In the development of ordinary and even scientific knowledge, however, scepticism is a curiosity, not a rite de passage. Undermining scepticism is not a requirement on knowledge. Many traditional * I am indebted to John Carriero and Barbara Herman for improvements in Section 2.1; to David Chalmers for catching a non-deductive transition in a purportedly deductive argument in an oral presentation of Section 2.2; and to Gary Ebbs and Michael Rescorla for improvements in Section 2.3. I am especially indebted Johannes Burge for valuable comments on several predecessors of Section 2.5. Tyler Burge, Entitlement: The Basis for Empirical Epistemic Warrant In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0002

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     philosophical accounts of knowledgeable perceptual beliefs build in purported defenses against scepticism, to their detriment. Whatever the causes, most of the philosophical tradition badly mischaracterizes wherein beliefs most immediately obtained from sense perception are knowledgeable. Much current philosophy still does so. Here I focus on a key normative property that beliefs must have if they are to constitute knowledge. The property is epistemic warrant. The work here comprises six sections. Section 2.1 states some normative principles of epistemic warrant and distinguishes two types—justification and entitlement. Justification is warrant through reason. Entitlement is warrant without reason. It is the primary type that applies to perceptual beliefs. Section 2.2 discusses the role of reliability in warrant, and the extent to which entitlements to perceptual beliefs depend on factors external to a believer’s psychology. Section 2.3 considers some issues about scepticism. Section 2.4 criticizes an argument that believing that we are entitled to perceptual beliefs would commit us to an unacceptable way of validating the reliability of those beliefs. Section 2.5 rebuts an argument that believing that we are entitled to perceptual beliefs is inconsistent with intuitions about confirmation and with Bayesian principles of subjective probability. Section 2.6 concludes.

2.1 Epistemic Norms, Justification, and Entitlement Much work on the epistemology of empirical belief in the last sixty years has focused on its epistemic structure. A modest foundationalism has been a common and salutary upshot. Little of this work has discussed the psychological capacities and the functions and norms for those capacities that make the structure what it is. Form follows function and norms, which follow capacities. In this section, I hope to take some steps toward remedying this situation. I begin by locating epistemic warrant in a normative structure.¹ Norms are standards for success, or for contributing to success, in fulfilling a function, purpose, or goal. Goods, in the broad sense relevant here, are successes, or contributions to success, in fulfilling a function, purpose, or goal. Hence goods are fulfillments, or contributions to fulfillments, of norms. ‘Epistemic’ is etymologically rooted in a Greek word for knowledge, and earlier in an Ionic Greek word for understanding things about the world.² Epistemic warrant is an unreducible, normative, epistemic property of instances of propositional psychological states or events—prototypically beliefs, and occurrent thoughts with some

¹ Several of the ideas in the account in this section are present in my ‘Perceptual Entitlement’, Philosophy and Phenomenological Research 67 (2003), 503–548. The account here is more fully developed. ² Here, I take ‘understanding’, in the relevant sense, to be factive, like ‘knowledge’. Indeed, it is a type of knowledge that makes explanatory connections, especially to relatively basic principles. Understanding in this sense should not be identified with minimal competence understanding of representational content. It should not be identified with rich explicative understanding of representational content, though it is closely related. Like ordinary propositional knowledge, understanding here is understanding the world, not understanding representational content. Understanding the world, however, consists in understanding approximately true, relatively deep, propositions or systematic theories.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



assertive or at least positive-leaning valence by the thinker. Epistemic warrant can also attach to instances of transitions that lead to such beliefs and thoughts.³ It can, yet further, attach to instances of reliance on psychological capacities that make use of such beliefs and thoughts—such as reliance on propositional memory or on reflection. Epistemic warrant is a normative property in that it signifies that beliefs, thoughts, inferences, or instances of reliance meet a standard for being good. It is an epistemic property in that meeting the standard constitutively contributes toward, or constitutively conduces to, having knowledge. Epistemic warrant constitutively conduces to having knowledge in at least these respects: (a) A state or event’s being epistemically warranted is constitutively necessary for its being knowledge. (b) A state or event’s being epistemically warranted supports that state or events’ being knowledge. Being epistemically warranted supports having knowledge through that state or event in that, if the epistemically warranted state or event (say, the belief ) is true, if no Gettier conditions obtain, if there is no internal incoherence, and if no further epistemic warrant is needed for its being knowledge, then the belief is knowledge.⁴ (c) It is impossible, in any case, that a belief ’s epistemic warrant could block the belief from being knowledge. In fact, in many cases, epistemic warrant conduces to knowledge in a stronger way. In many cases, epistemic warrant provides at least a prima facie certification of knowledge. Constitutively, for some true beliefs, an epistemic warrant certifies knowledge in that it suffices for knowledge, absent counter-warrants, internal incoherence, and Gettier failures. I think that canonical or basic warrants for epistemically basic beliefs certify true beliefs, prima facie, as knowledge, in the foregoing sense. Examples of basic epistemic warrants for epistemically basic beliefs are (i) warrants for perceptual beliefs through forming them directly from perceptual states and (ii) warrants for logical and mathematical beliefs through understanding them. In sum, basic epistemic warrant for a true perceptual belief certifies it as knowledge, under the stated conditions. In the cases of basic epistemic warrants for beliefs in simple truths of logic and mathematics—that is, warrants through understanding them—being epistemically warranted suffices for knowledge. The prima facie certifying condition does not apply to all cases for warranted true belief. It does not apply to some cases of warranted, statistically based, true beliefs. One can be warranted in a belief on the basis of purely statistical evidence. For example, in a random drawing of balls, nine-tenths of which one knows are red, one is epistemically warranted in believing that the first ball drawn will be red. Even though the belief is true and none of the undermining conditions applies, one may not know that the first ball drawn will be red. In such cases—where the certifying ³ By ‘instances’, I intend to signal, for example, instantiations of a psychological state—or event-type in a particular individual at a particular time. One can also think of an individual’s holding a belief at a particular time as a bearer of epistemic warrant. A given type of belief, or a given type of inductive inference that has premises and conclusions with the same representational-content types, can be warranted in one individual or time and unwarranted in another. Differences depend on background information available to the individual, or in the level of reliability in the individual’s competencies. I explain in more detail below what I mean by a belief type or transition type. See note 58. ⁴ For brief discussion of the lack-of-internal-incoherence condition, see note 82.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     condition does not apply—epistemic warrant still constitutively conduces to true belief and to knowledge. It conduces to knowledge in that it meets the conditions (a)–(c) above. In particular, more warrant that would support an explanation of why, in the particular case, a red ball will be formed would make the belief knowledge. Epistemic warrant always marks a belief as issuing from a competence that conduces to knowledge. Warranted, true beliefs that are based purely on statistics and that do not count as knowledge still conduce to knowledge and to being true. As I indicated, I think that for every epistemically basic belief, there is a type of possible epistemic warrant that certifies such beliefs, when true, prima facie, as knowledge. I also think that some epistemic warrants for some non-epistemicallybasic beliefs can certify true beliefs of those sorts, prima facie, as knowledge. The warranted true beliefs are knowledge, if the stated undermining conditions are not applicable. For example, I think that the relevant certification applies to certain beliefs (such as the belief that is an X-ray machine, applied on the basis of current perception) that are not strictly perceptual beliefs in the sense that I will explicate. I distinguish between epistemic warrant and generic, positive epistemic support. Epistemic warrant warrants a belief, at least prima facie, as belief-worthy. I think that a belief can have small epistemic support that is not enough to warrant belief, even prima facie. In being constitutively conducive to knowledge—specifically, in supporting the relevant state’s being knowledge in the sense of (b) above—epistemic warrant constitutively guides toward truth. Being epistemically warranted requires that a belief is formed or held in a way conducive to that type of belief ’s being true. This conduciveness must reside in a competence in forming or maintaining the belief. The competence must tend, non-accidentally, to connect beliefs of the relevant type, formed in the relevant way, to the subject matter that makes the beliefs true. These relations to knowledge and to truth are basic to the nature of epistemic warrant. In sum, a psychological state or event is epistemically warranted if it meets certain norms, or standards, for its being epistemically good. They are epistemically good in that beliefs or events of that type, formed or maintained in the relevant way, tend to yield knowledge and hence truth. Belief and occurrent analogs, such as judgment, are primary psychological states for understanding epistemic warrant—hence for understanding knowledge and understanding. They are primary because their representational function is to be true, and because the central epistemic goods—knowledge, understanding, and epistemic warrant—constitutively serve, in ways to be discussed, representational good or success—specifically, true belief. In elaborating these points, I start with representational goods and functions, as distinct from epistemic goods and functions. I focus on functions rather than purposes or goals. I think that beliefs have, by nature, a representational function. They do not literally have purposes, goals, or aims. Talk of the aim of belief is at best metaphorical. Every belief has the representational function of being true. Any belief undergoes a certain failure as a belief—what is naturally called a representational failure—if it is not true. Any belief enjoys a representational success as a belief, if it is true. Success and failure evince functions, goals, aims, or purposes. Beliefs do not themselves have

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



goals, aims, or purposes. So being true or false is a success or failure in fulfilling a function, a representational function. Similarly, every perceptual state has the representational function of being accurate. Truth and falsity are properties of propositions and propositional states. Perceptual states are not propositional. So they cannot be true or false. They can be accurate or inaccurate. Any inaccurate perceptual state undergoes a type of failure as a perceptual state. Accuracy is a type of success for perceptual states. Success and failure require presence of functions, goals, aims, or purposes. Perceptual states lack goals, aims, and purposes. So success and failure in perceptual states require functions. The functions are again naturally termed ‘representational functions’. Fulfillments of functions are goods. So being representationally successful—being true for a belief, being accurate for a perceptual state—is a representational good.⁵ Propositional inference—deductive and inductive—is a primary type of transition for understanding epistemic warrant—hence for understanding knowledge and understanding. Propositional inference undergoes a representational failure if it does not preserve truth.⁶ So propositional inference has a representational function of preserving truth. Transitions from perceptual states to perceptual beliefs have the representational function of preserving the veridicality (accuracy) of a perceptual state in the veridicality (truth) of the corresponding perceptual belief.⁷ Any such transition that does not preserve veridicality undergoes a type of failure, a representational failure. Propositional inferences and transitions from perception to belief that fulfill their preservational, representational functions are representational goods. Veridicality of belief and perception is success in fulfilling representational functions. Here is reasoning for taking representational functions, hence representational goods, to be distinctive types of functions and goods. As a matter of metaphysical necessity, in every case, beliefs and perceptions fail in some respect if they are not ⁵ Of course, I do not think that every true or accurate representational state or content is a representational good. The truth of a proposition about some arcane matter that no one has ever considered is not a good of any kind. Nor is the falsity of its negation a bad thing. Propositions do not in themselves have functions: a proposition does not fail in any way if it is false. I claim that truth and accuracy for certain committal states, like beliefs and perceptions, are representational goods. A committal state is one that has a representational function whose fulfillment hinges on whether its representational content is, was, or becomes veridical. Perception, belief, intention, memory, expectation, are committal. Supposing for the sake of argument and playful imagining are non-committal. Whether a modality is committal or not is apriori determinable from the nature of the modality. One cannot renounce the committal character of a state. I take the points in this note, and in the two paragraphs in the text that precede it, to be apriori. Understanding what a belief or perceptual state is requires understanding that it undergoes a certain type of failure, as a belief or perceptual state, if it is not true or accurate—and a certain type of success if it is true or accurate. ⁶ Inferences that meet deductive standards necessarily fulfill their representational function as a matter of form. From true premises, they necessarily lead to true conclusions by virtue of the inferences’ forms. No inductive inferential transition necessarily fulfills its representational function by virtue of its form. (Inductive inferences in mathematics and other inductive inferences necessarily preserve necessarily true conclusions from necessarily true premises, but not by virtue of the inferences’ forms.) Of course, neither deductive nor inductive inference can insure that premises are true. When representationally successful, both types conditionally preserve truth: if the premises are true, the conclusion is true. ⁷ Veridicality is the genus. Truth and accuracy are species. Truth is the propositional form of veridicality. Accuracy is the non-propositional form of veridicality. As noted earlier, beliefs are true or false; perceptual states are accurate or inaccurate.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     veridical. They do not necessarily fail with respect to any practical function, if they are not veridical. Any representational state metaphysically could be practically useful but non-veridical, or veridical but practically deleterious. So the function that perception and belief necessarily fail at, if they are not veridical, is not a practical function. Analogously, for relevant transitions. All biological functions are practical functions. They are practical in that they contribute, in some way, to survival long enough to reproduce. So there is a type of function for belief, for perception, and for relevant inferences and transitions that is associated with veridicality and that is not a biological or other practical function. I label this type of function a representational function. Meeting the norm for fulfilling this type of function is a representational good.⁸ I turn now to connections between representational goods and representational functions, on one hand, and epistemic goods and functions, on the other. Knowledge, understanding, and epistemic warrant are, of course, epistemic goods. They meet norms for different types of epistemic success. These epistemic norms are constitutively connected to representational goods, and to the representational function of belief. Knowledge and understanding are epistemically rich ways of achieving representational success for a belief—achieving truth. Epistemic warrant contributes to a belief ’s representational success: contributes to fulfilling the master representational norm of belief, being true. A belief ’s being epistemically warranted is conducive, in an epistemically good way, to the belief ’s being true. So all three epistemic goods are good in being epistemically good ways either of a belief ’s being representationally successful, or in contributing to a belief ’s representational success. In this very limited sense, true belief is a telos of these epistemic goods. A constitutive condition on knowledge and understanding is that they be true. Any belief undergoes an epistemic failure—a failure to constitute knowledge or understanding—if it is not true. A belief ’s being true is a constitutive element in central types of epistemic success—in fulfilling some central epistemic norms, in being epistemically good. True belief is a representational good regardless of how it is produced or maintained. It is not, however, an epistemic good apart from how it is produced or maintained. There is no epistemic good in forming a true belief by epistemically bad means. A belief ’s being produced or held in a way that is conducive to that type of belief ’s being true is a constitutive aspect of another epistemic good—epistemic warrant. Epistemic warrant is an epistemic good that is also a constitutive condition on knowledge and understanding. The epistemic goods, knowledge, understanding, and epistemic warrant, are good in being epistemically good ways of having true belief or, in the case of epistemic warrant, conducing to true belief. True belief is a representational good in itself. True belief is not in itself an epistemic good. The epistemic goodness of a true belief hinges partly on the epistemic goodness of how it is formed or maintained. True belief is

⁸ The first premise of this argument is set out in ‘Perceptual Entitlement’, op. cit., 508. The distinction between being true and being useful is also emphasized, 509–510. The argument in the text here, though implicit in the earlier paper, is not explicit there.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



epistemically good only through being supported or warranted—or through being knowledge or understanding—only through being formed or held in ways that meet distinctively epistemic norms for being epistemically good. I stress that, in discussing the epistemic goodness of true belief, I am not discussing what is good for “cognition” as a whole.⁹ I have, and have had, nothing to say on goods for cognition in aggregate. I make no claims about how or whether the specific distributed good that I discuss aggregates to make collective goods, or trades off to maximize some overall goodness. The goods that I specify attach to each belief or inference, one by one. In saying that being epistemically warranted is an epistemically good way to obtain truth, I do not imply that epistemic warrant is good only instrumentally, though it is instrumentally good. It meets a standard for having true belief in an epistemically good way—a way that contributes to knowledge and understanding. The epistemic goodness of epistemic warrant lies partly in how it conduces to true belief. It must conduce via a competence for forming or sustaining true beliefs of the relevant type, and that competence must be appropriately connected to its successes. Epistemic goodness derives from meeting distinctively epistemic norms. A telos of warrant, true belief, is not epistemically good independently of beliefs’ being formed or held in epistemically good ways.¹⁰ Unlike true belief, the epistemic goods, knowledge, understanding, and epistemic warrant, in themselves fulfill epistemic norms. Each of their instances is constitutively an epistemic good. All are epistemically good in being good ways of forming, maintaining, or conducing to true belief. True belief is a representational good that these epistemic goods serve. As noted, inferences are representationally good if they preserve truth. An inference’s preserving truth bears constitutive relations to certain epistemic successes. Any inference, and any transition from perception to belief, incurs an epistemic failure, a failure to yield knowledge, if it does not preserve veridicality. (I do not count

⁹ I think that the primary epistemic good of cognition is not to produce beliefs in as many “important” truths as possible while minimizing false beliefs. Here I differ with A. Goldman, Epistemology and Cognition (Cambridge, Mass.: Harvard University Press, 1986), 98; and W. Alston, Beyond “Justification”: Dimensions of Epistemic Evaluation (Ithaca, New York: Cornell University Press, 2005), 29. In the first place, the way in which one forms or maintains a belief is crucial to its epistemic goodness. In the second, having true beliefs that yield understanding through deep explanations is epistemically good in a way that just having true beliefs is not, even if these latter beliefs are “important”. Even if a belief is warranted, connecting it in explanatory ways to other warranted beliefs is clearly epistemically better than just holding them in a warranted way. (Needless to say, more needs to be said about what counts as being important, on the accounts that I am criticizing.) Connections among one’s beliefs and connections between perceptions and beliefs are crucial to the epistemic good and point of cognition. What is missing in these philosophers’ generalizations are roles for understanding and explanation–and more broadly, a role for specific, epistemically good ways of obtaining true belief, beyond just having true belief. ¹⁰ Truth is a representational good for belief independently of how the belief is obtained. I think it obvious that true belief is not an epistemic good independently of how the belief is obtained. For philosophers who may imply that obtaining true belief is in itself an epistemic good, see L. BonJour, The Structure of Empirical Knowledge (Cambridge, Mass.: Harvard University Press, 1985), 7–8; A. Goldman, ‘The Unity of the Epistemic Virtues’, in Virtue Epistemology: Essays on Epistemic Virtue and Responsibility, A. Fairweather and L. Zagzebski eds. (Oxford: Oxford University Press, 2001), 31–32.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     transitions from perception to belief as inferences, because I reserve the term ‘inference’ for propositional transitions that function, representationally, to preserve truth.) Fulfilling the representational function of preserving veridicality can be a constitutive condition on fulfilling certain epistemic norms for epistemic success— norms for having knowledge or understanding. As with true belief, an inference’s preserving truth is not in itself an epistemic good. There is no epistemic good in a truth-preserving inference that in no way supports the truth of its conclusion. To be epistemically good, truth-preserving inferences must meet distinctively epistemic norms. Analogously to the case of true belief, epistemic goodness in veridicality-preserving inferences and transitions derives from their fulfilling, or contributing to fulfilling, representational functions and norms in epistemically good ways. Relevant goods are fulfillments of a norm for fulfilling, or contributing to fulfilling, a function. Epistemic goods, for inferences and perception-to-belief transitions, fulfill or contribute to fulfilling epistemic norms. Doing so serves fulfilling, or contributing to fulfilling, representational functions and representational norms of veridicality-preservation. To summarize, the specific epistemic goods—knowledge, understanding, and epistemic warrant—are good in being epistemically good ways of forming or maintaining true belief. They are good ways of fulfilling or contributing to fulfilling a representational function. Being true fulfills a constitutive representational function of belief. Are there also constitutive epistemic functions for beliefs? In trying to answer this question, I re-use the method used in determining that beliefs constitutively have a representational function. All beliefs constitutively fail as beliefs if they are not true. That fact provides ground to think that all beliefs constitutively have the representational function of being true. A belief constitutively has an epistemic function—an epistemic function it has by virtue of being a belief, not by virtue of someone’s using it for some purpose, even an epistemic purpose—if and only if every belief constitutively fails as a belief if it does not succeed in fulfilling the function. Of course, any belief that fails to be understanding or knowledge fails in some respect. It fails to meet an epistemic norm. The issue here is whether it fails as a belief—whether it fails some constitutive function of belief. Not all beliefs fail as beliefs if they do not constitute understanding. Clearly, perceptual beliefs do not constitute understanding. (See note 2.) They do not thereby fail as beliefs. Many other beliefs do not constitute understanding. They do not fall short of norms that apply to all beliefs, as beliefs. Beliefs, per se, do not function to constitute understanding. What of knowledge? I believe that not all beliefs fail as beliefs if they are not knowledge. Beliefs that are not knowledge certainly fail to meet epistemic norms for knowledge. They fall short of realizing a central epistemic good. But failing to be knowledge does not seem to be a defect in a belief as a belief on a par with failing to be true. Imagine a person forming a belief about whether a red ball will be extracted blind from a bowl that is known to contain nine-tenths red balls and one tenth green balls. A true belief that a red ball will be extracted is not defective as a belief, if the belief is not knowledge. Similarly, for some beliefs about the future. I believe from weather patterns in Los Angeles that it will not rain in the first week of

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



October 2023. Suppose that the belief is true. Although it is not knowledge, it is not defective as a belief. Compare ‘I believe that p, but do not know that p’ with ‘I believe that p, but p is not true’. The latter evinces a defect in one’s system of beliefs. The former indicates that one’s belief that p does not achieve a certain epistemic good. But it does not necessarily evince a defect in a belief, as a belief. It may involve a wise judgment on the limits of one’s perspective. The first-level belief that p and the second-level judgment that one does not know that p may even both be epistemically warranted. So being knowledge is not a constitutive epistemic function of belief. Is being epistemically warranted a function of belief? Perhaps. But perhaps it is possible to have a non-defective belief with some epistemic support, but not enough support to count as warrant. Such a belief might be in a nether region in which it is permissible, perhaps given other considerations, to hold it tentatively. Epistemic warrant is a relatively strong type of epistemic support, solidly supporting belief, not just making it permissible. However that may be, any belief that lacks epistemic support suffers a failure as a belief, an epistemic failure. Of course an epistemic failure is compatible with a practical success and a representational success. I conclude that it is a constitutive epistemic function of belief to have some epistemic support. Similar, but more complex, points apply to propositional inferences and transitions from non-propositional to propositional states. In those cases, having an epistemic function hinges on whether the modality of the conclusion of the inference, or the end-point of the transition from a perception, is a belief—and on the point of the inference or transition. An inference that functions to form or maintain a belief that provides no support for the belief is defective. A deductively valid inference that functions to lead to a belief, but is circular, is epistemically defective as an inference. It is epistemically defective even though it fulfills its representational function of conditionally preserving truth. An inductive inference to a belief that preserves truth, but fails to support its conclusion fails as an inference. A transition from a perception to a belief is epistemically defective if the transition does not provide epistemic support for the belief. Non-defective inferences that function to yield conclusions that are not beliefs but are positive leanings must also yield some epistemic support. Otherwise, they fail in some way, an epistemic way, as inferences. Inferences whose point is not to produce a belief can be non-defective, even if they yield no support. One can make any inference for the fun of it. Transitions from perceptions that do not yield beliefs, but rather withholdings of belief, are non-defective, if, indeed because, the transitions yield no support. An individual could anticipate, because of background knowledge, that a belief based on a coming perception will be false. A transition to a suspension of belief or a dis-belief would not thereby be defective, if the transition yielded no epistemic support for the suspended or rejected belief. Inductive inferences that provide support for a belief, but that do not preserve truth, fail a constitutive representational function, but fulfill the relevant epistemic function. Is meeting a norm for knowledge or understanding a constitutive epistemic function of inductive inference? I think not. An argument analogous to the one

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     about the colored balls that showed that epistemically supported, even warranted, true beliefs need not fail as beliefs, if they are not knowledge, applies. A similar point applies for understanding. Some warranting truth-preserving inductive inferences do not fail as inferences by virtue of not yielding knowledge or understanding. Trivially, they fail norms for knowledge and understanding. They do not fail to fulfill a function of all inferences to beliefs. Our main positive conclusion is that there are epistemic as well as representational functions for beliefs, propositional inferences, and perception-to-belief transitions. Beliefs have a constitutive epistemic function of having epistemic support. Inferences that form or maintain beliefs, or positive leanings, have a constitutive epistemic function of providing epistemic support. Unlike deductive inference, inductive inference to belief can fulfill an epistemic function even if it fails its representational function. Inductive inferences with true premises and false conclusions can yield support, even warrant, for their conclusions. Thus there are constitutive epistemic goods for beliefs, inferences, and transitions from perception to belief. The toehold for a constitutive epistemic function of belief and relevant transition events grounds a more extensive set of constitutive epistemic functions. For example, it is a constitutive epistemic function of a belief to be maintained at a strength appropriate to epistemic support for it. Similarly, if an individual forms a belief that is prima facie epistemically warranted, and the individual has other accessible information that weighs against that belief, the formation or retention of the belief is epistemically defective if it does not take account of the information, in a way appropriate to its weight. So far we have discussed belief and inference per se. Constitutive functions in these cases must hold for even the most primitive levels of these capacities. Richer representational and epistemic psychological capacities carry richer epistemic functions. A psychology that not only has a capacity for beliefs but also a capacity to form beliefs about beliefs and to evaluate them using epistemic concepts, has wider scope for epistemically evaluable activity and processing. There are, I think, constitutive epistemic functions for different psychological levels of psychological capacity in which belief and inference are embedded. For example, an individual who can deliberate reflectively about beliefs has epistemically deficient beliefs if deliberation would bear on the epistemic warrant for a first-level belief and the individual does not deliberate. I think that there are epistemic functions and norms that are associated with adequately full use of one’s epistemically relevant powers. So I think that there are constitutive epistemic functions, norms, and goods not just for beliefs and inferences per se, but for beliefs embedded in specific psychological capacities to carry out explanatory justification. I do not pursue these matters here. I think that they may present a rich field for better understanding of epistemic functions, norms, and goods. Wherein is knowledge a good? The goods discussed here are good by virtue of fulfilling, or contributing to fulfilling, a function. I am focusing on representational and epistemic goods, not practical goods. So I lay aside the good of knowledge that derives from someone’s wanting it and succeeding in obtaining it or from yielding technology that helps people reach their goals. Being knowledge is not an epistemic function of belief or, I think, any other basic kind of

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



psychological state or event.¹¹ So knowledge’s being a good is not a result of fulfilling any representational or epistemic function of a basic psychological state or event. I return to the earlier point that knowledge’s being a good depends partly on contributing to fulfilling a representational function. By fulfilling its epistemic norms, knowledge fulfills the representational function and norm of belief in an epistemically good way. Knowledge is an epistemic good partly in being an epistemically good way of having representational success, a representational good—the representational good of forming or maintaining true belief. Knowledge meets its own norms for contributing to true belief in an epistemically good way. Of course, a belief ’s being knowledge fulfills not only the belief ’s representational function. It also fulfills belief ’s distinctively epistemic function—representing in an epistemically supported way. Obviously, fulfilling these functions does not suffice to fulfill norms for being knowledge. Similarly, epistemic warrant is a good not only in fulfilling an epistemic function of belief—the function of being epistemically supported. Epistemic warrant is a good also in contributing to fulfilling the representational function of belief, by involving conduciveness to the relevant type of belief ’s being true. Being representationally good or contributing to being representationally good is one constitutive point of being epistemically good—of fulfilling epistemic norms—for the central epistemic goods. Being representationally good is representing the world veridically. Being an epistemically good way of being or contributing to representational goodness is a distinctive type of goodness. Knowledge and understanding are such good ways of representing truly that they over-fulfill the representational function of belief. They over-fulfill the basic epistemic function of belief—the function of being epistemically supported. One can compare over-fulfillment here with over-fulfilling biological functions. The biological function of a heart is to pump blood well enough to contribute to survival long enough to reproduce. Pumping blood efficiently and healthfully over-fulfills this function. Life has a biological function similar to the function of a heart. Health, vitality, and living long are goods in that they over-fulfill this basic function. The ways in which knowledge and understanding go beyond true belief make the way in which true belief is formed, or held, especially firmly connected to being true. Knowledge goes beyond true belief partly in being epistemically warranted. An epistemically warranted true belief is formed or held through a psychological competence whose operation is conducive to forming or maintaining true beliefs of the same type. The competence must be constitutively related to the subject matter so as to make the truth of beliefs that are formed or maintained through the competence ¹¹ I do not count knowledge or understanding as basic psychological kinds, although I leave open whether they are basic in the sense of being undefinable. A basic psychological kind is a psychological kind that falls within psychological laws or law-like generalizations. Psychology does not treat knowledge or understanding as kinds that enter into psychological laws, though obviously they are, in a sense, kinds of psychological state. For more on this matter, see note 55. One might suggest that psychological systems have functions of producing knowledge or understanding. I doubt that there is any literal sense in which this is true. Of course, we may aim at knowledge or understanding. Knowledge and understanding are rich goods for both beliefs and psychological systems. And understanding ennobles believers.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     deeply non-accidental. Similarly, knowledge goes beyond true belief in avoiding Gettier cases. Without attempting a generalized account of Gettier failures, I think it fair to say that avoiding them necessarily involves one’s warrant’s being directly connected with the truth of one’s true belief. So in both respects—being warranted and avoiding Gettier cases—knowledge involves a way of forming or maintaining true belief that makes connection to truth especially strong. Understanding also overfulfills the representational function of belief. It over-fulfills it in all the ways that knowledge does. It adds a systematizing, explanatory epistemic value. Of course, it does not follow from any of this that the epistemic reduces to the representational. Meeting an epistemic norm does not reduce to fulfilling, or contributing to fulfilling, a representational function or representational norm. The point is, rather, that epistemic norms, in central cases, are standards for distinctively epistemic ways of being representationally good. The central epistemic goods are deeper and richer than representational goods. The reasoning about functions, norms, and goods in this section has been apriori. I think it interesting and valuable to have discovered, apriori, some basic representational and epistemic functions, norms, and goods for belief and for those inferences or other transitions that function to yield belief. The goods do not depend on anyone’s caring about them or pursuing them. They attach apriori and necessarily to relevant psychological capacities. I focus now on epistemic warrant, beginning with an overview of its fundamental aspects. Epistemic warrant fulfills a norm for an individual’s, or a psychology’s, operating competently and in a way that conduces to having knowledge, given the information available to the individual. Epistemic warrant attaches paradigmatically to beliefs and occurrent committal thoughts—relative to ways of forming or maintaining them—to inferences to committal conclusions, and to reliance on capacities that make use of beliefs or inferences. (See note 5.) The epistemic goods, knowledge and understanding, are factive. They constitutively involve true belief. Epistemic warrant is non-factive. A belief can be warranted without being true. An inference can be warranted even though it leads from true premises to a false conclusion. The distinctive aspect of epistemic warrant that allows a false belief to be warranted is that it allows for limitations on information available to individuals. For example, an individual’s false perceptual belief can be warranted if the individual’s belief falls into brute error. Brute error is error that derives from a well-working psychological system, such as a perceptual or belief-forming system, that happens to be in a situation that yields the error.¹² The error is not the fault of the system or individual. Every belief-forming system that relies on perception is subject to brute error. It can operate as well as possible, given the information that it has, and still yield a false belief. Similarly, a warranted inductive inference from true warranted premises may support a warranted but false conclusion. To be warranted, the inference type must tend, in normal conditions, to yield true belief given true

¹² ‘Perceptual Entitlement’, op. cit., 507.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



premises. But in the given case, there may be a factor that the individual lacks information about and that results in the inference’s being unsound. A second aspect of epistemic warrant is that it meets norms for operating so as to contribute to gaining knowledge. This second aspect is codified in conditions (a)–(c) above. Epistemic warrant’s contributions to knowledge are what make epistemic warrant an epistemic property. It contributes to knowledge by conducing to knowledge—(a)–(c). More strongly, as noted, epistemic warrant contributes by certifying certain types of true beliefs, prima facie, as knowledge. In the absence of countervailing warrants, internal incoherence, and Gettier failures, epistemically basic true beliefs, warranted in epistemically basic ways, are knowledge. A third aspect of norms for epistemic warrant is that they mark operating well in forming or maintaining true belief or preserving truth in inference. To operate relevantly well, the operation must tend toward forming true belief or preserving truth in inference, given the information available. If the relevant propensity to true belief is to contribute to knowledge, as epistemic warrant does, it must derive from the psychological competence. The formation of belief, whether inferential or not, must be part of a non-accidental pattern—capable of grounding a general law-like explanation—that tends to connect the belief-forming competence itself with the subject matter that makes beliefs of that type true.¹³ Epistemic warrant is not factive. But these connections between epistemic warrant and conduciveness to truth are just as constitutive of epistemic warrant as the factive connections are to knowledge and understanding.¹⁴ ¹³ I emphasize this point in section VI of ‘Perceptual Entitlement’, op. cit.. See further discussion in Section 2.2. ¹⁴ For an attempt to show that there is something wrong with taking epistemic warrant to have a teleological character, conduciveness to truth, see S. Berker, ‘Epistemic Teleology and the Separateness of Propositions’, Philosophical Review 122 (2013), 337–393; and ‘The Rejection of Epistemic Consequentialism’, Philosophical Issues 23 (2013), 363–87. The arguments hinge on taking, indeed defining, teleological approaches in epistemology so that they bear structural relations to simplistic, unattractive consequentialist ethical theories. The arguments are strong and insightful against the positions that they target. Some philosophers have written things that suggest the specific type of teleological position that the arguments target. But it is not clear to me that the philosophers whom Berker criticizes believe, reflectively, that being conducive to truth is a value for warrant and “cognition” in the way that his arguments assume. Specifically, the normal reading of ‘an epistemic warrant is conducive to true belief ’ is not that an epistemic warrant is conducive to true belief in general, regardless of what truth is at issue. The normal reading is that an epistemic warrant is conducive to the truth of beliefs of the type that the warrant warrants believing. Berker’s arguments hinge on opponents’ taking warrant’s conduciveness to truth in the former way rather than the latter. In any case, I am committed to none of the tenets of the teleological theories that he imputes to the philosophers whom he criticizes. I do not take true belief to be a final value in his sense. I propose no theory of overall value. I do not accept a deontic theory of the sort that his arguments require. I do not see an epistemic warrant as “promoting” truth in general, as opposed to marking a good way of forming or sustaining the relevant sort of belief. Still, contrary to his definition of a ‘teleological’ view, my view has teleological commitments. Epistemic warrant marks a way of producing or sustaining the belief as meeting standards for being an epistemically good way of meeting the belief ’s representational function. Berker presents his arguments as if they defeat all teleological conceptions of epistemic warrant (or “epistemic justification”–see note 17). He recommends giving up any requirement that a warrant conduce to truth. He gestures toward a guidance conception criticized below in Section 2.2. I believe that this position is conceptually wayward. Understanding what epistemic warrant is requires that one take it as marking conduciveness to both knowledge and true belief. As argued in Sections 2.1–2.2, it marks a certain sort of reliable true belief.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Although there are issues about when a competence can include a prosthetic device and remain the individual’s competence, I shall not investigate these issues here. I take it that perception through glasses counts as part of one’s competence, but a benevolent scientist’s hooking one up unawares to a special truth-finding machine cannot be included in one’s competence. The issues that I discuss do not hinge on how to adjudicate hard cases for determining what an individual’s competence consists in. An individual’s holding an epistemically warranted belief can be conducive to the belief ’s being true because of the nature of the belief itself, as in the case of an individual’s acceptance of obvious logical or mathematical truths through understanding them. Or more commonly, it can be because of how the belief is related to other psychological states and epistemically relevant conditions. An epistemically warranted belief is situated in repeatable psychological patterns. Relevant patterns include competent production of beliefs of the relevant type. Those patterns are part of larger patterns that involve relations of the competence to the belief ’s subject matter. The larger patterns conduce to beliefs of the relevant type being true, through the relevant competence’s relations to the belief ’s subject matter. I sloganize this point by saying that to be epistemically warranted, a psychological state or event must conduce to truth by being on a good route to truth. The route must involve a competence that is itself non-accidentally connected to forming true beliefs. Deductive inferences preserve truth by their form and nature. Warranted inductive inferences are conducive to preserving truth through their constituting psychological patterns that mirror and derive from environmental patterns. The environmental patterns tend to connect attributes represented by premises with attributes represented by conclusions. Warranted inductive inferences form parts of epistemically good routes to truth. Epistemically warranted belief must be on a route to truth that is relevantly reliable. Constitutively, epistemic warrant marks competencies for producing or sustaining belief as relevantly reliable in producing or sustaining true belief. This requirement derives from epistemic warrant’s role in marking a belief or inference as being conducive to knowledge, specifically (c). The formation of a belief or the carrying out of an inference that is not relevantly reliable could not be conducive to knowledge. No competence that formed or sustained a belief, or was involved in a belief-supporting inference, could be good enough to be conducive to knowledge if it were in every relevant way unreliable in yielding true belief or in preserving truth in inference. Indeed, if being epistemically warranted (per impossibile) allowed a competence that formed or sustained a belief to be relevantly unreliable, then being epistemically warranted could mark a competence that was a decisive hindrance to having knowledge—contrary to condition (c) on being conducive to knowledge. No belief that relied on the competence, even partially, to meet conditions on knowledge could possibly be knowledge. As indicated, I think that in certain cases—including epistemically basic types of belief—canonical or basic epistemic warrant for those beliefs certifies true beliefs prima face as knowledge. In the absence of stronger counter-warrants, of internal incoherence, and of Gettier failures, epistemic warrant marks many true beliefs as knowledge. In particular, under such conditions, the basic type of epistemic warrant for perceptual belief certifies true perceptual beliefs prima facie as knowledge. Again, such certification requires that the competence that supports the true belief be relevantly reliable.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



The most general point is this: Since, in all cases, epistemic warrant marks beliefs and inferences as being conducive to knowledge—as a necessary condition of being an epistemic warrant—epistemic warrant marks beliefs and inferences as formed or maintained in a reliably veridical way. I return to the relevant reliability, for empirically based beliefs, in Section 2.2. Being epistemically warranted also requires that the warranted state or event derive from a competence that functions well in the given case, in forming or maintaining the belief.¹⁵ The standards that epistemic warrant meets concern an individual’s psychological competencies’ contribution to getting things right in a way conducive to knowledge. To be part of a good route to truth that derives from a competence’s being connected to a belief ’s subject matter, the competence, as distinguished from a malfunction or an intrusion from the outside, must be the source of the belief. If a malfunction or intrusion figures essentially in producing or maintaining a belief, the competence is not the main psychological source. The exercise of the competence does not meet the standard of operating well in serving knowledge. The exercise of a well-functioning competence is the individual’s psychology’s contribution to being epistemically warranted. A further reason for taking a well-functioning competence to be a condition on epistemic warrant is the following: The norms for epistemic warrant are standards for forming or maintaining beliefs, or engaging in inferences, in ways that are conducive to gaining knowledge. A true belief that resulted from a malfunction of or intrusion onto the competence could never be knowledge, because of condition (c), even if the malfunction or intrusion reliably produced true belief. In ‘Perceptual Entitlement’, I distinguish two types of epistemic warrant— justification and entitlement.¹⁶ A justification is a warrant whose force derives at least partly from a reason. A psychological state or event is justified if and only if it is epistemically warranted, and the force of its epistemic warrant derives, at least partly, from a reason (or reasons) in the psychology.¹⁷ An individual is justified regarding a given state or

¹⁵ I get this point wrong in ‘Perceptual Entitlement’, op. cit., 537n24. ¹⁶ ‘Perceptual Entitlement’, op. cit., section I. ¹⁷ Many philosophers use ‘justification’ roughly as I use ‘warrant’. I use ‘justification’ as I do because the term is traditionally and etymologically tied to ‘justify’, which is to show something to be just or right, or to provide an acceptable explanation for something. Entitlements concern a certain status that does not in itself involve any such showing or explaining. Hence they are not naturally called justifications. Entitlements commonly attach to perceptual beliefs. Such beliefs need not be supported by argument. I think that use of ‘justification’ (and German cognates) goes fairly far back in the history of philosophy–it is certainly present in Kant. But use of the term to cover beliefs that are not “justified” by further reasons or by selfevidence, but rather by some default status, is more recent and less historically entrenched. Early influences in this unfortunate usage are articles by W. Alston from the 1970s and early 1980s, collected in his Epistemic Justification (Ithaca, Cornell University Press, 1989). Traditionally and etymologically, a warrant is a kind of pledge or guarantee. A warrant may or may not involve a showing or an explanation. ‘Warrant’ is better fitted to the broader category than ‘justification’ is. Incidentally, I hope that it is obvious that my use of ‘warrant’ differs from the use in A. Plantinga, Warrant: The Current Debate, op. cit. (Oxford: Oxford University Press, 1993). I think that Plantinga’s use, which entails that an individual is not epistemically warranted if Gettier conditions are not satisfied, is idiosyncratic. Some philosophers use ‘reason’ for any warranting factor. A pain, a perception, or even an environmental condition can, on that usage, count as a reason. Perceptual beliefs are counted ‘reasonable’ and ‘reason-based’. This usage perniciously blurs distinctions among psychological and normative kinds.

[Note 17 continues below next page]

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     event in his or her psychology, if the individual, or the psychology, connects warranting reasons to that state or event. A reason for a conclusion is a propositional state, actual or possible, that provides support for the conclusion through actual or possible reasoning from the state’s content to the conclusion. A reason can provide either prima facie or all-things-considered justification. It provides prima facie justification, if it provides some justification, and would provide sufficient justification for belief, inference, or what not, if no counter-warrants were involved. It provides all-things-considered justification, if in the actual case, including all relevant factors, it provides sufficient justification for belief, inference, or what not. I follow standard practice in distinguishing an individual’s having a justification for a state or event from that state or event’s being justified. A justification can be present in or easily available to an individual’s psychology, even if the justification is not operative in being appropriately connected in the psychology with what it is reason for. Then the individual has the justification. For the state or event to be justified, the individual must not only have the justification. That justification must be operative. It must be connected to the state or event in such a way as to support it in actual reasoning or in a psychological structure. I assume that reasons are propositional. They form part of a propositional sequence that in effect constitutes explanatory reasoning as to why the state or event that they are reasons for is epistemically good. In effect, reasons jointly answer possible normative why-questions with normative because-answers.¹⁸ Perceptual states lack propositional structure. They can be cited in reasons. They can figure constitutively in an entitlement. They cannot literally be steps in reasoning that explains anything. So they cannot themselves be reasons. Reasons need not involve a capacity for reflection. Reasoners need not have any epistemic or other meta-representational concepts. Hence reasoners need not have a capacity for self-consciousness. Propositional inferences by very young children who lack reflective capacities and consciousness of self can be reasoning and can yield justifications. Reasons need not even be accessible to consciousness. A structure or operation’s meeting rational norms does not require that it be accessible to consciousness. Having a reason requires only that the reason occur in the individual’s psychology—or be easily inferrable, given the individual’s psychological powers, from what is in that psychology.¹⁹ An individual’s belief ’s being justified requires [Note 17 continued:] I hope to discuss such issues elsewhere. I stress here that I take reasons to be propositional representational contents, or possible attitudes with such contents. ¹⁸ I say ‘in effect’ because no one need pose a question about why a conclusion is belief-worthy. A reasoner need not even be able to pose such a question. There is empirical reason to think that children reason before they have a concept of belief and can pose relevant questions. Their reasons nevertheless support the conclusion in a way that could ground an explanation of its belief-worthiness. ¹⁹ Some philosophers engage in a usage on which an individual “has” any reason that is applicable to the individual’s situation, whether or not the reason is in the individual’s psychology or otherwise available to the individual. See T. Scanlon, What We Owe Each Other (Cambridge, Mass.: Harvard University Press, 1998); Being Realistic About Reasons (Oxford: Oxford University Press, 2014). On this usage, an individual can “have” a reason even if he or she lacks concepts needed to think the content of the reason. I regard this

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



only that the believer have reasons, and that they operatively support, in the believer’s psychology, what they are reasons for.²⁰ As noted earlier, some epistemic support for usage as linguistically odd. I also think it detached from epistemology. It detaches reasons from potential reasoning by the individual who “has” the reasons. I think of reasons that are applicable to the individual’s situation, but that the individual does not have in his or her psychology, as potential reasons, or God’s-eye reasons. I hope to discuss these matters elsewhere. ²⁰ Here some remarks on the contrast between this characterization of justification and the one in ‘Perceptual Entitlement’, op. cit. are in order. The main contrast lies in a different position on accessibility. In ‘Perceptual Entitlement’, 505, I characterized justification as ‘warrant by reason that is conceptually accessible on reflection to the warranted individual’. I stressed (505, 519–521) that reflection is not necessary for having reasons. So I meant ‘is conceptually accessible, if reflection could be and were to be exercised’. In explicating entitlement, 504, I wrote that it is warrant that is not accessible even on reflection. Reflection is an especially strong tool for bringing something to consciousness. The relevant necessary condition for justification was simply a requirement that the individual be able–have a competence— consciously to think a propositional formulation of the warrant (504). For justification, the warrant consists in the warranting reasons together, in most cases, with an inferential transition from the reasons to a conclusion. So the access requirement on justification was merely that the individual be able consciously to think the reason (not necessarily think it as a reason). I emphasize that, even in ‘Perceptual Entitlement’, my access requirement on justifications was very spare–not at all demanding. To be justified, the individual also had to have a disposition to connect the reason, for example by inference, to the conclusion–an inference that could be made conscious. I contrasted justification with entitlement. I claimed that the former but not the latter requires conceptual access to the “warrant”. The former but not the latter required an ability to bring the warrant to conceptual consciousness. I later dropped the conscious accessibility requirement on justification. See ‘Epistemic Warrant: Humans and Computers’ (2011), in my Cognition Through Understanding: Philosophical Essays, Volume III (Oxford: Oxford University Press, 2013). I thought in 2003 that reasons must be accessible to being consciously thought. In 2003, I glossed the lack of access-requirement in the case of entitlement: ‘The individual need not have the concepts necessary to think the propositional content that formulates the warrant’ (504, italics added here). The warrant in the case of perceptual entitlement is the entitlement. For perceptual belief, the entitlement is the process of belief formation from registration of proximal stimulation through a perceptual state to a perceptual belief. To be an entitlement, such a process must reliably produce accurate perceptions and true beliefs generated from the perceptions; the reliability must be grounded in relations to the environment that help explain the connection between reliability and the warranted beliefs; and the process must be psychologically well-functioning in the particular instance. In fact, I believed, and continue to believe, that to be entitled to a perceptual belief, an individual need not be able, consciously or unconsciously, to formulate or specify a single aspect of the entitlement. Some commentators did not think seriously about this passage (‘Perceptual Entitlement’, 504). They produced badly mistaken glosses on my access requirement and on other aspects of my justification/ entitlement distinction. They imported notions into their explications that I nowhere used. And they overread passages, or read them out of context. See A. Casullo,‘What Is Entitlement?’, Acta Analytica 22 (2007), 267–279 and N. Silins ‘Explaining Perceptual Entitlement’, Erkenntnis 76 (2012), 243–261. A largely sound reply to Casullo occurs in B. Majors, ‘What Entitlement Is’, Acta Analytica, 30 (2015), 363–387. I hope to criticize these readings in more detail on another occasion. In ‘Perceptual Entitlement’, I used a conscious-access requirement on justification, because I thought that all reasons could in principle come to consciousness. I thought that warranting factors in an entitlement need not ever be capable of being consciously thought, or even conceptualized. My claim that an individual need not be able to conceptualize the individual’s own entitlements entailed that entitlement was warrant without reason. And I thought it obvious that much, or all, of the entitlement for perceptual beliefs was never consciously accessible. So consciousness was not a key issue for understanding entitlement. Subsequently, I realized that there is nothing apriori impossible about reasons in an individual’s psychology that are not consciously thinkable by the individual–in effect, modular reasons. Since I regard the distinction between warrants by reason and warrants without reason as fundamental, I dropped the access requirement on justifications. The distinction between justification and entitlement is now independent of issues about consciousness. See ‘Epistemic Warrant: Humans and Computers’, op. cit.. Of course, the accessible-reason/inaccessible-reason distinction remains philosophically important.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     a belief is too weak to constitute warrant or justification. The support may support only a slight favorable leaning. So not all reasons yield even prima facie justification. Operative support can be via actual inference or via structurally sustaining a reason’s conclusion. Since knowledge is propositional and perceptual states are not, perceptual states cannot be knowledge. Nor can memory that simply preserves perceptual representation. Since epistemic warrant is a property of states or events that can constitute knowledge, perceptual states and perceptual memories cannot have epistemic warrant.²¹ They can have an analog of epistemic warrant. By being reliable and the product of a well-functioning perceptual system, they meet an epistemically relevant norm for helping to form warranted perceptual beliefs. A psychological state, event, or transition has an epistemic entitlement if it is epistemically warranted and the force of its epistemic warrant does not derive, even in part, from a reason.²² A perceptual belief is warranted inasmuch as it fulfills the norm for being warranted. What fulfills the norm is a sequence of event types that begin with a sensory registration, lead to a perceptual state, and thence to the perceptual belief. Fulfilling the norm is meeting conditions that I will discuss. An individual is entitled to a belief or inference if the individual has an operative epistemic entitlement to it. Entitlement is warrant without reason.²³ An entitlement can always in principle be paralleled by a counterpart justification. A sufficiently sophisticated and knowledgeable individual could specify the entitlement in propositional form. The force of an entitlement does not, however, derive from any such specification. An individual whose state is warranted by entitlement need not be capable of thinking ‘the propositional content that formulates the warrant’, by thinking counterpart reasons. An entitlement derives from the epistemic goodness of the competent production of the warranted state or transition, where the competence is not that of producing a reason. The competence may be that of forming a reliably accurate perceptual state and transitioning reliably from it to a reliably true perceptual belief. Epistemic goodness resides in the warranted state’s being produced in a well-functioning, relevantly reliable competence. Most of our perceptual beliefs are warranted by

²¹ For further discussion of the point that epistemic warrant is a feature only of propositional states, which do not include perceptual states, see ‘Perceptual Entitlement’, op. cit., 516–518, 530–533, esp. 533, 535, 540n, 544n. ²² In ‘Perceptual Entitlement’, op. cit., 504, my primary explication of entitlement states that a warrant is an entitlement if an individual’s having the warrant does not require that the warrant be accessible to the individual or that the individual even be able to think any propositional content that formulates the warrant. Since I explicitly indicated that I conceived of reasons as constitutively propositional (for example, 528), it follows from the older explication that entitlements are warrants without reasons. Entitlements are warrants that do not derive from reasons that the individual has. In my view, the explication of entitlement in ‘Perceptual Entitlement’ (in contrast to the explication of justification–see note 20 above) is extensionally equivalent to the newer explication. The present explication of justification and entitlement is stated in ‘Epistemic Warrant: Humans and Computers’, op. cit., 490. ²³ Although my explications for justification and entitlement are for epistemic warrant, there are parallel notions for practical warrant: practical justification and practical entitlement. For work on practical warrants analogous to epistemic entitlements, see Y. Luthra, Embodied Rational Agency, UCLA Dissertation, (2013), chapter 6–retrieved from https://escholarship.org/uc/item/4q92g4w6.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



entitlement. To be warranted, they need not be backed by any reason. Similarly, the relevant competence may be that of forming a truth-preserving deductive inference or a truth-conducive inductive inference. Most of our inferential transitions, deductive and inductive, have entitlements, not justifications, at least until we attain a considerable level of sophistication. Giving reasons for the goodness of an inferential transition—perhaps showing that an inference rule is a good one—is a sophisticated enterprise. My discussion centers on the epistemic status of perceptual beliefs. Here is a good place to explain what I mean by ‘perceptual belief ’. Perceptual belief is belief formed directly from perceptual states. Being formed directly from perceptual states entails meeting four conditions. First, the formation must be not be through (propositional) inference. Second, all elements that function to represent in a de re way in a perceptual belief content depend for fulfilling their representational function on deriving from counterpart elements that function to be de re in the representational content of the perceptual state that underlies the perceptual belief.²⁴ I call representational, contextbound, referential elements that function to be de re in belief contents and other propositional contents, ‘referential applications in propositional thought’. I call representational referential elements that function to be de re in perception, ‘referential applications in perception’. In what follows, I often shorten ‘referential applications’ to ‘applications’. Applications that function to be de re (say, applications of singulardemonstrative-like representational elements) are applications that function to pick out a representatum in a non-inferential, not-completely-conceptual-or-perceptualattributional, context-dependent way. Applications of either type are individuated via events at a given time. They are semantically like actual occurrent uses of demonstratives. Applications in perception are distinguished from applications in propositional thought in that they are referential applications of perceptual attributives, rather than conceptual attributives. Perceptual attributives are attributives formed in a perceptual system. All applications in a perceptual belief that function to be de re depend for successful de re representation on corresponding successful applications in the perceptual state. Third, every concept in a perceptual belief conceptualizes a perceptual attributive in the underlying perceptual state.²⁵ Usually, not all such perceptual attributives are conceptualized. But every conceptual attributive in a perceptual belief must have a counterpart perceptual attributive in the perceptual state from which the belief (and ²⁴ A de re psychological state is (roughly) one that functions to pick out a representatum in a direct, noninferential way, and that does not depend fully on concepts or perceptual attributives for doing so. For perceptual states and perceptual beliefs, de re representation is representation that involves occurrent demonstrative—or indexical-like elements. For less rough discussion, see my ‘Postscript to “Belief De Re” ’ Foundations of Mind (Oxford: Oxford University Press, 2007). ²⁵ For discussion of conceptualization of perceptions, see ‘Perceptual Entitlement’, op. cit., section VIII. For discussion of conceptual and perceptual indication and attribution, see my Origins of Objectivity (Oxford: Oxford University Press, 2010), 30–38; and ‘Predication and Truth’, The Journal of Philosophy, 104 (2007), pp. 580–608. Determining what attributives are perceptual is a delicate empirical matter. See my ‘Reply to Block: Adaptation and the Upper Border of Perception’, Philosophy and Phenomenological Research 89 (2014), 573–583. Attributives for color, shape, motion, and body are clearly produced in visual perceptual systems. It is empirically plausible, but less clear, that hand is a perceptual attributive for

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     the conceptual attributive) is formed. The concept must indicate and attribute the same attribute that the perceptual attributive does, and must depend essentially on the perceptual attributive for doing so. I emphasize that conceptualizing perceptual representations is not representing them or being about them. It is producing concepts that indicate and attribute the same environmental entities that corresponding perceptual attributives in the perceptual state do, while making essential use of those perceptual modes of presentation to do so. Here is a concrete illustration of this third condition. A visual perception as of a body of a certain shape and size will figure in the formation of a visual-perceptual belief as of the body of that shape and size. The visual perceptual state has attributives for body and for a specific shape and specific size. The perceptual belief in which these perceptual attributives are conceptualized has counterpart concepts for body and for the same specific shape and specific size. The perceptual system will commonly also have more generic attributives for shape and size; and these attributives may also have counterpart perceptual concepts in perceptual beliefs. The representational content of the perceptual belief is partly made up of conceptual counterparts of visual perceptual attributives that indicate and attribute some of the same environmental attributes that the perceptual attributives do. The conceptualizations of perceptual-attributive counterparts represent by depending on representation by the underlying perceptual attributives. The conceptualizations differ from the perceptual attributives primarily in that they function to contribute to a propositional structure. No concept in a perceptual belief lacks a counterpart attributive in perception that indicates the same attribute that the concept does. So if a perceptual system produces no attributive that represents x-ray machines as such, there are no perceptual beliefs—in my sense—that represent bodies as x-ray machines. It is a substantive empirical question which beliefs are perceptual beliefs, in my sense. For it is an empirical question what attributives are formed in perceptual systems. Fourth, every attributive that is referentially applied in the perception and that is conceptualized in the belief is referentially applied in the belief. The perceptual belief has de re applications that match de re applications in the conceptualized perception. Imagine having a perception as of a red body. Imagine that one perceives the body and the instance of redness. There are perceptual applications both to the body and to the instance of redness. In English, we express what we colloquially call a ‘perceptual belief ’ formed from the perception as a belief that₁ body is red. It is understood that the subscript on ‘that’ marks a referential application, by or in the believer, of the attributive body that functions to refer to a particular body. However, the English here does not express what I count as a perceptual belief. It does not express the de re application purportedly to an instance of redness. The belief as expressed here could be true if one saw the body, but did not see its redness. Perhaps the body is too distant to discern its color. One might believe correctly that it is red through interlocution, or primates. I count it one in this chapter. I think it unlikely that mule—as opposed to animate body of such and such shape, where the shape is in fact mule-like–, playing card, x-ray machine, or particle accelerator is a perceptual attributive.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



on a whim. Then the belief would be a hybrid of perceptually dependent elements and elements from interlocution or whim. It would be a hybrid even if the conceptual attributive red conceptualizes a perceptual attributive red that had occurred in previous perceptions. To be a perceptual belief, in my sense, the conceptual attributive red must be applied in a de re way—given that the counterpart perceptual attributive red is applied de re. A way of expressing a relevant perceptual belief in English would be that₁ body is redthat2. Both the concept body and the concept red must be accompanied by applications that function to apply de re to particulars. So every concept in the perceptual belief not only has a counterpart perceptual attributive. Every concept in the perceptual belief that conceptualizes a de re applied perceptual attributive must also be de re applied in the belief, in a way that depends on the counterpart de re application in perception.²⁶ The foregoing explication of ‘perceptual belief ’ is strict. One might prefer to use the term ‘basic perceptual belief ’ or ‘purely perception-dependent belief ’. My motive for centering on this notion is to isolate those beliefs that depend on perception as directly as possible. I think that all empirically supported beliefs either are or are supported by perceptual beliefs in this sense. Perceptual beliefs are not formed by propositional inference. Nor are they formed from non-inferentially associating some perceptually represented attribute with a further attribute that is not indicated or attributed in the perceptual system. Such extensions go beyond converting perceptual representation into perceptual concepts and perceptual beliefs. They involve concepts that are not strictly perceptual concepts. A further system beyond that of conceptualizing perceptual attributives, such as a system of associative learning in long term memory, is involved. If visual perceptual systems do not produce attributives that represent x-ray machines as such, there are no perceptual beliefs, in my sense, that represent x-ray machines as such. By the third condition, a perceptual belief indicates and attributes no attributes not already indicated and attributed in the underlying perception. By the second condition, a perceptual belief makes no applications in thought that do not depend anaphorically on applications in the underlying perception. By the fourth condition, each de re application of a perceptual attributive that is conceptualized in the perceptual belief has a counterpart de re application in the perceptual belief. So perceptual belief can omit perceptual attributions and applications, but it does not generalize on, abstract from, or add to them. Such generalization, abstraction, and addition are not functions of perceptual belief. They have their own epistemic norms. We do say that the expert sees the object as an x-ray machine. We say this even if there is no perceptual attributive that applies specifically to x-ray machines as such. ²⁶ Some perceptual attributives are not referentially applied in occurrent perceptual states. These are perceptual attributives for relations, like spatial relations, whose instances lack relevant causal powers and are not, strictly speaking, perceived, although the relation types are perceptually attributed. One does not strictly see the spatial relation between two entities that one perceives as next to one another. One sees concrete particulars as next to one another. In such cases, the conceptualizations, in a perceptual belief, of the relational perceptual attributives need not accompany a referential application to an instance of the relational attributive in the perceptual belief. Such relational perceptual attributives always function, however, to relate particulars, the relata, that are picked out by applications, hence perceived.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Such talk may seem to be in conflict with the third condition. But we should not take the talk to imply a judgment about whether the expert’s perceptual system itself produces representations as of x-ray machines. What attributives are produced in a perceptual system are for empirical science to determine. An expert’s association might be so automatic and non-inferential that it is colloquially natural to call the result a ‘perceptual belief ’. I am interested in norms governing actual mental activity and actual psychological natural kinds—not what ordinary language allows us to say. A fair amount is known about what attributives perceptual systems produce. Epistemic norms are fitted to our specific psychological capacities. I want a term that tracks perceptual beliefs that are as close to perception as possible. I reserve the term ‘perceptual belief ’ for this narrow range of beliefs. I focus on entitlements to beliefs most closely derived from perception. Such focus not only limits complexity. It also tracks epistemic entitlements that are the basis for all empirical warrant. Most of what I say about entitlement for perceptual beliefs applies, however, with minor adjustments, to a wider class of beliefs that are colloquially called ‘perceptual’. The attributives in the beliefs in this wider class need not have counterpart perceptual attributives that indicate the same attributes—as long as they are non-inferentially, de re applied by way of perception. For example, suppose again that there are no perceptual attributives—attributives produced in a perceptual system—that indicate x-ray machines as such. That is, suppose that no perceptual system produces attributives specifically as of x-ray machines. Suppose that an individual’s psychology contains non-inferential connections between applications of perceptual shape-, size-, and body-attributives and applications of a conceptual x-ray-machine attributive. Suppose that if the individual has a perception as of a body of the relevant shape and size, the individual noninferentially forms a defeasible belief that₁ body is an x-ray machinethat2. (Note that the attributive function is an x-ray machine, marked by the copula, outranks in scope the second referential application.) Suppose that there is a strong associative relation between a conceptual attributive x-ray machine and a perceptual concept as of a body with a certain shape and size. The association is stored in long-term conceptual memory. So the referential application of x-ray machine finds its referent via perception as of a relevant body, shape, and size. The subscript on ‘x-ray machine’ indicates that although the conceptual attributive x-ray machine is the main predicate of the propositional content, it functions de re to betoken (roughly, refer to) a particular x-ray machine. It may or may not succeed in referring. Reference here is not infallible, since perceptual reference is not infallible. Suppose that the x-ray machine is perceived, though not strictly perceived as an x-ray machine. The conceptual attributive is applied through its non-inferential associative connection to a perception of and as of the body, where the body constitutes the x-ray machine. A condition on successful application of that₂ to an x-ray machine is that the application that₁ refers to the body. Successful application of x-ray machine in thought depends on successful application in the perceptual system of a perceptual attributive body. One can be entitled to the belief that₁ body is an x-ray machinethat2. The association relation is not one of propositional reason support. One’s warrant is not a reason. Reasons at least sketch explanations of the belief-worthiness of a conclusion.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Association is not explanation. Even where there is a transition from the perceptual belief as of a body and its size and shape to another propositional attitude, the x-raymachine belief, the transition need not be one of propositional inference or reason support.²⁷ The formation of the belief might be warranted, if it derives from a reliable, well-functioning, appropriately grounded, associational competence. So the notstrictly-perceptual belief, as well as the perceptual belief, could be warranted by entitlement. I shall not further explore the epistemology of such hybrids. I focus on starting points for empirical epistemology: entitlement to perceptual beliefs in my strict sense. A guiding idea for understanding entitlement—warrant without reason—is that perceptual beliefs can be, and commonly are, epistemically warranted. Perceptual beliefs are not conclusions of propositional inferences, because perceptions are not propositional. Perceptions are not premises in propositional inferences, and are not reasons, because they are not propositional. So perceptual beliefs are not formed by inference from reasons. They are commonly not supported by reasons. They need not be, if believers are to be entitled to them. There is no need for further support than that provided through a reliable, well-functioning, well-grounded competence for forming perceptions and perceptual beliefs from perceptions. That is the primary good route to truth for perceptual belief. That route is commonly sufficient for epistemic warrant. It is often sufficient, in normal situations, to make true beliefs knowledge. Like justification, entitlement can be prima facie or all-things-considered. As with justification, an individual can have an entitlement to a psychological state or event, even if it is not operative. For example, an individual can have an entitlement to believe something, but can doubt it—either without warrant or with a justification that defeats the prima facie entitlement. As with justification, we can speak of an individual’s merely having an entitlement or, by contrast, as being (operatively) entitled to a belief or inference. An individual may have both a justification and an entitlement for the same state or event. A state or event can also be justified and have an operative entitlement at the same time. The distinction between justification and entitlement has to do with the nature of types of epistemic warrants. Among the states that one can have a justification or entitlement for is the state of relying on a psychological competence, such as perception, memory, or inference. The term ‘justification’ is often used for what I call ‘warrant’. This usage is not ideal (see note 17). It need not, however, ruin theory, if the distinction between justification, as I explain the term, and entitlement is kept clear. Entitlement is not equivalent to what is sometimes called ‘immediate justification’.²⁸ On this usage, “immediate justification” is warrant that does not derive from

²⁷ One can certainly test whether a given transition is a propositional inference or an associative transition. Different timings might be involved. Different acquisition histories may be involved. I leave it empirically open whether the non-perceptual concept is directly non-inferentially associated to the perceptual body—and shape—attributives, or only to the conceptual body—and shape—attributives. ²⁸ In what immediately follows, I use phrases that involve ‘justification’ in double quotes–scare quotes– to mark others’ use of ‘justification’ in something like the way I use the term ‘warrant’.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     other beliefs.²⁹ Some beliefs are self-evident. A reason to believe them resides in their contents. Understanding their contents suffices to warrant believing them. Examples are I am now engaging in thought (thought occurrently), if something is an impala, it is an impala, and twice two is four. They are “immediately justified”, and also immediately justified, in my usage. Their warrant—their justification—depends for its force on no other propositional attitude or information. Such propositions yield sufficient reason for belief, when believed with minimal understanding. Such a reason is in effect an explanation of the belief ’s belief-worthiness. These beliefs are not warranted by entitlement. Perceptual beliefs are not self-evident. All entitlements are “immediately justified”—warranted without reason or propositional inference. Not all “immediately justifications” are warrants by entitlement. The distinction between warrant with and without reason is so fundamental that it should be clearly marked. Entitlement constitutively depends on both psychological factors and external factors. The external factors are often are not supervenient on the individual’s psychology. The psychological factors are the states or events occurring in a well-functioning competence, though none or these in themselves constitute an entitlement. For example, the perceptual state is not in itself the entitlement. The entitlement is a type of sequence of events or states—from initial registration via perceptual state to perceptual belief—in a reliable, well-functioning competence, that is appropriately connected to a subject matter. A specification of the sequence, its reliability, and its connection to the subject matter would be a reason-explanation of why the belief is epistemically good. To be entitled to the belief, the believer need not be able to produce any such specification. In the empirical case, the key external factor is reliability in veridical representation of a subject matter. That must be in place if the belief is to count as epistemically warranted. Nothing in the nature of the psychological states insures their reliability. Epistemic entitlement does not require that there be warranted meta-beliefs that the perceptions or perceptual beliefs are reliable. In depending on constitutive aspects of warrant, like reliability, that need not be represented, known, or otherwise coded in the mind, entitlement is a partly externalist type of warrant.³⁰ I focus on this matter next.

²⁹ For examples of this usage, see W. Alston, ‘What is Wrong with Immediate Justification?’ Synthese 55 (1983), 73–95; J. Pryor, ‘There is Immediate Justification’ in M. Steup and E. Sosa eds. Contemporary Debates in Epistemology (Oxford: Blackwell, 2005); and A. Goldman, ‘Immediate Justification and Process Reliabilism’, in Q. Smith ed., Epistemology: New Essays (Oxford: Oxford University Press, 2008). I think use of ‘justification’ in this way is unnatural, though common. See note 17. ³⁰ Even most justifications have an externalist aspect. They constitutively depend on reliability. Deductive inferences from self-justifying truths code this reliability into the nature of the reason-constituting transitions. Perhaps some inductions from self-justifying truths also do so. But most empirical justifications are only relatively internalist. The sequence of states/events in the individual’s psychology that constitutes the justification provides an explanatory reason-sketch of why the conclusion is belief-worthy. The sequence of states and events that constitutes a perceptual entitlement provides no such explanatory reason-sketch, because it is not propositional, as explanations must be. In neither empirical justifications nor empirical entitlements are all warranting factors metaphysically necessitated by the nature of the individual’s psychology. I develop these points about perceptual entitlement in Section 2.2.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



2.2 Empirical Warrant and Reliability The primary aim of this section is to explain why a modest type of internalism about epistemic warrant is unacceptable. First, I set out the relevant type of internalism. Second, I sketch some history. I explain why, beginning in the early modern period, internalism was a dominant view about epistemic warrant, and how, when its traditional forms of support collapsed in the twentieth century, many philosophers continued to believe it. Third, I criticize some recent misconceptions of epistemic warrant that aim to support internalism. Fourth, I explain why internalism is unacceptable. Finally, I criticize two stock arguments that have been taken in recent times to support internalism. The criticism of the second one explains the specific type of reliability required by epistemic warrant. Accompanying the critical focus of this section is a delineation of central positive aspects of epistemic entitlement for perceptual belief. The epistemic internalism that I discuss claims that all constitutive aspects of an epistemic warrant are coded in the psychology of the individual who has the warrant. More strictly, I take epistemic internalism to claim: (EI) An individual’s having an epistemic warrant, and being epistemically warranted, in having a belief or in engaging in an epistemically relevant transition, supervenes on the non-factive kinds of psychological states, and kinds of relations among them, that are present in the individual’s psychology.³¹

That is, if the non-factive kinds of an individual’s psychological states and events and their relations to one another are held constant, it is metaphysically impossible for the warrants or specific failures of warrant for those states or relations to vary. Warrant hinges entirely on what psychological states and events are present in the individual with the warrant. It depends not at all on relations between those psychological states and relations, on one hand, and the wider environment in which the individual is embedded, on the other. Most internalists regard this kind of internalism as modest. Many other internalist positions entail this one. All internalist positions that I know of, including some that do not entail this one, are subject to variants of the criticisms that I make. I discuss (EI) because it is widely taken to be a minimum core of epistemic internalism. (EI) and most other forms of epistemic internalism are logically consistent with believing that some type of reliable veridicality is constitutively necessary to being warranted. In fact, from Descartes to mid-twentieth century, holding both views was

³¹ The limitation to non-factive psychological states is meant to block trivializing internalism by letting such states as knowing count as psychological states. Most formulations of internalist positions are more committal than (EI). Many require that warranting internal psychological states are knowable by reflection from the inside. Many require that one have reflective access, from the inside, to knowledge as to whether a given state or event is epistemically warranted, or reflective access to some other aspect of the metaexplanation of the warrant. In ‘Perceptual Entitlement’, op. cit., 504, I characterized epistemic internalism as the view that the epistemic warrant must be conceptually accessible–thinkable–by the individual whose psychological states or events are warranted. I hope to discuss relations between this kind of internalism and the kind characterized by (EI) on another occasion.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     standard. If one is to hold both views consistently, one must take one of two approaches. Either one must take reliability to be necessitated by the natures of psychological states and events that are epistemically warranted. Or one must invoke some other metaphysically necessary guarantee of their reliability. I turn to some history. Beginning with Descartes, many philosophers sought a guarantee that we have knowledge, particularly scientific knowledge.³² This guarantee was to provide a bulwark against scepticism and a philosophical underpinning for science. In the early modern period, two ways of providing a guarantee dominated. One was to advocate idealism—the view that connection to the physical world just is following epistemically good procedures. Berkeley and Kant held versions of such a view. The other way was to invoke a metaphysically necessary connection between epistemic operations and the world. Descartes, Malebranche, and Leibniz grounded such connections in theology. Both strategies entailed that epistemic warrant is necessitated by an individual’s psychological states. As a matter of necessity, one need only form certain types of beliefs and follow certain rules, to be guaranteed to reliably track truth. Rejection of idealism and theologically grounded metaphysics was basic for Russell and Moore in the early twentieth century. Their rejections turned philosophy decisively away from such failed metaphysical and theological projects. The traditional bases of support for internalism vanished. These philosophers persisted, however, in seeking, against scepticism, a guarantee for empirical knowledge. They built such a guarantee into their accounts of initial empirically relevant psychological states. They took acquaintance with sense-data to ground empirical warrant. They regarded acquaintance as infallibly veridical. They and followers, like Broad and C. I. Lewis, and many positivists, such as Ayer and early Carnap, thought that there were meaning-rules that paved a guaranteed route from sense data to the physical world. The supposed rules caused persistent trouble. Still, the approach dominated epistemology in the first half of the twentieth century. So, like the early moderns, these philosophers tried to show that empirical knowledge, hence empirical warrant, is necessitated by having certain psychological states and following certain rules. In accord with (EI), all features of empirical epistemic warrant, including reliability, were taken to be necessitated by those types of states and rules. In mid-twentieth century, the ideas that acquaintance with sense data grounds empirically warranted belief, and that one can get from sense data to the physical world via epistemically golden rules, came under devastating attack. Once the strategies of Russell, Moore, and the positivists were abandoned, the historical line of support for (EI) from the early modern period into the twentieth century, was broken. (EI) was ripe for re-evaluation. At this critical juncture, however, philosophy lacked two things necessary for a successful re-evaluation. It lacked a clear conception of epistemic warrant. And it

³² As I remarked at the outset, the focus was on a kind of knowledge deluxe, or scientia. I am aware that modern uses of ‘knowledge’ do not match many traditional counterpart uses–uses of ‘cognitio’ or ‘scientia’, for example. For present purposes, I table these distinctions, although they are historically important. They do not affect the main points that I make here. See note 50. Incidentally, it is possible that for empirical scientia, Descartes was an externalist.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



lacked an understanding of perception. Realistic reflection on warrant, and indeed knowledge, had lapsed because of the misguided sense-data approaches and because of the battles over positivist theories of meaning. Perception had been ignored because of sense-data theory. Many post-sense-data philosophers assimilated perceiving to sensing. Sensing was taken not to be representational. So perception was taken to have only causal, not epistemic relevance. Prominent philosophers actually held that nothing in the causal chain before perceptual belief is epistemically relevant. Perception disappeared from accounts of empirical warrant and knowledge. Eager to avoid an infallible foundation for knowledge, many required all warranted belief to be backed by reasons. Coherence theories of knowledge and warrant flourished.³³ Coherence theories were right to emphasize the defeasibility of perceptual beliefs. They seriously underplayed the central role of perceptual belief, let alone perception, in empirical knowledge. They said little about how knowledge and warrant connect to their subject matters. Failure to regard perception and perceptual belief as prima facie starting points for empirical warrant led the conception not to mark a connection between empirical warrant and success in representing a belief-independent world. When opposition to coherence theories gained traction, and perceptual beliefs, broadly understood, regained priority in the structure of empirical warrant, philosophers lacked a thought-through conception of epistemic warrant and a realistic understanding of perception. Relations between environment and psychology that give perceptual states and perceptual beliefs their contents form patterns that knowledge and warrant capitalize on. Such patterns were commonly ignored in the narrow, intra-mural discussions that still dominate post-sense-data epistemology. Many felt free to claim that perceptual belief is intrinsically warranted. Many modest foundationalist approaches continued to accept internalist views of epistemic warrant.³⁴ Another factor bolstered this trend. In the post-sense-data period, scepticism’s specter continued to warp accounts of epistemic warrant. Many philosophers claimed that warrant must be determinable by resources accessible in one’s psychology, on pain of appealing to matters that the sceptic questions. Hasty reflection on sceptical scenarios often led to claims that warrant is independent of psychologicalenvironment relations, and even of any requirement of reliability.

³³ W.V. Quine’s epistemology in Word and Object (Cambridge, Mass.: MIT Press, 1960), chapter 1 and W. V. Quine and J. Ullian, The Web of Belief (New York: Random House, 1970) illustrate these themes. An equally important figure in the development just sketched was W. Sellars, ‘Empiricism and the Philosophy of Mind’ in Science, Perception, and Reality (London, Routledge & Kegan Paul, 1963). Both propounded coherence theories of knowledge. Neither had a clear understanding of perception. Both assimilated perception to sensing. Later, D. Davidson followed Quine and Sellars on these points. See his ‘A Coherence Theory of Truth and Knowledge’ (1983) in Subjective, Intersubjective, Objective (Oxford: Clarendon Press, 2001). More detailed discussion of mid-career Sellars and his followers occurs in ‘Perceptual Entitlement’, op. cit., section VII. ³⁴ R. Fumerton, Meta-epistemology and Skepticism (Totowa, New Jersey: Rowman and Littlefield, 1995); J. Pryor, ‘The Skeptic and the Dogmatist’, Nous 34 (2000), 517–549; R. Wedgewood, ‘Internalism Explained’, Philosophy and Phenomenological Research 65 (2002), 349–369; J. Cruz and J. Pollock, ‘The Chimerical Appeal of Epistemic Externalism’, in R. Schantz ed., The Externalist Challenge (Berlin: De Gruyter, 2004); S. Berker, ‘Epistemic Teleology and the Separateness of Propositions’, op. cit..

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Finally, much of the impetus to believe internalism derived from resistance to “reliabilism”. Reliabilism is widely understood as a view that takes reliability to be sufficient as well as necessary for epistemic warrant.³⁵ Reliabilism seemed to many to be an extreme form of naturalism that eschews normative, epistemic notions. I do not accept any such reliabilism. I now criticize some conceptions of epistemic warrant that were conceived to support internalism. These conceptions arose after mid-twentieth century, when the main historical lines of support for internalism collapsed. Traditional lines proposed a metaphysically necessary guarantee for the veridicality of the foundations of empirical knowledge. The approaches that I discuss gave up that project. But they have not been deeply thought through. I think it fair to say that they have not come close to understanding epistemic warrant as a knowledge-conducive epistemic good. Much discussion of epistemic warrant since mid-twentieth century is not guided by any acceptable conception. Confusion reigns. A view prominent in the heyday of coherence accounts of warrant, which has supporters today, is that all warranted beliefs, including perceptual beliefs, are justified through reasons.³⁶ Such a view is epistemically internalist only if it takes the warrantedness of a justification’s premises and transitions to depend purely on the individual’s psychology. The idea that perceptual beliefs are warranted only through an operative argument in the believer’s psychology is wildly implausible. Simple arguments like I saw (or had a perception as of) that red square; so my perceptual belief that that is a red square is (likely to be) true; so that is a red square are justifications. But they do not add to the strength of epistemic warrant for the conclusion beyond what strength entitlement to the perceptual belief already provides. Such arguments are normally neither needed nor relied on. If any more sophisticated arguments were required, few adults would have warranted perceptual beliefs. Young children who have warranted perceptual beliefs and perceptual knowledge (that is body is moving; that surface is red) lack a capacity to think about their perceptions or beliefs as such at all. So they could not think the just-cited argument. Such meta-representational capacities seem required for any perception-based justifications of perceptual beliefs. So even simple justifications like the one just cited are probably not available to, much less operative for, all who have warranted perceptual beliefs. More generally, it is wildly implausible that to ³⁵ D. Armstrong, Belief, Truth, and Knowledge, (Cambridge: Cambridge University Press, 1973); A. Goldman, ‘What is Justified Belief?’, in G. Pappas ed., Justification and Knowledge (Dordrecht: Reidel, 1979); E. Conee and R. Feldman, ‘The Generality Problem for Reliabilism’, Philosophical Studies 89 (1998) 1–29; and A. Goldman, ‘Reliabilism’, Stanford Encyclopedia of Philosophy, online. Incidentally, I do not discuss the “generality problem” here. I think it a pseudo-problem that derives from not focusing enough on the actual explanation-grounding kinds of psychological processes that yield given belief states. Most discussion of the issue betrays insufficient use of psychological science in doing epistemology. ³⁶ Coherence views (see note 33) tended to share with classical foundationalism the view that a warranted individual has a sufficient reason for every warranted belief and hence every piece of knowledge. A coherentist view inspired by Sellars’ ‘Empiricism and the Philosophy of Mind’, op. cit. is proposed in L. BonJour, The Structure of Empirical Knowledge, op. cit.. Classical foundationalism, unlike most coherence views, maintained that starting points for empirical knowledge, are states that are in effect reasons for themselves. In mid-twentieth century, the idea that there are starting points for empirical knowledge that provide reasons for themselves (or are self-evident or infallible) finally lost credibility.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



be warranted and knowledgeable in perceptual beliefs, one must have operative justifications that provide reasons. Most perceptual beliefs do not need argument, or any other justificational background. They are starting points among empirically warranted beliefs and for empirical knowledge. Another epistemically internalist conception that had a brief run is that epistemic warrant is a matter of being epistemically blameless.³⁷ This conception is internalist in that it assumes that blameworthiness hinges entirely on states that are in the individual’s psychology. Blamelessness is compatible with systematic malfunction in an individual’s psychology. Malfunction undermines warrant. It renders a state incapable of being knowledge. Further, as has often been noticed, benighted superstitions and bad inductive practice can be epistemically blameless. So epistemic blamelessness is not sufficient for having epistemic warrant. A similar conception is that being epistemically warranted is being epistemically responsible.³⁸ Epistemic responsibility is a genuine epistemic good. But responsibility, like blame, is only loosely connected to epistemic warrant. Responsibility and blame are relevant only inasmuch as an individual has or has had some control. Some warranted belief formation—including initial formation of perceptual beliefs—is not under an individual’s control. Then blame and responsibility are irrelevant to the belief ’s being warranted. They are not necessary conditions. What subsequent epistemic control individuals may exercise on such beliefs does not account for the beliefs’ initial prima facie conduciveness to knowledge. Using an unreliable psychological system responsibly also does not suffice for epistemic warrant. Indeed, a responsible but unreliable formation of belief would block knowledge, which is an impossible effect of epistemic warrant (see condition (c) in Section 2.1). In light of the difficulties about control, one could change the view to: being epistemically warranted is either being epistemically responsible or operating in an epistemically permissible way. Even laying aside doubts about whether epistemic permissibility is strong enough for epistemic warrant, the conception still allows cases in which an individual is epistemically warranted, the “warrant” blocks knowledge. Some internalists take warrants constitutively to provide guidance, recommendation, or advice.³⁹ They often add that the point of epistemology is to help an individual determine what to believe from the first-person point of view. Following good guidance is an epistemic good. However, these conceptions are too narrow to yield understanding of knowledge or epistemic warrant. Again, initial formation of perceptual beliefs is commonly not open to following guidance or recommendation. Often, with crude versions of naturalism and reliabilism in mind, proponents of guidance conceptions insist that epistemology is normative. It is. As I explained in ³⁷ C. Ginet, Knowledge Perception, and Memory (Dordrecht: D. Reidel, 1975); R. M. Chisholm, Theory of Knowledge, 2nd ed. (Englewood Cliffs, New Jersey: Prentice Hall, 1977). ³⁸ L. BonJour, ‘Externalist Theories of Empirical Knowledge’, Midwest Studies in Philosophy 5 (1980), 53–73. ³⁹ J. Pollock and J. Cruz, Contemporary Theories of Knowledge (Lanham, Maryland: Rowman and Littlefield, 1999); R. Wedgewood, ‘Internalism Explained’, op. cit.; S. Berker, ‘Epistemic Teleology and the Separateness of Propositions’, op. cit., and ‘The Rejection of Epistemic Consequentialism’, op. cit..

