Self-Reflection for the Opaque Mind: An Essay in Neo-Sellarsian Philosophy 2016033084, 9781138668829, 9781315618449

Table of contents :
Cover
Title
Copyright
Dedication
Contents
Preface
Part I Preliminaries
Preamble: Is Philosophy Anti-Scientific?
1 Introduction: How Is Rational Self-Reflection Possible?
2 The Empirical Case against Infallibility
Part II Knowledge of Thought
3 Infallibility in Knowing What One Thinks
4 Objection 1: It’s Apriori that Water Exists
5 Objection 2: Thought Switching
6 Content Externalism Does Not Imply Wayward Reflection
Part III Knowledge of Judging
7 Infallibility in Knowing What One Judges
8 Infallibility in Knowing What One Expresses
9 Objection 1: It’s Apriori that the Mental Exists
10 Objection 2: Attitude Switching
11 Attitude Confabulation Does Not Imply Wayward Reflection
Part IV Denouement
12 Conclusion: How Rational Self-Reflection Is Possible
Index

Citation preview

Self-Reflection for the Opaque Mind

Philosophical reflection on self-knowledge has recently bifurcated into two tendencies: on the one hand, there is increasingly strenuous effort from the armchair to resolve various a priori puzzles concerning the very possibility of self-knowledge, while on the other hand, there is an increasingly empirically informed effort to debunk our pre-­ scientific pretensions to self-knowledge. Parent’s manuscript provides a scholarly, detailed, and ultimately successful effort to reconcile these two tendencies, while also showing why the topic of self-knowledge is a topic of urgent practical importance. This is an important book. —Ram Neta, University of North Carolina at Chapel Hill, USA

This volume attempts to solve a grave problem about critical self-reflection. The worry is that we critical thinkers are all in “epistemic bad faith” in light of what psychology tells us. After all, the research shows not merely that we are bad at detecting “ego-threatening” thoughts à la Freud. It also indicates that we are ignorant of even our ordinary thoughts—e.g., reasons for our moral judgments of others (Haidt 2001), and even mundane reasons for buying one pair of stockings over another (Nisbett & Wilson 1977)! However, reflection on one’s thoughts requires knowing what those thoughts are in the first place. So if ignorance is the norm, why attempt self-reflection? The activity would just display naivety about psychology. Yet while respecting all the data, this book argues that, remarkably, we are sometimes infallible in our self-discerning judgments. Even so, infallibility does not imply indubitability, and there is no Cartesian ambition to provide a “foundation” for empirical knowledge. The point is rather to explain how self-reflection as a rational activity is possible. T. Parent is Visiting Assistant Professor in the Philosophy Department at Virginia Polytechnic and State University. Primarily, he works on the philosophy of mind, epistemology, and meta/ontology. His publications on such topics have appeared in Philosophical Studies, the Journal of Philosophy, and the Stanford Encyclopedia of Philosophy, among others. He lives with his wife in Blacksburg, Virginia.

Routledge Studies in Contemporary Philosophy For a full list of titles in this series, please visit www.routledge.com

76 Embodied Emotions A Naturalist Approach to a Normative Phenomenon Rebekka Hufendiek 77 Normativity and Naturalism in the Philosophy of the Social Sciences Edited by Mark Risjord 78 The Concept of Violence Mark Vorobej 79 A Social Theory of Freedom Mariam Thalos  80 The Cognitive Basis of Aesthetics Cassirer, Crowther, and the Future Elena Fell and Ioanna Kopsiafti 81 Interactive Justice A Proceduralist Approach to Value Conflict in Politics By Emanuela Ceva 82 The Epistemological Skyhook Determinism, Naturalism, and Self-Defeat Jim Slagle 83 Time and the Philosophy of Action Edited by Roman Altshuler and Michael J. Sigrist 84 McTaggart’s Paradox R. D. Ingthorsson 85 Perspectives on Ignorance from Moral and Social Philosophy Edited by Rik Peels 86 Self-Reflection for the Opaque Mind An Essay in Neo-Sellarsian Philosophy T. Parent

Self-Reflection for the Opaque Mind An Essay in Neo-Sellarsian Philosophy T. Parent

First published 2017 by Routledge 711 Third Avenue, New York, NY 10017 and by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Routledge is an imprint of the Taylor & Francis Group, an informa business © 2017 Taylor & Francis The right of T. Parent to be identified as author of this work has been asserted by him in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging in Publication Data Names: Parent, T., author. Title: Self-reflection for the opaque mind : an essay in neo-Sellarsian philosophy / by T. Parent. Description: 1 [edition]. | New York : Routledge, 2016. | Series: Routledge studies in contemporary philosophy ; 86 | Includes bibliographical references and index. Identifiers: LCCN 2016033084 | ISBN 9781138668829 (hardback : alk. paper) Subjects: LCSH: Self (Philosophy) | Self-perception. | Sellars, Wilfrid. | Philosophy of mind. Classification: LCC BD438.5 .P37 2016 | DDC 155.2—dc23 LC record available at https://lccn.loc.gov/2016033084 ISBN: 978-1-138-66882-9 (hbk) ISBN: 978-1-315-61844-9 (ebk) Typeset in Sabon by Apex CoVantage, LLC

To my mother and father, those great Catholic educators

Contents

Preface

ix

PART I

Preliminaries Preamble: Is Philosophy Anti-Scientific?

1 3

1 Introduction: How Is Rational Self-Reflection Possible?

35

2 The Empirical Case against Infallibility

55

PART II

Knowledge of Thought

81

3 Infallibility in Knowing What One Thinks

83

4 Objection 1: It’s Apriori that Water Exists

97

5 Objection 2: Thought Switching

111

6 Content Externalism Does Not Imply Wayward Reflection 129 PART III

Knowledge of Judging

153

7 Infallibility in Knowing What One Judges

155

8 Infallibility in Knowing What One Expresses

180

  9 Objection 1: It’s Apriori that the Mental Exists

223

10 Objection 2: Attitude Switching

242

11 Attitude Confabulation Does Not Imply Wayward Reflection

262

viii Contents PART IV

Denouement

281

12 Conclusion: How Rational Self-Reflection Is Possible

283

Index

289

Preface

Why a book on reflection? One reason is: Timing. These days in philosophy, there is a frenzy of interest in metaphilosophical issues, and philosophical method looms large in this. Reflection, moreover, is the paradigmatic philosophical method. (The book was almost entitled “Philosophical Method for the Opaque Mind.”) Another reason: A gap. The book presents an important problem about reflection which has been almost completely ignored in the literature, and the book is currently the sole attempt to resolve it.1 Third: Broad appeal. The issue about “reflection” is really a problem about reflective critical thinking more broadly. And the problem does not just consist in some esoteric philosophical riddle. Rather, it arises straightforwardly from concrete empirical results in psychology. I refer to evidence suggesting that we are largely ignorant of our own minds. The problem is that if we are largely self-ignorant, then self-reflection (i.e., critical thinking on one’s own thoughts) makes no rational sense. One would be ignorant of the thoughts one is supposed to reflect on. The book thus was written to defend the rationality of self-reflection despite the evidence for widespread self-ignorance. In this, the book also purports to show by example how philosophical work on a prima facie empirical topic can be not only appropriate but edifying as well. (Cf. the Preamble chapter.) Throughout, there is a preoccupation with philosophical method, not only in theory, but also in the book’s own practice. The book attempts to meet certain standards of “rigor,” which makes the work more difficult in some respects. But there is a trade-off I think that makes this necessary. Allow me to explain.

1 In fact, one version of the problem has been discussed extensively; I am referring to the literature spawned by Boghossian’s “inference argument” against content externalism. Yet the more daunting version—examined in chapter 11—is gestured at only in the work of Kornblith. (It is the more daunting version since a full solution, I think, requires a solution to Boghossian’s version.) Regardless, unlike myself, Kornblith largely capitulates to the difficulty. (For pertinent works by Boghossian and Kornblith, see the references to chapter 1.)

x Preface What does the theory of mind (hereafter, “TM”) hope to achieve? Uncontroversially, it is to advance our understanding of the mind. For my part, this requires developing TM into an increasingly scientific theory—but what does that mean? The immediate thought is to align TM more and more with empirical data. However, empirical data can be brought to bear on TM only if it is already scientific in certain ways. Most obviously, TM must already have definite, observable consequences that can be tested against the data. But to a significant extent, TM has not yet reached that point. The question, then, is how do we get from here to there? A mature TM, restricted to human minds, can be seen as a branch of human anatomy. (This is not to endorse reductionism; it is only a point about the relation between disciplines.) It’s not just that some brain anatomy is necessary for understanding our minds. More basically, it is that our mental state-tokens are token-states of the body (at least in part). And so, a complete inventory of bodily states, per human anatomy, will include an inventory of mental states (or the bodily components thereof). However, a salient feature of anatomy is its copious taxonomy. Even a relatively simple body-part, such as the forearm, has been mapped out in abundant detail. For starters, there are the two bones, the radius and the ulna, and the interosseous membrane connecting them. There are also numerous muscles, including the brachioradialis, the pronators, and ­supinators—plus the veins permeating the area, the main ones being the cephalic, median, antebrachial, and basilic veins. And this is not yet to mention the nerves. Already, we have a good amount of technical terms. But here, it is obvious that precision in language enables a more refined understanding; it is what permits (e.g.) a surgeon to focus an operation on “the annular ligament of the ulna.” These days, however, some theorists of mind proceed as if precise language is optional. Yet how much more subtle is the anatomy of the mental! If the aim is the scientific development of TM, then, we must not shy from precise terminology. Again, exact language is often what allows experts to make distinctions which novices cannot. I apologize, then, if my prose is not as readable as some TM-writers. Yet from where I sit, it is no longer excusable to write as if casual uses of ‘representation’, ‘content’, or ‘belief’ are sufficiently clear for TM.2 Part of what makes a theory empirically testable is that its predictions have a kind of determinacy to what they say. This helps guarantee that “accommodating the data” is not something the theory can fudge by exploiting the vagueness of its language. That means imposing precision. Precise terminology may make the work of TM more difficult, but it makes it better. (Yet to aid the reader, the definitions of key terms are repeated where relevant; this also helps make each chapter readable independent of the others.)

2 It is permissible to start by using such terms in the ordinary way, as is true in the present work. But my point is that the ordinary use is not where TM should end.

Preface  xi Having said all that, a wealth of specialized terminology in TM does not yet make sense. Thus the book is not as lexically adventurous as it might have been, and it is not yet primed for serious empirical testing. Ultimately the language is more meticulous than much philosophy of mind, but I hope to have explained why this heads in the right direction. I have incurred a number of debts to individuals and institutions in the course of writing. I am most grateful to my family, especially to BZ and to my parents, for their uplifting presence and therefore positive influence on the writing process. I am also particularly grateful to two colleagues: Ram Neta and Joseph C. Pitt. Although they may not realize it, their enthusiasm was tremendously encouraging at a professionally difficult time. I also must give special recognition to Dorit Bar-On, William G. Lycan, and (the late) Jay F. Rosenberg. Their philosophical influence on this work is unmistakable, and I thank them for all the erudition over the years. Also, special appreciation goes to Margo Irvin and Andrew Weckenmann at Routledge for their support of the project. My gratitude is sincerely extended to other individuals for feedback and/ or discussion on issues in this book, including Louise Antony, Jody Azzouni, Jim Baillie, Derek Ball, Gary Bartlett, Jason Bowers, Tyler Burge, Yuri Cath, David Chalmers, Matthew Chrisman, Larry Coulter, Chris Daly, Felipe De Brigard, Justin Fisher, Tamra Frei, Matthew Fulkerson, Brie Gertler, Ephriam Glick, Sandy Goldberg, Glen Hoffmann, Thomas Hofweber, Bryce Huebner, Ben Jantzen, Justin Jeffrey, Jim Klagge, David Landy, Douglas Lind, Dan Linford, Joe Levine, Doug Long, Eric Mandelbaum, Norah Martin, Aidan McGlynn, Tristram McPherson, Ruth Millikan, Greg Novack, Nico Orlandi, Mark Phelan, John Post, Jesse Prinz, Amber Ross, Dan Ryder, Sarah Sawyer, Keith Simmons, Kelly Trogdon, Meg Wallace, and Brian Weatherson. (I am certainly forgetting some names; this has been ongoing for several years. My deepest apologies.) The contents of the book have also benefited from audiences at meetings of the American Philosophical Association (2010 Central Division, 2004 Eastern Division), the University of Edinburgh (July 2014 conference on Mental Fictionalism), the University of Glasgow (June 2013 conference on the philosophy of Tyler Burge), Oregon State University (2012 Northwest philosophy conference), the University of Memphis (2004 Midsouth philosophy conference), the University of North Carolina at Chapel Hill (talk in Oct. 2008), Vanderbilt University (talk in Oct. 2009), and Virginia Tech (April 2015 workshop in the philosophy of mind). Also, my thanks to the publishers below for the use of material previously in print: •

Chapter 2 is a revised/expanded version of “The Empirical Case against Infallibilism,” Review of Philosophy and Psychology, 7.1 (2016): 223–242.

xii Preface •

• • •

Chapter 3 is mostly a revised/expanded version of “Infallibilism about Self-Knowledge,” Philosophical Studies, 133.3 (2007): 411–424. The exception is section 3.4, which is a revised/expanded version of “Infallibility Naturalized: Reply to Hoffmann,” Dialectica, 67.3 (2013): 353–358. Chapter 4 is a revised/expanded version of “Self-knowledge and Externalism about Empty Concepts,” Analytic Philosophy, 56.5 (2015): 158–168. Chapter 5 is a revised/expanded version of “Externalism and ‘Knowing What’ One Thinks,” Synthese, 192.5 (2015): 1337–1350. Chapter 9 is a revised/expanded version of “In the Mental Fiction, Mental Fictionalism is Fictitious,” The Monist, 96.4 (2013): 608–624.

The assertive mode of speech is needed for describing (and describing is what the book hopes to do). Yet asserting has the unfortunate feature of “representing the speaker as knowing,” as they say, even though what follows is highly tentative. But, rather than incessantly writing “it only seems to me that p,” I instead hereby cancel any implicature for knowing the propositions asserted.3 Hopefully I can rest easier. Finally, the reader should note that some technical sections of the book are marked with an asterisk ‘*’. These portions are intended mostly for specialists and can be skipped by the more casual reader without losing the main thread of the book. (Ditto for the footnotes and appendixes.) TSP Blacksburg, VA

3 Though per chapter 8, I do not cancel implicatures to knowing that I judge the propositions asserted. (That would create a formidable preface-paradox indeed.)

Part I

Preliminaries

Preamble Is Philosophy Anti-Scientific?

Initially, this book may seem to be philosophical excess of the worst kind. It can look like wishful thinking, a panicky response to legitimate scientific results about the limits of self-knowledge. For, despite ample evidence of these limits, the book argues that many beliefs about one’s own mind are not just reliable; they can be infallible in the way Descartes suggested. Such a view, if not anti-scientific, will at least seem quixotic. I shudder that this might be the impression left on the casual reader. Like many contemporaries, I think of myself as a naturalist: I not only reject ghosts and gods, but more generally accept that science is the highest authority on what mind-independent reality is like. (Yet this is not to say that science is perfect.) Thus, if science reveals that the mind is largely opaque to itself, one would expect a naturalist philosopher to let the matter lie. However, nowhere is any of the scientific data contested. Quite the contrary, the data puts an inviolable constraint on the project. Still, even if the data shows the mind is often self-ignorant, it does not follow that this is most often the case. It at least remains possible that in many important cases, we are attuned to our own minds. Naturally, to suggest the possibility is not to demonstrate its actuality. Yet this illustrates that some first-person authority may be possible, even given the science. (Though whether it is at all probable is another matter.)

0.1  A Neo-Sellarsian Approach Be that as it may, why exactly should this position be developed? (There is still the smell of wishful thinking.) The answer shall be developed in the next chapter. But my concern here is with a prior matter: Why is a philosopher suited to investigate the issue? Whether homo sapiens know their own minds may seem better left to empirical psychology, especially by the lights of a selfproclaimed naturalist. 0.1.1  Reconciling the Images I certainly welcome confederates from psychology. Nevertheless, some issues in the area are distinctly philosophical, in that they exist as conflicts between

4 Preliminaries what Sellars (1962) called the manifest image and the scientific image of the world. The leading question of the book is an example: How is rational self-reflection possible? From the manifest perspective, self-reflection stands as an important part of our intellectual and moral lives. Yet the activity presumes a reliable ability to identify one’s own beliefs, desires, etc. in order to evaluate them. And reliable access of this type appears non-existent in light of contemporary psychology.1 So is self-knowledge and self-reflection mere foolishness? Either way, the point is that we have now entered distinctly philosophical terrain. The question is: What should we do about this conflict between the manifest and scientific images? Should we try to resolve it? Or should we reject selfknowledge, and with it the practice of self-reflection? (Alternatively, should we just learn to live with the conflict?) Importantly, since the issue is a normative one, it seems appropriate to address it via (empirically informed) philosophy. A Sellarsian adopts a particular answer to the normative question— namely, that resolving the conflict is best.2 But this is contentious in two ways. First, it is doubtful whether the philosopher is qualified to discuss what exactly the data allows. Second, it may appear more scientifically serious just to reject what the manifest image says rather than prop it up somehow by massaging away the empirical difficulties. Regarding the first issue, a philosopher may well be unqualified to analyze the data for prototypical scientific purposes. Still, a Sellarsian is not necessarily out of line if she offers hypotheses to show how reconciliation is possible. Such hypothesis-building is certainly not the special province of philosophers. Yet if the Sellarsian recommends resolving conflicts between the images, she thereby incurs a burden to answer the “how possible?” question. She becomes dialectically obligated to do some amount of hypothesisbuilding, given that some of her peers answer the normative question differently. Whether her hypotheses are scientifically viable should be judged on a case-by-case basis. (And if they are not, that is reason to judge them philosophically lacking as well.) But even if they are de facto ignored by scientists, they can still serve a dialectical function vis-à-vis the normative question, a question which is indeed appropriate for philosophers. The book thus develops an increasingly nuanced hypothesis about selfknowledge, meant to vindicate the manifest value of reflection. But the second issue, again, is that it may be more scientifically respectable just to dismiss reflection as misguided. In fact, this is an instance of a very general 1 Also, self-knowledge is central to the manifest image since it is widely presupposed in ordinary conversation. We assume by default that a speaker knows what she is saying. (Without that, interpreting her is rather more difficult.) 2 Yet it is not as if the Sellarsian’s only job is resolving the conflicts. Often, her task is instead to show how other philosophers have simply glossed them over. For a fine example of such a Sellarsian at work, see Rosenberg (1999).

Preamble: Is Philosophy Anti-Scientific?  5 problem for Sellarsian metaphilosophy—namely, that the Sellarsian looks anti-science insofar as she advocates conserving the manifest image, despite its conflicts with the scientific image (hereafter, “MI” and “SI”). What follows is an attempt to respond to that worry. Even so, I do not endorse in an unqualified way Sellars’ metaphilosophy. Yet to fend off the anti-science objection, the main thesis shall be that engagement with the MI is simply non-optional. Further, if we are stuck with MI, we cannot be blamed for continuing to engage it, even if this is less than scientifically ideal. In which case, it is only understandable that we would try to reconcile the conflicts with SI. We could not be blamed for trying.3 0.1.2  Against Unconditional Unity Before elaborating, however, I want to register a few disagreements with Sellars—thus explicating the ‘neo-Sellarsian’ label. These concern three metaphilosophical ideas from Sellars (1962), which are often quoted approvingly by contemporary philosophers. The first is: (a) The aim of philosophy, abstractly formulated, is to understand how things in the broadest possible sense of the term hang together in the broadest possible sense of the term (p. 1). Dennett (2013), for instance, says, “That is the best definition of philosophy I have encountered” (p. 70). In the same paragraph, Sellars offers a related thought: (b) To achieve success in philosophy would be . . . to “know one’s way around” . . . in that reflective way which means that no intellectual holds are barred (1962, p. 1). But in addition to (a) and (b), there is also the main thesis of Sellars’ paper. Sellars regards problems of philosophy as problems, roughly, of: (c) how to reconcile the ultimate homogeneity of the manifest image with . . . the system of scientific objects (p. 36). Our interest is mainly in (c)—yet this section addresses (a) and (b), so to leave them behind later.

3 A suppressed premise here seems to be “rational ought” implies “psychological can.” However, I would not assume that this holds universally, though I think something like it holds in this instance.

6 Preliminaries On a natural reading, (a) is concerned to say that philosophy aims to unify, to give a systematic and cohesive theory of reality.4 Yet it seems one could equally say this of physics—and so pace Dennett, it is unclear how “definitional” of philosophy it is. But secondly, and much more importantly, it would neglect that philosophy is also in the business of de-unifying. Indeed, if we think of philosophy in the vein of Socrates, Cartesian skepticism, Humean skepticism, Schopenhauer, Quine, the later Wittgenstein, and feminist social criticism, philosophy seems more destructive than constructive. Here, “success” consists in identifying our pretensions, our illusions of thinking we “know our way around.” Even so, perhaps the aim of de-unifying is to ensure that things are ultimately unified in the right way. So perhaps philosophy is a unifying endeavor in the end. That is likely true in some cases. But other times, I think not. Often, pluralism seems like the most honest answer to, for example, the species problem in philosophy of biology. When we consider the many ways that actual biologists individuate species, attempts to “unify” this multitude can oversimplify matters considerably (cf. Kitcher 1989). Whether this pluralism about species is correct, however, is not really the point. Rather, it is just that pluralism per se is not contra-philosophical, simply because it surrenders the unifying goal. (There is a sense in which pluralism imposes a kind of unity [“E pluribus unum”], but that is the trivial sense in which any view “unifies.”) Such points bear on (b) as well. “Knowing your way around” is akin to knowing a map that unifies the disparate bits of reality, to facilitate easy travel across the philosophical terrain. But again, one central task of philosophy is also to show how “lost” we really are—in some cases, even irredeemably lost. Contemporary philosophers often have a distaste for such destructive or “skeptical” philosophy. Regardless, it is a strong current in the history of the discipline, and it oversimplifies the history if it is unified under the “unifying” goal.5

0.2  The Manifest Image and Quinean Scruples Perhaps Sellars can grant the points above, for his 1962 is not necessarily an attempt to give an exhaustive metaphilosophy. He may just be highlighting some particularly important features of some particularly salient

4 It seems (a)–(c) also introduce the idea that philosophical knowledge is a kind of knowinghow, rather than a knowing-that. I think there is something importantly right about this; however, I cannot pursue the matter here. 5 There is a passage akin to (b) in Sellars (1971/1975) which lacks the anti-pluralist overtone: “The ideal aim of philosophizing is to become reflectively at home in the full complexity of the multi-dimensional conceptual system in terms of which we suffer, think, and act” (p. 295, italics his). Yet this still seems to neglect that the ultimate point of some philosophy is to alienate us from what is familiar, rather than provide a sense of home.