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Section 2.1, epistemic warrant is a normative notion. It fulfills a standard for a type of epistemic success. This standard does not in itself guide. The relevant standard, or norm, is not in itself a prescriptive norm. A common mistake lies in thinking that all epistemic norms, indeed sometimes all norms, are prescriptive norms. Neither guidance nor prescription is essential to warrant or relevant to all types of knowledge.⁴⁰ Guidance is relevant to certain types of reasoning. But much epistemic warrant begins below psychological levels at which individuals can be guided. Guidance conceptions, like blamelessness and responsibility conceptions, take their cue from Descartes’ view that epistemology provides rules for the direction of the mind in the interests of gaining knowledge or cognition. (Descartes conceived belief as a product of the will, but proponents of guidance conceptions tend, rightly, not to follow him on this point.) Perceptual-belief formation does not admit of guidance. Moreover, if one’s belief-forming equipment is defective or unreliable, one’s attempts to follow good epistemic guidance can leave one without warrant or knowledge, even if one’s belief is true and not subject to Gettier failures. Indeed, following good epistemic guidance in such cases would still leave one in a condition that blocked having knowledge, which by condition (c) of Section 2.1 is incompatible with being epistemically warranted. Another internalist conception of epistemic warrant proposes that an individual is epistemically warranted in forming a belief if the belief is likely for that individual to be true, and so, epistemically appropriate for the individual to have the belief.⁴¹ The ‘for’ and ‘so’ are not clarified. They are supposed not to require reliability. The natural construal of ‘for’ is subjectivist. On that construal it simply means that the individual takes it to be reliable, or takes it to be likely true. A madman can meet the condition. So can a believer whose perceptual-belief-forming equipment is pathological. ‘For’ is perhaps intended to have a more objective status. The conception relies on an unspecified conception of epistemically “appropriate” states or procedures. What makes them appropriate (better, epistemically good), and whether any such good is

⁴⁰ Epistemology should set out other norms that guide. Discussion of reflective self-monitoring has a place in the epistemology of higher cognition. A common internalist mistake is thinking that all knowledge requires such guidance or in thinking that epistemic warrant must, in general, provide norms of guidance. Epistemic warrant and the basis of empirical knowledge begin below psychological levels at which individuals can follow rules for the direction of their minds. Arguments for deriving internalism from a guidance-based view of warrant occur in J. Pollock and J. Cruz, Contemporary Theories of Knowledge, op. cit., 130–137; and in R. Wedgewood, ‘Internalism Explained’, Philosophy and Phenomenological Research 65 (2001), 349–369, see 365–368. Both arguments state correctly that epistemic evaluation for warrant applies to states and events in an individual’s psychology. Wedgewood claims that a transition can be warranted (or “rational”, though this term misses entitlement) only if (a) it instantiates a rule that can be directly or non-derivatively followed, (b) it has the force of a “recommendation”, and (c) that it is followed because it is warranted (or is rational, or “makes sense”). The argument purports to show that reliability is not a requirement on warrant. The idea is that a recommendation to be reliable cannot be followed directly. I think that the argument provides no reason at all for the conclusion. (b) is not a reasonable condition on warrant. The epistemology of warrant is not confined to recommendation. Moreover, (c) does not accord with how either science or common sense construes formation of perceptual beliefs, many of which are clearly warranted. Nothing in either argument shows that reliability is not a framework-type, necessary condition on a psychological state or process’s being warranted. ⁴¹ J. Pryor, ‘What is Wrong with Moore’s Argument?’, Philosophical Issues 14 (2004), 349–378.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



the good of epistemic warrant are not explained. These are the cruxes of the matter. The relevant epistemic goodness must connect warrant both to truth and to knowledge in ways that I indicated in Section 2.1. Like the previous conceptions, this one does not do so. Internalist conceptions are often explicated by citing a list of ways of forming epistemically warranted or epistemically appropriate beliefs. Examples are forming perceptions, perceptual beliefs, perceptual memories, inductive inferences, and so on. Citations of such powers, whose exercises in us are typically warranted, do not suffice to illuminate what epistemic warrant is. A general gloss on wherein they constitutively conduce to knowledge is lacking. A serious conception of epistemic warrant— or “justification”—is often not developed at all. Most internalists over the last six decades have lacked any clear, guiding conception of what epistemic warrant is. Post-sense-data conceptions have provided no indication of wherein epistemic warrant is conducive to knowledge. That is the sense in which confusion reigns. In fact, an argument that applies to all the previously cited conceptions is that all allow epistemic warrant or justification to block knowledge in certain cases. They all fail to accord with condition (c) in the explication of conduciveness to knowledge that was set out in Section 2.1. Most of the internalist conceptions of epistemic warrant discussed so far do not directly face the task of defending (EI). This is the task of showing that, and why, an individual’s psychological states and events necessitate all features of epistemic warrant. Early modern philosophers confronted the problem directly. They constructed their views about why the mind, by virtue of God’s beneficence or by virtue of the mind-dependent nature of the world, necessarily tracks its subject matters. Many current philosophers are committed, explicitly or in an unarticulated way, to the view that something about the nature of perception or perceptual belief insures that perceptual beliefs are warranted. I do not know of a single substantial or interesting argument in recent philosophy for the position.⁴²

⁴² For a short time, deplorably and without argument, I claimed that there is such a connection. See T. Burge, ‘Interlocution, Perception, and Memory’, Philosophical Studies 86 (1997), pp. 21–47, esp. 31. Shortly thereafter, I rejected the claim. In ‘Perceptual Entitlement’, op. cit., 516–518, 532, I systematically avoid any claim that reliability is constitutive to perception, and I call attention to the contingency of the reliability of perceptual states. Ralph Wedgewood states that we can know that we are warranted in relying on perception purely by knowing the nature of perception. See ‘Primitively Rational Belief-Forming Processes’, in A. Reisner and A. Steglich-Petersen, eds., Reasons for Belief (Cambridge: Cambridge University Press, 2011), section III. The argument for this view seems to me both unclear and flawed in several ways. Either it moves illicitly from the necessity of some relevant perceptions’ and perceptual beliefs’ being veridical–a consequence of perceptual anti-individualism—to one’s having a reliable competence. Or it sets a requirement on warrant/ rationality that is clearly too weak. The too-weak requirement would be that warrant and rationality bear some constitutive, but not necessarily a reliable, relation to truth. This requirement falls afoul of condition (c) in the explication of conduciveness to knowledge, from Section 2.1. The argument that taking perception at face value is apriori warranted has the same flaws. There are other weaknesses in the argument that I will not go into. Nicolas Silins proposes that perceptual systems, or at least all ranges of perceptual attribution (such as the color range or shape range), are constitutively reliable. See his ‘Explaining Perceptual Entitlement’, op. cit.. No argument is offered.

[Note 42 continues below next page]

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     One prominent claim that an intrinsic aspect of an individual’s psychology is sufficient for warrant states that the phenomenology of perception and perceptual beliefs is the source of perceptual warrant.⁴³ A given type of perceptual phenomenology is supposed to make a perceptual state or perceptual belief with that phenomenology warranted. Perhaps some advocates think that phenomenology necessitates reliability. Many deny that reliability is required for epistemic warrant. They claim that perceptual phenomenology (constitutively) suffices for default epistemic warrant for perceptual belief. In any case, phenomenology itself does not necessitate reliability and does not constitutively (hence necessitate) conduciveness to knowledge, as epistemically warranted states must. Indeed, if a given phenomenology were not reliable, it would block knowledge, again contrary to condition (c) from Section 2.1. The only support given for this sufficiency view is that it is “intuitive”. It is intuitive that our conscious perceptual beliefs are warranted. It is intuitive that we commonly use perceptual phenomenology in our reliance on perceptual beliefs. We are right to do so. It is also intuitive that our perceptual beliefs and inductive inferences from perceptual beliefs are mostly prima facie warranted, without need for supporting argument or background beliefs or presumptions. None of these points implies that phenomenology insures epistemic warrant, or that conscious perceptual beliefs are by nature epistemically prima facie warranted. That is the issue. These internalists claim that necessarily epistemic warrant depends on nothing beyond the individual’s psychology. Here the relevant aspect of the [Note 42 continued:] I discuss reasons to deny such proposals below. Cf. also Origins of Objectivity, op. cit., chapters 8 and 9 and ‘Perception: Where Mind Begins’, Philosophy 89 (2014), 385–403. Ernest Sosa, in Reflective Knowledge (Oxford: Clarendon Press, 2009), 236, claims that when we think of a disposition as a competence, we require it to be reliable. He has a perceptual-belief-forming competence specifically in mind. See also his A Virtue Epistemology (Oxford: Clarendon Press, 2007), 106. This claim is mistaken. There is no oddity in thinking that someone has a competence in shooting a basketball—or in piano playing, or in archery—but that the competence is erratic and unreliable, and not just because it is interfered with by external factors. An ability is not a probability. More specifically, there is nothing impossible about a perceptual or a perceptual-belief-forming competence that is not reliably veridical in its natural habitat. Empirical science proceeds on this assumption. It investigates whether, to what degree and in what circumstances, perceptual competencies are reliable. ⁴³ Representative proponents are J. Pryor, ‘The Skeptic and the Dogmatist’, op. cit.; M. Huemer, Skepticism and the Veil of Perception (Lanham, Md.: Rowman & Littlefield, 2001) and ‘Phenomenal Conservatism and the Internalist Intuition’, American Philosophical Quarterly 43 (2006), 147–158; M. Steup, ‘Internalist Reliabilism’, Philosophical Issues 14 (2004): 403–425; and I. Dickie, ‘Visual Attention Fixes Demonstrative Reference by Eliminating Referential Luck’ in C. Mole, D. Smithies, and W. Wu eds., Attention: Psychological and Philosophical Essays (Oxford: Oxford University Press, 2011). Dickie declares that intentionality is ‘essentially a matter of subjective consciousness’, citing as opponents only naturalistic philosophers who envisage reduction of intentionality to non-intentional, non-mentalistic phenomena (299–300, 315). Such declarations are incompatible with what is known from the science of perception or perceptual-belief formation. Science freely postulates unconscious representational states (states with “intentionality”). It does not appeal to consciousness as the basis for explaining successful perceptual reference or perceptual attribution, even in cases of states that are conscious. There is an aberrant version of the view that phenomenology is sufficient for warrant that identifies phenomenology with environmental objects and attributes. I do not take this version seriously. See note 127.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



psychology is supposed to be the conscious phenomenology of the individual’s perceptual states or perceptual beliefs. An account of wherein the phenomenology bears a constitutive, or at least necessitating, relation to epistemic warrant—hence to conduciveness to knowledge—is completely lacking. The claim that phenomenology is what makes a perceptual state or perceptual belief epistemically warranted is hocus pocus. No proponent, to my knowledge, has attempted a serious explanation of how any specific empirical phenomenology, or consciousness, could in itself yield epistemic warrant. No one has explained wherein a phenomenology is constitutively conducive to knowledge. How phenomenology could constitutively determine these matters is mysterious. Perception is fundamentally a functional matter. A tendency to realize its representational function in representing the world correctly is its key contribution to epistemic warrant for perceptual belief. Although consciousness actually enhances veridicality and reliability, nothing about phenomenology is constitutive to this representational function. I have criticized the view that conscious phenomenal experience suffices for having empirical epistemic warrant. I shall return to that view. As an interlude, I discuss a claim that conscious phenomenal experience is necessary for having epistemic warrant. John Campbell argues that conscious phenomenal experience—indeed conscious focal attention—is necessary for having demonstrative reference to objects in thought, hence to having “justified” perceptual belief. He thinks that to “grasp” or “understand” a demonstrative in thought, one must consciously attend to the object that the demonstrative picks out in perceptual experience.⁴⁴ I believe this view to be mistaken. It is certainly not successfully argued.⁴⁵

⁴⁴ The view that focal attention is required for having demonstrative-governed perception-based beliefs was inspired by scientific work by Anne Treisman. In other work, Perception: First Form of Mind, forthcoming, I explain how the view has come to be undermined by psychological science itself. ⁴⁵ J. Campbell, Reference and Consciousness (Oxford: Clarendon Press, 2002), 7–9. D. Smithies, ‘What is the Role of Consciousness in Demonstrative Thought?’, The Journal of Philosophy 108 (2011), 5–34, especially 33ff., endorses Campbell’s argument and rests further arguments on it. I. Dickie, ‘Visual Attention Fixes Demonstrative Reference by Eliminating Referential Luck’, op. cit. also is influenced by Campbell’s claims. These authors assert that demonstrative reference in thought, at least when backed by categoreal attributives, can occur only consciously, and that no unconscious demonstrative-involving belief can be warranted. I do not accept these assertions. Unconscious perception can and does attribute categoreal attributes, like size and shape—for example in blindsight and in certain neuro-typical cases of human perception. See the citations in this note regarding blindsight. Unconscious perceptual beliefs sometimes are formed from such perceptions. Unconsciousness does not disqualify categoreal perceptual beliefs from involving demonstrative reference–from picking out entities in perception via contextual causal relation to them—or from being warranted, although actual blindsighters may form beliefs that are too weakly supported to count as warranted. Ian Philips doubts even the view that there is unconscious perception. Standard perceptual psychology postulates many cases of unconscious perception. Philips criticizes that standard view, and my and Ned Block’s acceptance of it, in ‘Unconscious Perception Reconsidered’, Analytic Philosophy 59 (2018), 471–514. No arguments are presented to show that consciousness is constitutively necessary to perception. The criticisms are empirical and case by case. Philips uses my characterization of perception as objective sensory representation by the individual as the framework for his criticisms (Origins of Objectivity, op. cit., 368–416). Philips’s main claim is that there is no unconscious perception by individuals. Philips invokes my notion by individuals. On my understanding, what it is for a perception to be by an individual is for it to be functionally integrated with exercises of whole individual functions. Such exercises can be active, such as turning one’s

[Note 45 continues below next page]

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     [Note 45 Continued:] head, or non-active, such as automatic freezing in response to danger. Philips’s holds that there is no evidence for unconscious perception that is not functionally isolated from exercises of whole individual functions. However, there is considerable evidence that unconscious perception guides action–say, in blindsight— and guides allocation of attention—whether active or “grabbed” attention. Philips assumes, I think mistakenly, that only active, guided attention is attributable to an individual. I think that in all relevant cases, active or “grabbed”, attention is attributable to an individual. There is also evidence that unconscious perception functionally interacts in systematic ways with later conscious perception by the individual. Such unconscious perception is thus closely integrated with whole-individual functions–with further perception and whole-individual responses to perception, including actions. I find Philips’ applications of my notion “by an individual” are often quite foreign. Philips’s focus on whether a phenomenon is perception by an individual is, I think, itself misconceived. Ned Block, who disagrees with most of Philips’s other contentions, shares this focus. See N. Block, ‘The Anna Karenina Principle and Skepticism about Unconscious Perception’, Philosophy and Phenomenological Research 93 (2016), 452–459; and I. Philips and N. Block ‘Unconscious Seeing’—a debate in B. Nanay (ed.) Current Controversies in Philosophy of Perception, (London: Routledge, 2016). Both authors take whether a visual state is attributable to an individual to be crucial to whether unconscious and conscious sensory representations are sub-species of a basic or important psychological natural kind. I disagree. I explicitly avoided taking being “by an individual” to be necessary to perception (Origins of Objectivity, op. cit., 369). ‘By the individual’ figures in my explication of perception because perception is paradigmatically by individuals. The phrase is a generic. It does not set a necessary condition. Perceptual constancies, conscious or not, and also by the individual or not, are necessary and sufficient for perception (Origins of Objectivity, op. cit., 413). Such constancies mark a basic, important psychological natural kind. The basic natural kind is perception! Conscious and unconscious species, whether by the individual or not, are sub-kinds. Some blindsight clearly guides individual action. So some blindsight is clearly by individuals. In supporting criticisms of what I had written on blindsight, Philips’s reporting of science is very unreliable. He writes, ‘we lack evidence that blindsight involves genuine perception [perceptual constancies] as opposed to mere sensory registration.’ He cites two studies of blindsight that suggest that specific types of visual detection— sensing motion not tracked in three-dimensional space, and sensing certain two-dimensional shapes—do not involve perceptual constancies. I think that these cases do not suggest, even prima facie, the presence of perceptual constancies. Philips generalizes from these cases as if they were the only relevant ones. He passes over, unmentioned, the very cases that I cited (Origins of Objectivity, op. cit., 374–375). Blindsight subjects’ grasping 3-dimensional objects, via smooth motion and with fitted aperture of grasp, constitute the sort of of evidence that supports attribution of perceptual constancies. A large scientific literature claims or strongly suggests perceptual constancies in blindsight. See ‘J. Danckert and Y. Rossetti, ‘Blindsight in Action: What Can the Different Sub-Types of Blindsight Tell Us about the Control of Visually Guided Actions?’, Neuroscience & Biobhavioral Reviews 29 (2005), 1035–1046; M. Perenin, and Y. Rossetti, ‘Grasping Without Form Discrimination in a Hemianopic Field’, Neuroreport 7 (1996), 793–7. M. Christensen, L. Kristiansen, J. Rowe, and J. Nielsen, ‘Action‐blindsight in Healthy Subjects after Transcranial Magnetic Stimulation’ PNAS 105 (2008), 1353–1357; R. Whitwell, C. Streimer, D. Nicole, and M. Goodale, ‘Grasping the Nonconscious: Preserved Grip Scaling to Unseen Objects for Immediate but not Delayed Grasping Following a Unilateral Lesion to Primary Visual Cortex’, Vision Research 51 (2011), 908–924; R. Whitwell, A. Milner, C. Cavina‐Pratesi, M. Barat, and M. Goodale, ‘Patient DF’s Visual Brain in Action’, Vision Research 110 (2015), 265–276. There are direct arguments for perceptual constancies in blindsight: I. Sperandio and P. Chouinard, ‘The Mechanisms of Size Constancy’ Multisensory Research 28 (2015), 253–283; R. Whitwell, I. Sperandio, G. Buckingham, P. Chouinard, and M. Goodale, ‘Evidence for a Functional and Anatomical Dissociation in the Use of Size Constancy for Perceptual Report and Goal-directed Grasping’ (abstract), Journal of Vision 15 (2015), 187. Philips gives back-up arguments that purport to show that none of the evidence used in science to support taking blindsight subjects to be relevantly unconscious really does so. I find the arguments unpersuasive. They join a long line of unsuccessful, rear-guard attempts to show or suggest that blindsight patients always perceive consciously. I think that the science will not be moved by Philips’s attempts along this line. Moreover, as Block indicates, there are many cases of unconscious perception by individuals who are neuro-typical.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Campbell’s argument goes as follows: You might acknowledge that ordinarily we would use visual information to interpret the demonstrative, but question whether it has to be conscious. The idea of visual information that is not conscious is made vivid by cases of blindsight. A blindsight patient is one who . . . has no awareness of objects in one half of his visual field. Nonetheless, when forced to guess about what is in the blind field, he may be reliably correct about, for example, the orientation, direction, and sort of object in the blindfield . . . .Couldn’t this subject use such visual information to achieve an understanding of a visual demonstrative? . . . we could try supposing as a thought experiment that we have a subject who has all the relevant visual information but is not yet conscious of the object in the blind field ( . . . ‘superblindsight’). [Campbell goes on to claim that super-blindsight is just guessing.]⁴⁶ . . . consider an ordinary case in which you and I are sitting at a dinner table with a large number of people around and you make a remark to me about ‘that woman’ . . . I cannot consciously single out the person you mean. All I get consciously is a sea of faces. But now we add some of what the blindseer has. You refuse to give me any further clues as to which person you mean, but you say, ‘Try to point to the woman I mean’. At first I protest that I can’t do that . . . but I do try to point, and to my surprise you say I’m pointing right at the person you mean. Suppose now that my conscious experience remains a sea of faces, but extend the reach of my reliable guessing so that it encompasses everything the blindseer can do. So I can make reliable guesses about what the person is eating, wearing, and so on, as well as reaching and pointing appropriately. But so long as my conscious experience remains a sea of faces, there is an ordinary sense in which I do not know what you mean. The problem here does not have to do with whether I am reliable. We can suppose that I am quite reliable in my guesses and we establish this over a series of such cases. The point is rather that I do not know who you mean until I finally look at where my finger is pointing or look to see who is wearing the clothes I described in my guesses. It is only when I . . . single out the woman in my experience of the room, when it ceases to be a sea of faces, and in my experience I focus on the person, that I would ordinarily be said to know who was being referred to. So it does seem to be compelling to common sense that conscious attention to the object is needed for an understanding of the demonstrative.⁴⁷ ⁴⁶ J. Campbell, Reference and Consciousness, op. cit., 144. ⁴⁷ J. Campbell, Reference and Consciousness, op. cit., 7–9. Campbell introduces this passage in misleading ways. He writes, ‘The issue is whether the blindseer has the very same way of interpreting the demonstrative as the ordinary subject has. That is, the question is whether for the ordinary subject consciousness of the object is not completely idle in an understanding of the demonstrative’ (8). These formulations are not equivalent to one another. And neither formulation is equivalent to the issue that the argument addresses–whether consciousness is constitutively necessary to having demonstrative thought. Anyone should agree that an unconscious “interpretation” (or mode of presentation or way of perceiving) differs from a conscious one. Moreover, the issue is not whether consciousness is ‘completely idle’ in a conscious individual’s understanding the demonstrative. The issue is whether it is constitutively necessary for demonstrative, de re thought. Consciousness quite obviously plays a role in actual competencies with conscious de re perceptual thought. It is empirically obvious that consciousness sharpens and enriches perceptual information in many uses. Its psychological role is hardly idle. Conflation between being

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     The argument fails in two ways. One centers on characterizing the super-blindseer as guessing. The super-blindseers are supposed to have reliably veridical perception and to make judgments with confidence. Actual blindseers take themselves to be guessing. But both they and super-blindseers have a real competence. A lot is known about the competence, and about the perceptual constancies involved. Given superblindseers’ competence at perceiving bodies and shapes as such, their reliability, and their confidence, they are not guessing. Campbell insists that even ‘fast, accurate, and reliable’ super-blindseeing ‘is still said to be ‘only guessing’.” (See note 46.) But where is it “said”, outside this insistence, that a super-blindseer would be guessing? Dictionaries explicate guessing as taking something to be correct without sufficient information to be confident about it. Super-blindseers’ competent, confident, information-based judgments fail this condition. Campbell’s argument cannot draw on shared common sense to say that a competent, confident, reliably veridical super-blindseer would be guessing. I and perceptual psychology (see citations note 45) take the ordinary blindseer to be perceptually picking out an object and attributing properties like shape and location to it. The ordinary blindseer has unconscious demonstrative thoughts as of categoreal properties formed from unconscious perceptions of, and as of, the object. I think that the super-blindseer’s thoughts—given their high reliability, confidence, and naturalness—would be knowledge. No attributions of guessing to the super-blindseer have any force. A second failure of the argument lies in its comparison of super-blindsight with attempts to understand the linguistic demonstrative references of another individual who gives no cues as to which face is meant or demonstratively referred to in thought. There is no mechanism or competence that would, in the absence of such cues, enable one to track which face the other individual intends. We do not read others’ minds except through perceived cues. So there is no way for the hearer in the sea-of-faces thought experiment to understand and competently track the speaker’s demonstrative usage, or to replicate it in thought. Here the situation differs crucially from blindsight and super-blindsight. Blindseers and super-blindseers have a known competence for tracking objects that they unconsciously perceive. The goal of Campbell’s argument is to show that conscious attention is necessary for the blindseers’ use of demonstratives in thought. To be successful, the argument must exhibit a case in which demonstrative thought is absent and the only feature that can account for its absence is absence of conscious attention. In the sea-of-faces thought experiment, conscious attention is absent. Its absence is supposed to explain why demonstrative thought is (supposed to be) absent. Also absent is a mechanism or competence for tracking and replicating the other’s demonstrative reference. This latter absence suffices to explain the hearer’s failure to think demonstratively about what the speaker means. The argument does not show that consciousness or a conscious mechanism is what explains the failure of the hearer to secure

constitutive to de re reference and being non-idle in normal, conscious de re reference figures in other philosophical discussion of these issues.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



demonstrative reference in the sea-of-faces case. Inferring the conclusion from the thought experiment is a non-sequitur.⁴⁸ Campbell frames the discussion as if it were analogous to understanding another’s speech. But relevant understanding for perceptual beliefs is different. It is a competence to rely on one’s own perception in contextually picking out objects in thought. The super-blindseer has relevant understanding of such thoughts—competence in thinking them. Campbell has not given the slightest reason to think that the superblindseer lacks relevant competence grounded in unconscious perception. Since ordinary blindsight is a defect, the individual is disinclined to rely on it. Ordinary blindsight is not strongly reliable. Hesitancy to form beliefs on the basis of ordinary blindsight and lack of strong reliability prevent it from yielding knowledge. The hypothetical super-blindseer, whose blindsight is natural, confident, and reliable, understands his or her thoughts in the only way that is relevant. Relevant understanding of demonstratives in thought consists in cognitive competence to think thoughts that pick out objects via perception. The demonstrative-like referential applications in the thoughts and perceptions pick out objects contextually, in ways that are embedded in perceptual constancies.⁴⁹ In sum, the argument that consciousness is constitutively necessary to having warranted, demonstrative-governed, perception-based thoughts clearly fails. Although actual blindseers lack knowledge, because their perceptions and perceptual beliefs are not reliable enough and because they lack confidence in their (forced) judgments, hypothetical super-blindseers’ perceptual judgments have not been shown to lack warrant or knowledge. I have criticized both the view that consciousness in perception is sufficient and the view that consciousness in perception is necessary for epistemic warrant in perceptual belief. The claim that conscious phenomenology suffices for epistemic warrant is by far the more pernicious. The hopelessness of the position follows from the fact that not even the natures of perception and perceptual belief, or any given perception or perceptual belief—let alone consciousness in them—constitutively insures epistemic warrant. I shall discuss this point as part of a general explanation of why internalism is unacceptable. ⁴⁸ Here the weakness in Campbell’s argument is similar to a weakness in Laurence BonJour’s argument from clairvoyance, which I discuss later in this section. ⁴⁹ Some of Campbell’s argument hinges on the entirely armchair claim that a blindseer only perceives functional properties, “affordances”, Ibid, 138–145. It is an open empirical question whether any functional attributives are formed in the visual system, as distinguished from the actional system. I assume that some are. Campbell’s armchair claim that blindsight involves perception only of affordances is, however, flatly incompatible with science. Shape constancies and categorizations of shape and shaped-body are attributed in the science to ordinary blindseers. See R. Whitwell, C. Streimer, D. Nicole, and M. Goodale, ‘Grasping the Non‐conscious: Preserved Grip Scaling to Unseen Objects for Immediate but not Delayed Grasping following a Unilateral Lesion to Primary Visual Cortex’, op. cit.; I. Sperandio and P. Chouinard, ‘The Mechanisms of Size Constancy’, op. cit.; R. Whitwell, I. Sperandio, G. Buckingham, P. Chouinard, and M. Goodale, ‘Evidence for a Functional and Anatomical Dissociation in the Use of Size Constancy for Perceptual Report and Goal-directed Grasping’, op. cit.. Moreover, Campbell’s claim is incompatible with the main shape of modern vision science (see note 45). Modern (post-1970) mainstream vision science never takes perceptual attribution of functional properties to be independent of perceptual attribution of categorical properties. One perceives something as prey only by perceiving it as having such attributes as shape, size, and body.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     The explanation is what is most important. I first state an argument that it frames. Then I explain and motivate its component steps: (1) Many individuals are epistemically warranted in, entitled to, many of their perceptual beliefs. (2) An individual’s perceptual belief ’s being produced or sustained in a way that reliably engenders true beliefs of that belief ’s content-type is constitutively, hence metaphysically, necessary to the belief ’s being epistemically warranted. (Reliability is to be assessed in the perceptual and belief systems’ normal environment, in a sense of ‘normal’ to be discussed later in the section.) (3) For a psychological system of states that actually has a warranted perceptual belief that has actually been produced or sustained in a way that reliably engenders true beliefs of that belief ’s content-type, it is not metaphysically necessary that a belief of the same content-type that is embedded in the same configuration of psychological state types, and that is produced or sustained in the same way, is reliably true. (Nor therefore is the perceptual belief reliably true by its nature or content.) Hence (4) It is metaphysically possible for the whole configuration of psychological state types that embeds an epistemically warranted perceptual belief to remain the same, while a perceptual belief of the same type is (would be) epistemically unwarranted. (4) entails that the epistemic warrant, the entitlement, to a perceptual belief does not supervene on the system of psychological states in which it is embedded, contrary to (EI). I have already supported (1). Rejection of (1) in the interest of requiring propositional buttresses for perceptual beliefs is deeply implausible. Young children have warranted, knowledgeable, perceptual beliefs, such as that round body is red, or that is a face. They lack reasons and background beliefs that function to support, protect, or buttress their perceptual beliefs. Adult epistemic practice presumes no need for propositional support or protection for perceptual beliefs. Denial of these points often evinces bad theory’s swamping good, common sense judgment. Or denial evinces a focus on a special type of knowledge or epistemic warrant that not all individuals with knowledgeable perceptual beliefs have. I discuss these matters further at the end of this section and in Sections 2.3–2.5. Although I think (1) clearly true, a more complex argument could be refashioned from steps (2) and (3) alone. In this section, I focus on (2) and (3). I believe that they are the central claims at issue. I think that most epistemic internalists would be disposed to dispute either (2) or (3), or both. I explain why such dispositions should be reconsidered. I begin with the claim in (2): An individual’s perceptual belief ’s being produced or sustained in a way that reliably engenders a true perceptual belief of that belief ’s content-type is constitutively, hence metaphysically, necessary to the belief ’s being epistemically warranted.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



I argued for this claim in Section 2.1. I elaborate the argument here. Later in this section, I develop the relevant type of reliability required for epistemic warrant. Epistemic warrant’s being epistemically good consists in its conducing to knowledge—in its marking a belief or inference as being produced or maintained through a competence that tends to be, or to contribute to having, knowledge. If a belief were produced or sustained by a competence that was in every relevant respect unreliable, the belief would not be formed or maintained in a way that could be knowledge, much less tend toward having knowledge. Its being formed in that way would in fact prevent the belief from being knowledge. Epistemic warrant would in those cases mark a decisive hindrance to having knowledge. Such a result—a violation of condition (c) on conduciveness to knowledge from Section 2.1—prevents any property that could possibly block knowledge from being epistemic warrant. Constitutively, epistemic warrant marks a state or process as being situated in a pattern of psychological competencies that conduces to having knowledge. It is impossible that one lacks knowledge by being epistemically warranted. One cannot ignore the connection between the concepts epistemic and knowledge. So all epistemic warrant constitutively requires that the relevant competence for producing or sustaining a warranted belief be relevantly reliable in yielding true belief. An analogous argument applies for epistemically warranted inferences, or other transitions, to a belief. In many cases, an epistemic warrant certifies true beliefs prima facie as knowledge. Among these cases are epistemically basic beliefs—including perceptual beliefs. In such cases, the basic or canonical warrant (perceptual entitlement, for perceptual beliefs) certifies true beliefs as knowledge, in the absence of certain undermining conditions. I conjecture that these conditions are an absence of counter-warrant, an absence of internal psychological incoherence, and an absence of Gettier conditions. A true belief could never be certified as knowledge by epistemic warrant, even if it met such absence conditions, if it were produced or maintained by a competence that was relevantly unreliable. So basic epistemic warrants for basic beliefs could not certify those beliefs, even prima facie, as knowledge if epistemic warrant did not require that the competence for producing and maintaining the beliefs be relevantly reliable. Many internalists cheerfully deny that any sort of reliability is a condition on a belief ’s or inference’s being epistemically warranted. Some offer substitute conditions—that the individual not believe that a warranted state is produced in an unreliable way, or that the individual have a warranted belief or presumption that the state is produced in a reliable way. Denials that reliability is a constitutive aspect of epistemic warrant derive from confused failure to recognize that epistemic warrant is an epistemic good in that it constitutively conduces to knowledge. The epistemic good, epistemic warrant, cannot in any situation block or prevent knowledge of the warranted belief by the believer. A belief ’s being produced in a way that does not reliably engender true belief blocks knowing it. This consequence is apriori unacceptable. Such a belief could not be epistemically warranted. Such denials are historically aberrant. No philosopher in the tradition would have taken seriously a denial that his analog of warrant or justification must be reliable in

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     marking true belief.⁵⁰ Each would have taken it as basic that a good procedure for gaining knowledge, or cognitio, or scientia, or Erkenntnis must be reliable. Each would have recognized that a way of producing or sustaining a belief or inference that is relevantly unreliable could not mark a true belief as knowledge, and could not conduce to having knowledge. Epistemic warrant constitutively marks a true belief as conducing to knowledge. So an epistemically warranted procedure for gaining knowledge must be relevantly reliable, in producing truth. Denial that a warranted belief must be produced or sustained in a reliably veridical way betrays conceptual confusion about what epistemic warrant is. It would be seriously wrongheaded to respond in the following way. The early-modern internalist tradition of guaranteeing that analogs of epistemic warrant are conducive to knowledge and reliable in yielding truth was an optional aspect of the tradition. The commitments to idealism and metaphysical schemes that claimed that certain good ways of empirical thinking must necessarily yield knowledge and understanding were optional. What mattered to the early modern philosophers was their internalism and their articulating epistemically responsible procedures. So the right understanding of their analogs of epistemic warrant is an internalist conception that centers on responsibility. Making such a position explicit should bring out how misguided, and unhistorical it is. One needs to know only a little of the history of early modern philosophy to recognize that this gloss is uninformed. The early moderns aimed at understanding the new science—understanding how thinking scientifically contributed to arriving securely at an understanding of the structure of the world. Understanding our scientific tracking of the world was Descartes’ primary objective. With the exception of Hume, the early moderns overestimated how much of this tracking could be understood through reflection alone. Many of them felt more confident than we can be in appealing to theological or other metaphysical strategies to explain such tracking. They thought that by articulating procedures that fitted with their metaphysical schemes, they were showing how minds were guaranteed to track the world, at least in its larger patterns. Internalism was a consequence. But the primary objective in exploring analogs of epistemic warrant was to understand what psychological states and procedures conduced to their analogs of knowledge. ⁵⁰ As I mentioned earlier (note 32), the philosophical tradition rarely focused on knowledge. It focused on scientia or (for Kant) Erkenntnis (“cognition”). The former is very roughly analogous to scientific knowledge. Kant distinguished Erkenntnis (“cognition”) from Wissen (knowledge). For him Erkenntnis is a complex notion. In the form that interested Kant most, it is also very roughly analogous to a high-level type of knowledge, in that it requires an ability to prove objective validity (Critique of Pure Reason, Bxxvin). I do not know whether before Kant, there was use of terms like ‘warrant’ or ‘justification’. Kant has a close, counterpart notion for justification–proof of objective validity. Before Kant in the early modern period, the analog of warrant or justification was good procedure for obtaining scientia. Descartes certainly focused on such procedures. He and Kant regarded their analogs of warrant as reliable in producing truth and in producing their analogs of knowledge. A twentieth century recovery of the insight that reliability is a condition on warrant, and on any good procedure for yielding knowledge or scientia, occurs in Alvin Goldman, ‘What is Justified Belief?’, op. cit.. Without endorsing all of Goldman’s positive view, I believe that his emphasis on reliability as a necessary condition on being warranted (“justified”) has been important and salutary.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



The ethics of epistemically responsible practice was set out by Descartes to serve an understanding of science’s tracking the world. Indeed, except for Descartes, none of the early moderns featured notions of epistemic responsibility in their accounts of (analogs of) knowledge. The central focus, even in Descartes, was how thought’s, especially scientific thought’s, tracking the world could be explained and understood. Modern internalists who think that an individual or state can be epistemically warranted, but be produced or sustained in a way that is unreliable in producing true beliefs, propose a notion of epistemic warrant that could consistently be a hindrance to knowledge. The early moderns would have and should have regarded any such notion as apriori unacceptable as a notion of epistemic warrant. It is certainly true that there are coherent internalist notions of being responsible, or of being blameless, or of using one’s psychology as best one can, in serving knowledge and true belief. These notions do have uses in understanding norms for some epistemic practices. They are not central to understanding the factors that make one’s psychological states good in knowing the world. As I have noted, the blame/responsibility notions do not even apply naturally to the most basic way of forming knowledgeable beliefs—entitled production of perceptual beliefs. Initial formation of perceptual beliefs, and epistemic warrants for them, begin below the level of norms of blame, responsibility, or guidance. How does the deep connection between epistemic warrant and knowledge bear on the controversies over defining knowledge? Does such a connection block attempts to analyze the concept of knowledge in terms of truth, epistemic warrant, and other concepts? Does it support the idea that epistemic warrant is to be defined in terms of knowledge? No, to both questions. I do not purport to analyze or define concepts as complexes of other concepts. I doubt common understandings of these enterprises. I think that most important philosophical concepts are not definable. They are among the conceptual starting points for understanding the world. However, the claims set out here are not incompatible with defining knowledge partly in terms of epistemic warrant. The connections between epistemic warrant and knowledge that I have articulated are certainly not obviously definitional or analytic connections. They are also not incompatible with an attempt to analyze epistemic warrant in terms of knowledge. I do not propose definitions of either knowledge or epistemic warrant. Specifically, I do not hold that the concept knowledge is part of the definition of the concept epistemic warrant, or that knowledge is otherwise prior to epistemic warrant in some order of understanding.⁵¹ I tend to think that the concepts of knowledge and ⁵¹ Thus my view is not intended as a form of “knowledge first” approach to epistemology. See T. Williamson, Knowledge and Its Limits (Oxford: Oxford University Press, 2001). On such approaches, epistemic warrant and the psychological competencies, states, operations, and structures that fall under epistemic norms tend to be ignored. Such approaches tend to center on counterfactual, “non-luck” accounts of relations between beliefs and the world in attempting to illumine the concept knowledge. While these accounts can be valuable in understanding counterfactuals that underlie conduciveness to truth, they provide little understanding of the psychological bases for knowledge–how psychological competencies, states, operations, and structures contribute to gaining knowledge. Nor do they tend to provide insight into epistemic norms that govern the relevant psychological phenomena. Epistemology

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     epistemic warrant are primitive, but apriori reciprocally inter-dependent. I am doubtful about definitions of important concepts (or analysis of the meaning of terms) that play such rich roles in our understanding of the world. I do believe that it is apriori that knowledge requires epistemic warrant and true belief. I also believe that it is apriori that epistemic warrant marks conduciveness to knowledge and reliability regarding truth. Most important apriori truths are synthetic in the sense of not being purely products of analysis of concepts, on an analogy with analysis of meaning. (I think that all truths are synthetic in the sense that their truth depends on being made true by a subject matter.) What is clear is that no property that does not conduce to knowledge and reliably signal truth could be the epistemic good that notions of epistemic warrant, entitlement, and justification in fact mark. No property that would prevent knowledge could be epistemic warrant. Epistemic warrant constitutively marks psychological states as deriving from competencies or procedures that yield, sustain, or contribute to knowledge. I turn to step (3) in the argument against (EI): For a psychological system of states that actually has a warranted perceptual belief that has actually been produced or sustained in a way that reliably engenders true perceptual beliefs with that belief ’s content-type, it is not metaphysically necessary that a belief of the same content-type that is embedded in the same configuration of psychological state types, and that is produced or sustained in the same way, is reliably true. (Nor therefore is the perceptual belief reliably true by its nature or content.) Reliability is coded into the very nature of a few types of psychological states and events. A belief in a logical or mathematical truth is reliably true by its nature. Of course, belief in such a truth on a whim does not derive from competence that is connected to the belief ’s subject matter in a reliable way. Accepting a mathematical truth on a whim would be reliably true by the nature of the belief, but not through the nature of a competence that reliably connects to the belief ’s being true. So a belief in a logical or mathematical truth on a whim is not epistemically warranted. However, understanding the representational content of such a belief can constitute a good, reliable route to true belief. For relatively simple logical and mathematical truths, their content is self-evident: understanding their content suffices to warrant belief in them.⁵² So believing on the basis of understanding the content makes one’s belief

must illumine both matters. Although I share Williamson’s doubts about twentieth-century attempts to define the concept of knowledge and his rejection of epistemic internalism, I do not find his arguments for the “knowledge first” view clear or persuasive. Beyond these remarks, I will not discuss that work or attempts by others to develop it. I mainly caution against taking my point that the concept of epistemic warrant bears apriori relations to the concept of knowledge to be associated with the slogan, “knowledge first”. Although there is certainly value in approaches motivated by this slogan, I regard it, and them, as uncongenial in their way of pursuing epistemology. ⁵² I support this view in Truth, Thought, Reason: Essays on Gottlob Frege (Oxford: Oxford University Press, 2007), 65–72, 301–316; and ‘Logic and Analyticity’, Grazer Philosophische Studien 66 (2003), 199–249.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



formation not only reliably true. It makes it reliably true by a route that is reliably connected to the subject matter that makes the belief true. Reliability is coded into accepting self-evident necessary truths on the basis of understanding them. Similar points apply to deductively valid inferences. An inferential transition made via a competence individuated by a deductive inference rule is warranted by its nature. Epistemic warrant is necessitated by these types of psychological states and events. Empirically warranted psychological states and events are different. There is no intrinsic guarantee of their reliability. In the first instance, reliability is a property of ways of producing kinds or types of beliefs. It is not constitutive to the nature of any perception or perceptual belief that it is reliably veridical. A natural competence for producing any of these states could be relevantly unreliable. Sustaining such beliefs in memory depends for its warrant partly on the initial belief state that is retained. There is nothing about perceptual belief that makes it constitutively, for every possible believer with a perceptual system, prima facie knowledge. Parallel points apply to most types of empirical inductive inference. There may be no perceptual systems in nature that are naturally unreliable in their natural habitats. There is, however, no metaphysically necessary bar to a perceptual system’s being naturally unreliable in yielding veridical perceptions or perceptual beliefs. There are almost certainly perceptual state types in nature that are relevantly unreliable. Accuracy versus speed trade-offs are a prominent source of such cases. The competence in generating such state types is not reliably accurate in normal, natural conditions. There may be, in some individuals, whole ranges of types of perception, perhaps attributions of color shades, that are unreliable. The key terms for evaluating (EI) are ‘constitutive’ and ‘metaphysically necessary’. By its nature—hence constitutively—epistemic warrant marks a perceptual belief type, produced or sustained in a given way, as conducive to being knowledge. If any perceptual belief is epistemically warranted purely by virtue of being that psychological type of perceptual belief, there must be something metaphysically necessary that makes it conducive to being knowledge. I think, to the contrary, that perceptual beliefs are epistemically warranted by virtue having a metaphysically contingent property. The property is being produced or sustained through a well-functioning belief-forming competence that reliably produces true beliefs of the same content-type by reliably connecting them to their subject matters. Strictly speaking, most perceptual states and perceptual beliefs are not veridical.⁵³ Human perceptual states tend, however, to be approximately accurate. The human visual system has been shown to be amazingly approximately accurate in its natural domains of application. Such “domains” include normal viewing conditions of macro-objects at relatively close range. Various optimality theorems have been proved, indicating that given such biological constraints as limitations on acuity, the human visual system is close to optimally accurate in several visual tasks, in ecologically normal circumstances.⁵⁴ Checks of perceptual states for accuracy use ⁵³ I discuss this issue in detail in Section 2.5. ⁵⁴ C. Blakemore, ‘The Range and Scope of Binocular Depth Discrimination in Man’, Journal of Physiology 211 (1970), 599–622; W. Geisler, J. Perry, B. Super, and D. Gallogly, ‘Edge Co-occurrence in Natural Images Predicts Contour Grouping Performance’, Vision Research 41 (2001), 711–24;

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     converging, mutually supportive tests. The conclusions about optimality are empirically grounded. They do not derive from considerations regarding the nature of perception. There is nothing impossible about any perceptual state type’s not being reliably accurate, even reliably approximately accurate. Reliability hinges on propensities of inter-relation between perceptual state types and features of the environment that they represent and that molded the natures of those perceptual state types. Perceptual systems, and other representational systems, evolved by serving biological goods—principally, survival long enough to reproduce. Accuracy is not in itself a biological good. Nature did not select for accuracy per se. Accurate perceptual states evolved as part of a practically fit package. Speed and quick-trigger avoidance mechanisms can and sometimes do outweigh accuracy to yield practically valuable perception-forming routes that are not reliably accurate. Or energy saving in producing just-good-enough perception might outweigh accuracy, to yield just-fit-enough perceptual systems. It is metaphysically possible that perceptual systems be reliably useful without being reliably accurate. As a matter of empirically established fact, reliable approximate accuracy made enough of a contribution to fitness to be a feature of most perceptual systems, and probably most perceptual states. But no one who reflects on the evolutionary process can reasonably believe that there is a metaphysical necessity that guarantees even reliable approximate accuracy in perception. Reliable approximate accuracy is not constitutive to or metaphysically necessary for perception or perceptual belief.⁵⁵ To be a perception-forming mechanism, it is enough to be embedded in a perceptual constancy.⁵⁶ Perceptual systems are constitutively associated, somewhere in their existence or evolution, with formation of some approximately accurate perceptions. The representational contents of perceptual states cannot be explained unless some of the states that are produced in the system are related, somewhere in the evolution or content-determining maintenance of the perceptual system, to perceptual states that were approximately accurate regarding environmental entities.⁵⁷ This point does not imply the reliability of any perceptual states or systems. What we know about evolution of perceptual states implies that it is possible, and sometimes actual, D. D’Antona, J. Perry, and W. Geisler, ‘Humans Make Efficient Use of Natural Image Statistics’, Journal of Vision 13 (2013), 1–13; J. Burge and W. Geisler, ‘Optimal Disparity Estimation in Natural Stereo Images’, Journal of Vision 14 (2014), 1–18. ⁵⁵ Factive psychological states might be thought to be trivially reliable. One might infer that seeingbased beliefs, factively construed, are warranted by nature. Cheaply bought, cheaply made. The norm of warrant is applicable to fallible types of perceptions and perceptual beliefs. Building veridicality into the taxonomizing of psychological types bypasses one of the basic features of warrant. Warrant allows for false beliefs whose falsity hinges on incomplete information that does not result from a cognitive or perceptual defect. Warrant concerns epistemic good use of available information. Epistemic warrant fulfills a norm for our psychological capacities. Those capacities are scientifically type-individuated as fallible perceptual states and fallible perceptual beliefs. For more discussion of this issue, see my ‘Disjunctivism and Perceptual Psychology’ Philosophical Topics 33 (2005), 1–78; and ‘Disjunctivism Again’, Philosophical Explorations 14 (2011), 43–80. For further argument against taking factive states as basic for epistemology, see Ralph Wedgewood, ‘Internalism Explained’, op. cit., 361. ⁵⁶ See Origins of Objectivity, op. cit., chapter 9; and ‘Perception: Where Mind Begins’, op. cit.. ⁵⁷ See Origins of Objectivity, op. cit., chapter 3, esp. 67ff..

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



that perceptual states are not reliably accurate, even in conditions in which they evolved or were maintained. The same evolutionary considerations indicate that it is metaphysically possible that there be perceptual systems that are systematically unreliable in yielding veridicality, even in the conditions that they evolve to deal with. So neither perceptual states nor perceptual systems are relevantly reliable constitutively, or by any other type of metaphysical necessity. It does not follow by necessity from what it is to be any perceptual state type that formation of instances of the type is reliably accurate, or a contributor to epistemic warrant.⁵⁸ An analogous point applies to perceptual systems. Since perceptual beliefs depend on perceptual states for their natures and for being warranted, and since there is no metaphysically necessary guarantee that perceptual states are reliably accurate, there is no metaphysically necessary guarantee that perceptual beliefs are reliably true. Of course, in many counterfactual situations in which actually reliable perceptual belief types are unreliable, reflective individuals with enough experience, might develop caution that the actual individuals do not need. But (3) only depends on a metaphysical possibility. One can imagine an individual that does not get the relevant experience. Either the individual dies soon after forming the first relevant perceptual belief, or the individual is lucky in being exposed only to cases in which the perceptual beliefs are in fact true, even though their mode of production lacks a propensity to yield true belief of the relevant kind in the individual’s normal environment. Step (1) assumes that the relevant perceptual belief is warranted through entitlement. So its being prima facie warranted does not depend on background beliefs that buttress the perceptual beliefs. So the presence of either buttressing or cautionary background beliefs is irrelevant to the initial prima facie entitlement to the perceptual beliefs. By (3), one can hold a given psychological system of states fixed, while the reliability of the fixed way of producing and sustaining a perceptual belief could vary. By (2), perceptual beliefs that are epistemically warranted in the most basic, canonical way must—as a matter of what it is to be epistemically warranted—be formed through a competence that reliably produces true beliefs. So perceptual beliefs are not epistemically warranted by metaphysical necessity.⁵⁹ Specific perceptual and perceptual-belief kinds could remain fixed while the epistemic warrant, the entitlement to the perceptual beliefs, could vary. So, contrary to (EI), epistemic ⁵⁸ A perceptual state type, or a perceptual belief type, is obtained from an instance of the type by abstracting from the occurrence of the state type in any given person at any given time, and correlatively by extracting de re referential applications that occur in the representational content. ⁵⁹ D. Smythies in ‘What is the Role of Consciousness in Demonstrative Thought?’, op. cit. and I. Dickie, in ‘Visual Attention Fixes Demonstrative Reference by Eliminating Referential Luck’, op. cit. claim that whether demonstrative reference in perceptual thought is possible depends on whether such thought has a justification. Nothing in the science of vision or the psychology of perception-based belief suggests any such condition on successful context-dependent perceptual reference (that is, on perception of particulars). Perceptual belief can refer to what the underlying perception can refer to. Epistemology cannot reasonably issue such armchair constraints on psychology. Neither perceptual, context-dependent reference (perceiving a particular in a given situation) nor demonstrative reference in perception-based belief depends by metaphysical necessity on any sort of warrant. To have a representational psychological competence, there must be a capacity for representational success. But the nature of the competence does not metaphysically guarantee that successes meet standards of justification or warrant.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     warrant for perceptual beliefs does not supervene on, or derive purely from, the natures of the psychological states or the natures of the psychological operations that figure in the entitlement. Perceptual beliefs’ being epistemically warranted depends partly on metaphysically contingent relations between the psychology and the environment. It depends on the deeply non-accidental, but metaphysically contingent fact that reliability is biologically valuable.⁶⁰ Since empirical warrant must be grounded in perceptual belief, all empirical warrant constitutively depends on factors beyond the individual’s psychology. Contrary to Epistemic Internalism (EI), psychological states’ being epistemically warranted does not supervene on the psychology of the individual in those states. Reliability is not coded into the natures of the states. A perceptual belief ’s being on a good route to truth is not insured by metaphysical necessity. Arriving at this conclusion has depended on two types of reflection. One is, I think, apriori. It is reflection on the nature of epistemic warrant. There are many epistemic goods that do not require reliability. Such goods can be, and have been, confused with epistemic warrant. But the role of epistemic warrant (or “justification”) in epistemic practice and the traditional role of its ancestors—Descartes’s good rules for the direction of the mind and Kant’s proofs of objective validity—requires that epistemic warrant mark those competencies that produce or sustain belief as conducing to knowledge. Conducing to knowledge requires that the competencies be reliable in producing or sustaining true belief. In the philosophical mainstream that ran from Descartes to mid-twentieth century, the requirement of reliability on epistemic warrant is unquestioned. Many have doubted the requirement since then. I believe that these doubts derive from a deficient understanding of the notion of epistemic warrant, and its ancestors. I have tried to show that conceptions of epistemic warrant that have been meant to motivate internalism (EI), or to doubt the reliability requirement, have straightforward defects. The other type of reflection centers on perception. This reflection differs from anything provided in the tradition. Relevant reflection features the place of perception in evolution. The early moderns could not have reflected on evolution, of course. Philosophical work in the twentieth century can invoke no such excuse. Sense-data theory was a poorly grounded dead-end in philosophy. Post-sense-data theory of knowledge largely failed to use perceptual psychology to isolate perception as a central psychological kind. And it failed to use evolutionary biology to recognize that the reliable (approximate) veridicality of perceptual states and perceptual systems is metaphysically contingent. Since evolution does not select for veridicality, veridicality is hostage to biological usefulness. Usefulness of reliable veridicality in ⁶⁰ Natural empirical inductive/abductive inferential patterns are probably not by nature, or constitutively, reliable or warranted. Most do reliably match environmental patterns. But the reason that shows that perceptual beliefs are not constitutively warranted seems to show that empirical, non-demonstrative inferences are not either. Most of our inductive/abductive patterns probably do not constitute a logic that, purely by the form and content of the representational transitions, necessarily supports truth. This issue is, however, complex. I doubt that we know enough about empirical inductive/abductive inferences to warrant confidence on the matter. For more discussion, see ‘Epistemic Warrant: Humans and Computers’, op. cit..

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



perception is not metaphysically guaranteed by any force, natural or otherwise. A consequence of these two forms of reflection is that epistemic internalism is unacceptable. I turn to this section’s final task. Recent epistemic internalism has leaned on two stock arguments. One tries to show that epistemic warrant without reason is impossible, and directly infers internalism. The other purports to show that being produced by a reliably veridical competence is not required for being epistemically warranted. I criticize these arguments. The first argument, by Laurence BonJour, denies warrant to a reliable clairvoyant. So reliability is not sufficient for warrant. Reliable clairvoyance is supposed to be analogous to reliable perceptual belief for which the perceiver lacks a reason-based justification. The argument concludes that reliable perceptual belief is not warranted without support by reasons.⁶¹ The argument’s main point is to show that mere reliability is not sufficient for epistemic warrant. To support this conclusion, BonJour claims that a clairvoyant would be in the rationally incoherent position of not being able to make sense of his/ her own competence. This latter claim is buttressed by suggesting that because the reliability relation is ‘outside the ken’ of the clairvoyant, it cannot help make the clairvoyant’s beliefs warranted. It is also claimed that from the clairvoyant’s subjective perspective, it is an accident that the belief is true. It is proposed that the warrantedness of the beliefs should be judged ‘from the clairvoyant’s perspective’, rather than from a perspective that is unavailable to him/her. As the argument claims, reliability is not sufficient for epistemic warrant. Not just any reliable route to truth yields epistemic warrant. The reliability must derive from a well-functioning competence that is reliably veridical, and connected in a natural, non-accidental, systematic way to the subject matters that make the warranted states or transitions representationally successful. And there must be no defeaters. Most of us think that clairvoyance is not a real, or even nomologically possible, psychological competence. No known or plausibly conjectured mechanism connects a putative clairvoyant to relevant subject matters. Intuitions behind the argument are biased by these attitudes. If clairvoyance is to parallel perceptual belief, one must suppose that clairvoyance is a real, reliable competence—backed by some causal mechanism—and that the clairvoyant lacks reason to doubt possession of the competence. The clairvoyant would thereby be like many ordinary human perceivers, including all young children. Perhaps this supposition is impossible. Then the appeal to clairvoyance has no standing in the discussion. Let us suppose clairvoyance of this sort to be possible, for the sake of argument.

⁶¹ For the argument from clairvoyance, see L. BonJour, ‘Externalist Theories of Empirical Knowledge,’ op. cit., 59–65. In using the argument, philosophers often move as does BonJour, without comment, from its conclusion to epistemic internalism. However, empirical reasons are epistemic warrants. To be good reasons, they must be reliable in supporting what they are reasons for. The premises for empirical inductive arguments, and almost surely the inductive transitions from them to empirical conclusions, are not reliable by metaphysical necessity. As it stands, the argument does literally nothing to support (EI). The argument is more relevant to entitlement–to opposing step (1) of the argument set out above.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Well-functioning empirical competencies’ tracking of truth is metaphysically contingent, but not accidental. To yield warrant, the competencies’ reliable veridicality must be grounded in a patterned connection between (a) environmental conditions that render the beliefs true and (b) the competence’s operations. So a psychological competence cannot constitutively involve help from benevolent angels, or bear a mere de facto correlation with its successes. Suppose that we stipulate that clairvoyance is a psychological competence that meets these conditions. Suppose that there is a scientific explanation of the clairvoyant competence. Suppose that the clairvoyant individual lacks grounds either to believe or to doubt possession of it. Then contrary to the suggestion of the argument, despite lack of meta-beliefs about their situation, neither an individual with perceptual beliefs nor an individual with reliable clairvoyance would be in a rationally incoherent position. If one supposed that clairvoyance were a natural capacity whose reliability were grounded in real systematic relations to the subject matter, and that there were no antecedent presumptions against its viability, clairvoyants would be fair analogs to people who accept their perceptual beliefs without relying on metarationalizations. BonJour gives no cogent reason to believe that unsophisticated clairvoyants, under these conditions, would not be entitled to their clairvoyant beliefs. The competence’s deliverances could be prima facie epistemically warranted.⁶² Sophisticated individuals who knew their clairvoyance to be reliable would, of course, also not be in an incoherent position. Our background knowledge that clairvoyance is actually not a competence—and possibly could not be a real competence—supports our intuitive judgment that a reliable clairvoyant would lack warrant and knowledge. But our background knowledge also disqualifies actual clairvoyance as an analog to unreasoned perceptual belief. BonJour begs the question in suggesting that since reliability is outside the ken of the clairvoyant, it cannot help make his or her beliefs warranted. Further, it would be no more an accident from the natural, competent clairvoyant’s perspective that his or her beliefs are true than it is an accident from a child’s perspective that his or her perceptual beliefs are true. The fact that neither the child nor the natural clairvoyant has a reason for the belief hardly makes the beliefs arbitrary or accidental to the believer. BonJour’s claim that the warrantedness of the beliefs should be judged from the clairvoyant’s perspective, rather than from an unavailable perspective, again begs the question. Perceptual believers need not have meta-perspectives on their perspectives in order to be warranted. They need not have reasons. Norms for knowledge and warranted belief do apply to individuals’ perspectives. They need not be applied or even applicable within those perspectives. The other stock argument for epistemic internalism is more influential. It maintains, plausibly, that in sceptical scenarios, an individual could be epistemically warranted, but not reliable. An individual in a matrix, having proximal stimuli that ⁶² A. Plantinga, Warrant: The Current Debate, op. cit. (Oxford: Oxford University Press, 1993) gives an example similar to BonJour’s clairvoyance example to show that a reliable process need not in itself yield warrant. Plantinga notes the example’s impotence against a view that sets as a necessary condition on knowledge a reliable well-functioning competence.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



yield the same types of perceptual beliefs that would be produced ordinarily, would have warranted, unreliable beliefs. So, the argument claims, reliability is not necessary for warrant.⁶³ This argument is solemnly intoned in article after article, without any attempt to think from opponents’ points of view.⁶⁴ No serious philosopher who thinks that reliability is constitutive to epistemic warrant could require reliability in all situations. It is obvious, apart from sceptical scenarios, that there are situations in which one’s warranted beliefs would be reliably mistaken. One could wander unawares into a natural situation in which illusions naturally and reliably occur. In such a situation, one would be reliably mistaken. One does not lose warrant simply by wandering into such situations. So one can be epistemically warranted in situations in which one’s beliefs are reliably mistaken. The reliability that is relevant to epistemic warrant is reliability in certain normal situations.⁶⁵ A contribution of ‘Perceptual Entitlement’, Section VI, was to specify what the normal situations are. A motive for doing so was to answer the claim that taking reliability to be a necessary condition on epistemic warrant is refuted by sceptical scenarios. The natures of all empirical beliefs are determined through interactions with the environment. The contents of perceptual states and empirical beliefs are fixed by direct or indirect causal relations to an environment. The natures of most empirical inductive inferences are also so determined. This point shows the way to determining conditions in which reliable veridicality is necessary for on entitlement to perceptual beliefs. The idea is that, for a perceptual belief to have an entitlement, the formation process that leads to the perceptual state, and from the perceptual state to the belief, must be reliably veridical in conditions in which the perceptual state, the belief forming process, and the perceptual belief were constitutively made to have the contents that they have.⁶⁶ ⁶³ As far as I know, the first arguments for epistemic internalism from sceptical scenarios occur in J. Pollock, ‘Reliability and Justified Belief ’, Canadian Journal of Philosophy 14 (1984), 103–114; and S. Cohen, ‘Justification and Truth’, Philosophical Studies 46 (1984), 279–95. ⁶⁴ For representative examples, see R. Foley, ‘What’s Wrong with Reliabilism?’, The Monist 68 (1985), 188–202; J. Pryor, ‘Highlights of Recent Epistemology’, British Journal of the Philosophy of Science 52 (2001), 95–124; R. Wedgewood, ‘Internalism Explained’, op. cit.; Matthias Steup, ‘Internalist Reliabilism’, op. cit.; M. Huemer, ‘Phenomenal Conservatism and the Internalist Intuition’, op. cit.; A. Moon, ‘Three Forms of Internalism and the New Evil Demon Problem’, Episteme, 9 (2012), 345–360. ⁶⁵ A. Goldman, Epistemology and Cognition, op. cit., 107, proposes that the process for forming an epistemically warranted belief must be reliable in normal situations. There and in ‘Strong and Weak Justification’, Philosophical Perspectives 2 (1988), section VI, he sets out what I think are unpromising conceptions of what count as normal situations. In the latter article, he himself criticizes the conceptions. He abandons specifying relevant reliability as being tagged to normal situations, claiming that envatted individuals’ beliefs are not warranted. I think that this view has no appeal. Specifying reliability by relation to the actual world, a move Goldman later made, is close to right. However, there are many situations in the natural world. An individual can be warranted but unreliable in some and not others. Individuals might stay longer in the unfavorable situations than in favorable situations. Such individuals’ warrants do not automatically lapse. An individual’s relation to the actual world can change in ways that affect warrant. Goldman’s later specification is inadequate because, although it is on the right track, it is insufficiently specific. ⁶⁶ For a fuller account see ‘Perceptual Entitlement’, op. cit., section VI. The following paragraphs in the text are derived from the argument pp. 531–537 of that article. Environmental conditions that sustain the perceptual content are understood to be included in normal conditions. Obviously, also, a perceptual

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Here is reasoning for this conclusion. (I) Reliable veridicality of a perceptual-belief type—and of the underlying perceptual-state type and of the systems for forming both—are constitutively necessary conditions on an instance of the perceptual belief ’s having an epistemic entitlement. These requirements derive from the fact that epistemic warrant necessitates that a warranted perceptual belief be conducive to knowledge and hence on a good route to truth. (II) Every prima facie entitlement for a perceptual belief attaches to the belief via its nature or type, relative to a mode of formation or sustenance by a wellfunctioning psychological competence that reliably connects to the subject matter of the belief. The relevant nature of the belief is its representational content type. The force of an entitlement to a perceptual belief does not depend on the belief ’s relation to any other belief. No other beliefs need support it. No background considerations need protect it.⁶⁷ (III) Thus the reliability constitutively required by an epistemic entitlement to a perceptual belief is a tendency of instances of the perceptual belief ’s representational content type to be true, given the formation or sustaining process. (IV) The epistemic entitlement, hence reliable veridicality, of a type of perceptual belief, relative to the underlying formation processes, must be nonaccidental. This non-accidentality must ground an explanation of why a perceptual belief of that type is conducive to knowledge and on a good route to truth. Types of perceptual belief and their formation processes are not constitutively reliable. Any type that is in fact reliable could have been unreliable. So the reliability that is predicable of a perceptual belief type must be non-accidental, but not metaphysically necessary. (V) What underlies this non-accidental, explanation-grounding relation between reliability and the type of perceptual belief—relative to its formation process—is the pattern of relations between the belief (and the process) and the environment that constitutively made the belief what it is. Those patterns are deeply regular, deeply systematic, and constitutively related to the perceptual belief ’s identity. Inasmuch as those patterns (contingently) yield a reliably veridical perceptual belief, they constitute a good, systematic connection between normal occurrences of the belief type and their being true. They constitute a deeply non-accidental basis for explaining wherein true instances of such belief types are conducive to knowledge, and prima facie certification of knowledge, in the senses explained in Section 2.1.

system could be formed with one set of contents; and through long-term association with another set of environmental conditions, it could take on other contents. ⁶⁷ If having a perceptual belief constitutively requires having some other beliefs, this point still stands. Having the other beliefs does not ground explanation of the entitlement to the belief, or contribute to the warranting force of the entitlement.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



(VI) Because of perspectival limitations on individuals who have perceptual beliefs—and the inevitable possibility of brute error—and because no perceptual belief is constitutively, by its nature, reliably veridical, no perceptual belief type can be reliably veridical under all conditions. Every type could be reliably non-veridical in various possible conditions that would be perceptually indiscernible to the believer. (VII) So warranted/entitled perceptual beliefs will be unreliable in various possible conditions other than those in which their natures—and the natures of their formation processes—were constitutively determined. (VIII) The only reliability relevant to the warrant/entitlement of perceptual beliefs is the reliability that obtains in conditions that ground explanation of their natures.⁶⁸ That is the only reliability that non-accidentally, but contingently, attaches to their types or natures in a way that grounds explanation of those beliefs’ being conducive to knowledge and on a good route to truth. The application of these ideas to sceptical scenarios is straightforward. Individuals who are moved unawares from normal environments to matrix-type situations remain entitled to their perceptual beliefs, even though they are reliably mistaken in the new environment. Being moved from one situation to another does not in itself affect epistemic warrant. Similarly, individuals born into matrix-type environments can be entitled to their perceptual beliefs. If their perceptual systems evolved in an ordinary environment, or if demon scientists copied stimuli from an ordinary environment, individuals’ perceptual states and perceptual beliefs got their representational contents through causal relations to an ordinary environment. If perceptual beliefs would be reliable in the ordinary environment, the envatted individuals are entitled to their perceptual beliefs. Individuals in vats that bear no relation to the environmental attributes that determined the contents of our empirical beliefs cannot share our perceptions or beliefs. In sum: although reliability is required for epistemic warrant, individuals lack knowledge in sceptical scenarios but can be warranted in locally unreliable perceptual beliefs. Sceptical scenarios should not distract one from the consideration that shows reliability to be constitutively necessary for empirical epistemic warrant. The connection between being warranted and being conducive to knowledge—cited in Section 2.1 and in this section—shows that epistemic warrant requires reliability. Any belief whose formation is relevantly unreliable would, far from being conducive to knowledge, prevent knowledge.⁶⁹ ⁶⁸ Reliability in other conditions besides the content-determining conditions may be relevant to determining other entitlements. For example, a newly discovered inductive method may have to be reliable in conditions relevant to those in which it is discovered, even though no new contentful states are established. As far as I can see, however, only conditions under which the nature of a perceptual belief is fixed or sustained are relevant to determining the kind of reliability that figures in its having prima facie epistemic entitlement. ⁶⁹ James Pryor ‘The Skeptic and the Dogmatist’, op. cit., 519 (see also 532), advocating “dogmatism”, writes: ‘The dogmatist about perceptual justification says that when it perceptually seems to you as if p is the case, you have a kind of justification for believing p that does not presuppose or rest on your justification for anything else, which could be cited in an argument (even an ampliative argument) for p.

[Note 69 continues below next page]

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     The decisive fact that makes Epistemic Internalism unacceptable is that it offers no appropriate constitutive relation between epistemic warrant of perceptual beliefs and knowledge. Relying on certain internal matters, like perception-formation or on phenomenology, is a good thing, because they are in fact reliable. Nothing constitutive, or otherwise metaphysically necessary, about them makes them conducive to knowledge. What makes epistemic warrant for perceptual beliefs conducive to knowledge and a good guide to truth is a deeply non-accidental but metaphysically contingent pattern of causal relations between operation of sensory systems and the environment. These relations gave perceptual states the content that they have. They established, in the formation or maintenance of most actual perceptual beliefs, a reliably veridical relation between competence in forming those beliefs and the subject matter of the beliefs. Internalism made sense in idealist and theological, guaranteed-harmony frameworks. For idealism, following good cognitive practice just is getting things right: the world just is the mind, operating according to rational or phenomenal rules. Idealism has shrunk to an archaic museum piece—a sign that philosophy can progress. Similarly, arguments for believing in a force that metaphysically guarantees reliable veridicality for perceptual states have been recognized as unacceptable. Sense-data accounts that took empirical knowledge to have an infallible base and an intrinsically reliable set of inference rules are also a dead issue. Empirical attitude contents are molded by interaction with a world that is largely independent of us. The molding is not geared to veridicality per se. It is geared to practicality—to fitness. No argument from the nature of perception guarantees its reliability, or hence epistemic warrant of perceptual beliefs. Most perceptual and perceptual-belief-forming systems are reliable in their central domains of operation. Being approximately veridical at basic levels of interaction with the world does tend to serve fitness. A central feature of epistemic warrant is that it is constitutively, by nature, conducive to knowledge and a good route to truth—and, in basic cases, a prima facie certification of knowledge. Reliability is part of epistemic warrant’s constitutive connection to knowledge and truth. Reliability is not guaranteed by anything internal. Internalism about epistemic warrant is, realistically, a lost cause. [Note 69 Continued:] To have this justification for believing p, you need only have an experience that represents p as being the case.’ Pryor adds that reliance on perceptual beliefs can be justified, even if they are not produced by a reliable belief-forming system. Entitlement is not a dogmatist type of warrant. ‘Perceptual Entitlement’ op. cit. develops a view that is similar to the dogmatist view in holding that an individual can be warranted in perceptual beliefs without presupposing, or resting the warrant on, anything else that could be cited as an argument, or protector, for the beliefs. However, it is not the case that, to be entitled to a perceptual belief, one need only have an experience that represents something as being the case. The experience (or perceptual state) must derive from well-functioning, reliable perceptual and perceptual-belief-forming competencies, whose reliability is grounded in the environmental conditions in which the nature and content of the competence were determined. Pryor allows, as sufficient for warrant, that it is ‘apt’ from the inside to believe something. (See the text following footnote 41.) Such “aptness” is some a kind of cognitive good. It is not epistemic warrant. A belief formed by an “apt”, unreliable system cannot be relevantly conducive to knowledge. By condition (c), Section 2.1, one is prevented from having knowledge if one’s belief is on an “apt”, unreliable route. Since epistemic warrant constitutively is conducive to knowledge and hence truth, no unreliable “apt” route could count as epistemically warranted.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



2.3 Entitlement and Moore’s Anti-sceptical Argument A guiding idea for understanding epistemic entitlement—epistemic warrant without reason—is that perceptual beliefs are epistemically warranted, on a good route to truth, conduce to knowledge, and commonly knowledgeable. Perceptual beliefs derive from psychological transitions from perceptions. The transitions are not reason-giving propositional inferences. Perceptions are not reasons, partly because they are not propositional. Perceptual beliefs are not reasons for themselves. So perceptual beliefs are not supported by reasons. They need not be, if believers are to be warranted in holding them. Entitlement to them does not depend on support from reasons. It does not depend on background beliefs. Perceptual beliefs need only be products of a belief-forming competence that reliably produces true beliefs, through connection to their subject matters, in their normal content-determining or content-sustaining environment. The process that forms those beliefs must function well when an instance of the belief type is formed. That is the main good route to truth for perceptual belief. This route suffices for prima facie epistemic warrant. It suffices to make true belief knowledge, in the absence of counter-warrants, internal incoherence, and Gettier conditions. To be epistemic warrant, entitlement need not borrow force from other considerations. Warrants for perceptual beliefs are not infallible or indefeasible. Perceptual beliefs can be shown wrong by warranted theory or by other perceptual beliefs. Perceptual beliefs are basic in that warrant for them is a starting point for empirical knowledge. Normally, they have weight. Perceptual beliefs are the source of warrant for all empirical beliefs, including empirical theory. This position accords perception its natural, central place in empirical knowledge.⁷⁰ The view that perceptual belief is defeasibly default-warranted, and an unaided source of knowledge, became a major position in mid-twentieth century from three sources. One was the natural idea that perceptual belief is the main basis for empirical warrant and knowledge. A second was recognition that it is a pipe-dream to think that an empirical foundational belief is infallible, or impervious to pressure from other empirical considerations. The structure of knowledge should acknowledge ordinary, fallible perceptual belief about the environment as a starting point. A third was loss of interest in tailoring epistemology to answering scepticism. Whereas Russell postulated an infallible empirical foundation to resist scepticism, ⁷⁰ This broad structural position is, of course, not new. Thomas Reid advances something like it: Essays on the Intellectual Powers (Edinburgh.: Edinburgh University Press, 2002), II 5, 96; II, 22, 251–252. Midtwentieth century versions are R. Chisholm, Theory of Knowledge (Englewood Cliffs, N.J.: Prentice Hall, 1st edition, 1966); A. Danto, Analytical Philosophy of Knowledge (Cambridge: Cambridge University Press, 1968); A. Quinton, The Nature of Things (London: Routledge & Kegan Paul, 1973), chapters 6–7; D. Armstrong, Belief, Truth, and Knowledge, op. cit.; J. Pollock, Knowledge and Justification (Princeton: Princeton University Press,1974); W. Alston, ‘Two Types of Foundationalism’, The Journal of Philosophy 73 (1976), 165–185; ‘What is Wrong with Immediate Justification?’, op. cit.; both reprinted in Alston’s Epistemic Justification, op. cit.; A. Goldman, ‘What is Justified Belief?’, in G. Pappas ed., Justification and Knowledge (Dordrecht: Reidel, 1979). I subscribe to this sort of view in ‘Individualism and Self-Knowledge’, The Journal of Philosophy 85 (1988), 649–663–see 655; reprinted in Cognition Through Understanding, op. cit., 60.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     by mid-twentieth century this strategy had come, to many, to seem not only hopeless but overwrought. I accept these ideas. They provide some motivation for the position. They do not rule out the view that perceptual beliefs need independent apriori guarantees in order to be warranted. One might think that a perceptual belief cannot be warranted unless one has a general apriori warrant to rely on perception.⁷¹ Or one might think that to protect against scepticism, one must hold that to be warranted one must have apriori warrant to believe that the scenarios are not in play.⁷² I think neither of these views correct. They are incompatible with our having entitlement to perceptual belief.⁷³ Entitlement derives from a reliable, well-functioning, perceptual-belief-forming system, grounded in content-determination that derives from causal interaction with the environment. No conditions on having specific background warrants need be fulfilled. Even if some individuals have such general background warrants, the warrants are not needed for one to be prima facie epistemically warranted in having perceptual beliefs. Once a believer can formulate sceptical hypotheses, the believer is warranted in rejecting them out of hand, unless they are shown to be reasonable, not just conceivable, threats to entitlements. Sceptical hypotheses are known to be incompatible with perceptual beliefs. The individual is entitled to the beliefs. Lacking warrant of their own, sceptical hypotheses have no force to counteract an individual’s entitlement. Lacking warrant of their own, they also cannot show an individual to be warranted in accepting perceptual beliefs, but unwarranted in rejecting sceptical hypotheses that the individual knows to be incompatible with those beliefs. At least two considerations back these points. The first is that very young children are epistemically warranted in their perceptual beliefs. They have knowledge through them. A young child knows that that body is round. Young children cannot think reasons for such beliefs. They cannot think even such relatively trivial reasons as I saw the F, because they cannot yet think reasons that make reference to their perception—much less to their belief-forming systems’ being ⁷¹ Examples of proponents are C. Wright, ‘(Anti)‐Sceptics Simple and Subtle: Moore and McDowell’, Philosophy and Phenomenological Research 65 (2002), 330–348; S. Cohen, ‘Basic Knowledge and the Problem of Easy Knowledge’, Philosophy and Phenomenological Research 65 (2002), 309–329; and ‘Why Basic Knowledge is Easy Knowledge’, Philosophy and Phenomenological Research 70 (2005), 417–430. ⁷² Proponents of the view that we have a general apriori warrant to reject sceptical scenarios are J. Vogel, ‘Cartesian Skepticism and Inference to the Best Explanation’, Journal of Philosophy 87 (1990), 658–666; L. BonJour, In Defense of Pure Reason (London: Cambridge University Press, 1998); S. Cohen, ‘Contextualism and Skepticism’, Philosophical Issues 10 (2000), 94–107; Christopher Peacocke, The Realm of Reason (Oxford: Oxford University Press, 2004); C. Wright, ‘The Perils of Dogmatism’, in S. Nuccetelli and G. Seay eds., Themes from G. E. Moore: New Essays in Epistemology and Ethics (Oxford: Oxford University Press, 2007). I find unpersuasive each of these claims about the content of an apriori reason for rejecting scepticism. We may have apriori reason to reject it, but such reasons are not easily found. Except for Peacocke, all these authors insist that we need an antecedent apriori reason, if we are to be warranted in perceptual beliefs and/or in rejecting scepticism. For an individual to be warranted (as opposed to just having warrant), the apriori warrant would, I think, have to be operative. Here I disagree with Wright. To be operative, a background warrant would have to be the content of an actual attitude. Such a requirement would leave much or all of humankind without epistemically warranted perceptual beliefs. ⁷³ They are also incompatible with how most of the authors listed in note 70 understand defeasible default warrant.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



reliable. Requiring even trivial reasons as a condition on perceptual knowledge would betray a lack of epistemic perspective, or a conflation of perceptual knowledge with some higher type of knowledge.⁷⁴ A second consideration is reflection on undefeated, mature human epistemic practice. Such practice does not require that individuals lean on, or even have, reasons for their perceptual beliefs as a condition for counting their perceptual beliefs epistemically warranted or knowledgeable. They need not lean on, or have, background beliefs that protect their perceptual beliefs. Even such simple reasons as I saw the F are glosses on an antecedent warrant. Such reasons are supplementary. They are not necessary for having knowledge. The perceptual beliefs are warranted and normally counted as knowledgeable, on their own. When human adults understand sceptical scenarios, they normally reject them as not worth taking seriously. Scepticism dramatizes the in-principle possibility of widespread error. One need not defend against other merely possible errors to be warranted in relying on perceptual beliefs. No reason has been given for thinking that one must defend against the wild possibility that one is in a sceptical scenario. Sophisticated scientists, as well as ordinary adults, commonly dismiss such scenarios. It is perhaps correct to criticize this reaction as uncurious and unphilosophical. It would be philosophical hubris to count it as not epistemically warranted. It would be hyper-intellectualized to count it warranted only because the individuals implicitly know an anti-sceptical argument, or implicitly rely on purportedly apriori warranted meta-beliefs like My perceptual beliefs are reliably formed. Sceptical scenarios have not been shown to be warranted threats. Defending against merely possible illusions has never been shown necessary for having perception-based warrant or knowledge. Ignoring illusions as threats, when the illusions are considered merely possible, is standard in science and ordinary thinking. Scepticism gives no reason to reject this practice.⁷⁵ Sceptical scenarios are dismissed because they are incompatible with beliefs licensed by excellent sources of empirical warrant and knowledge: perception, perceptual belief and reasoning from perceptual belief. Claiming that the scenarios must be blocked if one is to have ordinary perception-based warrant and knowledge is wildly out of step with our best bases for forming beliefs. It is out of step without the slightest positive reason to be so. It is a prima facie datum that sceptical scenarios are unwarranted. Showing why they are unwarranted is of real philosophical interest. Emphatically, I do not dismiss philosophical interest in scepticism. One cannot show scepticism wrong merely by relying on one’s perceptual beliefs. It is of philosophical interest to show scepticism wrong and to explain its error. Still, no theory should count the unphilosophical scientist’s disbelief in such scenarios unwarranted, even if the disbelief rests purely on the scenarios’ incompatibility with perceptual beliefs. ⁷⁴ This point is the primary emphasis of ‘Perceptual Entitlement’, op. cit., sections I-II. ⁷⁵ Many sceptical challenges implicitly depend on a sensitivity principle, to the effect that (necessarily) if A is warranted in believing that p (or knows that p), then A would not believe p, if p were false. I think that the common sense points adduced above entail rejection of any such principle. For good extensive critical discussion of the principle, see T. Williamson, Knowledge and Its Limits, op. cit., chapter 8.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     More generally, no one has shown that, for one’s perceptual beliefs to be warranted or knowledgeable, one must, in the absence of a prima facie warranted threat, have a reason that defends against the possibility of an illusion—sceptical or ordinary. Entitlements to perceptual beliefs require no help from general beliefs. Other things equal, one can reason from warranted perceptual beliefs to rejection of propositions known to be incompatible with them. It is a mistake to think that this response begs the question against the sceptic. The mistake lies in thinking that one needs to argue against the sceptic. A philosopher may want to give such an argument. Then it would not be enough just to appeal to one’s entitlements. But I am not arguing against scepticism. I am not explaining the irrationality of scepticism or trying to defeat it. Those are projects for other occasions. I claim that scepticism need not be argued against, for one to be warranted in rejecting it. One can stand by one’s warranted perceptual beliefs and warranted inferences, unless the sceptic gives a reason that undermines the warrant. Presentation of the mere possibility of undetectable sceptical scenarios is not such a reason. Philosophical approaches that flout these points suffer, I think, from serious hyper-intellectualization. Tailoring an account of epistemic warrant to fit conditions of philosophical understanding is the wrong way around. Philosophy is concerned with understanding. We want to understand warrant and knowledge—place them in a wider scheme of things. In the breadth of its ambition, this ideal goes beyond what even science aims at. It is a good ideal. It is not, however, a constraint on knowledge or warrant. Ordinary, non-philosophical individuals—both scientists and the unsophisticated—are entitled to their perceptual beliefs. They are entitled to use them to dismiss sceptical scenarios. They are entitled to ignore, reject, or dismiss putative threats from in-principle possible illusions, unless they have reason to do otherwise. Their having these entitlements does not require having the slightest understanding of why they have them. Having the entitlements does not require that one understand why sceptical scenarios, or indeed the possibilities of ordinary illusions, are not reasonable threats.⁷⁶ The tendency toward hyper-intellectualization derives partly from the ritual of initiating novices into philosophy by exposing them to scepticism. Such rituals are valuable in encouraging reflection. They should not be allowed to distort accounts of empirical warrant, by granting the mere possibilities raised by scepticism the status of threats to perceptual beliefs. ⁷⁶ Barry Stroud, ‘Understanding Human Knowledge in General’ in M. Clay and K. Lehrer eds., Knowledge and Skepticism (Boulder: Westview, 1989), challenges certain traditional attempts to understand knowledge in the light of scepticism. Specifically, he thinks that trying to use only materials from outside a domain questioned by a sceptic will inevitably fail. He thinks that this failure is endemic to epistemology and will inevitably leave epistemologists in an unsatisfactory position. Stroud’s challenge is a good one. Still, the idea that epistemology must follow the route that he challenges, and the idea that not following it must leave one in an unsatisfactory intellectual position, seem to me unpersuasive. Philosophy is flexible about methods that it can employ to gain understanding. For further discussion of these issues, see E. Sosa, ‘Philosophical Skepticism and Externalist Epistemology’, Proceedings of the Aristotelian Society, Supplementary Volumes 68 (1994), 263–290; B. Stroud, ‘Scepticism, “Externalism”, and the Goal of Epistemology’, Proceedings of the Aristotelian Society, Supplementary Volumes 68 (1994), 291–307; E. Sosa, Reflective Knowledge volume II (Oxford: Clarendon Press, 2009), chapters 8 and 9. Incidentally, the doctrines that Stroud and Sosa call ‘externalism’ are not what I count as externalism.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Some philosophers have held that we lack warrant to believe (nor do we know) that we are not in sceptical scenarios.⁷⁷ I think the claim purest philosophical affectation. This preposterous view derives from letting philosophical rituals distort philosophical judgment. An idea with similar pedigree is that to be warranted in rejecting sceptical scenarios, one must have a reason to block them that is independent of one’s perceptual beliefs. Granting philosophy a protected space, or believing that sceptical questions have a special status that innoculates them from standard dismissive responses, is an occupational disease in philosophy.⁷⁸ No one has given reason to believe that sceptical challenges are special in being immune to the demand for positive rational support—not simply presentation of a possibility—if they are to undermine warrant for perceptual beliefs. It is absolutely crucial in epistemology to distinguish what is needed to understand knowledge and epistemic warrant from what is needed to have them. The intuitions just criticized commonly underlie the idea that Moore’s best-known anti-sceptical argument is no good. Here is Moore’s argument adapted to our issues: (a) That [pointing] is a spherical body [warranted perceptual belief ]. (b) That’s being a spherical body is incompatible with my being in a matrix situation with no spherical bodies that could be perceived or otherwise picked out in thought [based on logic and apriori semantics]. So (c) I am not in a matrix situation with no spherical bodies . . . [by deductive inference from the premises].⁷⁹ ⁷⁷ R. Nozick, Philosophical Explanations (Cambridge, Mass.: Harvard University Press, 1981). ⁷⁸ Contextualism about knowledge-claims is a recent instance. See K. DeRose, ‘Solving the Skeptical Problem’, The Philosophical Review 104 (1995), 1–52; D. Lewis, ‘Elusive Knowledge’, Australasian Journal of Philosophy 74 (1996), 549–567. This approach has its ancestry in ideas about relations between ordinary and philosophical questions proposed by Kant and Carnap. Clarke and Stroud pursue a similar approach, without contextualist apparatus. They distinguish ordinary “internal” from “external” questions. They take scepticism to raise a doubt to which ordinary or scientific counter-claims are irrelevant. See T. Clarke, ‘The Legacy of Skepticism’, The Journal of Philosophy 69 (1972), 754–769; B. Stroud, The Significance of Philosophical Scepticism (Oxford: Clarendon Press, 1984), esp. chapters 2 and 3. The analogies that they use to show that the sceptic’s doubt survives standard dismissals all involve actual defeaters of purported knowledge. For example, Clarke’s airplane spotters’ purported knowledge is trumped by their superiors’ actual knowledge that there are unusual airplanes that the spotters cannot distinguish from the more usual planes that they do spot. Sceptical doubts are disanalogous on just this point. Clarke fails to notice the disanalogy. Sceptics have no warranted or knowledgeable counterclaims. I believe that an epistemically relevant distinction between ordinary and external (scepticism-relevant) questions or contexts has never been well supported. ⁷⁹ Here and in some of what follows, I omit de-re-functioning applications on secondary attributives–as in the case of spherical in this argument. For the original argument, see G.E. Moore, ‘Proof of an External World’, Proceedings of the British Academy 25 (1939), 273–300. I express sympathy for Moore’s premise and his use of it to deny scepticism in ‘Individualism and Self-Knowledge’, op. cit., 59–60; in ‘Reply to Stroud’, in M. Hahn and B. Ramberg, Reflections and Replies: Essays on the Philosophy of Tyler Burge (Cambridge, Mass.: MIT Press, 2003); and in ‘Reply to Martin Davies’, in M. Frapolli and E. Romero eds., Meaning, Basic Self-Knowledge, and Mind: Essays on Tyler Burge (Stanford: CSLI Publications, 2003). Others have defended Moore’s argument in some detail, notably J. Pryor, ‘What is Wrong with Moore’s Argument?’ op. cit.. Pryor’s defense of Moore’s argument is skillful and insightful. Pryor rightly takes prima facie warrant for perceptual belief not to depend on further beliefs, or on background propositional information available in the psychology of the believer. On this point, his view is congenial to mine, and to the views cited in note 70. On the other hand, as remarked in Section 2.2 and notes 34, 41, 43, and 64 his

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Moore’s argument tracks adult human, including scientific, epistemic practice. The perceptual belief in the first premise, (a), has undefeated prima facie epistemic entitlement. Premise (b) is obviously true. The premises support the conclusion. I discuss two responses to Moore’s argument. Both derive from excessive concern with finding an argument that explains why scepticism is wrong.⁸⁰ One response claims that accepting Moore’s argument makes knowledge “too easy”. The other claims that the argument is circular.⁸¹ The warranted belief, and knowledge, obtained from the argument is “too easy” only relative to an expectation that a warrant must yield understanding of why scepticism is mistaken. Moore’s argument does not explain why the sceptical scenario can be dismissed. It ignores or dismisses it. Normal sophisticated epistemic practice, indeed sanity, urges against accepting sceptical scenarios as actual defeaters. The argument sides with sanity without explaining why the scenarios are not justified threats to the entitlement, or why scepticism is irrational. Warranted dismissal of sceptical scenarios is compatible with fascination about why they should be dismissed. Such fascination can motivate good philosophical inquiry. But warranted dismissal is also compatible with not caring why sceptical scenarios should be dismissed. Similar points apply to doubters of ordinary perceptual beliefs who produce no ground for doubt. Consider ‘Is it really red?’, when one is looking at a surface in a good light and no special condition triggers the question. The idea that one needs to take such doubts seriously to be warranted in rejecting them is unsupported and out of keeping with standard conceptions of warrant and rationality. Such doubts are a natural focus for philosophical explanation or for teaching the young. For warrant and knowledge, such groundless doubts can be ignored. As I have indicated, young children have warranted perceptual beliefs and knowledge without knowing what an illusion is. It seems to me wildly mistaken to believe that a young child, who cannot even think sceptical scenarios, does not know or is not epistemically entitled to the belief that that is red or that is a body. Generalized denial that epistemic warrant (or “justification”) requires reliability and his claim that epistemic warrant for perceptual belief derives from phenomenality, or from the mere fact that one has a perceptual experience that represents what the perceptual belief represents, are, I think, serious mistakes. ⁸⁰ Arguments that warrant is not passed from perceptual belief to the conclusion of Moorean arguments because of failure of a relevant closure principle all seem to me to be defective. For examples of anti-closure arguments, see R. Nozick, Philosophical Explanations, op. cit.; C. Wright, ‘(Anti)‐Sceptics Simple and Subtle: Moore and McDowell’, op. cit.; M. Davies, ‘Externalism and Armchair Knowledge’, in P. Boghossian and C. Peacocke eds., New Essays on the A Priori (Oxford: Oxford University Press, 2000); N. Silins, ‘Basic Justification and the Moorean Response to the Skeptic’, in T. Gendler and J. Hawthorne eds. Oxford Studies in Epistemology 2 (2008), 108–142. I will not canvass the large literature on this matter. I criticize some arguments for rejecting one such closure principle in ‘Reply to Martin Davies’, op. cit.. Most such arguments assume that Moorean anti-sceptical arguments are circular or beg the question. Silins’ argument does not make these mistakes. His Bayesian-based argument is, however, undermined by the discussion in Section 2.5. ⁸¹ The “too easy” intuition is developed in S. Cohen, ‘Basic Knowledge and the Problem of Easy Knowledge’, op. cit. and ‘Why Basic Knowledge is Easy Knowledge’, op. cit.. The circularity claim is developed by C. Wright, ‘(Anti)-Sceptics Simple and Subtle: Moore and McDowell’, op. cit.; ‘Warrant for Nothing (and Foundations for Free)?’, Aristotelian Society, Supplementary Volume 78 (2004), 167–2012; and M. Davies, ‘Externalism and Armchair Knowledge’, op. cit.. Davies subsequently changed his mind.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



apriori knowable propositions that really would give reasons for rejecting sceptical scenarios are hard to come by. Entitlement to perceptual beliefs does not depend on anything so grand. It normally survives groundless doubts of all kinds.⁸² A philosopher’s gambit is to claim that although children may have warranted perceptual beliefs, philosophically initiated adults lose warrant unless they can block sceptical scenarios, or even ordinary illusions. I find such a view rationally threadbare and out of step with good epistemic practice. The mere in-principle possibility of a sceptical scenario, or an ordinary illusion, does not begin in young children as threat to an entitlement that must be addressed if the entitlement is to be retained. No reason has been given to think that it ever becomes one. Just conceiving a possible way in which one can be wrong does not undermine a warrant in ordinary cases. No reason has been given for thinking that conceiving sceptical scenarios creates an underminer of entitlements to perceptual beliefs. Entitlements to perceptual beliefs can yield so-called easy knowledge to the effect that one is not being fooled. If an individual has concepts of illusion and of looking red, he or she can reason from an entitled, knowledgeable, though defeasible, perceptual belief as of a red surface to a warranted conclusion that the surface is not blue but lit so as to look red.⁸³ The Moorean argument that I laid out follows such a procedure. The reasoning would not answer a doubter. In the absence of warranted doubt, the reasoning is, however, correct and justified. It can constitute knowledge. The bases for the warrant and for the knowledge lie already in the perceptual system’s relation to its subject matters. I return to this matter in Section 2.5. Claims that Moore’s argument is circular rests on assumptions that I have already criticized. One such assumption is that one’s perceptual beliefs are warranted only if one has an antecedent warrant to believe that one is not under an illusion. Another is that warrant requires antecedently knowledge that perceptual beliefs are reliable. Such views hyper-intellectualize warrant and derive, I think, from living within a bubble of philosophical dialectic with scepticism. Moore’s argument begs the question only if it is taken to answer doubts. In fact, the argument simply exhibits a route to knowledge. Perhaps Moore took it to answer doubts. I think that he aimed just to embarrass those who think that skepticism presents a serious threat to knowledge. Moore’s argument does not beg the question against sceptical doubt. It ignores such doubt. It can legitimately do so, because the doubt is not a warranted, or even supported, threat. Sceptical doubt is epistemically interesting, but epistemically insubstantial. A basic mistake in epistemology is to tailor an account of epistemic warrant to fit a response to scepticism. ⁸² J. Pryor in ‘What is Wrong with Moore’s Argument?’, op. cit. takes merely having an irrational belief that is incompatible with a perceptual belief, to undermine warrant for the perceptual belief. I see no reason to think that the irrational belief affects entitlement to the perceptual belief in any way. The irrational belief does make the individual’s views incoherent. The individual cannot be warranted in both the perceptual belief and the irrational belief, and perhaps cannot have knowledge through the perceptual belief. However, the irrational belief does not undermine the relevant epistemic goodness of holding the perceptual belief. Since the irrational belief presents no warranted defeater, lack of warrant lies purely in it and in the incoherence of harboring both beliefs. ⁸³ S. Cohen, ‘Basic Knowledge and the Problem of Easy Knowledge’, op. cit. claims that one cannot know that one is not having an illusion in that way.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     Moore’s argument uses premises that are vastly more powerful than any premises that motivate scepticism. Moore makes this point in a separate meta-argument.⁸⁴ The meta-argument should remind philosophers that scepticism is no threat to knowledge. For well-functioning, reliable systems of perceptual-belief formation, entitlement comes first. Defense of the entitlement must be mounted only if there is a real threat of danger—a threat with warranting force. At the heart of the structure of empirical epistemic warrant is entitlement—warrant without reason. Ordinary and scientific empirical reasoning build on that basis.⁸⁵

2.4. Entitlement and Bootstrapping I turn to the first of two arguments that purport to undermine the idea that we have entitlement to perceptual beliefs—prima facie epistemic warrant without reason. This argument is known as a track-record or bootstrapping argument. The argument proceeds in two stages. First, it displays an unattractive form of reasoning. Then it claims that any view that perceptual belief is prima facie warranted in a way that does not depend on reasoned support, or on warranted antecedent beliefs, is committed to the form of reasoning. Here is the argument. Suppose that one reflects on one’s perceptual state and judges, (1’)

This₁ body appears redthat2.

Suppose that one also forms the perceptual belief, (2’)

This₁ body is redthat2.

⁸⁴ G. E. Moore, Some Main Problems of Philosophy (London: George Allen & Unwin Ltd, 1953), 119ff., 141ff.. ⁸⁵ Ernest Sosa’s notion of animal knowledge bears a potentially misleading, superficial similarity to knowledge by entitlement. See his ‘How to Resolve the Pyrrhonian Problematic: A Lesson from Descartes’, Philosophical Studies 85 (1997), 229–249; A Virtue Epistemology, op. cit., 104–105 and Reflective Knowledge, op. cit., chapters 7–8. Roughly, for Sosa, animal knowledge is knowledge that does not require metareflection on belief, knowledge, reliability of resources, or the like. Animal knowledge includes first-order knowledge based on reasoning, hence justified knowledge, as well as unreasoned knowledgeable perceptual belief. It thus includes knowledge by entitlement, but is not equivalent with it. Sosa places metarequirements on first-order knowledge (what is defined as animal knowledge) in humans, if it is to count as knowledge. These requirements entail that no humans have animal knowledge. He writes ‘Human knowledge always requires a degree of meta-aptness of some minimal level. (Brute animal cognition requires a substantially lower level.)’, Knowing Full Well (Princeton: Princeton University Press, 2011), 92–93. Meta-aptness ‘requires . . . that the believer aptly believe, at least implicitly, in the aptness of his first-order belief ’ (94). Aptness, for Sosa, is accuracy on account of a reliable well-functioning competence. Contrary to these claims, human children have entitlement-based perceptual knowledge before they have relevant meta-competencies. They have such knowledge before they have concepts of belief, reliability, competence, well-functioning, or accuracy on account of a competence. The term ‘implicit belief ’ in Sosa’s requirement of meta-aptness cannot cover cases in which an individual lacks the concepts to have the belief. The requirement is not only mistaken about human children. It hyperintellectualizes human adult knowledge. Even unsophisticated human adults might not understand the relevant concepts well enough to be disposed toward believing in meta-aptness. Even if they do, their entitlement-based knowledge through perceptual belief is, constitutively, independent of meta-cognitive dispositions. For criticism of similar statements of Sosa’s, see J. Greco, ‘Virtue, Luck and the Pyrrhonian Problematic’, Philosophical Studies, 130 (2006), 9–34.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



So (3’) This₁ body appears redthat2 and is redthat2. An obvious consequence of (3’), together with the assumption that one believes (2’) and (3’), is (4’) My color vision provided a true perceptual belief—the belief, this₁ body is redthat2. One can reason similarly regarding other bodies and other colors, each time concluding that one’s color vision yields a true belief. One can then reflect on the reasoning and conclude, (5’) My perception-/perceptual-belief forming system yielded true perceptual beliefs each time: no errors were committed. One then infers inductively, (6’)

My perception-perceptual-belief forming system is reliable.

Suppose that the following view, (E), were correct. (E) We are prima facie epistemically warranted—entitled—to such beliefs as (2’). The warrant need not be supplemented. We are entitled to (2’), independently of any antecedent warrant that we may have for believing in our reliability. It is enough, for our being prima entitled to such beliefs, that the processes that formed them function well on the particular occasion, and that the belief types and their formation types are reliable, where the reliability is grounded in the competencies’ relation to the subject matter. (7’) (E) is committed to (1’)–(6’). (8’) Since the reasoning in (1’)–(6’) is unacceptable, (E) is mistaken.⁸⁶ There are many moving parts in (1’)–(8’). Let us go over some of them. The reasoning in (1’)–(6’) does not match (E)’s account of default warrant for perceptual belief. (E) says nothing about an entitlement’s making reference to beliefs or appearances ((1’)–(3’)), or to one’s psychological system or its reliability ((4’)–(6’))—except to say that beliefs about these matters is no constitutive part of individuals’ being entitled to their perceptual beliefs. Young children with entitlements to their perceptual beliefs are probably not even capable of meta-beliefs about

⁸⁶ This sort of argument traces to R. Fumerton, Metaepistemology and Skepticism, op. cit., 173–180. Variants are given by J. Vogel, ‘Reliablism Leveled’, The Journal of Philosophy 97(2000), 602–623; Stewart Cohen, ‘Basic Knowledge and the Problem of Easy Knowledge’, op. cit.; R. White, ‘Problems for Dogmatism’, Philosophical Studies 131 (2006), 525–557; and S. Cohen, ‘Bootstrapping, Defeasible Reasoning, and A Priori Justification’, Philosophical Perspectives, 24 (2010), 141–159. White (553n1) takes ‘Perceptual Entitlement’ to be undermined by this argument and by the one discussed in Section 2.5. White mainly targets the view, ‘I am justified in believing that a card is red if it appears red, even if I am not justified in believing that my color-vision is reliable. All that is required is that I have no reason to suspect that it isn’t reliable’. This view is roughly James Pryor’s. It is not mine. For me, more is required: reliability, grounding of reliable operations in content-determiners and truth-makers, and well-functioning of the operations. I will, however, consider the bootstrapping argument in relation to my view, (E).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 :     their perceptions, beliefs, or psychological systems. They are not committed to (1’)–(6’) because they cannot think it. Some proponents of (1’)–(8’) concede that (1’)–(3’) does not represent how (E), and other default-warrant views, regard formation of perceptual beliefs. According to (E), there is a transition from a perception as of the body’s being red to a perceptual belief that the body is red. The transition is not a propositional inference. Much less does it involve a belief about one’s perceptual state—about how things appear—as (1’) and (3’) do. The more sophisticated proponents of (1’)–(8’) hold that we adults can know (1’) by reflection on our perceptual experiences. Once we know it, we can make such correlations as (3’). These proponents state that once one has concepts like psychological system, reliability, and well-functioning, one can reason (1’)–(5’). I grant these points. Proponents of (1’)–(8’) want to place the blame for the odd argument (1’)–(6’) on the view that (2’) is default prima facie warranted. I think that the default prima facie warrant for (2’) is correctly described in (E). Proponents of (1’)–(8’) claim that if (2’) is thus warranted, the rest of the argument to (6’) is inductively sound and inductively warranted.⁸⁷ But the procedure is not a good induction to a conclusion about the reliability of one’s perceptual belief. The problem does not reside in any view about how (2’) is warranted. The form of (1’)–(6’) is parallel to the following. Suppose that I note (1’’) I believe that Istanbul is more populous than Vienna. (2’’) Istanbul is more populous than Vienna. So (3’’) I believe that Istanbul is more populous than Vienna, and Istanbul is (in fact) more populous than Vienna. So (4’’) My belief system has supplied me with a true belief. The procedure can be repeated for very many of my beliefs, yielding (5’’): My belief system yielded true beliefs each time: no errors were committed. So (6’’): My belief-forming system is reliable. (1’’)–(6’’) is no better as inductive reasoning than (1’)–(6’). Its failure to provide good inductive reason for the conclusion does not depend on how (2’’) is warranted. I suppose that the warrant is not by entitlement. One believes (2’’) from taking a census, reading travel books, asking inhabitants for population numbers, and so on. Listing beliefs that one has come to while maintaining those beliefs does nothing to support one’s reliability inductively. So applying (E) to (2’) is not responsible for the unattractiveness of (1’)–(6’). Normal inductive evidence or argument for one’s general reliability must derive from checking one’s perceptual beliefs against independent indicators of their truth, from various standpoints. In effect, (1’)–(3’) merely avoids a variant on Moore’s paradox.⁸⁸ In (1’)–(3’), one aligns one’s belief about a perception, or perceptual appearance, with a belief that ⁸⁷ Cohen, ‘Basic Knowledge and the Problem of Easy Knowledge’, op. cit. believes that the induction must be accepted by anyone one who thinks that one can have warranted perceptual belief independently of having warrant that one is reliable. White, ‘Problems for Dogmatism’, op. cit., 546, endorses the inductive inference, given that (2’) is taken to be immediately warranted. White asks rhetorically, ‘what else could account for my long run of successful color judgments if not the reliability of my color-vision?’. Cohen, ‘Bootstrapping, Defeasible Reasoning, and A Priori Justification’, op. cit., 143, calls the move from (5’) to (6’) ‘straightforward induction’ and endorses White’s endorsement, 144. ⁸⁸ G. E. Moore, ‘Moore’s Paradox’, in T. Baldwin ed. G. E. Moore: Selected Writings (New York: Routledge, 1993).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



directly derives from the perception. One is warranted in the matching as long as one is entitled to (2’)—and as long as no counter-warrants upset one’s entitlement to (2’). Both perception and belief are committal: They have a representational function whose fulfillment hinges on whether their representational contents are veridical. (See note 5.) The matching procedure consists in simultaneous acknowledgment of the commitments of the perception and the perceptual belief. Contrary to proponents of (1’)–(8’), the series of matches is best explained not by the conclusion that one’s perceptual system or belief system is reliable. It is best explained by the observation that one has aligned one’s meta-judgments about one’s states (whether these be perceptual states or belief states) with those committal perceptual or perceptual-belief states themselves.⁸⁹ Each line in (1’)–(5’) is warranted, assuming that (2’) is warranted.⁹⁰ Yet (1’)–(5’) does not inductively support (6’). The serial matching provides no evidence for (6’). (5’) is guaranteed in advance by the Moore-like procedure. The premises are not independently generated. As an induction from the series of matches, the argument is empty.⁹¹ There is, I think, good reasoning that justifies belief in the reliability of specific beliefs. Suppose that (2’), now re-labeled, is warranted. Here is the reasoning:

⁸⁹ There have been other diagnoses of an inductive failure in (1’)–(6’). One diagnosis is that the argument rests on an inductive test that cannot be disconfirmed. This idea is discussed by White, op. cit., 544ff., who credits Pryor. White endorses the claim that good inductions must risk disconfirmation, but claims that the view about warrant that he targets (for present purposes, (E)) is committed to there being an induction in (1’)–(6’) anyway (546–548). I think that his argument for this latter claim is unpersuasive. Incidentally, there are counterexamples to the idea that good inductions must risk disconfirmation. See B. Weatherson, ‘Easy Knowledge and Other Epistemic Virtues’, online http://brian. weatherson.org/ekoev.pdf, sourced 2014. In any case, one would have to qualify (1’)–(6’) to make it strictly the case that the purported induction is not disconfirmable. Actual perceptual systems do not engender perceptual beliefs if perceptual appearances contain internal signs of fishiness. For example, a perception in which there is an appearance as of a given color may be unstable, or may yield auxiliary cues that lighting conditions are too tricky to rely upon. As a result, the probability assigned by the perceptual system to that perception type’s being accurate is low. Perhaps no belief is formed on the basis of the appearance. Then an instance of the move from (1’) to (2’) would fail. Thus if the belief about color appearances were not paired with a belief about the corresponding color’s actually occurring–and not because of background propositional defeaters, but because of internal perceptual cues–then the “test” would tend to be disconfirmed. Since some of these internal cues can be cues to malfunction, the view that perceptual states are by their nature supportive of belief is shown here, again, to be mistaken. Absence of internal disconfirmatory perceptual cues over a series of runs may provide, to a reflective reasoner, some slight supplementary evidence of reliability, assuming that the reasoner’s perceptual beliefs are mostly reliable and warranted. I ignore this point in the text of this section. Absence of disconfirmatory cues cannot provide evidence by itself. Normal evidential support for reliability requires independent checks on each perceptual belief. ⁹⁰ This claim needs qualification. Given that we know that individuals commonly neglect defeaters, it may be reasonable to suspend belief about (4’) if the list of cases is long enough. One can perhaps avoid this point by taking easy cases, and by not allowing the list to be long. ⁹¹ I think that the argument does not even meet standards for enumerative induction. The premises are not appropriately independent of one another. All pairings instantiate a single type of rational commitment regarding the relation between one’s meta—and first-level commitments.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

(1’’’) This₁ body is redthat2. One can reflect and realize with warrant that one believes (1’’’). So (2’’’)

I believe this₁ body is redthat2.

I think that this meta-belief is empirically warranted, since one’s awareness of the occurrence of one’s own sensations and perceptions is empirically warranted. One judges correctly and with warrant, (3’’’)

My perceptual belief this₁ body is redthat2 is epistemically warranted.

Warrant for (3’’’) differs from warrants for (1’’’) or (2’’’). I offer no account of the warrant for (3’’’). I believe, however, that if one has no warrant to believe that there are defeaters for (1’’’) or for (3’’’), if one has the notion of warrant for perceptual belief, and if one is warranted in the first-level (here perceptual) belief (1’’’), one is normally warranted in believing that one’s perceptual belief is warranted. At any rate, I think that, when we have a warranted perceptual belief, we are normally warranted in believing that it is warranted. To hold the perceptual beliefs, with warrant, and doubt that one is warranted in believing them, would normally involve some of the irrationality elicited by Moore’s paradox. By the apriori warranted principle: (4’’’) If a perceptual belief is warranted, the way in which it is formed or sustained is reliable and well-functioning, I conclude, with justification, (5’’’)

My belief this₁ body is redthat2 is formed in a reliable, well-functioning way.

In this argument, one does not learn anything about the track record of one’s perceptual-belief-forming system from (1’’’)–(5’’’).⁹² One learns only what is implicit in one’s warranted acceptance of the perceptual belief. One’s reasoning is not inductive. It depends on one’s actual entitlements and a general apriori principle. In each case, one simply matches one’s warranted perceptual belief with a warranted belief about one’s being warranted in that belief. To get an inductive reason to believe in one’s reliability, one must use other considerations. One must use such considerations as whether one is diseased, whether there are defeaters, cross-checks with other perceptual beliefs, reasoning about one’s circumstances, independent indicators of the truth of one’s perceptual beliefs, studies of the workings of the perceptual and perceptual-belief formation systems, and so on. The argument (1’’’)–(5’’’) is a meta-epistemological analog of Moore’s antisceptical argument. The argument is sound. It justifies believing its conclusion in a specific, instantiated case. But the justification rests mainly on one’s entitlement to (1’’’). As with Moore’ anti-sceptical argument, the argument is not circular. The warrant for none of its steps requires being warranted in believing that one’s beliefforming mechanism is reliable. The argument can display structure in one’s warrants.

⁹² With the qualification in the last paragraph of note 89.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



It does not explain, or provide evidence for, one’s reliability, even in the given case. It does not answer doubts about one’s reliability. The argument form does not work for every possible individual arguer. If an individual has an unreliable or badly operating perceptual system, the individual is not warranted in accepting (1’’’) or (3’’’), and thus not in accepting (5’’’). One may not be able to discern one’s lack of warrant from the inside. One could innocently carry out unsound reasoning. The argument is not apriori, because not all of its premises are apriori warranted.⁹³ The argument is not independent for its warranting force from sense experience. The force of the warrant for (5’’’) depends on empirical entitlement to (1’’’). As emphasized in the previous paragraph, the argument is not shown to be sound purely through understanding it. (6’) and (7’) are further errors in the bootstrapping argument. Against (7’), accepting (E) does not require accepting to (1’)–(6’) as an inductive argument. (1’)–(6’) is a bad inductive argument regardless of how (2’) is warranted. Against (6’), one cannot infer inductively from instances of the Moore-like procedure that one’s epistemic equipment is generally reliable.

2.5 Entitlement and Confirmation In this section, I criticize a more interesting argument against the idea of epistemic warrant without reason. The argument aims to show that such warrant for perceptual beliefs is incompatible with ordinary intuitions about confirmation and with Bayesian ⁹³ S. Cohen, ‘Bootstrapping, Defeasible Reasoning, and A Priori Justification’, op. cit., 150–155, argues that the notion of undefeated empirical default warrant for perceptual beliefs is ‘incoherent’. He starts by arguing that perceptual states “provide one” with reasons for perceptual beliefs–reasons like That₁ body looks red, and I see that that₁ body is red. I think this point correct as long as two further points are kept in mind. First, such “provision” occurs only if one has meta-psychological concepts like looks and see– concepts that not all perceptual believers have. Second, the force of an entitlement to a perceptual belief does not depend on such reasons–even in adults who have such reasons. I argue both points in ‘Perceptual Entitlement’, op. cit., sections I–II. Next, Cohen postulates defeasible suppositional inference rules of the form a looks C; so a is C. He takes reasoning by such rules to be apriori. He infers that such reasoning (from a reason like That₁ body looks red to a perceptual belief like That₁ body is red) shows that the idea of undefeated empirical default warrant for perceptual beliefs is incoherent. The argument fails with respect to entitlement in two ways. One is that since entitlements to perceptual beliefs do not depend on reasons, no point about reasoning from propositions about how things look bears on the nature of such entitlements. The other is that it is mistaken to take the inference rule, and reasoning according to it, to be apriori. (See note 110.) By Cohen’s own lights, the position that he is arguing against is incoherent only if the suppositional inference rules are apriori applicable. But where such rules provide warrant, the force of the warrant depends on reliable connection between having perceptions with certain looks and having true beliefs. This connection, expressed by the schema’s ‘so’, is empirical. For it depends on the contributions to warrant of sensory connection to the environment, of perception, and of perceptual-belief formation. Although the perceptual entitlements do not depend on the relevant reasons, the relevant reasons depend on underlying perceptual entitlements. Entitlement is here, as often elsewhere, more basic than reason. Cohen takes his argument to tell against ‘evidentialist’ theories—theories that claim that empirical knowledge has its basis in evidence. I suppose that beliefs of the form a looks C can be taken to cite evidence. Then the first mistake cited in the previous paragraph would not apply. (E) is not an evidentialist account. However, most of the theories that Cohen does discuss do not construe these “evidential” reasons as basic. Cohen’s argument is defective in any case, because of its assumption that the relevant inference rules are apriori.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

approaches to confirmation. These issues are rather technical. However, I think that following at least the gist of the discussion should open up some important nontechnical, intuitive points about how entitlement applies within actual psychologies. Suppose that one is entitled to one’s perceptual belief: (M)

that₁ body part is a handthat2.⁹⁴

Suppose that via this warranted perceptual belief, and the obvious truth, (A)

that₁ body part is a handthat2 only if it is not the case that I am in a sceptical scenario with a mistaken belief as of a body part and hand,

One infers: (not-V)

it is not the case that (V) I am in a sceptical scenario with a mistaken belief as of a body part and hand.

Thereby, one can know (not-V). This is a variant of the Moorean anti-sceptical argument, (a)-(b)-(c), from Section 2.3. There are parallel arguments for knowledgeably rejecting ordinary illusions. A sample conclusion is that it is not the case that I am under an illusion as of a surface’s being red, whereas it is white but illuminated so as to cause illusion. Here is the argument against entitlement—warrant without reason. (1B)

If I gain warrant for believing a proposition, my confidence in the proposition should increase. So (2B) If (M) and (A) provide warrant for (not-V), my confidence in (not-V) should increase. (3B)

Given the sources of entitlement for (M)—the perceptual states, (P-B) that₁ body part [with an attendant relevant shape, also perceptually referred to]

(P-H) that₂ hand my confidence in (not-V) should decrease.⁹⁵ ⁹⁴ I have slightly altered Moore’s proposition to have a content of a perceptual belief. The representational content is false (or not true) if and only if either demonstrative fails to secure a referent, or either attributive fails to be true of a referent picked out by the demonstrative that it guides, or the attributive hand fails to be true of the picked out body part. Referential applications are representational contents whose identity is fixed by some occurrent event that functions to refer deictically. Subsequent applications of this initial application in the argument (M)-(L)-(not-V) connect anaphorically to this one. Ultimately, the applications in (M) connect to an occurrent application in the perceptual state (PH) cited below. The perception and the belief have different occurrent applications, but they are anaphorically connected. If they pick out anything, they pick out the same thing. I represent them as one application. Perceptual applications succeed in referring if and only if the perceiver actually perceives the referent. There are no such things as veridical perceptual illusions. So if the perceptual applications refer to anything, they refer to a particular that is perceived via the perceptual state that underlies the belief expressed in the first premise. For discussion, see Origins of Objectivity, op. cit., 381–383 and Section 2.1. ⁹⁵ I cite two perceptions here (actually two aspects of a single perception) to allude to the complexity of transitions that lead to perceptual belief. The highest-level perceptual attribution that engenders the

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Both a sceptical-scenario illusion and an ordinary illusion as of a hand could consist in the same perceptual states (P-B) and (P-H), or would at any rate be indiscernible from them. So the degree to which I suspect that that particular illusion is occurring should increase. (4B) (5B)

The incompatibility of (3B) with (2B) undermines the idea that (P-B) and (PH) help provide an entitlement to believe that there is no illusion—(not-V). One needs independent ground to believe (not-V) in order to be warranted in believing (M) through having perceptions (PB)-(PH).

(5B) runs contrary to the view that one can be entitled to perceptual beliefs like (M), or other beliefs formed non-inferentially from perception, without support from further beliefs or presuppositions. Bayesian probability theory is invoked, prominently by Roger White, to yield the same result. Here I center on the Moorean argument: (M)-(A)-(not-V). The probability of (P-B)-(P-H) given (M) is greater than the probability of (PB)-(PH). The probability of (PB)-(PH) given (V) is greater than the probability of (PB)-(PH). By Bayes’s Theorem, the probability of (M) given (PB)-(PH) is greater than the probability of (M); and the probability of (V) given (PB)-(PH) is greater than the probability of (V). So the probability of (not-V) given (PB)-(PH) is less than the probability of (not-V). (PB)-(PH) raise the probability of (M) and (V). Perceptual experience’s raising the probability of (V) is incompatible with its giving one warrant to believe (not-V). Here is a corollary of the foregoing argument: (not-V) follows from (M), assuming, as we are, that both (not-V) and (M) are thought by the same person in a specious present on the occurrence of the same perceptual state instance. The probability of (PB)-(PH) given (V) is approximately 1. So how high the probability of [M given (PB)-(PH)] is depends mathematically on how high the probability of (not-V) is. The probability of (M) given (PB)-(PH) is less than or equal to the probability of (not-V) given (PB)-(PH). And the probability of (not-V) given (PB)-(PH) is less than the probability of (not-V). So the probability of (M) given (PB)-(PH) is less than the probability of (not-V). So (PB)-(PH) can make one warrantedly confident that (M) only if one is already warrantedly confident that (not-V). One needs independent ground to believe (not-V) in order to be warranted in believing (M) through having perceptions (PB)-(PH). This result runs contrary to the view that one can be entitled to hold perceptual beliefs like (M)—independently of support from further beliefs, or from one’s having access to further propositional considerations.⁹⁶ perceptual belief (M) is (PH). (PH) is inseparable from perceptual states like (PB) and from earlier-formed perceptions as of surfaces and shapes. When a perception as of a hand occurs (see note 25), that very perception involves attribution of 3-d body part that makes up the hand, attribution of the body part’s visible surfaces, shape, direction, orientation, distance, and so on. The arguments wash out this complexity, taking a single perceptual attribution to be the “justification” for the perceptual belief. This oversimplification is not crucial. It does, however, encourage ignoring probability assignments in the perceptual system that are epistemically basic to the perceptual belief. ⁹⁶ The foregoing reconstructs two arguments in R. White, ‘Problems for Dogmatism’, op. cit., 531–534. Analogs occur in J. Hawthorne, Knowledge and Lotteries (Oxford: Oxford University Press, 2004); S. Cohen, ‘Why Basic Knowledge is Easy Knowledge’, op. cit.; and N. Silins, ‘Basic Justification and the Moorean

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

Standard applications of Bayesian probability theory do not always coincide with intuitive judgments regarding warrant. Acquisition of beliefs against a background of lottery-like confidence assignments can yield cases in which one acquires new evidence that disconfirms a proposition even though one’s subjective probability in the proposition does not go down. Similarly, one can acquire evidence that supports a proposition although one’s confidence in the proposition does not go up.⁹⁷ John Maynard Keynes called such cases ‘situations of uncertainty’. He contrasted them with cases in which new evidence is acquired against a background of differentiated prior confidence assignments. He called the latter cases ‘situations of risk’. Situations of uncertainty do show that the intuitions in the relevant arguments do not apply to all cases. So it is important to get right the application of Bayesian considerations. Some authors think that lottery-like cases or cases of extreme uncertainty are key to rejecting the Bayesian-type arguments. I think that lotteries and other situations of great ignorance before forming perceptual beliefs are not good models for understanding relations between entitlements to perceptual beliefs and degrees of confidence. My reasons will emerge in due course. Here is another case in which the intuitions underlying the Bayesian-type arguments do not apply. It is plausible that we have both apriori and empirical warrant to believe arithmetical propositions. Suppose that we assign an apriori subjective probability of 1 to 2+2=4. We can also gain empirical evidence for believing the proposition through counting, or through the role of arithmetic in empirical science. The subjective probability assigned to the proposition, given the empirical evidence, should be lower than 1. The overall probability assignment should remain 1. Here overall subjective probability is not raised by new evidence. But the new evidence does seem to add to our warrant, though at a lower probability than our initial probability. This situation is again disanalogous to the perceptual case. The empirical warrant for believing 2+2=4 depends on understanding of, and belief in, the mathematical proposition. Entitlement to perceptual beliefs does not depend on capability of understanding any eternal proposition. The example does issue a relevant reminder that new evidence may not raise overall subjective probability if strong prior warrant is in place. And it again brings out that one must pay careful attention to context in applying general Bayesian updating intuitions. The relevant context here is, of course, formation of perceptual states and perceptual beliefs. I begin by saying a little about what is known about such formation. The intuitive and Bayesian arguments depend on assigning levels of confidence to psychological states. It will be well to understand the nature of relevant states. I focus on vision and visual perceptual belief.

Response to the Skeptic’, op. cit.. White takes the argument to undermine, among others, my account of perceptual entitlement (553n1). He focuses on Pryor’s “dogmatist” view not on mine. See note 86. ⁹⁷ This point is well made by B. Weatherson, ‘The Bayesian and the Dogmatist’, Proceedings of the Aristotelian Society 107 (2007), 169–185; P. Kung, ‘On Having No Reason: Dogmatism and Bayesian Confirmation’, Synthese 177 (2010), 1–17; and E. Zardini, ‘Confirming the Less Likely, Discovering the Unknown’ in D. Dodd and E. Zardini eds., Scepticism and Perceptual Justification (Oxford, Oxford University Press, 2014). All cite Keynes as having made the point first.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



In formation of visual perceptual states, the proximal stimulus—the light array just before it hits the retinal receptors—is the retinal image. The initial sensory registration of the retinal image is the first-occurring informational state of the perceiver on which the formation of perceptual states depends. After that initial registration, a series of transformations leads to a perceptual state—say, a visual perceptual state that represents as of a surface’s being at a certain distance. Further transitions from this perceptual state lead to others. For example, a perceptual state as of a surface might lead to a perceptual state as of a three-dimensional body. This latter perception might engender a further perceptual state as of a body part and hand (see note 25). In some perceivers, many perceptual states transition to perceptual beliefs. There is a transition from a singular perception as of a surface (functioning to pick out a particular surface) to a belief that that is a surface. There is a transition from a singular perception as of a hand to a belief that that is a hand. The three primary types of state here—initial sensory registration, perceptual state, perceptual belief—have different formats. The initial sensory registration carries information about the proximal stimulus— the array of light intensities that strike the retina. The initial sensory registration is not a perception and does not by its nature have veridicality or accuracy conditions. The sensory registration is caused by, correlates with, and functions to correlate with the proximal stimulus. It is the basis for perceptual processing and all subjective probability assignments. Still, subjective probabilities regarding accuracy or truth apply only to perceptual states and beliefs, not to registration of retinal information.⁹⁸ Second, perceptual states have veridicality conditions for accuracy, as an aspect of the psychological kinds that they are. They represent non-propositionally.⁹⁹ My formulations of (PB) and (PH) are non-propositional. There is a transition from (PB) to (PH) in the perceptual system. Assuming, as I am, that there are states produced by a perceptual system that are as of hands (see note 25), there is a transition from the perception as of a certain shaped body part (PB) to a perception as of a hand (PH). Antecedent perceptual states led to (PB). Perceptions as of bodies and body parts are formed from perceptions as of shapes, textures, motion, surfaces. This broad order of formation is well-known in perceptual psychology. There are

⁹⁸ For discussion of lack of veridicality conditions in sensory registrations, see Origins of Objectivity, op. cit., chapter 8. Despite the fact that subjective probability regarding veridicality does not apply to information registrations, Bayesian assignments of prior and conditional probabilities of occurrences of such registrations do apply. ⁹⁹ Work in perceptual psychology does not treat perceptual states as having propositional form. I believe that this treatment is correct. The representational function of perception is to identify particulars, using attributives entirely to serve identification. Function determines form. The representational contents of perceptual states have the form of singular representations guided by attributives. The form is analogous to that of a noun phrase governed by a context-dependent determiner applied on a particular occasion. All perceptual attributives for perceivable particulars accompany applications that function to refer to instances of relevant attributes. The form differs from that of a noun phrase in language in that it is iconic. The iconic format of visual perceptual representational contents still, however, includes contextually applied determiners and attributives. For further discussion, see Origins of Objectivity, op. cit., 537–546 and ‘Origins of Perception’, Disputatio 4 (2011), 1–38; ‘Steps Toward Origins of Propositional Thought’, Disputatio 4 (2011), 39–67; also Elizabeth Camp, ‘Thinking with Maps’, Philosophical Perspectives 21 (2007), 145–182. Further discussion occurs in Perception: First Form of Mind, forthcoming. See note 21.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

also transitions from perceptions to beliefs. Less is known about these transitions. I assume that they tend to be automatic and direct. Third, the representational contents of perceptual beliefs are propositional. Recall from Section 2.1 that perceptual beliefs take over perceptual attributives and embed counterparts in a propositional structure. They include occurrent applications of demonstrative-like determiners that take over, anaphorically, referents of occurrent applications of the perceptual determiners. Proponents and opponents of the intuitive and Bayesian arguments consistently ignore basic features of perception and perceptual belief. This practice gives the arguments, and discussions of them, an overly abstract, almost make-believe character. I cite two facts about the formation of perceptual states and perceptual beliefs that the relevant arguments commonly ignore. One is that there is a change in representational format from non-propositional perceptual state to propositional perceptual belief. Some forms of the arguments take (M) to be justified by a further belief about a perceptual appearance. Passing over the distinctive way in which perceptions figure in yielding warrant for perceptual beliefs raises questions about whether the arguments apply to the epistemology that the arguments purportedly target. Ignoring differences in format and function between perceptual states and perceptual beliefs runs together justification and entitlement.¹⁰⁰ Entitlement to perceptual beliefs is grounded in non-propositional transitions within the perceptual system and between the perceptual system and perceptual belief. Epistemic warrant attaches only to propositional attitudes and propositional inferences. States warranted through entitlement are starting points in the structure of empirical epistemic warrant. States that are empirically warranted through justification are always warranted either through self-evidence or from inference from other warranted propositional states. Blurring the distinction between propositional attitudes and perceptual states tends to distort understanding of base-line conditions for entitlements to perceptual beliefs.

¹⁰⁰ White not only takes the content of the warrant for (M) to be propositional. He takes it to be a belief about perceptual appearance: the belief it appears to me that there is a hand, ‘Problems for Dogmatism’, op. cit.. At 534–535, he remarks that in taking such a meta-belief as the initial warranting state, he overintellectualizes the actual situation. He notes that in being warranted in a perceptual belief, one need not reflect (I add, or even be able to reflect) on the relevant perceptual state. He claims, however, that the transposition does not matter. He writes, I think plausibly, ‘ . . . it surely should not make any significant difference whether I choose to consider how things appear to me, and form the belief that this appears to be a hand, or I don’t consider the matter. If the rational response to its appearing that this is a hand, when I also believe that it appears that this is a hand, is to decrease my confidence that it is a fake-hand, then surely this is the rational response to the same experience when I do not even consider how things appear to me.’ What I think plausible here is that whatever epistemic force the judgment about the appearance has in favor of believing (M) derives mainly from the formation of the perceptual state that the judgment is about. One can let the appearance judgment, to which one is entitled, count as a reason for (M). The reason is a proxy for the epistemic force present in the entitlement to the perceptual belief. While I do not think that White’s transposition is the crucial error in the argument against an entitlement-based account of warrant for perceptual belief, I do think that it obscures the difference between how things work in a subpropositional perceptual system, and among perceptual beliefs, on one hand, and how individuals operate with background knowledge, on the other. These points are elaborated below.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



I think, however, that the Bayesian arguments can be formulated in light of differences between perceptual states and perceptual belief, so as to remain prima facie threatening. I have tried to do so above. Subjective probabilities for veridicality apply, both intuitively and in science, to both propositional states—the beliefs—and to non-propositional states—the perceptions.¹⁰¹ In perceptual psychology, perceptual processing, which involves no propositional attitudes in the perceiver, is often modeled in a Bayesian framework.¹⁰² Bayesian subjective probabilities are assigned to perceptual states with no assumption that the states are propositional.¹⁰³ Perceptual states and beliefs are committal: success and failure of their representational functions depends on whether or not the states are veridical. Subjective probability assignments for confidence in veridicality make sense with respect to any committal state in a psychology. Subjective probabilities that are assigned to perceptual states are not conscious and do not result from individually controlled acts. They are embedded in the operations of the perceptual system. They model degrees of confidence, reliance, or anticipation. Terms like ‘credence’, which are essentially associated with belief, are inappropriate. The assignments are subjective in that they bear on the perspective of the individual or the individual’s states, sometimes concern the accuracy of such states, and may fail to coincide with objective probabilities.

¹⁰¹ L. Moretti, ‘In Defense of Dogmatism’, Philosophical Studies 172 (2015), 261–282, argues that White’s argument fails because of its assimilation of perception to a judgment about appearance. As note 100 indicates, I do not agree. (Incidentally, Moretti treats perceptual content as propositional and, like White, does not note the de re applied character of perceptual, or perceptual-belief-, content.) His position hinges on claiming (273–276) that the appearance judgment lacks the warranting force of the relevant perception, and that when one makes the judgment, one loses warrant for perceptual beliefs. That is, one cannot with warrant move from the appearance judgment to the perceptual belief. This claim is, I think, implausible and unjustified. In fact, the transition from the appearance judgment to the perceptual belief is a good one because (and only because) the transition from the perception to the belief is a good one. Moretti’s position is of a piece with the philosopher’s gambit criticized at the end of Section 2.3. If the position were correct, it would provide a pyrrhic victory for the view that perceptual beliefs are warranted through perception, unaided by background propositional resources. One’s non-reflective warrants would be undermined by the Bayesian considerations, as soon as one reflected on one’s perceptual states. The position also does not explain well why the Bayesian arguments cannot be reformulated to apply to warrant for perceptual belief through perception. Moretti does make some sound critical points against other defenses of “dogmatism”. ¹⁰² See A. Yuille and H. Bulthoff, ‘Bayesian Decision Theory and Psychophysics’ in D. Knill and W. Richards eds. Perception as Bayesian Inference (Cambridge: Cambridge University Press, 1996); W. Geisler and R. Diehl, ‘A Bayesian Approach to the Evolution of Perceptual and Cognitive Systems’, Cognitive Science 27 (2003), 379–402; W. Geisler, J. Perry, B. Super, and D. Gallogly, ‘Edge Co-occurrence in Natural Images Predicts Contour Grouping Performance’, op. cit.; L. Maloney and P. Mamassian, ‘Bayesian Decision Theory as a Model of Human Perception: Testing Bayesian Transfer’, Visual Neuroscience 26 (2009), 147–155; J. Frisby and J. V. Stone, Seeing (Cambridge, Mass.: MIT Press, 2010), esp. 308–322; J. Burge and W. Geisler, ‘Optimal Disparity Estimation in Natural Stereo Images’, op. cit.. Similar Bayesian methods have been used in sensori-motor psychology. See R. Shadmehr and S. MussaIvaldi, Biological Learning and Control (Cambridge, Mass.: MIT Press, 2012), chapters 5, 10–12; D. Wolpert and M. Landy, ‘Motor Control is Decision-Making’, Current Opinion in Neurobiology, 22 (2012), 1–8. For good overviews, see M. Rescorla, ‘Bayesian Perceptual Psychology’, The Oxford Handbook of the Philosophy of Perception, M. Matthen ed. (Oxford: Oxford University Press, 2015); and ‘Bayesian Sensorimotor Psychology’, Mind and Language 31 (2016), 3–36. ¹⁰³ The science tends to focus on states, as do I. Some accounts focus on transitions.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

Subjective probabilities assigned in a perceptual system can be discovered by inferring from reports, if the perceiver has language. Discovery need not depend on reports. It can center on action patterns—including hesitancy and speed of action—and in interactions among perceptual systems. For example, perceptual systems make compromises attuned to the relative probabilities of accuracy among perceptual states in different modalities (say, vision and touch).¹⁰⁴ Perceptual systems factor in noise, acuity, distance, lighting conditions, and so on, in determining the degree to which given types of perceptual states are relied upon for being accurate. Any epistemology that connects with the actual psychology of perception and perceptual-belief formation must accommodate scientific ways of describing these matters. Bayesian descriptions are not ubiquitous in perceptual psychology. However, they are prominent enough that a reasonable epistemology should accommodate them. The Bayesian arguments in current epistemology do not connect to Bayesian accounts in perceptual psychology at all. A second feature of perception and perceptual belief is the singular, occurrent, de re character of their representational contents. This feature is never remarked. Relevant representational contents are treated not only as propositional, but as eternal. This feature is the most important one to bear in mind if one is to understand wherein the arguments mislead. I express the representational contents of (P-B), (P-H), and (M) with subscripts on the demonstratives. The subscripts mark occurrent events at particular times—the actual occurrence of a perceptual state or instantiation of a perceptual belief. These occurrent events are always present in the representational contents of perceptual states and perceptual beliefs. The events are applications.¹⁰⁵ Applications function to refer—to apply the guiding attributives, like hand, to particular instances of the indicated attribute. The function is realized successfully when a particular is perceived and the perceptual attribution to it is accurate of it. Here, the function is realized successfully when the application perceptually refers to a particular hand. The de re character of perception and perceptual belief is pervasive. Every property—and kind-attributive in perception and perceptual belief guides and is accompanied by an application that functions to be de re. (See note 26.) Of course, in an actual visual perception of and as of a hand, one perceives not only a hand, but the constituting body part, the instance of its shape, the instance of its color, the instance of its texture, the instances of its surface parts—all de re. The perceptual beliefs to which we are entitled are similarly pervaded by applications that function to be de re. Any property- or kind-attributive in a belief that lacks such an application is not being applied in perception. It might rather be the product of association in longterm memory, inference, supposition, interlocution, or the like. So to understand the empirical starting points for empirical theory, one must take the representational contents of perceptual beliefs not only to contain applications that function to be ¹⁰⁴ M. Ernst and M. Banks, ‘Humans Integrate Visual and Haptic Information in a Statistically Optimal Fashion’, Nature 415 (2002), 429–433. ¹⁰⁵ Strictly, the same application type can occur more than once, in different events at different times, if all the occurrences are anaphorically related to a single occurrence that is the primary event of referring. See earlier discussion of applications in Section 2.1 and notes 24, 26, 59, and 94.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



de re. One must regard every property- or kind-attributive to be referentially applied in a de re manner to a purportedly perceived particular. I emphasize that a perception as of a hand cannot occur separately from a cluster of accompanying perceptual attributions. Here perception contrasts with linguistic attribution. Linguistically, attribution of hand can be separated from attributions of hand-shape, hand-size, and so on. A perceptual attribution as of a hand that is part of a perceptual attribution as of a [hand-shaped] body part with a certain orientation and size may be assigned a different subjective probability than that assigned to perceptual attribution as of a hand that involves different attributives for bodily shape and size. A perception as of a hand as at a great distance or as partly occluded will receive a different subjective probability than a perception as of a hand that is close up and unoccluded. So there are many ways of perceiving something as a hand—and perceptually believing something to be a hand. The de re nature of perceptual states and perceptual beliefs forces a distinction between a kind or type of perceptual state and an actual occurrence of a perceptual state. The kind or type is marked by the schematic representational content without the applications. (See note 58.) It is like ‘that hand’ without an occurrent application of ‘that’. An occurrence of a perceptual state or perceptual belief always has applications in the representational content. This distinction yields structure in assignments of subjective probabilities that is missing for eternal, non-de-re contents. Different probability assignments attach to the occurrence of an instance of a perceptual state type and the veridicality of occurrences of perceptual state types. Occurrence and veridicality are importantly different matters. I turn to fuller discussion of formation of perceptual states, with an eye to its connection to confirmation. Patterns of perceptual-state generation were formed through eons of sensory interaction with the environment. For example, light patterns typical of borders of shadows differ from light patterns typical of edges of bodies or surface reflectances. Differences between darker and lighter light patterns that mark shadows’ edges tend to be fuzzy. Differences that mark a body’s edge or a change in surface reflectance tend to be sharp. Exceptions occur, but are rare in natural scenes. Perceptual-state formation capitalizes on this pattern. Perceptual systems tend to discount shadows in determining edges of bodies or edges of surface color boundaries. In effect, the environmental regularity is coded into laws governing perceptual state formation. Formation laws closely fit statistical distributions of properties in the environment. If one value of a scalar property is more common than another, and the property is visually important, such relative frequencies are likely to be encoded in subjective probabilities in visual processing. The development of such probabilities derives both from evolutionary selection and from feedback in individuals’ learning histories. Some subjective probabilities are innately coded, through selection, in perceptual systems. At any given time in the life of a perceiver, such innate codings will have been adjusted through feedback— adaptation, priming, learning, and so on. As indicated earlier, even though they do not have propositional structures, perceptual states are committal and have accuracy conditions. Many mainstream

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

theories in visual psychology treat perceptual states as having subjective probabilities for accuracy. The handling of accuracy and inaccuracy in visual psychology is much more nuanced than its handling in colloquial speech. In ordinary speech, we commonly take our perception of distances, at least at close ranges, to be accurate. In the science, where relatively exact measurements are made, strictly accurate perceptual attributions are relatively rare, for scalar attributes. A surface, even at a distance of 100 cm, is not likely to be perceived as exactly at that distance, largely because of noise in the system. Visual systems can discriminate very small, fraction-of-a-centimeter differences at close ranges. Errors, evinced in the course of actions, are corrected through constant feedback. Colloquially, one thinks of a single subjective probability for accuracy and a single one for inaccuracy, with the probability for accuracy relatively high. Simple yes-no perceptual attributions, such as whether a surface is foreground or background, will tend to be strictly accurate. With most attributions—for example, attributions of scalar attributes like distance, length, color, shape, and size—the objective and subjective probabilities for strict accuracy will be relatively low. The science focuses on a distribution of different errors, with different subjective probabilities for each. Inaccuracy might derive from malfunction, noise, limited acuity, adaptation, priming, interference from another system, or brute error.¹⁰⁶ So the objective and subjective probabilities of getting the distance of a surface exactly right are commonly low. Commonly, getting the distance exactly right is at the top of a bell curve, with slight errors grouped around it, and larger errors falling off down the curve with sharply decreasing probabilities. There is a range of objective and subjective probabilities for types of error, including different magnitudes of error. Higher probabilities tend to be assigned in perceptual systems for small errors—such as attributing distances of 99.95 cm to a surface that is 100 cm away. Lower probabilities tend to be assigned for large errors, such as attributing distances of 400 cm, or having an illusion as of there being a surface at all.¹⁰⁷ There is another layer of complexity in the science’s account of assignments of subjective probabilities in perceptual systems. Let us focus on a visual attribution of the distance of a surface. Given a registration of a retinal image, the system is modeled as producing a distribution of probabilities for different distances: 100 cm probability of .01; 100.05 cm/.009; 99.95 cm/.0095; . . . . The system then applies a cost function for the different degrees of errors, and an estimator function that selects a distance for the perceptual state to attribute—say 100 cm. It is natural to assume that the distance selected is paired with the highest probability. Because of anticipated costs for different types of error, the system cannot be expected always to select the distance attribution that has the highest probability of being strictly accurate. I will ¹⁰⁶ Brute error is error that derives from a well-working psychological system that happens to be in a situation that yields the error. Brute error can be caused by residual noise. The notion is discussed in ‘Perceptual Entitlement’, op. cit., 507ff.. See the text above associated with note 12. ¹⁰⁷ I do not assume that perceptual attributions as of distance involve measurement units like centimeters. They are likely to be unit-free magnitudes. For a fine discussion of this notion, see C. Peacocke, ‘Magnitudes: Metaphysics, Explanation, and Perception’, in Mind, Language and Action: Proceedings of the 36th International Wittgenstein Symposium ed. D. Moyal-Sharrock, V. Munz, and A. Coliva (Berlin: de Gruyter, 2015).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



count as approximately accurate any perceptual attribution that is either strictly accurate or a near miss.¹⁰⁸ The science’s assumption that, for scalar properties, strictly accurate perceptual attributions are not common provides no encouragement to philosophical error theories about perception. The objective probabilities of small errors that cluster around the correct attribution are in aggregate very high. The objective probabilities of serious mis-attributions are low. Optimality theorems (see note 54) show that, in cases so far tested, visual systems make the size of errors close to as small as is theoretically possible, given the systems’ natural limitations—such as noise and acuity. Subjective probability distributions for errors are close to objective probability distributions. Individuals’ degree of reliance on one modality (say, vision) in relation to another (say, touch or audition) matches the relative objective probabilities for accuracy of the different modalities in contexts where their representations of a situation conflict. (See note 104.) So it is appropriate to say that although most types of visual attribution tend to be strictly inaccurate, most tend to be approximately accurate. They tend to be as close to accurate as physical limitations allow. It is objectively probable that deviations from accuracy are small. Perception yields approximately true and approximately knowledgeable perceptual beliefs in many important cases. Given evolutionary time, reliance on a given type of visual perceptual state has come for the most part to closely match the objective probability of that type’s being accurate. In subsequent discussion, I write of approximate veridicality and radical nonveridicality. Approximate veridicality includes both strict veridicality and relatively small non-veridicalities. Radical non-veridicality includes large non-veridicalities and referential illusions, such as whether there is a surface at all at an attributed distance. I assume that every occurrence of a perceptual state or perceptual belief is either approximately veridical or radically non-veridical. When philosophers discuss sceptical and ordinary illusions, they are large errors, not near misses. My simplification will not, I think, materially affect the discussion. Subjective probability assignments for veridicality, for small non-veridicalities, for large non-veridicalities, and for referential illusions are all—when warranted— warranted empirically, not apriori.¹⁰⁹ Even assignments encoded innately in a perceptual or perceptual-belief-forming system are empirically warranted. What warrant those assignments have depends for its force on reliable sensory contact to relevant entities in the environment, in the system’s evolution. A warrant’s force is apriori if that force does not derive from sensory input. Whether a state—or ¹⁰⁸ For discussions of cost and estimator functions, see A. Yuille and H. Bulthoff, ‘Bayesian Decision Theory and Psychophysics’, op. cit.; W. Geisler, ‘Ideal Observer Analysis’ in L. Chalupa and J. Werner eds., The Visual Neurosciences (Boston: MIT press, 2003); M. Banks, J. Burge, and R. Held, ‘The Statistical Relationship between Depth, Visual Cues, and Human Perception’, in M. Landy ed. Sensory Cue Integration (Oxford: Oxford University Press, 2011). ¹⁰⁹ Although, epistemic warrant attaches to perceptual belief, not to perception, I allow myself here to write of warrant for perceptual states. Perceptual states have an analog of epistemic warrant. Warrant for perceptual beliefs depends mainly on the well-functioning and reliability of the perceptual states. What I write here about sub-epistemic “warrant” for perceptual states carries over to the genuine epistemic warrant (entitlement) for perceptual beliefs.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

probability assignment—occurs in an individual’s psychology before the individual has sensory input is irrelevant.¹¹⁰ Since the discovery of evolution, it has been clear that being determined in the mind before an individual’s sensory contact with the world is not being determined independently of sensory contact in the individual’s psychological systems’ evolutionary history. The basic types of perceptual attributives and basic forms of perceptual processing are set through interaction with the environment before the birth of any human individual. The degree to which a given type of perceptual state is likely to be accurate depends heavily on patterns of sensory interaction with the environment in the evolutionary pre-history of the individual. The force of one’s warrant for perceptual beliefs depends primarily on the reliability of perceptual accuracy. To take assignments of subjective probabilities to perceptual beliefs to be apriori because they are determined before an individual’s experience occurs would be to deem many of our specific perceptual beliefs apriori warranted—an absurd result. All warranted subjective probability assignments for specific perceptual beliefs that are discussed here are empirically warranted. Subjective probabilities are assigned in the first instance, for a perceiver and time, to types of perceptual state. Recall that actual perceptual states always have occurrent, de re elements—applications. State types are representational contents of a perceptual state, for a given perceptual modality, with occurrent applications removed. (See note 58.) For an individual at a time, all possible occurrences of a given perceptual state type receive the same subjective probability assignment, if they derive from the same antecedent state types.¹¹¹ Probability assignments concern the role of types of states in patterns in an individual’s psychology, and between the psychology and the environment. Few perceptual-state types occur at any given time. The system attaches subjective probabilities directly not to occurrent states, but to all and any occurrences of state types. To engage in a standard idealization: at any given time, every perceptual state type that could be produced in an individual’s perceptual system is assigned a subjective probability for approximate accuracy and for radical inaccuracy—strictly, for strict accuracy and for types and degrees of inaccuracy. The subjective probability for veridicality assigned to a perceptual state depends on the success record, in evolutionary history and as recorded in feedback during

¹¹⁰ Some have held that if a probability can be determined before an individual’s sensory contact with the world, that makes it apriori. Such views exhibit serious misunderstanding of what apriority is. Apriority has always concerned the nature of a warrant, even in pre-Kantian apriority concepts. See my ‘Frege on Apriority’ in New Essays on the A Priori, C. Peacocke and P. Boghossian eds., (Oxford: Oxford University Press, 2000); reprinted in Truth, Thought, Reason: Essays on Frege (Oxford: Oxford University Press, 2005). Before Kant, some ran together psychological priority, particularly innateness, and apriority, because they thought that innateness entailed independence of the force of a warrant from experience. Leibniz knew better. Kant, the father of modern conceptions of apriority, knew better. He stressed a distinction between psychological origin and justificational priority. See Critique of Pure Reason A1/B1. ¹¹¹ For a given individual and time, there could be different applications of the same content. In the actual world, applications are different if they apply to different particulars. An individual in the actual world could be presented with perceptually indiscernible particulars on different occasions. Then the individual’s perception—and belief-types involve different applications.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



the perceiver’s life, of that type of perceptual state. The same perceptual-state types in different individuals, with different sensory histories, usually have different subjective probability assignments. Different types within an individual tend to have different success records, hence different assignments. As noted, a visual system tends to assign a higher subjective probability for being approximately accurate to a perception as of a close surface than a perception as of a distant one. Another way of phrasing this point is that, built into the laws of a visual system are anticipations that attributions of close surfaces are likely to make smaller errors than attributions of distant surfaces. A huge number of probability assignments are present in a perceptual system at any given time—antecedent to the acquisition of an occurrent perception. A huge amount of information about subjective probability of approximate accuracy and radical inaccuracy is modeled as coded into a perceptual system at any given time. Such coding is in place before the system receives stimulation that causes an instance of a perceptual state type. Different subjective probability assignments for accuracy and inaccuracy of perceptual state types apply at each level of processing. Perceptual states formed at later stages depend on states at earlier stages. Glitches are possible at every transition. So the subjective probability of the approximate accuracy of a perceptual state type at a later stage is often slightly lower than that of a perceptual state type at an earlier stage. A system normally assigns a very slightly higher subjective probability for the approximate accuracy of a perception as of a 3-d shape than to that of a perception as of a body with that shape. It normally assigns a higher probability to the perception as of a 3-d shaped body part than to a perception as of a hand with that body-part shape. The points made about subjective probability assignments to perceptions carry over to those for perceptual beliefs. Like perceptual states, perceptual beliefs function to be de re. Perceptual beliefs are formed as direct conceptualizations of perceptions. For any given believer and time, all possible instances of a specific type of perceptual belief, with possibly different occurrent applications, receive the same distribution of subjective probability assignments for strict accuracy and for various types of errors. On our idealization, each perceptual belief type that can be generated in a perceptualbelief-forming system receives, for a given believer and time, a distribution of subjective probability assignments. The distribution aggregates, I am supposing, into a dual assignment for approximate truth and radical error. Those two assignments add to 1. Normally, the main determiner of the subjective probability for approximate truth of a perceptual belief is the subjective probability for approximate accuracy of the perceptual state that the belief conceptualizes. The perceptual belief ’s probability is normally very slightly lower—to allow for glitches in the transition from perception to belief. To fix ideas, I sketch a Bayesian model for a specific case. I begin by discussing objective probabilities. The environment contains attributes—kinds, properties, and relations—that the visual system functions to represent and attribute. Any given perceptual representation functions to represent a scene made up of instances of many attributes. I simplify, fixing on one attribute: the distance of a surface, symbolized by the variable: D. The psychologist, following in the footsteps of evolution, records the statistical distribution of various distances of surfaces from the perceiver’s

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

eyes at any given time. The nearest visible surfaces are several centimeters away. The farthest are at the horizon.¹¹² The probability distribution peaks at the most probable distance of a surface within this range and falls off from that peak. That is, there is a probability distribution that integrates to 1 that gives the objective probability that each discernible surface distance will occur in relation to the perceiver. Focusing here on objective probabilities, p(D) is the objective prior probability that distance D, for each D, is a surface’s distance from the perceiver. Environmental attributes cause proximal stimuli, which cause initial registrations of the proximal stimuli. For vision, initial registrations are of the retinal image. I symbolize the initial registration of a proximal stimulus I. There is a distribution of objective prior probabilities for the occurrence of the various possible initial registrations. As an idealization, for each possible initial registration, there is an objective prior probability that it will occur. I symbolize the objective probability of the occurrence of I: p(I). p(I) is a distribution of probabilities. For a particular value assigned to I, I write ‘Ir’. ‘Ir’ designates any member of a class of initial registrations that do not differ in ways that affect a perceptual state: all produce the same perceptual state (not merely perceptual states that represent the same attributes). So small differences among initial registrations that are within some criterion for unimportant differences are ignored in use of the symbol ‘Ir’. p(Ir) is a specific probability within the probability distribution. One could regard ‘Ir’ as a canonical name. Then p(Ir) would be a specific probability within the probability distribution. For scenes in which the stimulus is not impoverished and noise in the system is ordinary, some probabilities that given initial registrations will occur, for a given distance value Dj, will be higher than others. The probabilities are found by studying environmental properties, optics, and properties of sensory receptors. The science has done a lot of work measuring such probabilities.

¹¹² For detailed discussion of measurement of the statistics of various attributes in natural scenes, see W. Geisler, ‘Visual Perception and the Statistical Properties of Natural Scenes’, Annual Review of Psychology 59 (2008), 167–192; A. D’Antona, J. Perry, and W. Geisler, ‘Humans Make Efficient Use of Natural Image Statistics When Performing Spatial Interpolation’, op. cit.; J. Burge and W. Geisler, ‘Optimal Disparity Estimation in Natural Stereo Images’ op. cit.. Most popular philosophical discussions of objective environmental priors cite such priors as that light is more likely to come from above than below. Such illustrations tend to obscure the variety and complexity of environmental statistics that are coded in perceptual systems. A huge amount of information is coded through evolution. This information is tuned through learning. Statistics that are likely to remain stable, such as statistical facts about motion governed by the principle of inertia, are likely to be innately coded. Statistics that vary during the lifetime of the organism, such as the distribution of prey, are likely to depend more on learning. Measurement of some prior objective probabilities is much less difficult than that of others. Distance, being one dimensional, is relatively easy. Tilt, slant, binocular disparity values are harder but tractable. Hue is two dimensional. Size is two- or three-dimensional. 3-d shape is multi-dimensional. Higher-dimensional quantities are practically impossible to measure, because of the amount of data required to obtain reliable statistical measurements. Priors of any perceivable attribute are, however, in principle determinable, by way of dimensionality reduction—discussed in note 113. Similarly, for the idealization about initial retinal image registrations–which are extremely high-dimensional—cited in the next text paragraph. I ignore surfaces of heavenly bodies, in referring to the horizon. I doubt that evolution provided for them in the same way that it provided for distances of surfaces on earth. Heavenly bodies can be seen. I doubt that human visual systems attribute definite distances to them.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Given the optical laws and given that the perceiver’s eyes are open, the distribution of objective prior probabilities of distances of surfaces, together with a factor for degradation of the signal through noise, determines a distribution of probabilities that various initial registrations will occur for a surface at each given distance. From this joint distribution, a conditional probability distribution for the probability of any I for given a fixed Dj can be derived: p(I | Dj). So far, I have discussed objective probabilities relevant to Bayes’s theorem. The probabilities are objective in that they do not concern the perspective of the perceiver, and involve no assignments of probability in any psychological system to a representational state. Let us turn to interpreting Bayes’s theorem as being about subjective probabilities assigned in a visual system. The probabilities are subjective in that they are assigned in a psychological system and concern the perceiver’s perspective, in that they may not coincide with the objective probabilities, and in that some of them, at least, concern the probability of the occurrence or accuracy of a perceptual representational state. I write ‘ps’ for ‘the subjective probability of ’. Because I is not a representation with accuracy conditions, the meaning of ‘ps(I)’ is not: the subjective probability of an accurate occurrence of I. ‘ps(I)’ means: the subjective probability that I occurs, where I is any initial image registration possible in the system. The science presumes that visual systems can be modeled as establishing a probability distribution for occurrences of initial image registrations. Strictly, a visual system does not track full, complex initial image registrations themselves, but low-dimensional feature sets within those registrations that are relevant to the type of perceptual attribution at issue.¹¹³ Again, these are not assignments of probabilities by individuals, much less conscious assignments. The subjective probability is an anticipation setting or reliance setting. Different degrees of anticipation or reliance correlate with different subjective probabilities of an image registration’s occurring. The anticipations are usually closely correlated with the objective probability of the registration’s occurring (or a dimensionalityreduced aspect of its occurring). Discounting noise, a perception as of a surface-distance is a computational function of the occurrence of a relevant initial image registration for a given individual at a given time. So the subjective prior probability of an occurrence of a relevant image registration will yield a subjective prior probability for an occurrence of the corresponding perception as of surface distance. Any subjective probability of ¹¹³ Given the colossal number of possible retinal images, and because even local segments of images are very high-dimensional, image registrations may seem crazily unmanageable. The science presumes, however, that perceptual systems do manage. They are presumed to cut through the complexity of full image registrations via operations of dimensionality reduction. For an overview of such techniques, see Ali Ghosi, ‘Dimensionality Reduction: A Short Tutorial’, http://www.math.uwaterloo.ca/~aghodsib/courses/ f06stat890/readings/tutorial_stat890.pdf (sourced August 2015). For illustrations of application of such techniques in vision science, see J. Burge and W. Geisler, ‘Optimal Disparity Estimation in Natural Stereo‐ Images’, op. cit.; J. Burge and W. Geisler, ‘Optimal Speed Estimation in Natural Image Movies Predicts Human Performance’, Nature Communications 6, Article number 7900 (2015). These articles show that only a relatively small, statistically manageable number of dimensions (on the order of 6–8) is needed to produce ideal solutions to several problems and to predict human performance—for such features as binocular disparity and speed.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

the occurrence of a perceptual state—again discounting noise—will be close to and derivable from the subjective probability of the occurrence of the image registration that causes that type of state. We are primarily concerned not with objective probabilities regarding distances, but with subjective probabilities concerning perceptual state types that attribute distances. I symbolize a perceptual state type as of the distance of a surface: D*. Subjective probabilities apply to, among other things, perceptual (and belief) state types. A perceptual state that attributes a specific distance j is symbolized: D*j. There are many perceptual state types that attribute the same distance, most of which enter into different subjective probability distributions. To indicate a specific perceptual state type a that attributes distance j, I write D*aj. A specific attributive for distance will fix the attributed distance. So a determines j. I cite these items—a specific perceptual state type that attributes a distance and the distance itself—separately to emphasize their distinctness.¹¹⁴ I am interested in subjective probabilities for two features of perceptual state types. One is the probability of there being an occurrence of an instance of a perceptual state type, for a given individual at a given time. The other is the probability of an occurrence of a perceptual state type’s being approximately veridical for a given individual at a given time. I symbolize the subjective probability of there being an occurrence of (an instance of) perceptual type D*: ps(O(D*)). I symbolize the subjective probability of the approximate veridicality of any occurrence of perceptual type D*: ps(V(D*)). The subjective probability of the radical non-veridicality of any occurrence of D* is [1- ps(V(D*))]. All subjective probabilities are understood to be for a given individual at a given time. How does a visual system develop prior probabilities for there being an occurrence of a given type of perception and for any such occurrence’s being approximately veridical? As I indicated, a visual system can derive the probability of an occurrence of an instance of the perceptual state type from the probability of an occurrence of the initial image registration that computationally determines occurrences of

¹¹⁴ I discuss Bayesian models on the understanding that psychological laws, or law-like patterns, include specific perceptual state types. Different ways of representing a given distance have different psychological properties. They derive from different initial image registrations. They may be associated with different confidence levels. They may induce action at different rates of speed. They may have different priming or other associational histories. And so on. For more on this matter, see my ‘Reply to Rescorla and Peacocke: Perceptual Content in Light of Perceptual Constancies and Biological Constraints’, Philosophy and Phenomenological Research 88 (2014), 485–501. I think that these points are known and assumed in the science, by reflective scientists. However, even many Bayesian models in the science operate on an idealization–in effect, equivalence classes of psychological states. In our case, a relevant equivalence class would be visual perceptual states that represent a surface as being at a specific distance D*j (regardless of specific ways in which the distance is perceptually represented). Everything that I say about specific perceptual states, like D*aj, in what follows can be transposed to apply to coarse-grained equivalence classes of specific perceptual states and classes of particular initial image registrations. Strictly, the subjective probability assignments would be for averages of assignments to members of the equivalence classes, or paradigmatic assignments for the equivalence classes. The basic points that I will make apply under this transposition.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



the perceptual state type. Strictly, if one averages across the variation of state types caused by the initial image registration together with the noise, one gets an expected, computed state type. (Noise causes different state-types given a type of initial registration.) I will continue to use shorthand. I will write of the perceptual state type computed from the initial registration, ‘discounting for noise’. The system can determine a probability for the occurrence of an initial image registration (or rather the relevant low-dimensional aspect of it) by tracking feedback frequencies. A visual system builds estimates or assignments of probabilities for the approximate accuracy of an occurrence of a perceptual state type in a different way. Each possible initial image registration, or rather the relevant low-dimensional aspect it, has a distribution of subjective probabilities for the various distances. By applying a loss function and an estimator function, the system selects an attributive for one of these possible distances–j. (See note 108.) A loss function assigns penalties to various types and degrees of error. An estimator function operates on the loss function, according to some principle. The estimator function selects a perceptual attribution, taking into account the losses that would be incurred if the various respective attributions were inaccurate. The selection yields a perceptual state type a that attributes distance j. Given a probability distribution of distances, relative to an initial registration, and given an initial registration, the system can compute a probability distribution for strict accuracy and various degrees of error for instances of the perceptual state type that would be computed from that registration. The probability of approximate accuracy for instances of the perceptual state type is derived from this distribution. All and any instances of the state type receive the same subjective probability for being approximately accurate. The subjective probability assignments for approximate accuracy derive from feedback about the type, sign, and degree of error. The feedback helps calibrate the perceptual system, compensating for error, especially in action. Given an evoluationary and individual history of feedback, the system has a wealth of information from which to compute the probabilities of various types and degrees of errors.¹¹⁵ I emphasize two points here. First, the main shape of this distribution of prior probabilities is fixed by evolution, with adjustments in individual learning histories. This shape has a definite contour. Different types of errors are assigned different probabilities. Probabilistic relations between accuracy, different sorts of small errors, and larger errors are fairly well determined. Since the prior probabilities for perception are the main source of prior probabilities for perceptual beliefs, comparing prior probabilities for perceptual belief to those in a lottery case, or other cases in which one has no differentiating information, is off base. So the Keynesian distinction between uncertainty and risk is not central to understanding perceptual belief. Second, the information used by the visual system to produce these prior probabilities suffices to produce the conditional probability of an approximately accurate ¹¹⁵ There is evidence that human perceptual systems do build up subjective probabilities for veridicality, and the various types of possible error. See M. Ernst and M. Banks, ‘Humans Integrate Visual and Haptic Information in a Statistically Optimal Fashion’, op. cit..