Preamble: Is Philosophy Anti-Scientific?  7 philosophy. Regardless, a categorical metaphilosophy of “unity” is to be resisted. However, Sellars is quite right that “unifying” is one kind of philosophical endeavor, and (c) insightfully explicates this in terms of reconciling of the manifest and scientific images. But what exactly are MI and SI, and in what ways do they need reconciling? Sellars spends a good deal of time explaining what MI is, though as he admits, it remains sketchy. At least, MI is not the “original image”—it is not a pre-scientific vision of the world based in myth and religion. But nor is it thoroughly scientific; it instead seems to be a kind of “halfway house.” MI admits no gods or monsters, and is thus an empirical “refinement” of the original image (p. 7). Yet MI is importantly different from SI. For only MI posits things like intentions and agency along with standards of correctness for morality and for logical inference (pp. 16–17). Thus, Sellars is prompted to aim for reconciliation: [T]he task [is] of showing that categories pertaining to man [sic] as a person who finds himself confronted by standards (ethical, logical, etc.) which often conflict with his desires and impulses, and to which he may or may not conform, can be reconciled with the idea that man [sic] is what science says he is. (1962, p. 38) But the key question is: Why exactly should reconciliation be the goal? After all, there are at least two other options. One is pluralism about the images (not to be confused with more localized pluralisms about, e.g., species.) The image-pluralist accepts both images for their different advantages, without requiring consistency between them. Though in practice, she may want to side-step incompatibilities as much as possible. (My own view, in fact, is that MI and SI are occasionally irreconcilable, whence a kind of “theory dualism” becomes the best option.6) Sometimes, however, MI and SI seem reconcilable—and it may be apt, dialectically at least, for a philosopher to build hypotheses to reveal that. (But this is not to say that this is the only suitable task for philosophy, nor even its most characteristic task.7) Even so, my focus will be on defending Sellarsian reconciliation against an alternative response to MI–SI conflicts—one

6 See Parent (2015), where theory dualism is developed in response to (what is coincidentally) Sellars’ (1962) key example of an MI–SI conflict. This is roughly the problem of “qualia.” 7 After all, some philosophical problems are conceptual knots entirely within MI (e.g., conflicts of moral duties). And what Quine called “regimentation” into a rigorous idiom strikes me as a more characteristic philosophical task. (See section 1 of Parent ibid.) Yet I would second Haack’s (2005/2013) exhortation “to use whatever tools—formal, conceptual, linguistic, phenomenological, etc.—are appropriate to the task at hand [and to contend] with questions about the scope and limits of formalism” (p. 238–239).

8 Preliminaries where the conflicts prompt us to dispense with MI entirely, letting SI claim hegemony. This alternative can be associated with Quine (1948; 1960).8 For the Quinean, our ontological commitments should be fixed by our best, regimented scientific theory—specifically, by the range of its quantifier. The ontology of SI is thus the only one worth taking seriously. And this perspective is not to be taken lightly. After all, if one retains an attachment to MI, in spite of its conflicts with science, one risks being unscientific. Concurrently, any attempt to reconcile MI with SI can seem illegitimate. It would be like trying to reconcile the original image with SI, as some theologians are wont to do. Such things would differ only in degree and not in kind. Less dramatically, Sellarsian reconciliations can look unseemly akin to “domesticating metaphysics,” as criticized by Ladyman et al. (2007). Domesticating metaphysics uncritically assumes an atomist, mechanistic “folk physics”—and in so doing, it is patently unscientific. Now Sellars does not portray MI as adopting folk physics. But MI may be perturbing in an analogous way. For it assumes some substantive empirical hypothesis, without consideration of its scientific credentials. Yet in the case of MI, it is not folk physics that is assumed, but rather folk psychology and/or folk moral psychology (hereafter, “F(M)P”). MI assumes such things by portraying homo sapiens as having intentions to act, as being subject to norms, and so on. To be clear, F(M)P is not entirely at odds with science. Much of current psychology is within a folk psychological paradigm. But there are conflicts, as when F(M)P assumes that a person is authoritative in knowing her own mind. More obviously, F(M)P is committed to norms whereas SI (as Sellars portrays it) is not so committed. Even so, adopting MI need not be entirely unscientific. At one point, Sellars portrays MI as if it were empirically falsifiable, something that “might have to be rejected, in the last analysis, as false” (p. 14). But as things currently stand, the Sellarsian seems determined to integrate MI-ontology into SI, and to do so without antecedent scientific backing, and without plans for experimental confirmation. After all, it is not as if the Sellarsian has studied the viability of F(M)P from the point of view of cognitive science. Nor is she busy designing experiments to verify its posits. That can make F(M)P-ontology look like sheer bias—like a theologian putting God into SI without any serious scientific scrutiny.9 Based on such observations, Quineanism for many years seemed the more reasonable option. It seemed clear that ontology should be gleaned from our

8 Quine’s real views on ontology are much more complex than is usually appreciated. But I cannot pursue this here. 9 Unlike Sellars, the theologian denies that her paradigm may be overturned “in the last analysis.” So that is one disanalogy between them. Still, one wants some independent confirmation of F(M)P, beyond just that it helps unify MI and SI. Naturally, F(M)P has something like an empirical check: It must cohere with other scientific theories we accept. But usually, coherence with antecedent theory is not alone what we think of as empirical confirmation.

Preamble: Is Philosophy Anti-Scientific?  9 best science, and nowhere else. If F(M)P is thereby excluded, so be it. The background to such Quineanism, moreover, is a healthy respect for science, and a recognition that philosophers cannot muck about with SI without legitimate empirical reasons.

0.3  How Is an Inescapable Fiction Possible? Indeed, there is much to admire about such a stance, and in many cases, the Quinean ontological choices are the only respectable ones. Even so, I shall argue that MI and the framework of F(M)P end up being non-optional for creatures like us. That is so, even if F(M)P turns out to be a fiction.10 Contra the Quinean, we are thus unable to jettison MI entirely—any claim to do so is based on an illusion. Reconciling MI with SI, then, is the only alternative to jumping between incompatible ontologies, and reconciliation can thus be rationally motivated. But why think F(M)P is non-optional? There is the familiar, instrumentalist point that F(M)P is indispensable in praxis for predicting/explaining behavior. It provides user-friendly concepts that help us make sense of extremely complicated events, events that involve millions of neural firings in, for example, something as mundane as the family going out to dinner. However, F(M)P is non-optional for more than such practical purposes, or so I will argue. Further, there is a difference between being indispensable and being non-optional in the present context: ‘Non-optional’ has the sense of inescapable: If F(M)P is non-optional, we could not dispense with it even if we tried. That is so, no matter how inaccurate or non-utile it is. Hence, the “non-optional” issue is orthogonal to realism about F(M)P. To say that F(M)P is non-optional or inescapable is to say that it is inevitably used by creatures like us, fictitious though it may be. But: What does it even mean for a theory to be “inescapable?” Especially if F(M)P were known to be false, why might we be somehow forced into it? It is best to clarify this before going further, and I shall provide two illustrations. The first we can approach via an analogy. Suppose that while playing a game of football, Hetal asks, “How many points does our team have?” And suppose her teammate replies, “Actually, there are no such things as ‘points’, so neither team has any.” In refusing the question, the teammate violates a constitutive11 rule of the game—for a person must acknowledge points in order to play. This, moreover, would be contra his status as a participant. Understandably, then, Hetal may react negatively. That is so, even though she may know her teammate is right that, strictly speaking, there are no such things as “points.”12

10 For more on fictionalism about folk psychology, see chapter 9. 11 I borrow the distinction between “constitutive” vs. “regulative” rules from Searle (1965). 12 Points may exist as a kind of “social construct,” but talk of what exists strictly speaking is meant to exclude such things.

10 Preliminaries Analogously, Cog Sci Joe might ask, “What does the brain believe when it enters neurological state N?” If F(M)P were known to be false, one reply might be “there are no such things as ‘beliefs’, so the brain doesn’t believe anything.” But suppose we were antecedently committed to playing the “mentalistic language game.” Then, if we suddenly deny the existence of belief, we flout a constitutive rule of the game. So understandingly, Joe may react negatively—even if he knows that our denial is accurate, strictly speaking. The mentalistic language game is then akin to football, for both activities would feature a constitutive rule that invokes fictional entities: “points” or “beliefs,” as the case may be. Both would include a “mandatory fiction,” a pretense whose adoption is required for the activity. Now, as with football, one can quit playing the mentalistic language game—and one way to quit would be to forego the pretense. In refusing a pretense, moreover, a person can appear more rational for it. Yet the activity may still have value, and so the refusal may have a cost. If so, then even an eliminativist might have reason to commit to an activity where F(M)P is presupposed. Under such a commitment, the framework is then “non-optional” in a sense, regardless of its truth. In fact, later I shall argue we actually are committed to F(M)P in such a fashion.13 Though in Joe’s case, the relevant activity is cognitive science, and not everyone is committed to doing that. Regardless, if there is a commitment to the right sort of activity, F(M)P may be “non-optional” in a sense, even if it is just a fiction. Such things are rather tenuous—but at this stage, we are just addressing a “how possible?” question. And there is a second, more radical illustration of how F(M)P might end up being a “mandatory fiction.” Assume here that there is a neurological type denoted by ‘understanding’, whereby people “understand” chess or “understand” how an engine works. However, suppose we have a radically false F(M)P-account of it. (‘Understanding’ would then be like the term for lightning in ancient Greece: Even though the term refers successfully, speakers have a wildly false account of the referent.) Then, F(M)P might end up being an inescapable fiction in the following respect: Given our hardwiring, it is nomologically impossible for us to enter a state of ‘understanding’ toward ourselves, unless F(M)P is mobilized. To use the well-worn computer analogy, we may be so constituted that F(M)P is the “program” that the brain must use to understand itself. That may be so, even if F(M)P is largely a fiction.14

13 As will become clear, I have in mind the “activity of reason,” to use Korsgaard’s (2009) phrase. 14 I am reminded of a passage from, of all people, Nietzsche: “We have arranged for ourselves a world in which we can live—by positing bodies, lines, planes, causes and effects, motion and rest, form and content; without these articles of faith nobody could now endure life. But that does not prove them. Life is no argument. The conditions of life might include error” (The Gay Science §121).

Preamble: Is Philosophy Anti-Scientific?  11 This does not mean that understanding is achievable once we apply the false theory. It may be that whether we are familiar with F(M)P or not, “understanding” a cognitive system remains elusive. Yet things need not be entirely desperate. If F(M)P is jettisoned in a scientific psychology, then the science might be akin to Quantum Mechanics. QM has been extraordinarily successful as a predictive theory; nevertheless, some physicists say that to an extent, it is just not possible to understand QM on its standard interpretation. For QM tells us there is no fact of the matter concerning the location of an electron, on a given measurement of momentum. And that just boggles the mind. This limit on understanding, moreover, may owe to an adaptation of homo sapiens, where representing objects as determinately located is hardwired. But then, QM is just not translatable into our most primitive or intuitive conceptual scheme. It may then always feel unintelligible to a degree. That is so, even though QM is part of our best science. Analogously, if F(M)P were known to be false, there may be an inevitable sense that we are unintelligible to ourselves. That is so, even if neuroscience became a reliable predictor and explainer of our behavior. The sense of incomplete understanding, moreover, may owe to F(M)P-tokens being part of our most basic “machine language,” bestowed on us by evolution. So as with QM, we may then always be unintelligible to ourselves to some degree. Though also like QM, some degree of understanding would remain possible. At any rate, if F(M)P is needed to “understand” ourselves, then the intentional stance is not just instrumental for simplifying behaviors into predictable patterns. Viewing ourselves from this stance would also be nomologically necessary for a complete “understanding” of ourselves. That would be so, even if the intentional stance is mere story-telling. And if it is mere story-telling, F(M)P in that case may be an “inescapable fiction” for some psychological purposes.

0.4 Transitional But the foregoing is just illustrating possibilities. Why think that F(M)P actually is inescapable? There are at least three arguments here. The first begins by observing that F(M)P is part of MI, which is defined by how reality manifests or appears. Yet even science requires observation statements that declare how things appear to be, and “appearances” are F(M)P-objects. So MI, including F(M)P, is non-optional even in science. One problem, however, is that this distorts what Sellars means by the “manifest” image. Again, MI enjoys some degree of scientific sophistication; it is not just defined by however reality first strikes us. Still, there is evidence that Sellars counts “appearances” as objects of MI (even though his 1960; 1963a adverbialism is an attempt to fit them into SI). So MI, to that extent, remains inescapable even in science. Yet while this line may be defensible, it has limited effect: If appearances are

12 Preliminaries shown inescapable, this does not yet establish anything about the central F(M)P-objects like norms and belief. A second argument stresses that “inquiry” itself is an F(M)P-object. The goal of the “inquiry language game” is to advance knowledge, understanding or the like—and “knowledge,” “understanding,” etc. are F(M)P-objects. What’s more, inquiry breaks down into sub-activities also defined by F(M) P: One makes observations, records those observations for later assessment, and ultimately draws conclusions, etc. These belong to F(M)P insofar as they are intentional or representational. (These activities also are governed by norms, for there are correct and incorrect observations, records, assessments, etc.) The very notion of inquiry is shot through with F(M)P. The argument is not yet convincing, however, for the premises constitute a theory of inquiry. And no theory of inquiry is required for doing inquiry. An analogy: You don’t need a theory of carpentry to do some carpentry.15 So even if our present theory of inquiry is couched within F(M)P, that does not show that F(M)P is non-optional for first-order, theoretical goals. (But, rejoinders may be available; so the view is worth noting as a potential fallback position.) The third argument shall be developed in the next section. Briefly, however, the idea is that some inquiries—specifically, some normative (philosophical) inquiries—are compulsory for those who value rationality. These inquiries, moreover, have F(M)P as a background assumption. Insofar as this holds, F(M)P is inescapable. In a slogan, F(M)P is rationally required for some rationally required inquiries. (And since we value rationality, these requirements apply to us.) All this is so, even if F(M)P turns out to be a fiction. Some will feel that, ultimately, there is not much argument to the third argument. The next section can have the appearance of a “just so” story. But the problem is that the issues are large, and I cannot pursue them here in a rigorous way. (The notion of “rational requirement,” in particular, is not given the serious explication it deserves.16) Even so, I will be answering objections at various intervals. And one issue is important enough to flag beforehand. This is that the third argument is flatly question-begging against F(M)P-eliminativists—after all, it freely speaks of “valuing” and “rationality.” Yet that vocabulary comes directly from F(M)P. But the main question is not whether eliminativism is true; it is more whether one can really be an eliminativist. Still, the third argument has interest partly since it might justify inescapability, independently of whether F(M)P is true. So although eliminativism per se is not the issue, I do not

15 I owe the carpentry analogy to Ram Neta (in conversation). 16 Fortunately, the literature on “rational requirement” and “rational commitment” is growing. See Broome (1999; 2005; 2007), Shpall (2012), Hinchman (2013), and the papers in Verbeek & Southwood (2009).

Preamble: Is Philosophy Anti-Scientific?  13 want to prejudge the matter. Accordingly, later on, I propose a way of reformulating things in F(M)P-free terms. But for readability’s sake, I shall speak with the folk for now.

0.5  The “Categorical Interrogative” Orson Wells once said, “We’re born alone, we live alone, we die alone.” Happily, on the usual reading of ‘alone’, that is patently false. But it is true that in a fundamental respect, the individual experiences reality as an individual.17 Naturally, the individual’s experiences will include experiences of other individuals. Experience will also be significantly shaped by socially enforced rules and such. So nothing here should imply any kind of “methodological solipsism.”18 Yet it is true that, in some primitive or fundamental manner, the individual encounters reality as an individual. It is not just that her experience is uniquely hers; it is also that others do not bear the responsibility of understanding its significance for her. Thus, questions prompted by her experience are fundamentally confronted “alone.” Others may face analogous questions about their own experience, and they might advise her on the questions faced in her own case. But ultimately, it is her burden whether to accept their answers, to invent other answers, or to leave the questions unanswered. The following I also take to be axiomatic. The individual’s basic encounter with reality is such that, in one sense, the most urgent question that could be asked at any moment is: What should I do? The question features the ‘should’ of practical rationality, concerning effective means to one’s ends (whether or not these ends are “categorical”). The question thus would have no urgency for persons lacking interest in effectively achieving their ends. Yet that would be highly deviant. (For the sane individual, such interest is not even voluntary.) In any event, I shall presume that the individual values means-ends rationality up to some threshold, by which the urgency of the question is made real.19

17 To avoid complexities, “individuals” here are restricted to biologically normal, non-­juvenile minds. Also, I am considering ordinary waking life only, thus ignoring inter alia cases of “religious experience.” (But n.b., what follows is not meant to advocate any essential distinction between human and non-human animals.) 18 (This idiom is now associated with Putnam 1975, though in fact it occurs prominently in Carnap 1931, p. 461.) Also, nothing here precludes extended cognition, group mentality, etc. After all, “the individual” is a mind, and if minds are functional kinds, then even a group mind is an “individual.” 19 The individual’s valuing rationality beyond the threshold is sufficient for urgency, but I do not presume it is necessary. E.g., it may be enough if her community values rationality.

14 Preliminaries In different contexts, different versions of the question might arise. Thus, it can manifest as “What should I do right now?,” though sometimes a longer time-line is presupposed. Yet for those who have the requisite respect for rationality, the now-directed version is not just urgent—it is inescapable.20 For in brief, if one rejects or ignores the question—hereafter, “refuses” the question—that itself rationally commits one to an answer (regardless of whether one realizes it). As I will show, moreover, the individual is thereby committed to norms, beliefs, and in general, the framework of F(M)P. F(M)P in this fashion is inescapable. The point resembles William James’ (1896) view that the question of theism is “forced.” On James’ view, refusing the God-question is a way of answering it. After all, anyone who does not positively answer “yes” to “Do I believe in God?” is, by definition, not a believer (though it would remain open whether she is an atheist, agnostic, or some sort of quietist).21 However, the question of “What should I do?” is not exactly forced in James’ sense. One who refuses our question has not by definition answered it. The person might refuse the question while being undecided on whether she should refuse it. (She just instinctively refuses it, without judging whether this is apt.) But, long story short, one who refuses the question is still rationally committed to her refusal being what she should do. And such a commitment is an answer to the question. So in this respect, the question is “inescapable.” The individual will be committed to an answer not by definition, but by principles of rationality that she herself values.22 In pursuing this further, one complication is that the interrogative “What should I do?” can express a variety of questions in English. For instance, it can express, “What should I do morally speaking?,” or it might express, “What should I do prudentially speaking?” And each of these can be refined further. The first can be precisified as, “What is the morally best thing to do?”23—although among ordinary English speakers, it can have the force of, “What is a morally satisficing thing to do?” In the latter case, the aim is not to discern what is morally ideal, but rather what is “good enough”

20 Until now, statements/theories have been described as “inescapable,” but assume a question can be inescapable in a derivative way: A question is inescapable iff some answer or other to the question is inescapable. 21 Actually, a believer could refuse the question on fideist grounds. (She refuses the question since she thinks it is impermissible in general to raise questions about her religious beliefs.) But nevermind that. 22 Some might say that the principles of rationality alone can rationally commit someone to A, regardless of whether she values rationality. It is true that the principles suffice to define what is rationally required in a given case. But conceivably, an individual might not be committed to what rationality requires. In order to get that commitment, it seems she (or her community) also needs to respect principles of rationality to some minimum extent. 23 Despite the use of ‘the’, none of these questions really presuppose there is a unique action that figures into a correct answer. There is at least the possibility of a disjunctive reply: “the best thing to do is either A, B, or C.”

Preamble: Is Philosophy Anti-Scientific?  15 (given the constraints of the real world). Presumably, morally satisficing acts are morally permissible at least, yet sometimes the bare moral minimum is not enough to be “satisficing.” It depends on whether there is a commitment to going beyond the moral minimum. Similarly, the question about prudence could concern what is prudentially best, or just what is prudentially satisficing.24 But in this discussion, “What should I do?” is meant to express: “What is a satisficing thing to do, all things considered?” Here, “all things considered” (or “∀TC”) covers all moral considerations, prudential considerations, aesthetic considerations, considerations of etiquette—and in general, all considerations deriving from any rule of action that is valued at the time.25 Roughly, where A is an act that the individual is currently performing knowingly and voluntarily, A is ∀TC-satisficing if either it has a weighted-most considerations in its favor or it does so “nearly enough.”26 Caution is advisable regarding the phrase ‘rationally committed to x’. Here, it is used in expressing that one is rationally required to believe that refusal is the thing to do—though one could still fail in this requirement. Regardless, this prescription is applicable to the individual, given that she (or her community) sufficiently values rationality. And in the case under consideration, the individual is committed to: Refusing ‘What should I do’ is what I should do.27 And the irony, again, is that this commitment is an answer to that very question. The inescapability argument can now be made with added clarity. Suppose that someone (knowingly and voluntarily) refuses the question “What should I do?” when it concerns what is satisficing, all things considered.

24 There are other contrasts in what the interrogative expresses. Most obviously, “What should I do?” often means, “What am I morally obligated to do?” (As should be clear, what is obligatory is not necessarily the same as what is morally best or morally satisficing.) For additional distinctions of this sort, see chs. 9 and 10 of White’s excellent (1975). 25 N.B., my reading of “What should I do?” assumes that an act is practically rational iff it is ∀TC-satisficing. I cannot defend this assumption here—yet intuitively, calling an act “∀TCsatisficing” and calling it “rational,” in the practical sphere, are both ways of saying the act is a sufficiently effective means to an end. 26 Plausibly yet not uncontroversially, the weight of a consideration reflects how much it is valued (by the individual or community). I want to keep the above sketch of practical rationality relatively uncontentious, but naturally, objections exist. Unfortunately, I cannot delve into the controversies here. Yet n.b., the restriction to acts currently being performed allows us to side-step worries from Broome (2007). We need not decide if rationality means responding behaviorally to one’s reasons. My point is not that rationality commands a specific act given one’s beliefs. Instead: Given how one is currently acting, rationality commands a specific belief. 27 I also speak of the individual being ‘rationally committed to an answer’, but an ‘answer’ is just a kind of proposition. Another variation is the talk of, e.g., an ‘ontological commitment to x’. But as usual, this is shorthand for a rational commitment to believing the proposition that x exists.

16 Preliminaries Then, her refusing the question still rationally commits her to an answer. For there is a rule of rationality along the lines of: (R1) If an act A is not ∀TC-satisficing, then do not do A.

Hence, if refusal is what she does, (R1) means she is committed to her act of refusal being satisficing, all things considered.28 The question is in this way inescapable: She cannot avoid rationally committing to an answer, given the applicability of rules like (R1). The argument has a degree of semblance to Kant’s remarks on the categorical imperative. Kant said (something like) it is partly constitutive of being rational that one is morally bound to certain rules. Regardless of whether that is so, the parallel is that having sufficient respect for rationality means that one is committed to an answer to “What should I do?” (on the specified reading). It is a “categorical interrogative” if you like. Again, whatever one does, one is rationally committed via (R1) to the act being ∀TC-satisficing—and such a commitment is an answer to the question. What does this have to do with F(M)P? Well, an answer to “What should I do?” pretty clearly implies that there is some norm. (‘Should’ is the normative term par excellence.) Moreover, rationality prescribes a rule which says, roughly: (R2) If you are committed to an answer that is known to imply p, then commit to p.29 Hence, via such a rule, the individual’s answer commits her to the existence of a norm. So in sum: Thanks to the categorical interrogative, the individual is inevitably committed to an answer about what she should do, and that answer commits her to the existence of at least one norm. Thus far, this says nothing about F(M)P objects like belief (but see the next section). Yet even when our focus is limited to norms, one objection is that ‘should’ need not be used as a normative term. After all, “What should I do?” might be understood as a purely non-normative affair. For example, for a certain sort of utilitarian, it just concerns our naturally occurring preferences, plus facts about which actions maximize the satisfaction of those preferences. If this is right, then the ‘should’ in ‘A is what I should do’ is apparently non-normative—the sentence roughly means “my doing A will 28 This presumes she is bound to the rule “be committed to those things dictated by rules you value.” Yet plausibly, that itself is a rule of rationality—which would mean she is committed to that rule. (Admittedly, there is a rule-following regress in all this; on such matters, see section 0.7.2 below.) 29 I leave it to the reader’s discretion whether the individual herself must know the implication, or whether it is enough if her community is thus aware. Cf. n. 19.

Preamble: Is Philosophy Anti-Scientific?  17 maximize the satisfaction of such-and-such preferences.” So in committing to an answer, why must the individual be committed to norms? Actually, this non-normative explication of normative language is useful to us. For it can help illustrate that valuing rationality comes with a commitment to norms, even if there are no norms. (You are inescapably committed to norms, even if your concern with norms can be explained merely in terms of what maximizes preference-satisfaction.) And our objector is partially right: If the non-normative explication is allowed, then an individual herself can regard the categorical interrogative, plus her answer to it, all in non-normative terms. However, I would grant her this ability only at some non-fundamental level. For when we turn to our deepest commitments, I quite doubt that the individual is able to purge herself of every normative notion. Recall the bit of phenomenology that opened this section. The individual’s basic encounter with reality is such that, in one sense, the question “What should I do?” is urgent at every moment. Now as should be clear by now, I allow that there may ultimately be no real normative facts “out there.” However, in considering one’s deepest commitments about what one should do, the commitments employ unmistakably normative concepts. Still, one might ask whether these “fundamental” or “primitive” commitments to F(M)P really amount to ontological commitments. I am willing to concede that in one sense, the answer is “no.” It is well known that a professional philosopher’s “commitments” often do not reflect what s/he really believes. Indeed, I myself am not “committed” to F(M)P in such a fashion, for I “officially” allow that F(M)P may be fictitious. That is so, even though at a psychologically deeper level, I think we are all inescapably committed.30 But what do these two kinds of commitment really amount to? Sometimes, philosophy requires a bit of contrivance. Descartes is explicit about this in the opening pages of the Meditations. He admits he cannot really suspend all his commitments about the external world; that is not under his voluntary control. Instead, he effectively decides to pretend that they have been suspended. In practice, this means his arguments refrain from premises about the external world (unless established by previous arguments). Yet the practice does not mean that he succeeds in suspending all his beliefs about the external world. He is instead acting as if those beliefs are suspended, for some larger philosophical purpose.

30 It is heartening to know that the difference between a philosopher’s “official” commitments and her real commitments does not always show intellectual dishonesty—she is not necessarily arguing in a sophistical way for ideas that she herself finds absurd. Sometimes, she is instead openly adopting “pretend commitments” for some greater philosophical purpose. (This can range from the mundane supposing p for reductio—or the heroic advocating of p as the most reasonable view, contra an admitted prejudice to the contrary.)

18 Preliminaries In the present dialect, the distinction also applies. The objection, recall, is that perhaps John Mackie or A. J. Ayer can grasp the categorical interrogative and their answers to it in purely non-normative terms. But when we are discussing our deepest commitments, I doubt that homo sapiens are able to purge themselves of all normative thinking. Granted, we philosophers are accustomed to suspending judgment (or rather, pretending to suspend judgment) about wildly unintuitive ideas. This practice is advantageous in, for example, challenging dogma. But there can be disadvantages. In this instance, it creates unclarity about which commitments are real and which are just pretend. As in Descartes, there might be a real philosophical point in adopting the pretense. But one should not forget that it is a pretense. Descartes, Mackie, or Ayer might be “uncommitted” to certain sorts of entities, i.e., uncommitted according to a philosophical pretense. But it does not follow that they are uncommitted. The argument thus comes to this. Outside of all pretense, even the most die-hard moral anti-realist is inescapably committed to answers regarding “What should I do?” And with these answers comes an ontological commitment to norms. The anti-realist may explicitly deny that any of her answers are “normatively charged.” But in the end, this is a species of bad faith. At the most basic level, the individual engages reality in a way that makes “What should I do?” a distinctly normative question, owing to her valuing rationality. And the unavoidable answers come with the unavoidable ontological commitment to norms. This is not to say that there really are norms. It is rather to illustrate that an ontological commitment to norms is inescapable.

0.6  Commitment to Belief and to Self-Knowledge There are further questions here that should be addressed—but for expository purposes, let us first extend the argument beyond norms. What about other F(M)P-objects such as belief? Consider here a version of the question whose scope is limited to acts of believing:31 What should I believe?

31 Believing is assumed to be a (minimal) sort of act, yet it is normally not a voluntary or intentional act. However, one still bears responsibility for it, per norms of rationality. See Sosa (2011) and Toribio (2011) for related views. But see also Kornblith (2012, ch. 3), who is forceful in his attacks on Sosa. Yet I resist Kornblith’s argument (p. 106) that, since our cognitive processes are largely non-introspectable and involuntary, a responsibility to reflect on one’s beliefs could not be significant. For a responsibility to A may not imply an ability to A, on at least some readings of ‘ability’. (I have in mind a view of responsibility like that of Strawson 1962. See also Doris 2015.)

Preamble: Is Philosophy Anti-Scientific?  19 For those who sufficiently value rationality, this too is inescapable if read as “What is a satisficing act of belief, all things considered?” (N.B., some considerations—especially epistemic considerations—may loom larger than before. Accordingly, it may be best to model the semantics of ‘all things considered’ as context-sensitive, and in particular, as sensitive to the importance of different considerations on different occasions.32) Why think the second question is “inescapable” in the manner of the first? Grant that the individual is committed to an answer first, and assume this answer is of the form “A is an ∀TC-satisficing act.” Then, in virtue of being so committed, the individual is also rationally committed to the claim: “the act of believing that answer is an ∀TC-satisficing act.” That’s because among the principles of rationality, there is something like the general rule: (R3) If believing an answer is not ∀TC-satisficing, then do not commit to that answer. Hence, since the person is committed to the answer “A is an ∀TC-satisficing act,” (R3) commits her to: “believing that answer is ∀TC-satisficing.” And this latest commitment just is an answer to “What should I believe?” (on the intended reading). It is basically a commitment to “All things considered, I should believe this: A is what I should do, all things considered.” Therefore—commitments on which acts are ∀TC-satisficing rationally commit one to answers about which beliefs are ∀TC-satisficing. Now strictly speaking, it does not yet follow that one is committed to the existence of belief. One is just committed to a view about which beliefs are satisficing, in any world where belief exists. And thus far, it is left open whether our world is such a world. Nevertheless, if it arises in the primeval way, the question “What should I believe?” invariably presupposes that one is a believingsort of creature. It is presupposed that one already has beliefs, and that the concern is to identify which are ∀TC-satisficing, and which are not. In that respect, ontological commitment to belief is non-optional. The point can also be argued more directly, if the following is also seen as inescapable: What do I believe?

32 This allows the pleasing result that theoretical rationality is just a special case of practical rationality, wherein epistemic considerations are given more importance. Though if pressured, we need not insist on this. Some may think that only epistemic considerations should be relevant to answering “What should I believe?” However, this may oversimplify things (cf. Rescher 1987, ch. 1; Rinard, forthcoming; Sharadin 2016). Arguably, a person ought to trust her spouse’s declaration of innocence, even if the epistemic considerations somewhat favor a guilty verdict. (There are limits to such epistemic tolerance, however.)

20 Preliminaries Plausibly, we are inevitably committed to answers here as well. Argument: Grant that the individual is committed to an answer to “What should I do?,” and assume this answer is of the form “A is an ∀TC-satisficing act.” Consider also that principles of rationality likely include principles of charity, such as the following (where S is any individual): (R4) If S is rationally committed to p, then commit to S’s believing that p (unless there is sufficient evidence against her so believing). One can think of (R4) as a “presumption of innocence” regarding the individual’s rational commitments: She is presumed to believe what she is rationally committed to, unless there is special reason to say otherwise. Observe next that (R4) is applicable even to oneself. Given the inescapability of “A is ∀TC-satisficing,” then where S = you, (R4) forces a commitment to you yourself having the belief that “A is ∀TC-satisficing” (unless enough counter-evidence exists). This means that ceteris paribus you are rationally committed to: “I believe that A is ∀TC-satisficing.” Yet a commitment to that obviously comes with an ontological commitment to belief; cf. (R2).33 Though again, none of this implies that the ontological commitment is correct. The point is rather that the central posit of F(M)P is non-optional, owing to the value we place on principles of rationality. The argument can be reconstructed as follows, with such principles now playing the role of inferential rules. (In what follows, ‘Necessarily’ should be read as ranging over only those worlds where the individual has the same physical/psychological abilities as in this world): (1) Necessarily, the individual S values rationality sufficiently such that she is committed to conforming to the requirements of basic means-ends rationality.[Premise] (2) S performs some action A. [Assume for conditional proof] (3) S is rationally committed to “A is ∀TC-satisficing.” [From (1), (2), by (R1)] (4) S is rationally committed to S having the belief that “A is ∀TC-satisficing.” [From (1), (3), by (R4)] (5) The proposition committed to in (4) implies that belief exists. [Trivial] (6) S is rationally committed to the existence of belief. [From (1), (4), (5), by (R2)] (7) If S performs some action A, S is rationally committed to the existence of belief. [From (2)–(6), by conditional proof] (8) Necessarily, if S performs some action A, S is rationally committed to the existence of belief. [From (7), by modal necessitation] 33 Here too, knowing of your commitment to “A is ∀TC-satisficing” is unnecessary for incurring the ontological commitment. The commitment is forced on you, regardless of whether you are aware of it.

Preamble: Is Philosophy Anti-Scientific?  21 The conclusion (8) then naturally leads to the claim that for creatures like us (agents who value rationality), an ontological commitment to belief is inescapable. The inescapability of “What do I believe?” is of particular interest to the book’s overall interests. Normally, an answer would presuppose access to one’s own beliefs; it would manifest a kind of self-knowledge. Does this mean that, in being committed to “I believe that p,” one is committed thereby to a capacity for introspecting one’s own beliefs or the like? If so, then since the former commitment is inescapable, so too would be the latter! A rational commitment to this kind of self-knowing capacity would be unavoidable—even if you lack any such capacity. Moreover, rationality may well force this commitment. For the following plausibly articulates a rational requirement: (R5) If you believe p, then commit to your having some ability to detect whether p. In ordinary sorts of cases, (R5) seems sensible enough. For instance, if you believe that p on the basis of perception, where p = snow is white, (R5) rationally commits you to your being able to detect whether snow is white (presumably a perceptual ability). Yet in the same way, if you believe that p on the basis of reflection, where p = I believe that snow is white, then (R5) commits one to some kind of ability for detecting this belief. That amounts to a (minimal) commitment to self-knowledge. The folk may lack an explicit commitment to such an ability, even if some count as fully rational. Further, some philosophers and psychologists adopt an explicit commitment against such an ability (as far as their “official” commitments are concerned). Still, perhaps anyone who sufficiently values rationality is at least implicitly committed in the manner described. Note that if, per (R4), the individual believes her inescapable commitment on “What do I believe?,” then the person is capable of “metacognition” (belief in what beliefs she has). Similarly, if she believes an answer to “Am I able to know what I believe?,” she is capable of metametacognition (belief about whether she has knowledgeable beliefs about what she believes.) So does rationality require us to have metametacognitive abilities? No matter— “the individual” as understood here (cf. n. 18) is a typical adult, plausibly capable of metametacognition. Thus, a commitment to a modicum of self-knowledge may well be inescapable.34 That serves well our overall purposes. The aim, recall, is to defuse the charge that the Sellarsian is anti-science, when insisting on the manifest image. The reply being developed is that the individual is inescapably 34 This somewhat resembles the popular idea that rationality requires a self-knowing ability (Burge 1996, Shoemaker 1996, Moran 2001, Bilgrami 2006). Yet my point is not that rationality requires the ability itself, but rather a belief in the ability.

22 Preliminaries committed to norms and belief. And at least for metametacognizers, the same would hold for self-knowing, per the argument from (R5). If we are thus trapped in such commitments (outside any pretense), it is then understandable why the Sellarsian might try to reconcile them with science. She could not be blamed for trying. The individual’s primitive engagement with the world is such that some normative questions are categorical. This is surprising in one respect. Going by the scientific image, we tend to think of ourselves as fundamentally biological or physical entities—as entities in the “space of causes.” There is nothing wrong with this in itself. Nevertheless, we should not forget that—for better or for worse—we are also fundamentally denizens of the “space of reasons.” We are creatures who are inevitably committed to an answer to “What should I do?,” via principles of rationality that we ourselves value. The question “What should I believe?” is similarly inescapable, and with it, the ontological commitment to belief and to self-knowledge.

0.7  Some Clarifications At this stage, allow me to tidy up some stray issues that have arisen during the course of the discussion. 0.7.1  The Kantian Gambit First, we should distinguish the foregoing from a similar argument which, although alluring, is one I do not embrace. This begins with the idea that we value as an end actions that are ∀TC-satisficing. It then adopts the ­“Kantian” premise that to value x as an end rationally requires taking an effective means to x. Suppose further that an effective means to ∀TC-­ satisficing action is to explicitly answer (beforehand) the question “What is an ∀TC-satisficing act?” Then, one is rationally required to have an answer that is committal, re: norms. Also, for reasons given earlier, that in turn rationally commits one to an answer to “What do I believe?,” etc. Thus, one becomes inevitably committed to the F(M)P-framework. This is not wholly convincing, however one reason is that evidence from psychology suggests that reflection on a question like “What should I do?” is often not an effective means to satisficing action. In fact, it often can encourage anti-satisficing action35 So there is some empirical doubt on a premise of the argument. In contrast, the argument offered here does not rest on such a premise. “What should I do?” and “What should I believe?” are inescapable not 35 I mainly have in mind the data presented by T. Wilson and colleagues. For references, see n. 1 of chapter 1.

Preamble: Is Philosophy Anti-Scientific?  23 because asking such questions may promote satisficing action. Rather, the idea is that one who values rationality is already rationally committed to answers. My de facto action rationally commits me to the act being ∀TC-satisficing—which in turn, rationally commits me to an answer to “What do I believe?” Thus, the questions are inescapable not qua rational means to some end. Instead, the questions are inescapable insofar as our actions and our valuing rationality inevitably commit us to answers. *0.7.2  Is Folk Psychology Presupposed? Prior to launching into the argument, a warning was issued that folk psychological terminology like ‘valuing’ would be employed. And I issued a promissory note for rewording matters without such language. This section is where I follow through. Let it be clear, however, that the translations proposed are not ones I am ultimately wedded to. The aim is just to illustrate how similar points could be made without using the language of folk (/moral) psychology. Whether they are exactly the same points is not really the concern.36 As a start, talk of “valuing rationality” may just concern dispositions for specific kinds of behaviors—those we are disposed to tag with the term ‘rational’ (i.e., to utter ‘rational’ in the presence of the act). Similarly, talk of “commitment to p” may just tag dispositions to utter ‘yes’ when given as stimulus “p?” In which case, the inescapability thesis can be paraphrased as: Dispositions for specific kinds of behavior (those we tag as “rational”) yield dispositions to utter ‘yes’, given as stimulus an interrogative-token like ‘Are there norms?’, ‘Do you have beliefs?’, etc. And the argument for this is translated as: The disposition for rational behavior co-occurs with a disposition to tag your own behavior as “rational” (which intuitively reflects that valuing rationality commits you to your own behavior being rational). Whence, the latter disposition comes with a disposition for uttering ‘yes’ to a stimulus like ‘Are there norms?’, ‘Do you have beliefs’, and so forth (intuitively indicating ontological commitments to norms, beliefs, etc.). Notoriously, however, a disposition can be triggered incorrectly (Kripke 1982; cf. also McDowell 1984, Boghossian 1989, etc.). And it seems a dispositionalist explication of normative terms cannot acknowledge that. For that would suggest disposition-independent norms of correctness.

36 Relatedly, I am not concerned to give a dispositionalist metaphysics of valuing or rational commitment. I only wish to paraphrase my talk of “valuing,” “rational commitment,” etc., in dispositionalist terms. This may not yield synonyms for ‘valuing’, ‘commitment’, etc.; concurrently, it may not imply the right metaphysics.

24 Preliminaries An illustration: If Patricia utters ‘yes’ as a response to the stimulus ‘Are there beliefs?,’ then she obviously is disposed to utter ‘yes’ in that case. Yet it remains possible that she is, intuitively, not committed to beliefs. For instance, Patricia’s utterance might be insincere, or she may have misheard the question, or may have made a linguistic performance error, etc. Thus, it seems (e.g.) ‘commitment’ is inadequately glossed by the dispositionalist account. Still a dispositionalist may reply that when Patricia utters ‘yes’, this does not indicate a commitment if her utterance causes a disposition (in herself or in others) to respond negatively to her utterance. Basically, if her uttering ‘yes’ causes dispositions for negative feedback, then the utterance is “incorrect.” Conversely, if her utterance causes dispositions for positive feedback, then her response is “correct.” (If it causes both types of disposition, it is then “controversial;” if it causes neither sort of disposition, then the correctness of Patricia’s response is “unknown.”) The standard rejoinder would be that this just pushes the normativity issue back. For it is also conceivable that Patricia’s utterance of ‘yes’ causes dispositions only for negative feedback when she is committed to what she affirmed. (She is universally and wrongly accused of lying.) Dispositionalism, as I see it, should reply by denying the viciousness of the looming regress. A dispositionalist should respond that the subsequent negative feedback is itself “incorrect” if it causes dispositions for negative feedback to that negative feedback. (Similarly, it is “correct” if the initial feedback causes dispositions for positive feedback vis-à-vis that feedback.) And in like manner: The new, “second-order” feedback is incorrect if it causes dispositions for “third-order” negative feedback (and correct if it causes dispositions for third-order positive feedback). And so on, etc. In this manner, if there is ever a question about whether some response is “correct,” the dispositionalist can posit some higher-order disposition that makes things look ok. But in the end, does this mean 80,000,000 Germans can’t all be wrong? If the group is wholly co-ordinated to give positive feedback to positive feedback to positive feedback . . . etc. . . ., then the dispositionalist view seems to be that your response must be correct. However, couldn’t Patricia succeed in fooling everyone with her lies? (And couldn’t she be pleased with herself about this, so that there is not even a disposition in herself to react negatively to her utterances?) The envisioned dispositionalist could add that even if no one is presently disposed to react negatively to Patricia’s lie, the behavior is still “incorrect” if they have such dispositions at a different time. But what if the lie causes dispositions for positive feedback at all times? Well, the dispositionalist can say that, given that we are presently considering a mere hypothetical or a nonactual “possible world,” Patricia’s behavior is still “incorrect” in that world since we in the actual world are disposed to respond negatively to her

Preamble: Is Philosophy Anti-Scientific?  25 counterfactual behavior. That is, since we are talking about a counterfactual error, the error can be determined as such by the fact that we actual-world inhabitants deem it an error. (This, by the way, seems to be an unappreciated move in the dispositions literature.)37 This dispositionalist view remains tenuous, and there are many controversies that remain untouched. Regardless, if it is granted to be minimally viable, then that suffices for my purposes. The aim was just to forestall the objection that the language of the “inescapability” thesis and its argument requires the truth of F(M)P. The dispositionalist possibility, if the possibility is granted, shows that that is not so. And all this was to buttress the idea that even if F(M)P is false, it is still inescapable. From the dispositionalist angle, the inescapability thesis amounts to this. Even if there are no beliefs, desires, etc., we who are disposed to utter ‘yes’ to the stimulus ‘Do you value rationality?’ are also disposed to utter ‘yes’ to ‘Are there norms/beliefs?’ Or if the second disposition is lacking, then we face the prospect of negative conditioning within the communal feedback loop. (And if a hypothetical community-wide mistake is raised, the mistake can be fixed as a mistake, in virtue of our actual utterances to that effect.)

0.8  Closing Remarks The Sellarsian can stand blameless when she defends F(M)P, since some parts of the manifest image are simply non-optional. If we want to make the most intuitive sense of things, or at the psychologically deepest level, it is then understandable why reconciling conflicts with science might be a desideratum. Perhaps reconciliation is less than scientifically ideal—but there is some pretext, given the reality of who we are. The point can be illustrated in an alternative fashion. Recall from section 3 that cognitive neuroscience may end up resembling Quantum Mechanics in a way. In both cases, a full understanding of the phenomena may simply be beyond reach. And it is an undesirable feature of a theory if it leaves us with a sense of bewilderment. (Bewilderment is what theories hopefully dispel.) With QM, however, its empirical support is just too strong to render this a serious complaint. In contrast, as things presently stand, there is no eliminativist paradigm that commands empirical supremacy.