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

perception of the relevant type given an initial image registration. Very likely, in the evolutionary pre-history of perceptual systems, the prior and conditional probabilities were established together from the same information. I will not focus on the genesis of these settings. I focus on Bayes’s theorem as applying (a) to the already setup probability distributions in a given individual at a given time, and (b) the use of those distributions when specific image registrations or perceptual states occur. I have discussed symbolism and content of the prior subjective probabilities for the occurrence of an initial image registration and for the occurrence and veridicality of the relevant perception. We are preparing for an instantiation of Bayes’s theorem. We now need to discuss the remaining term on the right side of the equation that constitutes the theorem. ps(I | O(D aj*)) is the subjective probability distribution of the occurrence of an initial image registration I, given an occurrence of perceptual state D aj*. ps(I | V(Daj*)) is the subjective probability distribution for I given an approximately veridical occurrence of perceptual state D aj*. When the perceptual type is fixed, a given I– say, Ir–is fixed. Recall that a state type a is an attribution of a specific distance j, in a specific way. So the subjective probability of Ir, given either O(D*aj) or V(D*aj), will be near 1. For specific perceptual types derive only from corresponding specific initial registrations, (or from a class of initial registrations whose differences are psychologically irrelevant) for an individual at a time, discounting noise. The left side of Bayes’s equation is a subjective posterior conditional probability. We have been discussing two subjective probability assignments—one for occurrence, one for approximate veridicality. ps(O(D aj*) | I) is the subjective probability distribution of there being an occurrence of a perceptual state type a that attributes distance j in the specific way it does, given the occurrence of any initial image registration I. Again, for an appropriate registration Ir, this probability will be in the neighborhood of 1, since, discounting noise, a given perceptual state type will be computationally produced by a given type of initial image registration. ps(V(Daj*) | I) is the subjective probability distribution of the approximate veridicality of any occurrence of a perception D*aj attributing distance j in the specific way a, given I. ps(I | O(D aj*)) is the subjective probability distribution of the occurrence of an initial image registration I, given an occurrence of perceptual state D aj*. ps(I | V(Daj*)) is the subjective probability distribution for I given an approximately veridical occurrence of perceptual state D aj*. When the perceptual type is fixed as a result of a computation, a given I—say, Ir—will be fixed, discounting noise. (Again, a state type a is an attribution of a specific distance j, in a specific way.) The system registers this relation. So the subjective probability of Ir, given either O(D*aj) or V(D*aj), will be in the neighborhood of 1. For specific perceptual types derive only from corresponding specific initial registrations, for a given individual at a given time, discounting noise. These conditional probability relations are set up in the system, at a given time, in advance of any given retinal image registration or any given perceptual state. Given that Ir computationally leads to O(D*aj), discounting noise, and given that the system can register this fact, the system will attach approximately the same subjective probabilities to Ir and O(D*aj). Given either one of them, the subjective probability of the other will be close to 1, discounting noise.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



The left side of Bayes’s equation can be written: ps(O(D* aj) | I). This is the subjective posterior probability of an occurrence of a perceptual state type a that attributes distance j in the specific way a it does, given the occurrence of initial image registration I. Again, this probability will be in the neighborhood of 1 for specific image registration Ir. For, discounting noise, a given perceptual state type is computationally produced by a given type of initial image registration. ps(V(Daj*) | I) is the subjective probability of the approximate veridicality of any occurrence of a perception D* aj attributing distance j in the specific way a, given I. Certain instantiations of Bayes’s theorem will be important to our assessments of the arguments that began this section. This is the joint subjective probability of O(D*aj) and V(D*aj). It is written: ps(O(D*aj), V(D*aj)). This is the probability of the conjunction: there is an occurrence of D*aj and that occurrence is approximately veridical. By the laws of conditional probability, this is equivalent to the joint probability of there being an occurrence of D*aj and D*aj’s being veridical, given that it occurs. This is written: ps(O(D*aj), (V(D*aj) | O(D*aj))). I am interested in this instance of Bayes’s theorem    ps OðD* a j Þ; VðD* a j Þ jIr ¼     ps ðIr j OðD* a j Þ; VðD* a j Þ Þps OðD* a j Þ; VðD* a j Þ ps ðIr Þ: Let us put together two points already made. First, at any given time, the system attaches approximately the same subjective probabilities to Ir and O(D*aj), where Ir is the initial image registration (or dimensionality-reduced aspect of it) that computationally produces the specific way of attributing distance j, D*aj, discounting for noise.¹¹⁶ Given either O(D*aj) or Ir, the system attaches approximately 1 to the other. Second, all subjective probabilities attach, in the first instance, to perceptual state types that attribute distance j in a specific way. So at any given time, the individual’s system attaches the same subjective probability for approximate veridicality to all and any occurrent instantiations of a given perceptual state type, independently of whether there are occurrences of that type. These two points entail that the following four subjective probabilities are assigned approximately the same probability value, for a given individual at a given time:   ðaÞ ps VðD* a j ÞjOðD* a j Þ   ðbÞ ps VðD* a j ÞjIr    ðcÞ ps OðD* a j Þ; VðD* a j Þ jIr   ðdÞ ps VðD* a j Þ :

¹¹⁶ I think that beliefs that derive from direct stimulation of the central visual areas of the brain are not of the same psychological type as visual perceptual beliefs. If this view were mistaken, the points of this paragraph could be rephrased so as to maintain their main thrust.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

The significance of this fact is that the subjective probability for approximate veridicality of an instance of a perceptual state is not affected by occurrence of the type. A subjective probability value for approximate veridicality is coded in the system before any occurrence of the type. It is useful to distinguish two aspects of subjective probabilities in dynamic operations. That is, there are two aspects to how subjective probabilities are affected from the time before an image registration occurs, or perceptual state occurs, and the time after it occurs. The first is the way in which there is a change in the joint subjective probabilities for the conjunction of O(D*aj) and V(D*aj), from the subjective prior probabilities for the conjunction, once there is an actual occurrence of Daj*—that is, given O(Daj*). For simplicity, let us take ‘D*aj’ to stand for either the perception or the perceptual belief derived from it. Subjective prior probabilities for having an occurrence of a given type of perception and for having an occurrence of the associated perceptual belief will be almost the same. Similarly, the subjective probability for the veridicality of an occurrence of a perception of any given type, for a given attribution, will be almost the same as the subjective probability for the veridicality of an occurrence of a perceptual belief that attributes the same attribute and is derived from the perceptual attribution. They operate dynamically in the same way, absent background information. The subjective prior probability that is a base for the change that we are interested in is the subjective prior probability for there being an approximately veridical occurrence of a specific state type—ps(O(D*aj), V(D*aj)). How is there a change in subjective probability, from this base, with the occurrence of a perceptual state or perceptual belief D*aj? As the arguments that began this section claim, the subjective probability will be higher. For the occurrence of D*aj will be registered by the system. So the first “conjunct” will be established. Parallel considerations show, as the arguments claim, that the subjective probability for there being a radically nonveridical occurrence of D*aj will be higher than the prior probability. Low subjective prior probabilities are assigned both to there being an approximately veridical occurrence of the state type and to there being a radically non-veridical occurrence of the state type. These low probabilities derive largely from a low prior probability that an instance of the perceptual, or perceptual-belief, state type will occur at all. Given that one does occur, the subjective probabilities for there being an approximately veridical occurrence and for there being a radically non-veridical occurrence of the state type will both rise. The second aspect of the dynamics of subjective probabilities concerns how subjective conditional probabilities operate before and after a perceptual state and perceptual-belief state occur. Subjective conditional probabilities for perceptual states and perceptual beliefs are just as much in place, at a time for an individual, as subjective prior probabilities are.¹¹⁷ For Bayes’s theorem to be of use, the system ¹¹⁷ Again, both prior and conditional probability assignments are instilled through eons of interaction with the environment, with feedback on errors. The illusion (V) as of a hand that derives from one’s being in sceptical scenario is not specifically allowed for in the low subjective probability of there being a radically false occurrence of an instance of the perceptual-belief type. Innate selection and learning are influenced

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



needs, before the occurrence of the next perceptual state, a model that derives both prior and conditional probabilities from the joint probability distribution. What is the effect of an occurrence of an instance of a perceptual state or perceptual belief of the relevant kind on these antecedently established subjective conditional probabilities? The occurrence of the initial image registration causes an occurrent instantiation of the perceptual state and perceptual belief. The system registers these events. So ps(Ir), given Ir is 1 and ps(O(D*aj)) given O(D*aj) is 1. The subjective probability of Ir given an approximately veridical occurrence (or any occurrence) of D*aj is close to 1. So the subjective probability of the approximate veridicality of D*aj given a particular occurrence of D*aj is the same as the prior subjective probability of the approximate veridicality of any occurrence of D*aj. Thus (a), (b), and (c) have the same subjective probability value as (d). The system has already assigned subjective probabilities to the approximate veridicality and radical non-veridicality of all and any occurrences of given types of perceptual states and perceptual beliefs before they occur. The occurrences do not change those probability values. They do make the individual warranted in forming commitments to de re representational contents in accord with those values. Assume that the subjective conditional probability (and subjective prior probability) for the approximate veridicality of D*aj is very high. Assume that the subjective conditional probability for radical non-veridicality is very low. Then the occurrence of the initial registration Ir and the occurrent instantiations of the perceptual state-type and perceptual belief-type D*aj leave the subjective probabilities for radical nonveridicality of these occurrent state instantiations well below the subjective probabilities for the approximate veridicality of the state occurrences. These probability values do not change with the occurrence of the perceptual state. The high subjective posterior conditional probability of the approximate truth of any occurrence of a perceptual belief type, given an occurrence of the corresponding perceptual state and initial image registration, is set before the occurrence of the image registration. Equivalently, the low subjective posterior conditional probability of the radical falsehood of an occurrence of a perceptual belief, given an occurrence of the registration of the retinal image, is in place before the occurrence of the image registration. If the image registration occurs, the relevant perceptual states and perceptual beliefs occur. So given the probabilities in place before the occurrence of instances of the perceptual state types and perceptual belief types, the subjective probability of the approximate veridicality of any instances of those types will be high. The subjective probability of there being an occurrence of the perceptual belief type is higher after the occurrence than prior to the occurrence. So the subjective only by feedback from errors that actually occurred in the molding of the system. Sceptical scenarios are not among such occurrences. So they are never among the specific radical falsehoods that a perceptual system allows for. A corollary of this point is that sceptical scenarios are not specifically within the hypothesis space. They are, however, species of the generic category of radically non-veridical representations (say, as of hands). As stressed in Section 2.3, sceptical scenarios are not relevant threats to entitlements. Rejecting ordinary radically non-veridical perceptual beliefs does not require a special reason, unless there are specific reasons to think that the radical falsehoods are threats. The sceptic must show that rejection of sceptical scenarios is different. I think that there is no reason to believe that, to be warranted in rejecting the sceptical scenarios, one needs a reason to reject them.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

probabilities of there being an approximately veridical and of there being a radically non-veridical occurrence will both be higher after there is an occurrence of an instantiation of the perceptual belief type. It is true that the occurrence of the perceptual belief type is exactly what one would expect if one were to have a radically false occurrence of just that type. But all three subjective probabilities (for there being an occurrence of the perceptual type, for any occurrence’s being approximately true, and for any occurrence’s being radically false) are set independently of any occurrence. So the occurrence does not affect one’s subjective probabilities specifically for approximate truth/radical falsehood. The perception-forming and perceptual-beliefforming systems ground one’s entitlement to most perceptual beliefs as they occur. Since the approximate truth of the perceptual belief is logically incompatible with its radical falsehood, the psychological system will place a much higher subjective probability on denials of the radical falsehood of occurrences of the perceptual belief than on acceptances of their radical falsehood. So the psychological systems’ probability assignments, given that they are reliable, ground entitlement to reject an occurrence of the perceptual belief as radically false. Let us return to the point that after occurrence of the image-registration, perceptual state, and perceptual belief, there will be a rise from the subjective unconditional probability of there being a radically false occurrence of the perceptual belief. This suggests that there will be some “confirmation” of the proposition that there is a radically false occurrence of the relevant type of perceptual belief. Confirmation here is an appropriate increase in subjective probability. It does not amount to prima facie warrant. Prima facie warrant marks a belief ’s propositional content as worthy of belief and as prima facie knowledge. Nothing so strong can be claimed for the increase in probability that there is a radically false occurrence of that type of perceptual belief. Consider again: (not-V) it is not the case that (V) I am in a sceptical scenario with a mistaken belief as of a body part and hand. (not-V) is equivalent to: (not-V’) it is not the case that (V’) I am having a belief as of a body part and hand and that belief is an illusion by virtue of my being in a sceptical scenario. (not-V) is also equivalent to: (DeMorganized not-V) Either I am not having a belief as of a body part and hand or I am having such a belief and it not an illusion by virtue of my being in a sceptical scenario. These unapplied schemas do not themselves have truth-value. Hence they do not themselves have subjective probabilities for being true or false. The assigned probabilities apply to all and any instantiations of the schemas—all possible applications of them. In particular, they apply (at a time, for an individual) to all actual or possible applications of the progressive tense and applications within the perceptual belief content that₁ body part is a handthat2. Recall, yet again, that a visual perceptual belief of type that body part is a handthat is a very specific way of believing—with

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



specific body-part-, size-, and shape-attributions. (See note 114.) The schemas leave open whether there is an occurrence of the relevant belief-type. A high subjective unconditional prior probability in these cases is largely grounded in a low prior probability of there being an occurrence of a relevant perceptual belief type at all— radically non-veridical or not. Suppose now that the relevant perception and perceptual belief do occur. There are de re applications of the propositional forms. Applications occur both in the perceptual belief and in self-attributions of it. Applications of tense in (not-V), (not-V’), and (DeMorganized not-V) occur. The first disjunct of (DeMorganized not-V) is falsified. Parallel points apply to (not-V) and (not-V’). The subjective probability of all three decreases. The subjective probabilities that had been schematically assigned to there being occurrences of that is a hand and I am under a brain in vat illusion in the belief that that body is a hand increase. In shorthand, when there is an applied occurrence of the perceptual-state type and perceptual-belief type, the subjective probabilities of all three conjunctions [occurrence and approximately veridical], [occurrence and not radically non-veridical] and [occurrence and radically non-veridical] all go up. They go up because the psychological system has information that shows that the first conjunct is true. These points accord with the claim in step (2B) of the first argument. The occurrences of the relevant perceptual-state type and perceptual-belief type are exactly what one would expect if one were to have a radically non-veridical perceptual belief of the relevant type (even a brain-in-vat-illusion). But these changes in subjective probability do not bear in any way on the subjective probabilities already assigned to the approximate truth and radical falsehood of any application of the schematic forms that body part is a handthat or I am under a brain in vat illusion in the belief that that body part is a handthat. The changes in probability so far discussed derive entirely from the belief ’s propositional content’s coming into existence. The changes in subjective probability that the arguments fix upon are not changes that focus on the subjective probability of the approximate truth or radical falsehood of any occurrent perceptual belief. All perceptual beliefs are occurrent. As I remarked, there are subjective probability assignments that do assess the approximate truth and radical falsehood of any occurrence of an instantiation of the perceptual belief type. These assignments are already in place before the occurrence of the perceptual belief. When the applied propositional content that instantiates the perceptual belief type does occur, those subjective probability assignments for approximate truth and radical falsehood attach to the occurrence. They attach to that₁ body part is a hand that2 and to I am₀ under a brain-in-vat illusion in the belief that₁ body part is a handthat2. One such assignment is the assignment that is conditional on the occurrence of a relevant image registration, or the occurrence of an instance of a relevant perceptual state type. Again, such a conditional assignment is in place in the psychological system before the occurrence of the de re applied perceptual state and perceptual belief. When the condition is met—when there is an occurrent instantiation of the perceptual state type—the relevant degree of subjective probability attaches to the occurrent perceptual belief content. There is also a subjective non-conditional probability assignment in the psychological system before the occurrence of the relevant

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

image registration, perceptual state, or perceptual belief. This assignment is for the approximate truth (and correspondingly, the radical falsehood) of all and any occurrences of the relevant perceptual belief type. This assignment has approximately the same value as that of the conditional probability assignment. For, roughly, discounting for noise and pathology, the perceptual belief content occurs if and only if the perceptual state type occurs. Conditional and unconditional assignments are typically very high for approximate truth and very low for radical falsehood. The assignments to occurrent, de re applied perceptual beliefs yield entitlement to the perceptual belief, and to rejection of its being radically false. The assignments (contingently but non-accidentally) track the objectively reliable approximate truth of instances of the perceptual belief types. They provide confirmation of the perceptual belief ’s representational content when it occurs. Being assigned in the same psychological system at the same time, they limit the slight rise in the probability of a radically false occurrence that derives simply from there being an occurrence of the relevant type. Although the levels of subjective probability are fixed in advance and independently of any actual occurrence, an individual receives warrant for believing the belief ’s content only with its occurrence. An occurrent, de re applied perceptual belief content that can be approximately true or radically false does not exist before, or independently of, its occurrence. Lacking at least an anticipation of a specific occurrence (a case that I shall return to), no occurrent application can be individuated before its occurrence. There is no full-fledged perceptual belief content to be committed to, or for a subjective probability to be assigned to. Again, at any given time, the levels of subjective probabilities for the approximate veridicality and radical non-veridicality of any and every occurrence of a perceptual state and perceptual belief are assigned in advance of occurrence. In this sense, the assignments are not only unconditional, but “prior”. It is not a subjective prior probability in the sense that one already has some degree of belief or disbelief for a specific propositional content in advance of the occurrence. For the content of the belief does not exist before its occurrence.¹¹⁸ The subjective probability assigned to an occurrent, de re applied perceptual belief can be changed by acquisition of new information beyond that encoded in the perceptual belief. It is not changed by the occurrence of the perception or perceptual belief. As noted earlier, the subjective probabilities for the approximate truth—and for the radical falsehood—are empirical. They apply to any occurrence of a perceptual state or perceptual belief of a given type before—and independently of whether—the state type is instantiated. They are empirical because they depend for the force of the warrant that they help establish on past sensory interactions between perceptual system and environment, partly in the evolution of the perceptual system. (See note 110.) The subjective probabilities are set in the perceptual—and perceptual-beliefforming systems, not from the armchair.

¹¹⁸ The occurrent de re applied perceptual beliefs are flat-out approximately true or radically false, once they do occur. Their truth or falsity is not relative to a time, or to anything else.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



These points ground assessment of the arguments set out at the beginning of the section. I begin with the intuitive argument (1B)-(5B). It is important, in assessing steps (2B) and (3B), that one be clear exactly what they mean. (not-V) must be interpreted with some care. If (not-V) is regarded as a schema that leaves open whether there is a de re applied perceptual state and perceptual belief as of a hand, then as the argument claims, there is a decrease in subjective probability when (PB)-(PH) and (M) occur. However, the schematic interpretation of (not-V) is not the one intended in the Moorean argument (M)-(A)-(not-V). In the Moorean argument, (A) and (not-V) are interpreted as de re applied in a context. They are interpreted: (A₁)

that₁ body part is a handthat2 only if it is not the case that the current belief that₁ body part is a handthat2 is a brain-in-vat illusion

and (not-V₁) it is not the case that (V’) the current belief that₁ body part is a handthat2 is a brain-in-vat illusion. These propositional contents (not schematic contents) occur only because there are occurrences of (PB)-(PH) and (M). The subjective probability of the truth of (not-V) so interpreted does not decrease. Although neither (not-V₁) nor (M) existed before their occurrence, a given subjective probability was set in advance to be assigned to them. For there is a subjective probability assigned to the approximate truth, and correspondingly the radical falsehood, of every and any occurrence of the perceptual belief type that body part is a handthat for any applications of the thats. Indeed, since body part and hand are not scalar attributes, one can perhaps carry out the reasoning on truth and falsity (not their approximations) simpliciter. The relevant subjective probability does not change with the occurrence of applications. So, given the way that (not-V) is interpreted by the Moorean argument—as (not-V₁)—the subjective probability assigned to the truth of (not-V) does not change with the occurrence of (PB)-(PH) or (M). Prior to its occurrence, strictly speaking, one had no entitlement to believe (M). One could not have an attitude toward (M) or (not-V₁) themselves, since the propositional contents, which include the occurrent applications, did not exist. With (M)’s occurrence, one’s subjective probabilities that a true and that a false occurrence occur do increase. At the same time, the antecedent, high subjective probability for the truth of any occurrence of (M)—and low subjective probability for the falsehood of any occurrence of (M)—kick in. Those assignments place truth of the occurrence of (M) well above .5 and the falsehood of the occurrence well below .5. One gains an entitlement to believe (M). One gains confirmation, but not against a preceding baseline of a different assignment of subjective probability. One gains prima facie default entitlement in the occurrence of the perceptual belief. What of the reasoning from (M) and (A₁) to (not-V₁)? One can use the entitlement to (M) and the reason to believe (A₁)—an obvious logical or semantical truth—to provide a reason for (not-V₁). Again, the subjective probability of (not-V₁) does not

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

strictly rise against a baseline that preceded the occurrence of (M) or (not-V₁). (notV₁) itself did not exist to have a probability. The argument does provide a transmission of warrant from (M) and (A₁) to (not-V₁).¹¹⁹ To simplify all this, let us walk through the intuitive argument (1B)-(5B) to bring out again the intuitive points about where it goes wrong. Step (1B) If I gain warrant for believing a proposition, my confidence in the proposition should increase does not apply straightforwardly to entitlements to perceptual belief. I gain warrant to believe (M) That₁ body part is a handthat2 only with the occurrence of the belief. Usually, one has no level of confidence in (M), the occurrent de re belief itself, before its occurrence. In that sense, there is no rise in confidence. There is, however, a level of confidence in the truth of any instance of the schematic belief-type, That body part is a handthat. That level is set before the occurrence of any instance of that schematic type. So with an occurrence, that level does not rise. What rises is the level of my antecedent confidence in there being an occurrence of an applied instance of the schematic belief-type That body part is a handthat. It is true that I had very low confidence of there being any occurrence of the schema before (M) occurred, but higher confidence afterwards. The belief that₁ body part is a handthat2 is essentially an occurrent belief. It is also true that I had very low confidence in there being an occurrence of the negation of (M) before its occurrence and higher confidence afterwards. But the relative levels of confidence, regarding veridicality, in (M) and in its negation, are fixed in advance, via assignments to any occurrence of the relevant schematic types. The confidence level in (M) is much higher than the confidence level in its negation. Step (2B) If (M) and (A) provide warrant for (not-V), my confidence in (not-V) should increase is problematic in the same way. My confidence in the truth of any instance of the schema (not-V) (it is not the case that (V) I am under a brain-in-vat [belief] illusion as of a body part and hand) is set in advance. That level of confidence is vastly higher than the level set in advance for any occurrence of the schema (V). Those levels do not change with the occurrences of the relevant perceptual state and perceptual belief. The Moorean argument hinges on confidence in the truth of an actual occurrence of (not-V). I gain warrant for believing (not-V) only when it comes ¹¹⁹ One can also reason from one’s entitlement to reject the falsehood of (M) back to accepting the truth of (M). For the perceptual-belief-forming system assigns a very low subjective probability to the falsehood of (M). So one is entitled to believe: (M) is not false. One has reason to believe: (M) is not false if and only if (M) is true. (Clearly, if needed, one could substitute ‘not true’ for ‘false’ in the argument.) One has reason to believe: (M) is true if and only if that₁ body part is a hand that2. (These latter two beliefs are obvious logical or semantical truths.) So one has reason to believe: that₁ body part is a hand that2. This is warranted reasoning that supports its conclusion. It is obviously less natural than the Moorean argument from (M) and (A₁) to (not-V₁). It is less natural because it starts not with the ground floor perceptual belief, but with a meta-belief about (M). I mention the argument to stress that the perceptual and perceptual-beliefforming systems simultaneously assign subjective probabilities to all and any instances of a perceptualbelief-type for a whole distribution of cases. For non-scalar properties like body part and hand, I assume that the argument need not appeal to approximate truth and radical falsehood.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



into existence—only with its occurrence. So strictly speaking, Step (2B) is mistaken. The level of confidence for any occurrence of (not-V) is set in advance of, and independently of, its occurrence, and hence in advance of my having warrant to believe that specific occurrent propositional content. (3B)

Given the sources of entitlement for (M)—the perceptual states, (P-B) that₁ body part [with an attendant relevant shape, also perceptually referred to] (P-H) that₂ hand— my confidence in (not-V) should decrease

is problematic in the same way. It is true that my confidence in there being an occurrence of a perceptual state and perceptual belief predicted by (V) goes up. But my confidence in the truth of any instance of (not-V) is set in advance of its instantiation or occurrence, hence in advance of my having warrant to believe that specific occurrent propositional content. Of course, occurrences of perceptions can alter subjective probabilities set for future occurrences. Such changes include perceptual learning and learning associated with perceptual belief. Usually, significant, stable changes in subjective probabilities, for perceptions and perceptual beliefs (understood in my strict way), are slow to take hold. Since Steps (1B)-(3B) produce no reason to be puzzled by entitlements to perceptual belief, conclusions (4B) and (5B) of the argument cannot be drawn. In particular, contrary to step (5B), one does not need independent warrant, beyond that provided by a reliable, well-functioning perceptual-belief-forming system, to believe (M)— That₁ body part is a handthat2. Insofar as the subjective probabilities of those systems track objective probabilities, as they have been shown in the relevant science to do, those systems themselves help generate warrant for perceptual beliefs. These remarks about argument (1B)-(5B) carry over to the first of the two Bayesian arguments set out at the beginning of this section. What of the second Bayesian argument? This argument suffers from difficulties that undermined (1B)-(5B) and the Bayesian counterpart of (1B)-(5B).¹²⁰ The argument is correct, strictly speaking, in claiming that ‘the probability of (M) given (PB)-(PH) is less than the probability of (not-V)’. For (not-V)—interpreted as (not-V₁)—can be made true not only by (M) but by any illusion besides the brain-in-vat illusion. The argument errs, however, in its claim, ‘So (PB)-(PH) can make one warrantedly confident that (M) only if one is already warrantedly confident that (not-V)’. One has a warrant to be confident in (M) ¹²⁰ One problematic claim is: ‘And the probability of (not-V) given (PB)-(PH) is less than the probability of (not-V)’. If (not-V) is interpreted as the Moorean argument interprets it–as (not-V₁)— then the probability of the truth of (not-V₁) and the probability of the truth of [(not-V₁) given (PB)-(PH)] are almost the same. For the current perceptual belief that₁ body part is a hand that₂, mentioned in (not-V₁) does not normally occur without (PB)-(PH), and the system is sensitive to this fact. The ‘normally’ is, of course, a hedge. (not-V₁) could have occurred without (PB)-(PH), if it were caused by non-standard perceptual states together with noise. So strictly, the argument’s claim here is correct, but not for the reasons it envisaged. The difference in probability that the claim is correct about can be factored in, in advance of the occurrence of the perceptual states.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

through the occurrence of (PB)-(PH) and (M). Normally, one cannot be warrantedly confident in (not-V₁) before the occurrence of (M) or (not-V₁). Those contents do not exist before their occurrences. On the other hand, the subjective probability values to (M) and (not-V₁) are assigned to any occurrences of the propositional schemas in advance. So the relevant subjective probabilities for (not-V₁), as well as (M), are fixed independently of the occurrence of (M). The argument’s conclusion— (5B)

One needs independent ground to believe (not-V) in order to be warranted in believing (M) through having perceptions (PB)-(PH)

—is, however, mistaken. The subjective probabilities for (not-V₁) and (M) are fixed independently of the occurrence of (M). They are not fixed independently of one another. The perceptual-belief-forming system carries probabilities for strict truth, small errors, and radical errors all together. One can reason from (M)’s being true to its not being false, and from its not being false to its not being false in the particular way envisaged by (not-V₁).¹²¹ The force of the reasoning does not depend on any propositional support, beyond the entitlement to (M). I have remarked earlier (note 69) that my account of entitlement is not a dogmatist account. I want to expand the point. Pryor’s explication of dogmatism is as follows: The dogmatist about perceptual justification says that when it perceptually seems to you as if p is the case, you have a kind of justification for believing p that does not presuppose or rest on your justification for anything else, which could be cited in an argument (even an ampliative argument) for p. To have this justification for believing p, you need only have an experience that represents p as being the case. (See note 69.)

¹²¹ White claimed that the effect of his arguments was to require that one must have independent ground to believe that one is not having a brain-in-vat illusion. He claimed that the ground had to be independent of the occurrence of the perceptual state. Roger White, ‘Problems for Dogmatism’, op. cit., 533–534: ‘So its appearing to me that this is [a] hand can render me justifiably confident that it is a hand, only if I am already confident that it is not a fake-hand.’ White cites his point as supporting the view of C. Wright, ‘(Anti-) Sceptics Simple and Subtle: G. E. Moore and J. McDowell’, op. cit.. White took his claim to show that one needs an independent reason, a warranted propositional basis, to accept (or apply a high probability to) the negation of a claim that the perception is radically inaccurate and the belief is radically false. In fact, the relevant probabilities, though independent of the actual occurrence of the perceptual state and perceptual belief, are assigned in the perceptual system and perceptual-belief-forming system themselves. They are empirical, even fundamentally perceptual. The individual need not be able to think the assignments. They are part of the entitlement system, not part of a reason-based, justificational system, or a system of evidence citation. Contrary to what seems to be suggested in the quote, the confidence ratings of the [at least approximate] truth of the perceptual belief and of the denial of radical error in the perceptual belief are set together. Epistemically, neither the assignment of a high subjective probability to the approximate truth of the belief nor the assignment of a high subjective probability to the belief ’s not being radically false is prior to the other. Neither occurs before the other. The subjective probability assignments that favor approximate truth and disfavor radical falsehood are set independently of, and in advance of, the actual occurrence of the perceptions and beliefs. But both are set within the perceptual system and perceptual-belief-forming system themselves. White and Wright miss this fact. The relevant subjective probability levels are independent of and prior to the relevant de re applied beliefs’ and perceptual states’ occurrences. They are independent of its actually appearing to one that there is a hand. They yield entitlement only with the occurrences. But they are set within the perceptual system and perceptual-belief-forming systems themselves in advance of occurrences. They are not independent in the sense that they are background beliefs, supplementary to what goes on in the perceptual or perceptual-belief-forming systems.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



I agree that one normally has a warrant for a perceptual belief if one has a perceptual state that represents, non-propositionally, what the propositional content of the belief represents. I also believe that the warrant does not presuppose or rest on one’s warrant for anything else that could be cited in an argument for that propositional content. What I reject is that the warrant requires nothing other than having the corresponding perceptual state or experience. A perceptual state does not in itself provide any warrant at all for a perceptual belief. An entitlement requires something more than having the corresponding perceptual state. The distinction between the distinctive claim of dogmatism and just the view that, typically, warrant for perceptual belief does not depend for its force on reasons or propositional background protection is widely elided in the literature. Pryor does not blur the distinction. The occurrence of the perceptual state is not in itself the source of warrant. Again, in itself, the occurrence in itself provides no warrant at all for the belief. The perceptual state must derive from a well-functioning, reliable perception-forming competence, whose reliability is grounded in the environmental conditions in which the content-nature of the competence is determined. Reliability is not determined by the perceptual state itself. (See Section 2.2.) The same type of perceptual state metaphysically could be reliable or not in different content-constituting conditions. In some possible content-constituting (or content-sustaining) conditions, for a given state, practical evolutionary forces metaphysically could have engendered the same representational content, but yielded a different degree of reliability as regards veridicality. Reliability is determined by deeply non-accidental but metaphysically contingent connections of the system of production to its actual content-constituting conditions. Individuals rely, in actions and responses, on perceptual states’ and perceptual beliefs’ being veridical. Their psychologies have grades of reliance on or anticipation of the veridicality of these states and beliefs. The numerical subjective probabilities for veridicality postulated by Bayesian models are just numbers put on these grades of reliance or anticipation. Subjective probabilities derive from feedback over eons of evolution. If grades of reliance correspond well to objective probabilities of being right, and those objective probabilities are high, relevant perceptual beliefs are typically warranted. Connections between reliance and objective reliability, in a well-functioning system for producing perceptual states and perceptual beliefs, are the keys to entitlement—not the perceptual states in themselves. I doubt that, in holding that the occurrence of a perceptual experience in itself warrants the belief, a dogmatist can correctly resist the arguments that began this section. At any rate, none of the defenses against them that I have seen seem to me to work. It is important to the present counter-argument, and to understanding the nature of warrant for perceptual belief, that warrant derives from empirical subjective probability assignments that are in place independently of the occurrence of relevant perceptual beliefs. Those assignments are consequences of eons of feedback regarding results (mainly in action and reaction) of sensory interactions with the environment. Dogmatism was right to hold that perceptual beliefs do not need help from

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

further actual or potential beliefs. It was wrong to hold that perceptual beliefs are warranted by nothing more than occurrences of perceptual states (or experiences). White was right to reject dogmatism. That is, he was right to hold that the warrant for perceptual belief does not reside in the perceptual state’s occurrence in itself. He was wrong to apply the criticism to my entitlement account. White, following Wright, thought that support, beyond occurrence of the perception in itself, had to come from background, propositional reasons available to the believer. This view has never been plausible. What I have tried to do is to show in detail how warrant comes from aspects of the perceptual system and perceptual-belief-forming system, and their relations to the environment. The warrant is not present unless one has a perception. But the connections that give the warrant force are independent of whether the perceptual state occurs. They are in place prior to its occurrence. I briefly discuss two issues associated with the account that I have given. The first concerns the point that usually the representational content of a perceptual belief does not exist before the occurrence of the belief. I want to discuss this point in relation to the subjective prior probabilities assigned by individuals on the basis of considerations that antedate the occurrence of a perceptual state. These are the assignments that are normally discussed in the literature on subjective probability and confirmation. The representational contents of specific perceptual beliefs are individuateable only if their occurrent applications are individuateable. An individual or psychological system usually cannot individuate applications in advance of their occurring. So subjective probability assignments typically do not apply to specific propositional contents for perceptual beliefs until they occur. In this sense, there is no subjective prior probability for the approximate truth or radical falsehood of the specific occurrent propositional content—prior to its occurring. This point is rendered less momentous by the point that the subjective prior probabilities for approximate veridicality and radical non-veridicality are set for any and every propositional content of the relevant type, independently of whether it occurs. The individual gains entitlement to accept the propositional content of a specific perceptual belief only when the full content, including applications, occurs, even though the subjective probability that underlies the entitlement is set before the occurrence. Let us consider two cases in light of these points. First, suppose that I know somehow that 80% of the spherical bodies in a large bowl of such bodies are red. I know that the others are green. Suppose that I perceive, by touch, one of the bodies, but do not have a visual perception of its color. Then I can assign a conscious prior subjective probability of .8 to the veridicality of the propositional content: (*)

This₁ body is red.

After one looks at the body and has a perception as of its being red, one assigns a probability for the veridicality of (*) that is considerably higher than .8. The conditional probability of (*) given a visual perception as of the body’s being red is presumably much higher than .8. For the probability of any occurrence of the perceptual state’s (and the visual perceptual belief ’s) being approximately veridical

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



is much higher than .8, and one can register the occurrence of the perceptual state and use this registration to provide confirmation for (*). The intuitions set out by the arguments at the beginning of the section seem to apply. The subjective probability for the veridicality of (*) goes up with the occurrence of the relevant perception. What of the probability of there being a radical illusion as of a red body? The subjective probability for the veridicality of (*illusion) I have a radically illusory perceptual state as of a red body, when I look at that₁ body also goes up, relative to the prior for the schematic version of (*illusion), when the relevant perception as of a red body occurs. For (*illusion) could be falsified both by a perception (veridical or not) as of a green body and by a veridical perception as of a red body. The points already made reapply. There is a prior assignment in the visual system of a very high probability to the approximate veridicality of any occurrence of the relevant perception as of a red body. There is a prior assignment of a very low probability to the radically non-veridicality of any occurrence of the perception. So the rise in subjective probability from the individual’s conscious prior probability assignment to (*illusion) is limited to a level vastly below .5. So (*illusion) is not confirmed, in the sense of warranted. In fact, after the perception, one has a warrant to reject (*illusion). It is perhaps worth noting that (*) is not the content of a perceptual belief. A perceptual belief takes only concepts that are conceptualizations of perceptual attributives, and it takes over all demonstrative-referential applications that are perceptual applications of perceptual attributives that it conceptualizes and incorporates analogs of those perceptual applications in the perceptual belief. (*) does not meet these conditions. (*) is not a possible perceptual-belief content because red is not applied demonstratively in the way that it is in a perceptual belief (and in the underlying perception). That is clear from the example. The representational content does not function to refer perceptually to an instance of red in a de re perceptual application. So (*) does not have the logical form of a perceptual belief. The logical form must have a demonstrative application that red guides. The probability is assigned so as to presuppose that no instance of red is perceived. Hence there is no perceptual demonstrative-like referential application. (*) is confirmed and warranted not just by a perception as of red, but by a perceptual belief that₁ body is redthat2 or by some other empirical belief that traces back to a perceptual belief. I said that ‘typically’ propositional contents of perceptual beliefs cannot be individuated in advance of their occurring. There are probably exceptions.¹²² Let us consider a second case. Suppose again that one feels a spherical body as a spherical body, but does not see it. One might form a belief about the body’s color and have a subjective probability for the attribution of the color. Such a belief will likely lack the representational content of a perceptual belief (as with (*)). For it will likely lack the visual attributives, and applications to instances of the color, that

¹²² I owe the main outlines of this case to Michael Rescorla.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

would occur in a visual perceptual belief. Still, one might form an anticipatory belief that does have the same representational content as an anticipated visual perceptual belief. To have the same content, the anticipatory belief would have to involve only conceptualizations of visual perceptual attributives. And it would have to have the same referential applications as the visual perceptual belief. These would have to depend for their identity and reference on applications in the anticipated perception whose attributives are conceptualized in the belief. Perhaps one could visually imagine the anticipated body, shape, and color shade and bring up the same visual conceptual attributives that later occur in conceptualizations in the content of the visual perceptual belief. One could anticipate the events of de re application to the instances of body, sphericality, and color shade. Perhaps one could individuate the perceptual-belief applications before they occur by knowingly anticipating their occurrence, and by putting oneself in circumstances to cause their occurrence. A subjective probability could then be assigned to this content. The content is that of an anticipation of a perception belief, not yet that of a perceptual belief. The basis for the probability assignment might be in memory, interlocution, or background information. Then one has a subjective prior probability that attaches to that very content, before the perceptual belief is formed. Such a case is rare, but I conjecture possible. The account that I have given applies to this case. Suppose that one has the anticipated visual perception as of the color shade. Suppose that one defeasibly forms the corresponding perceptual belief. Then the subjective probability of there being an occurrence of an approximately true attribution of the color shade will go up. The subjective probability of there being an occurrence of a radically false attribution of the color shade will also go up. The subjective probability of the approximate truth of any occurrence of the propositional content—and correspondingly, the probability of the radical falsehood of any occurrence of that content—that is contributed by the perceptual-beliefforming system are fixed in advance. Assume, that visual perception prima facie outweighs memory, interlocution, and other background information. The prima facie warrant for the approximate truth of the visual perceptual belief ’s content outweighs the warrants for the content when the perception is just anticipated. And the subjective probability assigned to the content after the occurrence of the perception is higher than the assignment to the content when the perception was just anticipated. The subjective probability for the content’s approximate truth therefore increases with the occurrence of the visual perception. Since the subjective probability for any occurrence of the anticipated content’s radical falsehood is capped well below .5, one is warranted after the visual perception not only in believing in the approximate truth of the visual color shade attribution, but also in denying its radical falsehood. If, in the abnormal case in which memory, interlocution, or other background information outweigh perceptual belief, the subjective probability of the representational content’s approximate truth does not change with its vision-based occurrence. The second issue concerns the relation between subjective probability assignments that occur unconsciously in the perceptual or belief-forming systems, and

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



assignments that are attributable to the individual. The latter are commonly elicited by testing how an individual is inclined to bet. Some might worry that unconscious, possibly “sub-personal” (or, to allow for animals, “sub-individual”), assignments are not relevant to epistemology. Probability assignments to occurrences of non-representational sensory registrations are sub-individual. Veridicality assignments to perceptual states and perceptual beliefs are largely unconscious and not deliberately controlled. The anticipations and reliances that the subjective probabilities mark are broadly like an individual’s willingness to bet. Assignments to perceptual states posited in perceptual psychology are more ingrained, perhaps less accessible to conscious estimates, than Bayesian epistemologists tend to represent subjective probability assignments as being. Perceptual psychology’s use of such assignments is to that extent more realistic. When and whether unconscious subjective probability assignments to perceptual states or perceptual beliefs are ascribable to individuals or only to their sub-systems requires further clarification of the individual/sub-individual distinction. I think that the account that I have given does not hinge on how to draw and apply the distinction. Assignments set in psychological sub-systems are subject to normative standards. The degree of reliance on specific perceptual beliefs depends primarily on the degree of reliance on corresponding perceptual states. Degrees of reliance on these states for their approximate veridicality appear to be nearly optimal in a range of central cases. (See notes 54, 104.) The assignments to perceptual states are, in those cases, very close to what, from an epistemic perspective, they should be, in normal conditions. Deviations from strict veridicality mostly result from physical limitations of acuity and noise. Individuals who rely for the strength of their perceptual beliefs on the specific natures of their perceptual states are likely to have degrees of confidence that fit well with objective conditions. Normative epistemology applies to psychological capacities. Discounting the subjective probability assignments that I have discussed would leave epistemology inapplicable to actual psychologies. The arguments that began this section are ingenious. But they illustrate the hazards of doing the epistemology of perceptual belief without reflecting on the psychology of perception. The argument and previous replies to it are unrelated to applications of subjective probability theory in the relevant domain. Epistemology must be better informed by psychology.

2.6 Conclusion Entitlement to perceptual beliefs is the basis for all empirical epistemic warrant. This simple statement is part of a large shift that has been occurring over the last several decades away from much twentieth-century philosophical work on mind and warrant. The shift displaces rational capacities, and consciously accessible aspects of psychologies, from being nearly the sole focus of a philosophical account of mind and knowledge to being just two important factors. The shift forces conscious rational capacities to share focus with largely unconscious psychological sub-systems and

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

their relations to a wider non-psychological environment. Similar displacements have figured in the earlier histories of science and philosophy.¹²³ The shift regarding mind, embodied in anti-individualism, is a return to an older position, initiated by Aristotle and dominant in philosophy until the twentieth century. This view is that the natures of most psychological states are constitutively determined by relations that the individual or the individual’s psychological systems bear to a wider environment. Through causal-interactive patterns, some of which reach over evolutionary time, physical subject matters stamp themselves into the representational natures of psychological systems and psychological states.¹²⁴ Development of the view and good arguments for it were not prominent in the tradition. But the view itself was a majority position in philosophy, until the twentieth century. The shift regarding warrant—and its ancestor, following rules for obtaining scientific cognition—began in late-mid-twentieth century.¹²⁵ The idea that epistemic warrants for an individual are not entirely determined by the natures of and relations among an individual’s psychological states is not a return to a dominant view. At least since the early modern period, epistemic internalism dominated epistemology. To specify the displacements: Anti-individualism shows that representational mind is constitutively embedded in a wider world. The physical environment has a lead role in making perceptual states and empirical beliefs what they are. It does so partly through patterns of causal interactions with sub-representational sensors. These patterns help type-individuate the representational states. They also give such states their subject matters. The environment has a further lead role in yielding epistemic warrant for perceptual beliefs. It contributes in two ways. One is to set the conditions in which reliable veridicality counts toward warrant. The other is to help set a knowledge-inducing and warrant-inducing route back to the subject matter of perception and perceptual belief. So environment-mind connections that constitutively helped determine the representational natures of perceptual states and ¹²³ Obviously, the Copernican, Darwinian, and Einsteinean revolutions all, in various ways, displaced humans and ordinary human experience from a domineering role in determining truths about a subject matter. Parallel changes in philosophy–the separation of philosophy from religion, the recognition of the brute non-rational character of the physical world, the acknowledgment that philosophy is not the queen of the sciences, and the decline of idealism—participate in this large current of displacement. Ironically, Kant appealed to the Copernican revolution as an analog of his regressive defense of a form of idealism about the scientifically cognized world. However, Kant made major contributions to the other three shifts in philosophy just cited. ¹²⁴ See Foundations of Mind, op. cit., essays 4–9 and Origins of Objectivity, op. cit., chapters 1 and 3. Anti-individualism applies, in a different way, to psychological states about mathematics and logic. ¹²⁵ See the authors cited in note 70. W. V. Quine urged that epistemology use scientific knowledge regarding perception and discovery. See, for example, his ‘Epistemology Naturalized’, in Ontological Relativity (New York: Columbia University Press, 1969). He saw himself as suggesting that science replace traditional epistemology. I think that this view was deeply misconceived. He reflected too little on the difference between epistemic norms and the topics of ordinary empirical scientific knowledge. And he did not understand the relevant psychology. Still, he contributed by advocating connecting epistemology with science. After Quine, Alvin Goldman in ‘What is Justified Belief ’, op. cit.. and Epistemology and Cognition, op. cit. and Fred Dretske, Knowledge and the Flow of Information (Cambridge: MIT Press, 1981) played perhaps the primary roles in leading epistemology to connect with science. Dretske, almost like Quine, allows the science to swamp the normative focus of epistemology. And the science that he uses, information theory, is, in my view, too far from the psychology of perception and perceptual belief to undergird epistemology. But his initiative, like Goldman’s, was important.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



perceptual beliefs also help provide a good epistemic route back to the subject matter. They do so when they yield reliably veridical perceptual states and perceptual beliefs. This route helps ground perceptual entitlement and commonly yields knowledgeable perceptual beliefs. These displacements imply that neither the nature of representational mind nor the nature of empirical epistemic warrant can be explicated purely by reference to what goes on in the individual. Neither individuation of mental kinds nor empirical warrant is self-contained. Both constitutively depend on patterns of relations between individuals and the physical environment. These dependencies crucially engage unconscious, non-rational aspects of mind. As indicated, broadly anti-individualistic views about mind were dominant in philosophy before the twentieth-century. The views about warrant were not, at least not since the early modern period. The explanation for this fact is probably more complex than anyone yet understands. I cite some factors. The original and perhaps primary factor was confidence that Euclid’s geometry modeled the structure of all knowledge or scientific-cognition. On the Euclidean model, there is a system of basic beliefs whose natures provide self-support. These beliefs support all other warranted beliefs. The starting points and inferential connections are supposed to be warranted by their natures. The model was often applied to empirical as well as mathematical and philosophical knowledge. In fact, in a persistent distortion, many prominent philosophers in the tradition took epistemically basic empirical beliefs to be approximately as secure from revision as beliefs in simple mathematical truths. Even since this delusion was exploded in mid-twentieth century, many philosophers remain in the grip of the Euclidean model. Most acknowledge that, under sufficient pressure from theoretical considerations, supported by other basic empirical beliefs, any given basic empirical belief can be rejected. But some cling to the idea that basic empirical beliefs are prima facie warranted by their natures. Others take perceptual contents to be intrinsic sources of prima facie warrant and to be analogs of Euclidean premises—except that they are not propositions and their contribution to warrant is prima facie.¹²⁶ Such positions prescribe doing epistemology by reflecting only on epistemic norms and individuals’ mental states.¹²⁷ But reliability relations to the normal environment that are key elements in any empirical warrant are not coded into the natures of perceptual states, perceptual beliefs, or many inductive transitions.

¹²⁶ This analogy is thin. Perceptual states are indeed representational states that cause perceptual beliefs. They are the first representational states relevant to empirical knowledge. They are necessary for entitlement to perceptual beliefs. They bear normative relations to formation of perceptual beliefs. But nothing about the representational content of a perception itself provides any basis at all for the belief-worthiness of the belief. The representational content of a perceptual belief is nearly identical with part of the content of the perceptual state that underlies it, except for being in propositional form. The content’s being perceptual can help explain belief-worthiness, but only if reliable relations to the environment are in place. These relations are not metaphysically determined by the very nature of the perception, as all warranting factors (including reliability) were supposed to be, on the Euclidean model. ¹²⁷ The peculiar view, naive realism, is a special case of this tendency. Like some of the more mainstream views that I have focused on, naive realism is out of step with what is known from the science of perception. For discussion, see ‘Disjunctivism and Cognitive Psychology’, op. cit. and ‘Disjunctivism Again’, op. cit..

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

A second factor is the human tendency to rely too much on familiar perspectives to set the limits of a subject matter. The tendency is strong in philosophy. It tends to hyper-intellectualize such subject matters as mind and knowledge, so that they are directly accessible to philosophical reflection. The focus on knowledge deluxe and insufficient reflection on perceptually based knowledge in young children have obscured how unsophisticated empirical knowledge can be. A third factor is the idea that scepticism must be answered if epistemic warrant is to be obtained. The idea has dominated philosophical thinking since Descartes. Moore’s on-point counter-initiative has been widely decried as obtusely irrelevant. A fourth factor is a tendency to think of epistemic norms as always prescriptive— on an analogy with norms in ethics. Norms are standards for fulfilling functions, goals, or purposes. The most basic epistemic norms do not require any awareness of the norms or any ability to be guided by them. They apply not only to deliberative reasoning but to operations in partly unconscious psychological systems that are not under the control of the knower. A fifth factor is failure to connect epistemic warrant constitutively to being conducive to knowledge and on a good route to truth. This failure arose only in mid-twentieth century. A sixth factor is the practice of offering epistemic accounts of perceptual belief without having a scientifically grounded understanding of perception. This factor has dominated the history of philosophy. It has been less excusable over the last halfcentury, during which science has begun to understand perception in some depth. It is no longer defensible to do epistemology of perceptual belief, or philosophy of perception, without understanding perceptual psychology. A scientific understanding of perception has made several contributions to the present account. I highlight two here. One centers on the relation between veridicality and the nature of perceptual states. The science gives accounts of how both approximately accurate and illusory perceptual states are formed. It individuates perceptual states in terms of their veridicality conditions. It presupposes anti-individualism—which holds that what environmental features perceptual states attribute hinges partly on what environmental features figured prominently in the evolutionary pre-history of individuals with the perceptual systems. On the other hand, we know from biology that evolution does not directly select for accuracy. It selects for fitness to reproduce. Indeed, loss functions that help determine what perceptual states are generated from given stimulation are not set to maximize accuracy. They maximize avoidance of errors that lead to disadvantage or death. Although cases in which perceptual systems are approximately accurate are constitutive to determining the natures of the states, being reliably approximately accurate is not constitutive to the natures of the states. It is metaphysically possible that reliable radical inaccuracy, even in a contentdetermining environment, could yield environmental fitness. So reliable approximate veridicality is not constitutive of perceptual states or perceptual beliefs. Hence being epistemically warranted is not constitutive of perceptual beliefs. A second contribution that a scientific understanding of perception makes is to elicit sources of our entitlements to perceptual beliefs. Anticipations and reliances in perceptual systems are calibrated to the probabilities of distributions of errors.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



This fact shows how well perceptual systems respond to past errors and how reliable they in fact are, at least for a significant range of attributes in many normal circumstances. Actual perceptual states and perceptual beliefs seem almost as reliably veridical, for most types of attribution important to individuals’ survival, as physical limitations allow. These points show in detail how perceptual systems enable perceptual beliefs to track approximate truth and gain approximate knowledge. Such systems monitor upshots of the same types of environment-perceptualsystem connections that formed the attributives in the first place. The anticipatory dispositions and perceptual commitments coded into perceptual systems and perceptual-belief-forming systems are reliable in ways that contribute to entitlement. To be warranted, one need not be able to buttress perceptual beliefs with background thoughts. The source of warrant lies in pre-rational competencies that connect perceptual beliefs reliably to the world. Mindless forces molded our perceptual and perceptual-belief-forming systems and made them epistemically good. Reason is not needed to justify the basic empirical products of representational systems. By fulfilling its very different functions, nature fulfilled some of reason’s epistemic functions. Nature provided good routes to truth and knowledge for creatures with perceptual beliefs. Nature mindlessly formed an epistemic foundation for empirical reason.

Bibliography Alston, W. (1976), ‘Two Types of Foundationalism’, The Journal of Philosophy 73, 165–185. Alston, W. (1983), ‘What is Wrong with Immediate Justification?’, Synthese 55, 73–95. Alston, W. (1989), Epistemic Justification: Essays in the Theory of Knowledge (Ithaca, New York: Cornell University Press). Alston, W. (2005), Beyond “Justification”: Dimensions of Epistemic Evaluation (Ithaca, New York: Cornell University Press). Armstrong, D. (1973), Belief, Truth, and Knowledge (Cambridge: Cambridge University Press). Banks, M., J. Burge, and R. Held (2011), ‘The Statistical Relationship between Depth, Visual Cues, and Human Perception’, in M. Landy ed., Sensory Cue Integration (Oxford: Oxford University Press). Berker, S. (2013a), ‘Epistemic Teleology and the Separateness of Propositions’, Philosophical Review 122 (2013), 337–393. Berker, S. (2013b), ‘The Rejection of Epistemic Consequentialism’, Philosophical Issues 23, 363–387. Blakemore, C. (1970), ‘The Range and Scope of Binocular Depth Discrimination in Man’, Journal of Physiology 211, 599–622. Block, N. (2016), ‘The Anna Karenina Principle and Skepticism about Unconscious Perception’, Philosophy and Phenomenological Research 93, 452–459. BonJour, L. (1980), ‘Externalist Theories of Empirical Knowledge’, Midwest Studies in Philosophy 5, 53–73. BonJour, L. (1985), The Structure of Empirical Knowledge (Cambridge, Mass.: Harvard University Press). BonJour, L. (1998), In Defense of Pure Reason (London: Cambridge University Press). Burge, J. and W. Geisler (2014), ‘Optimal Disparity Estimation in Natural Stereo Images’, Journal of Vision 14, 1–18.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

Burge, J. and W. Geisler (2015), ‘Optimal Speed Estimation in Natural Image Movies Predicts Human Performance’, Nature Communications 6, Article number 7900. Burge, T. (1988), ‘Individualism and Self-Knowledge’, The Journal of Philosophy 85, 649–663, reprinted in T. Burge, Cognition Through Understanding (Oxford: Oxford University Press, 2013). Burge, T. (1997), ‘Interlocution, Perception, and Memory’, Philosophical Studies 86, 21–47. Burge, T. (2000), ‘Frege on Apriority’ in C. Peacocke and P. Boghossian eds., New Essays on the A Priori, (Oxford: Oxford University Press); reprinted in Truth, Thought, Reason: Essays on Frege (Oxford: Oxford University Press, 2005). Burge, T. (2003a), ‘Perceptual Entitlement’, Philosophy and Phenomenological Research 67, 503–548. Burge, T. (2003b), ‘Reply to Stroud’, in M. Hahn and B. Ramberg, Reflections and Replies: Essays on the Philosophy of Tyler Burge (Cambridge, Mass.: MIT Press). Burge, T. (2003c), ‘Reply to Martin Davies’, in M. Frapolli and E. Romero eds., Meaning, Basic Self-Knowledge, and Mind: Essays on Tyler Burge (Stanford: CSLI Publications). Burge, T. (2003d), ‘Logic and Analyticity’, Grazer Philosophische Studien 66, 199–249. Burge, T. (2005a), ‘Disjunctivism and Perceptual Psychology’, Philosophical Topics 33, 1–78. Burge, T. (2005b), Truth, Thought, Reason: Essays on Gottlob Frege (Oxford: Oxford University Press). Burge, T. (2007a), ‘Postscript to “Belief De Re” ’, in Foundations of Mind (Oxford: Oxford University Press). Burge, T. (2007b), ‘Predication and Truth’, The Journal of Philosophy, 104, 580–608. Burge, T. (2007c), Foundations of Mind: Philosophical Essays, Volume II (Oxford: Oxford University Press). Burge, T. (2010), Origins of Objectivity (Oxford: Oxford University Press). Burge, T. (2011a), ‘Epistemic Warrant: Humans and Computers’, in Cognition Through Understanding: Philosophical Essays, Volume III (Oxford: Oxford University Press, 2013). Burge, T. (2011b), ‘Origins of Perception’, Disputatio 4, 1–38. Burge, T. (2011c), ‘Steps Toward Origins of Propositional Thought’, Disputatio 4, 39–67. Burge, T. (2011d), ‘Disjunctivism Again’, Philosophical Explorations 14, 43–80. Burge, T. (2013), Cognition through Understanding: Philosophical Essays, Volume III (Oxford: Oxford University Press). Burge, T. (2014a), ‘Reply to Rescorla and Peacocke: Perceptual Content in Light of Perceptual Constancies and Biological Constraints’, Philosophy and Phenomenological Research 88, 485–501. Burge, T. (2014b), ‘Perception: Where Mind Begins’, Philosophy 89, 385–403. Burge, T. (2014c), ‘Reply to Block: Adaptation and the Upper Border of Perception’, Philosophy and Phenomenological Research 89, 573–583. Camp, E. (2007), ‘Thinking with Maps’, Philosophical Perspectives 21, 145–182. Campbell, J. (2002), Reference and Consciousness (Oxford: Clarendon Press). Casullo, A. (2007), ‘What Is Entitlement?’, Acta Analytica 22, 267–279. Chisholm, R.M. (1966), Theory of Knowledge, 1st edition (Englewood Cliffs, N.J.: Prentice Hall). Chisholm, R.M. (1977), Theory of Knowledge, 2nd edition (Englewood Cliffs, New Jersey: Prentice Hall,). Christensen, M., L. Kristiansen, J. Rowe, and J. Nielsen (2008), ‘Action-blindsight in Healthy Subjects after Transcranial Magnetic Stimulation’, PNAS 105, 1353–1357. Clarke, T. (1972), ‘The Legacy of Skepticism’, The Journal of Philosophy 69, 754–769. Cohen, S. (1984), ‘Justification and Truth’, Philosophical Studies 46, 279–295. Cohen, S. (2000), ‘Contextualism and Skepticism’, Philosophical Issues 10, 94–107.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Cohen, S. (2002), ‘Basic Knowledge and the Problem of Easy Knowledge’, Philosophy and Phenomenological Research 65, 309–329. Cohen, S. (2005), ‘Why Basic Knowledge is Easy Knowledge’, Philosophy and Phenomenological Research 70, 417–430. Cohen, S. (2010), ‘Bootstrapping, Defeasible Reasoning, and A Priori Justification’, Philosophical Perspectives 24, 141–159. Conee, E. and R. Feldman (1998), ‘The Generality Problem for Reliabilism’, Philosophical Studies 89, 1–29. Cruz J., and J. Pollock (2004), ‘The Chimerical Appeal of Epistemic Externalism’, in R. Schantz ed., The Externalist Challenge (Berlin: De Gruyter). Danckert, J. and Y. Rossetti (2005), ‘Blindsight in Action: What Can the Different Sub-Types of Blindsight Tell Us about the Control of Visually Guided Actions?’, Neuroscience & Biobehavioral Reviews 29, 1035–1046. Danto, A. (1968), Analytical Philosophy of Knowledge (Cambridge: Cambridge University Press). D’Antona, D., J. Perry, and W. Geisler (2013), ‘Humans Make Efficient Use of Natural Image Statistics’, Journal of Vision 13, 1–13. Davidson, D. (1983), ‘A Coherence Theory of Truth and Knowledge’ in Subjective, Intersubjective, Objective (Oxford: Clarendon Press, 2001). Davies, M. (2000), ‘Externalism and Armchair Knowledge’, in P. Boghossian and C. Peacocke eds., New Essays on the A Priori (Oxford: Oxford University Press). DeRose, K. (1995), ‘Solving the Skeptical Problem’, The Philosophical Review 104, 1–52. Dickie, I. (2011), ‘Visual Attention Fixes Demonstrative Reference by Eliminating Referential Luck’ in C. Mole, D. Smithies, and W. Wu eds., Attention: Psychological and Philosophical Essays (Oxford: Oxford University Press). Dretske, F. (1981), Knowledge and the Flow of Information (Cambridge: MIT Press). Ernst, M. and M. Banks (2002), ‘Humans Integrate Visual and Haptic Information in a Statistically Optimal Fashion’, Nature 415, 429–433. Foley, R. (1985), ‘What’s Wrong with Reliabilism?’, The Monist 68, 188–202. Frisby, J. and J. Stone (2010), Seeing (Cambridge, Mass.: MIT Press). Fumerton, R. (1995), Meta-epistemology and Skepticism (Totowa, New Jersey: Rowman and Littlefield). Geisler, W. (2003), ‘Ideal Observer Analysis’, in L. Chalupa and J. Werner eds., The Visual Neurosciences (Boston: MIT Press). Geisler, W. (2008), ‘Visual Perception and the Statistical Properties of Natural Scenes’, Annual Review of Psychology 59, 167–192. Geisler, W. and R. Diehl (2003), ‘A Bayesian Approach to the Evolution of Perceptual and Cognitive Systems’, Cognitive Science 27, 379–402. Geisler, W., J. Perry, B. Super, and D. Gallogly (2001), ‘Edge Cooccurrence in Natural Images Predicts Contour Grouping Performance’, Vision Research 41, 711–724. Ghosi, A. (2015), ‘Dimensionality Reduction: A Short Tutorial’, http://www.math.uwaterloo. ca/~aghodsib/courses/f06stat890/readings/tutorial_stat890.pdf (sourced August 2015). Ginet, C. (1975), Knowledge, Perception, and Memory (Dordrecht: D. Reidel). Goldman, A. (1979), ‘What is Justified Belief?’, in G. Pappas ed., Justification and Knowledge (Dordrecht: Reidel). Goldman, A. (1986), Epistemology and Cognition (Cambridge, Mass.: Harvard University Press). Goldman, A. (1988), ‘Strong and Weak Justification’, Philosophical Perspectives 2, 51–69.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

Goldman, A. (2001), ‘The Unity of the Epistemic Virtues’, in A. Fairweather and L. Zagzebski eds., Virtue Epistemology: Essays on Epistemic Virtue and Responsibility (Oxford: Oxford University Press). Goldman, A. (2008), ‘Immediate Justification and Process Reliabilism’, in Q. Smith ed., Epistemology: New Essays (Oxford: Oxford University Press). Goldman, A. ‘Reliabilism’, Stanford Encyclopedia of Philosophy, online. Greco, J. (2006), ‘Virtue, Luck and the Pyrrhonian Problematic’, Philosophical Studies, 130, 9–34. Hawthorne, J. (2004), Knowledge and Lotteries (Oxford: Oxford University Press). Huemer, M. (2001), Skepticism and the Veil of Perception (Lanham, Md.: Rowman & Littlefield). Huemer, M. (2006), ‘Phenomenal Conservatism and the Internalist Intuition’, American Philosophical Quarterly 43, 147–158. Kung, P. (2010), ‘On Having No Reason: Dogmatism and Bayesian Confirmation’, Synthese 177, 1–17. Lewis, D. (1996), ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549–567. Luthra, Y. (2013), Embodied Rational Agency, UCLA Dissertation, (2013), chapter 6–retrieved from https://escholarship.org/uc/item/4q92g4w6. Majors, B. (2015), ‘What Entitlement Is,’ Acta Analytica, 30, 363–87. Maloney, L. and P. Mamassian (2009), ‘Bayesian Decision Theory as a Model of Human Perception: Testing Bayesian Transfer’, Visual Neuroscience 26, 147–155. Moon, A. (2012), ‘Three Forms of Internalism and the New Evil Demon Problem’, Episteme 9, 345–360. Moore, G.E. (1939), ‘Proof of an External World’, Proceedings of the British Academy 25, 273–300. Moore, G.E. (1953), Some Main Problems of Philosophy (London: George Allen & Unwin Ltd). Moore, G.E. (1993), ‘Moore’s Paradox’, in T. Baldwin ed., G. E. Moore: Selected Writings (New York: Routledge). Moretti, L. (2015), ‘In Defense of Dogmatism’, Philosophical Studies 172, 261–282. Nozick, R. (1981), Philosophical Explanations (Cambridge, Mass.: Harvard University Press). Peacocke, C. (2004), The Realm of Reason (Oxford: Oxford University Press). Peacocke, C. (2015), ‘Magnitudes: Metaphysics, Explanation, and Perception’, in D. MoyalSharrock, V. Munz, and A. Coliva eds., Mind, Language and Action: Proceedings of the 36th International Wittgenstein Symposium (Berlin: de Gruyter). Perenin, M. and Y. Rossetti (1996), ‘Grasping Without Form Discrimination in a Hemianopic Field’, Neuroreport 7, 793–797. Philips, I. (2018), ‘Unconscious Perception Reconsidered’, Analytic Philosophy 59, 471–514. Philips, I. and N. Block (2016), ‘‘Unconscious Seeing’—A Debate’, in B. Nanay ed., Current Controversies in Philosophy of Perception (London: Routledge). Plantinga, A. (1993), Warrant: The Current Debate (Oxford: Oxford University Press). Pollock, J. (1974), Knowledge and Justification (Princeton: Princeton University Press). Pollock, J. (1984), ‘Reliability and Justified Belief ’, Canadian Journal of Philosophy 14, 103–114. Pollock, J. and J. Cruz (1999), Contemporary Theories of Knowledge (Lanham, Maryland: Rowman and Littlefield). Pryor, J. (2000), ‘The Skeptic and the Dogmatist’, Nous 34, 517–549. Pryor, J. (2001), ‘Highlights of Recent Epistemology’, British Journal of the Philosophy of Science 52, 95–124. Pryor, J. (2004), ‘What is Wrong with Moore’s Argument?’ Philosophical Issues 14, 349–378.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Pryor, J. (2005), ‘There is Immediate Justification’ in M. Steup and E. Sosa eds., Contemporary Debates in Epistemology (Oxford: Blackwell). Quine, W.V. (1960), Word and Object (Cambridge, Mass.: MIT Press). Quine, W.V. (1969), ‘Epistemology Naturalized’, in Ontological Relativity (New York: Columbia University Press). Quine, W.V. and J. Ullian (1970), The Web of Belief (New York: Random House). Quinton, A. (1973), The Nature of Things (London: Routledge & Kegan Paul). Reid, T. (2002), Essays on the Intellectual Powers of Man (1785) (Edinburgh.: Edinburgh University Press). Rescorla, M. (2015), ‘Bayesian Perceptual Psychology’, in M. Matthen ed., The Oxford Handbook of the Philosophy of Perception (Oxford: Oxford University Press). Rescorla, M. (2016), ‘Bayesian Sensorimotor Psychology’, Mind and Language 31, 3–36. Scanlon, T. (1998), What We Owe Each Other (Cambridge, Mass.: Harvard University Press). Scanlon, T. (2014), Being Realistic About Reasons (Oxford: Oxford University Press). Sellars, W. (1963), ‘Empiricism and the Philosophy of Mind’ in Science, Perception, and Reality (London, Routledge & Kegan Paul). Shadmehr, R. and S. Mussa-Ivaldi (2012), Biological Learning and Control (Cambridge, Mass.: MIT Press). Silins, N. (2008), ‘Basic Justification and the Moorean Response to the Skeptic’, in T. Gendler and J. Hawthorne eds., Oxford Studies in Epistemology 2, 108–142. Silins, N. (2012), ‘Explaining Perceptual Entitlement’, Erkenntnis 76, 243–261. Smithies, D. (2011), ‘What is the Role of Consciousness in Demonstrative Thought?’, The Journal of Philosophy 108, 5–34. Sosa, E. (1994), ‘Philosophical Skepticism and Externalist Epistemology’, Proceedings of the Aristotelian Society, Supplementary Volume 68, 263–290. Sosa, E. (1997), ‘How to Resolve the Pyrrhonian Problematic: A Lesson from Descartes’, Philosophical Studies 85, 229–249. Sosa, E. (2007), A Virtue Epistemology (Oxford: Clarendon Press). Sosa, E. (2009), Reflective Knowledge (Oxford: Clarendon Press). Sosa, E. (2011), Knowing Full Well (Princeton: Princeton University Press). Sperandio, I., and P. Chouinard (2015), ‘The Mechanisms of Size Constancy’ Multisensory Research 28, 253–283. Steup, M. (2004), ‘Internalist Reliabilism’, Philosophical Issues 14, 403–425. Stroud, B. (1984), The Significance of Philosophical Scepticism (Oxford: Clarendon Press). Stroud, B. (1989), ‘Understanding Human Knowledge in General’ in M. Clay and K. Lehrer eds., Knowledge and Skepticism (Boulder: Westview). Stroud, B. (1994), ‘Scepticism, “Externalism”, and the Goal of Epistemology’, Proceedings of the Aristotelian Society, Supplementary Volumes 68, 291–307. Vogel, J. (1990), ‘Cartesian Skepticism and Inference to the Best Explanation’, Journal of Philosophy 87, 658–666. Vogel, J. (2000), ‘Reliablism Leveled’, The Journal of Philosophy 97, 602–623. Weatherson, B. (2007), ‘The Bayesian and the Dogmatist’, Proceedings of the Aristotelian Society 107, 169–185. Weatherson, B. (2014), ‘Easy Knowledge and Other Epistemic Virtues’, online http://brian. weatherson.org/ekoev.pdf,%20sourced%202014. Wedgewood, R. (2002), ‘Internalism Explained’, Philosophy and Phenomenological Research 65, 349–369. Wedgewood, R. (2011), ‘Primitively Rational Belief-Forming Processes’, in A. Reisner and A. Steglich-Petersen eds., Reasons for Belief (Cambridge: Cambridge University Press).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



:    

White, R. (2006), ‘Problems for Dogmatism’, Philosophical Studies 131, 525–557. Whitwell, R., C. Streimer, D. Nicole, and M. Goodale (2011), ‘Grasping the Non-conscious: Preserved Grip Scaling to Unseen Objects for Immediate but not Delayed Grasping Following a Unilateral Lesion to Primary Visual Cortex’, Vision Research 51, 908–924. Whitwell, R., A. Milner, C. Cavina-Pratesi, M. Barat, and M. Goodale (2015a), ‘Patient DF’s Visual Brain in Action’, Vision Research 110, 265–276. Whitwell, R., I. Sperandio, G. Buckingham, P. Chouinard, and M. Goodale (2015b), ‘Evidence for a Functional and Anatomical Dissociation in the Use of Size Constancy for Perceptual Report and Goal-directed Grasping’ (abstract), Journal of Vision 15, 187. Williamson, T. (2001), Knowledge and Its Limits (Oxford: Oxford University Press). Wolpert, D. and M. Landy (2012), ‘Motor Control is Decision-Making’, Current Opinion in Neurobiology 22, 1–8. Wright, C. (2002), ‘(Anti)-Sceptics Simple and Subtle: Moore and McDowell’, Philosophy and Phenomenological Research 65, 330–48. Wright, C. (2004), ‘Warrant for Nothing (and Foundations for Free)?’, Aristotelian Society, Supplementary Volume 78, 167–2012. Wright, C. (2007), ‘The Perils of Dogmatism’, in S. Nuccetelli and G. Seay eds., Themes from G. E. Moore: New Essays in Epistemology and Ethics (Oxford: Oxford University Press). Yuille, A. and H. Bulthoff (1996), ‘Bayesian Decision Theory and Psychophysics’ in D. Knill and W. Richards eds., Perception as Bayesian Inference (Cambridge: Cambridge University Press). Zardini, E. (2014), ‘Confirming the Less Likely, Discovering the Unknown’, in D. Dodd and E. Zardini eds., Scepticism and Perceptual Justification (Oxford, Oxford University Press).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

3 Perceptual Entitlement and Skepticism Anthony Brueckner and Jon Altschul

3.1 Anti-individualism and Veridicality: A Priori vs. Empirical Burge distinguishes between two kinds of epistemic warrant: entitlement and justification. Entitlement is epistemically externalist, in that it is a form of warrant which need not be fully accessible, even on reflection, to the entitled subject S. S may even lack the concepts which are required to think the content that formulates the warrant. Justification, by contrast, is “conceptually accessible,” on reflection, to S. Burge’s (2003) main goal in “Perceptual Entitlement” is to elucidate the connection between anti-individualism about mental content and perceptual entitlement. The notion of perceptual entitlement which Burge is seeking to elucidate attaches to perceptual beliefs—beliefs which are formed on the basis of perceptual experiences. For Burge, the relation between a perceptual experience and an associated perceptual belief is not like the relation between a reason for belief and a belief. Reasons have propositional content, and perceptual experiences do not. When someone believes that there is fire on the basis of the belief that there is smoke and where there is smoke there is fire, this reason-based belief is justified. Things are different in the case of perceptual belief, according to Burge. When someone believes that there is a cat in her lap on the basis of a perceptual experience of a cat, she has an epistemic entitlement to rely upon the non-propositional perceptual representation in forming the belief. What is the source of this sort of perceptual entitlement? How do perceptual states contribute to perceptual entitlement? Burge’s answer is that the thesis of antiindividualism about perceptual content explains how perceptual beliefs can possess perceptual entitlement on the basis of associated perceptual representations. In “Content Externalism, Entitlement, and Reasons,” one of us (Brueckner) presented a reconstruction of an argument thought to be discernible in Burge’s “Perceptual Entitlement.”¹ It was supposed to be an a priori argument for the existence of perceptual entitlement. Though it was not touted as such, the argument ¹ See Brueckner (2007) and Burge (2003a). Thanks to Mikkel Gerken and Tamar Weber. Anthony Brueckner and Jon Altschul, Perceptual Entitlement and Skepticism In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0003

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



   

would be a kind of a priori anti-skeptical argument. For if it could be demonstrated a priori that we have entitlement to hold our perceptual beliefs, then this would answer a skeptic who claimed that our perceptual beliefs have no positive epistemic status, given our inability to rule out his skeptical counterpossibilities in a manner that itself possesses positive epistemic status. Before considering that reconstructed argument, let us look at two seemingly related theses: (1) Anti-individualism about perceptual content (2) The veridicality thesis We take it that the following is a formulation of (1), or at least of an important consequence of (1): A condition of particular representational states’ having the content that they have is that there have been both causal-formative interactions (which are not in themselves representational) and representationally successful interactions between instances of types of relevant perceptual referents and aspects of the individual’s perceptual system (in either the individual’s history, or in evolution of the system in his evolutionary ancestors, or in some other way).²

The veridicality thesis (2) can be formulated in this way: an individual could presume with near immunity from error that some individual or other has (at some time) perceived instances of some of the perceptual kinds or types that his perceptual system represents things as having.³

An earlier stronger formulation of (2) is this: most perceptual representations . . . represent what, in some complex sense of ‘normally’, they normally stem from and are applied to.⁴

Here is the argument for the existence of perceptual entitlement that was reconstructed from “Perceptual Entitlement” (it is assumed that the premise should be read as being knowable a priori): (I) Anti-individualism holds for perceptual content. (Assumption) (II) An individual’s perceptual state types are reliably veridical in his perceptual system’s normal environment. (I) (III) This reliable veridicality is explained by the nature of the perceptual states. (I) (IV) If (i) an individual’s perceptual state types are reliably veridical in his perceptual system’s normal environment, and (ii) this reliable veridicality is explained by the nature of the perceptual states, then the individual, absent reasons for doubt, is entitled to hold beliefs appropriately based on the states. (Assumption) (V) Perceptual entitlement exists. [(II),(III),(IV)] Let us note one key feature of Burge’s overall position. Suppose that S is landed in an abnormal environment that is quite different from the normal environment

² See Burge (2003a: 531).

³ See Burge (2003b).

⁴ See Burge (1986).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    



“by reference to which the nature of [his] . . . perceptual states is explained” (Burge 2003: 33). Suppose that S’s perceptual state types are not reliably veridical in the abnormal environment. Still, Burge wants to maintain that S has an entitlement to form perceptual beliefs on the basis of his experiences while in the abnormal, unfriendly environment. There is conclusive reason to believe that the reconstructed argument (which is criticized in a paper by Frank Barel (2012), to be discussed later) is not faithful to Burge’s intentions, though we will discuss in this section some connected questions about the logic of Burge’s overall position.⁵ Burge rejects premise (II) as being too strong—just think of the rabbit and his “Danger!” perceptual content, which has predominantly false positive tokens. Burge says, “I don’t see why an explanation of perceptual content could not get started that took important accuracies to form the basis of an explanation even though they are in a minority—and this might be so, as far as I can see, for all of a poor creatures perceptual categories.”⁶ More importantly, Burge did not intend in the first place to give an a priori argument for the existence of perceptual entitlement. He assumes in “Perceptual Entitlement” that we are entitled to our perceptual beliefs and tries to understand the structure and nature of the entitlement. So the basic intended position in “Perceptual Entitlement” could hardly be brought to bear against a skeptic about the existence of perceptual entitlement.⁷ Let us return to (1) and (2)—anti-individualism about perceptual content and the veridicality thesis. The starting point of the reconstructed argument was that on Burge’s view, (1) can be known a priori and that therefore (2) (at least in its weaker formulation) can also be known a priori. The main criticism of that argument centered around premise (IV): why should reliable veridicality of perceptual states that is grounded in their very nature give rise to perceptual entitlement? Burge is clearly not a pure reliabilist who connects reliability (a truth-relevant notion) with positive epistemic status. What of significance is added to the mix if reliable veridicality flows from the very nature of the perceptual states? Never mind whether that criticism was worth making (since Burge does not in fact endorse any a priori argument for the existence of perceptual entitlement, and since it is doubtful that his views yield the resources for such an argument). However, we would like to revisit the question of whether (1) and (2)— which Burge holds to be true—are knowable a priori. To begin with, Burge says that “anti-individualism is supposed to be asserting necessary truths.”⁸ He continues, “our having attitudes with the relevant contents about the world necessarily derives from some interaction between individuals with attitudes and the world.”⁹ Presumably, the same necessity is supposed to attach to (1) and (2). But these remarks are of course compatible with the view that the necessities in question are only knowable a posteriori and therefore of no anti-skeptical utility. But why deny, as Burge seems inclined to do, the a prioricity of (1) and (2)? Aren’t these closely related claims ⁵ ⁶ ⁷ ⁸

Our reason is personal communication with Burge. From email correspondence. See Barel (2012). See also Burge (2003a: 513). Thanks to Burge here, in email correspondence. See also Burge (2003a: 537). See Burge (2003b: 339). ⁹ See Burge (2003b: 339); our emphasis.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



   

established entirely by philosophical reflection on thought experiments? If so, our access to the truth of the claims would seem to be a priori. Indeed, regarding the veridicality thesis, Burge has said that it is a priori that perceptual content depends upon some interactions with instances of some perceptual categories at some time or other.¹⁰ We will return to this. Let us back-track a bit, to Twin Earth. One can run through the Twin Earth thought experiment even if one acknowledges that one may well be in a Cartesian Evil Genius world with no water, no twater, and no physical objects whatsoever. One can nevertheless compare in thought two worlds replete with superficially indistinguishable but chemically distinct liquids (H2O vs. XYZ) and then come to the antiindividualist judgement that the pertinent individualistic twins’ ‘water’-thoughts differ in content. So it seems that one can know a priori that the individualistic properties shared by the twins can be held fixed while their content-properties vary. One can know this anti-individualist thesis without making any empirical assumptions about what sort of world one is in fact in. One would think that the same sort of reasoning applies to (1), anti-individualism about perceptual content. The weaker statement of (2) formulated in the first relevant quote above could then be deduced from (1). But see Burge: the precise epistemic status of anti-individualism . . . is not obvious. It seems to begin by reflecting on the relation between mental states and a physical or social environment that is simply assumed to exist.¹¹

But the actual existence of such a relation need not be assumed in order to appreciate the force of anti-individualist thought experiments. That is the point of the foregoing Evil Genius/Twin Earth discussion. Still, Burge has been reticent about whether anti-individualism can provide a non-question-begging answer to skepticism since his 1988 “Individualism and Self-Knowledge” (and his unpublished 1993 Locke Lectures).¹² In the 1988 paper, Burge says in an often-quoted footnote (largely about a paper of one of the authors (Brueckner)), “there is no easy argument against skepticism from anti-individualism and authoritative self-knowledge.”¹³ We would like to discount a reason that Burge gives for his pessimism about the anti-skeptical power of anti-individualism. Burge proceeds from the assumption that skepticism is about knowledge or justification. However, he says: Neither anti-individualism nor the slogan that error presupposes veridicality says a word about knowledge or justification. . . . Anti-individualism is a metaphysical view about the nature of certain mental states—what having such states necessarily presupposes.¹⁴

Fair enough to a degree: skepticism is a view about our lack of knowledge, or of justification. But it seems that an answer to skepticism need not establish that we are ¹⁰ In email correspondence. ¹¹ See Burge (2003b: 339). ¹² See Burge (1988). Burge refers to his Locke Lectures in his reply to Stroud at n. 4. ¹³ See Burge (1988: 655, n. 6). Burge there discusses the criticism of Hilary Putnam’s anti-skeptical strategy in Brueckner (1986), reprinted in Brueckner (2010). ¹⁴ See Burge (2003b: 338).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    



justified in our external-world beliefs. It seems that it would be enough for an answer to block a skeptical argument purporting to establish that we are not so justified. We will return to this point later.¹⁵ We agree with Barel (2012) that premise (II) in the reconstructed argument is stronger than anything Burge is committed to knowing a priori: (II) An individual’s perceptual state types are reliably veridical in his perceptual system’s normal environment. Barel has read Burge correctly on this point. Nevertheless, the weaker (1) and (2) can plausibly be held to be knowable a priori. We could extract from (1) and (2) the following a priori knowable kernel: (K) Perceptual content depends upon there being some interactions with some instances of some perceptual types at some time or other, and for there to be any perceptual representation, there must be some veridical perceptual representation (at some time or other).¹⁶ Note that (K) does not say that one’s own possession of perceptual content depends upon one’s own interactions with instances of my own perceptual types. It is a weaker kernel than that, allowing for the possibility that one’s conspecifics are suitably related to their environment though one is not (maybe one is a brain in a vat). Notwithstanding this, (K) is prima facie sufficient to block some important forms of skepticism, such as Evil Genius based skepticism and radical Putnamian BIV based skepticism. A BIV never has any ordinary perceptual contact with anything in its world, in contrast to a creature who begins as a normal embodied human and is later rendered into his envatted condition. In the Putnamian BIV scenario, all sentient creatures are BIVs. The skeptic specifies counterpossibilities to our external-world knowledgeclaims (Evil Genius and BIV) and argues that since we cannot rule them out as non-actual, our knowledge-claims are thus seen to be false.¹⁷ According to the skeptical counterpossibilities used to generate the pertinent skeptical arguments, we have states bearing perceptual content and yet there have been no interactions between us and instances of my perceptual types (and none between anyone else and their perceptual type instances), and yet there is perceptual representation in the Evil Genius and vat worlds even though there is no veridical perceptual representation. In contrast to the reconstructed argument that was criticized by Burge and Barel, the currently proposed way of wielding perceptual anti-individualism against the foregoing forms of skepticism does not involve a proof of the existence of perceptual

¹⁵ In personal communication, Burge has said that a really powerful and satisfying a priori answer to the skeptic should deliver some such conclusion as that many of my external-world beliefs possess a positive epistemic status, such as entitlement. But he is pessimistic about the possibility of such an answer that is non-question-begging. ¹⁶ Thanks here to Tyler Burge, personal communication. He does not straightforwardly endorse the a priori knowability of (K), since the content-forming interactions that figure in (K) might well not themselves be representational and hence might not themselves be veridical. ¹⁷ An appropriate closure principle for knowledge is required, connecting knowledge that P with the ability to knowledgeably rule out propositions that are known to be counterpossibilities to P: propositions that are logically possible and yet incompatible with P.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



   

entitlement. Instead, the anti-skeptical maneuver is “metaphysical” in nature: it works by denying the possibility of certain skeptical hypotheses (Evil Genius and BIV) that are used to launch the skeptical arguments. These alleged possibilities are incompatible with (K) (the incompatibility arising from the idea that error presupposes at least some veridicality). There are some worries about this anti-skeptical maneuver, even if we grant the a prioricity of (K). First, one might worry that we cannot know a priori that our states indeed have representational content (BIV and Evil Genius victims are standardly thought to have contentful states—but maybe this is not possible). Second, one might worry that we cannot know a priori that the representational content they have is as of physical entities—maybe the content in question is as of non-physical, mental entities.¹⁸ We will return to these worries below.