37 The idea is not that the behavior is incorrect if in some world people have negative dispositions toward my counterfactual lies. It is rather that the behavior is incorrect if in this world people are thus disposed toward them. But couldn’t this group of thisworldly dispositions be inappropriate also? Sure, but again, that is because there are this-worldly negative dispositions against some counterfactuals involving the first set of dispositions.

26 Preliminaries (No cognitive paradigm of any sort has such command.) At this stage, then, the instinctual sense of bewilderment left by eliminativism is worth admitting as a drawback. It is, thus far, an excuse for trying different ways to incorporate the data into our manifest conceptual scheme, in the manner of the Sellarsian.

Appendix: Philosophy as of Categorical Value

The foregoing puts us in an excellent position to discuss the value of philosophy. This is not wholly continuous with the discussion above—but it is obviously related, and I could not pass the opportunity by, given all the relevant infrastructure in place. However, I shall approach the issue a bit indirectly. There is a sense in which metaphilosophy is inevitably “first philosophy.” The idea is not that one’s metaphilosophy is an unrevisable epistemic “foundation” for the rest of one’s philosophical views. It is more that rationally speaking, the metaphilosophy should be a “background assumption” to one’s philosophical work. This is not to say that each argument would depend on some idea from the metaphilosophy. Rather, the sense and purpose of the philosopher’s work would be expressed by the metaphilosophy. The metaphilosophy would explain what the point is of her philosophical efforts. It would answer: Why bother with these philosophical issues? What benefit is to be gained? If her metaphilosophy does this in relation to all of her philosophy, then the metaphilosophy does it in relation to the metaphilosophy itself. Her metaphilosophy would give sense and purpose to her metaphilosophy (and it is only right and natural that the metaphilosophy should be rationally motivated). But if so, then one type of question is inevitably begged. A metaphilosophy can conceive of philosophical work as either valuable in itself or as valuable for what it yields. If the former, then there is nothing beyond the work itself that explains the value of the work. So if a skeptic were to challenge its value, there is nothing the philosopher can point at to persuade the skeptic otherwise. For the value of the work is explained, if at all, by the value of the work. And that would hardly convert a skeptic. Suppose, on the other hand, the metaphilosophy explains the value of philosophy by its distinctive results (philosophical insight, wisdom, or the like). Then, if the skeptic rejects the value of those results, we are again trapped in a question-begging situation. Note well: The view should say that philosophy is valuable for the distinctive things it yields, i.e., things that one

28 Preliminaries cannot obtain elsewhere (or at least not as efficiently).38 Otherwise, philosophy would still seem expendable. So it is best not to cite something other than its distinctive results in the attempt to convert the skeptic. Yet merely reiterating the value of those results would be question-begging. In recent decades, the whole of philosophy seems caught in this standoff, with the skeptic being played by more “practical” members of society. What I am suggesting, however, is that this is to be expected. For the skeptic, nothing will justify the value of philosophy except some non-philosophical benefit, like helping the economy, advancing technology, or the like. There is nothing wrong with those things per se. Yet the distinctive benefits of philosophy are (of course!) philosophical. Thus, the standoff. The solution, I think, is to convince the skeptic that there are other things of value, besides the economy, technology, etc.39 The argument is rather like the phenomenological strategy against the eliminativists. To an extent, it is exactly like it. Questions like, “What should I do?” and “What should I believe?” are ones that animate us all, fundamentally. The skeptic can verify from his own phenomenological perspective. What this represents, moreover, is not just some petty fancy. Rescher (2014) puts the point forcefully: The discomfort of unknowing is a natural component of human sensibility. To be ignorant of what goes on about us is almost physically painful for us—no doubt because it is so dangerous from an evolutionary point of view. . . . For us, cognitive orientation is itself a practical need: cognitive disorientation is actually stressful and distressing. (p. 3)40 And to repeat, since “What should I do?” and “What should I believe?” are normative questions, they fall within the proper purview of philosophy.

38 To be sure, philosophy may offer a variety of benefits. I think it is especially effective at increasing general I. Q. But that is not why the philosopher falls in love with philosophy. 39 Cf. Russell (1912): “we must first free our minds from the prejudices of what are wrongly called ‘practical’ men. The ‘practical man’ . . . is one who recognizes only material needs, who . . . is oblivious of the necessity of providing food for the mind. . . . It is exclusively among the goods of the mind that the value of philosophy is to be found; and only those who are not indifferent to these goods can be persuaded that the study of philosophy is not a waste of time” (pp. 238–239). 40 Psychologists have long held that well-being requires nonmaterial goods (e.g., social support, autonomy, etc.). Further, the evidence is that material goods do not trump all. (Cf. Nietzsche: “He who has a why to live can bear almost any how.”). E.g., in the third-world, people are measured with above-average “subjective well-being” despite unsatisfied basic needs (Tay & Diener 2011). More broadly (contra Maslow), psychologists now typically say that “all human needs are interrelated and interactive. With the sole exception of the need . . . to remain alive, no hierarchies exist within the system. On the contrary, simultaneities, complementarities and trade-offs are characteristics of the process of needs satisfaction” (Max-Neef et al. 1989, p. 17, italics mine).

Preamble: Is Philosophy Anti-Scientific?  29 Further, the categorical interrogatives do not stop there. In the end, answers to these two questions rationally require answers to further questions. For instance, if one answers, “I should believe that Fa, all things considered,” the norm is to have a reason for believing Fa. Hence, one is compelled to have an answer to a third question: What is my reason for believing Fa? But so far, none of this need involve the distinctly esoteric questions studied by professional philosophers. More, it is beyond obvious that many of those questions are not categorical. (In all likelihood, some are not worthwhile at all.) However, despite what commonsense may say, I suspect that several of the so-called “esoteric” questions are categorical. An illustration. Suppose I respond to “What is my reason for believing Fa?” by appeal to my belief that Gb. As is familiar from epistemology, a regress starts to loom. For rationality appears to enforce the following rules as well: (R6) No dogmatism: Do not believe p without a reason. (R7) No circularity: A sequence of reasons for believing p cannot include a belief that p. In short, these make compulsory an answer to: What is my reason for believing Gb?, etc. And thus rationality seems to demand infinitely many beliefs to have even one rational belief. Yet since I seem not to possess infinite chains of reasons, my beliefs apparently fail to be rational. Even now, this may seem of interest only to philosophers. Nonetheless, there is a rule of rationality which says something like: (R8) If one’s believing that p appears irrational—then take some rationally permissible means to change that appearance, or stop believing that p. This plausibly is a rule that applies even in everyday contexts. Ordinarily, if a belief starts to seem irrational, I must either give it up, or explain away the apparent irrationality by some rationally permissible means. But obviously, I cannot stop believing that p, for an arbitrary p. Yet the regress makes each of my beliefs look irrational. So in some rationally permissible way, (R8) requires me to change that. This means, effectively, that rationality forces me into a heavyweight philosophical question, one that normally arises only in the philosopher’s study: In light of the regress, how can any belief be rational? Now regardless of where you stand on this question, the present point is metaphilosophical. Even though this is a “philosopher’s” philosophical

30 Preliminaries question, it is starting to look like a categorical interrogative, where an answer would be demanded of anyone who values rationality. That is so, assuming that (R6)–(R8) are among the rules of rationality. Yet surely this is preposterous. Not every man, woman, and child should be faulted if they lack a solution to the regress problem! Perhaps the fault instead lies with (R8). Perhaps a superior version of (R8) would be: (R8') If one’s believing that p appears to be irrational—then if one is able, take some rationally permissible action to change that, or stop believing that p. The worry, however, is that (R8') would let us off the hook for too much. Suppose Chet was raised by Nazis, and has been so conditioned to be incapable of seeing Jews as equals. Naturally, his stance will often appear irrational. Yet if it is impossible for him to surrender the belief (and since it cannot be made to look rational by any rationally permissible means), then (R8') would not impugn his rationality. Chet could still count as fully rational. But that is absurd.41 There are other issues with (R8), yet suppose we allow it. How then could the ordinary person remain blameless in ignoring the regress? I hypothesize that a division of intellectual labor may be operative (akin to that from Putnam 1975). Yet here the labor is not individuating natural kinds or whatnot, but rather meeting the demands of rationality. Solving the regress problem thus would be allotted to the epistemological “experts.” If the ordinary person is willing to defer to them, she would then be discharged with the burden. Indeed, one can see this as her “taking rationally permissible means to change” the appearance of her own irrationality, as per (R8). She outsources the labor. If this hypothesis is correct, it leaves intact that the “esoteric” philosophical question is a categorical interrogative, a question forced upon us by principles of rationality. Yet it is not merely that rationality puts pressure on us to answer it. As with the other categorical questions, an answer would be inescapable: The general commitment to rationality would already commit us to an answer. With the regress question, however, an ordinary person’s commitment would be nondescript, given that she leaves the matter to the experts. In the typical case, she would be rationally committed to just “whatever answer the experts think is best.”42

41 Perhaps Chet is not to be rationally blameworthy if he is psychically incapable of changing. But he still fails to be fully rational—thus (R8') cannot suffice for full rationality. (An analogous point is what I suggest for the Sellarsian: She is not blameworthy for defending MI, even if it is less than fully rational. Yet she more plausibly avoids blame, since “Rational ought implies psychological can” seems more clearly applicable there than to Chet’s case.) 42 Unfortunately, epistemologists are not agreed on the solution to the regress problem. Though we might still commit the folk to “whatever answer is favored by a weighted-most of experts.”

Preamble: Is Philosophy Anti-Scientific?  31 If the central questions of philosophy are categorical interrogatives, then the value of philosophical inquiry is categorical. It would owe to the value assigned to rationality. For it would address urgent questions forced on us by basic principles of rationality that we all value. One feature of this view I especially like is the egalitarian implication. Philosophy would not just be an amusing pastime for the elite; its answers would be of fundamental interest to everyone.43 But here is an entry point for metaphilosophical skepticism. Suppose that philosophy can achieve what I’ve advertised. Still, who really cares whether we conform to the more esoteric demands of rationality? Why should it matter? Well, why should anything matter? My own inclination is to say that nothing ultimately matters (including the fact that nothing ultimately matters). Regardless, in non-ultimate terms, what counts as “mattering?” Earlier, I discussed a skeptic who holds that the only things that matter are non-philosophical goods and services, having to do with technology, the economy, and the like. I admitted, moreover, this skeptic cannot be dismissed without begging a question. However, the skeptic also begs in speaking of the philosopher’s “esoteric” rational demands. And our desire to be rational is not some trifling fetish. One is “thrown into the world,” as they say, and the hope is to live under the direction of good beliefs, motivated by good desires, to act according to good choices. Norms of rationality define the pertinent sense of ‘good’. In the case of belief, the norms advise us on choosing beliefs in the most truth-conducive and useful ways. The regress suggests, however, that it is impossible to follow this advice. The threat is then life without real epistemic guidance, in which inquiring agents are lost and attempts to find one’s way are absurd. No actual human being is indifferent to this. Actual human beings recoil at the absurd, and that discloses the value given to rationality. As we saw, moreover, the value of even some “esoteric” philosophy derives from that. So in the end, the retort to any actual skeptic is: You are already committed to the value of philosophy. It derives from the value we all put on rationality. For one sort of skeptic, however, this too is a question-begging response. Yet fighting the rationality-indifferent skeptic is like trying to justify NASA to those who have no spirit of exploration. But human beings are interested in outer space. They just are. In the same way, actual human beings are

43 Yet, to an extent, the value of philosophy lies not in the answers it proffers, but in drawing attention to the questions themselves. Here again is Russell: “The value of philosophy is, in fact, to be sought [partly] in its very uncertainty. The man who has no tincture of philosophy goes through life imprisoned in the prejudices derived from common sense, from the habitual beliefs of his age or his nation. . . . In such a life there is something feverish and confined, in comparison with which the philosophic life is . . . free” (1912, pp. 242–243; 244).

32 Preliminaries interested in exploring the “space of reasons,” including the pro-reasons for our self-image as the “rational animal” (despite what the regress may suggest).44 Philosophy is what serves such interests. And more broadly, philosophy addresses the “categorical interrogatives” animated by the value we place on rationality.45 Throughout, the Kantian themes have been obvious, yet I had not realized until recently the extent of it. The concern of the first Critique is metaphysics rather than philosophy per se. But ignoring that, many essentials of the preceding are stated in the very first paragraphs of Kant’s Preface (and ‘philosophy’ might have just as well replaced Kant’s term ‘metaphysics’): Human reason has the peculiar fate in one species of its cognitions that it is burdened with questions which it cannot dismiss, since they are given to it as problems by the nature of reason itself. . . . It begins from principles whose use is unavoidable in the course of experience and at the same time sufficiently warranted by it. With these principles it rises (as its nature also requires) ever higher, to more remote . . . principles that . . . seem so unsuspicious that even ordinary common sense agrees with them. But it thereby falls into obscurity and contradictions. . . . The battlefield of these endless controversies is called metaphysics. (The Critique of Pure Reason, pp. Avii–Aviii, my italics, his boldface)

References Bilgrami, A. (2006). Self-knowledge and resentment. Harvard UP. Boghossian, P. (1989). The rule-following considerations. Mind, 98(392), 507–549. Broome, J. (1999). Normative requirements. Ratio, 12, 398–419. ———. (2005). Does rationality give us reasons? Philosophical Issues, 15, 321–337. ———. (2007). Does rationality consist in responding correctly to reasons? Journal of Moral Philosophy, 4(3), 349–374. Burge, T. (1996). Our entitlement to self-knowledge. Proceedings of the Aristotelian Society, 96, 91–116.

44 The self-image of animal rationale is perhaps too flattering—though the regress still forces us to defend the more modest self-image of rationis capax (cf. Letter from Jonathan Swift to Alexander Pope, 29 Sept. 1725). 45 The categorical interrogatives raised here pertain mostly to value theory and epistemology (“What should I do?;” “What should I believe?”). So admittedly, the argument of this appendix directly concerns the value of only those two subdisciplines. It can partly justify other areas of philosophy, insofar as they serve the first two, such as the epistemology of self-knowing, as we saw. Regardless, this still leaves a lot of philosophy out in the cold. To repeat, I would not defend all of philosophy (perhaps not even most philosophy). Nevertheless, there are other philosophical inquiries which deserve justification; see again section 1 of Parent (2015) as a partial remedy; see also Pitt (2016) for another excellent Sellarsian angle. But this is also something I hope to address further in future work.

Preamble: Is Philosophy Anti-Scientific?  33 Carnap, R. (1931). Die physikalische Sprache als Universalsprache der Wissenschaft. Erkenntnis, 2(1), 432–465. Dennett, D. C. (2013). Intuition pumps and other tools for thinking. New York: W. W. Norton. Doris, J. (2015). Talking to our selves: Reflection, ignorance, and agency. Oxford UP. Haack, S. (2005/2013). Formal philosophy? A plea for pluralism. In her Putting philosophy to work: Inquiry and its place in culture (pp. 235–250). Amherst, NY: Prometheus Books. Hinchman, E. (2013). Rational requirements and “rational” akrasia. Philosophical Studies, 166, 529–552. James, W. (1896). The will to believe. The New World, 5, 327–347. Kitcher, P. (1989). Some puzzles about species. In M. Ruse (Ed.), What the philosophy of biology is: Essays dedicated to David Hull (pp. 183–208). Kluwer. Kornblith, H. (2012). On reflection. Oxford UP. Korsgaard, C. (2009). The activity of reason. The Proceedings and Addresses of the American Philosophical Association, 83(2), 23–43. Kripke, S. A. (1982). Wittgenstein on rules and private language. Harvard UP. Ladyman, J. & Ross, D., w/ Spurrett, D. & Collier, J. (2007). Everything must go: Metaphysics naturalized. Oxford UP. Max-Neef, M., w/ Elizalde, A. & Hopenhayn, M. (1989). Human scale development: Conception, application and further reflections. New York: Apex. McDowell, J. (1984). Wittgenstein on following a rule. Synthese, 58, 325–364. Moran, R. (2001). Authority and estrangement: An essay on self-knowledge. Princeton UP. Parent, T. (2015). Theory dualism and the metalogic of mind-body problems. In C. Daly (Ed.), Palgrave handbook to philosophical methods (pp. 497–526). Basingstoke, UK: Palgrave. Pitt, J. (2016). The future of philosophy: A manifesto. In M. E. Franssen, P. Vermaas, P. Kroes & A.W.M. Meijers (Eds.), Philosophy of technology after the empirical turn (pp. 83–92). Springer. Putnam, H. (1975). The meaning of “meaning”. In his Mind, language, and reality: Philosophical papers vol. 2 (pp. 215–271). Cambridge UP. Quine, W.V.O. (1948). On what there is. Review of Metaphysics, 2, 21–38. Reprinted in his (1961) From a logical point of view: Nine logico-philosophical essays, 2nd edition (revised). Harvard UP, pp. 1–19. ———. (1960). Word and object. MIT Press. Rescher, N. (1987). Forbidden knowledge. Dordrecht, Netherlands: D. Reidel. ———. (2014). Metaphilosophy: Philosophy in philosophical perspective. Lanham, MD: Lexington Books (Rowman & Littlefield). Rinard, S. (forthcoming). No exception for belief. Philosophy and Phenomenological Research. Rosenberg, J. (1999). How not to be systematic: Three case studies. Proceedings of the Twentieth World Congress of Philosophy, 1999, 217–225. Russell, B. (1912). The problems of philosophy. New York: Henry Holt and Company. Searle, J. (1965). What is a speech act? In M. Black (Ed.), Philosophy in America (pp. 221–239). London: Allen and Unwin. Sellars, W. (1960). Being and being known. Proceedings of the American Catholic Philosophical Association, 35, 28–49. Reprinted in his (1963b), pp. 41–59.

34 Preliminaries ———. (1962). Philosophy and the scientific image of man. Reprinted in his (1963b), pp. 1–40. ———. (1963a). Phenomenalism. In his (1963b), pp. 60–105. ———. (1963b). Science, perception and reality. London: Routledge and Kegan Paul. ———. (1971/1975). The structure of knowledge. In H. Castañeda (Ed.), Action, knowledge and reality: Studies in honor of Wilfrid Sellars (pp. 295–347). Indianapolis: Bobbs-Merrill. Sharadin, N. (2016). Nothing but the evidential considerations? Australasian Journal of Philosophy, 94(2), 343–361. Shoemaker, S. (1996). The first-person perspective and other essays. Cambridge UP. Shpall, S. (2012). Moral and rational commitment. Philosophy and Phenomenological Research, 88(1), 146–172. Sosa, E. (2011). Knowing full well. Princeton, UP. Strawson, P. (1962). Freedom and resentment. Proceedings of the British Academy, 48, 187–211. Tay, L. & Diener, E. (2011). Needs and subjective well-being around the world. Journal of Personality and Social Psychology, 101(2), 354–365. Toribio, J. (2011). What we do when we judge. Dialectica, 65, 345–367. Verbeek, B. & N. Southwood (Eds.) (2009). Practical reasoning and normativity [special issue]. Philosophical Explorations, 12(3). White, A. (1975). Modal thinking. Cornell UP.

1 Introduction How Is Rational Self-Reflection Possible?

Back in Catholic elementary school, reflection was mandatory. Time was set aside for regular prayer (to reflect on our blessings and shortcomings), and we were taught that the Watchful Eye of God was always upon us. The latter encouraged a certain habit of self-monitoring, to gauge how one’s thought and action appeared from His point of view. Thus: “Catholic guilt,” the feeling that whatever good one can do, one ought to do. I do not necessarily recommend raising children this way. Regardless, Catholics are right that reflection is an important part of our intellectual and moral lives. Philosophers are generally sympathetic to this. Even so, philosophers have rightly abandoned Cartesian reflective inquiry as hopeless. It is not as if apriori philosophical reflection could justify science (or empirical inquiry in general). If anything, science reveals how unjustified philosophy is. Nonetheless, many of us believe that critical reflection has some benefit.1 It is typically thought that such reflection can expose skewed priorities, inconsistencies, non-sequiturs in reasoning, and so forth. The practice is prized in many academic disciplines, but it seems especially central to philosophy—and philosophy, I would argue, serves critical reflection particularly well.

1.1  The Problem of Wayward Reflection Nevertheless, anyone who champions self-reflection2 faces a grave ­difficulty. The activity presumes epistemic access to one’s own thoughts: I cannot

1 Still, there is empirical evidence against the benefits of self-reflective inquiry. See, e.g., Wilson & Schooler (1991), Wilson et al. (1993), Wilson & Hodges (1993), Wilson & Kraft (1993), Damasio (1994), Wilson & Dunn (2004), Epley & Gilovich (2005), and Halberstadt & Wilson (2008). Kornblith (1999, 2002, ch. 4; 2010; 2012; 2013; 2016) is one who appeals to such evidence to ask pointedly why reflection should be such a revered practice. But while these doubts are quite important, I am unable to address them in the present work. Yet see Tiberius (2008; 2013) for a response to such doubts. 2 How does reflection differ from self-reflection? I often use the two terms interchangeably, but strictly speaking, only the latter directly targets one’s own beliefs. Thus, not all reflection

36 Preliminaries critically reflect on my thoughts in ignorance of what those thoughts are. But unfortunately, there are strong reasons to say that each of us is largely ignorant of our own thoughts. For one, many argue that content externalism (sometimes called anti-individualism3) precludes knowing what one thinks just while seated “in the armchair” or without empirical investigation.4 Also, there is a second reason, less discussed in the philosophical literature. Experimental psychology indicates that we are ignorant, to a large degree, about our own thoughts—especially those thoughts that act as our reasons for action or belief.5 Yet such widespread self-ignorance would confound self-reflection. One would be ignorant of the thoughts one is supposed to reflect on. These two points—re: externalism and experimental data—each motivate what I call “The Problem of Wayward Reflection” (cf. Parent 2007, p. 421). Confined to the armchair, how can we reliably discern our own thoughts, rather than just confabulating them? If the answer is “we can’t,” then armchair self-reflection makes no sense. If one cannot reliably know what one’s reasons are, then so much the worse for the hope to evaluate them.6

is self-reflection proper—I can reflect on the beliefs of others, e.g., our political leaders. (Yet such reflection still operates on my beliefs about what the other person believes.) That may suggest, however, that self-reflection proper takes a metacognitive form, e.g., “I believe that social equality is just; therefore . . . ” But not so: It can just feature my first-order beliefs: “Social equality is just; therefore . . . ”. Still, it counts as self-reflection only if they are understood as my beliefs. (If I see them just as Obama’s beliefs, there is a clear sense in which I am not self-reflecting, even if I happen to share his beliefs.) 3 See Putnam (1973; 1975), McGinn (1977), Stich (1978), Burge (1979; 1982; 1986). Burge prefers the term ‘anti-individualism’, yet that thesis is not really about content at all. (See Burge 2006.) In conversation, Burge clarifies that he holds Frege’s view where contents exist as abstracta. Thus, anti-individualism more concerns individuating one’s concepts rather than their contents. But Fregean contents are not assumed here; thus, I avoid ‘anti-­ individualism’, since the relevant non-internalist thesis is not really Burge’s. 4 At times, the book vacillates between talk of “armchair,” “apriori,” and “introspective” selfknowledge. “Armchair” self-knowledge covers the other two, though it is not exhausted by them. Indeed, in Part III, I discuss a type of self-knowledge which, although attainable from the armchair, is somewhat empirical. (This is “expressive” self-knowledge; it is achievable in linguistically expressing one’s judgments.) 5 For a list of pertinent works in experimental psychology, see n. 1 of chapter 2. N.B., I assume throughout that reasons for belief are themselves beliefs. (Yet I am neutral on whether reasons qua reasons are causes.) An objection: If I judge that the mail carrier has not arrived yet, my reason is not that I have the belief that the little flag on the mailbox is still raised. Rather, it is the fact that the flag is still raised. My preference, however, is to say that the latter is a piece of evidence, re: the mail carrier’s arrival. My access to this evidence still consists in a belief, and that is all I mean when speaking of my reasons as beliefs. 6 A Kantian ethicist should be doubly concerned; knowing her reasons is also needed to morally assess her own acts. (For the Kantian, the worth of her act depends on her “maxim.”) Ignorance here would thus force ignorance of, e.g., to what extent the “dear self” drives her act. (Further still, Lehrer 1990 argues that access to one’s beliefs is key to many phenomena of philosophical interest—including free action, rational acceptance, and even intentionality itself.)

Introduction  37 The basic Problem can be made vivid by an analogy. Suppose neuroscience has advanced to the point where we can use a “brain scanner” to detect what your thoughts are. When at peak performance, the scanner is even sensitive enough to detect thoughts that are otherwise non-conscious. Suppose, then, that for several months your therapist has used the scanner to identify a number of thoughts, and has focused your therapy toward acknowledging and evaluating them. But suppose it is discovered that the scanner has been malfunctioning for several weeks—and that the thoughts reported included a lot of false positives. Now although s/he is alarmed, suppose your therapist recommends continuing with this method anyway, as if nothing were wrong. Wouldn’t you find this laughable? The thoughts dictated by the scanner might be interesting to think about independently, but they may be irrelevant to your therapeutic goals. You may end up dealing with thoughts that have no relevance to your aims and interests. The Problem of Wayward Reflection suggests that we are actually in this dire relationship to our own mind. But the difference is that each person is dealing with a deficient internal scanner as it were. To repeat, externalist arguments suggest that armchair methods cannot enable discriminations between one’s thoughts and any number of “twins.”7 This seems to imply that one might regularly and unwittingly equivocate during armchair reflection, due to a thought being “switched” with a twin (see Boghossian 1992a, b, 1994, 2010, 2014.). Call this “Wayward Reflection via Thought Switching.” Further, studies show not just that we are bad at introspecting our own ego-threatening beliefs à la Freud, e.g., “My mother never loved me.” The data also reveals that people are bad at introspecting ordinary sorts of beliefs—including one’s reasons for moral judgments of others (Haidt 2001), and even mundane reasons for buying one pair of stockings over another (Nisbett & Wilson 1977)! Thus, a person might identify “p” as one of her beliefs during armchair self-reflection, even though she does not really believe p—and perhaps even denies p. The conclusions she infers then may deviate from what should be inferred from her beliefs. And thus, reflective reasoning will take her further, rather than nearer, to

7 Though it is little noted, the externalism/self-knowledge debate reflects a historically important clash between philosophical paradigms before and after the linguistic (/conceptual) turn. On one hand, Descartes held that one can know one’s thoughts independently of knowing the external world. On the other hand, Frege held that content fixes reference (or in Carnap’s terms, that intension determines extension). Yet as Putnam (1975, p. 218ff.) argues, Frege’s idea seems to lead to externalism: If content determines reference, then a difference in reference between Oscar and Twin Oscar shows a difference in content. That holds, even if narrow psychological states are the same. So: If Frege’s view implies externalism and precludes Cartesian self-knowledge, then apparently one of these paradigm-defining ideas has to go. Either Descartes was wrong to say we can know contents apriori, or Frege was mistaken that content determines reference. The externalism/self-knowledge debate thus commands our serious attention.

38 Preliminaries appreciating what she ought to believe. Call this “Wayward Reflection via Attitude Switching.” Both versions of the Problem suggest we self-scrutinizers are all in “epistemic bad faith” given what psychosemantics and experimental psychology suggest. Again, critical self-reflection presupposes knowing what one’s beliefs are. So if the mind is largely opaque to itself, why attempt selfreflection? The activity would just show naivety about psychology. Granted, I think there are cases of reflection that are rather sad, due to misjudging one’s own beliefs or desires. I might even be persuaded that in most cases, argument amounts to confabulating reasons post hoc, to rationalize whatever one has said or done.8 And so, what was supposed to be “a way of screening our beliefs in order to make them more accurate turns out, instead . . . to be little more than self-congratulation” (Kornblith 2012, p. 3). Even so, my hope is to show that self-reflection can be a reasonable activity, and that it is reasonable to believe as much. It is to show that despite findings from psychology, self-reflection can be a rational pursuit.9

1.2  Methodological Crisis in Philosophy and Psychology There is a special case of the Waywardness issue worth singling out, of particular concern to contemporary philosophers. These days, there is much discussion about intuitions in philosophical methodology. The usual view is that philosophers rely on intellectual “seemings” about, for example, the workings of language, the metaphysics of objects, the ethics of cloning, etc. to arrive at considered philosophical positions. Indeed, the prominent role of intuitions is (arguably) what distinguishes philosophical inquiry from the usual sort of scientific inquiry.10 Now there is some debate on whether intuitions play the role of evidence in philosophy (see Cappelen 2013). But at

  8 I have even seen instances of the Dunning-Kruger effect re: self-knowing. A person claims to be acutely self-aware, yet this ends up being another example of her/his self-ignorance. (A puzzle: How do I know I am not such a person?)   9 This is not to endorse what Doris (2015) calls “reflectivism,” the view that moral agency requires deliberation on one’s act. I agree with Doris that much less is required. (See also Kornblith 2012.) Yet even if moral agency is not under fire, Wayward Reflection is still worrisome, assuming it deprives us of some notable moral/epistemic benefits. 10 There is a case to be made that intuitive judgments are ubiquitous in science also. Williamson (2009) says as much, in discussing Weinberg’s (2009) empirical data against the reliability of intuitions: “Weinberg’s paper is itself full of informal qualitative epistemological judgments, for example about whether we are justified in believing that armchair methods in philosophy are reliable. Nor could any current natural science proceed without such judgments. Even statistical data need to be interpreted; the judgment that they render some hypothesis untenable remains an informal, qualitative one, whatever formal and quantitative considerations it draws on” (p. 474). I doubt the worry about intuitions is comparable for science as for philosophy, yet the issue indeed exists for science to a notable degree.

Introduction  39 the least, inquiry has to begin somewhere, and if “intuitions” just concern how things seem at the start of inquiry, intuitions are where we must start. (However, this certainly does not mean they have axiomatic status or the like.) Yet “intuitions” sounds suspicious. Why think one’s intuitions are at all reliable in discovering philosophical truths? (For a good overview of the issue, see Nagel 2007.) Wherever you fall on this, however, there is a prior matter—one which is underappreciated in the intuitions literature. Namely, in light of current psychosemantics and experimental psychology, how could a philosopher know what her intuitions are to begin with? Nevermind reliability—the more basic question is whether a philosopher can recognize her intuitions in the first place.11 The presumption that the philosopher knows her own intuitions may just show naivety about psychology. So Waywardness does not concern merely the type of self-scrutiny that (hopefully) everyone engages in. It also concerns the type of reflection utilized in professional philosophy. But before moving on, let me note a further motive for confronting Wayward Reflection. This should be of particular concern to psychiatrists and professional psychologists, and shall be explored in more detail in chapter 2. Yet briefly, the problem is that experimental method within psychology seems to have the same naivety about the opacity of mind! Very often, experimental design in psychology gives central importance to subjects’ self-reports of mental states. Participants in an experiment are first subjected to some condition, and then are asked to describe something about their own mental state. For example, a subject is asked what she thinks when she sees the ink-blot, what she believes of the person shown in the photograph, or what she judges to be the best decision in a hypothetical situation. And standardly, the method presumes the subject is able to discern what she herself thinks, believes, or judges in the first place. Yet here is where the Problem arises. In a bit of irony, the evidence gathered by that very method suggests subjects are not reliable in identifying their own mental states. Thus, there seems to be a crisis of methodology in experimental psychology. The method used to gather psychological data rests on an assumption which the self-same data falsify!12 Again, this shall be discussed more thoroughly in chapter 2, where we will look in detail at specific psychological studies, to see how this plays out

11 Relatedly, there is a concern about “conceptual analysis:” Nevermind whether conceptual analysis is a reliable method—the prior question is whether one can know about one’s own concepts in the first place. 12 By the way, what goes for psychology also goes for experimental philosophy. There too, experimental design gives central evidential weight to self-reports of beliefs, desires, etc. In a further twist, Wayward Reflection also bears on the experimental brand of philosophy that tries to avoid problems with intuitions.

40 Preliminaries exactly. For now, the point is just that the opacity of mind may present a serious problem for methodology in psychology as well, besides philosophical methodology and self-reflection generally.

1.3  The Start of an Answer The book was written to confront these matters. The positive claim will be that, under some recognizable conditions, we can indeed armchairknow our own minds. In fact I will develop the radical view that, in some identifiable circumstances, we are infallible about our own occurrent beliefs—judgments about what we judge cannot possibly be false! Even so: I reject the Cartesian idea that higher-order judgments are indubitable. (I’m inclined to think the range of what can be doubted is limitless.) Relatedly, I endorse Quine’s (1951) view that any belief is subject to empirical revision, if the evidence available warrants it. Even so, if some reflective judgments are de facto infallible (and we can recognize when that is so), then that can go a long way toward quelling anxiety about Wayward Reflection. Before infallibility is dismissed as a Cartesian pipe dream, however, here is a quick and dirty argument for why it has a chance. Consider that we all accept one type of infallibility thesis: (INF0) Necessarily, if a subject S judges that S is in mental state M, and that judgment is true, then S is in M. This of course is trivial. Nonetheless, (INF0) highlights two points. First, it suggests that infallibilism is true if it is restricted to the right subset of second-order judgments. Second, and in light of that, it prompts the question whether this subset—or a subset of that subset—can be identified in some other, substantive way. It is not obvious that the answer is “no.” And if the answer is “yes,” then a non-trivial infallibilism holds. There will be objections of course. But this argument is just meant to soften up the reader. Let me also repeat that the motivation is not to secure a foundation for empirical knowledge. Again, that sort of Cartesian endeavor seems futile. Rather, infallible self-knowledge is sought in order to counter doubts about reflection—and in particular, about self-reflective critical thinking. It is meant to vindicate the practice by suggesting that, in some recognizable circumstances, reflection infallibly targets one’s own occurrent thoughts and beliefs. An important caveat: The infallibilism to be defended is not seen as an exhaustive view of all self-knowledge. On this, I instead align with Schwitzgebel’s (2012) pluralism, which holds that most contemporary theories of self-knowledge are right about some self-knowing phenomena. (See also Coliva 2016.) Schwitzgebel writes:

Introduction  41 [Self-knowing]13 is not a single process but a plurality of processes. It’s a plurality both within and between cases: Most individual [selfknowing] judgments arise from a plurality of processes (that’s the withincase claim), and the collection of processes issuing in [self-­knowing] judgments differs from case to case (that’s the between-case claim). (p. 29) But despite such pluralism, the book shall focus on only one type of selfknowledge, viz., the allegedly infallible type. Yet this should not suggest an anti-pluralist stance. I concur that many self-knowing processes can be operative in a given instance (the within-case), and that other self-knowing judgments owe to other processes (the between-case).14 Nevertheless, I focus on infallibility mechanisms since they yield a particularly striking kind of self-knowledge, one which provides the strongest antidote to Wayward Reflection. 1.3.1  Opening Objections There may be a few basic worries about the project as a whole. For one, it can be unclear how any sort of infallibilism can be made empirically respectable.15 Empirical data shows us only what is the case, not what is necessarily the case. So how can an evidence-based approach defend the infallibility of some judgments? The question is understandable, but fortunately the reply is straightforward. Empirical data can indeed justify nomological ­necessities— including, perhaps, that the occurrence of some judgments nomologically necessitates their truth. Granted, this would not make for the Cartesian infallibility of a judgment, defined as truth in all epistemic/doxastic possible worlds (or the like). But as we shall see, an empirically minded philosopher can support that some judgments are true whenever executed, as a matter of psychological law. This would make them “infallible” in one sense, albeit not in the Cartesian sense.

13 Schwitzgebel speaks of “introspection” rather than self-knowing as such. But he soon incorporates, e.g., “agency-theoretic” and “expressivist” views into his pluralism, even though such views avoid talk of introspection. (See Moran 2001, Bar-On 2004, etc.). (Still, introspection in Schwitzgebel is not the usual sort; it more resembles Evans-style transparency. See Schwitzgebel 2011, p. 137 along with the attending endnotes.) 14 Indeed, Part III is rather in line with Carruthers’ (2011; 2015) view that self-knowledge is based on interpreting one’s own behavior (along with data from proprioception, interoception, etc). Yet while I think Carruthers is right in many ways, I do not think his view respects all self-knowing phenomena. I discuss his view more critically in chapter 2. 15 My thanks to an anonymous referee for raising this concern.

42 Preliminaries A different objection is that the project is undermotivated. Some may feel that reflection holds enough intrigue, whether or not it targets one’s own beliefs. For example, a philosopher may be undecided on modal fictionalism, but she can be motivated to engage the arguments regardless. This of course is quite right. Yet for one, she would likely assume some armchair self-knowledge, viz., that she regards fictionalism as a live option at least. But more importantly, if reflective inquiry is to have anything more than entertainment value, it must be rationally motivated and not just accidentally amusing. In particular, it must be reasonable to believe that reflection can achieve its goals, including that of self-evaluation. So we need reason to think that, at least in some recognizable circumstances, the beliefs reflected upon are indeed one’s own. The endeavor of the book also faces a kind of self-referential objection. My plan is to use armchair reflection to defend armchair reflection. But this seems viciously circular. This can be elaborated in two ways (corresponding to the two versions of the Waywardness Problem). First, if content externalism is true, then my reasoning may equivocate in ways that I cannot detect from within the armchair. Thus, premises about reflection may be used to infer conclusions about “twin-reflection” without my realizing it. This may strike you as idle skepticism, however. Yet I do not wish to be so dismissive toward the objector. But the matter cannot be settled quickly, and it is best addressed in chapter 6, where I discuss externalism and equivocal reasoning more broadly (see esp. the final footnote of the chapter). The second version of the self-referential charge is that, per experimental psychology, I cannot know whether my stated assumptions on critical reflection capture my own judgments.16 But to this, I say “fair enough.” I will proceed in ignorance of whether various claims express what I believe.17 I shall even feign neutrality on whether a view strikes me as a live possibility. (I will just be agnostic on why I find some proposition worth thinking about.) Fortuitously, I am still interested enough in the issues that I don’t need a rationalizing motive to get started. Though again, if reflective inquiry is to have anything more than entertainment value, the hope is that reflective inquiry can be rationally vindicated. And in attributing this hope to myself, I am perhaps begging a question. But my reasons for interest in the issues can be separated from my arguments on the first-order issues themselves. So there need not be any question-begging at the ground level. (I may inevitably beg a question concerning why I should care about the arguments. Yet that is not to say that the arguments themselves beg.)

16 Such circularity, I take it, is closely related to Kornblith’s (2012, ch. 1) regress problem. 17 I may occasionally speak of “what I believe” for rhetorical or stylistic reasons. But the point is that no argument will depend on any premise about what I, TP, judge to be one of my thoughts/judgments.

Introduction  43 *1.3.2  Internal- and External-World Skepticisms But admittedly, there remains a more difficult aspect of the self-referential objection. Consider: If my modus operandi is largely philosophical, then per section 1.2, I apparently must draw upon philosophical intuitions in reasoning about various claims. And that requires recognizing what my intuitions are. Yet the ability to recognize one’s own intuitions is partly what’s in doubt. So if the project of the book requires such an ability, then there indeed seems to be a worrisome circularity built into the endeavor. Of necessity, the reply here is more programmatic—for it is meant to counter a more programmatic kind of skepticism. Concurrently, as an expository tactic, it proves useful to consider briefly the problem of externalworld skepticism. In an ordinary debate, a question about question-begging can be perfectly apt; for ordinarily, the arguments on each side should start from common ground. But things take an odd shape when debating the external-world skeptic. One such skeptic, by definition, concedes no ground about the external world at all. The shared ground is thus invariably insufficient for vanquishing him. So, simply by being sufficiently uncooperative, the skeptic remains undefeated. Mooreans thus ignore this skeptic, insisting that anyone that skeptical is intellectually disingenuous, if not insane. (And one should ignore the disingenuous or insane!) Regardless, our skeptic can teach us something. The lesson is apparently that one cannot have reasons for beliefs about the external world, without making presumptions about the external world. Let us suppose, nonetheless, that this does not threaten our knowledge of the external world. Still, the skeptical lesson at least raises an explanatory challenge for, e.g., cognitive science. Namely: Given that we know the external world, partly on the basis of mere presumptions, how do we manage to do that? By what means are we able to know in this presumptuous manner? In focusing on this explanatory question over traditional skepticism, I show an allegiance to Quine’s (1969) naturalized epistemology. Yet in the present context, Quine’s program is understood a bit differently than is typical. After all, the explanatory question here is not answered by studying the neurology of the brain or the various perceptual mechanisms (though that may be relevant). For what animates the question is a puzzle about our reasons for belief, and reasons do not show up on an fMRI scanner (at least, not qua reasons). The puzzle again is how we can know about the world despite our reasons being oddly ungrounded or circular. To repeat, it is taken for granted that we do know; the question is merely how, given the structure of our reasons.18

18 I take this to be a puzzle only about reflective knowledge, not about so-called animal knowledge. Cognitive neuroscience alone might very well explain animal knowledge. (The

44 Preliminaries For my part, Sellars (1956/1963) answered this in a single, brilliant brushstroke: “[E]mpirical knowledge, like its sophisticated extension, science, is rational, not because it has a foundation but because it is a self-correcting enterprise which can put any claim in jeopardy, though not all at once” (p. 170, his italics). Here, the “foundation” is basically the common ground that we would share with the skeptic. It would consist in certainties, beliefs that one is unable to doubt regarding the external world. The explanatory puzzle arises since we tend to assume that, to know the world, we need noncircular reasons starting from such common ground. But by definition, our skeptic does not concede enough ground for that. Thus, the puzzle. Sellars teaches us, however, that this foundationalist conception of knowledge is a mistake. We can suppose that knowledge is true rational belief, yet rational belief need not be skeptic-immune belief. It is instead enough if the belief is situated within a “self-correcting enterprise,” a system of beliefs where nothing is adopted dogmatically. Self-correction means “checking” one belief against other beliefs, presumably by how well the one coheres with the others. Accordingly, if a belief creates enough conflict, it is removed. The proposal, then, is that such coherentism is the correct account of what makes a belief “rational,” and shows how true rational belief possible.19 Sellars’ idea still leaves some anxiety. Most notably, it leaves us with (what Bonjour 1985 calls) the “alternative coherent systems” objection. The objector notices that, in principle, vastly different systems could each count as having fully justified or rational beliefs as members. If all that’s required is a certain kind of coherence among the members, that hardly picks out a unique system. This, consequently, raises the concern that a belief’s being “rational” is ultimately not a truth-conducive property—it is unclear whether that feature increases the objective likelihood of a belief being true.20 However, the intuition is that justification must render a belief

distinction is from Sosa 1985, 2011, etc., yet I would revise Sosa’s way of drawing the distinction a bit, in light of sound criticism from Kornblith 2004, 2016.) 19 On coherentism, see Quine & Ullian (1970/1978), Sellars (1973), Rescher (1973; 1979), Lehrer (1974; 1989a, b; 2000; 2003), Bonjour (1985; 1989), Harman (1986), and Lycan (1988; 1996; 2002; 2012). See also Foley (1987), who explicitly speaks of rational rather than justified belief. I first became worried about Wayward Reflection in thinking about Bonjour’s (1985, p. 101) “Doxastic Presumption.” Yet strictly speaking, the Presumption concerns an individual’s grasp of the entire belief-system, and such global coherentism seems unattractive (see n. 21). But localized coherentism would still need to contend with Waywardness. By the way, “checking” a belief against other beliefs admittedly engages in the circularity as well, re: self-knowing. For the process requires knowing what these other beliefs are. (The circle also exists in knowing your intuitions about what coheres with what.) Ultimately, the reply to internal-world skepticism is meant to address all instances of the circle. 20 The most powerful critique of this sort is from Klein & Warfield (1994).

Introduction  45 more objectively probable. After all, we want “justification” to corroborate our choice of belief, vis-à-vis the goal of attaining truths and avoiding falsities. Without going into detail, the objection is essentially stating the underdetermination of theory by evidence, as made famous by Duhem and Quine. Quine (1951) puts the matter thus: The totality of our so-called knowledge or beliefs . . . is a [hu]manmade fabric which impinges on experience only along the edges. . . . A conflict with experience at the periphery occasions readjustments in the interior of the field. . . . But the total field is so undetermined by its boundary conditions, experience, that there is much latitude of choice as to what statements to re-evaluate. . . . No particular experiences are linked with any particular statements in the interior of the field, except indirectly through considerations of equilibrium affecting the field as a whole. . . . Any statement can be held true come what may, if we make drastic enough adjustments elsewhere in the system. . . . Conversely, by the same token, no statement is immune to revision. (pp. 39–40) It may overstate matters to say that the whole system is implicated when encountering recalcitrant experience (cf. Quine 1991).21 However, it is widely accepted that our theories of the world are undetermined by the evidence, along the lines Quine indicates. So it seems the “alternative coherent systems” objection is not to be resisted as much as admitted, as an important philosophical truth about our epistemic position. Yet the objection stands that a belief’s being “rational” now fails as a truth-conducive property. In reply, however, I follow Quine, who construes the rationality of belief not by its connection to truth, but rather in pragmatic terms: Each man [sic] is given a scientific heritage plus a continuing barrage of sensory stimulation; and the considerations which guide him in warping his scientific heritage to fit his continuing sensory promptings are, where rational, pragmatic. (Quine 1951, p. 43)

21 Relatedly, at no point do I say that to properly acquire beliefs, the epistemic agent must be guided (consciously or not) by considerations of global coherence. I concur with Kornblith (1989) that such things require a computational complexity that is unrealistic. Even a coherentist “check” of a belief can operate more locally, using only a proper subset of domain-relevant beliefs. (This is what Kornblith 1989 himself accepts in his Quine-style view of confirmation; see p. 212.)