3.2 Barel’s Rationalistic Anti-skeptical Argument Barel would, we think, be sympathetic to the spirit of some of the foregoing ideas, as he himself attempts to construct an anti-skeptical argument from perceptual antiindividualist materials. But we think that his so-called Rationalistic argument is problematic. The argument proceeds from the premise that the thought that I have purported representations of objects and kinds in the world (P) is “prima facie intelligible” to oneself. But it is hard to see how we can get any anti-skeptical mileage out of this premise, which just about anyone who has thought about skepticism has felt constrained to attribute to the skeptic. How could the skeptic even attempt to show that we do not know that the representations that P concerns are not known by us to be veridical if the skeptic does not agree that P is intelligible? Otherwise what could the skeptic and the anti-skeptic even be arguing about? Which representations are we talking about? This is to say, in effect, that we agree with Barel that the worries raised at the end of the previous section may well be extravagant. Barel’s basic move is to argue from (A) the intelligibility-to-me of P and my ability to reason about P, to (B) the conclusion that this rational ability regarding P has a “rational source” that yields “true presentations-as-true.” Though Barel does not say so in the course of his argument, he indicates later that P itself is among the true presentations-as-true in question. But Barel gives no explanation of how to get from (A) to (B). It seems to be a sort of classic verificationist move: we begin with the intelligibility or meaningfulness of a thought (better, of a sentence seeming to express a thought) and then move to a conclusion about the knowability of the truth of the thought. But to make such a move, we need a general principle connecting meaningfulness with knowability of truth (or with knowability of truth or falsity). As Stroud (1968) pointed out in his classic paper “Transcendental Arguments,” appeal to such a principle would render any fancy anti-skeptical argumentation superfluous.¹⁹ A verification principle linking meaningfulness and intelligibility, on ¹⁸ See Brueckner (2010: 87–106) for similar worries about anti-individualist-based transcendental arguments. Burge has also expressed these worries in personal communication and, we think, in the Locke lectures. ¹⁹ See Stroud (1968).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

    



the one hand, with knowability of truth (say, under the right sensory conditions), on the other, would itself be sufficient to argue from P’s intelligibility to its truth, regardless of any anti-individualist considerations. So although we are sympathetic to Barel’s aim of using perceptual anti-individualism in arguing against skepticism, we think that his Rationalistic argument is problematic.

3.3 Concluding meditation Assume that perceptual entitlement exists, as per Burge’s (2003) procedure in “Perceptual Entitlement”. What explains its existence, Burge asks? “Anti-individualism about perceptual content” appears to be the root of the answer. But if such antiindividualism (APC) is knowable a priori and constitutes an entailing explanation for the existence of perceptual entitlement (PE) then a closure principle for a priori knowledge would underwrite the following inference (where ‘KAPφ’ stands for ‘φ is knowable a priori’): (A) KAP(APC) (B) KAP (APC entails that there is PE) (C) KAP(There is PE) [(A),(B),KAP closure] But Burge denies that his views afford the a priori knowledge in (C). In order to avoid (C), Burge can deny (A) or (B) or both. He accepts, (A), as it turns out, but denies the entailment in (B).²⁰ On his view, if S is entitled to his perceptual belief that φ in virtue of having an associated perceptual representation E, then other tokens of E’s type must reliably give rise to true φ-like beliefs. Since APC says nothing about the matter-of-fact existence of such reliability, APC cannot entail that there is perceptual entitlement. To expand a bit, for Burge there are two conditions that must be met in order for a perceptual belief that Fa to have entitlement. First, the individuation-conditions tied to the state-type Fa are anti-individualistic. Second, the state-type Fa is reliably veridical in the normal content-forming environment.²¹ Let us grant that the first condition can be known a priori to hold. The problem is that there is no way to know a priori that the second condition holds, even given the a priori knowledge just supposed. The thesis of anti-individualism simply makes no claim, nor needs to, about whether perceivers are in fact reliable in certain circumstances. Burge does say that when the states are reliable, this reliability is explained by the anti-individualistic nature of the states. But this does not imply that anti-individualistically individuated perceptual states must be related to some or other reliably veridical state-types that share the pertinent content. In the end, the most that the anti-skeptic can hope for in his deployment of APC is the rather slender kernel K that we have considered. Returning to the worries expressed by Burge in the first section of this chapter, consider

²⁰ In personal communication. ²¹ See Burge (2003a: 531–2). So anti-individualism by itself does not explain the existence of perceptual entitlement.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



   

(#) There are some things that exist apart from my mind (or were at some point things that existed apart from my mind) Does (K) yield a priori knowledge of (#)? We think that the answer may well be “Yes”, though Burge would not second the affirmative answer. We think that one may well know a priori that some of one’s perceptual states are contentful and that some of these states represent there to be things apart from one’s mind (though knowledge of # is not sufficient to yield knowledge of the nature of the things). Whether the successfully represented representata (there must be some, according to (K)) are, e.g., cats or instead computer states (as in the Putnamian BIV scenario) is not knowable a priori. As against this, the following troubling questions resurface. First, why couldn’t there be thoroughgoing “illusions of content”: states that phenomenally seem to be representing an F but are in fact representing nothing? This, however, is an inapt expression of the pertinent worry. The worry cannot be that one could have a mistaken thought as follows: I have contentful perceptual states. For if one had such a thought, then it would follow that one has some contentful mental states. In that case, it would seem odd to hold that none of one’s perceptual mental states are content-bearing. Second, suppose that some of one’s perceptual states do have representational content, as it seems that they do. If it is suggested that these states do not represent there to be physical things apart from one’s mind, then how do they represent the world to be? Do they represent there to be mental contents that are all alike one’s own? In particular, does one’s current perceptual state (in fact caused by a mind-independent cat in my lap) represent itself (as opposed to representing there to be a cat in one’s lap)? This would seem to be implausible. So in the end, plausible, minimal assumptions about a priori access to one’s contents and their general representational character seem to yield a priori knowledge of (#) above, in the presence of (K).

References Barel, F. 2012. Perceptual Entitlement, Reliability, and Scepticism. International Journal for the Study of Skepticism 2: 21–43. Brueckner, A. 1986. Brains in a Vat. Journal of Philosophy 83: 148–67. Brueckner, A. 2007. Content Externalism, Entitlement, and Reasons. In Internalism and Externalism in Semantics and Epistemology, ed. S. Goldberg, 160–76. Oxford: Oxford University Press. Brueckner, A. 2010. Essays on Skepticism. New York: Oxford University Press. Burge, T. 2003a. Perceptual Entitlement. Philosophy and Phenomenological Research 67: 503–48. Burge, T. 2003b. Some Reflections on Scepticism: Reply to Stroud. In Reflections and Replies: Essays on the Philosophy of Tyler Burge, eds. M. Hahn and B. Ramberg, 335–46. Cambridge, MA: MIT Press. Burge, T. 1986. Cartesian Error and the Objectivity of Perception. In Subject, Thought, and Context, eds. J. McDowell and P. Pettit, 117–36. Oxford: Oxford University Press. Burge, T. 1988. Individualism and Self-Knowledge. Journal of Philosophy 85: 649–63. Stroud, B. 1968. Transcendental Arguments. Journal of Philosophy 65: 241–56.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

4 Epistemic Entitlement—Its Scope and Limits Mikkel Gerken

On the basis of a pluralist background, I argue that the notion of entitlement is indispensable in epistemological theorizing. Hence, I argue against monist internalist views according to which there is no room for externalist notions of warrant—i.e., entitlement. However, I also argue that the explanatory scope of entitlement is limited. Hence, I also argue contra monist externalist views that the internalist notion of justification is indispensable in epistemological theorizing. In particular, I will argue that the notion of justification plays a significant role in areas of epistemology that are concerned with social cognition. Once epistemic pluralism is assumed, the internalism–externalism dispute in epistemology may be replaced by the question: How to draw the entitlement– justification distinction? I argue that the “traditional” accessibility criteria may be seen as distinguishing between subspecies of justification. But I argue that no accessibility criterion draws the most fundamental distinction between entitlement and justification. Consequently, I set forth an alternative criterion for distinguishing between internalist and externalists species of warrant: The Reason Criterion. On this basis, I partly characterize entitlement by articulating some minimal cognitive requirements on it. Finally, I take stock of the characterized notion of entitlement by considering its scope and limits in individual and social cognition.

4.1 Introduction This chapter explores epistemic entitlement in relation to the epistemic internalism– externalism dispute. I have three specific aims and each of them contributes to an overarching one. My first aim is to situate epistemic entitlement in an epistemically pluralist framework. Epistemic pluralism is the view that there are both internalist and externalist species of epistemic rationality. Following Burge’s terminology, I use ‘warrant’ to denote the genus of epistemic rationality that harbors internalist species—labeled ‘justifications’ and externalist ones—labeled ‘entitlements’ Mikkel Gerken, Epistemic Entitlement—Its Scope and Limits In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0004

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

(Burge 2003, 2013).¹ Given epistemic pluralism, the internalism–externalism dispute in epistemology may be replaced by the question: How to draw the entitlementjustification distinction? My second aim is to answer this question by a new way of drawing distinction between justification and entitlement. Consequently, I set for a Reason Criterion according to which justification and entitlement are characterized as follows: Reason Criterion (Justification) S’s warrant, W, for her belief that p is a justification if and only if W constitutively depends, for its warranting force, on the competent exercise of S’s faculty of reason. Reason Criterion (Entitlement) S’s warrant, W, for her belief that p is an entitlement if and only if W does not constitutively depend, for its warranting force, on the competent exercise of S’s faculty of reason. Although this Reason Criterion is a novel way of articulating the justificationentitlement distinction, I will argue that it classifies cases better than accessibility criteria which are best seen as characterizing subspecies of justification. My third aim is to partly characterize entitlement by articulating by considering its scope and limits in individual and social cognition. Thus, my overarching aim is, on the one hand, to argue against monist internalist views that leave no room for externalist notions of warrant—i.e., entitlement. On the other hand, I argue against monist externalist views according to entitlement can replace internalists notions of warrant—i.e., justification. I have given more direct arguments for epistemic pluralism with regard to testimonial warrant and the kinds of warrant required for assertion (Gerken 2012, 2013b). But since epistemic pluralism is a structural assumption, it may be motivated abductively by considering its explanatory advantages in epistemological theorizing. This will be my approach here. I proceed as follows: In the remainder of Section 4.1, I briefly consider entitlement in relation to the traditional internalism-externalism dispute and its contemporary incarnations. In Section 4.2, I consider the case for pluralism in the light of a number of important cases and raise the problem of how to draw the entitlement-justification distinction. In Section 4.3, I criticize a prominent way of drawing the distinction between internalist and externalist theories of warrant—namely, an accessibilist criterion. I argue that while such a criterion may distinguish between subspecies of justification, it is inapt to distinguish justification from entitlement. ¹ The term ‘entitlement’ is used in other ways as well. See for example (Wright 2004; Smith, this volume; and, in particular, the editors’ introduction to the present volume). Likewise, the term ‘warrant’ is used in multiple ways. See, for example (Plantinga 1993; Wright 2004). The term ‘justification’ is used in an even wider array of more or less technical senses often corresponding more closely the broad sense in which I use ‘warrant.’ These terminological discrepancies are irksome but unavoidable. To narrow down the discussion somewhat, I restrict it to doxastic warrant which, as opposed to propositional warrant, accrues to entertained beliefs and similar doxastic attitudes.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



In Section 4.4, I set forth an alternative way of drawing the justificationentitlement justification distinction: The Reason Criterion. On this basis, I reinforce the case for epistemic pluralism. In Section 4.5, I consider the scope and limits of epistemic entitlement by indicating the relevance of justification theorizing about both individualistic and social cognition. In Section 4.6, I conclude.

4.1.1 Entitlement and the tradition As mentioned, I use the phrase ‘epistemic entitlement’ to denote an externalist species of warrant. Thus, the notion of entitlement may be seen as emerging from Dretske’s and Goldman’s developments of externalist theories of knowledge and justification (Dretske 1970, 1971; Goldman 1967, 1976, 1979). As a first approximation, externalism challenges the traditional focus on the perspective of the individual’s faculty of reason. In consequence, epistemic externalism amounts to an ambitious reorientation of epistemology. Indeed, the externalist turn in epistemology has sometimes been thought of as an externalist revolution that dispenses with internalist notions of warrant altogether. Some such epistemic externalists are of a broadly naturalist orientation (Quine 1969; Kornblith 2002). Naturalistically inclined externalists may regard the notion of entitlement as pivotal in a broader externalist revolution of epistemology that is supposed to free us from internalist preconceptions that have generated more problems than they have resolved. Radical externalist revolutionaries may be inclined to de-psychologize epistemic rationality. But few if any epistemologists regard the perspective of the cognizing subject as entirely irrelevant for an account of epistemic rationality. At any rate, I assume that a truth-connection that is non-lucky in virtue of cognitive competence is a constitutive feature of all types of warrant (Gerken 2013a, 2018a). Thus, entitlement also requires the exercise of a cognitive competence. The brand of epistemic externalism that I prefer has it that the truthconduciveness of the cognitive competence is assessed relative to its normal environment where this is constrained by the environment that partly individuates the subject’s mental state types (Burge 2003; Sawyer and Majors 2005; Graham 2012; Gerken 2013a). In consequence, epistemic externalism is unified with a version of externalism in the philosophy of mind. However, here my main aim is not to develop this brand of entitlement in further detail but to argue for pluralism and to provide a novel way of distinguishing entitlement from justification. Given that both entitlement and justification involve a cognitive competence, it is natural to distinguish these two kinds of warrant by the nature of the competence in question. Indeed, this is another way to integrate epistemology with the philosophy of mind. So, I will propose that the cognitive competence associated with justification is the faculty of reason whereas this is not the case for entitlement. Understood broadly, as denoting externalist warrant, the notion of entitlement has established itself as a key epistemological concept. However, its significance for epistemological theorizing cannot be sufficiently appreciated without reflection on its proper scope and limits.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

4.1.2 Entitlement and justification—contemporary debates The contemporary internalism–externalism debates in epistemology have at times presupposed monism about warrant. More often than not, monist commitments are presupposed rather than argued for.² In some cases, the presupposition of warrant monism is rationally required as the ‘tacit premise’ of an enthymematic argument against internalism or externalism. In other cases, an explicitly monist internalism or externalism is taken to be established by positive arguments that appear to be perfectly compatible with pluralism. In yet other cases, alternative species of epistemic warrant are officially accepted but regarded as inferior or ultimately irrelevant. For example, epistemic internalists may suggest that epistemic externalists “change the subject” (Fumerton 1995; Stroud 1994; Pollock and Cruz 1999). Epistemic externalists, in turn, sometimes take internalist approaches to be expressions of misguided traditional preconceptions that must be overcome (Goldman 1979, 1999b; Kornblith 2002). The lack of arguments against epistemic pluralism indicates a lack of explicit articulation and discussion of it. One source of monism about epistemic warrant is the internalism–externalism dispute in the theory of knowledge. This dispute has revolved around the question of whether S can know that p without meeting some internalist requirement (typically accessibility). Externalists say answer in the affirmative and internalists answer in the negative (Kornblith 2008). Given that the question about the necessary conditions of knowledge is a yes/no-question, it may suggest monism about knowledge. (However, it does not entail it. One can be a pluralist about knowledge by recognizing both internalist and externalist species of knowledge.) The presupposition of monism in the theory of knowledge may well have led to monist presuppositions in the theory of warrant as well. But this is a mistake since the question as to whether there are one or more genuine kinds of epistemic warrant is different from questions concerning the necessary conditions for knowledge. In consequence, I will argue for pluralism in the theory warrant in abstraction from the theory of knowledge.³ Although epistemic monism has at times been presupposed in trench wars between internalists and externalists, the ranks of epistemic pluralists are growing—at least among those who have explicitly considered the view in relation to the internalism/externalism debate. For example, theorists as diverse as Burge, Dretske, Chalmers, and Sosa subscribe to some form of epistemic pluralism (Burge 2003; Dretske 2000; Chalmers 2011; Sosa 2007). To reinforce such a pluralism, I will indicate areas in which an epistemology that only invokes entitlement will be inadequate. That is, I will argue that human cognition, and in particular social cognition, cannot be understood in terms of entitlement alone. Although the development of epistemic externalism (i.e., entitlement) should be regarded as philosophical progress, it is a supplement to rather than a replacement of internalist notions of

² For the relevant notion of presupposition, see Gerken 2013a Chap II. ³ This approach exemplifies my general skepticism about a “knowledge-first” program in epistemology (Williamson 2000, 2013; Gerken 2011, 2017, 2018a, 2018b Forthcoming a.).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



justification. Before I am in a position to argue for these critical points, a more constructive approach is required.

4.2 Pluralism about Warrant The most generic rationale for epistemic pluralism is that a belief can enjoy a positive epistemic status that can be classified as either internalist or externalist. If that’s the case, we need a theory of each of these epistemic statuses. So, pluralism about epistemic statuses that a belief might enjoy motivates pluralism in the theory of warrant. Such a pluralism may be either exclusive or inclusive. According to exclusive epistemic pluralism, justification and entitlement are compatible in the sense that a token belief may possess exactly one of those species of warrant. But the two species are mutually exclusive in the sense that a token belief cannot simultaneously be entitled and justified. In contrast, inclusive epistemic pluralism allows the two types of warrant to be co-instantiated: a token belief that p may be entitled or justified or both. I’ll argue for inclusive pluralism about most types of warrant.⁴ In order to outline and motivate the epistemic pluralism about warrant that I am adopting, I will first consider some cases of entitlement without justification and then turn to the historically important case of Descartes’ cogito (Descartes 1641/1996).

4.2.1 Entitlement without justification The simplest cases of entitlement without justification are basic perceptual beliefs held by unreflective creatures such as animals and young children. Their perceptual competences are truth-conducive in the normal environment. But animals and young children do not have the ability to conceptualize such facts about epistemic reasons. Thus, their perceptual beliefs are not supported by the faculty of reason and do not depend on it for their warranting force. According to the Reason Criterion, then, the relevant perceptual beliefs are entitled but not justified. This is the diagnosis that an epistemic pluralist should want. A similar rationale may be given for epistemic pluralism about memory and testimony (for the latter, see Gerken 2013b). More complex cases have also been set forth. For example, the case of chicken sexing has been suggested as a case of in which chicken sexers acquire a reliable method that is inaccessible to them (Brandom 1998; Pritchard 2003). The case is illustrative (although it is not altogether empirically realistic). Likewise, there are ⁴ A general account is complicated by the fact that a monism or pluralism may be domain-specific. It is consistent to uphold, for instance, externalist monism about perceptual warrant and pluralism about testimonial warrant. A further complication for the inclusive/exclusive distinction between versions of epistemic pluralism arises for certain cases. Assume, for the sake of discussion, that S can be both justified in and entitled to the belief that theorem T follows from a set of axioms, A. S’s belief is entitled when she forms a reliable intuitive judgment that T follows from A. It is justified when she consciously carries out the proof that T follows from A. However, it is not clear whether the justification in this case “swamps” the entitlement. If it does, we should be exclusive epistemic pluralists about this type of judgment. If the entitlement is merely complemented by the justification, we should be inclusive epistemic pluralists. The case is difficult. I mention it not to adjudicate on it but for illustration of the point that the inclusive/ exclusive distinction is non-trivial (see also Gerken 2013b).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

cases of innate cognitive abilities with an inaccessible basis, such as proprioception. However, in such cases, the individual may have access to her track-record. Finally, there are complicated cases of idiot savants (Treffert 1988). The class of cases suggests entitlement without justification. However, although the complex cases have sometimes figured more prominently in the internalism–entitlement debates, they are much harder to diagnose. One complex case that has figured very prominently is Bonjour’s much-discussed clairvoyant case: CLAIRVOYANT Norman, under certain conditions that usually obtain, is a completely reliable clairvoyant with respect to certain kinds of subject matter. He possesses no evidence or reasons of any kind for or against the general possibility of such a cognitive power or for or against the thesis that he possesses it. One day Norman comes to believe that the President is in New York City, though he has no evidence either for or against this belief. In fact the belief is true and results from his clairvoyant power under circumstances in which it is completely reliable (Bonjour 1985: 41). Bonjour sets forth the case as a counterexample to epistemic externalism (especially reliabilist versions). Thus conceived, the case illustrates the monist presupposition that I mentioned above. Reflection on the case suggests that Norman’s beliefs lack an important epistemic property. Epistemic internalists often take it to be accessibility to the truth-conducive faculty (Bonjour 1985). But even if it is granted that Norman’s belief lack an epistemically important property, a cogent anti-externalist argument furthermore requires monism about warrant. For Norman’s belief also possess some important epistemic properties such as reliability. So, it cannot be assumed without argument that Norman isn’t (given an appropriate specification of the case) entitled to his president-belief although he is not justified in it. The parenthetical point in the previous sentence is important because it is in fact very hard to give a specification of the clairvoyant case such that Norman is entitled. Many externalists do not regard reliable belief-generation as sufficient for entitlement (cf. Burge 2003; Lyons 2009; Graham 2012). Graham, for instance, requires that Norman’s inaccessible cognitive power is a reliable cognitive competence with the etiological function of forming true beliefs. I agree that the origin of the clairvoyant capacity and its integration in Norman’s cognitive life matters for a final assessment. Further, it should be stipulated that Norman has forgotten his successful track-record of beliefs in the relevant domain. Otherwise, he would be in a position to warrant his clairvoyancegenerated beliefs by induction or abduction. Finally, Bonjour’s case is unfortunate insofar as the actual non-existence of a reliable clairvoyant faculty may influence our judgments about the beliefs of the protagonist in the thought experiment.⁵ Discussions of Bonjour’s ingenious thought experiment have suffered from the fact that the case is

⁵ In particular, our judgments may be hampered by a hindsight bias (Nagel 2008, 2010 argues that attributions of knowledge are prone to hindsight bias although she does not discuss the case at hand).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



insufficiently described in important respects. Consequently, it is not the epistemic pluralist’s best candidate case of entitlement without justification. Rather, it is worth mentioning because it clearly illustrates the unarticulated internalist monist presupposition of Bonjour’s argument.

4.2.2 Understanding and the case of self-verifying thoughts An externalist monist might object to the present approach by arguing that entitlement swamps justification. The objection may be cast as a dilemma: Either justification is not truth-conducive and, hence, not a genuine form of epistemic warrant or any increase in justification can be fully accounted for in an externalist manner—e.g., in terms of reliability. To counter this objection by addressing the second horn of the dilemma, I will consider the Cartesian cogito. I will argue that this is a case in which the thinker is maximally entitled but in which she may nevertheless improve her epistemic position by acquiring justification. The issue calls for a treatment of its own. So, here I will just state my basic idea in abstraction from some of the complex issues that it raises. Consider the Cartesian cogito—viz. the thought I am thinking or what Burge calls a “cogito-like” thought such as I am hereby thinking that water is wet (Burge 1988).⁶ These thoughts are self-verifying (Audi 1999). If someone is thinking one of them, she is ipso facto guaranteed to think something true. Competent cogito-like thoughts are maximally reliable in virtue of their self-verifying structure. They are what I call well-functioning infallible—optimal performance in thinking the thought entails that the thought meets its cognitive goal of truth (Gerken 2013a: 79ff ).⁷ However, it is psychologically possible to competently think cogito-like thoughts without realizing that they are of a self-verifying structure. The cognitive requirements on competently thinking the thoughts fall short of the cognitive requirements for understanding their self-verifying structure and, thereby, understanding the thoughts’ particular infallibility. It is no great cognitive achievement to think a cogito-like thought. But it was a significant cognitive achievement to understand its self-verifying nature (in case of the Cartesian cogito) or structure (in case of Burge’s cogito-like thoughts). However, if a thinker competently thinks the cogito, then she is maximally entitled to the thought I am thinking. Since it is guaranteed to be true, its reliability cannot be increased. Yet, the thinker may still improve her epistemic position vis-à-vis the thought by understanding the self-verifying nature. So, the thinker can improve on an epistemic dimension that is not tracked by degree of entitlement. This motivates for the idea that there is an epistemic species of warrant—justification by understanding—that is distinct from the epistemic entitlement that all competent thinkers of cogito thoughts enjoy. So, epistemic pluralism about warrant may be motivated by reflection on warrant by understanding and the cogito. As mentioned, the issue is very complex and requires an investigation of its own. For now I rest content with setting forth the basic idea. ⁶ I am using underlining to mention thoughts and their constituents such as concepts. ⁷ I leave open whether cases of incompetent thinking may render the thought fallible. It is, at the very least, hard to come up with such cases. I compare the notion of well-functioning fallibility to Burge notion of immunity from brute error in (Gerken 2013, Sect. II.viii.a: 78–81).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

4.2.3 Epistemic pluralism in epistemological theorizing The arguments for epistemic pluralism above can be supplemented by arguments pertaining to more specific cases. For example, I have provided more specific arguments in the epistemology of testimony (Gerken 2013b). But, as mentioned, it is hard to provide conclusive, non-question-begging arguments for a view that amounts to a cornerstone in a theoretical edifice. Although I have sought to motivate epistemic pluralism, its tenability will ultimately be decided by reflection on whether it earns its explanatory keep in epistemological theorizing. So, I will, for the present purpose, remain content with the present motivation for epistemic pluralism. However, I hope that developing the view will augment it by exemplifying some of the ways in which it enriches epistemological explanations of otherwise problematic explananda.

4.2.4 How to draw the entitlement–justification distinction? Given epistemic pluralism, the pivotal question becomes that of drawing the entitlement–justification distinction. Standardly, a criterion of demarcation is provided via an internalist restriction—i.e., a restriction on justification (i.e., internalist warrant). Externalism, then, is standardly defined as the negation of the internalist claim that only the factors or conditions which meet the restriction contribute to warrant. For example, in an encyclopedia article, William Alston thinks of externalism as follows: “We may think of externalism as simply the denial of this restriction” (Alston 1998).⁸ This modus operandi does not entail that justification is explanatorily primary to entitlement. In contrast, it requires that entitlement receives an independent positive characterization. A positive characterization of entitlement is central to the drawing of the distinction if it specifies the truth-conducive factors that may or may not meet the internalist restriction. Consider, for concretization of this abstract point, the following simplistic model: Entitlement is articulated fully in terms of reliability of a belief-generating competence and justification is articulated fully in terms of accessibility to it. Given such a model, entitlement is explanatorily primary to justification since justification requires that the individual has access to the basis of the entitlement. Assuming pluralism, then, the standard modus operandi does not entail or even suggest that entitlement is a derivative species of warrant. Given this qualification, the standard modus operandi articulated by Alston may be used to draw an internalist–externalist distinction. But it does not answer the substantive question of how to draw it. Specifically, it does not answer how to characterize the internalist restriction. In the next two sections, I’ll consider two answers to this question. First I criticize accessibilist criteria (Section 4.3). Then I set forth the Reason Criterion as an alternative (Section 4.4).

4.3 Accessibility as an Internalist–Externalist Criterion The most common way of specifying an internalist constraint by which the internalism–externalism distinction is drawn is called ‘accessibilism.’ The view is, ⁸ Alston discusses accessibilist versions of the internalist restriction.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



roughly, that justification requires cognitive access to one’s epistemic reasons. Thus Bonjour (in an encyclopedia article): The most generally accepted account . . . is that a theory is internalist if and only if it requires all the factors needed for a belief to be epistemically justified for a given person be cognitively accessible to that person . . . a strong version of internalism would require that the believer actually be aware of the justifying factors in order to be justified; while a weaker version would only require that he be capable of becoming aware of them . . . (Bonjour 1992: 132ff)

Monist internalists shy away from an overly strong internalist requirement. Consequently, a weaker modal variety of the requirement is typically upheld. Accessibilist criteria alone do not specify what sort of factors must be accessed for a belief to be epistemically justified. Furthermore, an articulation of the criterion must involve a specification of the notion of accessibility. Typically, it is required that cognitive accessibility involves first-person methods (Bonjour 1992; Burge 1993, 2003; Fumerton 1995). But sometimes accessibilism is articulated in terms of higher-order justification (Smithies 2012, 2015) and the literature contains further notions (see Pappas 2014 for a survey). But before turning to the varieties of epistemic access, it will be worth noting some properties of accessibilist approaches that will continue to be important even if accessibilist criteria are not the best way to draw the entitlement–justification (externalism–internalism) distinction.

4.3.1 Degrees of access Warrant generally comes in degrees and this is something that theories of entitlement as well as theories of justification must explain. Roughly, degrees of entitlement may be characterized in terms of degree of truth-conduciveness of the type of cognitive competence involved. An accessibility criterion provides a dimension in terms of which degrees of justification may be partly specified. One thing that matters for a belief ’s degree of justification is, roughly, the degree to which an agent has access to the factors that matter for the truth-conduciveness of it. But accessibilists should hold that the specification of degree of justification is determined only partly by the degree of accessibility. This is because the factors that are accessible remain partial determiners of justification. Access to the fact that I am reliable does not provide as strong justification as similar access to the fact that I am extremely reliable. So, accessibilists should assume S’s degree of justification may depend on both matters pertaining to degree of reliability and degree of accessibility to the degree of reliability in question. Consequently, a belief may be only minimally justified if it is based on a barely accessible but highly reliable mode of belief generation. But it will also possess a low degree of justification if it is based on a highly reliable mode of belief formation that is barely accessible. Further qualifications and distinctions are required. For example, it is unclear whether access justification is always based on entitlement or whether a belief may be justified without being entitled. It may even be that some subject matters are such that beliefs about them can only be justified. These are intriguing issues that arise for accessibilists once the pluralist stance is adopted. Here I merely suggest that in an epistemically pluralist framework, the degree of access justification can be determined by at least two parameters: The truthconduciveness of the basis of a subject’s belief and the extent to this basis are

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

accessible to the believer. There is a caveat, however. The fact that justification may be accessible to a limited degree suggests that it makes sense to talk about accessible and inaccessible justification. This, in turn, suggests that justification cannot be given an exhaustive characterization of justification in terms of accessibility.

4.3.2 Kinds of access and subspecies of justification: Accessing, articulating and arguing An accessibility approach does not merely allow for degrees of accessibility. It also allows for different kinds of cognitive accessibility. Important kinds of accessibility may mark subspecies of justification. The fact that there are various kinds of accessibility and that some of them reflect distinctive cognitive competences, may indicate that ‘cognitive accessibility’ denotes a phenomenon that is too disparate to constitute the most explanatorily fundamental criterion of justification. I take a criterion to be explanatorily fundamental if it can characterize justification in a principled manner that explains other candidate criteria as merely characterizing subspecies of justification.⁹ But it is not clear that any notion of accessibility is apt to do so. Assume, for illustration, that one kind of accessibility consists in the ability to phenomenologically discern one’s operative cognitive competences whereas another kind of accessibility consists in the ability to conceptualize certain epistemically relevant factors. We may also distinguish between the ability to merely conceptualize these factors and the ability to conceptualize them as epistemic reasons. We may furthermore distinguish between the ability to generate warranted beliefs about the epistemically relevant factors by first-person methods and the ability to do so by being taught. While each of these different kinds of accessibility might mark important properties of justification, it is not clear why either one should be privileged as the internalist restriction that yields an externalist–internalist criterion. So, it is correspondingly unclear why any accessibilist criterion is the most explanatorily fundamental way to draw the entitlement–justification distinction. I return to this critical point below. Here I will focus on a positive lesson. Even if accessibility is not the defining aspect of justification, it remains epistemically important. For example, different subspecies of justification may be distinguished between by reference to different kinds of accessibility. Since I am mainly concerned with considering accessibility as a criterion for the justification–entitlement distinction, I restrict the discussion to a brief illustration of this approach to taxonomizing subspecies of justification. In previous work on the epistemic norms of assertive speech acts, I have suggested that S may, in some conversational contexts, meet the epistemic conditions on asserting that p by being entitled to a belief that p (Gerken 2012, 2018a). However, I also suggested that in many conversational contexts, entitlement is insufficient. In particular, this is so in discursive contexts. Roughly, discursive conversational ⁹ A comparison here is Evans’ attempt to draw the distinction to draw the Type 1/Type 2 distinction in dual process theory in a principled manner. In doing so Evans’ argues that his working memory criterion is more fundamental than other proposals (fast/slow, conscious/unconscious, effortless/effortful etc.) because it explains why the different types of cognition has these features (Evans 2008. See Evans and Stanovich 2013 for a development).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



contexts are those in which speakers are reasonably presupposed to be sensitive to reasons for and against assertions. In fact, there are discursive contexts in which not even accessibilist justification is enough. In some discursive contexts it is required that S can articulate reasons in support of her assertion. Assertion within a scientific collaboration is an important example (Gerken 2015). On the basis of these assumptions, I characterize a species of justification—discursive justification—as follows: Discursive Justification S’s warrant for believing that p is a discursive justification iff S is able to articulate some epistemic reasons for believing that p (Gerken 2012: 385). I do not mention this approach to reiterate the case for my proposed norm of assertion. Here my aim is just to indicate how kinds of accessibility may fruitfully be seen as distinguishing between subspecies of justification. Reflection on norms of assertion suggests a useful distinction between access justification requiring access to reasons and discursive justification that requires the ability to articulate them.¹⁰ I’m inclined to suspect that discursive justification requires access justification but not vice versa. I doubt that the mere ability to articulate an epistemic reason is sufficient for discursive justification. The epistemic reason has to be discursively relevant. But the issue requires detailed discussion. My key proposal here is that accessibilist criteria may serve to distinguish between explanatorily important subspecies of justification. Importantly, however, this approach requires assuming that a particular subspecies of justification is inaccessible in some senses but not in others (inaccessible by introspection, inaccessible to linguistic articulation and so on). This, in turn, suggests distinguishing between accessible and inaccessible justification makes sense. If so, there is a category of inaccessible justification. But if there is a general category of inaccessible justification, the central lesson should be this: While accessibilist criteria are apt to distinguish between subspecies of justification, no accessibilist criterion captures what distinguish justification from entitlement. More specifically, I will argue that there are two senses in which accessibilist criteria are less fundamental than the Reason Criterion that I will set forth. First, the accessibilist criteria will misclassify important subsets of cases that the Reason Criterion can classify in a principled manner (Sections 4.3.3–4.3.4). Second, the Reason Criterion is explanatory more fundamental in virtue of characterizing accessibilist justification as a (prominent) subspecies of justification (Section 4.4).

4.3.3 Challenges for accessibilist criteria The fact that there are various kinds of accessibility of epistemological significance suggests that accessibility is inapt as a general criterion to draw the internalism– externalism distinction. The fact that an accessibility criterion does not speak to the base or source of warrant suggests the same thing. I will make these suggestions more ¹⁰ This is in contrast to, for example, Graham who uses “ ‘justification’ and ‘justifications’ for the various beliefs one offers or would offer when asked why one’s belief is likely to be true” (Graham 2012: 454). Thus, Graham’s otherwise congenial approach does not distinguish justification from what I take to be a subspecies of it—namely, discursive justification.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

concrete by considering two specific challenges for accessibilist criteria. The problem of non-access (Section 4.3.3.1) and the challenge from inaccessible reasoning (Section 4.3.3.2). ...    - The problem of non-access is a problem for weak accessibilists according to which the modal notion of accessibility is the characteristic feature of justification. Assume that there are factors that are epistemically relevant for the belief that p that are accessible to Sandra and that she accesses those factors. Given that my argument is structural we can leave it open what the factors and the accessing amounts to. Compare Sandra to Sally to whom a type-identical set of epistemically relevant factors are equally accessible. But Sally fails to access any of these factors. Thus, the epistemically relevant factors are merely accessible to Sally. Compare, finally, Sally to Samson to whom the epistemically relevant factors in question are not accessible. However, Samson generates the belief that p via cognitive processes that are type-identical to those that Sally exercises. Let us assume that the type cognitive process that generates Sally’s and Samson’s respective beliefs that p are reliable in a manner that suffices for entitlement. Thus, Sally and Samson are on a par regarding their operative cognitive states and processes. They differ only in that Sally fails to access the accessible epistemically relevant factors whereas Samson does not access them because they are not accessible to him. Sally and Sandra, in contrast, differ regarding their operative cognitive processes, whereas they are on a par regarding the epistemically relevant factors that are accessible to them. Of course, Samson and Sandra differ regarding both accessibility and operative cognitive processes. The question is now this: Is Sally justified like Sandra or is she merely entitled like Samson? Is the relevant feature the access to reasons that play no role in her formation and maintenance of the belief? Or is it the nature of the cognitive competence that is in fact operative? According to the weak accessibilist, we should regard Sally as justified (like Sandra) rather than merely entitled as (like Samson) because of the non-operational difference in accessibility between her and Samson. I think this is problematic. The challenge for this view is to explain why a difference in mere accessibility marks a principled epistemic difference for the belief given that it makes no difference as to how it is generated and sustained. One answer is that someone who has accessible but inoperative reasons for her belief that p is epistemically better off than one who does not because the belief is more robust in a range of counterfactual circumstances. I agree that this modal property of belief is epistemically significant (Gerken 2013b and below in Sections 4.5.1–4.5.2). Yet, it seems wrongheaded to regard it as the explaining the nature of doxastic justification that Sally’s belief enjoys. We can capture the significance of the modal property by saying that Sally is entitled like Samson although she is, contrary to Samson, in a position to become justified like Sandra. A mathematician who has access to a proof of a theorem, T, in the sense that she could prove T if she spent time and effort on it but who does not do so, is not justified in believing that there is a proof of T (Gerken 2011). The nature of the cognitive

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



processes that are in fact operative in the generation and sustainment of the belief in question are far more pertinent to assessing whether Sally is justified or merely entitled. At least, in the case in which there is a conflict between considerations of accessibility and operative processes, the latter seems more important to determine the epistemic status as one of justification or entitlement. ...      The challenge above may be supplemented with a purported case of inaccessible justification. Such a case amounts to a counterexample to accessibilist criteria for the entitlement–justification distinction. The case is one of unconscious reasoning. Assume that S is holding a set of fully conceptualized beliefs, U, that are not only unconscious but inaccessible to consciousness. (Let your Freudian fantasies rip!) Although S is incapable for bringing the beliefs in U to the level of consciousness, we may assume that they are nevertheless governed by reason. For example, we may assume that S will unconsciously detect inconsistencies in U and unconsciously revise the belief set in a truth-conducive manner. Likewise, she will generate new subconscious beliefs if they are clearly entailed by U. Furthermore, beliefs in U will figure as premises in reasoning that from the perspective of consciousness appears to be enthymematic.¹¹ Assuming that such a case is psychologically possible, it appears to be one in which certain beliefs are not merely entitled but justified in virtue of being warranted by unconscious reasoning. The beliefs are warranted by the exercise of a sophisticated high-level propositional competence that is ascribable to S. To say that such beliefs are merely entitled fails to respect the significance of the difference between truth-conducive support by reasoning and by sub-propositional cognitive competences, respectively. A case that does not concern a set of already held beliefs may be a case of abductive reasoning. Surely S may be a competent abductive reasoner within some domain without having any representation of the relevant abductive inference rules. Even those who study reasoning have a hard time articulating abductive inference rules. But furthermore, it seems psychologically possible that even though the premisebeliefs of someone’s reasoning are unconscious and inaccessible, she may, thereby, become warranted in the conclusion-belief. However, although the premise-beliefs are inaccessible, they operate as epistemic reasons. Moreover, given that S is in fact competently following an abductive inference rule, she is competently exercising a reasoning competence (for discussion, see Gerken 2013a: ch. II). I think these facts speak strongly in favor of regarding S as justified in the conclusion-belief. Of course, it must be admitted that a belief justified by inaccessible reasoning lacks some epistemologically significant properties. However, this fact may be accounted for by noting that such a belief only enjoys justification but not access justification or discursive justification. Thus, the case of inaccessible reasoning supports the diagnosis that accessibility marks important distinctions between subspecies of justification but not the distinction between justification and entitlement. ¹¹ On one account enthymematic reasoning simply is reasoning in which some premise-beliefs are unconscious (Gerken 2013a). But an account of enthymematic reasoning is not required for the purpose of the present case.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

4.3.4 Accessibilist criteria in conclusion The challenges above are inconclusive considered as counterexamples to an accessibilist way of drawing the justification–entitlement distinction. The distinction is at the core of a broad epistemological framework which should not be adopted or abandoned on counterexamples alone. Nevertheless, the challenges indicate that although accessibility marks important epistemological distinctions, may not mark the justification-entitlement distinction. So, I will consider another criterion.

4.4 The Faculty of Reason as the Internalism–Externalism Criterion? The considerations above provide some hints for an alternative entitlementjustification criterion.¹² They suggest that the nature of the operative cognitive competences that constitute a warrant for a belief is central to determining whether the warrant is an entitlement or a justification. Entitlements derive from the exercise of sub-propositional cognitive competences. Justifications, in turn, derive from sophisticated propositional competences that are traditionally conceived of as instantiating faculties of reason. Consequently, the entitlement-justification distinction may be regarded as derivative from a fundamental distinction in the philosophy of mind: The distinction between the faculty of reason and more primitive cognitive competencies. I will mainly consider epistemic reasoning and understanding as sources of warrant by reason. Given that reasoning, and perhaps even understanding, may ¹² Warning: Most readers will want to skip this overly nostalgic footnote. My 2000 BA Thesis at the University of Copenhagen concerned externalist responses to skepticism. So, when I arrived at UCLA, as a visiting undergraduate in 2000, I was already defending epistemic pluralism. Between 2000 and 2004 I wrote several sophomorically entitled term papers on the topic. One (‘Gone Mental’) targeted Conee and Feldman’s mentalist criterion (Conee and Feldman 1998), another (‘Access Denied’) targeted accessibility criteria and yet another (‘A Reasonable Externalism’) set forth a psychological constraint on entitlement. Finally, in my UCLA Proposition (Gerken Ms 2005) I argued for the following criterion (in slightly different notation): A warrant-determiner contributes to the justification for S’s belief that p iff it is a proposition, r, and S uses r in reasoning which supports S’s belief that p. The 2004 criterion excludes warrant by understanding as internalist warrant unless understanding is a brand of reasoning. I recall that Burge criticized the proposal on these grounds. After 2001 my thinking was influenced by Burge’s work although I argued against the accessibility criterion that he favored at the time: “Justifications . . . involve reasons that people have access to” (Burge 1993: 230). “Justification is warrant by reason that is conceptually accessible on reflection to the warranted individual” (Burge 2003: 505). Burge now draws the entitlement–justification distinction differently: “As I now use the terms, justification consists in warrants by reasons; an individual has a justification only if a relevant reason is present in an individual’s psychology; entitlements are warrants that do not consist, even partly, in reasons . . . ” (Burge 2013: 28). I don’t know whether my early attempts had influence on Burge’s turn from an accessibility criterion to a brand of a reason criterion. Thanks to UCLA’s ethos favoring oral over printed philosophy, I never sought to publish the early material. But descendants of it figure in later articles. For example, the suggestion that an accessibility criterion indicates the underlying idea that justification is warrant “by reason” figures in (Gerken 2013b, n. 3). This resembles Burge’s current formulation that I discuss in the main text. Speaking of the main text, I should really get back to it before the present ego-trip gets out of line.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



occur unconsciously, in what sense is it epistemically internal? What is the difference between, for example, unconscious epistemic reasoning and the computations of a properly functioning perceptional system? A clue to an answer is provided by the familiar distinction between unconscious reasoning and subconscious computation. The distinction between justification and entitlement is an internal–external distinction because the capacity of reasoning— even unconsciousness reasoning—is attributable to the person. In contrast, it is a sub-system which computes non-propositional representations. The epistemological internal-external distinction is in some respects analogous to the distinction between eating and digesting. It is me who eats the cake, but it is my digestive system which digests it. Of course, my stomach is part of my body and hence in one sense ‘internal.’ But there is another sense in which it is ‘external.’ It is a sub-system which operates independently. The internal-external distinction in epistemology may be understood in broad analogy.

4.4.1 The Reason Criterion The fact that one of S’s belief-generating systems is truth-conducive in her normal environment is not her reason for belief.¹³ However, we might label such a fact which contributes to the entitlement an epistemic ground. So, epistemic grounds for belief need not be attributable to the believer. In contrast, reasoning is attributable to the reasoning individual. Even when unconscious, reasoning is a distinctively internal, propositional mental activity that requires the exercise of fairly sophisticated cognitive competences. So, let us consider the following Reason Criterion (where ‘reason’ denotes sort of high-level cognitive competence that is operative in reasoning and understanding). Reason Criterion (Justification) S’s warrant, W, for her belief that p is a justification if and only if W constitutively depends, for its warranting force, on the competent exercise of S’s faculty of reason. We may say that meeting the Reason Criterion provides basic justification in the following sense: S may be justified by the competent exercise of her faculty of reason (i.e., her cognitive competence) although S does not have a reason (i.e., an attitude or content lending epistemic support). The point is not merely that there is no discursive justification that would convince someone who doubted the proposition in question (Gerken 2012). The point is that a principle of logic or metaphysics may be basic in the sense that it is impossible to provide a reason for it that provides a justification that is stronger than the justification obtained by understanding the principle. Someone may have such basic justification without possessing access justification or discursive justification. In accordance with the adopted modus operandi, entitlement may be negatively characterized in a derivative manner: ¹³ Of course, S may have beliefs about the reliability of her belief-generating system. But this does not render the proper functioning of a belief-generating system an epistemic reason. Every (epistemically relevant) fact may become a reason by being the subject of belief.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

Reason Criterion (Entitlement) S’s warrant, W, for her belief that p is an entitlement if and only if W does not constitutively depend, for its warranting force, on the competent exercise of S’s faculty of reason. Although the Reason Criterion is novel, it is not an especially revisionist one. For example, the Reason Criterion classifies cases that have been thought to mark the internalist–externalist distinction in a standard manner. For example, epistemic pluralists may by the Reason Criterion classify an appropriately specified case of child and animal cognition as involving entitlement but not justification. As mentioned, the case of Norman the clairvoyant is considerably harder. But if Norman is to be regarded as warranted at all, he is to be regarded as entitled and not justified. However, if Norman competently reasons about his successful track record, then inclusive epistemic pluralists may classify Norman’s belief as enjoying both entitlement and justification. In contrast, exclusive epistemic pluralists would argue that Norman’s pre-reflective entitlement has been replaced with justification. In general, the Reason Criterion may serve to classify other standard cases— children’s perception, appropriate specifications of chicken sexing, idiot savants— in a similar manner. Cases of inaccessible reasoning are less straightforward but reflection on unconscious reasoning suggests that they do occur. Given the Reason Criterion, such cases may be classified as exemplifying basic justification but not access justification or discursive justification. This captures the tension between various traditional epistemic internalist ideas that such cases exhibit. Of course, the Reason Criterion does not resolve all borderline cases. For example, there are difficult questions pertaining to what the faculty of reason is and what it is to exercise it competently. These are unresolved difficult questions in the interface between epistemology and philosophy of mind. But given that the epistemological distinction is informed by the philosophy of mind, it should be expected to inherit some unresolved issues in this area. Moving from cases to broader considerations, entitlement is widely thought to be shared with fairly unsophisticated non-human animals. In contrast, justification is distinctive to rational animals (see Burge 2003; Graham 2011, 2012. See also Sosa 2007 for a similar distinction in terms of animal and reflective knowledge). It is a difficult question whether the distinction tracks a broad historical distinction antedating the internalist-externalist dispute that arose in post-Gettier analytic epistemology. But at least it is historically prominent to distinguish between fundamentally different kinds of warrant in terms of reason.¹⁴

¹⁴ A preliminary glance indicates that the Reason Criterion has interesting but imperfect parallels to historical precursors of the internalism-externalism distinction. Aristotle distinguish between perceptual cognition (empeiria, Posterior Analytics ii 19) and scientific cognition (episteme). This distinction is drawn in terms of the capacity for reasoning (logos). However, there are further requirements on episteme. In particular, Aristotle assumes that at least one of the premises in the syllogism must be necessary. This is not part of the present distinction. Aristotle includes a distinctive category of cognition that consists in a direct intellectual grasp of first principles (noûs). The contemporary distinction between reasoning and understanding may derive from this distinction. As mentioned, I think that warrant by reasoning and warrant by understanding are usefully conceived of as species of warrant by reason.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Moreover, the Reason Criterion classifies contemporary theories in a nonrevisionist manner. For example, the criterion provides a principled explanation as to why evidentalism is a paradigm of an internalist theory (Conee and Feldman 2004). Since evidence is propositional, the key doctrine of evidentialism—roughly, that S’s justification is determined by her (response to her) evidence—requires that S competently exercises one of her a truth-conducive propositional-level competencies. That is, evidentialism requires that the agent exercises her faculty of reason and consequently it is classified as an epistemically internalist theory. Likewise, according to the Reason Criterion, it is not an accident that process reliabilism is the paradigm of an epistemically externalist theory (Goldman 1979). Process reliabilism concerns the sub-propositional processes of belief-generation such as perceptual processes. Indeed, a distinctively externalist claim of process reliabilism is that no exercise of the faculty of reason is required to supplement the normal reliable functioning of such processes insofar as they operate reliably in their normal environment (Burge 2003; Graham 2012b; Gerken 2013a). The Reason Criterion also tracks divisions within the internalist and externalist camps. For example, Kornblith is categorized as an externalist monist insofar as he thinks that reasons and reflection have no role in epistemology that goes beyond reliability (Kornblith 2002, 2008, 2015). But other externalists may be classified as pluralists who simply work on the externalist area of epistemology focusing less on warrant by reason than on more primitive species of warrant (Goldman 1986, 2009; Burge 2003; Lyons 2009; Graham 2012). The Reason Criterion is apt to capture a pluralist conception of warrant because it allows for different specifications of the non-lucky truth-connection that is constitutive of epistemic warrant. It allows that the truth-connection can be cashed out in terms of reliability for simple sub-propositional belief-generating processes. But it also allows for theories that articulate a more indirect and abstract truthconnection as internalist theories tend to do. Thus, the difference in the nature of the cognitive competence yields explanatorily different truth-connections. For example, the epistemic support and defeat structures for a sub-propositional system/competence with an etiological function of producing true beliefs are different Descartes famously distinguishes between cognitio and scientia. This distinction is drawn in terms of certainty rather than in terms of awareness: “The fact that an atheist can be “clearly aware (clare cognoscere) that the three angles of a triangle are equal to two right angles” is something I do not dispute. But I maintain that this awareness (cognitionem) of his is not true knowledge (scientiam), since no act of awareness (cognitio) that can be rendered doubtful seems fit to be called knowledge (scientia).” (Descartes, Replies 2, AT 7:141). Since Descartes requires the exercise of reason to establish scientia, the Reason Criterion may approximate the traditional cognitio-scientia distinction better than an accessibility criterion. Yet the parallel is broad and substantive differences should not be ignored. For example, both justification and entitlement are non-factive warrants and as such neither provides certainty. Kant distinguishes between cognition (Erkenntnis) in a broad sense of purportedly objective conscious representation and a narrow sense of purportedly objectively valid conscious representation whose objective validity is provable in way revealing grounds of its objective validity. Kant requires consciousness for both types of cognition. Narrow and broad cognition differ in terms of provability, which, in turn, requires the exercise of theoretical reason. Of course, this unduly broad and brisk historical perspective skips over both historical context and substantive detail. I include it only to note it has been historically common to distinguish between fundamentally different kinds of warrant by reference to reason considered as a cognitive competence.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

from the case of a proposition-level competence that concerns the structure of reasons. So, a truth-connection that is non-lucky in virtue of cognitive competence remains the unifying feature of all types of warrant. The distinction between species of warrant is psychological but it is not merely psychological. This is because the different psychological competences align with categorically different manifestations of the truth-connection. E.g., reliability vs. more abstract supporting/defeating relations between propositions. All in all, the Reason Criterion tracks important aspects of the internalism– externalism dispute. Of course, it does not align with every aspect of these extensive debates. But I take deviations, such as the ones discussed above, to be motivated. More importantly, the Reason Criterion tracks a substantially fruitful distinction. In consequence, I suggest that it is apt to draw the epistemic internalist–externalist distinction—i.e., the justification–entitlement distinction. As noted above, this does not render accessibilist approaches misguided. In particular, kinds of accessibility should be thought of as differentiating different sub-species of justification rather than as differentiating between justification and entitlement.

4.4.2 Warrant by reason without a reason? Thinking of access justification as a sub-species of justification may be in contrast to Burge’s most recent conception. As mentioned, Burge’s current conception of the justification–entitlement distinction is close to the Reason Criterion. On the one hand, he characterizes the distinction as follows: “A justification is a warrant that consists partly in the operation or possession of a reason. An individual is justified if and only if the reason is operative or relied upon in the individual’s psychology” (Burge 2013: 3). “An entitlement is a warrant whose force does not consist, even partly, in the individual’s using or having a reason”. (Burge 2013: 3–4)

I have articulated the Reason Criterion in terms of reason (i.e., a cognitive competence) whereas Burge casts the distinction in terms of a reason (i.e., a propositional content). On the other hand, he claims that justification does not require that the justifying reason is a fully formed state in the individual’s psychology: “But individuals can have justifications even though they lack the justifying reason as the content of a fully formed state in their psychologies—for example, if they could through prompting or reflection come to form such states relatively easily”. (Burge 2013: 3)

One way to understand this idea goes back to the slave boy scene of the Meno (Meno 82 B). Given this understanding, accessibility involves what the individual could relatively easily come to form propositional contents about—perhaps given Socratic probing.¹⁵ On this understanding, Burge appears to retain an element of the accessibility criterion. So, the Reason Criterion above should be seen as an alternative according to which S may lack justification although S has access to a justification. The Reason Criterion allows both cases of inaccessible justification (e.g., unconscious ¹⁵ However, Socratic probing must be characterized as to exclude teaching. I will set aside the interesting question as to whether Socrates is teaching the slave boy. I mention the broad analogy for orientation. Of course, the view does not involve any commitment to a Platonic theory of concepts or the like.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



reasoning) and cases where the agent is merely entitled but has a justification accessible to her. This illustrates the significance of accessibility to justification without reducing the justification to accessibility. Setting aside the question of accessibility, Burge’s remark raises the independently interesting question as to whether warrant can constitutively depend on the individual’s reason (the faculty) although “no relevant reason [content—MG] is present in an individual’s psychology”? This is a hard question. Cases of rational insight or warrant by basic understanding are candidates for an affirmative answer. Candidate examples include the Cartesian cogito, cogito-like thoughts and basic logical truths. The case of the cogito discussed above is also important in this regard. Descartes himself rejected that the warrant for the cogito was warrant by reasoning: “When someone says “I am thinking, therefore I am, or I exist,” he does not deduce existence from thought by means of a syllogism, but recognizes it as something self-evident by a simple intuition of the mind” (Descartes, Replies 2, AT 7: 140). I am sympathetic to the idea that warrant, indeed justification, by understanding is sometimes basic in the mentioned sense that S may be justified by the competent exercise of her reason (competence) although S does not have a reason (attitude or content lending epistemic support). I do not merely mean that there is no discursive justification that would convince someone who doubted—methodologically or doxastically—the proposition in question. The sense in which one cannot give reasons for basic principles may go deeper insofar as there is no reason one can give in support of it that is stronger than the warrant obtained by understanding the principle. Theorem proving in mathematics may deploy axiomatic methodology in reverse order. Mathematicians may not start with a small set of axioms and seek to prove theorems that they have not antecedently conceived of. Rather, they may first form a warranted belief that some theorem is true or provable.¹⁶ Then they try to “backward engineer” a proof for the theorem from accepted axioms. An additional axiom may even be postulated on the basis that it affords proof of plausible and otherwise improvable theorems. (This may be a candidate way of justifying an axiom.) Of course, constructing the proof provides additional warrant for the theorem. Indeed, doing so provides a justification. My point concerns the epistemic status of the theorem that is selected for (attempted) proof on its basis of being believed to be true or provable. Something like this modus operandi does not seem exclusively restricted to the context of discovery. Rather, mathematical insight may consist in part in fully or partly understanding a theorem, and thereby acquire warrant for believing that it is true or provable. If this is a correct description of some cases, it is not clear that the mathematicians in question have a reason for believing that the theorem is true or provable when they set out to construct a proof. There may not, prior to the proof, be any specific proposition (or set of propositions) that supports the belief that the theorem is true or provable. Nevertheless, such warrant appears to

¹⁶ I will not take a stand on the aims or metaphysics of mathematics. So, I will use the formulation ‘true or provable.’

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

be constituted by reason—i.e., by a sophisticated proposition-level cognitive capacity. I think that this makes it reasonable to classify the warrant in such cases as an internalist species—i.e., as a justification. Of course, this candidate case requires considerable further investigation and part of it concerns a psychologically plausible account of the mind of the mathematician in question.¹⁷ More generally, I do not have a conclusive answer to the question as to whether warrant by reason (the cognitive competence) can be manifest although no relevant reason (the propositional content) is present in an individual’s psychology. Even though understanding does not reduce to reasoning or accessible reasons, it might be that understanding requires the capacity for reasoning or access to reasons. Rather than trying to settle this question, I will make a conditional suggestion. Even if warrant by basic understanding does not require access to reasons, it is best characterized as a justification rather than as an entitlement. Such warrant is within the “realm of reason” and therefore attributable to sophisticated propositional aspects of the individual’s psychology that go beyond more primitive beliefgenerating processes. It would be problematic to regard an individual who is warranted due to the operation of an unconscious and inaccessible reason as justified and regard someone who is warranted by basic understanding as merely entitled. The assumption that one can be warranted by reason although without having a reason, is consistent with the Reason Criterion as stated above. However, the Reason Criterion and an accessibility criterion will very often coincide. Sub-propositional cognitive competences are often inaccessible by first-person methods whereas propositional level cognitive competencies are often accessible. So, accessibility is often a good proxy for the underlying Reason Criterion although it is not the most fundamental characterization of justification conceived most broadly. Rather, the importance of accessibility lies in the fact that it may be used to categorize and distinguish between different subspecies of justification and in the distinctive truth-conducive properties that it marks. (I have argued for the former above and will argue for the latter below.) Given the prominence of first-person access justification in these debates, this will be the central brand of justification that I will contrast with entitlement in the ensuing discussion. However, I will also emphasize the significance of discursive justification which I think has received too little attention. In particular, I will argue that both first-person accessibility and articulability are epistemically significant properties that lie outside of the scope of entitlement. ¹⁷ In many cases a reason may be available or operative. For example, it may be that the theorem has certain structural features that the mathematician reasonably takes to support the belief that it is true or provable. In such cases, the proposition expressing those structural features may be part of her understanding. But it is not clear that this class of cases is exhaustive. A further class of cases concern “intuitive mathematicians” such as Ramanujan (see Burge’s discussion in Burge 2013: 497–8). As Burge remarks, we know too little about Ramanujan’s psychology to know whether his “intuitive” beliefs are guided by the presence or availability of reasons, by basic understanding or by perception-like intuition. While I am skeptical of the latter suggestion, I do not seek to rule out that such cases are possible. Nor do I reject that in such cases the individual’s warrant would be best characterized as an entitlement. Rather, my point is to suggest that if there are cases of warrant by understanding where the individual does not have access to a reason but nevertheless exercises her faculty of reason, the warrant is a justification rather than an entitlement.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



4.5 Entitlement—Its Scope and Limits Entitlement of the sort that I have characterized, and distinguished from justification, has earned its keep in epistemological theorizing. I regard the development of externalist species of warrant as an instance of philosophical progress and I am not alone in doing so. In fact, epistemic externalism is such a successful reaction to traditionalist internalist concerns that it is, at present, a more urgent task to address externalist overreactions according to which entitlement should replace, rather than supplement, justification. In consequence, I will consider the scope and limits of epistemic entitlement. The cogito argument of Section 4.2.2 has already indicated some such limits. At the individual level, I will argue that access justification furthers epistemically rational belief revision (Section 4.6.1). At the level of social cognition, I will argue that discursive justification furthers epistemically rational communal belief revision in ordinary life as well as in scientific collaboration (Section 4.6.2).

4.5.1 The scope and limits of entitlement in individual cognition Why should an agent seek to achieve justification rather than mere entitlement? After all, if a belief is maximally entitled, it is maximally truth-conducive. Of course, there may be practical benefits of adding justification by, for example, appreciating the reliability of the sources of one’s belief. But externalist monists might object that this is not an epistemic benefit. In response to this challenge, will argue that many of the ways in which justification may be epistemically beneficial are manifest in social cognition. But I will begin by considering the epistemic benefits of justification at the individual level. Specifically, I will argue that in some cases access justification allows for truth-conducive belief revision in a manner that mere entitlement does not (see also Weiner 2005). Consider a case in which S meets the minimal conditions on entitlement, i.e., cases in which S competently exercise a cognitive competence in the formation of each of the following three beliefs: p, if p, then q and non-q. Given non-factivity of entitlement, we may assume that had each of the beliefs been held in isolation, it would have been entitled. However, from the perspective of overall epistemic rationality, S is in a wanting epistemic position. My point is not that a theory of entitlement is at a loss to account for this. Internal incoherence may be regarded as a defeater of entitlement. So, S is not entitled to the belief set and this may defeat her prima facie entitlement for each individual belief. My point is that the prima facie but defeated entitlement of the beliefs in the inconsistency triad does not put S in an optimal position to revise her belief set in a truth-conducive manner. Given only minimal entitlement, S has no reasons for any of the beliefs and she lacks access to the grounds that underwrites the entitlement for each belief. In such cases, entitlement provides a very sub-optimal basis for critical reasoning and rational belief revision. Note that if the inconsistent beliefs rest on other entitled beliefs, in a manner that S lacks access to, the entitlement for larger subsets of S beliefs may be defeated. Likewise, if all three beliefs are fundamental to S’s larger belief sets (perhaps in the manner of Wright’s cornerstones Wright 2000, 2004). So, prima facie entitlement for

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

rather large sets of beliefs may be defeated by a small set of inconsistent beliefs. So, the issue is not inconsequential. It is natural to think that insofar as S notices the inconsistency, she should either suspend each of the beliefs or pursue justification for them. The above considerations suggest that the latter is often an epistemically good idea. This, in turn, indicates the epistemic value of supplementing basic entitlement with justification. If S has justification of the accessibilist variety for the beliefs in the inconsistency triad, she may be in better a position to rationally revise them in a truth-conducive manner (Harman 1986).¹⁸ Access to the reasons for the beliefs in the inconsistency triad may allow S to identify various defeaters or structural asymmetries. For example, it might be that the epistemic reasons for holding the belief that p are equally good reasons for regarding the prima facie reasons for holding the belief that non-q as faulty. If the situation is not symmetric, i.e., if the reasons for believing that non-q do not tell against the reasons for believing that p, then it might be epistemically reasonable to dispense with the belief that non-q. I do not suggest that this is always the case or that access justification will always allow S to revise her beliefs in a truth-conducive manner. There may be cases in which the warrants for the beliefs in the inconsistency triad are independent and equally strong. In such cases, access to the relevant epistemic grounds may be to no avail. My claim is that there are some cases where mere entitlement puts the subject in a sub-optimal position for truth-conducive belief revision and where (access) justification improves the subject’s position in this regard. But this makes for an epistemically significant contribution of justification. Importantly, this line of reasoning is compatible with the idea that the mechanisms that resolve inconsistencies are unconscious and relying on tacit reasons that may even be inaccessible to conscious reflection. Nevertheless, such mechanisms operate on propositional contents of S’s beliefs and revise those beliefs on the basis of the (tacit) reasons supporting them. So, the mechanisms would manifest the subject’s faculty of reason. Moreover, for human agents there will, as a matter of contingent psychological fact, be a subset of cases in which the only mechanism for truthconducive belief revision involves access to the reasons for the inconsistent beliefs. If, for example, the faulty belief is produced by a biased belief-generating process, there may be no subconscious mechanisms that recognize this fact. Our biases are especially problematic when we are unaware of them. Indeed, it is only once we submit the beliefs and presuppositions that they generate to critical reflection that we make cognitive progress. So, given that our biased subconscious processes often lack selfcorrecting mechanisms, having access to the basis for the beliefs that they produce is often paradigm for truth-conducive correction. So, given that we pursue a theory warrant for humans, the theory should admit access justification. The present considerations do not represent a monist internalist line of reasoning. Indeed, the present line of reasoning is compatible with the idea that entitlement is also central to rational belief revision (Burge 1996). I am merely arguing for the ¹⁸ Smithies also argues that his strong brand of accessibilist justification is valuable in virtue of contributing to belief revision but he fails to emphasize the truth-related value of this contribution (Smithies 2012, 2015).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



pluralist conclusion that access justification may provide an additional and genuinely epistemic contribution to belief revision. So, once we cease to only focus on the generation of individual beliefs and turn our attention to the agent’s sets of beliefs, (access) justification appears to be an important genuinely epistemic property. In retrospect, this upshot may not be too surprising. After all, many externalist theories, such as processes reliabilism, concern the properties of beliefs—in particular, perceptual beliefs—in isolation. In contrast, internalists, such as coherentists, have typically emphasized the rational perspective of the agent’s total set of beliefs (Bonjour 1985). However, the line of reasoning of this section does not require any substantive coherentist views. The key requirement is only that warrants may be interrelated insofar as they may exhibit defeating relations. Given suitable access to these relations, reflection upon them may be used to revise beliefs in a truthconducive manner. If so, access justification is an important species of justification. However, it does not show that access or accessibility is a necessary condition on justification. Again, the exercise of rational belief revision requires the exercise of the faculty of reason. Without the exercise of reason, access is of little use. I take this fact to augment the Reason Criterion. The epistemic value of (access) justification is not merely valuable in supporting truth-conducive belief revision. It may also render warranted beliefs more robust against, for example, misleading evidence. If one has access to the grounds for one’s belief, one is a better position to retain it in the face of misleading evidence to the contrary. I have, in the context of testimonial justification, given examples of cases in which an access justified individual may rebut or ignore such defeaters in a truthconducive manner (Gerken 2013b: sect. 3.2). But since space is limited and I do not want to repeat myself I will move on to the realm of social cognition.

4.5.2 The scope and limits of entitlement in social cognition Agents who are merely entitled to their beliefs will have a hard time defending their views when challenged in discourse and to provide reasons for holding their views in communal deliberation. In consequence, analogs of the above case of belief revision may occur in intersubjective cognition (here I elaborate on Gerken 2013b: sect. 3.2). Consider a group consisting of S¹, S², and S³. Each of them is entitled to a belief and together their beliefs form an inconsistent belief set. For example, S¹ is entitled to the belief that p, S² is entitled to the belief that p entails q and S³ is entitled to the belief that non-q. If the members of the group come to realize that they hold jointly inconsistent beliefs, it is, at least typically, epistemically rational for them to engage in communal belief revision. But if they are entitled and lack any access to the grounds of their beliefs, they are in a bad position to determine who among them should give up her belief. In contrast, if the group members are discursively justified, they can articulate their reasons for their respective beliefs. Hence, they have a better basis for making a truthconducive communal deliberation. In such a case, discursive justification seems to be genuinely epistemically important. What enables the group deliberation to proceed truth-conducively is that the reasons for and against the beliefs in question are set forth. As in the individual case, this is particularly so if, for example, some of the

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

reasons for believing that p also defeats the reasons for believing that non-q (and not vice versa). Importantly, S¹’s degree of entitlement may be held fixed throughout the process of communal belief revision. To see this, assume that S¹’s and S²’s respective justifications for their respective beliefs that p and that p entails q are far stronger than S³’s justification for the belief that non-q. Assume, moreover, that S¹’s and S²’s reasons undermine S³’s justification (and not vice versa). If the communal deliberation goes well and S³ gives up her belief that that non-q, S1 is not any more or any less entitled to believe that p. But nevertheless the group’s epistemic position is improved. More subtly, S1’s own epistemic position is also improved insofar as she may avoid suspending an entitled belief. The epistemic contribution of justification may go beyond the social analog of belief revision. Justification may also contribute to more constructive cognition. For example, a group might improve its epistemic position on some subject matter through communal deliberation. Doing so can not only expend the data set, it can also provide novel perspectives and methods. There is some evidence that on a Wason card selection task, a deliberating group can outperform its best members (Moshman and Geil 1998). Other studies suggest similar results for mathematical tasks (Stasson et al. 1991). Further studies have led to the idea that, in Hong and Page’s phrase, “diversity trumps ability.”¹⁹ Given the available evidence, I will only assume that in some conditions and for some tasks, a deliberating group can outperform its best members. When this occurs, however, it is a distinctively epistemic benefit of group deliberation. While more empirical evidence is required to assess the extent of this epistemic benefit, there is good reason to think that it is non-negligible. The conditions under which communal deliberation is truth-conducive is primarily an empirical question. However, I venture to suggest that discursively justified deliberators are better suited to engage in truth-conducive group deliberation than merely entitled deliberators. After all, truth-conducive group deliberation consists, in significant part, in offering epistemic reasons and perspectives and reevaluating them in the light of the reasons and perspectives that other’s set forth. This process requires not only that the grounds for the warrant be accessible to the members but also that they are capable of articulating them as epistemic reasons. In contrast, it is hard to see how a group consisting of members who only possess basic entitlement for their belief can make much progress. The cases sketched above are idealizations. But they are idealizations of situations that may be reflected in humanity’s most successful mode of cognition: Science. This is not the place for a detailed discussion of the nature of scientific warrant. So, I will restrict myself to a couple of broad suggestions (developed further in Gerken 2015). My key suggestion is that reflection on scientific virtues suggests that access justification and even discursive justification are central to the scientific enterprise. Someone who has no access to the basis of her results, hypothesis and theories is unlikely to be taken seriously in the scientific community. In general, scientific results,

¹⁹ See, e.g., Hong and Page 2004: 16385. See also, Page 2007; Fishkin and Luskin 2005; Mercier and Sperber 2011; Mercier and Landemore 2012; Landemore 2013.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



hypotheses and theories are supposed to have a transparent basis that is not idiosyncratic to some individual. This overarching virtue of transparency manifests itself in the requirement of replicability. Replicability, in turn, requires that the data for some scientific hypothesis as well as the methodology relevant to collecting and assessing it be made available. Science is an increasingly collaborative affair and this sets forth important demands on scientific practice (Thagard 1997; Wray 2002). The fact that the scientific enterprise is collaboratively driven suggests that discursive justification is central to scientific warrant. Indeed, part of what one learns when one learns a science is a number of dialectical and epistemic norms pertaining how to conduct and present scientific research. Epistemic norms of action and assertion should accommodate this (see Gerken 2011, 2012, and, especially, 2015). While the norms are highly discipline specific, it is fairly general that it must be made explicit what epistemic support that a given scientific result, hypothesis or theory enjoys. So, although the nature of scientific warrant is a complex issue, I mention it briefly here because it is a central example of how entitlement has important limits in highly organized collaborative cognition. My point is not to deny that there is a place for entitlement in an account of scientific warrant. Rather, my point is that both access and discursive justification are indispensable in the scientific enterprise (see Gerken forthcoming b). Given a robust scientific realism, the internalist kinds of warrant are not merely pragmatic but genuinely epistemic. Science is, when things go well, a highly truth-conducive mode of cognition in large part because it has a social structure that permits for rational continuity based on criticism, refinement and reassessment of both data and theory. If scientific warrant was merely entitlement, this idea of continuous rational progress would rarely be approximated. Of course, the example of scientific warrant is a grand case that requires a separate treatment (see Gerken 2015 for an elaboration). I mention it as a candidate example of truthconducive inquiry in which entitlement only plays a limited role because it is too important a case to ignore. In fact, it seems reasonable to suppose that the entitlement-justification debates have a fertile application at the intersection of epistemology and philosophy of science. Here I restrict myself to calling attention to these issues and to suggesting that scientific warrant exemplifies some important epistemic scope and limits of entitlement.