46 Preliminaries Pragmatic desiderata include maximizing “super-empirical virtues” like simplicity, conservativeness, fertility, scope, etc. Such concerns can guide a rational decision among “alternative coherent theories,” even though the superempirical virtues might not always be truth-conducive. Yet if they allow for rational theory-choice regardless, knowledge again seems possible, despite the circular structure of our reasons. And the point would remain that knowledge does not require a skeptic-immune foundation, but rather just a place in a system that implements coherentist self-correction, where the rationality of the whole system ultimately has to do with its pragmatic advantages. What then of truth and falsity? “Justification” that is not truth-conducive can sound like a contradiction in terms. But this may just indicate we should follow Sellars and Quine in speaking of “rational” rather than “justified” belief. Still, if the rationality of a belief is largely orthogonal to truth, then our means to maximizing truth and minimizing falsity seems not very effective. (Prisoners of The Matrix can just as easily achieve fully “rational” belief, despite being systematically mistaken about the outside world.) However, rational belief is not wholly independent of truth, since consistency is the minimum required on a true theory. (I would argue that that ontological parsimony, judiciously applied, helps us avoid falsity as well. Perhaps the success of science also assures us that we are not entirely off track.) Regardless, a less-than-effective means to truth may be our lot in life, given the ungrounded or circular structure of our reasons. And perhaps truth is not the goal of inquiry as much as a regulative ideal. The more immediate goal, apparently, is to maximize coherence and the super-empirical virtues.22 The preceding sketch is just a sketch; even so, it contains much that is controversial. However, I cannot pursue the debates at this time. The Quine-Sellars view at least offers a minimally tenable way for responding to external-world skepticism (although it is admittedly fairly concessive). Nevertheless, if knowledge is a species of true rational belief, it still reveals how knowledge of the external world is possible, so that despair is not inevitable. My suggestion, furthermore, is that the view also indicates a minimally viable tactic against skepticism about the internal world. Our internalworld skeptic is similarly pointing out a circularity in our reasons—that one must presume access to one’s intuitions to show how it is possible to access (inter alia) one’s intuitions. But if we extend the Quine-Sellars view, this circularity need not preclude that some beliefs remain more rational than others, insofar as some beliefs better cohere with the others (in a self-checking

22 This may lead some to conclude that maximizing truth and minimizing falsity cannot really be our goal in inquiry. The matter likely depends on what “the goal of inquiry” means exactly. But there is a sense, I think, in which inquiry typically has truth as a goal (perhaps inter alia). Though whether that should be a goal is yet a further question which I do not address here.

Introduction  47 system). For coherence makes for rational belief at least in a pragmatic sense of ‘rational’, and also in a sense not wholly orthogonal to truth. Just to be clear, coherentism was not invoked to explain reasons for belief in the “context of discovery.” Typically, when I recognize that I am hungry, my reasons do not relate to how well “I am hungry” coheres with my belief-system. The discovery is usually instead based on introspective evidence (yet per Schwitzgebel’s pluralism, it can be based on other things as well). Instead, I appealed to coherentism only in a specific “context of justification,” namely, that of combating internal-world skepticism.23 (Though coherence likely plays a role in other contexts as well.) We must acknowledge an important disanalogy between internal- and external-world skepticism. Namely, the internal-world skeptic has non-idle reasons for doubt, viz., reasons gleaned from psychology. Thus, compared to the external-world skeptic, the internal-world skeptic has a better case for suspending judgment about her domain of interest. This is why, as in chapter 2, I call her not a “skeptic,” but rather an “empirically-based antagonist” toward self-knowledge or knowledge of the internal world. (Though n.b., in chapter 2, antagonism concerns only a particular kind of self-knowledge, viz., the supposedly infallible kind. But see chapters 10 and 11 on antagonism more generally.) Thus, even if the Quine-Sellars view shows how one can know the external world, does it adequately buttress the possibility of self-knowledge? There is a mistaken presupposition in the question. Strictly speaking, it is not simply coherentism-cum-pragmatism that shows how we might have external-world knowledge. If we are being careful, that view just sets the terms on what would count as such knowledge. But it does not yet follow that our beliefs satisfy those terms. An analogy: ‘x is round and x is square’ defines what it is to be a “round square,” but it does not follow that some object satisfies that definition. In the case of the external-world knowledge, of course, it is natural enough to say that our beliefs often do satisfy those conditions. In contrast, however, beliefs about one’s internal world may not cohere adequately with other beliefs, specifically, beliefs about psychological findings. So in brief, when antagonism’s “evidence-based” feature is stressed, it is unclear if beliefs about the internal world qualify as knowledge. Nevertheless, the psychological results are not a direct challenge to the conditions themselves. (For that, normally one offers a counterexample or the

23 ‘Justification’ as it appears in ‘the context of justification’ is used to denote a process (“the game of giving and asking for reasons”). This, by the way, affords a clear distinction between the two types of context: When acquiring new evidence, one can discover p (= know p for the first time) sans a process of justifying p to someone else. Yet continuing to know p may depend on your ability to quell doubt in subsequent contexts.

48 Preliminaries like.) The Quine-Sellars idea per se is not affected by stressing the empirical difficulties. But no doubt, the results from psychology feed antagonism toward selfknowledge. Again, knowing these results, our beliefs about the internal world seem not to cohere sufficiently with the rest of the system. The coherentist is thus lead to ask about the rationality of these beliefs. The eventual goal, however, is to show that hypothesizing self-knowledge is not just permissible, but also can be made rationally respectable, despite what psychology tells us. So although I cannot cut off antagonism at this point, that will be the ultimate aim in what follows. 1.3.3  Epistemic Internalism? There is a follow-up objection, however, and it is a broader concern for the book as a whole. The sense may be that epistemic internalism is being assumed, a view where knowledge requires evidential reasons to be introspectively accessible. After all, the previous sub-section favors a kind of coherentism (which is usually typed as internalist)—and besides, the book’s central concern is with armchair self-knowledge. However, epistemic internalism is highly controversial. Apparently, it precludes reliabilist epistemology (cf. Goldman 1986), along with various kinds of social epistemology currently on the scene (cf. Goldman 1999; see also the papers in Goldman & Whitcomb 2011). Yet I myself am a social epistemologist. To me, it is obvious that even Descartes needed peer-feedback on The Meditations before he could see it as establishing anything. (The isolated individual is just too unreliable, especially in philosophy.) Thus, I hardly mean to presuppose epistemic internalism.24 But the subsequent chapters do not assume epistemic externalism either. Neutrality on the matter is possible since, surprisingly, the book is largely unconcerned with the evidential reasons that enable self-knowledge. In this respect, ‘self-knowledge’ may be misleading. After all, the book’s basic contention is that specific kinds of second-order judgment are infallible. That is to say, these judgments have a distinctive semantic ­feature— they are invariably true whenever executed. Given that the thesis is semantic, one could see the book as more a study in the philosophy of mind than in epistemology. (That is so, especially since the ancillary themes are

24 Still, I see the appeal of internalism. Per the Preamble, the individual faces “alone” the question “What should I believe?,” and internalism seems apt in that setting. Given such epistemic responsibility, I thus balk at Kornblith’s (2013) verdict that the internal perspective “is wholly underserving of the special place which . . . philosophers would give it” (p. 126, my italics). Yet if the individual is responsible for answering the question, her methodology of answering need not be individualistic. She is usually better off conferring with others. (I thus also agree with Kornblith 2012, chs. 2 and 3, that the role of individualistic inquiry within epistemology has been overblown.)

Introduction  49 content externalism, methodology in psychology, and the possibility of eliminativism.) Nonetheless, ‘self-knowledge’ is not truly a misnomer, for chapters 5 and 10 argue that infallibility can help secure knowing what occurrent mental state I am in. Yet knowing-what is markedly different from the traditional topic of epistemology, knowing-that. For whether true belief suffices for “knowing what” depends crucially on one’s contextually determined goals or purposes—specifically, on whether one’s beliefs are adequately informative for those purposes. To be sure, knowing-what also depends on one’s reasons. And in these chapters, I explain how it is possible to armchairjustify second-order beliefs to either a Cartesian skeptic (chapter 5) or to an evidence-based antagonist (chapter 10).25 However, at no point do I require introspective access to one’s reasons in order to know-what. But nor is such a requirement dismissed. Neutrality on the internalism question is therefore sustained.26 As for the coherentism invoked, I can similarly feign neutrality on whether it is an internalist or externalist view. Note well, it is quite possible to be an externalist coherentist. At least, the “beliefs” in question might be those of a group, rather than of the individual.27 A social coherentist might even invoke the reliable “cognitive processes” used by the group in arriving at its beliefs, for example, peer review, the replication of experiments, etc. (Indeed, this is probably how to best understand the Sellarsian “selfcorrecting enterprise.”) Finally, the book’s interest in armchair self-knowledge does not require assuming epistemic internalism. For one, a process-reliabilist can take an interest, given that some reflective judgments are defended as highly reliable (nay, infallible). Second, it will emerge in Part III that linguistic

25 I think of justification here as a social activity (vs. a static property), though I will not demand this in what follows. One may object that a social epistemologist should deny the possibility of justification if Cartesian skepticism is live. For then, justification should be possible even if no one else exists. But in fact, “justifying” remains a social activity even if it involves only two individuals (a skeptic and myself). Yet I would resist the idea that in an utterly solipsistic world, I can “justify” beliefs to my own imaginary skeptic. (I’m not even sure what that would mean.) 26 In chapter 5, the reasons attending an infallible self-attribution take the form of introspective judgments—but again, I do not require a role for introspection in self-knowing. In chapter 10, however, the relevant reasons end up being perceptual rather than introspective. Yet this too is compatible with either epistemic internalism or externalism. Internalism says not that one’s reasons must be introspectings; rather, it must be possible for one’s reasons to be introspected. And presumably, both perceivings and introspectings can be introspected. 27 See Huebner (2014) for a stellar defense of group cognition. Social coherentism can seem dubious given that coherentism usually requires introspective access to reasons. And “group introspection” is obscure at best. But again, coherentism here does not require introspectively accessible reasons for rational belief. Though granted, this means it is not the most familiar sort of coherentism.

50 Preliminaries communication enables a crucial type of self-knowledge—and communication in the knowing-process is of central interest to social epistemology. Third, a key concern of the book is to show that externalism about content is compatible with armchair self-knowledge. This bears important connections with social epistemology, for it concerns the most basic ingredient of knowledge—thought—and how that ingredient is shaped by the community. Indeed, if one is an externalist about thought-content, then one is an externalist about knowledge-content. And so to that extent, a content externalist is a social epistemologist.

1.4  Overview of the Book Often an introductory chapter contains what are, in effect, abstracts for the chapters. But the first section or two of each chapter should adequately introduce its concerns and contentions. In what remains of this chapter, the aim is instead to describe the book from a bird’s eye point of view, to give a sense of what (if all goes well) the chapters achieve collectively. The book is divided into four parts (although Part IV is just a short recapitulation and conclusion). Part I consists of preparatory work for what happens later; it includes the preamble, this chapter, and the next. Part II is designed to counter Wayward Reflection via Thought Switching, while Part III is meant to allay Wayward Reflection via Attitude Switching. The layouts of Parts II and III are parallel: Each begins with one or two chapters articulating some infallibilist thesis about second-order judgment, followed by two chapters addressing objections, and ending on a chapter about how the newly minted infallibilisms subvert Waywardness (chapters 6 and 11, respectively). At bottom, the issue is whether a person can know what she judges (that is, occurrently believes) for self-reflecting, self-reporting (e.g., in a psychological experiment), or philosophical inquiring. Throughout, it is assumed that in order to self-know what one judges, two things must be in place. First, it must be known what thought is being judged, or to say much the same, what the content of one’s judgment is. Second, it must be known that one has the judging-attitude toward that thought (or, as I like to call it, the “alethic pro-attitude”)—as opposed to the wishing-attitude, the doubtingattitude, the denying-attitude, etc. With this in mind, Part II defends infallibilism about self-knowledge of thought, whereas Part III details an infallibilism about self-knowledge of judgment. For those in the know, both parts take after Tyler Burge’s (1988) view, where knowledge of one’s occurrent thoughts is “selfverifying.” (See also Davidson 1987 and Heil 1988.) However, Burge’s account can seem open to serious objections, and he does not extend the “self-verifying” idea to self-knowledge of attitude. These matters are remedied in Parts II and III, respectively. Such infallibilist views are then shown to help stall (the respective versions of) the Waywardness Problem.

Introduction  51 The two Parts thus jointly support the possibility of armchair-knowing one’s own judgments, and thus of rationally self-reflecting, despite contraindications from psychology.28 It is worth remarking that, besides this chapter, chapter 2 is a “must read,” for it shows that some form of infallibilism about self-knowledge is consistent with the existing psychological data. This is crucial, since many will antecedently think that the empirical evidence debunks such a view. Not so, as we shall see. Yet one may want to read the Preamble chapter prior to that. It concerns doubts about philosophy in general. But beyond this, the Preamble could be skipped, as its details shall not be crucial elsewhere. Nonetheless, the Preamble is essential to grasping the larger, metaphilosophical orientation of the book (which I myself see as important). As the final matter, I might stress the importance of chapter 8 against Wayward Reflection. For a select class of judgments, the chapter defends an infallibilist variant of a “Neo-Expressivist” view (cf. Bar-On & Long 2001, Finkelstein 2003, Bar-On 2004). Yet it borrows from other views on the scene as well, including constitutivism (Shoemaker 1996, Wright 1998, Coliva 2009), “agency-theoretic” views (Bilgrami 2006, O’Brien 2007, Soteriou 2013), and “transparency” views (Byrne 2005, Fernàndez 2013).29 However, this is where the book is most engaged in the philosophy of language, which generally tends to be more difficult. But chapter 8 is at the heart of the book, and my belief is that the attentive reader will be rewarded.30

References Bar-On, D. (2004). Speaking my mind: Expression and self-knowledge. Oxford UP. Bar-On, D. & Long, D. (2001). Avowals and first-person privilege. Philosophy and Phenomenological Research, 62(2), 311–335. Bender, J. (Ed.) (1989). The current state of coherence theory: Critical essays on the epistemic theories of Keith Lehrer and Lawrence Bonjour, with replies. Dordrecht: Kluwer. Bilgrami, A. (2006). Self-knowledge and resentment. Harvard UP.

28 It may be said that self-reflection also requires knowing one’s own desires, intentions, etc. Then, even if the book upholds self-knowledge of judgment, the rationality of reflection still has not been shown viable. But for one, the accounts of chapters 7 and 8 can be extended to self-knowledge of other mental states (and I mention how in the appendices of each chapter). Second, paradigm reflection just consists in inferring one judgment from others. So if self-knowledge of judgment is shown possible, that at least establishes that paradigm self-reflection is possible. 29 In such ecumenism, the chapter 8 account has the closest affinities to Moran (2001) and Heal (2002). Though unlike myself, none of these authors defend any sort of infallibilism. 30 I suspect that Part III is in many ways the most impressive part; it is unfortunate that exposition forces it to occur third. But its basics can be grasped sans Part II, and for the time-pressed reader, I might advise visiting it first.

52 Preliminaries Boghossian, P. (1992a). Externalism and inference. Philosophical Issues, 2, 11–28. ———. (1992b). Reply to Schiffer. Philosophical Issues, 2, 39–42. ———. (1994). The transparency of mental content. Philosophical Perspectives, 8, 33–50. ———. (2010). The transparency of mental content revisited. Philosophical Studies, 155(3), 457–465. ———. (2014). Reply to Otero’s “Boghossian’s inference argument against content externalism reversed”. Philosophy and Phenomenological Research, 89(1), 182–184. Bonjour, L. (1985). The structure of empirical knowledge. Harvard UP. ———. (1989). Replies and clarifications. In J. Bender (Ed.), (pp. 276–292). Burge, T. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, 73–121. ———. (1982). Other bodies. In A. Woodfield (Ed.), Thought and object (pp. 97– 120). Oxford UP. ———. (1986). Intellectual norms and the foundations of mind. Journal of Philosophy, 84, 697–720. ———. (1988). Individualism and self-knowledge. Journal of Philosophy, 85, 649–663. ———. (2006). Introduction. In his Foundations of mind: Philosophical papers volume 2 (pp. 1–31). Oxford UP. Byrne, A. (2005). Introspection. Philosophical Topics, 33, 79–104. Cappelen, H. (2013). Philosophy without intuitions. Oxford UP. Carruthers, P. (2011). The opacity of mind: An integrative theory of self-knowledge. Oxford UP. ———. (2015). The centered mind: What the science of working memory shows us about the nature of human thought. Oxford UP. Coliva, A. (2009). Self-knowledge and commitments. Synthese, 171, 365–375. ———. (2106). The varieties of self-knowledge. Basingstoke, UK: Palgrave. Damasio, A. (1994). Descartes’ error: Emotion, reason, and the human brain. New York: Grosset/Putnam. Davidson, D. (1987). Knowing one’s own mind. Proceedings and Addresses of the American Philosophical Association, 60(3), 441–458. Doris, J. (2015). Talking to our selves: Reflection, ignorance, and agency. Oxford UP. Epley, N. & Gilovich, T. (2005). When effortful thinking influences judgmental anchoring: Differential effects of forewarning and incentives on self-generated and externally provided anchors. Journal of Behavioral Decision Making, 18, 199–212. Fernàndez, J. (2013). Transparent minds. Oxford UP. Finkelstein, D. (2003). Expression and the inner. Harvard UP. Foley, R. (1987). The theory of epistemic rationality. Harvard UP. Goldman, A. I. (1986). Epistemology and cognition. Harvard UP. ———. (1999). Knowledge in a social world. Oxford UP. Goldman, A. I. & Whitcomb, D. (Eds.) (2011). Social epistemology: Essential readings. Oxford UP. Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment. Psychological Review, 108, 814–834. Halberstadt, J. & Wilson, T. (2008). Reflections on conscious reflection: Mechanisms of impairment by reasons analysis. In J. Adler & L. Rips (Eds.), Reasoning: Studies of human inference and its foundations (pp. 548–565). Cambridge UP.

Introduction  53 Harman, G. (1986). Change in view: Principles of reasoning. MIT Press. Heal, J. (2002). First person authority. Proceedings of the Aristotelian Society, 102, 1–19. Heil, J. (1988). Privileged access. Mind, 98, 238–251. Huebner, B. (2014). Macrocognition: A theory of distributed minds and collective intentionality. Oxford UP. Klein, P. & Warfield, T. (1994). What price coherence? Analysis, 54, 129–132. Kornblith, H. (1989). The unattainability of coherence. In J. Bender (Ed.), (pp. 207–214). ———. (1999). Distrusting reason. Midwest Studies in Philosophy, 23, 181–196. ———. (2002). Knowledge and its place in nature. Oxford UP. ———. (2004). Sosa on human and animal knowledge. In J. Greco (Ed.), Ernest Sosa and his critics (pp. 126–134). Blackwell. ———. (2010). What reflective endorsement cannot do. Philosophy and Phenomenological Research, 80(1), 1–19. ———. (2012). On reflection. Oxford UP. ———. (2013). Naturalism versus the first-person perspective. Proceedings and Addresses of the American Philosophical Association, 87, 122–142. ———. (2016). Replies to Boghossian and Smithies. Analysis, 76(1), 69–80. Lehrer, K. (1974). Knowledge. Oxford UP. ———. (1989a). Coherence and the truth connection: Reply to my critics. In J. Bender (Ed.), (pp. 253–275). ———. (1989b). Knowledge reconsidered. In M. Clay & K. Lehrer (Eds.), Knowledge and skepticism (pp. 131–154). Westview. ———. (1990). Metamind. Oxford UP. ———. (2000). Theory of knowledge, 2nd edition. Boulder, CO: Westview. ———. (2003). Coherence, circularity, and consistency: Lehrer replies. In E. J. Olsson (Ed.), The epistemology of Keith Lehrer (pp. 309–356). Dordrecht: Kluwer. Lycan, W. G. (1988). Judgement and justification. Cambridge UP. ———. (1996). Plantinga and coherentisms. In J. Kvanvig (Ed.), Warrant and contemporary epistemology (pp. 3–23). Totowa, NJ: Rowman and Littlefield. ———. (2002). Explanation and epistemology. In P. Moser (Ed.), Oxford Handbook of Epistemology (pp. 408–433). Oxford UP. ———. (2012). Explanationist rebuttals (coherentism defended again). Southern Journal of Philosophy, 50(1), 5–20. McGinn, C. (1977). Charity, interpretation, and belief. Journal of Philosophy, 74, 521–535. Moran, R. (2001). Authority and estrangement: An essay on self-knowledge. Princeton UP. Nagel, J. (2007). Epistemic intuitions. Philosophy Compass, 2(6), 792–819. Nisbett, R. & Wilson, T. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 8, 231–259. O’Brien, L. (2007). Self-knowing agents. Oxford UP. Parent, T. (2007). Infallibilism about self-knowledge. Philosophical Studies, 133(3), 411–424. Putnam, H. (1973). Meaning and reference. Journal of Philosophy, 70(19), 699–711. ———. (1975). The meaning of “meaning”. In his Mind, language, and reality: Philosophical papers vol. 2 (pp. 215–271). Cambridge UP.

54 Preliminaries Quine, W.V.O. (1951). Two dogmas of empiricism. Philosophical Review, 60(1), 20–43. Reprinted in his (1961) From a logical point of view: Nine logicophilosophical essays, 2nd edition (revised). Harvard UP, pp. 20–46. ———. (1969). Epistemology naturalized. In his Ontological relativity and other essays (pp. 69–90). Harvard UP. ———. (1991). Two dogmas in retrospect. Canadian Journal of Philosophy, 21(3), 265–274. Quine, W.V.O. & Ullian, J. (1970/1979). The web of belief, 2nd edition. McGraw-Hill. Rescher, N. (1973). The coherence theory of truth. Oxford UP. ———. (1979). Cognitive systematization. Blackwell. Schwitzgebel, E. (2011). Perplexities of consciousness. MIT Press. ———. (2012). Introspection, what? In D. Smithies & D. Stoljar (Eds.), Introspection and consciousness (pp. 29–48). Oxford UP. Sellars, W. (1956/1963). Empiricism and the philosophy of mind. In H. Feigel & M. Scriven (Eds.), Minnesota studies in the Philosophy of Science vol I (pp. 253– 329). Reprinted with additional footnotes in his Science, perception, and reality (pp. 127–196). London: Routledge & Keegan Paul. ———. (1973). Givenness and explanatory coherence. Journal of Philosophy, 70, 612–624. Shoemaker, S. (1996). The first-person perspective and other essays. Cambridge UP. Sosa, E. (1985). Knowledge and intellectual virtue. The Monist, 68, 226–245. ———. (2011). Knowing full well. Princeton UP. Soteriou, M. (2013). The mind’s construction. Oxford UP. Stich, S. (1978). Autonomous psychology and the belief-desire thesis. The Monist, 61, 573–591. Tiberius, V. (2008). The reflective life: Living wisely within our limits. Oxford UP. ———. (2013). In defense of reflection. Philosophical Issues, 23, 223–243. Weinberg, J. (2009). On doing better, experimental-style. Philosophical Studies, 145, 455–464. Williamson, T. (2009). Replies to Ichikawa, Martin, and Weinberg. Philosophical Studies, 145, 465–476. Wilson, T. & Dunn, E. (2004). Self-knowledge: Its limits, value, and potential for improvement. Annual Review of Psychology, 55, 493–518. Wilson, T. & Hodges, S. (1993). Effects of analyzing reasons on attitude change: The moderating role of attitude accessibility. Social Cognition, 11, 353–366. Wilson, T. & Kraft, D. (1993). Why do I love thee? Effects of repeated introspections about a dating relationship on attitudes toward the relationship. Personality and Social Psychology Bulletin, 19, 409–418. Wilson, T. & Schooler, J. (1991). Thinking too much: Introspection can reduce the quality of preferences and decisions. Journal of Personality and Social Psychology, 60, 181–192. Wilson, T., Schooler, J., Hodges, S., Klaaren, K. & LaFleur, S. (1993). Introspecting about reasons can reduce post-choice satisfaction. Personality and Social Psychology Bulletin, 19, 331–339. Wright, C. (1998). Self-knowledge: The Wittgensteinian legacy. In C. Wright, B. Smith & C. Macdonald (Eds.), Knowing Our Own Minds (pp. 13–46). Oxford UP.

2 The Empirical Case against Infallibility

This book aims to develop an increasingly nuanced infallibilist view of self-knowledge—a Cartesian-inspired view where a subject cannot be wrong, when sincerely self-attributing specific kinds of mental states. But to be clear: The premises will always include some substantive, falsifiable, empirical hypothesis, thus leaving room for rational doubt. So the book in no way attempts to prove such infallibilism. It instead develops the best sort of infallibilism that can stand as a viable albeit empirically corrigible hypothesis.1 However, it is generally thought that the psychological evidence has already debunked such a view. Now it is, indeed, beyond question that many self-attributions of mental states are false. I would also agree that, especially in light of the psychological data,2 it is possible to doubt virtually any selfattribution. Still, if the data shows that self-attributions are often wrong, it does not follow that they are most often wrong. In fact, it remains conceivable that in some cases, the truth of a self-attribution is nomologically necessitated by the attribution itself. That would make the self-attribution “infallible” in one sense (although not in the Cartesian sense) of ‘infallible’.

1 Despite our opposing views, I thus emulate Carruthers (2011) when he declares that his work “doesn’t by any means fit the mold of contemporary analytic philosophy. It contains very little that is recognizable as conceptual analysis, and hardly any claims are intended to be apriori. Indeed, the book can just as well be thought of as an exercise in theoretical psychology (Compare theoretical physics, which uses other people’s data to develop and test theories.) But this is an activity that Hume and many other philosophers . . . would have recognized as a kind of philosophy” (p. xiii). 2 See, e.g., Dutton & Aron (1974), Tversky & Kahneman (1974; 1996), Wason & Evans (1975), Nisbett & Wilson (1977), Nisbett & Ross (1980), Rubinoff & Marsh (1980), Gazzaniga (1985; 1995), Garrett & Brooks (1987), Devine (1989), Merikle (1992), Gopnik (1993), Bargh et al. (1996), Wegner & Wheatley (1999), Haidt (2001), Pronin et al. (2002), Wilson (2002), Briñol & Petty (2003), Lucas & Ball (2005), Johansson et al. (2005; 2006; 2008), Pronin (2008), Bortolotti & Cox (2009), Bortolotti (2010), Hall et al. (2010; 2013), and Bortolotti et al. (2012). An anonymous referee reminds me that there are prominent infallibilists regarding occurrent phenomenal states (see e.g., Chalmers 2003, Horgan & Kriegel 2007). But in my own work, the focus is on self-attributions of occurrent thoughts and judgments. Infallibilism here seems pretty atypical.

56 Preliminaries Yet suggesting even this milder infallibilism may provoke an “Incredulous Stare” that is usually reserved for modal realists. The force of such incredulity is admittedly very powerful. However, my aim is to demonstrate that it is misplaced. The existing psychological studies do not discredit all forms of infallibilism. Not only do the warranted inductions from the evidence fail to rule out infallibilism, but also, for somewhat subtle reasons, the relevant studies would be self-incriminating if they licensed such a strong antiinfallibilist stance.

2.1  What Is Evidence-Based Antagonism? The basic claim in dispute may be called “evidence-based fallibilism” about self-attributions. The view is roughly: (EBF) The psychological evidence shows that self-attributions of mental states are fallible. In this, a “self-attribution” is any second-order belief, expressible by the subject by a sentence whose normal form begins with [‘I’ + Ψ-VERB]—i.e., the concatenation of the first-person pronoun ‘I’ with a verb expressing a psychological property or relation. Yet the linguistic expression of a selfattribution I often call a “self-ascription.” Thus, self-attributions include beliefs expressed by self-ascriptions like ‘I feel cold’, ‘I am thinking’, ‘I am thinking that water is wet’, ‘I wish it were summer’, and so on. Evidence-based fallibilism is not merely fallibilism about self-attributions, for (EBF) claims that the empirical evidence has basically shown fallibilism.3 This is stronger than the claim that fallibilism is true (and this distinction will matter late in the game). I also do not call (EBF) a kind of “skepticism.” (EBF) does not merely consist in doubts about infallibility (though such doubts are implied). It is more condemning than that. Besides, unlike many skepticisms, the view is evidence-based. It is not an idle skeptical stance, based on remote possibilities that no normal person would ever take seriously (deceiving demons and such). Instead, it is explicitly based on strong evidence from psychology. And plenty of people take this sort of stance seriously, including Kornblith (2002, ch. 4; 2012; 2013), Schwitzgebel (2011), and Carruthers (2011; 2015). Even so, there still remains more than one way to interpret (EBF). For instance, it could be read as “the evidence shows that some self-attributions are fallible.” Yet that is uncontentious—no sophisticated science is needed

3 Terms like ‘show’ may be unclear, but I shall not demand much of them. At most, I assume that if p is shown, then it is irrational to deny p. (There are other terms in this work that I don’t take the time to define, despite their nebulous meanings. But my goal shall be to apply those terms only in uncontroversial instances.)

The Empirical Case against Infallibility  57 to know that some self-attributions are false. (I doubt Descartes needed Freud to teach him that people sometimes fool themselves, or don’t know what they want.) Nevertheless, (EBF) seems quite incorrect on a different reading, where it says that the evidence shows that all self-attributions are fallible. For there seem to be truistic cases of infallibility, e.g., when one thinks “I am thinking a thought.” If one manages such a thought at all,4 then the thought is true—its occurrence suffices for its truth. Plausibly, then, we should interpret (EBF) as opposing infallibilism about self-attributions of specific types of mental states. This results in a fallibilist view specific to a mental type T: (EBFT) The psychological evidence shows that all self-attributions of mental tokens of the type T are fallible. Traditionally, values for T may include: T = occurrent phenomenal state T = occurrent thought T = occurrent belief Some terminology: ‘Thought’ shall denote any state that hosts a content— whereas ‘belief’ denotes a specific type of thought, one where the subject adopts the “believing-attitude” toward a content (or as I like to say, the “alethic pro-attitude”). Finally, I refer to an occurrent belief as a judgment. Mental states like those above are ones that Descartes’ infallibilism supposedly concerned. Accordingly, the evidence-based antagonist says that for at least one of these Ts, the psychological evidence shows that every selfattribution of a T-state is fallible.5 Now I would not oppose the evidencebased fallibilist for most choices of T. The only evidence-based fallibilist

4 As discussed in the Preamble and chapter 9, I think it is an open question whether thought exists. Ditto with the existence of the self. Still, the point remains that if the self thinks “I am thinking a thought,” then it is invariably true. 5 There is a serious question about what ‘occurrent’ means (although this is rarely noticed in the literature). One might have thought that “occurrent” states are just those mental states that are datable, or occur at some specific time. Yet some dispositional beliefs are held only for a short duration, despite being non-occurrent (Bartlett ms.). Suppose you see me wearing my favorite green shirt. That I am wearing a green shirt is your occurrent belief, but it might simultaneously cause a dispositional belief that, e.g., I am not wearing a black shirt. Further, if you see me change into a black shirt, then the occurrent and dispositional beliefs terminate at the same time. Since the duration of the two beliefs is the same, why does only one count as “occurrent”? (One hopes that the answer is not: “Because only conscious states can be occurrent”—after all, the word ‘conscious’ has at least 17, and possibly as many as 40 different meanings; see Lycan 1996, ch. 1; Vimal 2009.) Still, I allow the fallibilist the use of the word ‘occurrent’, as long as I can do the same. (We have some grasp of the term, and we can work out details if the need arises.)

58 Preliminaries of concern is one who targets self-attributions of occurrent thoughts or of judgments, which we may speak of jointly as “o.t.j. self-attributions.” And even then, you would not find me saying that all o.t.j. self-attributions are infallible, but only those of a certain subclass. Yet let us leave open what the class is. (I do not hold that the only evidentially admissible infallibilisms have the same class as I.) But it includes more than just the truistic selfattributions like “I am thinking a thought” and its ilk. (A non-truistic case I would include, under certain provisos, is “I am judging that water is wet;” see chapter 7.) To sharpen this, let us henceforth call the non-truistic cases “OTJ self-attributions.” Then, the only evidence-based fallibilist we need to consider is one we can call the “antagonist:” (A) The psychological evidence shows that every OTJ self-attribution is fallible. According to our antagonist, the psychological evidence demonstrates the fallibility of any OTJ self-attribution. Note that (A) is still relatively modest. It even permits believing in the actual truth of all OTJ self-attributions. It only badmouths the idea that they are infallibly true, where the ‘-ible’ in ‘infallible’ expresses some type of necessity. To make this plain, let K be the set of worlds on which the relevant type of necessity is defined (epistemic, doxastic, nomological, or what have you). Then, one may explicate (A) as: (AK) The psychological evidence shows that no OTJ self-attribution is true at every world of the kind K. We shall leave open which worlds are relevant, to allow the antagonist the most latitude. (But assume that K-worlds include our world at least.) Regardless, I eventually argue that if something along the lines of (AK) is true, then so is the following ‘no presumptions’ principle: (NP) The psychological evidence means it is irrational in an SA-testing context to presume, of some OTJ self-attribution, that it is true.6 In this, an “SA-testing context” is a context where the truth of a subject’s Self-Attributions are being tested, for example, the experimental contexts of those studies cited in n.1. As for ‘presume’, we will need it to be defined only with respect to OTJ self-attributions, and only with respect to SA-testing

6 Observe I am speaking of presuming de re rather than presuming de dicto. The former makes for a logically stronger condition; presuming de re that (∃x)Φ secures (in virtually all cases) the corresponding de dicto presumption, but not vice-versa.

The Empirical Case against Infallibility  59 contexts. So it will be easiest to give a partial definition of ‘presume’ in those terms: (Df) S “presumes,” of a subject’s OTJ self-attribution, that it is true in an SA-testing context if S believes without independent evidence, of the self-attribution, that it is true. Above, “independent evidence” for an OTJ self-attribution is (at minimum) evidence that the subject has the thought/judgment self-attributed—where this evidence is probabilistically independent of the subject’s own understanding of what she thinks/judges. Putting (Df) together with (NP), we then arrive at: (NPDf) The psychological evidence means it is irrational in an SA-testing context to believe without independent evidence, of some OTJ selfattribution, that it is true. My suggestion now is that (NP), interpreted as (NPDf), is entailed by (A) when (A) is understood in terms of (AK). I shall simply assume the entailment for now, and give an argument later, once I am in a better position to do so.

2.2  Revisiting the Experiments Thus, we are presently assuming that the evidence-based antagonist holds it is irrational to presume the truth of some OTJ self-attribution, at least in an SA-testing context, given the evidence from psychology. Although I have yet to argue this, the basic thought is rather intuitive. In a context where the subject’s self-attributions are being tested, it would be odd if researchers simply presumed the truth of some OTJ self-attribution (especially if results from other SA-tests show that these attributions are often false).7 Even so, I will demonstrate that under this norm, the relevant psychological studies are self-incriminating. That’s because the experimenters adopt exactly those kinds of presumptions. To be clear, I myself do not regard the studies as self-incriminating—the experiments indeed show significant limits on self-knowledge. Yet that is because I suspect the experiments (unwittingly) exploit the infallibility of some self-attributions, despite revealing other self-attributions to be quite

7 Take heed that although (NP) concerns self-attributions of occurrent first-order states, there is no assumption that the self-attributions are themselves occurrent (higher-order) beliefs. This is as it should be—after all, it still would be odd for experimenters just to presume the truth of a subject’s higher-order belief in an SA-testing context, even if that belief is a dispositional belief rather than an occurrent one.

60 Preliminaries fallible. (But again, this chapter attempts no positive argument for infallibilism; infallibilism here is only playing defense against a type of fallibilism.8) Still, we shall see that our antagonist should have a serious objection to the experiments. That is so, even though they provide the empirical basis for her antagonism. In brief, that’s because (NP) is violated in those experiments. What’s more, we shall appreciate that the violation of (NP) will seem like a pseudo-problem. This will indicate that evidence-based antagonism is not only justified in a self-prohibited manner, but also that it must be wrong in some fashion, insofar as it insists on (NP). The moral will be that the antagonist is incorrect—the evidence does not discredit infallibilism about some OTJ self-attributions. Such a view may be rightly rejected in the end, but my claim is that our antagonist currently lacks proper grounds for that. Matters shall be clearest if the issue is related to some actual psychological studies. For this purpose, I have selected a classic one by Nisbett and Wilson (1977), and a more contemporary study by Haidt (2001). (However, the issue is hardly unique to these studies.9) The claim, then, is that (NP) would mean that there is a serious methodological problem for both studies. There are a variety of experiments carried out by Nisbett and Wilson, but in the interest of brevity, I shall just discuss one. In this experiment, 81 male introductory psychology students were provided a list of word pairs, with the instruction to commit them to memory. Some of the pairs were designed to facilitate certain associations with other words. For example, when subsequently asked to “name a detergent,” the memorized pair ‘ocean-moon’ was meant to encourage subjects to name “Tide.” And indeed, when interviewed, such associations were exhibited. Target responses (e.g., “Tide”) doubled from 10% to 20% when subjects had memorized the list of wordpairs. However, in subsequent interviews, subjects failed to cite the memorized word-pairs as a reason for their answers. Instead, subjects “focused on some distinctive feature of the target [as in] ‘Tide is the best-known detergent’ . . . or . . . ‘I like the Tide box’ ” (p. 243).10

  8 Moreover, I grant that one need not take this infallibilist line to see the experiments as legitimate. It is enough to say that they exploit the high reliability of some self-attributions. But again, my present aim is not to give a positive case for infallibility; it is only to rebut the empirical case against it.   9 Indeed, the issue seems present in all studies that I am aware of. For it concerns any methodology that takes subjects’ verbal reports as accurate expressions of occurrent beliefs (even if those beliefs are deemed mistaken, re: the mental states they are about). This shall become clearer in later sections. 10 Smith and Miller (1978) claim, contra Nisbett and Wilson, that the suggested responses on the part of the subjects may well be true—if the subject’s task was to identify some causal influence on why she responded with ‘Tide’. But while this possibility strikes me as real, I shall bracket it in what follows.

The Empirical Case against Infallibility  61 As for Haidt’s (2001) experiment, subjects first read a short passage about a brother and sister agreeing to have sex in secret while vacationing. The story adds that she is already on the pill, and that he uses a condom as well. Further, although the two have a fine time, they decide not to do it again. And the experience makes them feel closer by sharing in the secret. Haidt reports that, nonetheless, most people judge that what the siblings did was wrong. Yet they are hard-pressed to say why (“moral dumbfounding”). Haidt reports that his subjects: point out the dangers of inbreeding, only to remember that Julie and Mark used two forms of birth control. They argue that Julie and Mark will be hurt, perhaps emotionally, even though the story makes it clear that no harm befell them. Eventually, many people say something like ‘I don’t know. I can’t explain it. I just know it is wrong.’ (p. 814) Here, subjects seem engaged in post-hoc rationalization, rather than introspecting the actual mental precursors to their verbal response. In so doing, they portray themselves as having replied because of antecedent beliefs x, y, and z, although the reply was often just a knee-jerk reaction.11 The experiments support a similar hypothesis, namely, that subjects are unreliable in identifying why they respond in the way they do. So they apparently provide evidence for some form of antagonism, re: self-attributions. Even so, consider that subjects expressed their faulty self-reports along the following lines: (1) [I responded with ‘Tide’ because] “I like the Tide box.” (2) “They [= the memorized word-pairs] didn’t affect my responses.” (3) [I responded that the siblings did something wrong because] “It could result in emotional damage or birth defects.” Crucially, this data may not be of the right type to serve antagonism about OTJ self-attributions. Prima facie, (3) does not even express a self-attribution of any sort. After all, the subject’s explicit response just concerns the siblings’

11 Pace Haidt, it is unclear whether subjects meant to describe the actual reasons that lead to their responses. Possibly, they might have seen themselves as producing reasons to defend their intuitive judgment. (A similar issue can be raised for Nisbett and Wilson, though it is less obvious in that case.) But I shall ignore this in the discussion above.   Question: Might the Haidt-subject be right that her response owes to a belief in the harms of incest? In fact, the belief she reports is not that in general incest can result in emotional damage or birth defects. It is rather that this particular act of incest between Julie and Mark could result in harm. Yet since that overtly contradicts what is known from the story, the subject is likely confabulating. (Thanks to an anonymous reviewer for pressing me to clarify this.)

62 Preliminaries act; it does not concern a mental state of the subject herself. Even so, we can plausibly interpret (3) otherwise, where it is shorthand for: (3.1) [I answered that the siblings did something wrong because I believe] “It could result in emotional damage or birth defects.” The insertion of ‘I believe’ in (3.1) makes plain that the response is meant to communicate her belief about the story (rather than the facts of the story per se). We could concurrently suggest that (3.1) is a conversational implicature of (3) (cf. Grice 1975). But suffice it to say that one is a shortened version of the other, and that this is understood in context. Seen in this way, (3.1) is of the right type to bear on antagonism about OTJ self-attributions. But what of the other responses? (1) seems to selfascribe a preference or liking, rather than an occurrent thought or judgment. Even so, let us suppose for discussion’s sake that the subject uses the utterance to mean: (1.1) [I answered with ‘Tide’ because I believe] “The Tide box is likable.” Here, the utterance would express an OTJ self-attribution. As for (2), however, this less readily fits the mold. For it primarily concerns the absence of a causal relation between the subject’s earlier answer and other mental states. In this respect, it does not obviously involve OTJ self-attributions. It is true that (2) (like all assertions) is an expression of an occurrent thought. But it is not about the subject’s occurrent thought or judgment—which is essential to OTJ self-attributions. I thus suggest we leave (2) aside, since we can at least work with (1.1) and (3.1). Still, (1.1) and (3.1) concern the causal basis of earlier answers as well. Yet they seem to remain relevant—for in these two cases, the subject identifies her belief as causing her earlier answer.12 (In contrast, the causal influence in (2) was a somewhat non-descript state of memory.)13 But: Are (1.1) and (3.1) about occurrent beliefs? It seems not. In each case, the belief is identified as one from the past, as the cause of the subject’s answer to an earlier question (White 1980, p. 106). Yet if so, then ultimately none of the actual responses from the experiments justify “evidence-based” antagonism. That is so, given that the antagonism is restricted to OTJ self-attributions.

12 There is a question about whether these beliefs are causes vs. reasons (cf. Anscombe 1957/1963, Davidson 1963). Well, the experiments assume that they are at least causal influences, and whether they are also reasons is a question we can leave open. I thus speak of causes above, but this should be read as neutral on the reasons-question. 13 The memorized word-pairs may constitute a group of associative beliefs, in which case, the subject’s reference to the word-pairs might ultimately be spelled out as some kind of OTJ self-attribution. However, (2) at most implies that some beliefs or other of that type exist. It does not self-ascribe any of the specific beliefs involved.