4.6 Concluding Remarks Epistemic entitlement has come to stay. It marks an important and, I believe, lasting counter to traditional requirements on epistemic rationality that have emphasized individual responsibility, higher-order cognition or awareness, the phenomenality of appearance and so forth. However, the step from recognizing externalist conceptions of epistemic rationality to rejecting the traditional internalist conceptions is faulty insofar as it presupposes a problematic monism. A pluralist alternative provides a superior framework. But given epistemic pluralism, a host of novel and intricate questions arise. Principal among them is the question as to how to distinguish entitlement from justification. I have answered this question in a novel manner by setting forth the Reason Criterion. However, epistemic pluralism raises further questions including questions

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

about the proper domains of entitlement and justification, respectively. Addressing these issues may be more productive than the internalist-externalist “trench war” that is based on monist presuppositions. I have begun to address them by noting some of the limitations of entitlement in individual and social cognition. For example, I have emphasized a range of cases in which discursive justification appears to be required, or at least beneficial, for social cognitive processes. For dramatic effect, these points may be sloganized: Entitlement is socially inapt. The slogan is, as slogans tend to be, an overstatement. Merely entitled agents are capable of some social cognition. And, of course, entitlement is important to theorizing about social cognition (Goldman 1999a, 2010, but see also Shogenji 2012). Nevertheless, the slogan contains a grain of truth. It is very doubtful that intricate social cognition—such as science—is possible for groups with members lacking a faculty of reason and the ability to articulate epistemic reasons. Since I am nearing the end, I’ll allow myself yet a slogan to sum up this idea: Humans are social animals in virtue of being rational animals. In conclusion, the development of epistemic entitlement or, more broadly, externalism in epistemology is an example of philosophical progress. To make further progress, however, it is paradigm that we recognize and respect the scope and limits of epistemic entitlement.²⁰

References Audi, R. 1999. Self-evidence. Philosophical Perspectives 13 (suppl.): 205–28. Audi, R. 2001. The Architecture of Reason. New York: Oxford University Press. Bonjour, L. 1985. The Structure of Empirical Knowledge. Boston: MIT Press. Bonjour, L. 1992. Internalism/Externalism. In A Companion to Epistemology, eds. J. Dancy and E. Sosa, 132–6. Oxford: Blackwell. Brandom, R. 1998. Insights and Blindspots of Reliabilism. The Monist 81: 371–92. Burge, T. 1988. Individualism and Self-Knowledge. The Journal of Philosophy 85: 649–63. Burge, T. 1993. Content Preservation. The Philosophical Review 102: 457–88. Burge, T. (1996). Our entitlement to self-knowledge. Proceedings of the Aristotelian Society 96(1): 91–116. Burge, T. 2003. Perceptual Entitlement. Philosophy and Phenomenological Research 67: 503–48. Burge, T. 2013. Cognition through Understanding: Philosophical Essays Vol. 3. Oxford: Oxford University Press. Chalmers, D. 2011. Verbal Disputes. Philosophical Review 120: 515–66. Conee, E. and Feldman, R. 1998. The Generality Problem for Reliabilism. Philosophical Studies 89: 1–29. Conee, E. and Feldman, R. 2004. Evidentialism. Oxford: Oxford University Press. Descartes, R. 1641/1996. Meditations on First Philosophy: With Selections from the Objections and Replies, transl. J. Cottingham and ed. B. Williams. Cambridge: Cambridge University Press. ²⁰ I grateful to Peter Graham and Nikolaj Jang Lee Linding Pedersen and for helpful feedback on the chapter and many discussions about entitlement before that. I am also grateful to Tyler Burge for criticism of my early work on this topic and to two OUP referees who endured a way too early draft and provided excellent comments nevertheless. Some of the material was presented at Danish Epistemology Network (University of Copenhagen), at a Basic Knowledge Workshop (Northern Institute of Philosophy, University of Aberdeen) and at an NYU Epistemology Group. I am grateful to the audiences for helpful discussion. Dedikeret til min lille ven Teo.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Dretske, F. 1970. Epistemic Operators. Journal of Philosophy 67: 1007–23. Dretske, F. 1971. Conclusive Reasons. Australian Journal of Philosophy 49: 1–22. Dretske, F. 2000. Entitlement: Epistemic Rights without Epistemic Duties? Philosophy and Phenomenological Research 60: 591–606. Evans, J. 2008. Dual Processing Accounts of Reasoning, Judgment, and Social Cognition. Annual Review of Psychology, 59: 255–78. Evans, J. St. B. T. and Stanovich, K. E. 2013. Dual-process Theories of Higher Cognition: Advancing the Debate. Perspectives on Psychological Science, 8: 223–41. Fishkin, J. and Luskin, R. 2005. Experimenting with a Democratic Ideal: Deliberative Polling and Public Opinion. Acta Politica 40: 284–98. Fumerton, R. 1995. Metaepistemology and Skepticism. Rowman & Littlefield. Gerken, M. 2011. Warrant and Action. Synthese 178: 529–47. Gerken, M. 2012. Discursive Justification and Skepticism. Synthese 189: 373–94. Gerken, M. 2013a. Epistemic Reasoning and the Mental. London: Palgrave Macmillan. Gerken, M. 2013b. Internalism and Externalism in the Epistemology of Testimony. Philosophy and Phenomenological Research 87: 532–57. Gerken, M. 2015. The Epistemic Norms of Intra-Scientific Testimony. Philosophy of the Social Sciences 45: 568–95. Gerken, M. 2017. Against Knowledge-First Epistemology. In J. A. Carter, E. Gordon, and B. Jarvis eds., Knowledge-First Approaches in Epistemology and Mind. Oxford: Oxford University Press. Gerken, M. 2018a. On Folk Epistemology. How We Think and Talk about Knowledge. Oxford: Oxford University Press. Gerken, M. 2018b. The New Evil Demon and the Devil in the Details. In V. Mitova ed., The Factive Turn in Epistemology. Cambridge: Cambridge University Press. Gerken, M. Forthcoming a. Truth‐Sensitivity and Folk Epistemology. Philosophy and Phenomenological Research. Gerken, M. Forthcoming b. The Significance of Scientific Testimony. Gerken, M. Ms (2005). Justification: Warrant by Reasoning. UCLA Proposition. Unpublished manuscript. Goldman, Alvin I. (1967). A causal theory of knowing. Journal of Philosophy 64 (12): 357–72. Goldman, Alvin I. (1976). Discrimination and perceptual knowledge. Journal of Philosophy 73: 771–91. Goldman, A. 1979. What Is Justified Belief? Justification and Knowledge: New Studies in Epistemology, ed. G. Pappas. Dordrecht: D. Reidel Publishing Company. Goldman, A. 1986. Epistemology and Cognition. Cambridge, MA: Harvard University Press. Goldman, A. 1999a. Knowledge in a Social World. Oxford: Oxford University Press. Goldman, A. 1999b. Internalism Exposed. Journal of Philosophy 96: 271–93. Goldman, A. 2009. Internalism, Externalism, and the Architecture of Justification. Journal of Philosophy 106: 309–38. Goldman, A. 2010. A Guide to Social Epistemology. In A. Goldman and D. Whitcomb eds., Social Epistemology: Essential Readings. Oxford: Oxford University Press. Graham, P. 2011. Psychological Capacity and Positive Epistemic Status. In J. Hernandez ed., The New Intuitionism, 128–150. London: Continuum Press. Graham, P. 2012. Epistemic Entitlement. Noûs 46: 449–82. Harman, G. 1986. Change in View. Cambridge, MA: MIT Press. Hong, L. and Page, S. 2004. Groups of Diverse Problem Solvers Can Outperform Groups of High-ability Problem Solvers. Proceedings of the National Academy of Sciences 101: 16385–9. Kornblith, H. 2002. Knowledge and its Place in Nature. Oxford: Oxford University Press. Kornblith, H. 2008. Knowledge Needs No Justification. In Q. Smith ed., Epistemology: New Essays, 5–23. Oxford: Oxford University Press. Kornblith, H. 2015. The Role of Reasons in Epistemology. Episteme 12: 225–39.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 :   

Lyons, J. 2009. Perception and Basic Beliefs: Zombies, Modules, and the Problem of the External World. Oxford: Oxford University Press. Mercier, H. and Sperber, D. 2011. Why Do Humans Reason? Arguments for an Argumentative Theory. Behavioral and Brain Sciences 34: 57–74. Mercier, H. and Landemore, H. 2012. Reasoning Is for Arguing: Understanding the Successes and Failures of Deliberation. Political Psychology 33: 243–58. Moshman, D. and Geil, M. 1998. Collaborative Reasoning: Evidence for Collective Rationality. Thinking and Reasoning 4: 231–48. Page, S. 2007. Diversity Trumps Ability Theorem. In The Difference: How the Power of Diversity Creates Better Groups, Firms, Schools, and Societies, 131–74. Princeton, NJ: Princeton University Press. Pappas, G. 2014. Internalist vs. Externalist Conceptions of Epistemic Justification. In Edward N. Zalta ed., The Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/archives/ fall2013/entries/justep-intext/ Plantinga, A. 1993. Warrant and Proper Function. Oxford: Oxford University Press. Pollock, J. L., and Cruz, J. (1999). Contemporary Theories of Knowledge (Vol. 35). Rowman & Littlefield. Pritchard, D. 2003. Virtue Epistemology and Epistemic Luck. Metaphilosophy 34: 106–30. Quine, W. V. 1969. Epistemology Naturalized. In Ontological Relativity and Other Essays. New York: Columbia University Press. Smithies, D. 2012. Moore’s Paradox and the Accessibility of Justification. Philosophy and Phenomenological Research 85: 273–300. Smithies, D. 2015. Why Justification Matters. In D. Henderson and J. Greco eds., Epistemic Evaluation: Point and Purpose in Epistemology, 224–44. Oxford: Oxford University Press. Sawyer, S. and Majors, B. 2005. The Epistemological Argument for Content Externalism. Philosophical Perspectives 19: 257–80. Shogenji, T. 2012. Internalism and Externalism in Meliorative Epistemology. Erkenntnis 76: 59–72. Sosa, E. 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, Volume I. Oxford University Press. Stasson, M. F., Kameda, T., Parks, C. D., Zimmerman, S. K., and Davis, J. H. (1991). Effects of Assigned Group Consensus Requirement on Group Problem Solving and Group Members’ Learning. Social Psychology Quarterly 54: 25–35. Stroud, B. 1994. Scepticism, ‘Externalism’, and the Goal of Epistemology. Proceedings of the Aristotelian Society 68 (suppl.): 291–307. Thagard, P. 1997. Collaborative Knowledge. Noûs 31: 242–61. Treffert, D. A. 1988. The Idiot Savant: The Review of a Syndrome. American Journal of Psychiatry 145: 563–72. Williamson, T. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, T. 2013. Knowledge First. In M. Steup and J. Turri eds., Contemporary Debates in Epistemology, 2nd ed., 1–9. Oxford: Blackwell. Weiner, M. 2005. Why Does Justification Matter? Pacific Philosophical Quarterly 86: 422–44. Wray, K. B. 2002. The Epistemic Significance of Collaborative Research. Philosophy of Science 69: 150–68. Wright, C. (2000). Cogency and question-begging: Some reflections on McKinsey’s paradox and Putnam’s proof. Philosophical Issues, 10: 140–63. Wright, C. 2004. Warrant for Nothing (and Foundations for Free)? Aristotelian Society 78 (suppl.): 167–212.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

5 Why Should Warrant Persist in Demon Worlds? Peter J. Graham

5.1 Introduction Suppose you perceive as of a red ball on a table and thereby believe that there is a red ball on the table. According to mainstream moderate foundationalist theories of epistemic warrant, your perception as of a red sphere on a flat surface contributes to a prima facie warrant for your perceptual belief.¹ Though nearly everyone in epistemology agrees perceptual states contribute to the warrant for perceptual beliefs, epistemologists disagree over why this should be so. One major disagreement surrounds so-called “demon-world” cases.² Imagine an ordinary perceiver reliably forming true perceptual beliefs on the basis of a normally ¹ I use ‘warrant’ in line with Burge’s use, where warrant constitutively though fallibly contributes to, and conduces, truth and knowledge. Warrant generically is not, as Plantinga would have, that which converts true belief into knowledge. On Plantinga’s use, there is no such thing as a warranted true belief that does not amount to knowledge, or any such a thing as a warranted false belief for that matter. I shall not be using the word this way. As an English word, ‘warrant’ goes back at least to the fourteenth century, with a number of different uses. Plantinga didn’t invent the word; he stipulated a particular use for his own purposes. Burge uses ‘warrant’ largely synonymously with the field’s broad use of the term ‘justification.’ Warrants are epistemic accomplishments, successes or achievements that underlie the individual’s epistemic right to form or hold a belief, epistemic bases or grounds that are constituents or enablers of knowledge. Burge then distinguishes two kinds of warrant: entitlement and justification. Justifications are associated with reasons. An epistemic reason is a propositional attitude that contributes to the explanation of the belief-worthiness of the conclusion belief. A self-evident proposition contains its own reason for belief; understanding the proposition involves understanding the reason for belief. Deductive and inductive arguments in the psychology of a believer paradigmatically include warranted premise beliefs that support the belief-worthiness of the conclusion belief. The premise beliefs are then reasons for the conclusion belief. Entitlements are warrants without reasons. Perceptual warrants are paradigmatically, if not exclusively, entitlements, for perceptual belief formation is not a case of reasoning, but instead a psychological transition from a perceptual state to a conceptualization of that state. Importantly for Burge, as we’ll see, the warrant for a perceptual belief is not the perceptual state considered in isolation, but involves the normal functioning of the perceptual system in the formation of the perceptual state and the normal functioning of the transition to perceptual belief in the conceptualization of the perceptual state. For elaboration and discussion of these points, see Graham (forthcoming). ² I shall use the phrase ‘demon world’ broadly for not just those circumstances where a powerful demon is systematically fooling his victim, but also cases where massive undetectable error occurs because of a cunning scientist, undetectable abnormalities in the environment, or even a bizarre cosmic accident. I do Peter J. Graham, Why Should Warrant Persist in Demon Worlds? In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0005

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

functioning perceptual system and perceptual belief forming process. As described so far, it is more than reasonable to assume that this perceiver is forming warranted, and mostly true, perceptual beliefs. Now imagine a psychological duplicate of this individual undergoing the very same type of perceptual states and forming the same type of perceptual beliefs, but also imagine that this duplicate is not forming mostly true perceptual beliefs but entirely false perceptual beliefs. Maybe the individual has been unknowingly kidnapped and is now the subject of a virtual reality experiment at Cal Tech, or perhaps the individual is a massively deceived disembodied brain in a vat on Alpha Centauri, or perhaps an evil demon is systematically causing undetectably misleading perceptual states. We now have two cases: the normal case where a perceiver forms mostly true perceptual beliefs, and “demon-world” cases where the psychological duplicate (the individual who, from the inside, as it were, has exactly the same states, phenomenology, beliefs and so on) forms massively false perceptual beliefs. If the ordinary perceiver has warranted perceptual beliefs, does the psychological (but massively fooled) duplicate also have warranted perceptual beliefs? One position in epistemology—internalist essentialism—says yes. Perceptual states necessarily possess the epistemic power to defeasibly warrant perceptual beliefs in all possible worlds, no matter the chances of truth. Perceptual states not only (contribute to a prima facie) warrant (for) perceptual beliefs in worlds where those states lead the individual to form mostly true perceptual beliefs, but perceptual states also (contribute to) warrant (for) perceptual beliefs in worlds or circumstances where those states, unbeknownst to the individual subject, lead the individual to entirely false perceptual beliefs. There is something “intrinsic” or “essential” about perceptual states so that they contribute to prima facie warrant for beliefs appropriately based on those experiences, never mind the reliability of those experiences, or the reliability of the beliefs they appropriately cause and sustain. Forming true perceptual beliefs, on this view, has nothing to do with the warranting force of perceptual states. True, warranted perceptual beliefs in normal circumstances are mostly true, but that is neither here nor there given the nature of warrant. On this view, warrant, qua epistemic property, is not by its nature a reliable route to truth. Warrant, qua epistemic property, does not by its nature contribute to truth.³ An opposing position in epistemology—reliabilism—says no. Simple process reliabilism asserts that in all possible worlds or circumstances a belief-forming process contributes defeasible warrant on a belief appropriately based on the process if and only if the belief-forming process reliably induces true beliefs in that very possible world or circumstances of use. Warrant by its nature is a reliable route to truth. Applied to perception, in all possible worlds or circumstances a perceptual state or perceptual competence defeasibly warrants a perceptual belief appropriately derived from the state or competence if and only if the state or competence is reliably

not grant the metaphysical possibility of disembodied spirits or always “envatted” brains with no causal or explanatory relations to a wider environment. See Graham 2012 and Gerken 2018. ³ Cohen and Lehrer 1983; Cohen 1984; Pollock 1984; Foley 1985; Wedgwood 2002; Cruz and Pollock 2004.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



veridical and so reliably induces true perceptual beliefs in that very possible world or circumstances of use.⁴ On this view, our ordinary perceiver has warranted perceptual beliefs, for we’ve assumed he’s relying on reliable perceptual states when forming perceptual beliefs, but our duplicate does not have warranted perceptual beliefs, for the duplicate’s perceptual states are completely in error. Whether perceptual states defeasibly warrant perceptual beliefs seems to be, on this view, an entirely “extrinsic” or contingent, nonessential feature of perceptual states and perceptual competencies. Some do, some don’t. It depends on the perceiver’s environment, among other things. “Internalists” of the moderate foundationalist variety think perceptual states contribute to warrant by their nature, never mind truth. That’s why warrant may easily persist in “demon-worlds.” “Externalists” think warrant contributes to truth by its nature, never mind the essential nature of the warranting state. That’s why warrant might just lapse in “demon-worlds.” This dispute between so-called “internalists” and “reliabilists” began in the late 1970s and the early 1980s. It is still very much alive as I write. Who currently has the upper hand? It is hard to say. On the one hand, the intuition that the massively deceived psychological duplicate has warranted perceptual beliefs is very strong and widespread. On the other hand, the general claim that warrant and truth are somehow constitutively related—the general claim that warrant is, in some sense, a reliable route to truth—is equally very strong and widespread. How could warrant, after all, be epistemic warrant, if it didn’t somehow reliably conduce truth and knowledge? Since it doesn’t initially seem as if both sides can be correct, the disagreement continues. The work on this issue that interests me the most looks for middle ground. One strategy starts from the internalist side and tries to accommodate the opposing intuition: insist that warrant does not require reliability and then explain the connection with truth in other terms. The other strategy starts from the reliabilist side: insist that warrant requires reliability and then explain why warrant would persist in demon worlds even so. My preference is to start from the reliabilist side, for reliability theories of warrant purportedly lay bare the connection between warrant and truth, where “internalist” theories seem to leave it a mystery, at least to me. Even though I think starting from the reliabilist side is the right way to go, this is not an easy task. The best-known work along these lines comes from Alvin Goldman and Ernest Sosa.⁵ Though the details differ, their main move goes like this: as a long as a belief-forming process (like perception) is reliable in a special set of worlds or circumstances (which then need to be specified to fill out the details of the theory), then the process defeasibly warrants beliefs based on that process in all possible worlds or circumstances. Then our ordinary perceiver has warranted perceptual beliefs (for our ordinary perceiver is in special circumstances, reliably forming true perceptual beliefs), and our massively deceived duplicate in “demon” circumstances also has warranted perceptual beliefs (for the duplicate is using a process that is

⁴ Goldman 1979, 1986.

⁵ Goldman 1986, 2011; Sosa 2001.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

reliable in special circumstances, even if not reliable in the duplicate’s current circumstances). According to “special circumstances” reliabilism, one can thereby “capture” the intuition that demon victims have warranted perceptual beliefs; in other words, one can concoct an extensionally adequate theory “immune” to demon world type counterexamples. Though highly sympathetic, I am not persuaded by Goldman and Sosa’s efforts. They both face a general problem that applies to any instance of so-called “special circumstances” reliabilism: if warrant, qua epistemic property, is constitutively a good route to truth and knowledge, then how could warrant persist outside of “special” circumstances where the belief-forming process isn’t a route to truth and knowledge? Just because the process is reliable here, in these “special” circumstances, why should it confer warrant there, in those circumstances? If it contributes to truth here, but not there, how could it warrant there? If it does warrant there, why does it warrant there? What explains why warrant persists outside of special circumstances; what explains why warrant should persist in demon worlds? In my opinion, Goldman and Sosa have done little to answer this explanatory question, let alone directly address it. An extensionally adequate theory is not the same as an explanatorily adequate theory.⁶ In “Perceptual Entitlement” (Philosophy and Phenomenological Research, 2003) Tyler Burge, on the other hand, directly addresses this question: One can ask why should reliable success in . . . [the] normal environment bear on the contribution of perceptual states to entitlement in abnormal environments that the individual may have landed in . . . [for] it is the individual at a given time for which warrant is supposed to provide a route to truth . . . In many such environments, a perceptual state or competence . . . would not be a good guide to veridicality or truth . . . Why does [the contribution of the normal environment] carry over to other environments that the individual might form perceptual beliefs within?. (Burge 2003: 533, emphasis added)

Burge’s answer to this question, to my knowledge, has received little attention in the epistemological literature, despite its potential to answer a question that has gone largely unaddressed among those working in the reliabilist camp. Given the potential payoff, and the overall philosophical sophistication of his account of perceptual warrant in particular and epistemic warrant in general, critical examination of Burge’s account should prove well worth the effort.⁷

⁶ I have discussed this strategy in Graham 2010, 2012, 2014b, 2016 and 2017. I criticize Goldman in Graham 2012, 2017. I criticize Sosa in Graham 2016. Other proposed solutions come from Henderson and Horgan 2011 and Lyons 2009, 2013. I criticize Henderson and Horgan in Graham 2014c. I criticize Lyons in Graham 2011, 2014a, 2017. ⁷ I regard “Perceptual Entitlement” as a landmark paper in the philosophy and epistemology of perception and perceptual belief. It addresses the nature of warrant, the nature of sensation, perception, perceptual agency, perceptual belief, and the epistemology of perception, while effectively criticizing seminal positions on these topics from Sellars and others. Dense, detailed, empirically well-informed and highly original, it presents one of the most sophisticated accounts of these topics to date. We all have a great deal to learn from this paper. I know I have. Despite my admiration for Burge’s paper, it has not been widely discussed, though it has been frequently cited. This may be due to its difficulty; as I said, it is dense and detailed, covering a wide scope of interrelated topics.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



That’s exactly what I plan to do in this chapter. I first expound Burge’s account of perceptual warrant before turning to his direct answer to our question. Burge gives his direct answer in a “transcendental argument” that is easy to find difficult to understand. After reconstructing his argument as an explanation for why warrant persists in demon worlds, I argue (unfortunately for our purposes) that it falls short, that it doesn’t provide a satisfactory explanation. A change of focus, however, combined with a distinction between two grades of warrant, on the other hand, does the trick. Consistent with Burge’s overall account of perceptual warrant, we can explain, I shall argue, why warrant should persist in demon worlds.

5.2 Burge’s Normal Circumstances Reliabilism I now turn to Burge’s view. Burge is no simple reliabilist; the sheer reliability of a belief-forming process is not on its own sufficient for warrant. Even so, he’s clearly as much of a reliabilist as Goldman and Sosa, for Burge also advances a version of special circumstances reliabilism, just as they do; reliability in special circumstances is (largely) necessary and sufficient, absent counter-considerations, for warrant. On Burge’s version of special circumstances reliabilism, special circumstances are normal circumstances: a perceptual state or competence defeasibly warrants a perceptual belief appropriately derived from the perceptual state in all possible circumstances when functioning normally if and only if the perceptual state or competence is reliably veridical in normal circumstances when functioning normally (Burge 2013: 223, 260). An individual massively fooled in a demon scenario may thus enjoy warranted perceptual beliefs, provided the individual’s normally functioning perceptual competence is reliable in normal conditions: Where a perceptual state is reliable in the normal environment, and certain other conditions on entitlement are met, the individual is entitled in any environment to perceptual beliefs that are appropriately derived from the perceptual state. (Burge 2003: 532, 539) Thus one does not become unwarranted by landing in an abnormal environment, or by being systematically tricked by a demon or a scientist. (Burge 2003: 507) [R]eliability . . . in relation to the environment the entitled individual happens to be in . . . is not necessary for entitlement. (2003: 509, n. 6)⁸

Burge must then face the same question any special circumstances reliabilist must face: why should the fact that a perceptual state is reliably veridical in special (and so normal) conditions entail that it defeasibly warrants perceptual belief in any circumstances? As I just said, what’s so appealing about Burge’s account is that he squarely faces this question and provides an answer worth interrogating. He does not settle for extensional adequacy; he sets out to explain the persistence of warrant in demonworlds. And so the issue before us is not extensional but explanatory adequacy. Does Burge explain why warrant should persist in demon worlds? ⁸ Other authors that emphasize the importance of reliability in normal conditions include Bergmann 2006; Graham 2012, 2016, 2017; Leplin 2009; and Majors and Sawyer 2005.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

Before turning to Burge’s answer, we’ll first have to run through the way he motivates his normal circumstances theory of perceptual warrant. If you don’t understand his motivation, you won’t understand his answer.

5.2.1 Two conditions on perceptual warrant To motivate his view, Burge places two conditions on perceptual states to defeasibly warrant perceptual beliefs.  :  Perceptual warrant depends on the reliable veridicality of the perceptual state, for a condition on warrant is that it be a good route to truth. The “fundamental function of epistemic entitlement [is] serving as a good route to truth” (Burge 2003: 533). To contribute to perceptual warrant, a perceptual state type must be reliably veridical (Burge 2003: 506, 531–6). “Reliability is necessary for epistemic entitlement because all epistemic warrant is fundamentally an epistemic good inasmuch as warrant is a good route to truth and knowledge” (Burge 2003: 532).  :      Perceptual warrant also depends on the nature of the perceptual state, for a condition on perceptual warrant is that the perceptual state contributes to truth and knowledge through its nature. Here Burge has in mind the fundamental foundationalist and traditional idea that perceptual states defeasibly and empirically contribute to warrant for perceptual beliefs, and do so on their own, independently of supplementary support from the individual’s background beliefs; perceptual states are elements in an individual’s perspective or point of view that provide empirical starting points for warranted belief about one’s external surrounds. But they do not simply do so; they do so non-accidentally, through or by their natures. Perceptual states, by their nature, empirically warrant perceptual beliefs. And so there must be something about the nature of the state that explains why it contributes to empirical warrant: [A] perceptual state makes a contribution to entitlement that is not derived from positive contributions of other states . . . [Perceptual states can do that because perceptual] epistemic entitlements are goods that partly depend, in a principled or explanatory way, on the nature of the individual’s representational competence or perspective—hence on the nature of the individual’s perceptual states. (Burge 2003: 534)

This second condition separates Burge’s view from simple reliabilist theories of perceptual warrant, as we’ll see in a moment. Burge then puts two and two together: to contribute to perceptual warrant, a perceptual state must have a nature (second condition) that is reliably veridical (first condition); this makes the state, through its nature, a good route to truth and knowledge: Any contribution that a perceptual state makes to perceptual entitlement must depend on its having a nature that is reliably veridical . . . a perceptual state makes a contribution to entitlement that is not derived from positive contributions of other states. An empirical contribution derives from the state’s having a nature, as marked by its representational content, that is

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



reliably veridical. (Burge 2003: 534) [W]arrant requires of warranted perceptual states and competencies that their being part of a good route to truth be explained in terms of their natures. (Burge 2003: 536)

So to contribute to warrant, a perceptual state must be reliable (first condition), and the explanation for its reliability must be associated with the nature of the state (second condition). A perceptual state, to contribute to warrant, must be reliable through its nature. That’s Burge’s core idea: perceptual states contribute to perceptual warrant (on their own, without positive contributions from other states) for they have a nature that is reliably veridical. The first condition puts Burge squarely in the “reliabilist” camp. The second condition falls within the broad traditional foundationalist view that sees something epistemically special or relevant about the contributions perceptual competencies and perceptual states make to perceptual warrant for perceptual beliefs, independently of backing from other warranted beliefs.

5.2.2 Having a nature that is reliably veridical What is it for a perceptual state or competence to be reliable through its nature? That’s the core idea, as I just said. What does it mean? Is Burge somehow embracing the internalist idea that perceptual states contribute to perceptual warrant necessarily or essentially? Is Burge then somehow embracing the idea that perceptual states are necessarily reliable, that they necessarily or essentially contribute to warrant through being necessarily or essentially reliable? Burge rejects both ideas. He resolutely rejects the idea that any perceptual state is, or must be, reliable in all possible circumstances. He grants, consistent with his antiindividualism about perceptual states, the possibility of massive error, and so the possibility of demon world cases. Perceptual states do not essentially warrant beliefs appropriately derived from them for perceptual states are not essentially reliable. Instead of insisting that perceptual states are necessarily reliably veridical in all possible circumstances, could Burge be arguing that perceptual states are necessarily reliably veridical in normal circumstances, and that’s how perceptual states meet the second condition on warrant, combined with the first? Though Burge once asserted that perception was necessarily reliable in normal conditions, for some time he has resolutely rejected that idea too.⁹ Perceptual states need not be reliable, even in normal conditions. So for those that are reliably veridical in normal conditions, they are not essentially or necessarily reliable in normal conditions. What then does Burge mean when he says that a perceptual state must be reliable through its nature to contribute to warrant if not necessarily reliable in normal conditions? What is it for a perceptual state to have a nature that is reliably veridical? What is it for a perceptual state or competence to be reliable through its nature? Burge’s answer derives from his anti-individualist (semantic externalist) account of perceptual states (perceptual representations). It will take a few paragraphs to explain how he uses his anti-individualism to answer our question. ⁹ For discussion of these points, see the Appendix.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

Perceptual states are individuated by their contents, so to explain the nature of a perceptual state type, one explains how the state acquired its capacity to represent the properties—attributes—of external objects. According to Burge’s perceptual anti-individualism, perceptual state types acquire their contents partly through entering into patterns of causal interactions with physical particulars and their attributes distributed across the wider environment. A state acquires its content partly through “wide” patterns of causal interaction— environment-to-perceptual state and perceptual state-to-environment patterns—and partly through “narrow” causal interactions—patterns of interaction involving the perceptual state and other states within the organism. To uncover the explanation for why a perceptual state type has the content it has, look to evolutionary and learning history of the perceptual system as it serves the organism in a broader environment. These patterns thus help form (and so constitute) the natures of perceptual states. These patterns are formative patterns, patterns that constitute the natures of the states. They are explanatorily (and so non-accidentally) associated with the natures (the contents) of representational states. Burge then defines normal environments (circumstances, conditions) as environments where the perceptual system was formed (Burge 2003: 531, 533; cf. Millikan 1984). Normal environments are formative environments. Normal environments are then explanatorily and non-accidentally associated with the natures of representational states. Normal functioning (normal working, operating or well-functioning) is then working or operating the way the perceptual system worked or operated when the state acquired its representational content. Normal functioning is functioning via the “narrow” formative pathways are associated with the formation of the perceptual state type. We can now say what it is for a perceptual state to be reliable through its nature. It is to be reliable in normal conditions when normally formed. Non-accidental, explanatory reliability is reliability through normal functioning in normal conditions.¹⁰ A perceptual state type that is reliable in normal conditions through normal functioning is a state whose reliability is explanatorily associated with the nature of the state, reliability explained through conditions that formed the nature (the content) of the state. What about our perceptual competencies and perceptual states? Are they reliable through their natures, in this sense? Yes. We know empirically that human perception on the whole is reliably veridical in our normal environment when functioning normally. Our perceptual states reliably represent particulars and their attributes in our normal environment when normally formed; our perceptual states (for the most

¹⁰ Reliability in normal conditions is not merely being in normal conditions and being reliable. Reliability in normal conditions entails being reliable through normal (wide and narrow) pathways in the normal environment. An individual organism might be in normal conditions and its perceptual states might be reliably veridical, but if the reliability results from artificial stimulation of the perceptual system or disease or even cosmic accident, then the reliability is not reliability in normal circumstances—it is not reliably explained through the normal narrow and wide pathways, reliability through the normal functioning of the competence in normal conditions.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



part) are reliable through their natures, though pathways explanatorily associated with the formation of their representational contents. Burge asserts being reliable in normal conditions is the only way to be reliable through one’s nature. Being reliable in normal circumstances is the only way to meet the second condition on perceptual warrant while also meeting the first: The . . . reliability that is relevant to the contribution of perceptual states to perceptual warrant is one that attaches to the states’ normal functioning in the conditions that explain their natures. That is the only reliability that is non-accidental relative to the natures of the perceptual states . . . The account of perceptual entitlement must take as privileged the normal environmental conditions that help individuate perceptual states and competencies. For they alone connect reliable representational success to the natures of the states in a principled way. (Burge 2003: 536)

In other words, the only way to explain the reliability for perceptual states that is explained through the natures of perceptual state types is reliability through patterns of formative interaction—viz., reliability in normal conditions when functioning normally. We can now summarize in argument form Burge’s motivation for his account from the two initial conditions on perceptual warrant: (i) Condition one: warrant is a reliable route to truth and to knowledge. (ii) Condition two: perceptual warrant depends on the nature of the perceptual state. (iii) Hence perceptual warrant depends on the perceptual state having a nature that is reliably veridical. (iv) To have a nature that is reliably veridical is to be reliably veridical when occasioned through normal wide and narrow formative pathways. (v) To be reliably veridical when occasioned through normal wide and narrow formative pathways is to be reliably veridical in normal conditions when normally formed. (vi) Hence a perceptual state contributes to warrant when and only when reliably veridical in normal conditions when normally formed. We thereby reach Burge’s view that a perceptual state defeasibly warrants a perceptual belief appropriately derived from the perceptual state when and only when the perceptual state is reliable in normal conditions when normally formed.

5.2.3 A clear consequence: accidental reliability does not confer warrant Burge now has an easy answer to a commonly asked question from the simple reliabilist: why isn’t in situ, de facto reliability enough for prima facie warrant? After all, if a perceptual capacity—even a broadly perceptual capacity like clairvoyance that results from a bolt of lightning—leads to mostly true beliefs, then it is a route to truth for the individual. Why isn’t that enough for warrant? Easy answer: the reliability in question is not associated with the nature of the warranting state or capacity; hence the reliability is “accidental” or “non-explanatory” relative to the nature of the reliable state or capacity. Only the first condition is met, not the second.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

There must be something about the nature of the individual’s capacities or point of view that helps explain the individual’s psychological states’ being a good route to truth. Brute, “accidental” reliability does not suffice for warrant. (Burge 2013: 187–188, emphasis added)

The reliability is accidental, deviant. That’s why one “does not become warranted because a bolt of lightning or some puppeteer sets up a reliable connection to a subject matter” (Burge 2003: 507). So reliable, the state would not be a route to truth through its nature; the state would fail to connect the first condition on perceptual, empirical warrant with the second. Simple Reliablism is wrong for it ignores the second condition on warrant. To put it another way, an exercise of a belief-forming process contributes to warrant only if it is an exercise of a belief-forming competence. A belief-forming competence is reliable in normal conditions when functioning normally. If nothing explains why the process is reliable in normal conditions when functioning normally, then the process is not a competence. If not a competence, then even if it reliably produces true beliefs, it doesn’t produce warrant. Not any kind of reliability will do. Before returning to our question about demon worlds, we can now explain what Burge means by the slogan “warrant is a good route to truth.” By the first condition, warrant is a reliable route to truth; by the second condition, warrant is a good route, a route that non-accidentally and explanatorily connects the warranted belief with reliable truth and thereby conduces towards knowledge. Good routes to truth are routes that contribute to truth in normal conditions when functioning normally. Good routes are explanatorily connected to the nature of the state. Good routes are competent routes, routes that manifest competence. Good reliable routes contribute to truth, reliably so.

5.3 Burge’s Explanation for the Persistence of Warrant in Demon Worlds I now turn to Burge’s answer to our main question: why should warrant persist in demon worlds? If warrant is a good route to truth and knowledge for the individual, then how could perception contribute to warrant in demon worlds?

5.3.1 Burge’s text Here is Burge’s answer from section VI of “Perceptual Entitlement” (Burge 2003: 534–6). I have numbered the sentences in brackets for reference. [1] It is a fundamental feature of perceptual warrant, hence perceptual entitlement, that it allows that an individual can be fooled while retaining warrant. [2] Warrant is retained if the error is accidental, relative to the nature of the perspective. [3] Perceptual errors or unreliabilities that are perceptually indiscernible and derive from brute abnormalities in the environmental conditions that cause perceptual states do not undermine warrant. [4] Both unreliability and reliability in conditions other than those that played a role in explaining the nature of the perceptual state and the exercise of the perceptual competence are accidental relative to those natures. [5] So reliability and unreliability in such conditions are irrelevant to the connection between warrant and veridicality. [6] The only reliability that is relevant to the

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



contribution of perceptual states to perceptual warrant is one that attaches to the states’ normal functioning in the conditions that explain their natures. That is the only reliability that is nonaccidental relative to the natures of the perceptual states. [7] Thus the normal environment— the conditions in which content is explained and established—is privileged in explanation of entitlement. [8] Its privilege derives from the fact that it plays a central role in making the individual’s states what they are—a role that abnormal environments do not play. [9] This privilege entails the irrelevance to entitlement of reliability and unreliability in other conditions. [10] So it extends to indiscernible abnormal environments that the individual might contingently find himself in. (Burge 2003: 536, emphases added)

If you are like me, you found this reasoning difficult. I shall provide a reconstruction that, if successful, would explain why a perceptual state or competence that is reliable in normal conditions would continue to contribute to warrant even in demon worlds. I shall not rely heavily on the latter half of Burge’s paragraph, for that repeats points I just made in the previous section that flow from his conclusion that states must be reliable through their natures to contribute to warrant. Instead I’ll rely on the first half where I think the main ideas reside. You’ll be the judge.

5.3.2 Burge’s argument reconstructed Here is my reconstruction: (1) It is a platitude that warrant is compatible with error, that not all warranted beliefs are true. [1] (2) Warrant is compatible with accidental error, in particular. [2] from [1]. (3) Accidental error is (or includes) error due to abnormalities in environmental conditions (such as a burst of radiation before the eye, or strange lighting conditions, or the presence of distorting mirrors, etc.). [3] (4) So, warrant is compatible with error due to environmental abnormalities. [2], [3] (5) Systematic accidental error (massive error, and so accidental unreliabilities of a perceptual state type) is unreliability due to systematic abnormalities in environmental conditions, such as imagined in demon world scenarios. [3], [4] (6) There is no principled distinction between occasional accidental error (a few mistakes due to abnormalities) and extensive, even systematic, accidental unreliability (mostly mistakes due to abnormalities) that matters to the compatibility of warrant and error (there is no principled difference between some error and massive error in this context; both are compatible with warrant). (7) Warrant is then compatible with massive, systematic accidental error (accidental unreliability), due to systematic abnormalities in environmental conditions, such as demon worlds. [9] (8) So as long as a perceptual state or competence is reliable in normal conditions (reliable through its nature), its warrant conferring power thus “extends to indiscernible abnormal environments that the individual might contingently find himself in.” [6–9] This appears to be a valid argument from plausible premises. If successful, it explains why warrant should persist in demon worlds.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

5.4 Three Criticisms: Doubting Burge’s Explanation Though I agree with Burge’s conclusion, I am not satisfied with this explanation. I shall make three criticisms. The first two criticisms rely on a distinction between two versions of “Burgean” normal circumstances reliabilism: bounded normal circumstances reliabilism versus unbounded normal circumstances reliabilism.

5.4.1 Unbounded vs. bounded normal circumstances reliabilism Taking Burge’s two conditions on perceptual warrant at face value (as well as the rest of his derivation of normal circumstances reliabilism), we still leave open the possibility (if not the plausibility) of a view very close to Burge’s that nevertheless diverges on the issue at hand. Here’s what I mean: granting Burge’s two conditions on perceptual warrant, an epistemologist might advance (what I’ll call) “bounded” normal circumstances reliabilism, where Burge advances (what I’ll call) “unbounded” normal circumstances reliabilism. Here are the two rivals: Unbounded, Normal Circumstances Reliabilism. In all possible worlds W, perceptual beliefs are prima facie defeasibly warranted when, and only when, the perceptual system is reliable in normal conditions when functioning normally. However, warrant does not require de facto reliability in the circumstances of use, only in normal conditions. The subject need not reside in normal conditions, and so the process need not be in situ reliable; the subject’s perceptual states need not be occasioned through wide normal pathways. A massively deceived brain-in-a-vat may then enjoy prima facie warranted beliefs, provided the massively deceived brain-in-a-vat’s perceptual states are reliably veridical in normal conditions. Warrant is not bounded by normal circumstances. Bounded, Normal Circumstances Reliabilism. In all possible worlds W, perceptual beliefs are prima belief defeasibly warranted when, and only when, the perceptual system is reliable in normal conditions when functioning normally. Warrant requires that the individual subject reside in normal conditions, for warrant requires that the perceptual state not only be occasioned through normal narrow pathways (and so the perceptual system is functioning normally) but also through normal wide pathways (so that the subject is in normal conditions), for warrant is a good route to truth and knowledge through the nature of the warranting state. A massively deceived brain-in-a-vat does not enjoy prima facie warranted beliefs, even if the deceived individual’s perceptual states are reliable in normal conditions, for while envatted perceptual states are not reliable routes to truth. Not in normal conditions, the beliefs are not on a narrow and wide route to truth. Warrant is bounded by normal circumstances.

Both bounded and unbounded reliabilism think that accidental reliability won’t do for warrant; mere de facto, in situ reliability is not enough. They agree that warrant is a good and reliable route to truth; the bounded reliabilist accepts Burge’s explanation for why this is so. But bounded reliabilism requires that the perceptual state not only be reliable when occasioned through its formative pathways, it also requires that it be in normal conditions so that it is also in situ reliable, so that as a matter of fact the perceptual state makes the perceptual belief more likely to be true. For the bounded reliabilist, perceptual warrant in situ non-accidentally raises the conditional probability of truth; no raising, no warrant. For the bounded reliabilist, the first condition is just as paramount as the second. Perceptual competencies and states

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



do not, on this view, contribute to prima facie warrant for perceptual beliefs in demon worlds; no contribution to a tendency to form true beliefs in the present environment, no contribution to warrant. In a demon world, though many have the strong intuition that warrant persists, the bounded reliabilist doubts the persistence of warrant. Indeed, taking Burge’s two conditions at face value, the most reasonable conclusion is that warrant does not persist in demon worlds. I raise the possibility of bounded reliabilism not because I believe it is true—on the contrary I side with Burge—but rather because I want to understand why it is false, and understanding why it is false just is to understand why warrant should persist in demon worlds. And so to evaluate Burge’s explanation for why warrant should persist in demon worlds, we should find that Burge’s explanation provides good grounds for rejecting bounded reliabilism. And so I will ask whether it does.

5.4.2 Accidental vs. normal error To evaluate Burge’s explanation, let’s see where the bounded reliabilist might pushback. I think there are two places to pushback: the argument from (1) to (3) for (4), and step (6); the bounded reliablist can deny either (4) or (6) or both. (If you have forgotten my reconstruction of Burge’s argument, you may want to quickly review it again before continuing.) Burge presents (2) as a natural extension, application, or even as a gloss, on (1). I accept (1) as a platitude, as a “fundamental feature” of warrant. But the conclusion (4) from (2) and (3) is not a simple extension of (1). Here’s why. There are at least two kinds of perceptual error. One kind is the kind Burge emphasizes: accidental error, error due to abnormalities in environmental conditions. Abnormal conditions would be conditions dissimilar to the kinds of conditions that were present during the formation of the perceptual state type in the perceptual system; they will be errors occasioned through non-formative pathways. Paradigm cases of error in abnormal conditions involve strange highly unlikely bursts of radiation just before the eye, triggering a perceptual illusion, or cases of irregular colors of light, or distorting mirrors, or new undetectable particles in the air that distort the refraction of light, and so on. These are errors when the state is not on a good route to truth. But error due to circumstantial abnormality is not the only cause or category of perceptual error. There are also perceptual errors that arise through the normal functioning of the belief forming process in normal conditions. For there are cases where the perceptual system, though reliable in normal conditions when functioning normally, is not infallible in normal conditions when functioning normally, so there are bound to be cases where the perceptual system misrepresents the environment while in normal conditions when functioning normally; being occasioned through normal wide formative pathways is not a guarantee of veridicality. These are errors when the state is on a good route to truth. If you find yourself puzzled about this, note first that Burge believes that not all perceptual states and competences are even reliable in normal conditions when functioning normally. He imagines that a rabbit’s perceptual system represents the presence of danger. The rabbit’s perceptual attributive danger is clearly not infallible in normal conditions when normally formed, for it is not even reliable in normal

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

conditions when normally formed. And if there are perceptual state types that are not even reliable when normally formed in normal conditions, then there are sure to be many perceptual state types in a perceptual system that, though they are reliable in normal conditions when functioning normally, are not infallible in normal conditions when functioning normally. Hence there will be perceptual errors that occur when the system is functioning normally in normal conditions. Good and reliable routes are not necessarily reliable, let alone infallible. Indeed, think of nearly any device you’ve ever worked with. Some may indeed be infallible (they always work) in normal conditions when functioning normally. But many are made to be just good enough, so that even when they are functioning normally in normal conditions, they don’t always work. Most can openers I have owned are like that. Textbook cases of perceptual illusions can also be interpreted as errors due to normal functioning in normal conditions. True, some illusions are generated when there is something unusual or abnormal about the circumstances (such as colored lighting or the presence of distorting mirrors), but not all textbook cases of perceptual illusions work like that. Cases like the Müller-Lyer illusion arise when the perceptual system is in normal conditions (the conditions in which the perceptual system was formed) and is functioning normally. The illusion arises because the perceptual system was formed through tradeoffs for accuracy and efficiency. The efficient system results in a highly reliable perceptual system in normal conditions when functioning normally for the most part. But given the built-in biases and formation principles within the perceptual system, there are bound to be errors that occur even when the perceptual system is functioning normally in normal conditions. So the bounded reliabilist is free to distinguish two cases: error that occurs when the perceptual system is functioning normally in normal conditions (errors on a good route), and error that occurs when the perceptual system is functioning normally but in abnormal conditions (errors off the good route). The bounded reliabilist is free to argue that perceptual warrant persists only when all is normal, and so free to agree that warrant is compatible with perceptual error when all is normal, but to deny that warrant persists when conditions are abnormal; just as “narrow” abnormality (malfunctioning, disease, overload) undermines warrant, “wide” abnormality (bursts of light, distorting mirrors) undermines warrant: warrant arises through the normal functioning of the belief-forming process while situated in and through normal conditions. Warrant does not entail truth, but for all that warrant entails a higher chance of truth. The bounded reliabilist can thus assert the fundamental feature of warrant in (1) but deny (2) (in the light of (3)) and so free to deny (4) and thereby stop Burge’s explanation just as it starts to get off the ground. Warrant is not compatible with some accidental error, so it is not compatible with massive accidental error. That’s the first way the bounded reliabilist can accept Burge’s overall point of view—Burge’s two conditions on perceptual warrant—but deny that warrant persists in demon worlds.

5.4.3 Some error vs. massive error I now turn to the second way the bounded reliabilist can pushback. This second way abstracts away from the point just made, viz. the difference due to errors through

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



normal wide pathways and errors through abnormal wide pathways. In other words, it is compatible with accepting Burge’s argument for (4). The second way to pushback questions (6), for doing so is in the very spirit of the bounded reliabilist position. Recall premise (6): (6) There is no principled distinction between occasional accidental error (a few mistakes due to abnormalities) and extensive, even systematic, accidental unreliability (mostly mistakes due to abnormalities) that matters to the compatibility of warrant and error (there is no principled difference between some error and massive error).

Now the bounded reliabilist may agree that she doesn’t know exactly how to draw a principled distinction here, but like the difference between being bald and having a full head of hair, the bounded reliabilist feels confident that there is a difference between occasional accidental error (so that not every warranted perceptual belief is true, even when accidentally caused) and massive accidental error (so that nearly every warranted perceptual belief is false, even when in abnormal and so accidental conditions). Surely there is a difference, the bounded reliabilist asks, between occasional error (due to either abnormalities or normalities in the environment) and massive unreliability? Error once in a while is still compatible with being right most of the time, and so compatible with warrant, for warrant is a reliable route to truth. But error nearly all the time? How could that, the bounded reliabilist believe, be compatible with warrant? Where is the epistemic good in that? How could that certify a belief as true and knowledge? The point I’ve just made suggests we can’t simply infer that warrant may persist in demon worlds from the premise that warrant is fallible. True, warrant does not guarantee truth (some warranted beliefs are false), but that does not ipso facto show that most warranted beliefs may be false (massive error is compatible with warrant). It may show that, but further considerations are surely required to seal the deal. Not knowing how to answer these two ways the bounded reliabilist might push back, I find myself dissatisfied with Burge’s explanation for the persistence of warrant in demon worlds. Burge’s explanation does not explain to me why warrant should persist in demon worlds.

5.4.4 Removing a requirement vs. explaining persistence I think I can go a step further and say more about why I am dissatisfied. What leaves me feeling short on understanding why warrant should persist on Burge’s explanation is its “negative” character. By that I mean that Burge’s explanation works (if it does) by arguing (in effect) that since warrant does not entail truth, it doesn’t entail (require) reliability either (at least reliability in abnormal conditions when the individual is in abnormal conditions). That is, it works by saying there isn’t a condition on warrant that would require warrant in demon worlds. In other words, it removes an obstacle to the persistence of warrant in demon worlds. It doesn’t explain what it is about warrant that persists, or that is clearly present, in demon worlds. It doesn’t identify the “positive” epistemic property that perceptual beliefs clearly possess in demon worlds, the kind of property the bounded reliabilist needs to see to be moved to abandon her position, for it was exactly her inability to see such a positive epistemic property that led her to believe that warrant does not persist in

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

demon worlds. All we are told is that the perceptual state type, because it is reliable through its nature (through narrow and wide pathways), would be a reliable guide to truth were the state “over there” in normal conditions, instead of “here” in abnormal, massively deceptive conditions. But how does that explain why warrant should persist? If warrant is a good and reliable route to truth, and so a reliable route to truth, then how could it persist in demon worlds, where it is not a route to truth at all, where it will never take you to your destination? “[I]t is the individual at a given time for which warrant is supposed to provide a good route to truth” (Burge 2003: 533). It can’t do that in demon worlds. So why should warrant persist in demon worlds? Why should reliability in normal environments mean warrant in abnormal environments? After all this work, we seem to be back where we started. We still don’t know why being reliable in normal conditions should have a force that carries “over to other environments that the individual might form perceptual beliefs within” (Burge 2003: 533). And so I am not satisfied with Burge’s direct answer to our question. More still needs to be said.

5.5 Two Grades of Warrant: Normal Functioning Persists in Demon Worlds To see whether we can answer our central question within Burge’s framework, we need to rethink the weight that’s been placed on the phrase “route to truth” and the role that reliability plays in our overall account of warrant, especially perceptual warrant. This should come as no surprise, for the very idea that warrant requires reliability—that warrant is a reliable route to truth—is the source of the conflict with the idea that warrant might persist in demon worlds. So if warrant requires reliability, we need to reassess what we mean when we say that. Here’s an imaginary conversation between Burge (or someone speaking in defense of Burge) and an interested Interlocutor mutually engaging in this very rethinking: : Warrant is an epistemic good. : Why? : Because truth and knowledge are epistemic goods. Warrant contributes, in an epistemically good way, to truth and knowledge; warrant is a good route to truth and knowledge. : But what about accidental reliability? Why isn’t it a contributor to truth and knowledge? : Though a contributor to truth, an accidentally reliable state does not contribute to truth through its nature. The state’s accidental reliability does not explain why the state, on its own, should make an explanatory, non-accidental contribution to warrant; accidental reliability isn’t a good route to truth. Non-accidental routes exercise and sometimes manifest competence. Nothing counts as exercising or manifesting competence for routes that are but accidentally reliable. “Clairvoyance” cases are ruled out. : But what about a demon world situation? Why should that individual have warranted perceptual beliefs, for perception there is not a route to truth, let

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 

:

:

:

: :

: :



alone a reliable route? Perception there does not contribute to truth and knowledge. Let’s rethink a little. Maybe we’re placing too much weight on the phrase “route to truth.” We should think instead of warrant, as such, as a part of a good route to truth and knowledge, as a constituent or contributor to knowledge, and not as a route to truth and knowledge all on its own (modulo Gettier failures, internal incoherence, and counter warrants). To be a good and reliable route to truth and knowledge, maybe the warranting state and exercise of the competence must be in normal conditions, but the warranting state and exercise of the competence is still a part of a good route to truth even when it is not in normal conditions. Warrant is then part of a good route to truth. Construct the road to the destination and then factor out a part. Call that warrant. What do you mean? Why is the exercise of a perceptual competence in forming a perceptual state part of a good and reliable route to truth, even when the individual is not in normal conditions? How could it be a part of a route to truth? Well, a perceptual state type that is reliable in normal conditions is a perceptual state that is reliably veridical in normal conditions when the perceptual system is functioning normally. That’s why perceptual beliefs appropriately derived from the normally formed state and are reliably true in normal conditions, non-accidentally so. The way a perceptual competence contributes to truth and knowledge is by being reliable in normal conditions when the perceptual system is functioning normally. The perceptual state type is a good route to truth through its nature (a) when normally formed, (b) reliable-in-normal-conditions, and (c) in normal conditions, and so in situ reliable through “narrow” and “wide” normal pathways. I understand all that. But what has this got to do with warrant while in a brain-in-a vat or demon scenario, when the individual is not in normal conditions? Once we have a good, explanatory pathway from competence to truth, we can factor the way a perceptual competence contributes to truth and knowledge into two elements. One: formation of the perceptual state and perceptual belief along “narrow” formative pathways, viz. normal functioning of the perceptual system and the transition to perceptual belief. Two: formation of the perceptual belief along “wide” formative pathways, viz. the normal functioning of the perceptual system in normal conditions. A good route runs on both parts. Yes, I remember you said all of that when you were explaining what it was for a perceptual state within a perceptual system to be a reliable guide to truth through its nature. When both parts obtain, a perceptual state is a good route to truth and conduces to knowledge. So each part contributes to a good route to truth and knowledge; each part on its own is a contribution to a good route to truth and knowledge. Since we can factor these two contributions—normal

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

:

:

:

:

:

functioning (formation along narrow formative pathways) and being in normal conditions (formation along wide pathways)—we can see that the first dimension—normal functioning—is one part (one factor) of the overall good route to truth. So normal functioning (in the formation of the perceptual state and the transition to perceptual belief) is part of a good route to truth and knowledge. (Warrant contributes to fulfilling representational functions; you can contribute to a goal without necessarily, or even reliably, achieving that goal). Like I said, build the road to your destination and then factor out a part. That part, like the other, contributes to a route to truth and knowledge. OK, so normal functioning when the state is reliable in normal conditions is a part of a good route to truth and knowledge: it is not a good route to truth all on its own. But why does that explain why warrant persists in demon world scenarios? That’s easy. In the vat fooled by a scientist, or fooled by a demon, or even in undetectably non-normal circumstances, the formation of the perceptual state by the perceptual system and the transition to perceptual belief may still function normally. You don’t have to be in normal conditions for your perceptual system to function normally. For example, normal conditions for a car involve the car sitting comfortably on the road so it can take you from A to B. But you can take your car into the mechanic’s shop and raise it up off the ground for inspection. The mechanic may find that the major systems in the car work just fine—they are functioning normally—even if the car isn’t on the road and so isn’t in normal circumstances. The brainin-a-vat scenario is a pretty clear parallel. Your brain and visual system can function normally in such circumstances, even if you end up forming entirely false visual beliefs as a result. The normally functioning perceptual system won’t know the difference. Normal functioning can persist in demon world scenarios. I see that. I remember Peter Graham using the car analogy in his paper “Epistemic Entitlement.” But I still don’t get it. Since reliability lapses, why does warrant persist? Shouldn’t it lapse? Warrant persists because normal functioning persists, provided warrant consists in the normal functioning of the perceptual belief-forming process, provided the process is reliable in normal conditions when functioning normally, which is what I am now claiming. That’s why warrant persists in demon world scenarios. Warrant isn’t as such a good route to truth and knowledge; warrant as such contributes to truth and knowledge by being a part of a good route to truth and knowledge; warrant is part of a good route to truth and knowledge. Hmm, so warrant consists in the normal functioning of a perceptual competence? But if warrant consists in normal functioning, then why do you need to add the qualification “provided the state is reliable in normal conditions”? Why should that matter to warrant? Why shouldn’t a normally formed perceptual state that isn’t reliable in normal conditions also warrant an appropriately derived perceptual belief in a demon scenario?

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



After all, in both cases, the perceptual system is functioning normally. Why not go internalist, as it were? : Good question. Though now I’ve claimed perceptual warrant as such isn’t on its own a good route to truth, but only part of a good route to truth, you’re right to ask why normal functioning alone isn’t enough for warrant. I think the answer is easy. If a perceptual state or system isn’t reliable in normal conditions, then the normally functioning system isn’t part of a reliable route to truth, even in normal conditions. Hence if a perceptual state or system isn’t reliable in normal conditions, normal functioning isn’t itself a part of a good and reliable route to truth and knowledge; it is not a part of a reliable route to truth through its nature. For warrant to consist in the normal formation of the state and the transition to perceptual belief—normal functioning—then the state must be reliable in normal conditions, for then it is part of a reliable route to truth and knowledge through its nature. : I get it now. You’ve in a way retreated to a weaker conception of warrant, where it necessarily connects to truth in normal conditions, but not necessarily outside of normal conditions. If normal functioning in normal conditions didn’t reliably result in true beliefs, then normal functioning wouldn’t warrant in any conditions, for it wouldn’t be part of a good and reliable route to truth. Warrant consists in normal functioning provided the process is reliable in normal conditions, even if the process isn’t in normal conditions.¹¹

¹¹ One might object to this interpretation of Burge, as holding that warrant consists in the normal functioning of the transition from a reliably veridical perceptual state to perceptual belief, and the normal formation of the perceptual state by the perceptual system, for in “Perceptual Entitlement” Burge seems to say that normal functioning in the formation of the state by the system is not required for warrant in footnote 24 (pp. 537–8): “There is a subtle question here about malfunction. The question is whether the entitlement lapses if the system malfunctions in the particular circumstances, but the perceptual system and the particular perceptual state in the individual remain generally reliable. Suppose that, on one occasion, in the normal environment, a ray of the sun, included in the stimulation of the retina, causes a momentary neural malfunction that produces a perceptual state inappropriate to the environmental circumstances, and that the individual could not discern this case from a normal case of having the perceptual state-type . . . Would the individual’s epistemic entitlement lapse? . . . On the one hand, the particular path that normally leads to the perceptual state has been circumvented. This suggests that the entitlement might be undermined. For the functioning that supports reliable belief is not present. On the other hand, the perceptual state-type is what the entitlement derives from. It is produced by a reliable perceptual system; it is itself reliable in normal conditions, and has a nature that normally connects it to appropriate perceptual objects. This suggests that the entitlement remains intact, in something like the way it does in the case of brute error. An unlucky token event blocks knowledge but not warrant” (Burge 2003: 537–8). Burge then distances himself in an earlier footnote from “proper functioning” views like Plantinga’s on this basis: “There are cases in which a state is warranted in my sense, but is not even an instance of proper functioning (relative to the function of producing true beliefs) in his” (Burge 2003: 509, n. 6). So it seems that there’s pretty strong textual evidence that this can’t be Burge’s view. Even so, I’m pretty sure this is Burge’s view. Firstly, it makes sense of all the texts, especially the post “Perceptual Entitlement” writings. Secondly, Burge appears unsure about the issue, even in the footnote. Thirdly and most importantly, Burge tells me in personal correspondence that he realized later that the footnote was mistaken. Burge really does hold a normal or “proper” functioning account of warrant, provided the normally formed perceptual state type is reliable in normal conditions.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

In order to reconcile the apparent conflict between your first condition on warrant and the persistence of warrant in demon worlds, you’ve made use of what’s required to satisfy the second condition (perceptual states contribute to warrant through their natures, and so contribute to truth and knowledge when normally formed in normal conditions) to reassess what counts as satisfying the first condition (warrant is a now a part of a good route to truth and knowledge). Pretty clever. Can I make a suggestion? Instead of “re-thinking” warrant so that warrant consists in normal functioning, provided the competence is reliable in normal conditions (even if the competence is not in normal conditions), why not divide warrant into two grades? The first grade would consist in normal functioning, as you are now framing warrant, and the second would consist in normal functioning plus being situated in normal conditions, so that the process, as a matter of fact, is indeed reliable in the circumstances, so that warrant is, as a matter of fact, clearly contributing to truth and knowledge? The first grade constitutively depends on the second grade (the first grade would not be a grade of warrant unless it conduced towards truth in normal conditions), but the first grade can persist beyond normal conditions. The first grade would be “unbounded,” the second grade would be “bounded.” The first grade would contribute to warrant’s function of conducing truth and knowledge, while the second would, absent Gettier failures, internal incoherence, and so on, really would conduce truth and knowledge. How does that sound? : I like the way it sounds. Maybe that’s what I was thinking all along, an implicit conception that may have been guiding my thinking. Hmm, maybe we even need a further grade, to capture Planting’s use of ‘warrant.’ Someone should write a paper called “Two (Maybe Three) Grades of Warrant.” : Hmm, not a bad idea. And now that we’ve had this discussion, my memory of Peter Graham’s paper “Epistemic Entitlement” is coming back to me. He offers a proper functionalist view that also purports to explain why warrant persists in demon worlds. I think he’s even said there are two grades, or two kinds, of warrant or entitlement, in at least one or two of his papers. But unlike you, I don’t think he relies on the thesis that it is a priori that perceptions and beliefs have representational functions (an idea we haven’t had a chance to talk about yet). Instead he tries to derive their functions from their history. I think he also then claims directly that the competence has the function of reliably inducing true beliefs, without arguing that the states or beliefs themselves have functions. He then, if memory serves, argues that normal functioning by its nature is constitutively associated with reliably inducing true beliefs, when the system has forming true beliefs reliably as a function. What do you think about these differences? Where do you think Graham goes wrong, if he does? : Well, I’m glad you’ve asked. I think he makes a number of grave mistakes. Just let me get a glass of water to clear my throat, and then I’ll be happy to discuss those at length.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



: My goodness, is that the time? I’m afraid I have to go. Thank you so much for bearing with me for so long. I’m sure we’ll have a chance to continue our conversation on another occasion.¹²

Appendix. Misunderstanding Perceptual Anti-Individualism Some wrongly read Burge’s perceptual anti-individualism as the thesis that perceptual states are necessarily reliable (reliable in all possible circumstances). This is not correct. Perceptual states can be massively in error outside of normal conditions. That’s the brain-in-a-vat scenario, provided the perceptual system previously interacted with entities in a broader environment so as to underwrite perceptual representational content in the first place: “In principle, any perceiver could be placed in a situation in which any given perceptual state-type could be regularly mistaken—if the perceiver already has perceptual states with definite content. In other words, every perceptual state type would be unreliable in a variety of possible abnormal environments that are perceptually indiscernible, at the time of a given perception, from the normal environment” (Burge 2003: 535; 2005). “There is no metaphysical necessity that [a] perceiver be in an environment that makes perception reliable” (Burge 2013: 357; cf. Burge 1986). One source of misunderstanding derives from misunderstanding the externalist slogan “contents are causes.” According to a crude reading of the slogan, the content of a token perceptual representation just is the token distal cause of the representation; as such perceptual representations are at least referentially infallible if not also attributively infallible. This is far from Burge’s understanding of the slogan. According to Burge, the attributive content (the red or moving part of the content) of perceptual representations derives from patterns of causal interaction between the perceptual system and the wider environment; representational types are partly fixed by types of causes. Burge does not hold that perceptual representations are either referentially infallible nor attributively infallible. Far from it. (See Burge 2005 for elaboration.) Does Burgean perceptual anti-individualism entail that perceptual states are necessarily reliable in normal conditions? Burge once argued for an affirmative answer. Here are some passages from papers from the 1990s that are included in his recent collection (2013): I believe that a perceptual system in any agent . . . is constitutively associated with reliable perceptions in a range of cases. (2013: 79, n. 11) I think that being reliable in normal circumstances can be shown to be necessary to a capacity . . . to register perceptual appearances . . . (2013: 294, emphasis added) There is no metaphysical necessity that the perceiver be in an environment that makes perception reliable (though one’s perceptual system must have been formed in

¹² I have presented earlier versions to audiences at the University of California, Irvine; University of Copenhagen, Denmark; University of Geneva, Switzerland; Oxford University; University of Edinburgh; National Chung Cheng University, Taiwan; National Ching Huang University, Taiwan; the University of California, San Diego; and Notre Dame University. I benefitted from the discussions on all of these occasions. I recall helpful comments from Robert Audi, Sven Bernecker, Tyler Burge, David Brink, Nevin Climenhaga, Pascal Engel, Elizabeth Fricker, Alvin Goldman, Mikkel Gerken, Liz Jackson, Andrew Moon, and Ralph Wedgwood, in particular. I am grateful to Zach Bachman for comments on the penultimate draft.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

such an environment). (2013: 357, emphasis added) Being a perceiver necessarily involves certain reliabilities in perceiving normal perceptual objects in normal circumstances. (2013: 347, emphasis added) But he abandoned this in “Perceptual Entitlement”: An individual can have perceptual beliefs, but no warranted ones. The nature of perceptual states does not require that they be reliably veridical in their normal content-determining environment. . . . I have been misunderstood to hold . . . that it is apriori and constitutively true that perceptual systems are reliable in normal circumstances—the circumstances in which their contents were formed. I do not hold this view, and consciously wrote around it in [“Perceptual Entitlement”]. (2013: 202, n. 89, emphasis added; cf. 2013: 262, n. 14) Burgean perceptual anti-individualism does not entail that perceptual states are necessarily reliable in normal conditions. Perceptual anti-individualism requires explanatory connections with a wider environment for perceptual content. Whether those patterns of interaction underwrite reliable or unreliable perceptual state types depends on how the perceptual system in the organism uses those perceptual states and other contingent features of the organism’s normal environment. For example, as I have noted in the chapter, perceptual representations of danger in some animals like the rabbit are not reliable in the organism’s normal environment. Burge noted this possibility in 1996: A tempting oversimplification is to claim that . . . constitutive veridical causal relations are always reliably veridical. Such a claim is tempting because in so many cases or perceptions are reliable . . . [T]he claim is oversimplified because some perceptual intentional types in some systems are established through the systems’ reliable avoidance of false negative rather than through their reliable achievement of true positives. It is more critical to a hare’s perceptual system that it not fail to register a predator when one is there than it be reliable in its registrations of predators. (Burge 1996: n.11) He later concluded that the point generalizes: An individual can be perceptually wrong more often than right even in its normal environment. The value of veridicality may pay for many errors. One veridical representation of a predator may pay for a lifetime of false positives. (Burge 2005: 5) Representation can in principle be quite unreliable, even in normal circumstances. An animal’s representation of danger might be reliably inaccurate but still serve the animal’s biological needs. (Burge 2010: 299–300) For the most part rabbits accurately represent danger when danger is present, but more often than not inaccurately represent danger as present when it is not. Danger is not a reliable perceptual state type within the rabbit’s normal environment. Various creatures in the animal kingdom might have a score of perceptual attributives that are not reliably applied to entities in their normal environment when the systems are functioning normally. Lobsters, covered with armor with no natural predators, may have little use for reliable perceptual competences. A perceiver may possess no reliable perceptual attributives, but still a perceiver for all that. I conjecture that Burge would also accept the following: even if a perceptual state type in an organism is reliable in normal conditions, it would still be the same perceptual state type even if, counterfactually, it was formed and explained in an environment where the perceptual state type is not reliable. Take a perceptual attributive F in a perceptual system that represents the attribute F-ness. Suppose the representational state was formed through input-output patterns

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 . 



of interaction FP-1 with entities distributed across the wider environment. As a contingent matter, the attributive F in this system was formed in this manner. The state, so formed, may be reliable in that environment. But the very same perceptual attributive type in another system, or even the very same system in a counterfactual setting, might have been formed through different input-output patterns of interaction FP-2 with the environment, possibly a very different environment, where the state type is not reliable. Though explanatorily associated with the nature of perceptual states, particular formative patterns and particular formative environments are not necessarily associated with perceptual state types. Perceptual state types, when they are reliable in normal conditions, are only contingently so reliable. Consider again the rabbit. There may be rabbits whose danger representations are reliable in their normal environments, just as there are rabbits with danger representations that are not reliable in their normal environments.

References Bergmann, M. 2006. Justification without Awareness. Oxford: Oxford University Press. Burge, T. 1986. Cartesian Error and the Objectivity of Perception. In Subject, Thought, And Context, eds. P. Pettit and J. McDowell. Oxford: Clarendon Press. Burge, T. 1996. Our Entitlement to Self-Knowledge. Proceedings of the Aristotelian Society 96: 91–116. Burge, T. 2003. Perceptual Entitlement. Philosophy and Phenomenological Research 67: 503–48. Burge, T. 2005. Disjunctivism and Perceptual Psychology. Philosophical Topics 33: 1–78. Burge, T. 2010. Origins of Objectivity. Oxford: Oxford University Press. Burge, T. 2013. Cognition through Understanding: Self-Knowledge, Interlocution, Reasoning, Reflection, Philosophical Essays, Volume 3. Oxford: Oxford University Press. Cohen, S. 1984. Justification and Truth. Philosophical Studies 46: 279–95. Cohen, S. and Lehrer, K. 1983. Justification, Truth, and Coherence. Synthese 55: 191–207. Cruz, J. and Pollock, J. 2004). The Chimerical Appeal of Epistemic Externalism. In The Externalist Challenge, ed. R. Schantz, 125–42. Berlin: De Gruyter. Foley, R. 1985. What’s Wrong with Reliabilism? The Monist 68: 188–202. Gerken, M. 2018. The New Evil Demon Problem and the Devil in the Details. In The Factive Turn in Epistemology, eds. A. Carter, E. Gordon, and J. Jarvis. Cambridge: Cambridge University Press. Goldman, A. 1979. What is Justified Belief? In Justification and Knowledge, ed. G. Pappas. Dordrecht: D. Reidel Publishing Company. Goldman, A. 1986. Epistemology and Cognition. Cambridge, MA: Harvard University Press. Goldman, A. 2011. Toward a Synthesis of Reliabilism and Evidentialism? Or: Evidentialism’s Problems, Reliabilism’s Rescue Package. In Evidentialism and Its Discontents, ed. T. Dougherty, 254–280. Oxford: Oxford University Press. Graham, P. J. 2010. Testimonial Entitlement and the Function of Comprehension. In Social Epistemology, eds. D. Pritchard, A. Millar, and A. Haddock, 148–74. Oxford: Oxford University Press. Graham, P. J. 2011. Perceptual Entitlement and Basic Beliefs. Philosophical Studies 153: 467–75. Graham, P. J. 2012. Epistemic Entitlement. Noûs 46: 449–82. Graham, P. J. 2014a. Against Inferential Reliabilism: Making Origins Matter More. Philosophical Analysis: Journal of the Korean Society for Analytical Philosophy 15: 87–122. Graham, P. J. 2014b. Functions, Warrant, History. In Naturalizing Epistemic Virtue, eds. A. Fairweather and O. Flanagan, 15–35. Cambridge: Cambridge University Press.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



       ?

Graham, P. J. 2014c. Against Transglobal Reliabilism. Philosophical Studies 169: 525–35. Graham, P. J. 2016. Against Actual World Reliabilism: Epistemically Correct Procedures and Reliably True Outcomes. In Performance Epistemology: Foundations and Applications, ed. M. A. Fernandez. Oxford: Oxford University Press. Graham, P. J. 2017. Normal Circumstances Reliabilism: Goldman on Reliability and Justified Belief. Philosophical Topics 45: 33–61. Graham, P. J. Forthcoming. What is Epistemic Entitlement? In Virtue Theoretic Epistemology, eds. C. Kelp and J. Greco. Cambridge: Cambridge University Press. Henderson, D. and Horgan, T. 2011. The Epistemological Spectrum: At the Interface of Cognitive Science and Conceptual Analysis. Oxford: Oxford University Press. Leplin, J. 2009. A Theory of Epistemic Justification. Berlin: Springer. Lyons, J. 2009. Perception and Basic Beliefs: Zombies, Modules, and the Problem of the External World. Oxford: Oxford University Press. Lyons, J. 2013. Should Reliabilists Be Worried About Demon Worlds? Philosophy and Phenomenological Research 86: 1–40. Millikan, R. 1984. Language, Thought, and Other Biological Categories. Cambridge, MA: MIT Press. Pollock, J. 1984. Reliability and Justified Belief. Canadian Journal of Philosophy 14: 103–14. Majors, B. and Sawyer, S. 2005. The Epistemological Argument for Content Externalism. Philosophical Perspectives 19: 257–80. Sosa, E. 2001. Goldman’s Reliabilism and Virtue Epistemology. Philosophical Topics 29: 383–400. Wedgwood, R. 2002. Internalism Explained. Philosophy and Phenomenological Research 65: 349–69.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

PART II

Extending the Externalist Project

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

6 Knowledge, Default, and Skepticism Ernest Sosa

Virtue epistemology takes its own approach to the questions of traditional epistemology. What follows will focus on a distinctive notion of default assumptions, which enables a fresh treatment of philosophical skepticism. The response to the skeptics will explain how they’ve mistaken what’s required for the epistemic quality of ordinary judgments and beliefs. This treatment of skepticism is enabled by an analogy between epistemic and athletic performance, and between episteme and praxis more generally.¹ Our virtue epistemological approach relies on a category of default assumptions that is different from any “default justification” or “entitlement” already in the literature, if only because ours is embedded in virtue theory and is to be understood thereby.² Wittgenstein comes closest in On Certainty, though his own ideas are also unhinged from any broader virtue epistemology.

6.1 What Lies Ahead The epistemology here is focused on knowledge itself, not primarily the word, nor the concept.³ It is knowledge-friendly in that it takes the phenomena of epistemology to ¹ And this is really more than an analogy: telic virtue epistemology takes judgment and judgmental belief to be forms of praxis, of intellectual praxis. ² Two comments about this sentence: First, a stylistic observation about the occasional use of the first person plural and its grammatical variations, in this sentence and elsewhere in “our” main text. My use of that device is meant to encompass my auditors or readers, so that for example ‘our virtue epistemological response’ denotes the virtueepistemological response that will be of interest to “us,” meaning to those who do or will consider it. Secondly, in writing that sentence I have had in mind mainly the work of Tyler Burge, Fred Dretske, Peter Graham, Christopher Peacocke, and Crispin Wright, whose respective ideas of “entitlement” or “default justification” bear surface similarities to the one developed in our main text (as in their work cited in our bibliography). Despite those surface similarities, the proposal here is quite different, as it is deeply embedded in virtue theory, which bears on the assessment of epistemic performance (such as judgment and belief ) only as a special case. In our virtue theory such default assumptions are in place across domains of human performance generally. ³ The discussion to follow will focus on one main side of epistemology: its judgmental, reflective side. But the teleological orientation of our account and its telic normativity can be extended to the epistemology of functional, animal knowledge. The guiding concepts here are the telic concepts of attempt, success, Ernest Sosa, Knowledge, Default, and Skepticism In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0006

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

be best delineated through our concept of knowledge. Truth is important, but it will not do the job on its own. We would of course like to understand the ways in which human beings harvest truth so as to flourish. But that project is distinct from the project of understanding the nature, limits, and value of human knowledge, the project of epistemology that goes back to Plato’s Theaetetus and Meno. Our inquiry is in this latter line. Let’s begin with a question.

6.2 What is a Background Condition? Athletic attempts have a likelihood of success normally determined by the athlete’s Skill, Shape, and Situation, or “triple-S profile.” The success of an attempt is creditable only if sufficiently owed to the agent’s pertinent triple-S competence. In managing their performances, athletes must heed their level of skill, along with how tired they are, how far from the target, and so on, for the various skill, shape, and situation factors that are known to affect performance. Still, many factors they can properly ignore. In a night game, for example, they can ignore the lighting system, even when the lights are in fact poised to fail without warning. As an athlete, one is not negligent for ignoring such factors, with a default assumption that they are of no concern, absent indication to the contrary. What would be negligent is to reduce one’s attention to the ball and the field so as to ponder such imponderables. As for the relevant background conditions, one need only assume that they are in place, which one can do if no defeater intervenes, such as flickering lights, or an alert over the PA system. Background conditions come in three sorts, corresponding to the triple-S structure of competence. They can concern either the skill, or the shape, or the situation of the performer. Archers can retain their skill, for example, even when drunk or in the dark. A background condition is one that must hold if the relevant S factor is to be in place at the time of performance. The presence of the pertinent skill, shape, or situation will thus entail respective background conditions. What puts such conditions in the background is that, although they must hold, one can perform aptly without knowing that they hold. Nor need they hold safely. Background conditions thus pertain to competences generally. When an agent performs, in whatever domain of performance, it is often the case that the relevant background conditions might easily have been missing. They might have been present just by luck and unbeknownst to the agent, with no detriment to the quality of that performance, nor to the credit earned by the agent for its success.

6.3 Radical Skepticism and Relevant Alternatives Our account enables a response to the skeptic independent of the modal remoteness shared generally by radical skeptical scenarios. competence, aptness, and achievement, all of which span the distinction between attempts that are consciously intentional and those that are functionally teleological.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Consider again the quality of an athlete’s performance and the credit earned for its success. As noted, imminent background danger might have zero bearing on either quality of performance or credit for its success. Nor need the agent rule out all such danger, some of which may be “irrelevant.” Analogously, epistemic agents may not be required to rule out irrelevant skeptical alternatives, apart from assuming that no such danger will be realized in fact. Take for example the “brain in a vat” scenario. The actuality of that scenario would of course affect one’s epistemic performances, just as the lights going out would affect the performance of players in a night game. But the merely modal fact that the scenario might easily occur would seem harmless to the quality of epistemic performance, just as the merely modal fact that the lights might easily fail is harmless to the quality of athletic performance. Nor need one knowledgeably rule out that skeptical scenario, any more than nighttime athletes need knowledgeably rule out the malfunction of the lights. Our initial response to the skeptics is that they’ve mistaken what’s required for the epistemic quality of our ordinary judgments and beliefs. Their mistake is like disparaging a superb baseball catch as due to discrediting luck because, unbeknownst to the fielder, the lights could so easily have failed.⁴ A distinctive response to the radical skeptic is thus enabled by our analogy between athletic and epistemic performance, and between episteme and praxis more generally. Our approach also suggests a way to understand the appeal to “relevant alternatives” in responses to skepticism of years ago. At that time, the notion of relevance remained obscure, where the present approach might now shine its light. In our framework, an alternative is relevant if and only if it is one that the subject needs to rule out specifically (either through explicit judgment or through implicit belief ), as its absence cannot be taken for granted by default. I take that notion to apply pervasively in human life, and to epistemic performance as a special case.⁵

6.4 Adversity Ahead Our analogy encounters an objection. Consider the fielder’s belief that he will make the catch if he tries. Suppose this belief to be no less apt than is the catch itself, despite the grave danger posed in each ⁴ And that’s also how one might go astray in denying that Barney can know they see a barn, when in fake barn country. ⁵ Moreover, when restricted to human epistemic performance, our background conditions are also akin to Wittgenstein’s hinges. For a discussion of the relevant alternatives approach to skepticism and to the analysis of knowledge, see section 5.3 of the article on the analysis of knowledge in the Stanford Encyclopedia, by Matthias Steup and Jonathan Ichikawa. (References are found there to the relevant work of Stine, Goldman, and Dretske.) Our present approach also entails a crucial distinction between safety and relevance, but there is no real rapprochement between the two approaches. In my view, safety is not required for knowledge nor for apt performance generally. A performance of a nighttime athlete can be apt even if extremely unsafe (because of how unsafe is the lighting system). And a similar conclusion may be drawn about epistemic performances, where again grave danger, great risk of failure, seems compatible with success that is apt. Only the danger involving “relevant” alternatives can deny us knowledge.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

case by the fragility of the lights. If knowledge is apt belief, that stadium belief then amounts to knowledge. But this seems absurd, given how probable it is (nearly certain, by hypothesis) that the lights will fail. We thus face a reductio of the virtue-theoretical thesis that apt belief is knowledge. Consider the analogy between the fielder’s attempted catch and his belief that he will succeed if he tries. Suppose the catch is apt despite how lucky it is that its background conditions hold. Suppose the fielder’s belief that he will succeed if he tries is similarly apt, despite the parallel luck in the holding of its background conditions. Since that belief is then an apt belief, our virtue theory declares it a case of knowledge. But, again, that verdict is very hard to sustain. We need a step back to reassess.

6.5 The Place of Default Reconsidered Suppose we awaken some morning with our body of stored knowledge containing its multitude of stored beliefs that we would severally endorse. That would be as normal as daybreak. But what if an alien force could have recently spoiled or thwarted our epistemic faculties? Suppose it came to a coin flip whether we would all be placed in a Matrix setting. We’re supposing that we were actually in close danger of that Matrix disaster, but that the coin flip turned out favorably for us, and the alien force moved away harmlessly. Once we knew all this, how would we properly react? Would we say: “right now we no longer know nearly anything”? After all, even if post-Matrixing beliefs matched earlier beliefs, this could only be by coincidence, not through retentive memory, as (by hypothesis) the later contents then come from the Matrix. Here let us assume a radical Matrix scenario wherein victims’ brains are envatted and massively manipulated, with all their beliefs removed and new ones inserted arbitrarily, in one fell swoop.⁶ The point is that the post-insertion beliefs are related to the pre-insertion beliefs through nothing like retentive memory. Instead of retentive memory, what is operative is just the random procedure implemented by the Matrixers. Does the mere danger of that epistemic disaster remove our ability to know as we normally do? Note well: in hosting our later beliefs we run the great risk of having been thus Matrixed. But this mere risk could hardly entail that massive loss of knowledge. Rather, even in full awareness of the danger we had lived through, we would still have attributed to ourselves the knowledge that we attribute in a normal morning unaffected by such danger. What is more, we would still have attributed to ourselves our body of background knowledge and I say we would have done so properly and correctly. We could even suppose that rather than a coin flip, the aliens ran a lottery, only by winning which would we have been allowed to retain our body of standing beliefs. ⁶ Such massive manipulation is supposed for simplicity and effect. But even highly local manipulation will still give us a radical enough skeptical threat. And such local manipulation will not be blocked by externalist assumptions concerning theory of content or other plausible enough views concerning language or mind.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Otherwise they would have thoroughly fooled us by Matrixing. Suppose we are informed by NASA, which provides powerful evidence that we had been besieged by those aliens and had luckily won their lottery. Should we conclude that our body of standing beliefs no longer constitutes knowledge? Given the picture that we would then have from NASA, if our body of standing beliefs is indeed largely true, this is a massive bit of luck. Only by luck are those beliefs largely true, as it was far more likely that our body of standing beliefs would have been massively false. Only by luck were we not radically Matrixed. That is what we then believe, if we are convinced by NASA, as we are by hypothesis. Imagine yourself in that situation. Is it really an option to abandon the whole world view that we live by, by means of which we place ourselves in a surrounding jointly shared world, and based on which we act at every turn? Surely that is not an option. Must we then say that, despite our lack of any feasible alternative, continuing faith in our world view is wrong or inappropriate? Are we doomed to such incoherence? Not if we can find an alternative account of appropriateness, such as the account based on our default assumptions. On that account, the quality of our epistemic performances seems unaffected by the NASA-recognized danger, in just the way nighttime athletic performance is unaffected by grave danger that the lights will fail.⁷ Full aptness is important on both sides of the episteme/praxis divide. Skilled athletes need to assess how likely they are to succeed, and they must be guided thereby. In properly making that assessment, moreover, subjects are allowed to rely by default on the relevant background conditions, absent specific reasons for doubt. Why would that not be so on both sides of the episteme/praxis divide?⁸ How could it not be so if the episteme side is just a special case of the praxis side?

6.6 Default Assumptions in Domains of Performance Agents in a performance domain aim at certain outcomes with their movements and affirmations (a baseball catch, say, or a soccer goal). We aim at attaining those ⁷ It might be argued (with some initial plausibility) that the very knowledge provided by NASA acts as a sort of defeater-defeater that restores our background knowledge in face of the knowledge that we ran such dire epistemic risk. However, this involves complexities that would need to be pondered. (For example, how can we take seriously the testimony of NASA, if it undermines itself, given that if there was indeed such massive luck in the workings of our retentive memory, including the mnemonic faculties of the NASA personnel, how then can we even be sure that NASA is really to be trusted? Aren’t we now left with suspension as the only epistemic option, as against our supposed knowledge?) And in any case the supposed remedy seems insufficient when we reflect that our whole body of standing background knowledge would still have been lost even if the interlopers had been undetected by NASA. Does this not still seem absurd? Note well: given the dependence of agential normativity generally on its basis in the knowledge of agents, all of life across the world would have lost its agential worth with any such massive loss of background knowledge. ⁸ The kinship between some at least of Wittgenstein’s hinges and our own background conditions comes again to the fore. Both are assumed by default to buttress our coherent commonsense perspective as well as the particular judgments that we make from that perspective. Hinges can thus enable our body of knowledge without our knowing them to be true, and without our reaching them as conclusions (through “ratiocination”). As with hinges, background conditions can be assumed to obtain, with assumptions that stand fast for us, and can properly do their enabling even when they are just assumed to be true, an assumption that is both unsafe and in ignorance.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

outcomes not just by luck but by competence. We aim at the apt success distinctive in the given domain, through corresponding triple-S competences. And these competences allow, and require, proper default assumptions, even when these are unsafe and not known to be true, as in our athletic examples. It is part of the competence of nighttime athletes to focus on their athletic performance, while ignoring whether the lights might fail. It is part of athletic competence to take for granted background conditions like the stability of the lights. Compare what pilots can take for granted with no negligence, and how this affects credit for success. Pilots have a check list to run through prior to takeoff, but they need not kick the tires. A pilot can trust others on the airline staff to have checked the tires. Still, it is not only what pilots know with safety that they can trust without negligence to be true. They might trust without negligence that no terrorism is in the offing even when it is a near certainty. If by luck the terrorist misses that flight, so that disaster is avoided, the mere danger involved casts no doubt on the quality of that pilot’s performance, nor does it reduce in the slightest the credit earned for successfully piloting the plane to its destination. However, the pilot’s performance would not have retained that quality if the pilot had known about a credible threat of terrorism, so that he should have aborted the flight; which takes us to a more general claim covering ordinary performance. There is a commonsense, ordinary domain, beyond more specific performance domains, be they athletic, artistic, professional (such as piloting, medicine, or the law), and so on. Even when we perform in more specific domains, we constantly also perform in the domain of commonsense practice that coexists with specific domains. Take the physical movements whereby we make our moves in a chess match. These might also be aimed at numerous and varied ends, both internal and external to chess. Thus, one might use one’s left hand despite being right-handed, so as to avoid the arthritic pain of a right-hand move. Intentionally protecting the right hand, in order to avoid the pain, is an ordinary performance that is not strictly a chess move. What of human attempts proper to such ordinary praxis? Do background conditions and default assumptions have a place in the competences of that broader domain? An athlete can properly assume by default that the lights will stay on, as he considers whether to run for a fly ball, or to dribble towards the soccer goal, and he will attain full credit even if his attempt attains success only by luck, since only by luck do the lights stay on. What is more, even while knowing that the lights are fragile, the athlete can still rely on the assumption that they will not fail. No matter how unsafe his performance is at that point, no matter how likely it is that his attempt would fail with the failure of the lights, the performance can be fully apt despite the known risk involved.⁹ ⁹ This still seems plausible even if there might be special circumstances where knowledge that the lights are so likely to fail might have some proper influence on the best choice of play. However, the knowledge that the lights will almost certainly fail, so that the fielder’s all-out running attempt to catch that flyball is almost certain to fail, has nearly zero effect on the aptness and the full aptness of his great catch, so long as he does act with proper sensitivity to the factors that do matter, such as the trajectory of the ball above him,

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



By contrast, a walker outside the stadium in a dangerous neighborhood cannot assume by default the truth of those same propositions. Yes, if he attempts to walk harmlessly, while assuming that the lights will stay on, and if this assumption turns out by luck to be true, the walker does attain the desired success of his walk. But, unlike the fielder’s catch, this success is owed excessively to the luck that the lights stay on. Had the lights gone out, as they were nearly determined to do, then surely the walk would have ended badly. When we walk at night in a neighborhood that turns unsafe in the dark, we cannot properly, competently rely on a default assumption that the lights will not fail, with no thought or care about any potential failure of the lights. Here we do seem guilty of negligence or recklessness if we are entirely heedless of that problematic possibility, so that not even implicitly do we rely on its exclusion. Essential to the success of that walker’s walk is hence an implicit belief that is too probably false, so that the success is by luck. And that is why the walker manifests insufficient competence of the sort required to make the success of the walk attributable to them. Compare again the fielder’s catch. This too is owed to a large element of luck (that the lights will stay on). However, despite the luck that makes the fielder’s assumption true, and despite his reliance on that assumption in making his attempt, the success of the catch is attributable to the fielder’s triple-S competence. His attempt is apt, does sufficiently manifest competence, and earns full credit. Accordingly, the fielder’s performance, within the domain of baseball, is allowed a default assumption that the lights will stay on normally. Unlike the walker, moreover, the fielder can properly rely on that assumption even in the knowledge that it is unsafe. That is to say, he can rely on it with domaininternal propriety, with limited or zero detriment to the domain-internal quality of his baseball catch. So long as the lights do not fail in fact, the catch may attain full credit anyhow, will certainly count as an out, and will redound to the baseball credit his knowledge of his tracking ability and foot speed, his knowledge of the best alternative option (letting it bounce off the wall), and of the comparative baseball value of these two alternatives, etc. The fielder’s ability to take all of this into account instantaneously is certainly relevant to the full aptness of his attempt. Factors like these he cannot neglect to consider at least quickly and implicitly. He would be negligent if he ignored them, and reckless if he did not give them their proper weight. But the fragility of the lights he can ignore or dismiss with no domain-internal negligence or recklessness. His great catch would be equally admired and credited as a baseball catch, even despite the lights being so fragile and his knowing them to be so. Consider the lights’ degree of fragility, the probability that they will fail, with its huge bearing on likelihood of success. The bearing of this, even when known, is discountable in assessing domain-internal credit for athletic success. Success remains about as creditable to an athlete even when they make their athletic attempt in the knowledge that they are running such risk. Compare the risk recklessly dismissed by a basketball player who sinks a basket from the opposite end of the court when there was plenty of time to dribble or pass instead. That success (so bittersweet, certainly to the coach) will hardly earn full, unalloyed credit. The coach will not be pleased with the play selection. But does that affect the quality of the shot? Well, the shot was poorly selected, too risky, dumb, outrageous really. The coach will be unhappy with the player, and specifically with those qualities of the shot itself, which make the shot itself deplorable, for how poorly chosen it was, how much worse than many easily available alternatives. Its guiding intention is constitutive of that intentional action which is the shot, and the quality of that intention is dependent, surely, on the agent’s operative rationale. If the rationale is bad, that rubs off on the quality of that intention, and it rubs off in turn on the quality of that intentional action which is the shot itself, so that if the shot succeeds, which will be through luck, then on the whole it will have limited athletic worth creditable to the agent.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

of that fielder in the record books. No credit-reducing or credit-blocking luck much affects any of that. The fielder can attain full baseball credit for his catch and need incur no proper charge of negligence or recklessness. And here the contrast with the walker is stark. The walker who knows that the lights might so easily fail must take this seriously into account on pain of negligence or recklessness. What determines the assumptions allowed in a certain domain? Those assumptions come packaged in the competences recognized in that domain. The competences pertinent to archers, for example, are determined by skills defined as sufficiently reliable dispositions (or the seats thereof) to attain pertinent success in certain conditions, within certain parameters of shape and situation. The wind and the light must lie within certain limits, as must the sobriety of the archer. An archer does not show himself to be less skillful when he fails while drunk, or when he suddenly feints, or when a gust of wind or strike of lightning intervenes. Failure in those conditions of shape and situation does not affect the presence of true skill. Your failures outside relevant shape/situation parameters do not affect your possession of the skill. What matters is the success rate that you would have relative to shape/ situation parameters proper to that domain, set by the participants whose practice it is, whether the domain is formalized, as is competition archery, or less formal, as is the archery hunt. Parameters of competence can thus be set in a cultural practice.¹⁰ A default assumption is an assumption that a certain background condition obtains, one that (a) is required for a corresponding triple-S competence to be in place, but (b) can be in place unsafely, and unbeknownst to the agent. Agents may thus risk the failure of a background condition, while running that sort of risk need affect neither the quality of the agent’s domain-internal performance, nor the credit earned for its success. Epistemology’s main domain is that of attempts to get it right, and to do so aptly, along with proper storage of such beliefs. Thus, consider epistemic risk that triple-S epistemic competences may be diminished or lost through failure of background conditions. Such risk need not affect the epistemic quality of our attempts to get it right, nor the credit we earn for their success, and in particular need not affect whether we know. The risk that does affect quality and creditability of successful epistemic performance is risk that resides in competence below a certain threshold. Degree of such risk is inversely proportional to degree of corresponding competence. High risk is here tantamount to low competence, which surely does affect relevant quality of performance. Creditable achievement requires success by competence, not by a fluke.

6.7 Knowledge and Default In a performance domain, agents aim at certain outcomes with their movements, representations, and affirmations (a baseball catch, say, a soccer goal, a medical diagnosis, or a legal conviction), based on proper default assumptions. ¹⁰ But they might be determined, more functionally, by our basic human capacities and our species-wide need to coordinate action and share information, already present in early humans.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Our ordinary knowledge is also based on a proper assumption that no abnormal spoilers will intervene. That implicit assumption helps to guide our ordinary epistemic navigation. This all takes place within our epistemic domains, whether of expertise or of ordinary common sense. Suppose that assumption itself is put in question, however, as we consciously question its truth. That is sometimes thought to import an epistemically consequential shift. We are warned that we can then no longer insist on ruling out spoilers by default assumption. This is because, given our new, more deeply reflective context, no such assumption is thought to be sustainable simply by default. The plausibility of such schizophrenia has been questioned by those who reject the doubts of radical skeptics as mere “paper doubts.” And, I must say, if I consider how I would react to a NASA announcement about extraterrestrials bent on Matrixing us, I think I’d just proceed epistemically in pretty much the same ways I would have proceeded absent the warning. Would we really give up in such a case? Would we yield to universal suspension of judgment and belief? I say we would do no such thing. Those of us who took the announcement seriously would just continue to try to place its warning in proper epistemic perspective, based on the best judgment that we could muster, while those who dismissed the announcement would carry on with their normal routines, which they could hardly do if entirely deprived of judgment. And could we really be blamed, practically or epistemically? What other option would we have, really? It is hard to see how life in general could possibly go on otherwise, given our pervasive dependence on knowledge of many and various forms. Even for Robinson Crusoe, but especially in a complex social setting, we depend on judgmental knowledge, the knowledge that constantly fuels our practical syllogisms, and also the testimony that we need for coordinating action and sharing information. Sure, but what we would do is one thing, along perhaps with what we could not help doing. Why think that we would properly continue to attribute our ordinary knowledge and justification to ourselves? How can we avoid recklessness or at least negligence if we look away from the urgent NASA warnings? How is that case different from the case of the surgeon who is warned about the fragility of the lights in her high-tech operating room? Why should we think of our situation when we receive the NASA warning as like the situation of the fielder and unlike that of the surgeon? Is there perhaps a kind of pragmatic encroachment at work? Athletes risk little of much consequence by ignoring the warnings about the lights as they proceed with their game while assuming the lights won’t fail. By contrast, both the surgeon and the walker at night incur intolerable risk by ignoring similar warnings. Relevant here is the distinction between ludic practices where we aim at conventionally set objectives for the sake of amusement or enjoyment, and practices that are extra-ludic by aiming at objectives with inherent, extra-ludic importance. On this account, sports, including chess, are ludic, whereas the professions, including surgery, are extra-ludic. If this distinction between the ludic and the extra-ludic is indeed relevant to our concerns, as suggested, just how is that so more specifically?

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

How can we properly continue a game once we know of a credible and imminent terrorist threat? Pragmatic importance can bear by making relevant certain factors seemingly external to the domain. We saw how a surgeon needs to take the condition of the lights into account once she has been warned that they might fail. And the like goes for the walker at night in the bad neighborhood. Both performances are evaluable by reference to the danger posed by the fragility of the lights. In neither case can the fragility of the lights be bracketed by a default assumption that avoids negligence and recklessness. In the domains wherein the surgeon and the walker issue their respective performances, the safety imperiled by the lights’ fragility is essential to proper performance. Ludic domains differ from the non-ludic in that the engagement of ludic agents and spectators can properly ignore dangers like that of the lights’ fragility. Domaininternal performances, whether athletic or artistic, can attain full creditable achievement even with high danger in the wings. Such danger is then ignorable through default assumption with negligible detriment to the domain-internal quality of the performance. The difference then need not even be a matter of degree. Even if a terrorist threat (whether known or unknown) is ignored through default assumption by athletes and spectators, this might have zero effect on the athletic quality attainable by those athletes in that game. What is more, pragmatic importance can also block external factors from bearing on the quality of domain-internal performances. In a war zone, surgeons cannot postpone their surgery so as to check the condition of the lights. Cost/benefit analysis may immediately reveal that no sufficient advantage can be derived from any such delay. What would then be negligent or reckless is to delay the surgery so as to check the lights. Across domains of human performance, virtue theory recognizes default assumptions that can support excellent performance despite being unsafe and beyond the ken of performers who proceed on their basis. Virtue epistemology countenances a default assumption that radical skeptical scenarios are irrelevant, in just the way that the possible failure of the lights is irrelevant, both in the baseball game and in urgent surgery. Even if the lighting is in dire condition, this affects the quality neither of athletic performance, nor of the surgeon’s urgent procedure. We have seen how athletes in a game and surgeons in a war zone can properly assume by default that the lights will stay on. They might turn in unexcelled performances on that basis, even without knowing the truth of their assumption, and even when the assumption is unsafe. In particular, they incur no proper charge of negligence or recklessness in performing based on such sheer assumptions. And a similar conclusion may now be drawn about radical skeptical scenarios. Conscientiously enough, without negligence or recklessness, we normally assume ourselves safe from skeptical scenarios. And this assumption is proper even on the rare occasions when it is true but not known to be true and even quite unsafe, as on the occasion of the alien Matrixers. What of our cases of pragmatic encroachment that involve a vital, domain-external threat? Even then, it is hard to resist the following conclusion: that domain-internal performances are assessable by domain-internal standards. And such assessments seem properly unaffected by how unsafe the relevant background conditions may be.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Even threats of utmost urgency may still be harmless to the quality of performance in domains properly sealed against incursion by external values. What does this “proper sealing” amount to? It is not that the domain protects itself from external incursion because, all things considered, one can—properly, all things considered—continue domain-internal agency while disregarding urgent external threats. The point is rather the following. Suppose one does disregard the external threat or danger, and one does continue to make domain-internal attempts while dismissing such threats or dangers by default assumption. In that case, one’s continuing attempts may be unwise in the extreme (all things considered), while still attaining the heights of domain-internal quality, unaffected by the risks incurred. What relevantly distinguishes cases wherein one acts in pursuit of ordinary practical aims, such as the harmless walk at night or normal surgery in an operating room? Why are these domains so different from ludic domains like that of an athlete engaged in a sport? Because these practical domains build in the objective of physical safety for those involved, such as the walker and the patient under the knife. Why do we downgrade the performance of the walker and that of the surgeon, when we learn that they dismiss credible warnings? Why do we think that their success by luck is much less creditable than similar success by competence rather than luck? Why don’t we similarly downgrade the performance of the fielder who relies on assumptions that are unsafe and unknown and on occasion perhaps even when they dismiss credible warnings about the lights and even credible warnings about terrorism? Why is the credit to the surgeon and to the walker so diminished by their disregard of the danger, while the credit to the fielder for the outstanding catch is undiminished by his disregard of similar danger? Why is the catch deemed no less outstanding even when based on a default dismissal of a serious practical danger? How can that be so when that default dismissal is so very unwise? After all, the success of the catch is (by hypothesis) equally endangered, by the equally credible threat of terrorism, as is the success of the walk and of the surgery. Why are these latter successes less creditable to their respective agents than is the success of the fielder’s catch? Why is the success of the surgery and that of the walk diminished and less creditable to the full competence of the performers, given the guilty negligence or even recklessness in their disregard of authoritative warnings? Ordinary domains of praxis within which we pursue basic, generally valued goods should be distinguished from domains with internal constitutive goods, many of them ludic, whether athletic or artistic. There is a notable, plausible resemblance between epistemic and ludic assessment in the way both of these are sealed off from incursion by pragmatic external factors. And it is presumably the communities in charge of a relevant practice that properly determine whether and how the domain involved is thus sealed off.

6.8 Default Assumptions Again Our example of the Matrix-wielding alien force shows that our ordinary, commonsense domain contains its own default assumptions. Despite knowing that we had undergone such Matrix danger, we would have continued to attribute nearly all of the

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

knowledge that we ordinarily self-attribute, which we surely would have done properly and correctly. Could we have ruled out such Matrix intervention knowledgeably? In what domain? When we face our varied quotidian questions, we act in line with our epistemic competence if we assume that no abnormal spoiler will intervene. Among such spoilers is Matrix intervention by extraterrestrials, so we take it for granted that no such spoiler will intervene. When we turn out to be right on our quotidian questions, we thus manifest our competence, as we do also in making such default assumptions. What if space scientists warn about a real threat of spoiler dangers? In addressing that question, are we beyond our normal everyday framework, so that default assumptions are no longer competent?¹¹ If we persist in making the default assumptions that are pertinent to judgment and belief, and we turn out by luck to be right, could the success attendant on our luck still be competent and even apt? It could indeed, even when we properly reduce our pertinent degrees of confidence in the ordinary judgments and beliefs by which we act ordinarily. Within a virtue theoretic framework this means that we can still judge and believe with full aptness, even despite whatever reduction of confidence might also be appropriate. The “fullness” of our aptness is constituted by two things in combination: (a) the aptness of our success, as this manifests not just luck but sufficient competence exercised by the agent, and (b) the aptness of that apt attainment, as the aptness is itself attained through the agent’s competent assessment that they’re in a position to aptly make their attempt. Note that this can happen even when the agent could have attained an even higher level of competence in their attempt by heeding some further factors with some small influence on their likelihood of success. Compatibly with all this, their attempt may be sufficiently competent both on the first order and on the second order. Their attempt may thus be apt and “fully” apt, even while neglecting factors with negligible bearing, and even when attending to those factors might have slightly improved overall reliability by enhancing the competence manifest in that performance.¹²

6.9 Deductive Closure If we can rule out spoilers thus, if only by competent default assumption, can’t we now competently deduce that no such spoiler will intervene? Can’t we now thereby come to know after all that we need fear no such spoiler? Is this how we can agree ¹¹ This is reminiscent of another thought in Wittgenstein’s On Certainty: that hinge propositions are like a riverbed for the flow of our beliefs over time. They might at first be lodged in the stable riverbed but later come loose and form part of the flow rather than the bed. ¹² I grant that a probabilistic account is an alternative that some will find plausible. So, my proposed account assumes that we are explaining performances and their competence and aptness through on/off judgments or beliefs and the practical syllogisms that they fuel. Our view of the role of on/off knowledge—and its place in creditable, successful attempts in general—is set within that context. It is beyond our purview to defend our preferred framework against alternative frameworks that may dispense with on/off beliefs and with practical reasoning in terms of such beliefs and knowledge. To me it seems quite implausible that any fully satisfactory account of our agency and our cooperation can entirely dispense with such judgments and beliefs and with the corresponding assertions that enable us to share information and coordinate action. In any case, our reasoning presupposes a framework of outright judgment, belief, and knowledge.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



with G.E. Moore that, within ordinary contexts, we do know the sorts of facts listed in his defense of common sense. Let’s return to the hovering Matrixers, with the drawing of the lottery that will determine our epistemic fate. If we lose, we will all be Matrixed. At that moment, all across the globe there will be an immense welter of human activity, all based on a corresponding wealth of knowledge how and knowledge that. Consider the fact of the lottery, its complex setting, and its monumental epistemic threat. Does such danger spoil that immense complex of activity in its entirety, since it all now seems epistemically baseless? I press my car starter, say, expecting this to start the car. And that is what happens, but only because miraculously we win the lottery and the spoiler force moves on harmlessly. Is that enough to take away my credit for starting the car? No? But how can I keep that credit if my knowledge is gone? True, I do attempt to start my car, and my car does consequently start. But is that success properly, fully attributable to me? Can a success be through ignorance, through guessing or the like, while still creditable to the agent? Suppose we do find it plausible that, in such circumstances, people worldwide would still credit themselves and each other for the success of their attempts. In line with that, suppose we do keep our credit for the many things we do just as we win the momentous lottery. If we are to keep our credit, we must also keep our pertinent knowledge as the drama unfolds. I then know that I will start my car. And I could then deduce that I will not be Matrixed so as to preclude that anticipated outcome. So, could I thereby come to know after all that I will not be Matrixed, and also that we would not lose any such lottery? Undeniably, I can thus deduce that we’ll not be Matrixed, and later that we’ve not been Matrixed. But would that deduction give me knowledge of its conclusion, if I did not already know it to be true? How could I possibly know that no Matrix spoiler would intervene even when a spoiler was nearly certain to intervene? How could I know such a proposition to be true, based on the sheer assumption that it is true? Our way with the skeptic thus requires the rejection of epistemic closure and transmission, its rejection at least when such closure is stated in full generality.¹³ Our ordinary knowledge is based on a proper assumption that no abnormal spoilers will intervene. That implicit assumption helps to guide our ordinary epistemic navigation. This all takes place within our ordinary epistemic domains, whether of expertise or of everyday common sense. Putting that assumption in question, however, seems an epistemically consequential shift. Can we then still insist on ruling out abnormal spoilers by default? In that new, more deeply reflective context, can any such assumption still be sustained so easily? The reflective shift is said to take away our ordinary knowledge. In the ordinary domain itself we seem unable to knowledgeably rule out any such spoilers. Consciously reflective knowledge requires us to consider the possibility of spoilers, which would drive us out of the ordinary context and into a context of reflection where we can no longer just assume that we are free of spoilers. ¹³ Note how distinctive is this route to denial of closure. This is compatible with rejection of every one of the several other rationales for closure-denial that are found in the literature, such as that of Dretske, and that of Nozick, which differ from ours both in their scope and in their content.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

Recall the fielder who can aptly make his catch even in the knowledge that the lights are likely to fail. Can the domain of epistemology plausibly go that far? If the NASA report gives us knowledge that we do face that Matrix lottery, can we all still proceed with our welter of practical activities under a default assumption that we will not be Matrixed? Can we adopt a position like that of the athletes who carry on with their game even while knowing that the lights are about to fail. If they do carry on, they plausibly earn full credit for performing full well athletically despite the fact that their default assumptions are made in the teeth of contrary knowledge? Can we take an analogous view of our judgments? It is surprising how tempting it is to answer here in the affirmative. I do myself yield to that temptation. To me the domain of epistemology is a special case of human endeavor with a telic normativity of the sort we have explored. And it is then a feature of that normativity to allow default assumptions that can properly support the performance of agents with no negligence or recklessness. Epistemology is then similar to athletic domains in allowing default assumptions against abnormal spoilers, assumptions that fall far short of knowledge, and even of safety, and in allowing even default assumptions in the teeth of contrary knowledge! Allowing default assumptions in our epistemology has an important epistemic implication. Once we make that allowance, we must reject familiar and general principles of deductive closure. Default hinges are assumptions that stand fast by being faultlessly assumed. By standing fast, such competent assumptions can buttress knowledge even if they are not themselves known to be true or even safe. But we can’t possibly come to know that p by reasoning based essentially on the sheer prior assumption that p, not even if the assumption is true, nor even if it is true and competently assumed to be true. You cannot get to such knowledge through any deduction that relies essentially on a sheer assumption not itself already known to be true, not when the assumption has the very same content as the conclusion to be drawn. If exposed to the passing danger of Matrix intervention by extraterrestrials, we might be cut off not only from the external world but also from our past. Once Matrixed we would have beliefs coincident in content with our earlier beliefs, but the coincidence of such beliefs would be a sheer coincidence, as would be their correctness. But would such mere passing danger immediately deprive us of the manifold knowledge that we all normally retain in our ordinary lives? Not plausibly. Even once we became aware of the great risk we had endured, we’d continue properly and correctly to attribute to ourselves all the knowledge that we would have attributed even had we knowingly not been exposed to any such danger. The mere threat of the extraterrestrials would hardly deprive us of all our perceptual and stored knowledge, when the danger was fleeting and unrealized.

6.10 Are We Not Confusing the Epistemic with the Practical? Perhaps we would indeed properly continue to engage in quotidian judgment while disregarding the warning that possibly we are Matrixed by spatial interlopers, or indeed that this is likely, or even true. But might we proceed thus, on an assumption

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



of normality, just because we have no practical choice? Even if we take the probability that we are Matrixed to be astronomically high, perhaps we must carry on based on the same continuing assumptions, so that we’d go through the day making the very same judgments that we would have made even if we had not been privy to the NASA announcement. But why suppose this prudential decision to bear on the epistemic standing of the profusion of ordinary judgments that we’d still emit with abandon. Should we not then grant that any continuing justification we may have for rendering those ordinary judgments will be just pragmatic and not epistemic? A powerful intuition, granted.¹⁴ Can it still be resisted? Can we resist it so as to save our proposed epistemic role for default assumptions? Here follows an argument for the affirmative. Consider first an analogous example. Suppose I am about to play chess with an opponent at the public chess space in Washington Square Park. Night has fallen just prior to the match, as onlookers gather. Suddenly a stranger emerges from the darkness, eyes me intently, points with cocked thumb and index finger, and just then I see this text on my iphone: “I am looking at you. You must play. Play normally. If you do not play normally, we’ll know and you’ll be shot dead.” Please fill out the context so that, as I then see it, I have an irresistible rationale to play normally. As far as my selection of chess moves goes, I then restrict myself to the reasoning that I would have used in normal chess. Well, I do so to the extent possible, since normally I would not have been motivated to play normally by the fear of being shot dead. Still, I make my specific chess moves based purely on chess considerations, and I do all this through my chess competence. So, the AAA/SSS structure will apply in pure chess form, and I will earn chess-specific credit for any success that I may have. We can imaginatively fill in details so that, given my fully convinced view of the situation, I have no real option but to keep on playing. On that hypothesis, I must play normally, I must stay in the game, and this for powerful practical reasons, given how I see the situation with full conviction. Nevertheless, is it not obvious that these reasons have no bearing at all on the chess standing of any move that I may make in the course of that match? That being so, here is the proposed analogy to the epistemic case. Because we are essentially rational animals, we have no real option of how we proceed cognitively over the enormous span of the animal knowledge we rely on in any ordinary day. However, that applies equally to Mr Magoo and to normal perceivers.¹⁵ Any particular human, no matter how inept or adept, will be bound to proceed within some parameters of perceptual belief acquisition. But no-one is likely to entirely avoid illusion and every other perceptual error. Even when we are informed that there is an even chance that we will soon be Matrixed, Magoo and I will each continue to exercise our animal competences, perceptual and otherwise. And our respective cognitive performances will not be thought to be on a par. Magoo

¹⁴ My grateful thanks to Chris Kelp for pressing this concern. ¹⁵ The reference here is to the nearly deaf and blind cartoon character who miraculously escapes disaster at every turn.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

will perform poorly, or at least much more poorly than I will, so that I will attain a kind of knowledge, a first-order “animal” knowledge still denied to Magoo. True, the question is trickier when we turn to reflective knowledge full well. Here, taking a leaf from the Pyrrhonists, we might forbear judgment. In our alethic affirmations, we might forbear aiming for aptness as well as forbearing to aim for success. We would then continue to perform on the first order—the level of animal knowledge—however poorly we might fare in doing so, perhaps as poorly as Magoo. That would be an imperative of our animal nature. However, the very rationality of that nature at its best would move us to a general or even universal suspension of judgment, despite continuing to affirm alethically at every turn, as we are force-guided by our animal nature, by our perceptual competence. So, with the Pyrrhonists we would continue to be guided by our appearances and their inevitable first-order functional belief outcomes, but with a crucial difference emphasized by those skeptics: While refusing to endorse those outcomes rationally, we would also refuse to make the corresponding judgments, so as to attain through such judgments the status of reflective knowledge or knowledge full well. The inclusion of default assumptions in our virtue epistemology gives us an alternative to that dreary skeptical life. We can appeal instead to the analogy with the baseball fielder, by recognizing that there are things we can assume nonnegligently despite our not knowing them to be true and even when they are highly unsafe. That the lights will stay on is one such condition that the fielder can just trust to be in place even without knowing this and even when it is highly unsafe. The suggestion is further that when the space aliens are poised to Matrix us unless we win their lottery, we can still trust that we are not and will not be Matrixed, and can proceed without negligence to exercise our normal animal competences. What is more, we can also face the question of whether we are well placed to proceed that way—in good triple-S fashion—and we can take this question to be just a special case. In addressing it we can and must once again exercise our competences. Here again we can assume without negligence that we are not tripped up by space interlopers bent on Matrixing us, nor on any such highly abnormal spoiler. Of course, that leaves a lot of room for reflection requirements that must still be satisfied—on pain of negligence or even recklessness—in the exercise of our cognitive competences. So, there will remain an important distinction between bare animal knowledge and reflective knowledge full well.

6.11 An Application: Barnspotting About the large wooden structure that he sees before him, does Barney know it to be a barn, and does he know this by trusting his visual appearances when he happens to see the one real barn in an area rife with mere façades? Opinions differ on this. Among epistemologists there is widespread reluctance to allow that Barney does “know.” But many of the folk apparently disagree, and some epistemologists now take that side. How should we deal with this? Let us first generalize beyond the particular example.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Suppose you judge that p on a certain basis of visual appearance, so that it visually appears to you that p—abbreviated here as A(p). Can you thereby know that p, through a default assumption that reality here fits appearance? What if your assumption is not something you know to be true, nor is it even a safe assumption? What if it could very easily have been false, since in your modal vicinity A(p) comes mostly with ~p? Our approach allows, however, that you can know that p even if you rely on an assumption that is neither safe nor known by you to be true. In particular, you can know that p even if you base your belief that p on your knowledge that A(p) and your assumption that if A(p) then p, and even if initially this assumption is just a hinge that stands fast for you and is neither safe nor known to be true. What you do not know with priority in such a case is that your appearance is not misleading. In the familiar case of the fake barns, what you cannot thus rule out through conscious reflection, not knowledgeably, is the following: that you seem to see a barn without really seeing one. You cannot then know with priority that if you seem to see a barn, you do see a barn; in particular, you cannot know this so as to thereby obtain knowledge that you do see a barn, for you cannot know it prior to your knowing that you do see a real barn (nor can you even use it as a non-posterior basis for this latter knowledge). It might be argued that we can rely on the known fact that the overwhelming majority of barn appearances across the face of the earth are veridical. Can’t we then know that the one real barn in that vicinity is indeed a barn when it gives the corresponding appearance to us? Not plausibly. Compare a case where you enter a vast warehouse with millions of bags all of which contain only white marbles by the hundreds, with the possible exception of at most one bag. And you know that to be so. When you enter the room and pick up one of the bags at random you accordingly believe that the marble you pick is white, as you take it out of the bag, and you know that marbles picked at random from bags in that room tend strongly to be white. But what if that particular bag too has hundreds of marbles, all of them red, however, with the sole exception of the white marble that you happen to pick? Do you know that marble to be white as you take it out of its bag? Surely not. So, at least for now we have reason to favor the approach through default assumptions of the form: if A(p) then p. Your relevant default assumption is that if you visually seem to see a barn, then you do see a barn, which is a special case of our general default assumption, as we approach the world through our senses, that if A(p) then p, an ongoing assumption that reality here tends to correspond to (visual) appearance. This sustains the assumption made by epistemic agents, in their ordinary epistemic proceedings, on any given occasion, that they will likely enough succeed by believing in line with how things visually appear to them. And it is an assumption that according to our account of default assumptions, they can make with priority, and properly so. The ordinary epistemic visual domain is thus a special case of domains in which agents are allowed default assumptions. The following assumption is one made generally by default in that domain: that if here now one seems to see that p, then one does see that p.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



, ,  

6.12 Concluding Remarks This chapter has developed a notion of default assumptions and their normative roles. That notion makes room for a telic normativity that enables an enhanced account of epistemic assessment, one more realistic and plausible than earlier views. That anyhow has been the objective. What is the status of such “assumptions” as they play out in the first-order “animal” knowledge of humans and other animals? Neither lower animals nor even humans plausibly host these assumptions as conscious mental states that play a causal role in their mental economy. Rather, such assumptions are more plausibly made only implicitly. As for the reality of any associated states of knowledge or proper belief that may be hosted, these would be grounded in part on the sheer truth of our assumptions, even in the absence of any corresponding explicit thoughts deployed by the pertinent epistemic protagonists. We need to attribute such assumptions to protagonists and attributors in order to apply the model of valid reasoning in explaining the epistemic standing of beliefs and inferences by appeal to which we can understand human conduct and sustain it normatively. But what is the source of that kind of normative propriety for human agency in its many varieties across a great many domains? The source of such normativity is presumably to be found in the practices that constitute such domains of performance.

References Burge, T. (2003). Perceptual Entitlement. Philosophy and Phenomenological Research 67(3): 503–48. David Colaço, David, Wesley Buckwalter, Stephen Stich, and Edouard Machery (2014). Epistemic Intuitions in Fake-barn Thought Experiments. Episteme 11(2): 199–212. Dretske, F. (2000). Entitlement: Epistemic Rights without Epistemic Duties? Philosophy and Phenomenological Research 60(3): 591–606. Graham, P.J. (2012). Epistemic Entitlement, Nous 46(3): 449–82. Peacocke, C. (2004). The Realm of Reason. Oxford: Oxford University Press. Sosa, E. (2015). Judgment and Agency. Oxford: Oxford University Press. Sosa, E. (2017). Epistemology. Princeton: Princeton University Press. Steup, Matthias, and Jonathan Jenkins Ichikawa (2018). The Analysis of Knowledge, The Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/archives/sum2018/entries/knowledgeanalysis/. Wright, C. (2004). Warrant for Nothing (and Foundations for Free)? Aristotelian Society Supplementary Volume, 78(1): 167–212.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

7 Extended Entitlement J. Adam Carter and Duncan Pritchard

7.1 Extended Cognition, Epistemology, and Parity According to the hypothesis of extended cognition (HEC), there is no principled reason to preclude the possibility that one’s cognitive processes can extend beyond one’s skin and skull to include items in the world, such as instruments or even other cognitive agents.¹ This thesis should be distinguished from the related, but more contentious, extended mind thesis according to which mental states themselves, such as beliefs, can be similarly extended into the world.² Nonetheless, (HEC) represents a striking departure from ordinary received thinking about the ‘bounds of cognition.’ Even so, HEC has drawn a number of enthusiastic takers in contemporary philosophy of mind and cognitive science.³ While there are several ways to motivate HEC, a simple way to do so is via common-sense functionalism.⁴ This is at any rate the key idea driving Andy Clark and David Chalmers’s (1998) much-referenced parity principle, a principle that tells us under what conditions something should be accepted as part of a cognitive process: Parity Principle: If, as we confront some task, a part of the world functions as a process which, were it to go on in the head, we would have no hesitation in accepting as part of the cognitive process, then that part of the world is part of the cognitive process. (Clark and Chalmers 1998, 8)

¹ Following Menary (2012, 152), it can be helpful to think of the term ‘extension’ in terms of distribution. As Menary takes it, extended cognition ‘mean[s] distribution, cognition is distributed across brain, body and world.” For some other representative statements of the hypothesis of extended cognition, see Aizawa (2012, 92–3), Rowlands (1999, 22), Clark (2009), and Rupert (2004, 393). ² For some key defences of the extended mind thesis, see Clark and Chalmers (1998), Clark (2009), Menary (2007), Rowlands (1999), and Wilson (2004). For two notable lines of criticism, see Adams and Aizawa (2001, 2008) and Rupert (2004). ³ See Clark and Chalmers (1998) and Clark (2007). For defences of HEC from an epistemological perspective, see Pritchard (2010; 2018a; 2018b) and Palermos (2014a). For dissenting treatments of HEC, see Rupert (2004) and Adams and Aizawa (2008). ⁴ Another way to motivated HEC is by appeal to dynamic systems theory. See Chemero (2009), Froese, Gershenson and Rosenblueth (2013), and Palermos (2014b).

J. Adam Carter and Duncan Pritchard, Extended Entitlement In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0007

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

The parity principle is aimed at safeguarding against ‘bioprejudice’—viz., when it is just on account of the fact that something is external to the skin and skull that it is excluded as part of a cognitive process.⁵ To get a feel for the parity principle in action, it is helpful to consider here Clark and Chalmers’s hallmark case featuring ‘Otto’: Otto: Otto suffers from Alzheimer’s disease, and like many Alzheimer patients, he relies on information in the environment to help structure his life. Otto carries a notebook around with him everywhere he goes. When he learns new information, he writes it down. When he needs some old information, he looks it up. For Otto, his notebook plays the role usually played by a biological memory. (Clark and Chalmers 1998, 12–3)

Otto is clearly using his notebook in a way that it is on a ‘functional par’ with the way ordinary agents rely on a working biological memory, vis-à-vis storage and retrieval. By reference to the parity principle, then, Clark and Chalmers insist that if an ordinary agent (‘Inga’) relies on her biological memory and we count her biological memory as an element of the cognitive process she employs, then so must we count Otto’s notebook as part of the cognitive process that he employs. In this respect, we should view Otto’s cognitive process as a kind of ‘extended memory.’⁶ HEC is first and foremost a thesis about the metaphysics of cognition and, as such, not an epistemological thesis. But given that cognitive processes play indispensable roles in epistemological theory (which traffics in epistemic evaluations and appraisals), it is natural to consider the ramifications HEC would have if taken on board in mainstream epistemology. It would after all be a strike against HEC if the thesis could not be squared with ordinary insights in the theory of knowledge. One natural insight which is relevant here is that taking extended cognition seriously in epistemology involves a willingness to make symmetrical epistemic evaluations across pairs of cases that materially differ only insofar as one case in the pair is an extended analogue of the other. For example, insofar as we count Inga (relying on her biological memory) as knowing that MoMA is on 53rd street, then we should also be willing to treat Otto, who differs only in that he relies on an extended memory process, as knowing this proposition also. We can capture this idea as the epistemic parity principle: Epistemic Parity Principle: For any agent S and true proposition p, if S comes to believe that p by a process which, were it to go on in the head, we would have no hesitation in ascribing knowledge that p to S, then S knows that p.⁷ ⁵ Note that there are various issues related to how one might best understand the parity principle which, for reasons of space, we will be setting to one side here. For example, is how is one to understand the counterfactual aspect of this principle (i.e., the possibility that, in some sense, the very same process that occurs outside of the head is now occurring within the head)? For useful critical discussion of the parity principle on just this front, see Rupert (2009, ch. 2). ⁶ This is an example of a variety of extended cognition called artefact extension, where HEC is implied by external objects featuring in the cognitive processes of agents. See Menary (2012) for a different version of HEC—encultured cognition—according to which cognition is realized by mind, body and world not because the objects themselves are part of the cognitive process, but because some encultured transcranial processes are cognitive processes. ⁷ See here Carter (2013) and Carter, Kallestrup, Palermos, and Pritchard (2014).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



It might not be initially obvious that there would be any in principle difficulty posed for mainstream epistemology by respecting the epistemic parity principle. After all, why think there’s any in principle barrier (for an epistemological theory) to issuing exactly the kinds of symmetrical epistemic evaluations across cases that epistemic parity demands? In particular, don’t we have the intuition that both Otto and Inga know that MoMA is on 53rd street? But it turns out that there is trouble lurking. For notice that in order for the epistemic parity principle as just formulated to hold it is not enough merely for Otto and Inga both to have knowledge. Instead, it must also be the case that Otto’s and Inga’s beliefs have sufficiently similar epistemic properties to ensure that their knowledge co-varies across a suitable range of adapted cases. For example, notice that Otto and Inga’s beliefs can be epistemically alike in terms of both being cases of knowledge while also being importantly epistemically distinct in that one is susceptible to defeaters that the other isn’t. It would follow in such a case that although Otto and Inga in fact both have knowledge, there will nonetheless be possible cases, involving the relevant defeaters, where one of the pairing has knowledge while the other lacks it. This would of course be in conflict with the epistemic parity principle. Why might one hold that Otto and Inga’s beliefs have different epistemic properties? One reason—recently proposed by Åsa Wikforss (2014)—might be that there is a psychological difference regarding how the beliefs are formed. According to Wikforss, whereas Inga simply recalls a memory, Otto instead undertakes a more complex psychological process which involves a series of steps including a desire to consult a notebook, the intentional action of consulting a notebook, and the locating of the right entry using reason and perception.⁸ We think that this is a tendentious way of describing the two cases, and may in any case be epistemically irrelevant. On the first point, notice that part of the claim made by proponents of HEC is that in cases of extended cognition the only material difference between the subject and the counterpart who employs the corresponding non-extended cognitive process is that the extended cognitive process is taking place outside of the subject’s skin and skull. So whereas we can of course imagine variants of the Otto case where Otto is indeed undertaking relatively complex psychological processes in this manner, it is not this variation of the Otto case that we are interested in.⁹ Rather, our concern should be with a rendering of the Otto case where Otto’s belief formation is psychologically on a par with Inga’s memorial belief formation. ⁸ See Clark (2010, 7–8) for an attempt to turn the tables on the suggestion that Otto’s psychological process is more complex than Inga’s, in that it involves (at least) two steps rather than just the one step. Clark argues that an equally uncharitable characterization of Inga’s psychology would have her taking (at least) two steps as well. ⁹ It might be pointed out that a further potential difference between the cases is that Inga needn’t rely on her memory in order to use her memory (e.g., the issue doesn’t arise that Inga might have to rely on memory to recall that her own biological memory is in fact her own), whereas it seems Otto must use his memory in order to remember that his notebook is in fact his. While this looks troubling at first for the prospects of suggesting the two cases are ultimately on an epistemic par, we want to emphasize that while there are different ways one could spell out the details of the Otto case, the version we are interested in is one in which it is every bit as automatic for Otto to consult his notebook as it is for Inga to consult her memory. To the extent that we are focusing on such cases, the issue of whether Otto would or would not remember the notebook is his own is moot. Thanks to an anonymous referee for raising this point.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

In particular, we are to imagine that Otto has the same kind of psychological immediacy when consulting the notebook in order to find the relevant information that Inga displays when she consults her memory. (Note that we will be returning to this point in Section 7.3.) In any case, the second point—regarding the epistemic relevance of the putative psychological differences between Otto and Inga—is more important. The crux of the matter here is that even if there are such psychological differences their epistemic implications could well be moot. For what ultimately counts when it comes to the epistemic standing of a belief is what the epistemic credentials of that belief are and not the psychological processes involved in acquiring that belief. For example, a perceptual belief might enjoy an excellent epistemic pedigree in virtue of the fact that it is the product of a highly reliable cognitive process. That a complex psychological process was undertaken as part of the formation of the belief need have no bearing on its epistemic status. Conversely, we can imagine a perceptual belief which is accompanied by a very simple psychological process but which is nonetheless lacking in epistemic standing because the perceptual faculties in play were highly unreliable.¹⁰ The moral is that what counts in the Otto and Inga cases from an epistemological perspective is not so much the psychological processes in play, but rather the epistemic credentials of the epistemic basis for the belief so formed. More precisely, we should be wary about drawing any quick conclusions from the nature of the former to the nature of the latter. Henceforth, for the sake of argument we will grant to the proponent of HEC that there are no significant psychological differences in play when it comes to Otto and Inga’s belief. In particular, we will take for granted that the phenomenology of belief-formation is essentially the same in both cases. Even if we grant this point, however, there is still reason to reject the epistemic parity principle, at least as it currently stands. In order to see this, we just need to note that the cognitive process involved in the formation of Otto’s belief is undeniably different from that in play with regard to Inga’s belief, in that while the former involves factors outside of Otto’s skin and skull, Inga’s belief relies only on her biological memory. In short, even if we grant that these two subjects are on a psychological par, and even if we grant that they are both employing cognitive processes that are bona fide routes to memorial knowledge, it is still the case that they are distinct routes to memorial knowledge. While this point can at first sound epistemically inert, it raises some problems for the epistemic parity principle as it is currently formulated. The problem is that the epistemic source of one’s knowledge can have a bearing on which defeaters are relevant to the epistemic standing of one’s belief. Consider two agents who each form the perceptual belief that there is a sheep in front of them, and on this basis acquire perceptual knowledge of this proposition. Let us stipulate that these beliefs are psychologically on a par, in that they are both formed spontaneously in response to roughly the same visual stimuli and have a similar phenomenology. ¹⁰ See Lyons (2009) for a discussion of how, for example, a perceptual process can be psychologically inferential even if epistemically non-inferential (i.e., such that the epistemic standing of the beliefs so formed is largely independent of the psychological processes taking place).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



Imagine, however, that whereas the first agent views the sheep in natural light, the second agent instead views the sheep in artificial light. This difference becomes epistemically important once we consider undercutting defeaters—i.e., defeaters which undermine the epistemic pedigree of the belief.¹¹ For notice that what constitutes an undercutting defeater for the one agent’s belief need not constitute an undercutting defeater for the other agent’s belief. For example, being told that the artificial light in play is deceptive can be a misleading undercutting defeater for the second agent’s belief, but will have no bearing at all on the first agent’s belief since artificial light is not involved in the perception. What goes for the two agents in this case will also apply to Otto and Inga’s beliefs. Otto’s knowledge can be defeated by a misleading undercutting defeater to the effect that the notebook has been tampered with, but this won’t affect Inga’s knowledge. In contrast, Inga’s knowledge can be defeated by a misleading undercutting defeater to the effect that her (biological) memory is unreliable about this subject matter, but that won’t affect Otto’s knowledge.¹² The upshot is that there will be cases where Otto and Inga don’t both have knowledge (but where one of them does), contrary to the epistemic parity principle. That this issue can just as much affect two perceptual beliefs which differ only in whether natural light or artificial light is in play should give us pause for thought here, however, since if two perceptual beliefs of this kind can fail to satisfy the epistemic parity principle, then why would we expect a pairing of non-extended and extended cognition to satisfy this principle? This suggests that what we are interested in when it comes to epistemic parity is not the principle as formulated above, but rather a weaker principle which abstracts away from the question of defeaters. Consider this principle: Epistemic Parity Principle*: For any agent S and true proposition p, if S comes to believe that p by a process which, were it to go on in the head, we would have no hesitation in ascribing defeasible warrant to S’s belief that p, then S has defeasible warrant for her belief that p.

By ‘defeasible warrant’ here we mean an epistemic standing which suffices for knowledge absent any (undefeated) defeaters.¹³ This reformulation of the epistemic parity principle will evade the problem just mooted in that it is compatible with this

¹¹ Undercutting defeaters are usually contrasted with overriding defeaters, which are independent reasons against the truth of the target proposition. For more on this distinction, see Pollock (1986). ¹² It is after all already part of the details of the original case that Otto after recognizes that his biological memory is unreliable; hearing this will not effect his epistemic position. Interestingly, another kind of undercutting defeater which is particularly relevant to the Inga case concerns alcohol consumption. As Atkinson and Shiffrin (1968) note, the primary effect of alcohol consumption on biological human memory is that of impairing the transfer of information from short-term to long-term storage. Thus if Inga were to be convinced that she has (unbeknownst to her) regularly imbibed high levels of alcohol, this this could undermine the epistemic standing of her memorial beliefs. But of course this sort of undercutting defeater will not in principle affect Otto’s extended memorial beliefs, as he is not relying on his biological memory. For more on this point, see Carter and Kallestrup (2016). ¹³ Note that we can ignore Gettier-style epistemic luck in this regard, since we are keeping fixed the features of the subject’s environment (including her modal environment) and only switching between an extended and a non-extended version of the relevant cognitive process. Thus, the subject will either be subject to Gettier-style epistemic luck in both cases or in neither.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

principle that Otto’s knowledge and Inga’s knowledge are not epistemically alike with respect to their sensitivity to defeaters. Nonetheless, this principle captures the idea that there is a commonality of epistemic status to Otto and Inga’s beliefs, even though Otto’s belief is formed via an extended cognitive process. Plausibly, Otto and Inga’s beliefs will both satisfy the epistemic parity principle*. We have already stipulated that Otto’s belief-forming process is psychologically on a par with Inga’s. Moreover, in order for the epistemic parity principle* to apply to the pairing of Inga and Otto it must also be the case that Otto’s belief-forming process is also epistemically on a par, which means, for example, that it must be just as reliable. Putting these points together, and bearing in mind that defeaters are no longer relevant to our epistemic assessment in this regard, there seems no barrier to treating Otto and Inga as both enjoying defeasible warrant for their beliefs. Thus, if HEC merely has the epistemological ramification that the epistemic parity principle* holds, then there is no in principle problem involved in epistemology embracing HEC.

7.2 Basic Epistemic Sources That HEC can be made to square with our epistemological judgements in this way should surprise us. After all, insofar as contemporary epistemologists state a view in this regard (which is rarely), it is usually to endorse the position that cognitive processes take place entirely within the skin and skull of the agent.¹⁴ Would epistemologists who take this line really be so willing to endorse the epistemic parity principle* and its application to the Otto and Inga cases? Interestingly, they might well do just that. In order to see why, we need to note a crucial ambiguity in the epistemic parity principle* as it is currently formulated. This is that as it is presently characterized it only demands a parity of positive epistemic status should be respected across cases like that of Otto and Inga. But one can accept this much while nonetheless arguing that the positive epistemic status that accrues to the agents’ respective beliefs is completely different in character, such that Inga’s belief enjoys an epistemic pedigree that is not applicable to Otto’s belief. The reason why this is relevant for our purposes is that memory is standardly thought to be a canonical example of a basic epistemic source, where this means an innate belief-forming process which is epistemically privileged in the sense that it can be the source of non-inferential knowledge. More precisely, memory can be the source of non-inferential defeasible warrants which can in turn, absent defeaters, suffice for non-inferential knowledge. In this respect memory is held to be akin to perception and introspection and distinct from non-basic epistemic sources, such as testimony.¹⁵

¹⁴ See, for example, Goldman (1979, 13; 1986, 51) and Sosa (2007, 29). Elsewhere, one of the authors of this chapter has characterized this view as epistemic individualism—see Pritchard (2016; cf. Kallestrup and Pritchard 2012; 2013, 2014). ¹⁵ For more on basic epistemic sources, see Audi (2010). Note that there is some controversy over what counts as a basic epistemic source, and in particular whether testimony can qualify in this regard—Lackey (2008), for example, famously argues that testimony is as basic an epistemic source as perception or memory—though we will not be engaging with these issues here. As we will see in Section 7.4, the key issue for our purposes is not whether memory is epistemically on a par with, say, testimony, but rather whether

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



When epistemologists describe memory as a canonical basic epistemic source, however, they clearly have biological memory in mind. We might reasonably ask then whether extended non-biological memory is also a basic epistemic source. If it isn’t, then even though Otto and Inga’s beliefs are on an epistemic par sufficient to satisfy the epistemic parity principle*, there will still be a fundamental difference in their epistemic properties. In particular, Inga’s belief will have a privileged epistemic status that Otto’s belief lacks. One might wonder at this juncture whether the proponent of HEC should care whether there are epistemic differences of this level of fine grain between Otto and Inga’s beliefs. So long as they are epistemically on a par to satisfy the epistemic parity principle*, then why should this further level of difference matter? The reason why it matters is that if HEC is consistent with Inga’s belief having a superior epistemic standing relative to Otto’s belief in virtue of how it is formed,¹⁶ then it follows that extended cognitive processes are not on an epistemic par with their non-extended counterparts. So, for example, extended memory is not epistemically on a par with biological memory, extended perception is not epistemically on a par with biological perception, and so on. HEC when applied to epistemology thus ends up being a much less striking thesis. Moreover, once we concede that such extended cognitive processes are epistemically deficient relative to their nonextended counterparts, then one might start to naturally wonder whether they should be regarded as the same kind of cognitive processes at all. For example, if extended memory doesn’t share its essential epistemic properties with biological memory, then why is extended memory a genuine kind of memory, rather than simply being a nonmemorial epistemic process? Our goal in the remainder of the chapter is to argue that the proponent of HEC can resist this conclusion and continue to endorse the more striking claim that extended cognitive processes are fundamentally on an epistemic par with their non-extended counterparts. As we will argue, what is key here is to recognize that insofar as we understand Otto’s belief-forming process properly so that it is a plausible case of extended memory, then it likewise becomes credible that Otto’s belief enjoys exactly the same kind of basic epistemic standing that applies to Inga’s belief.

7.3 Cognitive Integration We noted in Section 7.1 that it is crucial to our evaluation of the Otto and Inga cases that we treat them as being psychologically on a par. For this to be the case, however, we need construe Otto’s situation in a very specific way. Otto’s extended memorial belief enjoys the very same kind of epistemic standing as Inga’s non-extended memorial belief. ¹⁶ By ‘superior’ we do not mean to suggest that, ceteris paribus, for two beliefs, a and b, where a is inferentially justified and b is justified via a basic epistemic source, the level of epistemic justification b enjoys exceeds the level of justification that accrues to b. Rather, the idea is that a basic source would be superior to a non-basic source in the sense that the deliverances of the former automatically enjoy a kind of defeasible epistemic support that is not automatically enjoyed by beliefs that are the deliverances of nonbasic sources.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

To begin with, note that Otto is very different from an individual who (say) consults a phone book on occasion.¹⁷ Rather, the claim is that Otto satisfies a number of ‘glue and trust’ conditions which someone who merely uses a phone book from time to time will not satisfy. Following Clark (2008), we can formulate these conditions as follows: (i) Availability: The information in the notebook must be reliably available and regularly consulted. (ii) Accessibility: The information in the notebook must be easy to access. (iii) Automatic Endorsement: The information retrieved from the notebook should be automatically endorsed and should not normally be subject to critical scrutiny.¹⁸ (iv) Past Endorsement: The information in the notebook must have been previously endorsed by Otto and be there as a consequence of this endorsement. Note that all of these conditions generally apply to normal biological memory, at least insofar as one’s biological memory can generate the basic defeasible warrant which is our present concern. In line with (i), one’s memory is reliably available and one regularly consults it—indeed, it is a matter of habit, in the sense that one’s decision to consult it can be spontaneous. In line with (ii), one’s memory is usually easy to access, at least when it is functioning in a way that is conducive to furnishing one’s memorial beliefs with basic defeasible warrant. In line with (iii), one does not normally subject the deliverances of one’s memory to critical scrutiny. Finally, in line with (iv), one’s memories are characterized by their relationship to one’s past experiences. Of course, biological memory doesn’t always function in these ways, even when it is a source of basic defeasible warrant. Memories can be hard to ‘locate’ sometimes, for example. But in order to keep matters simple, let’s focus on a particular way in which Inga uses her memory, such that it does satisfy these four conditions. In particular, we are interested in cases where Inga immediately consults her memory when called upon to do so, and can easily identify thereby the information she requires, information which is automatically endorsed by Inga and which was previous endorsed by Inga at some point in the past (leading it to form part of Inga’s memories). In order for Inga’s use of her memory, so understood, to lead to defeasibly warranted memorial beliefs, we will also need to impose some epistemic conditions on this belief-forming process. Interestingly, however, it seems that such conditions can be quite minimal, at least to the degree that they require very little by way of reflective reasoning on Inga’s part. In particular, what seems most important is that

¹⁷ If it were, then it would count against the plausibility of HEC that Otto’s case be ruled-in as a bona fide case of extended cognition. See Rupert (2004, 401–5). This point is part and parcel with the ‘cognitive bloat’ worry: if too much of the external world is ruled-in as part of the cognitive process, then HEC is implausibly liberal in a way that would not preserve a meaningful difference between agent and environment. See Palermos (2014b) for a reply to Rupert on this point. ¹⁸ That is, it ‘should be deemed about as trustworthy as something retrieved clearly from biological memory.’ (Clark 2010, 49)

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



this belief-forming process be in fact reliable. That is, in the right environment—e.g., where there are no defeaters present—the exercise of a reliable memorial beliefforming process, where the belief so formed is true, can lead to knowledge.¹⁹ In such cases, the agent’s cognitive success would be creditable to her cognitive agency (i.e., her exercise of relevant cognitive ability, in this case memory) as opposed to nonagential factors (such as purely environmental considerations), and would result in a true belief which could not have easily been false (i.e., which is safe). That makes it a prime candidate for knowledge.²⁰ Moreover, the knowledge in question is epistemically basic, in that it simply involves the exercise of a basic epistemic source. When defeaters enter the scene, then the epistemic situation changes, since it is now incumbent upon the agent to defeat the defeaters, and that requires a reflective rational process be undertaken by the agent, in contrast to the simple formation of memorial belief which may involve no reflection at all. This highlights the point that the basic warrant furnished by the reliably functioning memory is only defeasible.²¹ For now, let us set defeaters to one side and focus only on cases where no defeaters are present. Note that when we attributed basic knowledge to Inga above, we did so because the cognitive success in question was creditable to her cognitive agency. That is, we have a cognitive success that is explained by the subject’s exercise of relevant cognitive ability, and not by other non-agential factors (such as purely environmental factors). What does it take for a subject’s belief-forming trait to count as a cognitive ability? Reliability is surely very important in this regard, as noted above, but clearly not sufficient, since all manner of belief-forming processes could be reliable without thereby counting as cognitive abilities. For example, a cognitive malfunction can nonetheless be reliable, as Alvin Plantinga’s (1993) famous example of the brain lesion which reliably causes the subject to believe that she has a brain lesion demonstrates. Cognitive malfunctions are not cognitive abilities, no matter how reliable they might be, which is why they not routes to knowledge in the way that cognitive abilities can be. ¹⁹ Note that a further way (i.e., aside from defeaters) in which a reliable memorial belief-forming process can fail to lead to memorial knowledge is when there is environmental epistemic luck in play. Such luck can be present even where there are no defeaters and even where the actual environment is conducive for the reliable experience of the belief-forming process in question. As noted in note 13 we can set the issue of Gettier-style luck (of which environmental epistemic luck is a sub-species) to one side for our purposes. Environmental epistemic luck was introduced in Pritchard (2009a, 2009b, chs. 3–4, 2012). See also Pritchard, Millar, and Haddock (2010, chs. 2–4), Kallestrup and Pritchard (2012, 2013, 2014), and Pritchard (2016). ²⁰ In addition, the cognitive success in question manifests the subject’s cognitive agency, which is a condition that does not follow merely from the claim that this cognitive success is creditable to that cognitive agency. In what follows we will take it as given that this manifestation condition is also met, both in cases of biological memorial knowledge and extended memorial knowledge. See Sosa (2007) and Turri (2011) for more on this point. For a general defense of the idea that knowledge requires both the exercise of cognitive ability and safe cognitive success, see Pritchard (2009b, 2012) and Pritchard, Millar, and Haddock (2010, chs. 2–4). ²¹ An interesting question here is what happens to basic defeasible warrant that is subjected to a defeater which is subsequently defeated. Does the warrant continue to be basic, or is it now non-basic? We will be setting this question to one side in what follows. For helpful discussion of defeaters in this regard, see Plantinga (1993).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

The additional ingredient that is required is that the belief-forming process should be an integrated part of one’s overall cognitive character. That’s just what’s lacking in the cognitive malfunction case, for example, in that while there is a belief-forming process on display which is reliable, it is not a process which is integrated with the rest of the subject’s cognitive character. Indeed, it is in despite of the subject’s cognitive character that this belief-forming process is reliable, in that her cognitive success is not creditable to her cognitive agency but rather to the happenstance that the cognitive malfunction in play turns out to be reliable. In contrast, had the process been integrated within the agent’s cognitive character then it could potentially have led to knowledge. If, for example, the subject in the brain lesion case becomes aware that there are brain lesions of this kind and that she likely has one, then arguably she could come to know via this means that she had a brain lesion. This would be a case where a reliable belief-forming process which is not integrated within a subject’s cognitive character becomes integrated via a rational process. Insofar as the subject succeeds in this regard, then her cognitive success would be creditable to her cognitive agency. In cognitive integrating the cognitive process into her cognitive character, the subject thus effectively transforms it from a cognitive malfunction into a cognitive ability.²² One’s innate cognitive faculties—such as memory—are in a privileged position in this regard, in that they effectively define what is to constitute a minimal (i.e., knowledge-conducive) level of cognitive integration. Even prior to a subject being able to undertake reflective rational processes of any level of sophistication, she can nonetheless gain knowledge via her cognitive faculties, where this involves cognitive success which is creditable to her cognitive agency. That this is possible demonstrates that cognitive integration, while it may involve a reflective rational process as in the brain lesion case, need not require any substantive level of reflective rationality (much less the formation of a ‘perspective’ on one’s cognition in terms of a network of meta-beliefs). Rather, cognitive integration in this regard just requires that one’s innate cognitive processes be appropriately entwined with one another. For example, a subject whose memorial beliefs were not appropriately sensitive to her perceptual beliefs—for instance, where her recent memories about her environment are regularly completely at odds with her current perception in this regard, but both are automatically endorsed nonetheless—would not count as having properly functioning innate faculties. When we imagine Inga to have properly functioning memory, and thus being in the market for basic memorial knowledge, we are therefore treating her as having innate cognitive faculties which are cognitively integrated at least to this minimal extent. With all this in mind, let us return to the case of Otto. In imagining Otto as being on a psychological par with Inga we are supposing that Otto meets the ‘glue and trust’ conditions laid down above. Merely looking at a phonebook—or, indeed, one’s notebook—from time to time would not suffice in this regard. But notice now what would be required for Otto to count as meeting these conditions.

²² For further discussion of the brain lesion case in the context of cognitive integration, see Greco (1999) and Pritchard (2010, 2012).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



The fourth condition is trivially met in virtue of it being his own notebook, but it is the other three conditions which are interesting for our purposes. For in order to satisfy these conditions, we need to suppose that the notebook is a constant feature of Otto’s life, and habitually consulted, so that the information that it provides becomes automatically endorsed. Imagine, for example, that the notebook is attached to Otto’s body in order to ensure its easy access so that Otto can consult it immediately, and that it has been attached in this way for a significant period. Indeed, notice that these three conditions effectively entail that Otto’s has been using the notebook in this direct fashion for some time. This is important because it has a bearing on the extent to which the notebook has become cognitively integrated into Otto’s cognitive character. Unlike the mere occasional checking of a phone book, which isn’t psychologically on a par with Inga’s use of memory, Otto has reached a stage where consulting the notebook has become every bit as automatic as consulting one’s biological memory. For Otto to reach such a stage the notebook will have become integrated with his other belief-forming processes in the same way as his innate cognitive faculties like memory and perception are cognitively integrated. In particular, the information in the notebook will have accorded with the deliverances of Otto’s other existing cognitive abilities for a sustained period before it is treated as an automatic information-resource in this fashion. Now of course we can imagine a version of the Otto case where no such process of cognitive integration takes place, and where Otto immediately begins trusting the information in the notebook in the same way as Inga trusts her memory. But notice that we are now no longer keeping the relevant epistemic features of the two cases fixed. The epistemic parity principle* stipulates that the only difference between the cognitive processes employed by Inga and Otto is that only the former takes place within the subject’s skin and skull. But that means that insofar as we treat Inga as being in a position to acquire defeasible warrant from employing her memory, such that her memorial cognitive process is reliably and cognitively integrated, then we need to likewise suppose that Otto’s cognitive process is reliable and cognitively integrated too. (Note that, in line with the discussion in Section 7.1, we are setting aside all defeaters here.) With the Otto case so understood, however, then this reinforces the point made above that Otto’s beliefs ought to be credited with the same kind of positive epistemic standing that we would attribute to Inga’s memorial beliefs. In particular, by consulting the notebook Otto can acquire direct knowledge, where this means that his cognitive success in this regard is creditable to his cognitive agency and not to other non-agential factors (such as purely environmental factors). This still leaves the issue that we noted earlier, however, which is whether Inga’s belief enjoys a special kind of epistemic standing in virtue of being formed via an innate cognitive faculty that Otto’s belief lacks. Our claim, however, is that once we properly understand how the two cases are on an epistemic par, then we can also deal with this issue. For the special kind of epistemic standing that accrues to Inga’s belief is a form of entitlement, and we suggest that with Otto’s belief-forming process understood along the lines just proposed his beliefs are also in the market for entitlement, albeit entitlement which accrues from an extended cognitive process rather than from an innate one.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

7.4 Extended Entitlement ‘Entitlement’—it should be noted up front—is a term of art in epistemology with its own distinctive characterization in the hands of (among others) Fred Dretske (2000); Christopher Peacocke (2004); and Crispin Wright (2004). The nuanced notion of epistemic entitlement that we shall be interested in here has been developed in a series of influential works by Tyler Burge (1993, 1997 2003). On a first approximation, as Burge sees it, our perceptual beliefs (and other basic beliefs) enjoy the status of ‘entitlement,’ wherein (in the absence of positive reasons to the contrary) one has the right to take the state ‘as it represents the world as being.’²³ This positive epistemic status is understood as a kind of externalist warrant distinct from the kind of ‘internalist’ justification enjoyed by (for instance) inferentially acquired knowledge. In particular, Burge (1997) argues that we are entitled to accept the deliverances of memory, and our focus will specifically be on this claim. In order to engage with Burge’s account of memorial entitlements, however, it’s important to first come to grips with the particular way that Burge is thinking about memory and its epistemic significance. In particular, we need to note that Burge rejects generativism about the epistemology of memory, as defended by such figures as Robert Audi (1995) and John Pollock (1974; 1986). According to generativism, there is a close analogy between memory and perception such that, just as the phenomenology involved in perception generates positive epistemic status for one’s perceptual beliefs, so the phenomenology of recalling generates positive epistemic status for one’s memorial beliefs. Generativist-based arguments to the effect that we have memorial entitlements would thus parallel closely the kind of arguments we’d expect to find for perceptual entitlements. One problematic implication of generativism in the epistemology of perception, at least for generativist proposals the sort defended by Audi and Pollock,²⁴ is that the more we observe a phenomenon the greater the epistemic support our perceptual belief in this regard enjoys. Matthew McGrath (2007 19–22) calls this the ‘epistemic boost problem’ for generativism.²⁵ Such an implication is even more problematic when applied to generativism in the epistemology of memory. Would we really want to say that, ceteris paribus, a memorial belief that is regularly retrieved enjoys thereby a better epistemic status than one that is retrieved infrequently?²⁶ In opposition to the generativist view, Burge (1997) accepts the more standard line in the epistemology of memory, which is preservativism. Preservativism draws a ²³ Compare Burge’s ‘Acceptance Principle’ according to which: “A person is entitled to accept something as true something that is presented as true and that is intelligible to him, unless there are stronger reasons not to do so.” (Burge 1993, 467) ²⁴ Compare what Bernecker calls radical generativism, the view defended by Pollock and Audi, with moderate generativism, according to which, as Bernecker (2011, 11–12) puts it, ‘the only way for memory to function as a generative source of justification is by removing defeaters and thereby unleashing the justificatory potential that was already present at the time the belief was initially entertained.’ Unlike radical generativism, moderate generativism is not plagued by the epistemic boost problem. Thanks to an anonymous referee for requesting further clarification on this point. ²⁵ See Bernecker (2011, 11) for helpful discussion on this point. ²⁶ See McGrath (2007) for a critical discussion of this implications of generativism in the epistemology of perception and memory. For a useful overview of the issues in this regard, see Bernecker (2011).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



closer parallel between memory and testimony than between memory and perception. Here is Sven Bernecker’s characterization of preservativism: Just as testimony is said to transmit knowledge from one person to another, memory is said to preserve knowledge from one time to another. (Bernecker 2011, 11)

In terms of our terminology, preservativism is the idea that memory can preserve defeasible warrant from one time to another.²⁷ Since the epistemic standing is derived from the original memory, and not generated as part of the recall of the memory, such a view avoids the problematic implication that we just noted afflicts generativism in the epistemology of memory. A further advantage of preservativism is that it is better able than generativism to make sense of how it is that our memorial beliefs can have a kind of positive epistemic status even though (very often) we’ve simply forgotten our original warrants. As Burge remarks here: I believe that a person clearly can be entitled to believe a theorem she believes because of preservative memory even if she cannot remember the proof she gave long ago, and even if she cannot remember that she gave a proof. Most of what one is entitled to believe from past reading, past interlocution, past reasoning, or past empirical learning, derives from sources and warrants that one has forgotten. (Burge 1997, 38)

According to preservativism, the epistemic entitlement that accrues to a memorial belief is in virtue of the epistemic standing of the original belief, and this explains why an agent can have entitlement for a memorial belief even when she is unable to recollect the source of this memory. In contrast, on the generativist picture a defeasible positive epistemic status is generated for the memorial belief simply in virtue of it being a memorial belief, regardless of its original epistemic pedigree. It’s safe to assume Burge has biological memory in mind as undergirding the picture of memorial entitlement he is offering. However, there’s nothing to prevent us from allowing that the status of (Burgean) entitlement that Inga’s memorial beliefs enjoy won’t also be enjoyed by Otto. To see how this is so, let’s get clearer about why, according to Burge, Inga’s memorial beliefs enjoy the status of memorial entitlement. For one thing, Inga’s (innate) memorial process satisfies a functional requirement that Burge lays down, which is that biological memory, in the default case, has a characteristically veridical and reliable function, as established through a history of causal interactions with the environment in which Inga relies on her memory. Moreover, Inga’s biological memory has this veridical and reliable function in part because: (i) Inga’s biological memory represents and stores veridical content (and has a history of doing so via causal interactions with her environment); and (ii) the content Inga represents and stores is sufficiently causally related to the content that

²⁷ The general idea that memory preserves knowledge is known as the epistemic theory of memory. See, for instance, Malcolm’s (1963, 223) suggestion that one remembers a proposition just in case one knows it because one knew it. See Bernecker (2010) for a sustained criticism of this position, and see Moon (2013) for a recent defense of the epistemic theory of memory in light of Bernecker’s criticism.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

she later represents at the time of recall.²⁸ It is in virtue of meeting these Burgean conditions that Inga’s memorial belief enjoys the status of entitlement (i.e., she can take it for granted, absent any positive reason not to). Crucially, however, Otto’s extended memorial beliefs will satisfy the very same Burgean conditions. In order to see this, we first need to draw attention to how an obvious prima facie difference between Otto and Inga turns out to be entirely irrelevant to whether Otto’s extended memorial beliefs enjoy entitlement. This obvious difference is that Otto will (perhaps, invariably) forget the original warrant for each belief that is transferred into his extended memorial storage, whereas this eventuality will be rarer for Inga, who often enough will recall both the stored proposition as well as her original basis. After all, the information that Otto will be entering into the notebook will usually just be the target proposition, and not also the epistemic basis that he has for believing this proposition at the time of the entry. As we’ve seen, however, the details of Burge’s line make clear that Otto’s failure to recall his original grounds will be neither here nor there as regards the issue of whether Otto’s extended memorial beliefs enjoy Burge-style entitlement. What matters is that the original warrant is preserved through a characteristically veridical and reliable process, via a process of continual causal interactions with the environment. Now we’ve just considered how Inga’s reliance on her biological brain constitutes such a process. Consider now Otto’s extended memorial process, once his use of the notebook has been cognitively integrated into his cognitive character in line with our discussion in Section 7.3. Once cognitively integrated in this way, Otto’s extended cognitive process will exhibit a characteristically veridical and reliable function because: (i) like Inga’s biological memory, it has a causal-environmental history of representing and storing veridical content, and (ii) again just like Inga’s biological memory, the content that Otto represents and stores is sufficiently causally related to the content that he later represents at the time of recall.²⁹ It follows that there is nothing in Burge’s view about memorial entitlement which wouldn’t apply just as well to Otto’s extended memorial belief as it would to Inga’s non-extended memorial belief. What matters is whether Otto’s belief meets the conditions that Burge lays down for memorial entitlement, and so long as Otto’s extended memorial belief-forming process is cognitively integrated within his cognitive character then he will satisfy these conditions just as much as Inga does. In particular, we have seen that in order to for the Otto case to be relevantly analogous to the Inga case, such that the former differs from the latter only in as much as that an extended memorial cognitive process is involved, then we need to suppose that Otto’s extended memorial belief-forming process is cognitive integrated within his cognitive character. But, so construed, Otto’s extended memorial belief enjoys a defeasible warrant which is not merely on an epistemic par to that enjoyed by Inga’s non-extended memorial belief, but is of the same kind: Burgean entitlement.³⁰ ²⁸ Note that this condition is needed to safeguard against deviant causal chains between stored and recalled content as counting as bona fide memory. ²⁹ In particular, there isn’t, for instance, any kind of deviant causal chain that infects Otto’s causal connection between storage and retrieval in the default case. ³⁰ This article was written as part of the AHRC-funded ‘Extended Knowledge’ (#AH/J011908/1) research project that was hosted by the University of Edinburgh’s Eidyn research centre, and was presented

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



Bibliography Adams, F. and Aizawa, K. 2001. The Bounds of Cognition. Philosophical Psychology 14: 43–64. Adams, F. and Aizawa, K. 2008. The Bounds of Cognition. Oxford: Blackwell. Aizawa, K. 2012. Distinguishing Virtue Epistemology and Extended Cognition. Philosophical Explorations 15, 91–106. Atkinson, R. C., and Shiffrin, R. M. 1968. Human Memory: A Proposed System and its Control Processes. In The Psychology of Learning and Motivation: Advances in Research and Theory, ed. K. W. Spence, 89–195. York: Academic Press. Audi, R. 2010. Epistemology: A Contemporary Introduction to the Theory of Knowledge. London: Routledge. Audi, R. 1995. Memorial Justification. Philosophical Topics 23: 31–45. Bernecker, S. 2010. Memory: A Philosophical Study. Oxford: Oxford University Press. Bernecker, S. 2011. Memory Knowledge. In Routledge Companion to Epistemology, eds. S. Bernecker and D. H. Pritchard, 326–34. London: Routledge. Burge, T. 1993. Content Preservation. Philosophical Review 102: 457–88. Burge, T. 1997. Interlocution, Perception, and Memory. Philosophical Studies 86: 21–47. Burge, T. 2003. Perceptual Entitlement. Philosophy and Phenomenological Research 67: 503–48. Carter, J. A. 2013. Extended Cognition and Epistemic Luck, Synthese 190: 4201–14. Carter, J. A., and Kallestrup, J. 2016. Extended Cognition and Propositional Memory. Philosophy and Phenomenological Research 92: 691–714. Carter, J. A., Kallestrup, J., Palermos, S. O., and Pritchard, D. H. 2014. Varieties of Externalism. Philosophical Issues 24: 1–45. Chemero, A. 2009. Radical Embodied Cognitive Science. Cambridge, MA: MIT Press. Christensen, D., and Kornblith, H. 1997. Testimony, Memory and the Limits of the A Priori. Philosophical Studies 86: 1–20. Clark, A. 2007. Curing Cognitive Hiccups: A Defense of the Extended Mind. Journal of Philosophy 104: 163–92. Clark, A. 2008. Supersizing the Mind: Embodiment, Action, and Cognitive Extension. Oxford: Oxford University Press. Clark, A. 2009. Spreading the Joy? Why the Machinery of Consciousness is (Probably) Still in the Head. Mind 118: 963–93. Clark, A. 2010. Memento’s Revenge: The Extended Mind, Extended. In The Extended Mind, ed. R. Menary, 43–66. Cambridge, MA: MIT Press. Clark, A., and Chalmers, D. 1998. The Extended Mind. Analysis 58: 7–19. Dretske, F. 2000. Entitlement: Epistemic Rights without Epistemic Duties? Philosophy and Phenomenological Research 60: 591–606. Froese, T., Gershenson, C., and Rosenblueth, D. A. 2013. The Dynamically Extended Mind. http://arxiv.org/abs/1305.1958. Goldman, A. 1979. What Is Justified Belief? In Justification and Knowledge, ed. G. Pappas, 1–23 Dordrecht: D. Reidel Publishing Company. Goldman, A. 1986. Epistemology and Cognition. Cambridge, MA: Harvard University Press. Greco, J. 1999. Agent Reliabilism. Philosophical Perspectives 13: 273–96.

in 2014 as part of the research group affiliated with this project. We are grateful to the participants of this research group for valuable discussion, including Bolesław Czarnecki, Matt McGrath, Orestis Palermos, Tim Kenyon, and Kevin Wallbridge. Thanks also to Nikolaj Pedersen and Peter J. Graham for helpful comments on an earlier version of this paper.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



 

Kallestrup, J., and Pritchard, D. H. 2012. Robust Virtue Epistemology and Epistemic AntiIndividualism. Pacific Philosophical Quarterly 93: 84–103. Kallestrup, J., and Pritchard, D. H. 2013. Robust Virtue Epistemology and Epistemic Dependence. In Knowledge, Virtue and Action, eds. T. Henning and D. Schweikard, ch. 11. London: Routledge. Kallestrup, J., and Pritchard, D. H. 2014. Virtue Epistemology and Epistemic Twin Earth. European Journal of Philosophy 22: 335–57. Lackey, J. 2008. Learning from Words: Testimony as a Source of Knowledge. Oxford: Oxford University Press. Lyons, J. 2009. Perception and Basic Beliefs: Zombies, Modules, and the Problem of the External World. Oxford: Oxford University Press. Malcolm, N. 1963. Knowledge and Certainty. Ithaca: Cornell University Press. McGrath, M. 2007. Memory and Epistemic Conservatism. Synthese 157: 1–24. Menary, R. 2007. Cognitive Integration: Attacking the Bounds of Cognition. London: Palgrave Macmillan. Menary, R. 2012. Cognitive Practices and Cognitive Character. Philosophical Explorations 15, 147–64. Moon, A. 2013. Remembering Entails Knowing. Synthese 190: 2717–29. Palermos, S. O. 2014a. Knowledge and Cognitive Integration. Synthese 191: 1931–1951. Palermos, S. O. 2014b. Loops, Constitution, and Cognitive Extension. Cognitive Systems Research 27: 25–41. Peacocke, C. 2004. The Realm of Reason. Oxford: Oxford University Press. Plantinga, A. 1993. Warrant: The Current Debate. Oxford: Oxford University Press. Pollock, J. 1974. Knowledge and Justification. Princeton: Princeton University Press. Pollock, J. 1986. Contemporary Theories of Knowledge. Savage, MD: Rowman & Littlefield. Pritchard, D. H. 2009a. Apt Performance and Epistemic Value. Philosophical Studies 143, 407–16. Pritchard, D. H. 2009b. Knowledge. London: Palgrave Macmillan. Pritchard, D. H. 2010. Cognitive Ability and the Extended Cognition Thesis. Synthese 175: 133–51. Pritchard, D. H. 2012. Anti-Luck Virtue Epistemology. Journal of Philosophy 109: 247–79. Pritchard, D. H. 2016. Epistemic Dependence. Philosophical Perspectives 30: 1–20. Pritchard, D. H. 2018a. Extended Knowledge. In Extended Epistemology, (eds.) J. A. Carter, A. Clark, J. Kallestrup, S. O. Palermos and D. H. Pritchard, 90-104. Oxford: Oxford University Press. Pritchard, D. H. 2018b. Extended Virtue Epistemology. Inquiry 61: 632–47. Pritchard, D. H., Millar, A., and Haddock, A. 2010. The Nature and Value of Knowledge: Three Investigations. Oxford: Oxford University Press. Rowlands, M. 1999. The Body in Mind: Understanding Cognitive Processes. Cambridge: Cambridge University Press. Rupert, R. 2004. Challenges to the Hypothesis of Extended Cognition. Journal of Philosophy 101: 389–428. Rupert, R. 2009. Cognitive Systems and the Extended Mind. Oxford: Oxford University Press. Sosa, E. 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge. Oxford: Oxford University Press. Turri, J. 2011. Manifest Failure: The Gettier Problem Solved. Philosophers’ Imprint 11: 1–22. Wikforss, Å. 2014. Extended Belief and Extended Knowledge. Philosophical Issues 24: 460–81.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

.     



Wilson, R. 2004. Boundaries of the Mind: The Individual in the Fragile Sciences. Cambridge: Cambridge University Press. Wright, C. 2004. Warrant for Nothing (and Foundations for Free)? Proceedings of the Aristotelian Society 78 (suppl.): 167–212.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

8 Moorean Pragmatics, Social Comparisons, and Common Knowledge Allan Hazlett

8.1 Introduction A diverse family of epistemologists is united by a common commitment to the idea that, when certain conditions are met, you are entitled to rely on a source of information, even if you do not possess evidence of the reliability of that source. I use “source of information” in a broad sense, so that “sources of information” include cognitive faculties or processes (e.g. sense perception, memory), beliefforming methods or strategies (e.g. inference to the best explanation, deduction), and people or institutions (e.g. your friends, The Economist). The different members of the aforementioned family are distinguished by what they take to be necessary and sufficient conditions on entitlement. Some (“reliabilists”) maintain (roughly) that you are entitled to rely on a source of information if and only if beliefs formed on the basis of that source tend to be true.¹ Others (certain “internalists”) maintain (roughly) that you are entitled to rely on a source of information if and only if, so far as things seem from your perspective, beliefs formed on the basis of that source tend to be true.² Finally, others maintain (roughly) that you are entitled to rely on a source of information if and only if relying on that source is a necessary condition on the possibility of inquiry or theoretical rationality, either in general or when it comes to some particular domain.³ The members of the entitlement family are united in their opposition to those (“evidentialists”) who maintain that you are entitled to rely on a source of information only if you possess evidence of the reliability of that source. The members of the entitlement family can offer a powerful rebuke to one kind of philosophical skeptic. This skeptic argues that you do not possess evidence of the reliability of sense perception, for example, in virtue of the fact that you cannot rule out the possibility that you are deceived by a malicious demon of utmost power and ¹ Goldman 1979; Sosa 1991. ² Foley 1987, 1993. This formulation captures the spirit, but not the letter, of the view I defend in Hazlett 2013. Some versions of coherentism can also be interpreted as articulations of this idea. ³ Williams 1996; Wright 2004a, 2004b; Hazlett 2006; Wedgwood 2011; see also Wittgenstein 1969. Allan Hazlett, Moorean Pragmatics, Social Comparisons, and Common Knowledge In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0008

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



cunning who has employed all their energies in order to deceive you, and therefore that you ought to suspend judgment about the deliverances of sense perception. The members of the entitlement family reply: possession of such evidence is not necessary for our being entitled to rely on sense perception, and therefore the premise that we cannot rule out the possibly that we are deceived does not support the conclusion that we ought to suspend judgment. However, the members of the entitlement family seem drastically underprepared when it comes to another kind of philosophical skeptic. I mean the one who argues as follows: 1. You know that you have hands only if you know that you’re not deceived by a malicious demon. 2. You don’t know that you’re not deceived by a malicious demon. 3. Therefore, you don’t know that you have hands. The argument has two premises, and so there are four possibilities: either premise (1) is false, or premise (2) is false, or the argument is invalid, or the conclusion of the argument is true. Which of these four possibilities obtains, according to the members of the entitlement family? Here I shall argue is that there is good reason, having nothing to do with its being true, to expect premise (2) to seem true. If my argument succeeds, the members of the entitlement family would be in a position to explain why this premise seems true, even though it isn’t. To do this, they would need to give a positive explanation of why premise (2) is false (e.g. by appeal to our entitlement to rely on certain sources), which could then be supplemented with my account of why premise (2) seems true. There are other ways for a member of the entitlement family to go, in responding to the present skeptical argument. They might reject premise (1), and with it the “closure principle.”⁴ They might argue that the argument is invalid, a move typically associated with semantic contextualism. And they might also accept the conclusion of the argument. Wittgenstein (1969), who has been interpreted as a member of the entitlement family, sometimes suggests this; he says that he “should like to say: Moore does not know what he asserts he knows [i.e. that he has hands], but it stands fast for him, as also for me.” (§151; see also §137, §407) So perhaps we should understand Wittgenstein’s view to be that Moore’s belief (conviction, certainty) that he has hands is justified (warranted, permissible), perhaps in virtue of his being entitled to trust certain sources, but he does not know that he has hands. In any event, I shall explore here, from an entitlement-friendly perspective, the idea that Moore does indeed know that he has hands. I said that there are four possibilities when it comes to the present skeptical argument: either premise (1) is false, or premise (2) is false, or the argument is invalid, or the conclusion of the argument is true. Norman Malcolm (1949) argues that these are not the only four possibilities, on the grounds that it is sometimes neither the case that I know that p nor the case that I do not know that p, given that “Either I know or I don’t know” “is a rule which applies to these expressions when they occur in their normal ⁴ Harman and Sherman 2004; Sherman and Harman 2011; Coliva 2012.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

contexts,” but not in “unnatural contexts.” Compare “Either my desk is good-natured or it isn’t.” “It would be fantastic to insist that either my desk is good-natured or it isn’t and that it must be one or the other.” (218) I think Malcolm’s argument here is confused. His desk is not good-natured, but it seems false to say this when it is heard as having the implication that his desk has a personality. It is hard to see why you would say unless x had a personality, which you were evaluating, and so saying “Malcolm’s desk is not good-natured” seems to imply that Malcolm’s desk has a personality. But it is nonetheless true that Malcolm’s desk is not good-natured. I’ll proceed as follows. First (§8.2), I’ll review a debate between Malcolm and G.E. Moore about the correctness of Moore’s utterance of “I know that I have hands” in some of his lectures. We’ll take away the idea that knowledge attributions sometimes seem false when they are heard as having misleading implications. Second (§8.3), I’ll argue that knowledge attributions often function as social comparisons. We’ll take away the idea that knowledge attributions sometimes seem false when they are heard as having misleading implications about the would-be knower’s relative epistemic position. Third (§8.4), I’ll argue that these two ideas explain why attributions of common knowledge sometimes seem false. Moore’s utterance of “I know I have hands” is an example of this, but also the anti-skeptical philosopher’s utterance of “I know that I’m not deceived by a demon.” This explains why premise (2) of the present skeptical argument seems true, without appeal to its being true.

8.2 Moorean Pragmatics In his lectures and essays on idealism and skepticism, G.E. Moore frequently appeals to premises of the form , where the proposition that p is something obvious or “commonsensical.” For example: • In “A Defence of Common Sense” (1993: ch. 7), Moore says he knows, with certainty, that each of a list of “truisms” is true, including that “[t]here exists at present a living human body, which is my body.” (107) • In “Proof of an External World” (1993: ch. 9), Moore says that he knows the premises of an argument against idealism, one of which premises is “that two human hands [namely, his own] exist.” (166–7) • In “Hume’s Theory Examined” (1993: ch. 5), Moore says (of a pencil) “I know this pencil to exist” (77) and appeals to this as a premise in an argument against two Humean principles. • In “Four Forms of Scepticism” (1959: ch. 9), Moore says (of a pencil) “I . . . know that this is a pencil” (222) and appeals to this as a premise in an argument against four Russellian assumptions. Norman Malcolm (1949; 1963) and Wittgenstein (1969) criticize these appeals, and their criticism is linguistic in nature: Malcolm (1949) says that “there is something wrong with Moore’s assertions” and that Moore’s use of knowledge attributions “is contrary to their ordinary and correct use,” (202) and Wittgenstein (1969) describes “the wrong use made by Moore of the proposition ‘I know . . . ’.” (§178) There are several reasons that Malcolm and Wittgenstein draw these conclusions, but central to

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



their critique is the idea that Moore cannot give grounds for believing the propositions that he claims to know. Malcolm argues that, in paradigm cases of “ordinary and correct” use, “the person who asserts ‘I know that that’s a tree’ is able to give a reason for his assertion.” (1949: 203) And Wittgenstein (1969) says: “I know” often means: “I have proper grounds for my statement.” (§18) One says “I know” when one is ready to give compelling grounds. “I know” relates to a possibility of demonstrating the truth. (§243; see also §91, §111, §438, §504)

The evidence for this comes from the fact that, so Malcolm (1949) argues, in response to someone’s claim to know, “it would be natural to ask ‘How do you know?’ ” which would amount to “a request for a reason, for proof, for evidence.” (208) “In ordinary discourse,” he points out, “we are reluctant to say that someone knows that so and so is true if he cannot give some reason or some evidence for saying that so and so is true.” (210) But Moore, in the cases described above, by his own admission in some cases (1993: 169), cannot give grounds for the propositions that he claims to know. He insists that knowledge does not require the ability to give a proof (170), but Malcolm denies this (1949: 209–13). It seems then that we have a clash of intuitions about the nature of knowledge. Moore is on the right side of this clash. First, there are counterexamples to to the claim that S knows that p only if S can give grounds for believing the proposition that p. For one thing, there are ways of knowing—and thus ways of answering the question “How do you know?”—that do not involve the provision of reasons, proofs, or evidence. I can recognize my friends by seeing their faces, such that I know that Matthew has arrived at the party, even though I could not give any reason or evidence that it is Matthew who has arrived, and certainly not anything like a proof that he has arrived. (Under pressure, you might offer something like reasons or evidence—“It looks like him; I know his face”—but this is surely not what Malcolm and Wittgenstein are talking about, for the same kind of “reasons and evidence” could be given by Moore, in defense of his “truisms.”) For another thing, even though it may be the case that S knows that p only if there is some way that S knows that p, there are counterexamples to the claim that S knows that p only if S can articulate or describe the way that they knows that p. As A.J. Ayer points out in The Problem of Knowledge (1956): [W]e cannot assume that, even in particular instances, an answer to the question How do you know? will always be forthcoming. [ . . . ] Suppose that someone were consistently successful in predicting events of a certain kind[,] like the results of a lottery. If his run of successes were sufficiently impressive, we might very well come to say that he knew which number would win, even though he did not reach his conclusion by any rational method, or indeed by any method at all. We might say that he knew it by intuition, but this would be to assert no more than that he did know it but that we could not say how. (32–3)

So, again, is false that S knows that p only if S can give grounds for the proposition that p. Moore was right: you know things that you cannot prove. Second, and more important for my purposes in this chapter, we can offer a diagnosis of why it seems like knowledge requires the ability to give grounds. In a

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

letter to Malcolm, dated June 28, 1949 (1993: ch. 13), Moore argues that a distinction must be drawn between what someone implied by saying something and what they asserted (214). In addition to the argument articulated above, that “ordinary and correct” knowledge attribution requires that the person said to know is able to give grounds, Malcolm also argues (1949: 203) that “ordinary and correct” knowledge attribution (in a context) requires that (in that context) there exists (or existed) some doubt about the relevant proposition. Moore’s reply to this idea is that saying may often imply that there exists (or existed) some doubt about the proposition that p, but that this is no part of what is asserted. More generally, Moore insists (215), we must distinguish between a knowledge attribution’s being true or false and its comprising a senseless or sensible thing for a person to say. That a particular knowledge attribution comprises a senseless thing to say—if I were to arise during a solemn eulogy and say, “I know that I had cornflakes for breakfast this morning”—is orthogonal to its truth or falsity. Another way of putting this is that we must distinguish between two ways in which an utterance can be said to be “correct” or “incorrect”: in one sense, “correct” just means true and “incorrect” just means false; in another sense, “correct” means sensible, proper, or purposeful, and “incorrect” just means insensible, improper, or purposeless. Moore’s point has become familiar, in the form of the distinction between what is said and what is conversationally implicated. The idea we need for our present purposes is that of relevance implication. In H.P. Grice’s (1989: 32) familiar example, saying, to a motorist who has just confessed that they are out of petrol, “There is a garage around the corner,” would imply that the garage has, or at least may have, petrol for sale. It would be intentionally misleading to say this if you knew, for example, that the garage was closed for repairs. This is explained, Grice proposes, by the fact that conversational participants expect, and know each other to expect, each other to observe certain principles of cooperation, including the rule “Be relevant.” In the context of a conversation about where to buy petrol, the presence of a garage around the corner is irrelevant unless said garage has petrol for sale. Given the motorist’s expectation that their interlocutor is trying to provide relevant information, they conclude that their interlocutor thinks that the garage has, or at least may have, petrol for sale, and that they are trying to communicate this fact by saying, “There is a garage around the corner.” The idea of relevance implication can be put to work explaining why attributing knowledge sometimes implies that the person said to know has the ability to give grounds (cf. Rysiew 2001; Hazlett 2009).⁵ Just as there is an obvious but imperfect connection between garages and the sale of petrol, such that the mention of a garage in the context of a conversation about where to buy petrol will imply that said garage has, or at least may have, petrol for sale, there is an obvious but imperfect connection between knowledge and the ability to give grounds, such that the

⁵ Malcolm and Wittgenstein come close to agreeing (Malcolm 1949: 216; Wittgenstein 1969: §425, §468, §552). But they resist my conclusion. In response to the idea that Moorean utterances are “odd uses” but not “misuses,” Malcolm (1949) argues that they are nonetheless “not correct.” (215)

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



attribution of knowledge (that p) in a context of conversation about grounds (for believing the proposition that p) will imply that the person who is said to know (that p) has the ability to give grounds (for believing the proposition that p). What do I mean by “a conversation about grounds (for believing the proposition that p)”? We should not expect, in general, to be able to provide an account of what it is for a conversation to be “about” some subject (topic, question, issue). Consider the notion of a conversation about where to buy petrol. The explicit discussion of buying petrol seems sufficient to make a conversation “about” where to buy petrol, but Grice’s own example shows that the question of where petrol is to be bought can be the subject of a conversation, even when this question has not been explicitly discussed. It is enough, in that case, that the conversational participants mutually understand that a motorist out of petrol is likely to want to buy some. Despite our inability to give a general account of what it is for a conversation to be “about” some topic or question or issue, we are able to recognize the subject of a conversation in particular cases. In particular, we are able to recognize, in particular cases, that the subject of a conversation is grounds for some proposition. Consider: S. A team of mathematics students is working together on a proof. Some of them have established that a significant part of the proof can be competed, so long as they can prove a certain lemma, which they all agree seems true. “Does anyone know whether this lemma is true?” someone asks. A student replies, “I know the lemma is true.” And so they get to work on other parts of the proof.

The student implies that they can prove the lemma, because their utterance would have been irrelevant and misleading unless they had such an ability. The students are concerned with constructing a proof, and without proof of the required lemma, their project will fail. But the student does not assert that they can prove the lemma, because the ability to prove the lemma is not required for knowing that the lemma is true—you could know that a lemma was true on the basis of expert testimony, for example, which might leave you completely unable to give grounds for said lemma. We are also able to recognize what kinds of grounds are relevant in a particular conversation. J.L. Austin (1979: ch. 4) writes that “[w]henever I say I know, I am always liable to be taken to claim that, in a certain sense appropriate to the kind of statement (and to present intents and purposes), I am able to prove it.” (85) What I think is right about Austin’s claim is that, when attributing knowledge implies the ability to give grounds, the precise nature of these grounds will depend on the context of conversation. In the context of the conversation described in S, the ability to give grounds is the ability to articulate a mathematical proof. But imagine an analogous case involving lawyers, who recognize the utility of a particular premise for their case. In that context, the ability to give grounds would be something like the ability to establish the premise to the satisfaction of the relevant court. However, we can also recognize cases in which the subject of a conversation is not grounds for any proposition. Consider: G. Smith is a highly reliable predictor of lottery results (as in Ayer’s example, above), which I’ve confirmed by comparing their past predictions to subsequent draws. I’ve decided to

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

use Smith’s powers to my advantage by buying lottery tickets, and have their prediction for tonight’s draw: 12-22-17. You are my investment partner, and are worried about the reliability of my source. “Trust me,” I assure you, “Smith knows what the outcome of the draw will be.” And so we buy the tickets based on Smith’s prediction.

In this case, I do not imply that Smith can prove that the outcome of the draw will be 12-22-17, nor that they can give reasons or evidence in support of that proposition. Indeed, for all I know, Smith cannot give grounds for believing the proposition that the outcome of the draw will be 12-22-17, and we can easily imagine that I know they cannot give such grounds. Nonetheless, my utterance is neither irrelevant nor misleading, because our conversation in this case is not about grounds for believing the proposition that the outcome of the draw will be 12-22-17. We simply want to be sure that this proposition is true, before we place our bet. Although there are interesting differences between first-person (e.g. ) and third-person (e.g. ) knowledge attributions,⁶ there are cases in which first-person knowledge attribution does not imply the ability to give grounds. Smith might say: “I know that the outcome of the draw will be 12-22-17, but I cannot give grounds for believing this proposition. For I do not know how I know; I just know.” So I think Austin’s claim, that whenever I say I know something I am always liable to be taken to claim that I am able to prove it, is false. Now this is not yet to diagnose the appeal of the idea that knowledge requires the ability to give grounds. But that diagnosis can be constructed from these materials. The additional idea we need (Weatherson 2003; Hazlett 2009) is that a claim sometimes seems false when it is heard as having misleading implications. Consider the idea (above) that Malcolm’s desk is neither good-natured nor not good-natured. It seems false to say that Malcolm’s desk is not good-natured, when this claim is heard as implying that Malcolm’s desk has a personality. And so it can seem like Malcolm’s desk is neither good-natured nor not good-natured. The same, mutatis mutandis, when it comes to knowledge attributions. It seems false to say that someone knows that p, when they are unable to give grounds for believing the proposition that p, when this claim is heard as implying that they are able to give grounds for believing the proposition that p. And so it can seem like knowledge requires the ability to give grounds. Given that a claim sometimes seems false when it is heard as having misleading implications, it is easy to confuse something people commonly imply by saying

for a necessary condition on the truth of the proposition that p. ⁶ Consider, in particular, the promissory aspect of first-person knowledge attributions (Austin 1979: 99), which is not present in third-person knowledge attributions. Given the differences between firstperson and third-person knowledge attributions, we should avoid focusing on first-person knowledge attributions when theorizing about the nature of knowledge attributions in general, and especially when theorizing about the nature of knowledge. Compare: saying often implies that you lack first-hand knowledge that p, but this isn’t a condition on being sure that p; saying often implies doubt or hesitation about whether p, whereas this isn’t a condition on the proposition that p’s being consistent with everything you know.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



8.3 Social Comparisons What I want to take away from the discussion so far is that knowledge attributions sometimes seem false when they are heard as having misleading implications. We focused above on the implication that the person who is said to know (that p) is able to give grounds (for the proposition that p). In this section I’ll describe another sort of implication that knowledge attributions can be heard as having. One way of articulating the idea defended above (§8.2) is to say that knowledge attributions often, but not always, function as indicators of who has the ability to give grounds. This claim, about a common “function” of knowledge attributions, is just another way of talking about what people commonly imply when they attribute knowledge, or (which amounts to the same thing) about what implications knowledge attributions are often heard as having. Given this conception of a common function of knowledge attributions, it is obvious that knowledge attributions have a plurality of common functions. (We should bear in mind that the notion of a “common” function is vague, and that there is nothing that is not sometimes the function of a knowledge attribution.) Theorizing about knowledge attributions in functional terms has become popular in contemporary epistemology, and a plurality of functions have been discussed, including “flagging reliable informants” (Craig 1990; Lackey 2012) and signaling the end of inquiry (Kelp 2011; Rysiew 2012). These discussions may employ a different notion of “function” than the notion I’m employing here, but given my notion, the following two claims seem plausible: (i) knowledge attributions often, but not always, function to flag reliable informants, and (ii) knowledge attributions often, but not always, function to signal the end of inquiry. We have considered three plausible candidates for common functions of knowledge attributions: indicating who has the ability to give grounds, flagging reliable informants, and signaling the end of inquiry. To these three candidates I propose to add a fourth: knowledge attributions often, but not always, function as social comparisons. Let me explain. To engage in social comparison is to evaluate the relative standing of two or more people. “Evaluation,” in the relevant sense, requires a valenced comparison in which some are understood as better as or worse than others. Valence need not be explicit, so comparisons in terms of intelligence, where intelligence is tacitly or implicitly understood to be valuable, count as social comparisons. I am borrowing the notion of social comparisons from personality and social psychology, where social comparison activity is seen as central to the way most people represent themselves and other people. We can compare people in terms of their relative epistemic position relative to some proposition, question, or topic. Examples of such comparisons include comparisons in terms of: • Quantity or quality of evidence possessed (where it is assumed that in general more or higher quality evidence is better than less), relative to some proposition, question, or topic. E.g. “Smith has more evidence, relevant to this topic, than Jones.”

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

• Having justified beliefs, as opposed to having unjustified beliefs (where it is assumed that in general justified beliefs are better than unjustified beliefs), about some proposition, question, or topic. E.g. “Smith’s beliefs, about this topic, are justified, by contrast with those of Jones, which are unjustified.” • Amount of truth believed or information possessed (where it is assumed that in general more is better than less), about some proposition, question, or topic. E.g. “Smith knows more about this topic than Jones does.” • Reliability of sources (where it is assumed that in general higher reliability is better than lower reliability), relative to some proposition, question, or topic. E.g. “Smith is more reliable, when it comes to this topic, then Jones.” However, social comparisons need not be explicit. Saying, “Smith is more intelligent than Jones,” amounts to explicit social comparison in terms of intelligence, but it is easy to imagine contexts in which saying, “Smith is intelligent,” would also amount to social comparison in terms of intelligence. Consider: N. You are a university teacher, and are asked by the dean to nominate your most intelligent student for an award. “Smith is intelligent,” you reply.

In this context, you engage in implicit social comparison in terms of intelligence, by implying that Smith is more intelligent than your other students. The same point applies when it comes to social comparisons in terms of epistemic position. In particular, knowledge attributions can amount to implicit social comparisons in terms of epistemic position. This may be easiest to see when it comes to attributions of the form . Consider: H. You and a group of colleagues suspect, but do not know, that Kant never read Hume in English. You then point out that Smith, a student whose dissertation concerns Kant’s sources, knows about Kant’s sources.

In this context, you engage in social comparison in terms of epistemic position, by implying that Smith’s epistemic position, relative to the question of whether Kant read Hume in English, is better than that of you and your colleagues. The same, mutatis mutandis, when it comes to attributions of the form , e.g. saying, “Smith knows whether Kant read Hume in English,” in H. But this point applies to attributions of the form as well. Consider: Q. All the members of your pub quiz team suspect that Utah’s state motto is “Industry.” However, because a wrong answer will be penalized, the question is raised as to whether any of you know that Utah’s state motto is “Industry.” No one voices certainty, but you point out that Smith, despite their silence, surely knows that Utah’s state motto is “Industry,” since they wrote their dissertation on the concept of industry in Utahan identity politics.

In this context, this amounts to social comparison in terms of epistemic position, by implying that Smith’s epistemic position, relative to the question of whether Utah’s state motto is “Industry,” is better than that of you and the other members of the team. Knowledge attributions often function as social comparisons. How shall we explain this fact? Contrastivists in epistemology (Schaffer 2005, 2007; Morton

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



2013) maintain (roughly) that (propositional) knowledge attributions are contrastive: an attribution of propositional knowledge to S is true or false depending on whether S knows that p rather than that q, where the contrast proposition that q is often implicit and supplied by conversational context. We might follow their lead and say that an attribution of knowledge to S is true or false depending on whether S rather than S* knows, where the contrast person S* is often implicit and supplied by conversational context. However, it is easier (because simpler and more elegant) to understand the social comparisons (in terms of epistemic position) involved in knowledge attributions as instances of relevance implication (§8.2). Saying that Smith knows, in the cases just described, implies Smith’s superior epistemic position, relative to some salient question, in virtue of the relevance of epistemic position, relative to that question, in the contexts described. Was Smith’s position not superior, saying that Smith knows, in said contexts, would be misleading. These cases are not unusual or abnormal. And so, more generally, we can conclude that attributing knowledge often implies that the knower’s epistemic position is superior to that of salient others.

8.4 Common Knowledge I have argued that knowledge attributions sometimes seem false when they are heard as having misleading implications (§8.2) and that attributing knowledge often implies that the knower’s epistemic position is superior to that of salient others (§8.3). In this section I’ll put these two ideas together: knowledge attributions sometimes seem false when they are heard as implying that the epistemic position of the person said to know is superior to that of salient others. This predicts that true attributions of common knowledge will sometimes seem false. This, I shall argue, explains why the antiskeptical philosopher’s utterance of “I know that I’m not deceived by a demon” seems false. Just as knowledge does not require the ability to give grounds (§8.2), knowledge does not require enjoying superior epistemic position (§8.3). Whether you know that p depends on you and your relationship with the proposition that p, not on how that relationship compares to other people’s relationship with the proposition that p. Knowing that p does not require doing better, epistemically, than anyone else—it simply requires doing well enough to know. But this means that it might happen that a true knowledge attribution seems false, because it is heard as having the misleading implication that the knower’s epistemic position is superior to that of salient others. Consider a “lottery proposition”—e.g. the proposition that you, an ordinary punter with no inside information, will not win a fair lottery. In some contexts, assertions of lottery propositions are improper, because the person who asserts them misleadingly implies that they have some inside information about the outcome of the draw (cf. Weiner 2005: 232–3; Lackey 2008: 137–8). On my view, in exactly the same contexts, attributions of knowledge of lottery propositions seem false, because the person who attributes knowledge misleadingly implies that the knower has some inside information about the outcome of the draw. It is easy to hear the person who says, “Smith knows that their ticket is a loser,” as implying that Smith has inside

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

information about the outcome of the draw (e.g. that the lottery is rigged). In other words, it is easy to hear this knowledge attribution as implying that Smith’s epistemic position, relative to the outcome of the draw, is superior to that of the other ticket holders. In that case, the attribution of knowledge to Smith may seem false. But there are contexts in which lottery propositions are assertible without impropriety: your friend is maxing out her credit cards on the assumption that they will win the lottery, and you (without having any inside information) admonish them by insisting, “You will not win the lottery!” (Lackey 2008: 137). In this context, attribution of knowledge of a lottery proposition doesn’t seem false: you might just as well say to your friend (who has no inside information), “You know that you’re not going to win the lottery!” And, indeed, even in those contexts in which attributing knowledge of lottery proposition implies that the knower has some inside information about the outcome of the draw, this implication can be cancelled (Weiner 2005: 234). Now, the important thing to notice here is that our account of why attributions of knowledge of lottery propositions often seem false appeals at no point to their being false. What predicts for their seeming false is the fact that the person said to know does not enjoy an epistemic position superior to that of salient others. This, I want to argue, opens the door for an epistemological discussion of whether lottery propositions can be known, but one that sets aside the fact that attributions of knowledge of lottery propositions often seem false. In any event, I submit these cases as evidence that knowledge attributions can seem false when they are heard as implying that the epistemic position of the person said to know is superior to that of salient others. However, this point applies, mutatis mutandis, to attributions of knowledge where the proposition known is common knowledge. I understand “common knowledge” as follows: It is common knowledge that p among the members of some group G if and only if it is presupposed among members of G that (i) each member of G knows that p and (ii) the members of G are in an equally good epistemic position relative to the proposition that p.

Presupposition, in the intended sense, requires mutual awareness of belief, acceptance, assumption, or supposition. In the present sense, something is an item of common knowledge among the members of some group when they, with mutual awareness, take it to be known, and equally well known, by all group members.⁷ Attributing knowledge often implies that the knower’s epistemic position is superior to that of salient others (§8.3). We should expect, then, that attributions of knowledge where the proposition known is common knowledge will seem false when they are heard as implying that the knower’s epistemic position is superior to that of the other members of the relevant group (§8.2).⁸ But it is clear that knowledge

⁷ Note that this secures the non-factivity of “common knowledge,” in the present sense, as in “It was common knowledge among the Aztecs that only human sacrifice could appease Huitzilopochtli.” “Common knowledge” also has a factive sense (cf. Vanderschraff and Sillari 2013). ⁸ However, since relevance implications are understood here as a species of conversational implicature (§1), attributions of common knowledge do not always have this implication—in the jargon, this implication is “cancelable.” Consider, in particular, explicit attributions of common knowledge (e.g. “It’s just

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



attributions of this kind are sometimes true—so, at least, says the anti-skeptic—since it is sometimes true that S knows that p and that the proposition that p is common knowledge. This explains why Moore’s utterance, in his lectures, of “I know that I have hands,” seems false, to the extent that it does. For it is common knowledge, among the delegates at his lectures, that Moore has hands. As Wittgenstein (1969) points out, “[t]he truths which Moore says he knows, are such as, roughly speaking, all of us know, if he knows them.” (§100) That Moore’s utterances are part of a lecture thus has a certain significance, as does the demonstrative character of his famous “proof of an external world.” “Here is one hand,” Moore says, “and here is another.” As Wittgenstein points out, things would have been different had Moore asserted a proposition that only he knew, e.g. “that in such-and-such a part of England there is a village called so-and-so” (§462). A kind of doubt would be possible, on the part of the audience, in that case—a kind of doubt that is not possible when it comes to the proposition that Moore has hands. Indeed, things would have been different had Moore asserted a proposition such that his epistemic position, relative to that proposition, was better than that of the audience members—e.g. that his jacket is made of wool. Doubt would be possible in that case, in a way that it is not possible in the case of the proposition that Moore has hands. Moore’s reasons for asserting items of common knowledge need not concern us here; what matters for our purposes is the fact that their status as items of common knowledge makes it possible to hear Moore’s utterances as falsely implying that his epistemic position is superior to that of the delegates. But this also explains, I want to suggest, why the anti-skeptical philosopher’s utterance of “I know that I’m not deceived by a demon” seems false, to the extent that it does. For the proposition that I’m not deceived by a demon—or something close to it—is an item of common knowledge—or something very much like it— among philosophers, indeed, among people in general. Indeed, the propositions that Wittgenstein (1969) called “hinges upon which our doubts turn” (§341–3) seem to have this in common: they are all items of common knowledge among people in general (cf. Greco 2016). This is what makes it possible to hear the anti-skeptical philosopher’s attributions of knowledge as falsely implying that her epistemic position is superior to that of other people. If I am right, this is good news for the members of the entitlement family, when it comes to responding to the skeptical argument articulated above. The present account would help explain why premise (2) of the skeptical argument seems true, even though it is false. To defend this explanation, the fan of entitlement will need to do two things. First, they will need to defend the idea “hinges” are items of knowledge. This is no mean feat. Consider the idea that you are entitled to trust a source of information if and only if trusting that source is a necessary condition on the possibility of inquiry or theoretical rationality. This, let us suppose, implies that you are permitted to trust the deliverances of sense perception, and this, let us

common knowledge that Pat’s is better than Gino’s!”), which are often used to justify or explain one’s acceptance of some proposition. (Thanks to John Greco for this point.)

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

suppose, implies that you are permitted to believe that you are not deceived by a demon. Does this permission to believe plausibly amount to a way of knowing that you are not deceived? Nothing I’ve said in this chapter makes answering this epistemological question any easier. Second, the fan of entitlement will need to defend the idea that “hinges” are items of common knowledge. But this seems plausible—we do ordinarily take ourselves to know these propositions, and no one seems in a better epistemic position relative to them than anyone else.⁹ The proposed explanation, in terms of social comparisons and common knowledge, strengthens the dialectical position of the members of the entitlement family. We have reason to expect attributions of knowledge of “hinges” to sometimes seem false, even if they are not, if “hinges” are items of common knowledge. And this gives the members of the entitlement family one of the tools they need to challenge premise (2) of the skeptical argument described at the outset.¹⁰

References Austin, J. L. 1979. Philosophical Papers, 3rd edition. Oxford: Oxford University Press. Ayer, A.J. 1956. The Problem of Knowledge. New York: Penguin Books. Coliva, A. 2012. Varieties of Failure (of Warrant Transmission, What Else?!) Synthese 189: 235–54. Craig, E. 1990. Knowledge and the State of Nature. Cambridge University Press. Foley, R. 1987. The Theory of Epistemic Rationality. Cambridge: Harvard University Press. Foley, R. 1993. Working without a Net. Oxford: Oxford University Press. Goldman, A. 1979. What Is Justified Belief? In Justification and Knowledge, ed. G. Pappas, 1–23. Dordrecht: D. Reidel. Greco, J. 2016. Common Knowledge. International Journal for the Study of Skepticism. 6: 309–25. Grice, H.P. 1989. Studies in the Way of Words. Cambridge: Harvard University Press. Harman, G. and Sherman, B. 2004. Knowledge, Assumptions, and Lotteries. Philosophical Issues 14: 492–500. Hazlett, A. 2006. How to Defeat Belief in the External World. Pacific Philosophical Quarterly 87: 198–212. Hazlett, A. 2009. Knowledge and Conversation. Philosophy and Phenomenological Research 78: 591–620. Hazlett, A. 2013. Entitlement and Mutually Recognized Reasonable Disagreement. Episteme 11: 1–25. Kelp, C. 2011. What’s the Point of “Knowledge” Anyway? Episteme 8: 53–66. Lackey, J. 2008. Learning from Words: Testimony as a Source of Knowledge. Oxford: Oxford University Press.

⁹ But work is required here as well: why not think that the reflective person, who has considered philosophical or proto-philosophical reasons to reject premise (2) of our skeptical argument, is in a better epistemic position, relative to the proposition that they’re not deceived by a demon, than someone who, although they may very well know that they are not deceived, has never reflected at all on the problem of skepticism? If the reflective person enjoys a superior epistemic position, relative to “hinges,” then the proposed explanation is problematic. ¹⁰ I presented a version of this chapter at the Basic Knowledge Seminar, Northern Institute of Philosophy, University of Aberdeen, in 2011.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



Lackey, J. 2012. Group Knowledge Attributions. In Knowledge Ascriptions, eds. J. Brown and M. Gerken, 243–69. Oxford: Oxford University Press. Malcolm, N. 1949. Defending Common Sense. Philosophical Review 58: 201–20. Malcolm, N. 1963. George Edward Moore. Knowledge and Certainty: Essays and Lectures, 163–83. New York: Prentice-Hall. Moore, G.E. 1959. Philosophical Papers. London: Allen and Unwin. Moore, G.E. 1993. Selected Writings. London: Routledge. Morton, A. 2013. Contrastive Knowledge. In Contrastivism in Philosophy: New Perspectives, ed. M. Blaauw, 101–15. New York: Routledge. Rysiew, P. 2001. The Context-Sensitivity of Knowledge Attributions. Noûs 35: 477–514. Rysiew, P. 2012. Epistemic Scorekeeping. Knowledge Ascriptions, eds. J. Brown and M. Gerken, 270–94. Oxford: Oxford University Press. Schaffer, J. 2005. Contrastive Knowledge. Oxford Studies in Epistemology 1: 235–73. Schaffer, J. 2007. Knowing the Answer. Philosophy and Phenomenological Research 75: 383–403. Sherman, B., and Harman, G. 2011. Knowledge and Assumptions. Philosophical Studies 156: 131–40. Sosa, E. 1991. Knowledge in Perspective: Selected Essays in Epistemology. Cambridge: Cambridge University Press. Vanderschraff, P. and Sillari, G. 2013. Common Knowledge. In Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/ Weatherson, B. 2003. What Good Are Counterexamples? Philosophical Studies 115: 1–31. Wedgwood, R. 2011. Primitively Rational Belief-Forming Processes. In Reasons for Belief, eds. A. Reisner and A. Steglich-Petersen, 180–200. Cambridge: Cambridge University Press. Weiner, M. 2005. Must We Know What We Say. Philosophical Review 114: 227–51. Williams, M. 1996. Unnatural Doubts: Epistemological Realism and the Basis of Scepticism Princeton, NJ: Princeton University Press. Wittgenstein, L. 1969. On Certainty, trans. D. Paul and G.E.M. Anscombe. New York: Harper and Row. Wright, C. 2004a. Wittgenstein Certainties. In Wittgenstein and Scepticism, ed. D. McManus, 22–55. London: Routledge. Wright, C. 2004b. Warrant for Nothing (and Foundations for Free)? Proceedings of the Aristotelian Society 78 (suppl.): 167–212.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

9 Internalism and Entitlement to Rules and Methods Joshua Schechter

9.1 Introduction Here are two of the most important questions within epistemology: (i) What explains the fact that we are by-and-large reliable in believing truths and disbelieving falsehoods? (ii) What explains the fact that we are by-and-large justified in believing and disbelieving what we do? These two questions have somewhat different flavors. We know—more or less— what it is to be reliable. To answer the first question, what is needed is an explanation of why it is that our beliefs have this status.¹ In contrast, there is no uncontentious account of the nature of justification. To answer the second question, we must develop an account of what it is to possess this epistemic status. In answering these questions, progress can be made by focusing not on beliefs, but on reasoning (and on our transitions in thought more generally).² As has been argued by Williamson and by Hill, it is plausible that our modal beliefs are reliable because we are reliable in our subjunctive reasoning.³ As I have argued elsewhere, our beliefs about logic are reliable because we are reliable in our deductive reasoning.⁴ Similarly, our basic perceptual beliefs are reliable because the perceptual mechanisms that generate them are reliable. And so on for many other domains. Not only is this change of focus helpful for making progress toward answering the first question, it is helpful for making progress toward answering the second. It is plausible that our modal beliefs are justified (in part) because we are justified in reasoning subjunctively as we do. We are justified in our logical beliefs (in part) because we are justified in reasoning deductively as we do. We are justified in our basic perceptual beliefs because we are justified in moving from experiential states to perceptual beliefs as we do. And, again, the analogous claims apply for many other domains. ¹ This question is perhaps most pressing for a priori domains such as logic, pure mathematics, and (fundamental parts of) morality. ² In this chapter, I use “reasoning” to refer to theoretical rather than practical reasoning. ³ See Hill (2006) and Williamson (2007), ch. 5. ⁴ See Schechter (2010). Joshua Schechter, Internalism and Entitlement to Rules and Methods, Social Comparisons, and Common Knowledge In: Epistemic Entitlement. Edited by: Peter J. Graham and Nikolaj J. L. L. Pedersen, Oxford University Press (2020). © the several contributors. DOI: 10.1093/oso/9780198713524.003.0009

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



According to this line of thought, the epistemic credentials of our reasoning (and our transitions in thought more generally) are—at least in many cases—prior to the epistemic credentials of our beliefs. The nature of reasoning is a vexed topic. But on a very natural picture, reasoning is a rule-governed activity. Our reasoning is governed by rules of inference. More generally, our transitions in thought are governed by belief-forming methods. There are three principal grounds for this claim. First, appealing to rules and methods is the most promising strategy for explaining the difference between genuine reasoning and mere change in belief. Second, we are familiar with two different sorts of mistakes thinkers may make in their reasoning—errors of competence and errors of performance. Appealing to rules and methods can help to explicate this distinction: Thinkers may employ the wrong rules, or they may misapply the rules that they employ.⁵ Finally, there are good candidates for the rules and methods that we in fact employ—deductive rules such as Modus Ponens, ampliative rules such as Inference to the Best Explanation, as well as perceptual methods that tell us to believe what perceptually appears to be the case. A terminological clarification may be helpful here. In what follows, I’ll use ‘rules of inference’ as a subset of ‘belief-forming methods.’ Rules of inference are those beliefforming methods that are relevant to inferential reasoning. As a rough approximation, we can think of them as belief-forming methods where both the inputs and outputs are beliefs or belief-like states.⁶ It shouldn’t be pretended that there are no obscurities in a rule-governed picture of reasoning.⁷ On pain of infinite regress, it must be the case that we can follow rules of inference without explicitly representing them in our thought. It is difficult to understand how this could be possible. How is it that thinkers can count as genuinely following rules—and not merely conforming to them—without explicitly representing them? The answer to this question is not at all clear. Moreover, it is natural to think that certain rules are rules of permission. Other rules are rules of obligation. It is not clear how this distinction can be made out for rules that are followed but not explicitly represented. These are genuine difficulties for a rule-based conception of reasoning. But I am not aware of any alternative picture that can do the same work in explaining the nature of reasoning. There is an important distinction that ought to be raised here, one that will be important in what follows. This is the distinction between basic and non-basic beliefforming methods. Certain belief-forming methods are basic (for us). They are basic in an intuitive sense—they are the most fundamental methods we employ in reasoning. In other words, these methods are the ones that are employed but not in virtue of employing any other methods. Plausible examples of such methods include Modus Ponens, Inference to the Best Explanation, as well as the fundamental methods governing our perceptual, modal, and moral thought.⁸ ⁵ See Boghossian (2005). ⁶ This may need to be broadened to include reasoning with suppositions, as occurs in reductio arguments and in Reasoning by Cases. ⁷ See Kripke (1982) and Boghossian (1989, 2008) for discussion. ⁸ See Wedgwood (2002) for a discussion of basic methods. There are other basic/non-basic distinctions. For instance, a method may be basic in the sense that having justification to employ it does not require

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

Just as a thinker’s beliefs can be justified or unjustified, so too can her employment of belief-forming methods. These two notions are related in roughly the following way: A thinker is justified in employing a rule just in case the thinker is pro tanto justified in believing the output of the rule when it is applied to justified input beliefs, and is justified in believing the output because the belief was so formed. More generally, a thinker is justified in employing a belief-forming method just in case the thinker is pro tanto justified in believing the output of the method when it is applied to justified inputs (where the question of the justification of the inputs arises), and is justified in believing the output because the belief was so formed.⁹ Notice that this claim is not merely a biconditional. It also has an explanatory component. In particular, beliefs are justified because they were formed by applying justified belief-forming methods to justified inputs (where the question of justification arises). This explanatory claim is intuitively very plausible. To give a simple example, a thinker may be justified in believing a mathematical theorem because she inferred it from more basic mathematical principles she was justified in believing using deductive rules she was justified in employing. Similarly, a thinker may be justified in believing that there is an apple on the table because she had a certain visual experience and she transitioned from the experience to the belief using a beliefforming method she was justified in employing. This general picture of the relationship between belief, reasoning, and rulefollowing raises the following important questions about justification: In virtue of what are we epistemically justified in employing certain belief-forming methods and not others? In particular, in virtue of what are we justified in employing many of the beliefforming methods that we employ?

I take these questions to be central questions about epistemic justification. They are the primary questions in this area of epistemology. In this chapter, I’m not going to answer these questions directly. Rather, I’m going to discuss various desiderata on an acceptable answer. There are several desiderata on a satisfying account that can be agreed upon by (nearly) everyone. Here are six: (i) The account of justification should be extensionally adequate. In particular, its extension should seem reasonable—it should moreor-less fit with our pre-theoretic judgments. Where it doesn’t, there must be some story about how conflicting judgments are to be explained away. (ii) The account should show why we are justified in employing certain methods. That is, it should not merely provide necessary and sufficient conditions. It should also present an explanation of why it is that we are justified. (iii) The account should provide the fundamental explanation of our justification. It should identify the fundamental normative principles governing this normative status. Of course, one may be interested in discovering a correct account of justification, whether or not it is fundamental. But it would be most satisfying to provide the fundamental account. (iv) The being justified in believing that it is reliable. I believe that there are basic rules in this second sense, too. That is, in fact, part of the point of this chapter. But I’ll officially use ‘basic’ in a non-normative sense. ⁹ These theses must be further generalized to handle cases where the output is not a belief, but a disbelief or a different kind of mental state.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



account of justification should be unified, non-ad hoc, and possess the usual theoretical virtues. (v) On the account, justification should be appropriately connected with truth. There should be a sense in which we are more likely to arrive at the truth using justified methods than non-justified methods.¹⁰ (vi) Finally, on the account, justification should be appropriately normatively significant. In other words, the account should show justified methods in a positive rational light. To borrow a term used by Feigl, the account should vindicate justified methods.¹¹ The central issue of this chapter concerns another class of putative constraints. The issue is whether there are any internalist constraints on an adequate account of the justification of rules inference and belief-forming methods. What do I mean by ‘internalist’ in this context? Here are several candidate internalist constraints: (i) A thinker is justified in employing a belief-forming method only if the thinker has a (doxastically) justified belief that the method is reliable (or that it has some other positive epistemic status).¹² (ii) What makes it the case that a thinker is justified in employing a beliefforming method is solely a matter of what is internal to the thinker’s mind. (iii) Whether a thinker is justified in employing a belief-forming method supervenes on the thinker’s internal mental states. (iv) Whether a thinker is justified in employing a belief-forming method supervenes on what’s accessible to the thinker by reflection alone. In what follows, I’m going to argue against such constraints. There are no internalist constraints on an adequate account of our justification to employ rules of inference and belief-forming methods. ‘Rule internalism’ and ‘method internalism’ are false.

9.2 Justification, Entitlement, and Epistemic Responsibility Before I turn to rule and method internalism, it is important to get more of a fix on the target notion of justification. One of the lessons of contemporary epistemology is that there may be several different properties that are referred to by ‘justification.’ There are certainly different concepts of justification. For instance, some philosophers characterize justification as that which must be added to true belief to make knowledge (modulo some condition to handle Gettier cases).¹³ This is not the notion that I have in mind. The concept of justification that I will work with here may well be conceptually primitive. If you don’t already possess it, it may be difficult or impossible for me to communicate it to you. But I hope that you share the notion with me. ¹⁰ Similarly, justification should be appropriately tied to evidence. ¹¹ See Feigl (1952). ¹² An alternative constraint would require that the thinker has proposition justification to believe that the method is reliable (or otherwise has some positive epistemic status). As I argue later in this chapter, such a constraint lacks the intuitive motivation of the constraint requiring a doxastically justified belief. ¹³ This is what Plantinga (1993) calls ‘warrant.’

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

It is common to introduce this notion by talking about what a thinker epistemically ought to believe. By analogy to morality, a distinction is then made between objective and subjective oughts. It is the latter that is claimed to be of interest.¹⁴ A thinker is justified in believing a proposition just in case she subjectively ought to believe it. A thinker is justified in employing a method just in case she subjectively ought to employ it. That, it is claimed, is what epistemic justification is. I find this way of introducing the notion helpful, but also somewhat problematic. It is controversial whether there really is a principled distinction between subjective and objective oughts. More importantly, justification is more of a notion of permission than a notion of obligation. We are permitted to believe what we are justified in believing. At least in some cases, we are not also obligated to believe it. I find it more helpful to say that a thinker is justified in believing a proposition just in case the thinker is epistemically responsible in holding the belief. (For some reason, it is sometimes more helpful to say that the thinker is ‘not epistemically irresponsible’ in holding the belief.) A thinker is justified in employing a beliefforming method just in case the thinker is epistemically responsible in employing the method.¹⁵ To state the obvious: Justification is a positive normative status. It is also an epistemic status, not a moral or pragmatic status. To state the less obvious: Being justified in holding a belief is not the same as being epistemically blameless in holding it. This is for two reasons. First, we do not have a systematic practice of epistemically blaming people for their beliefs or reasoning prowess.¹⁶ We don’t typically have any of the Strawsonian reactive attitudes—blame, guilt, and resentment—to people who have a belief that is not properly based on the evidence or who otherwise reason poorly.¹⁷ There is a broader sense of ‘blame’ according to which we sometimes seem to epistemically blame people for their beliefs and reasoning. But it doesn’t seem that we do so in any very systematic way. Second, it can be natural to describe thinkers as blameless but unjustified. This can happen when they have a (good) excuse for their belief.¹⁸ For example, if someone reasoned incorrectly on some trivial matter because they were distracted by something important,

¹⁴ This is because it is tempting to think that thinkers objectively ought to believe all and only the truths. (Interestingly, the analogous claim for rules doesn’t hold—there is no sense of ‘ought’ on which thinkers objectively ought to reason in every truth-preserving way.) On an alternative view, familiar from Williamson (2000), thinkers objectively ought to have all and only those beliefs that constitute knowledge. ¹⁵ I should point out that adopting a responsibilist conception of justification makes my task of arguing against internalism harder. Many externalists give up on anything like a responsibilist conception. That is part of what’s behind the familiar charge that endorsing reliabilism about justification is a matter of changing the subject. ¹⁶ As an anonymous referee reminds me, we sometimes morally blame people for their beliefs—e.g., racist and sexist beliefs. And the appropriateness of such blame can depend on whether the thinker reasoned correctly. But the kind of blame in question is moral blame. ¹⁷ See Strawson (1962). ¹⁸ See Austin (1956) for the classic discussion of excuses. See Littlejohn (forthcoming) and Williamson (forthcoming) for a discussion of excuses as they are relevant to epistemology.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



a natural description of the situation is that they are blameless but unjustified in their belief.¹⁹ There is another clarification worth making: To responsibly believe a proposition or employ a method, a thinker need not have carried out the relevant inquiry in a fully responsible way. A person can be responsible in believing a proposition given her evidence despite the fact that she has been lazy or careless in collecting evidence and inquiring into the issue. The same holds true in the practical domain: A person can behave responsibly in solving a problem, even though the person is otherwise irresponsible, the problem was of her own making, etc. This, then, is the target notion of justification I’ll be working with. Of course, there are several questions and obscurities that remain. But I take it that the notion is at least tolerably clear for my purposes here. Finally, let me say a bit about entitlement. The term ‘entitlement’ is used in several different senses in the literature.²⁰ On one important usage, a thinker is entitled to believe a proposition just in case the belief is warranted but is not so warranted on the basis of being supported by evidence. In this chapter, I’ll use ‘entitlement’ in a related way, but in application to rules of inference and belief-forming methods. On my usage here, a thinker is entitled to employ a belief-forming method just in case the thinker is justified in employing the method as basic but is not so justified on the basis of having a doxastically justified belief (or has propositional justification to believe) that the method is reliable, justified, or has some other positive epistemic status.²¹ This characterization is not strictly analogous to the characterization of entitlement for beliefs—for instance, it does not directly appeal to what is (or is not) supported by evidence. But the central idea is similar. Most of the discussion to follow will concern justification rather than entitlement. But the nature of entitlement to rules and methods will loom large toward the end of this chapter.

9.3 The Argument from Examples Should we accept an internalist constraint on providing an account of the justified employment of belief-forming methods? I claim that the answer is no. To give a bit of the game away, there will be several general morals of this chapter. First, the intuitive motivations for internalist constraints are heterogeneous. (This is a familiar claim from the literature.) Second, these motivations directly motivate distinct internalist constraints. (This is also familiar from the literature.) What may be less familiar is a third moral: The main motivations for internalist constraints directly motivate extremely strong internalist constraints. In particular, they do not motivate ¹⁹ The same holds true for action. We do sometimes want to say of someone that they have acted irresponsibly—for instance, in failing to turn off the oven—without being blameworthy—for instance, because they were distracted by something important. ²⁰ See Burge (1993), Dretske (2000), and Wright (2004) for classic discussions of entitlement. ²¹ In the literature, the terms ‘entitlement’ and ‘justification’ are typically employed so that they stand for incompatible epistemic statuses. In this chapter, I use the term ‘justification’ broadly. I take entitlement to be a species of justification. I find this terminology more congenial. It also better fits with the terminology in Enoch and Schechter (2008).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

restricting the internalist theses to rules of inference rather than to belief-forming methods more generally. They also do not motivate internalist constraints that appeal to considerations of access. In what follows, I’m going to focus on three motivations for internalist constraints. These are the ones that I find the most compelling. In the remainder of this section, I will discuss a motivation based on simple cases. In Section 9.4, I will discuss a motivation based on a general conception of what it is to be a responsible thinker. In Section 9.5, I will discuss a motivation based on our reactions to skeptical scenarios.²²

9.3.1 Examples The first motivation for rule internalism is based on an examination of several simple examples. The most striking kind of case is that of our use of measuring instruments—for instance, our use of thermometers. The following claim is very intuitive: A thinker is justified in relying on her favorite thermometer to determine the local temperature only if she has an independently justified belief that the thermometer is reliable. The grounds justifying the belief need not be specific to the thermometer in question. For instance, it may suffice that the thinker justifiably believes that manufactured thermometers tend to be pretty reliable. But some independent justification of the reliability belief seems necessary. (The restriction to independent justifications is needed because it clearly would be no good if the thinker was convinced of the reliability of the thermometer by testing it against itself. This requirement is, in effect, a ‘no bootstrapping’ constraint.) An analogous claim is intuitive for more complex devices. Suppose you own a complicated computational device that outputs a grammatical sentence of English whenever you hit a particular button. A thinker is not justified in believing the output of such a device unless she has an independently justified belief that the device tends to output truths. Generalizing from these examples, the following principle is very attractive: Measuring Instruments: A thinker is justified in relying on a measuring instrument only if the thinker has an independently justified belief that the instrument is reliable. ²² There are other motivations for internalist constraints but they strike me as less powerful than the three I discuss in the main body of the text. Here are a few: (i) Justification is a matter of belief-guidance. Indeed, Goldman (1999) suggests that this is the main motivation for internalism. I do not have the space to discuss this motivation here, other than to say that I agree with much of what Goldman says in response to it: Either the constraint is way too strong or it is incapable of supporting internalism. (ii) We have a (rather inchoate) intuition that justification is purely a matter of what’s ‘going on’ with the thinker. I take it that this intuition should not be taken fully seriously, since (for example) the fact that we are justified in employing the rule Modus Ponens seems to also depend in part on the logical fact that Modus Ponens is valid, which is not a fact about the thinker. (iii) Justification for employing a method can be defeated by a justified belief that the method isn’t reliable. This, it might be argued, suggests that a justified belief that the method is reliable was somehow responsible for the justification. This suggestion is untenable for the case of deductive rules, for the reasons familiar from discussions of Carroll (1895). (iv) The unpalatable nature of Moore-like sentences—‘P, therefore Q but I don’t believe that that rule is good’—suggests that we adopt an internalist constraint. I take it that Moore-like sentences are not very probative, since they can be used to support absurdly strong constraints on justification and knowledge.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



There is some intuitive pull in favor of the view that an analogous claim applies to cases of testimony. Take, for instance, a thinker’s reliance on an oracle, say the oracle at Delphi. It is plausible that a thinker is justified in relying on the testimony of an oracle only if she has an independently justified belief that the oracle is reliable. Similarly, a member of a jury is justified in relying on the testimony of a witness at a trial only if she has an independently justified belief that the witness is reliable. Generalizing from these cases, we arrive at the following principle: Testimony: A thinker is justified in relying on the testimony of a speaker only if the thinker has an independently justified belief that the speaker’s testimony is reliable. This principle is somewhat less convincing than the first. Indeed, it is highly contentious in the literature on testimony. But the principle is prima facie at least somewhat plausible. Taking these examples seriously, there is a natural further generalization to make: A thinker is justified in relying on a belief-forming method in her thinking only if the thinker has an independently justified belief that the method is reliable. Notice that this constraint requires that the thinker actually have a justified belief in the reliability of the method. That is the constraint that the examples directly support. If a thinker doesn’t actually believe that a thermometer is reliable, for instance, she shouldn’t make use of it. It is easy to motivate further strengthening this generalization from a conditional to a biconditional. What more could be required to be justified in employing a method than to have a justified belief that the method is reliable? Having the justified belief that a particular thermometer is reliable would seem to suffice for being justified in relying on it. The examples also motivate a claim about explanation: The thinker is justified in relying on the thermometer because she has a justified belief that it is reliable. It is in virtue of her justified belief that she is justified in employing the method. It is worth pointing out that reliability may not be the only relevant status. If a thinker has a justified belief that she is justified in relying on her favorite thermometer—whether or not she has a justified belief that it is reliable—she may still be justified in relying upon it. So we can slightly liberalize our principles.²³ Putting this all together, we arrive at the following general thesis: Extreme Method Internalism: A thinker is justified in employing a belief-forming method just in case and by virtue of the fact that the thinker has an independently justified belief that the method is reliable (or is a method that she is justified in employing, or otherwise has a positive epistemic status). For simplicity, let us call a belief that a method is reliable (or is a method that one is justified in employing or otherwise has a positive epistemic status) a ‘backing belief ’ of the method. Extreme Method Internalism is the thesis that a thinker is justified in ²³ I owe this observation to Paul Boghossian.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

employing a method just in case and by virtue of the fact that she has a justified backing belief of the method. This is the thesis that is directly motivated by the examples. It is a very strong version of method internalism. One might try to further support this thesis by appealing to BonJour’s case of Norman, the reliable clairvoyant, or to Lehrer’s case of Mr Truetemp.²⁴ In these cases, the relevant thinkers lack justified backing beliefs for their methods. They also apparently fail to be justified in employing them. These cases, then, might be thought to be something like negative controls for Extreme Method Internalism. However, the BonJour and Lehrer cases are very tricky to evaluate. Our intuitions about them fluctuate when we make apparently minor modifications to them. Consider the case of a reliable clairvoyant for whom clairvoyance has a rich quasi-perceptual phenomenology. It is less obvious that such a clairvoyant is not justified in relying upon his clairvoyance. Similarly, modify the case so that the clairvoyant doesn’t have a visual faculty but only a faculty of clairvoyance. Our intuitions—or, at least, my intuitions— get muddier, still. Evaluating these kinds of cases is very tricky. Ultimately, I suspect that such cases are not very probative. Nevertheless, Extreme Method Internalism has an air of plausibility. It fits well with the simple cases we have considered and does not seem obviously false.

9.3.2 Against Extreme Method Internalism Despite its plausibility, Extreme Method Internalism is subject to serious problems. Consider the following two claims: (i) A thinker is justified in employing a belief-forming method just in case and solely by virtue of the fact that the thinker has an independently justified backing belief of the method. (ii) A thinker is justified in holding a belief just in case and solely by virtue of the fact that the belief was generated by the application of a belief-forming method that the thinker was justified in employing to justified inputs (where the question of the justification of the inputs arises). The first claim is a restatement of Extreme Method Internalism. The second claim is independently very plausible. Indeed, it already appeared in the Introduction under a slightly different guise. The trouble is that the two claims cannot both be true. The chief difficulty here is not the usual sort of infinite regress problem. The usual sort of problem would go like this: The two claims jointly entail that to be justified in employing a method, a thinker must have infinitely many justified beliefs. Given the finitude of our epistemic capacities, that is impossible. So at least one of the two claims is false. That may well be a problem, but I’m not going to focus on it here. This is because the line of thought turns on subtle issues concerning the nature of independent justifications. Instead, the difficulty that I’d like to present is that the two claims listed above launch an explanatory regress. Indeed, there are really two problems concerning ²⁴ See BonJour (1980) and Lehrer (2000).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



explanation here. The first is an infinite regress problem, just not an infinite regress of the usual sort. Formally put, the problem is that the explanatory grounding relation—the relation that holds between a fact and the facts that explain it—has no infinite chain.²⁵ In other words, tracing this relation from an explanandum to its explanans, to the explanans of each of the explanans, and so on, must terminate. Since the two claims above entail that there is an infinite explanatory chain, at least one of the two claims must be false. Notice that this problem does not turn on anything to do with independence. It does not turn on the finitude of our capacities. It also does not turn on my earlier claim that the simple examples directly motivate the requirement that the thinker actually holds a backing belief (and is not merely in a position to believe it). Rather, the problem turns on a very general feature of the explanatory grounding relation.²⁶ Why believe that the grounding relation has no infinite chains? As far as I can tell, there is no fancy argument for this claim. However, the claim has been taken for granted by many. It is deeply intuitive that explanations have to bottom out somewhere.²⁷ The second explanatory problem resembles a circularity problem. The two claims listed above jointly yield an explanatory structure with the following shape: The justification of belief-forming methods is completely grounded in the justification of beliefs. The justification of beliefs is completely grounded in the justification of methods (and the justification of beliefs). This is not a possible explanatory structure. To help explicate this problem, consider a clear case of an explanatory circularity. Consider a case in which the fact that p is supposed to totally explain the fact that q, and the fact that q is supposed to totally explain the fact that p. Such an explanatory structure is intuitively problematic. The explanatory grounding relation is plausibly transitive and irreflexive.²⁸ At the very least, its transitive closure is plausibly irreflexive. It cannot contain such loops. What we have here is not quite that strong. There need not be any pair of facts such that each is supposed to completely explain the other. There is no obvious violation of transitivity or reflexivity. But what we do have is a case in which there are two natural classes of facts such that each member of the first class is completely explained by members of the second class, and each member of the second class is completely explained by members of the first class (along with members of the second). This is not a problem with the logical features of the grounding relation, per se. But it is intuitively problematic, as well.²⁹ ²⁵ See Fine (2001, 2012), Rosen (2010), and Schaffer (2009) for discussion of the grounding relation. ²⁶ Notice that this problem still arises if ‘solely’ is replaced by ‘partially’ in the two above claims. ²⁷ In conversation, Jonathan Schaffer has suggested that relations lack infinite chains when they transmit an important status. For instance, since the relation between explanans and the explanandum they explain transmits the status of being explained, there must be minimal explanatory elements. This proposal has some plausibility—without minimal elements, it is difficult to see how the status of being explained can get off the ground. However, there are apparent counterexamples. For example, the relation between a cause and its effects also transmits an important status—that of being causally explained. This does not obviously entail that there are always first causes. ²⁸ But see Schaffer (2012) for argument that metaphysical grounding is not transitive and Jenkins (2011) for argument that metaphysical grounding may not be irreflexive. ²⁹ Notice that this problem also arises if ‘solely’ is replaced by ‘partially’ in the two claims above.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

It may be worthwhile to ward off some immediate concerns about this second explanatory problem. One might think that the explanatory structure is relevantly like that of a recursive definition. Thus, it might be thought that this structure is unproblematic. But the explanatory structure that is entailed by the two above claims does not resemble a recursive definition. To have a genuine recursive definition, there must be basis clauses. On the explanatory structure entailed by the two claims, there is no basis. One might instead think that the explanatory structure is relevantly similar to the structure of explanation posited by theorists who endorse reflective equilibrium or coherentist pictures of justification. Insofar as such pictures do not suffer from an explanatory regress problem, it might be thought that neither does Extreme Method Internalism. This, however, is also a mistake. On coherentist pictures, the justification of a belief-forming method is not explained by the justification of beliefs and vice-versa. Rather, on a coherentist picture, what’s directly justified is not an individual belief or method, but a package of beliefs and methods. What explain the justificatory status of a package are facts about the coherence of the package. On such a picture, there is no need to posit infinite explanatory chains or a quasi-circular structure to explanation.³⁰ I conclude, therefore, that we cannot accept both of claims (i) and (ii) above. If we have to chose between one of these claims, claim (ii) is the winner. It is extremely natural to accept that our beliefs are justified in virtue of being formed by the application of justified belief-forming methods to justified inputs. This is far more natural than accepting claim (i). These problems, of course, do not show that claim (i) is false when shorn of its explanatory implications. That is, I have not said anything that entails that the biconditional underlying claim (i) is false. However, the explanatory problems do show that the motivation for the biconditional goes astray. Since this motivation fails, we are left with no good reason to accept an internalist constraint, at least not on the basis of the simple examples. Why, then, do we feel the pull of Extreme Method Internalism? There is a natural diagnosis to offer: We overgeneralize from examples. Recall the distinction between basic and non-basic belief-forming methods. In our thinking about the justification of belief-forming methods, we need to carefully distinguish between these classes of methods. Extreme Method Internalism is plausible when restricted to non-basic methods. To be justified in employing one of those methods, we need to have a justified backing belief. For basic methods, no justified backing belief is required.

9.4 The Argument from Responsibility 9.4.1 Epistemic responsibility and defending one’s view The second motivation for rule internalism depends on a general conception of what it is to be a responsible believer. The central idea is that it is epistemically ³⁰ For a version of this point, see, for example, BonJour (1985).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



irresponsible to believe a proposition without having a sufficiently strong defense of the belief. This is an attractive thought about epistemic responsibility, at least upon first blush. What is it to have a defense of a belief? Let me say a few things to fill in the intuitive picture. One might make the following claim: It is irresponsible to believe a proposition without possessing an argument in favor of the truth of the proposition. This is much too strong as a construal of the idea that we should have defenses of our beliefs. A defense might be an argument in favor of the truth of the relevant proposition. But, alternatively, it might be an explanation of why it is epistemically acceptable to believe the proposition.³¹ A defense only needs to be something that supports—in either a justificatory or an explanatory way—the thinker’s belief or the proposition that is believed. In short, a defense must somehow legitimate holding the belief. For something to count as a defense, the relevant thinker must be able to recognize it as a defense. The motivating picture is that in believing a proposition, a thinker is putting something forward—she is staking a claim. To legitimately stake a claim, the thinker had better have a defense of it. But not only must she possess a defense of her claim, she had better be able to recognize the defense as such. If she isn’t able to recognize the defense, she doesn’t really have a defense at all. Indeed, it is plausible that the thinker has to do more than be in a position to recognize her defense as a defense. She had better actually recognize it as a defense. A defense of a claim is no good if you don’t in fact recognize it as such, whether or not you are able to do so. This requirement, however, is very strong. It quickly leads to a regress problem. In the interests of charity, then, I won’t insist upon the point. A defense need not be able to rationally convince a skeptic or neutral party. It need only be rationally convincing to the thinker in question. This is one of the lessons of contemporary epistemology—trying to convince a committed skeptic is a mug’s game. But it is a necessary part of having a defense of a belief that the thinker in question be satisfied with the defense and that she be rationally satisfied with it. For a defense to be rationally satisfying, it had better not be trivial or grossly circular. After all, one cannot defend—justify or explain—a belief by appealing to the very belief in question. This motivates a (perhaps weak) independence requirement on defenses. For a defense to be acceptable, it must itself be justified. In particular, the beliefs appealed to in the defense must be justified. The methods employed by the thinker in the defense must be ones that the thinker is justified in employing. I take it that this is straightforward: An unjustified defense is not a defense at all. Finally, it should be acknowledged that the need for a defense (at most) applies to sufficiently cognitively sophisticated thinkers. It is sometimes suggested that some forms of internalism are false on the ground that they wouldn’t enable children and non-human animals to count as justified—they are cognitively too demanding. While I have some sympathy for this general kind of complaint, this particular

³¹ There are other kinds of defenses, too. As an anonymous referee points out, certain kinds of explanations of how it is that I came to have a belief may also count as defenses.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

instance of it seems mistaken. We should think that with greater cognitive power comes greater responsibility.³² Children and animals need not have a defense to count as justified, since (roughly speaking) they couldn’t come up with a defense if they tried.³³ By contrast, we need a defense. This is a consequence of a plausible version of the principle that ought-implies-can. Given these clarifications, we can now state the following constraint on justification: Defense InternalismB: A thinker is justified in believing a proposition only if the thinker has an independently justified defense of the belief. This is a principle concerning the justification of beliefs. By analogy, one might extend this line of thought to rules of inference and to belief-forming methods more generally. This yields the following principle: Defense InternalismM: A thinker is justified in employing a belief-forming method only if the thinker has an independently justified defense of the method. Defense InternalismM is in some ways weaker than Extreme Method Internalism. For instance, possessing a defense doesn’t require a thinker to hold a backing belief—an explanation of how it could be acceptable to employ the method would suffice. Nevertheless, Defense InternalismM is also a very strong form of method internalism.

9.4.2 Against Defense Internalism Despite the appeal of a defense-based conception of epistemic responsibility, Defense InternalismM ought to be rejected. One concern with Defense InternalismM stems from the fact that Defense InternalismB is motivated by the idea that believing a proposition is broadly analogous to making a claim or staking a position. However, employing a rule or a method is not a matter of explicitly staking a position. So the analogy does not obviously succeed. Even if Defense InternalismB is correct, we need not accept Defense InternalismM. This problem is not decisive. Employing a method is not a matter of explicitly staking a position. However, employing a method does entail that the thinker implicitly treats certain inputs as supporting certain outputs. So there is a case to be made that employing a method is bound up with the implicit staking of a claim. Why doesn’t this need defense, too? Given this line of thought, I don’t want to lean on the disanalogy too strongly. The main lesson here is just that the intuitions supporting the need for defenses are murkier for the case of rules and methods than they are for the case of beliefs. A second problem is that a defense-based conception of responsibility does not merely motivate a conditional. As before, the line of thought motivates an explanatory connection. In particular, it motivates the following principle: A thinker is justified in employing a belief-forming method only if and (in part) by virtue of the fact that the thinker has an independently justified defense of the method. ³² Apologies to Stan Lee. ³³ Alternatively, one might claim that children and animals do not count as justified—at least not in the same sense that we do.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



As before, this yields an explanatory regress. Essentially the same two explanatory problems emerge for this constraint as for Extreme Method Internalism. There is one small difference—in some cases, the thinker’s justification for employing the method is not completely explained by the fact that the thinker has a defense of the method. More may be required for the thinker to be justified. But this is not a relevant difference. The two explanatory problems retain their bite. I take it that these problems are decisive. They show that there is something wrong with a defense-based conception of epistemic responsibility. This problem undercuts the motivation for Defense InternalismM. There is another serious problem with the view. For some justified methods—for instance, for the basic methods governing our moral and modal thought—we don’t possess any persuasive defenses. For other justified methods—for instance, for Modus Ponens and Inference to the Best Explanation—the only apparent candidate defenses are circular. For example, consider the following argument for the validity of Modus Ponens: Suppose both p and if p then q. So p. So if p then q. So q. So if both p and if p then q, then q. So if both p is true and if p then q is true, then q is true. Therefore, Modus Ponens is valid. This argument relies on an instance of Modus Ponens.³⁴ Similarly, it is natural to try to defend Inference to the Best Explanation as follows: Inference to the Best Explanation has worked well for us in the past. The best explanation of this is that Inference to the Best Explanation is more-or-less reliable. Therefore, it is acceptable to employ this rule of inference. This defense relies on an instance of Inference to the Best Explanation. It, too, is circular. In response to this circularity worry, one might claim that the circularity here is not vicious. The arguments are rule-circular and not premise-circular. And it may be thought that at least some rule-circular arguments are acceptable.³⁵ In some contexts, it is appropriate to appeal to this distinction. For instance, it may be acceptable to make use of Modus Ponens in justifying the belief that Modus Ponens is valid. The trouble is that we aren’t discussing the justification of a belief. These arguments were raised as defenses of rules. It is grossly circular to defend a rule by appealing to the very rule in question. Such a defense should not be satisfying to a rational thinker. Why not take another tack? Why not just say that we find all of these rules intuitive (or obvious or self-evident or . . . )? The idea is that we have the intuition that Modus ³⁴ Of course, we could present an argument that makes use of Reasoning by Cases or some deductive rule other than Modus Ponens. But that would just be to move the bump in the rug. ³⁵ See Boghossian (2000).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

Ponens, for instance, is reliable (or otherwise has a positive epistemic status). The belief-forming methods that govern our reasoning with intuitions permit us to conclude that Modus Ponens is reliable. This provides a defense of Modus Ponens. This suggestion relies on endorsing several contentious theses about the existence and nature of intuitions. Even putting such worries aside, it doesn’t seem to be very much of a defense to say that Modus Ponens is intuitively reliable. More importantly, we can ask the same question about our methods for reasoning with intuitions. What is our defense of these methods? On pain of vicious circularity, the defenses of the methods that govern reasoning with intuitions cannot themselves depend on reasoning with intuitions. Very generally, we cannot simultaneously defend all of our beliefforming methods, at least not in a non-trivial and non-circular way. This point is closely connected to the final problem I’d like to raise for Defense Internalism. Namely, Defense InternalismM leads to an infinite regress. This is because a defense of a method is justified only if the methods employed in the defense are themselves justified. This is a familiar sort of problem. I think that it is genuinely a problem. The trouble stems from the independence requirement on defenses. Given this requirement, to possess a defense of a method requires having infinitely many other defenses. Given our finitude, this is an impossible demand to satisfy. The reason I raise this problem here is that there is a putative way out of this problem that merits discussion. In particular, the suggestion is that the infinite regress can be blocked if Defense Internalism is restricted to rules of inference, rather than belief-forming methods more generally. This may be part of what motivates some philosophers to focus on ‘inferential internalism’ rather than internalism about both inferential and non-inferential reasoning.³⁶ The trouble with this suggestion is that there is no intuitive motivation for the restriction to inference rules. The defense-based conception of responsibility motivates a very general constraint. There is no obvious reason to think, for example, that inference rules require defenses whereas perceptual methods do not. Given all of these problems, I conclude that we ought not to adopt Defense InternalismM. Why, then, do we find the defense-based conception of justification so appealing? Why does Defense Internalism have such an intuitive pull? Again, there is plausible diagnosis that can be offered: As rational thinkers, we have several general rational obligations. For instance, we are rationally obligated to try to explain interesting phenomena and to evaluate our own patterns of thinking.³⁷ It follows from these general obligations that we ideally ought to try to defend all of our beliefs and beliefforming methods. This is one of our rational ideals. However, this is not a constraint on justification. Although we may sometimes confuse the two, we do not need to

³⁶ For discussions of inferential internalism see Boghossian (2003), Fumerton (2006: 101), and Leite (2008). I should note that inferential internalist theses are typically formulated not as theses about which rules of inference are justified but as theses about which arguments transmit justification from premises to conclusion. ³⁷ See Enoch and Schechter (2008) for discussion.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



possess a defense in order to be justified. And that is a very good thing since, as we have seen, the rational ideal is unsatisfiable.³⁸

9.4.3 Entitlement Before I turn to the third motivation for rule internalism, let’s take stock. If what I’ve argued is correct, Extreme Method Internalism and Defense InternalismM are false. So we must be justified in employing certain belief-forming methods as basic but not in virtue of having justified backing beliefs or defenses of the method. In other words, we are epistemically entitled to employ those belief-forming methods. Plausible examples of methods we are entitled to employ include Modus Ponens, Inference to the Best Explanation, and our basic perceptual methods. This raises a question: In virtue of what are we entitled to employ certain beliefforming methods and not other? I’ll have something to say about this question below. But, first, let me turn to the third and final motivation for rule internalism.

9.5 The Argument from Skeptical Scenarios The third motivation for rule internalism is based on our intuitive reactions to skeptical scenarios. Imagine someone who is just like you, with exactly the same experiences that you are currently having and with the same beliefs about the world, but who is being radically misled by an evil demon. Alternatively, imagine that your counterpart is not being misled by an evil demon, but rather has had her brain surreptitiously removed and placed in a vat, where it is being fed experiences by a complex apparatus controlled by an evil neuroscientist. Or, alternatively, imagine that your counterpart is stuck in a Matrix-like scenario. We have strong intuitions about each of these scenarios. We find it intuitive that our counterparts are just as justified as we are in their beliefs about the world. They are just as justified as we are in the belief-forming methods they employ. Moreover, they seem justified in exactly the same way and for exactly the same reasons that we are. These intuitive reactions support a supervenience claim. They make it plausible that justification supervenes on what is in common between us and our counterparts. This is another internalist thesis.³⁹ (We’ll come back to the question of how exactly to formulate the thesis.) There is reason, however, to be wary of this line of thought. In particular, the line of thought may be the product of confusing having a justified belief with having an excusable belief. Our counterparts in the skeptical scenarios are being manipulated by an outside malevolent (or, perhaps even, benevolent) being. They are not, in some sense, ‘to blame’ for their beliefs. This feature of the cases may play havoc with our intuitive judgments about justification. To address this worry, we can switch to another class of skeptical scenarios. Imagine someone who is just like you, with the very same beliefs and who—from the inside—seems to be having the same experiences, but who is not genuinely ³⁸ See Christensen (2007) for discussion of unsatisfiable rational ideals. ³⁹ This line of thought is closely related to the ‘New Evil Demon’ argument of Lehrer and Cohen (1983) and Cohen (1984).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

having the same experiences, because she is caught in a dream. Alternatively, imagine that your counterpart is not caught in a dream, but is having hallucinations with exactly the same character as your experiences. In each of these cases, the ‘problem’ (as it were) is not that some outside being is manipulating your counterpart’s mental states. Rather, the problem is internal to your counterpart’s mind. Scenarios such as these can be used to test whether we are confusing justification and excusability. It will be helpful to have a single example in focus. For the sake of concreteness, let me focus on the following case. Wishful Hallucination. Suppose that you justifiably believe that p on the basis of visual perception. Suppose that you have a counterpart that is exactly like you and is in exactly the same conditions as you are, with one exception. Namely, your counterpart believes that p on the basis of having a visual hallucination of its being the case that p. The causal explanation of the hallucination is (in part) that your counterpart subconsciously wishes it to be the case that p.⁴⁰ Interestingly, I find that I have equivocal intuitions about this case. On one way of thinking about the case—the ‘first-personal’ way—my counterpart seems to be just as justified as I am. On a different way of thinking about the case—the ‘third-personal’ way—my counterpart seems not to be as justified as I am. Perhaps I have idiosyncratic reactions to the case. But suspect that this is not so. Let me try to convince you. Think about the case first-personally: Project yourself inside the head of your counterpart. Seen from the inside, the hallucination is indistinguishable from a genuine perception. Now ask yourself what you should believe given your (apparent) experiences. The answer is clear: You should believe that p. That’s what you are justified in believing. Now consider the case third-personally—from the outside. Focus on the idea that your counterpart believes that p on the basis of wishful thinking. This is not an ordinary case of wishful thinking—in the case, the wish directly causes a hallucination rather than a belief. Nevertheless, it is a case of wishful thinking. It seems clear that one cannot be justified holding a belief that was formed in such a dubious way. How could this count as a good case of believing? Not only are my reactions equivocal, but I find that I can make them switch back and forth in a fairly systematic way. If I think about the case from the inside, my counterpart seems justified. If I think about the case from the outside, my counterpart seems unjustified. I also find that I have similar reactions to other dream, hallucination, and wishful thinking cases.⁴¹ I’m not entirely sure what to make of this phenomenon. Perhaps there are two epistemic notions in play—justification and something else. If so, the difficulty is in

⁴⁰ The wishful hallucination scenario is a case of cognitive penetration, as discussed in Siegel (2012). The case may be a useful test case for Evidentialism, as presented by Feldman and Conee (1985). The subject is, in effect, providing himself with evidence. That seems epistemically dubious. ⁴¹ It is worth thinking about evil demon, brain in vat, and Matrix cases where it is somehow the subject’s fault that she is being misled—for instance, the subject volunteered to be put in the vat and then was caused to forget this. These cases also seem to elicit equivocal intuitions.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



determining which of our two reactions genuinely reflects something about justification and not a different epistemic status. In any event, let’s proceed as if the first-personal way of thinking genuinely reflects our intuitions about justification. On this way of thinking, our counterparts are just as justified as we are in their beliefs and in their employment of belief-forming methods. This supports a supervenience claim. The claim is that justification supervenes on a subset of mental states—those mental states shared between a thinker and her counterparts in the relevant skeptical scenarios. As an approximation, we can take the supervenience base to consist of the conscious mental states that make up a thinker’s ‘point of view.’ Consider the case of a thinker with the very same point of view as you have, but with some other differences in their mental states. If you project yourself into your counterpart’s head, it will be intuitive that your counterpart is just as justified as you are.⁴² This line of thought motivates the following two supervenience claims: Mental InternalismB: Whether a thinker is justified in believing a proposition supervenes on the mental states that make up the thinker’s point of view. Mental InternalismM: Whether a thinker is justified in employing a belief-forming method supervenes on the mental states that make up the thinker’s point of view. These theses can be thought of as a generalization of Evidentialism, at least so long as a thinker’s evidence is completely included in his point of view.⁴³ Notice that Mental InternalismM is a very different sort of internalist constraint from the two we have so far considered. There is no appeal to backing beliefs, defenses, or anything similar. Notice, too, that this constraint does not suggest a positive account of the justification of belief-forming methods. If the motivation works, it provides something like a proof of the possibility of an internalist account. It does not directly provide a positive account.⁴⁴ I’d like to have a compelling objection to make against Mental InternalismM. In truth, I don’t have one. But I do think there are two important reasons to resist endorsing it. First, as we’ve seen, our intuitive reactions to skeptical scenarios are equivocal. They do not clearly motivate Mental InternalismM. Second, as we will see in the next section, there is reason to think that no plausible account of justification satisfies the constraint.

9.6 The Way Forward? Given the failure of the first two forms of internalism, what is needed is an account of how we can be justified in employing a belief-forming method as basic in our thought but not in virtue of having a justified backing belief or a defense of the method. In other words, we need an account of how thinkers can be epistemically entitled to ⁴² This supervenience basis may have to be slightly expanded. For instance, the basis may also have to include the thinker’s point of view over an interval of time (as opposed to an instant). In imagining a skeptical scenario in a first-personal way, we imagine being that person for a short while. It may also have to be expanded to include some of the thinker’s dispositions to form beliefs, and the like. ⁴³ See Feldman and Conee (1985). ⁴⁴ I owe this observation to Christopher Hill.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

employ belief-forming methods. Such an account must be appropriately normative— in particular, it must explain how employing such a method can be an epistemically responsible thing to do. A natural suggestion at this point would be to endorse something like process reliabilism restricted to basic belief-forming methods. The suggestion would be that a thinker is entitled to employ a belief-forming method as basic in thought just in case (and by virtue of the fact that) the method is reliable. However, such a view faces many familiar problems. The most central is simply that reliability and epistemic responsibility are very different statuses. The mere fact that a method tends to get at the truth does not make it epistemically responsible to employ the method. Just consider the rule that tells us to believe a particular highly complex mathematical truth on the basis of arbitrary inputs. This rule is completely reliable—its output is a necessary truth. Yet, a thinker would not be epistemically responsible in employing this method as basic in her thought.⁴⁵ I’d like to be able to say here that my own account of epistemic entitlement to belief-forming methods is the only view that has the resources to address this issue. But I cannot make that claim. There are a few accounts in the literature that make use of resources that show promise for the project of providing an explanation of epistemic entitlement to belief-forming methods. In the remainder of this chapter, I’d like to briefly sketch three such accounts. I cannot compare these accounts here. I believe that my own account is the best of the three, but arguing for this claim would require delving into the details of the accounts. In what follows, I would like to stay at a very high level of generality.

9.6.1 Proper function The first account is due to Alvin Plantinga. He writes: [A] belief has warrant for me only if (1) it has been produced in me by cognitive faculties that are working properly (functioning as they ought to, subject to no cognitive dysfunction) in a cognitive environment that is appropriate for my kinds of cognitive faculties, (2) the segment of the design plan governing the production of that belief is aimed at the production of true beliefs, and (3) there is a high statistical probability that a belief produced under those conditions will be true.⁴⁶

Plantinga is here interested in the nature of warrant—that which when added to true belief yields knowledge. My discussion has been focused on a different epistemic status—namely, justification. Plantinga’s view is also heavily reliabilist. It inherits all of the intuitive problems of process reliabilism. But I think that the governing idea can be freed from those problems. Here is what I take to be the core insight: There is some normativity that emerges from the fact that you are ‘firing on all cylinders’. More precisely, there is a kind of normative status that attaches to a cognitive faculty when it has an appropriate aim, is well-designed for its aim, and is properly functioning. ⁴⁵ See Schechter (2019) for a related case. ⁴⁶ Plantinga (1993: 46). Also see Bergmann (2006) for a related view.

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi

 



There are certainly some obscurities in this formulation. What is an appropriate aim? What is it for a faculty to have an aim? What is it to be well-designed for an aim? But at a high-level of generality, there is something appealing about this thought. Modifying this account for our purposes, we arrive at the following view: A thinker is pro tanto epistemically entitled to employ a belief-forming method as basic in thought if (and by virtue of the fact that) the method is part of a cognitive mechanism that (1) has an appropriately epistemic aim—e.g., the truth; (2) is welldesigned for that aim; and (3) is functioning properly.⁴⁷

9.6.2 Skillful performance The second account is due to Sosa. According to Sosa, a thinker has knowledge that p only if (1) the thinker’s belief is accurate; (2) the thinker’s belief is adroit—it manifests an epistemic competence; and (3) the thinker’s belief is apt—it is accurate because it is adroit.⁴⁸ Sosa’s thesis does not concern justification but knowledge. In particular, it concerns the sort of knowledge we can share with non-human animals (as opposed to reflective knowledge, which must satisfy further constraints). But this view makes use of resources that can again be applied to the problem that concerns us here. Like Plantinga’s view, Sosa’s view is heavily reliabilist. Sosa claims that manifesting an epistemic competence is a matter of reliably getting at the truth. But we abstract his view away from this commitment. Sosa’s core insight is that there is a sort of normative status that applies to things that are done skillfully whether or not they are done thoughtfully. There is a kind of normative status that attaches to a performance that is skillful or adroit. To repeat one of Sosa’s motivating examples: Consider the case of a skilled archer who shoots the arrow just so—that is, in a skillful manner. There is value in the archer’s shooting an arrow just so even if the archer has no conscious awareness of what he is doing or how he is able to do it. Sosa’s motivation for making use of the notion of a skillful performance is to answer the value problem—the problem of explaining why knowledge is more valuable than true belief. Sosa’s answer is that there is value in doing something skillfully in addition to the value of doing it successfully. My suggestion here is that

⁴⁷ Plantinga claims that the only appropriate epistemic aim is the truth. I do not think that this is so. He also presents a conception of function that requires that the function of a cognitive mechanism depends on it being literally designed for its function. I think this is mistaken as well. There are several other attractive conceptions of function in the literature. For example, on Wright’s (1973) view, a function of a mechanism is an effect of the mechanism that explains why it is there. For instance, a function of the heart is to pump blood, since this effect is what the heart was evolutionarily selected for. See Millikan (1984) for a related view of functions. On Cummins’s (1975) view, the function of a mechanism is an effect of the mechanism that enables the system of which it is a part to have some capability. For instance, on this conception, we can say that a function of a certain part of the brain is to enable one to do advanced mathematics, even if this played no role in its evolutionary history. Graham (2012, 2014) presents an account of epistemic entitlement that makes use of Wright’s conception of functions. For reasons to do with swamp people, I would be more tempted to try to make use of something like the Cummins conception. Interestingly, the resulting view would resemble Sosa’s skillful performance-based view discussed immediately below. ⁴⁸ See Sosa (2007).

OUP CORRECTED PROOF – FINAL, 21/1/2020, SPi



    

we might take the same basic idea and put it to a different use—that of explaining epistemic responsibility. Modifying his view for this purpose, we arrive at the following view: A thinker is pro tanto epistemically entitled to employ a belief-forming method as basic in thought if (and by virtue of the fact that) beliefs formed on the basis of employing the method are adroit—they skillfully manifest an epistemic competence.

9.6.3 Basic rational obligations Finally, there is my own view.⁴⁹ The guiding thought of this view is that there is a kind of epistemic status that attaches to a belief-forming method that is indispensable to one of the central projects of rationality. More precisely, the view has two main theses: (i) There are certain cognitive projects that rational thinkers ought to engage in irrespective of their goals and desires. These include the projects of explaining the world around them, deliberating over what to do, planning for the future, and evaluating their own patterns of thinking. (ii) A thinker is pro tanto epistemically entitled to employ a belief-forming method as basic in thought if (and by virtue of the fact that) employing the method is indispensable for successfully engaging in one of these rationally required projects. We have certain basic rational obligations—for instance, to explain, to deliberate, to plan, and to self-evaluate. These are the central projects of rationality. An agent who isn’t engaging in these projects is not doing what she ought to do. This is where epistemic normativity bottoms out. We have these rational obligations whether or not we are aware of them. Beliefforming methods that are sufficiently closely connected to these obligations gain a similar status—we are epistemically entitled to employ them even without a defense or backing belief.

9.6.4 Mental Internalism, again It is beyond the scope of this chapter to defend any of these views or to adjudicate between them. What I want to emphasize here is that these three views make use of resources that just might be able to help explain how a thinker can be epistemically responsible in employing a belief-forming method as basic in her thought despite lacking a defense or backing belief of the method. Functioning properly, skillfully manifesting a competence, or successfully engaging in one of the central projects of rationality are each properties that seem potentially able to explain normative

⁴⁹ See Schechter and Enoch (2006); En