The Empirical Case against Infallibility  63 (Admittedly, however, there are other ways of construing subjects’ responses, and we shall consider these in the next section.) To his credit, Carruthers (2011) recognizes this sort of issue, despite his clear antagonistic proclivities.14 However, he cites other experiments (e.g., Wegner & Wheatley 1999), where the time-lapse is quite short between the mental state and the self-report.15 He thus suggests that failures of working memory are not a factor in these cases—so as concerns reliability, verbal reports can be counted as if they co-occurred with the reported mental states (2011, p. 341). Yet the evidence here remains that subjects are quite fallible in what they report. However, Carruthers assumes that memory loss would be the reason for any decreased reliability. But this is not clear. Consider that a first-order state might occur without being recorded in memory at all. A stray thought might just have a momentary existence in the global workspace,16 disappearing without a trace. In such a case, if one subsequently fails to report the thought, it is not so much that memory failed to retrieve a record of it. It is rather that no record was ever stored in the first place. In contrast, as long as a thought is presently occurrent, then introspection at least has a chance at detecting it. Thus, there may be a significant difference in reliability between co-occurrent self-attributions and those that are even slightly delayed. It is tempting to say that a moment later, any thought will leave some trace. But this is dubious, in the same way that it is dubious to say that every facet of the visual field leaves some impression on memory. I am reminded of an anecdote, where Watson claims to recall well a certain building. Holmes then cleverly asks how many windows the front of the building has—whereupon Watson is unable to answer. This exemplifies a

14 Surprisingly, however, Carruthers is not particularly antagonistic toward knowledge of our own sensations: “[Regarding] the set of sensory or sensory-involving states, which include seeing, hearing, feeling, and so on . . . the model of self-knowledge that I present regards our awareness of these types of state as being relatively unproblematic” (p. xi). But see Schwitzgebel (2011) for counter-evidence on this. 15 Also, Carruthers (2011, p. 344) claims that Briñol and Petty’s (2003) study concerns occurrent judgments, viz., self-assessments of confidence. Yet Briñol and Petty’s experiment provides no evidence that those self-assessments are in error. I would grant that some amount of self-ignorance is shown, but only self-ignorance of a non-rational influence on an earlier judgment. (To be fair, Carruthers’ primary interest with Briñol and Petty is to confirm his Bem-style self-perception theory, and it may well do so. Still Briñol and Petty’s data do not show self-ignorance of an occurrent state.) 16 In the previously published version of this chapter, I misread Carruthers (2014) as denying altogether the existence of a cognitive “workspace.” But in fact, he just denies the existence of a workspace as we usually understand it. He thinks there is a sensory-based “stream of consciousness,” but with only two exceptions (ibid., p. 145), he holds that no propositional attitudes appear in the workspace. Regardless, the point above does not really depend on thoughts entering the workspace. The real point is just that memory is selective with respect to the mind’s own operations, just as it is selective with respect to external states-of-affairs.

64 Preliminaries well-known phenomenon, where one remembers x, without remembering every detail about x. In Watson’s case, memory never bothered to commit the exact number or arrangement of the windows. So it is not so much that he forgot these things. Rather, he never registered them in the first place. Plausibly, some momentary, stray thoughts are like the windows on Watson’s building. While a thought is occurrent, one can report it reliably enough, just like Watson can report the number of windows while gazing at the building. One can attend to the relevant feature. But once the feature is no longer present, such an attentional act is not possible, and memory will not have recorded every detail. In which case, one may be significantly less reliable about what was “in view”—even if only a moment has passed. Contra Carruthers, then, it would not be memory loss, but rather the selectivity of memory that would cause such unreliability.17 By the way, there is a second reason why co-occurrent self-attributions might be especially reliable, compared to those occurring even a millisecond after. On the view defended in subsequent chapters, sometimes the firstorder thought is literally an (ineliminable) part of the self-attribution. If so, then tokening the second-order judgment necessitates the occurrence of the first-order thought—just like writing ‘I am writing that water is wet’ necessitates writing ‘water is wet’. And in both cases, since the occurrence of the first-order bit is precisely what is contended, the second-order bit is thus infallible. It is clear, moreover, that this sort of “compositional” infallibility is possible only if the self-attribution temporally overlaps with the entire firstorder state. Any delay between them means that compositional infallibility cannot occur. So again, delayed self-attributions may well be significantly less reliable.

2.3  The Argument against Evidence-Based Antagonism Thus, evidentially speaking, there seems to be no substitute for selfattributions that co-occur (roughly) with the reported states. But unfortunately for the antagonist, the existing experiments all feature some timelapse. So the point remains that the existing evidence does not demonstrate the unreliability of OTJ self-attributions. Nevertheless, we might assume otherwise for discussion’s sake. Specifically, with (1.1) and (3.1), let us suppose that the belief being self-ascribed is both occurrent and a standing belief from earlier. (The belief is re-occurrent; the subject is reaffirming what she judged before.) To make this clearer, let

17 An anonymous reviewer worries that this indicates a quite general problem in the psychologist’s aim to test a subject’s self-knowledge. While this may be so, I do not want to delve into the matter at this point. However, see section 2.3.3 below for some pertinent considerations. Also, see the concluding section.

The Empirical Case against Infallibility  65 us rephrase the subjects’ responses as conjunctions, where the first conjunct patently expresses an OTJ self-attribution. In each case, we can then focus on these bits bearing directly on the antagonistic view. The rephrasings look like this: (1.2) I judge the Tide box to be likeable, and this is the same belief that caused me to respond earlier with ‘Tide’. (3.2) I judge that their act might result in emotional damage or birth defects, and this is the same belief that caused me to respond with ‘The siblings did something wrong’. Granted, such adjustments to the subjects’ responses mean we are deviating quite a bit from the raw data. But again, I am doing this out of charity. Besides, it is not entirely implausible that (1.2) and (3.2) just articulate more precisely what was actually said in loose terms. Imagine now that the Nisbett-Wilson respondent is somewhat repelled by the Tide box, and her verbal response is an ad hoc rationale for the earlier answer. Similarly, suppose the Haidt subject is mainly concerned to rationalize her initial snap judgment. Indeed, since Haidt’s story directly contradicts the beliefs the subject self-ascribes, we can easily imagine that she has the contrary belief, and momentarily forgot that fact. Seen from this angle, the experiments would seem to justify antagonism about (at least some) self-attributions of occurrent beliefs. Whether or not that is so,18 the point I wish to make now is that the experiments ought to be methodologically problematic by the lights of (NP). For, even though the context counts as an SA-testing context, the experimenters are nonetheless presuming the truth of some OTJ self-attributions. The key idea here is that, if subjects are seen as mistaken about their firstorder beliefs, then the experimenters must be presuming subjects are right about their second-order beliefs.19 For if subjects do not really have the second-order beliefs they express, then they are not mistaken in the way the

18 Since one’s beliefs/desires can be inconsistent, it does not follow from current suppositions that the subject’s self-ascription is false. This is especially clear with Haidt’s subject: It may be she has both the self-ascribed belief and the opposite one as well. Perhaps her self-ascription has a misleading implicature that she is being consistent in her judgment— regardless, the presence of the one belief is sufficient to make the ascription true, regardless of whatever else she believes. Even so, I shall not rely on this point in what follows. 19 There is a connection here with Davidson’s (1983) view that, in order for a speaker to be interpretable, she must be presumed to have mostly true beliefs. The case of SA-testing is arguably a special case of this: In order to interpret a speaker as being mistaken about her first-order beliefs, we must presume she correctly grasps her second-order beliefs. However, I mention it only for those who are interested—nothing above requires adopting Davidson’s views. Cf. also Wright’s (1998, p. 17) view that self-ascriptions must have a presumption of truth, in order to make sense of the subject as a thinker/agent. Shoemaker (1968) makes a similar claim as well.

66 Preliminaries experimenters suggest. Yet the evidence for attributing these second-order beliefs just consists in the subjects’ expressions of those beliefs. (At least, the studies mentioned no other evidence on this point.) However, that means there is no independent evidence for a subject having these second-order beliefs—no evidence that is independent of the subject’s own understanding of what she second-order believes (as manifested in her assertion). So there seems to be a violation of (NP).20 In developing this line, it will help to use Rosenthal’s (2005) distinction between a subject reporting a belief and expressing a belief (see Rosenthal 2005, ch. 2, §2.) To illustrate, suppose I utter ‘I believe that dinner is at six’. Normally, this reports a first-order belief that dinner is at six, i.e., it is a representation of me having that belief. Yet it also is an expression of a second-order belief. The utterance articulates a belief to the effect that I have the first-order belief. Similarly, if a Nisbett-Wilson subject utters ‘I believe the box is rather likable’, she reports a first-order belief. Simultaneously, however, the utterance is an expression of the subject’s second-order belief—the utterance stems from a belief about what she first-order believes. Now in the Nisbett-Wilson study, the question concerns the truth of the second-order belief, the belief allegedly expressed by her assertion. That’s because it is in doubt whether the subject really has the first-order belief, the one she reports herself as having. Yet in all this, the subject’s possession of the second-order belief is not in question. That belief is assumed to be present; the only concern is whether it is true. In taking its presence for granted, however, are we presuming that some OTJ self-attribution is true? On this matter, consider the initial conjuncts of (1.2) and (3.2): (1.25) I judge the Tide box to be likeable. (3.25) I judge that the sex act might result in emotional damage or birth defects. The experimenters claim that, when uttered by the relevant subjects, (1.25) and (3.25) are false. But notice this alone does not imply that subjects have

20 One reader remarked that this point was already made at the end of Bem (1967). But for one, Bem’s target is only the cognitive dissonance theorists, such as Festinger (1957). For another, Bem’s criticism is that these theorists impose much of their own psychology onto the subject when looking for cognitive dissonance. The present point, however, is that experimenters presume that a subject expresses her own second-order belief, versus a confabulated one. Moreover, the basis of this presumption is not that experimenters assume that they themselves reliably express their own second-order beliefs, a reliability which they then project onto subjects. (My suspicion is instead that the experimenters’ presumption is based on a need to interpret subjects in a way that makes their utterances relevant to the experiment. Cf. the connection with Davidson et al. in the previous footnote.)

The Empirical Case against Infallibility  67 made a mistake, for they must also believe them to be true. That is, if ‘I’ is indexed to an appropriate subject, it must also be true that: (1.3) I believe that I judge the Tide box to be likeable. (3.3) I believe that I judge that the sex act might result in emotional damage or birth defects. Yet what establishes that these second-order beliefs exist? Apparently, the assumptions in this are: (4) If a subject asserts “I judge that p,” then (other things equal) she is expressing her belief that she judges that p. (5) The Nisbett-Wilson subject asserts (1.25). (6) The Haidt-subject asserts (3.25). From these, it indeed follows that the subjects possess the relevant secondorder beliefs (if ceteris is paribus). And on this basis, experimenters see the evidence as showing that subjects have erred. But here is the crux of the matter. What justifies (4)? The premise is not trivial; it is entirely possible to speak without expressing one’s own belief (even if things seem entirely normal). The subject might misunderstand the terms she’s using; she might be making a performance error; she might be lying or sarcastic. Conceivably, she may even be second-order confabulating. Now if the case looks normal, it is right and natural to ignore such possibilities (or so we may allow for argument’s sake). But what exactly does this take for granted? When there is no defeating evidence, it takes for granted that the subject indeed has the second-order belief she understands herself as having, an understanding that her assertion is based on. Her assertion thus functions as evidence by which the subject is attributed the second-order belief. But as a result, the evidence for the second-order belief is not independent of the subject’s own self-understanding. In particular, it is not independent of her understanding of what she second-order believes, where this understanding is reflected in which (second-order) belief she chooses to express. So apparently, the experimenters are simply presuming the subject is right in her understanding of which second-order beliefs she has. And that violates (NP).21 As earlier noted, however, the violation of (NP) does not seem like a real problem. The psychological research still strikes me as entirely legitimate. But the problem is real if (NP) is true—and that suggests that (NP) is not true. Moreover, assuming that (NP) is a corollary of the antagonist’s view, it

21 Or rather, that is so assuming “representationalism” about the relevant metacognitions. See the next section.

68 Preliminaries follows that such antagonism is also mistaken somewhere, at least when it comes to OTJ self-attributions. Before moving on, one might ask if the violation of (NP) ought to be seen as a problem. For present purposes, this need not be settled—the antagonist is in trouble regardless. If the problem is indeed real, then the empirical justification of the antagonist’s own view is vexed. Granted, her view might not be sufficiently self-incriminating to be self-undermining. I.e., the difficulty might not be serious enough to disqualify the evidence as evidence. Still, it should put a damper on her insistence against any infallibilist hypothesis. After all, her own evidential basis treats some OTJ self-attributions as if they were infallible, as if their truth was guaranteed.22

2.4  Objections and Replies *2.4.1  Metacognitive Anti-Representationalism But in fact, the argument goes through only if the subjects’ self-understanding is constituted at least partly by a third-order belief about what she secondorder believes. This is because the subject’s self-understanding is what is presumed accurate by the experimenters—yet (NP) is thereby violated only if this amounts to presuming, of some OTJ attribution, that it is true— where an OTJ attribution is a belief. Thus, to get a violation of (NP), the subject’s (presumed accurate) understanding must suffice for a third-order belief about a second-order belief. Nevertheless, this additional supposition is not as dicey as it may seem, for several reasons: a) Such a third-order belief need not wholly constitute her selfunderstanding. b) The third-order belief would typically be non-occurrent or tacit (see ch. 6, §§5 and 6 for more on tacit belief). c) The third-order belief may end up being somewhat non-conceptual (in the sense of Drestke 1981, ch. 6; Evans 1982, ch. 6). That is, it might not be comprised entirely of conceptual representations (at least, of the sort expressed in public language). d) The third-order belief need not be a result of a mindreading—roughly, an act of interpreting proprioceptive, interoceptive, or behavioral data, 22 My self-incriminating objection should not be confused with another self-effacing issue (noted by Louise Antony, in conversation). This other issue arises from asking the evidencebased antagonist: How do you know you are an evidence-based antagonist? Ordinarily, her self-reporting as much would be accepted without independent evidence. But that too would be a violation of (NP). By the lights of evidence-based antagonism, then, one is ordinarily not warranted in self-attributing that view! But whether or not this objection is sound, my present point is just that the objection above is a different objection. It concerns whether the antagonist can consistently endorse the experiments providing the evidence for antagonism. It is not a point about self-identifying per se as an evidence-based antagonist.

The Empirical Case against Infallibility  69

e) f)

guided by the subject’s “theory of mind” (cf. Gopnik & Wellman 1992, Gopnik 1993). Instead, it could conceivably be the output of a simulation process (cf. Gordon 1986, Heal 1986). Even if the third-order belief results from mindreading, the process could still be an automated one that occurs entirely sub-personally (or as a “System 1” process, in the sense of Stanovich and West 2000). One need not assume that the experimenters themselves posit a third-order belief. When they presume the accuracy of the subjects’ selfunderstanding, they might have only a de re presumption, of the thirdorder belief, that is true.

So from this angle, the “representationalist” premise is really less contentious than it may seem. (Thanks to an anonymous referee for urging further clarify on this matter.) Nonetheless, Proust (2013) is one who still might balk, insisting that some metacognitive states are wholly non-representational. So the supposition of a third-order belief remains controversial, and admittedly, I cannot settle the controversy here. But in brief, consider that Proust sees her non-representational metacognitive states as capable of guiding thought/action, in the manner of a “non-intellectualist” kind of knowhow. And as Proust is aware, there is some difficulty in construing such “know how” as metacognition, assuming the relevant states do not represent any cognitive event or process. Proust glosses this by talking of her non-representational states as “being sensitive” to specific cognitive events. Yet I am hard pressed to understand why such sensitivity is not a representational affair. (See Langland-Hassan 2014 for a more detailed presentation of this objection.) *2.4.2  The Objection from the Intentional Stance A different line of objection grants that the experimenters believe (4) in the SA-testing context, but denies that (4) is merely presumed. That’s because experimenters have independent reasons for believing (4), reasons that concern the standard belief-desire explanation of behavior. The idea is that ceteris paribus a subject would not have made that assertive speech-act, as opposed to some other one, unless she really had the second-order belief she expresses. So in that respect, the experimenters have abductive grounds for accepting the subject’s expressions of her own second-order beliefs (hereafter, grounds for “trusting the subject”).23

23 ‘Trusting’ is not an ideal word-choice, especially since it has started accruing philosophical baggage in the literature. But I am unable to think of a better word without inventing nomenclature.

70 Preliminaries The relevant belief-desire explanations would be similar to more mundane ones. I have in mind explanations like: (7) She brought her umbrella, as opposed to leaving it at home, because: (i) She believes that rain is presently likely; plus (ii) she desires to stay dry, and (iii) she believes that bringing her umbrella is a way to stay dry. And insofar as the explanation is plausible, there are abductive grounds for attributing her the judgment ascribed by (7.i). As for the subjects of the experiments, the explanation would be along the lines of: (8) The subject uttered ‘I judge the Tide box to be rather likable’, as opposed to some other sentence, because: (i) She believes that she judges the Tide box to be likable; plus (ii) she desires to express that (second-order) belief, and (iii) she believes that her utterance is a way to express that (secondorder) belief. (9) The subject uttered ‘I judge that the sex-act may result in emotional damage and birth defects’, as opposed to some other sentence, because: (i) She believes that she judges the siblings’ act might have those results; plus (ii) she desires to express that (second-order) belief, and (iii) she believes that her utterance is a way to express that (secondorder) belief. Here, the explanations partly consist in (8.i) and (9.i). So adopting these explanations means attributing the second-order judgments that are expressed. Thus, in the style of abductive arguments, the experimenters have grounds for trusting the subject. There is nothing irrational about this. It illustrates what Dennett (1975) famously calls the “intentional stance,” and countless instances confirm that the strategy works well. Nonetheless, the issue is not whether there are some grounds for trust; it is whether there is independent evidence for trust, within an SA-testing context. The abductive grounds, moreover, do not count. The short argument here is that, if they did, then the experimenters would also have “independent evidence” to trust the subject regarding her first-order beliefs. That’s because the same type of abductive grounds could justify experimenters in attributing such beliefs. But in an SA-testing context, the subject’s grip on her first-order beliefs should remain in doubt. So the abductive grounds from the intentional stance are not “independent evidence” in an SA-testing context.

The Empirical Case against Infallibility  71 *2.4.3  The Rejoinder from Expressive Speech-Acts We shall approach this point again, by first considering a rejoinder at this stage. The rejoinder is basically this: Conceivably, the abductive grounds might suffice as evidence in one case, but not the other, because the evidence is stronger in one instance. In particular, one might argue that the subject’s second-order utterance means she is more likely to have the second-order belief, as compared to the first-order belief. That’s because a subject’s utterance of ‘I believe that p’ is normally an expression of her second-order state, viz., her belief that she believes that p. In contrast, it only reports that she has the first-order belief that p. This difference suggests, moreover, that the self-ascription will normally be caused by the second-order belief. And of course, if it is caused by that belief, then that belief must exist. No such guarantee exists for the reported first-order belief. But why exactly should we favor the second-order belief as the cause of the utterance? Why not say the subject’s assertion “I judge that p” has her first-order belief that p as its causal antecedent? In many cases, such a causal supposition may be tenable. Still, the evidence-based antagonist could contend, with some plausibility, that the second-order belief is the more proximate or powerful cause. This is so, given that the utterance is second-order. (If the first-order belief were the dominant cause, one would instead expect the subject to assert “p” rather than “I judge that p.”) Thus, says our antagonist, if there is less causal distance or a tighter nomological connection between the utterance and the second-order belief, then the utterance is a more reliable sign of the second-order belief. So the evidence is indeed stronger when the second-order belief is posited, in contrast to positing the first-order belief. And that is why, in the SA-testing context, the evidence-based antagonist can maintain that the experimenters have adequate evidence for trusting the subject on which second-order beliefs she has, but not on which first-order beliefs she has. Now it is still dubious whether our antagonist has identified independent evidence for trusting the subject, but waive that. It is clear that a beliefexpression should not count as adequate evidence (whether it is independent or not) in an SA-testing context. After all, in such a context, it is clearly inadequate for trusting her expression of a first-order belief. That is so, even though a first-order belief would likely be the more direct cause of that type of utterance. Put differently, if the context includes an assertion “p,” where a first-order belief is more plausibly the dominant cause, it should still be in doubt whether the subject believes p. And that suggests that in the SAtesting context, belief-expression is not adequate evidence for trusting the subject. Indeed, the response from the Haidt subject at (3) was an expression of a first-order belief. It was this feature that, initially, made the response seem irrelevant to antagonism about OTJ self-attributions. We decided that this was a superficial problem, however, since the utterance could plausibly be

72 Preliminaries seen as shorthand for such a thing. Even so, in its original form, her expression would give parallel reasons for attributing her the first-order belief expressed. The explanation of the first-order action-expression would be the same type of explanation as before: (10) The subject uttered ‘It [= the sex act] could result in emotional damage and birth defects’, as opposed to something else, because: (i) She believes that the siblings’ act might have those results; plus (ii) she desires to express her belief, and (iii) she believes that her utterance is a way to express that belief. And accepting the explanation means accepting that she has the first-order belief, as per (10.i). But in the SA-testing context, her first-order belief clearly remains in doubt. So in an SA-testing context, abduction from an apparently “direct” expression of a belief is per se insufficient. In such a context, belief-expressions are inadequate evidence for trusting that the subject has the belief expressed. And this contradicts the antagonist’s strategy for trusting the subject on her second-order beliefs. As indicated, however, I do not think the violation of (NP) is a genuine problem. (Although again, if it is a real problem, that is no better for the antagonist.) After all, belief in the subject’s second-order self-attributions still allows us to doubt meaningfully the first-order self-attributions. This is odd in one way, since there is no obvious reason why the “order” of the judgment should matter to its reliability. Indeed, the evidence-based antagonist may see our favoritism toward higher-order beliefs as ad hoc. That may be why antagonism—and in particular, the corollary at (NP)—did not allow such a double standard. (But see the next subsection for additional dialectics on this matter.) 2.4.3  So What? Jay Rosenberg used to joke that there were only two objections in philosophy. The first is “No way” and the second is “So what?” In the latter case, an objector will grant one’s thesis, but deny that it has any real significance. In the present context, the antagonist may be similarly unimpressed about the experimenters’ violation of (NP). In particular, the antagonist can argue that the subject’s self-understanding must be flawed in some substantive way regardless. After all, if the experimenters falsely presume that the subject gets her second-order beliefs right, the falsity of these presumptions just means the subject misapprehends some of her own beliefs (albeit second-order beliefs). On the other hand, if the experimenters’ presumptions are correct, then per usual, the evidence suggests that the subject misattributes what she first-order believes. So either

The Empirical Case against Infallibility  73 way, antagonism toward OTJ self-attributions would be vindicated. A clever move by the antagonist. As one reviewer put it succinctly, the point is that if the antagonist is not entitled to rely on the subject’s utterances, that already signals victory for fallibilism. So apparently, the antagonist is unharmed even if (NP) was de facto violated during the experiments. Still, this seems like the antagonist is abandoning (NP). But more generously, perhaps she is just tweaking (NP) to acknowledge an exception. This modified (NP) would say that in SA-testing contexts, there can be “no presumptions” that favor subjects’ self-attributions, except when such presumptions serve only to vindicate antagonism in the end. The modified (NP) would still mean that the experimenters cannot presume subjects have the self-ascribed first-order beliefs. (After all, in that case, antagonism would not be vindicated if such presumptions were true.) Yet it would mean experimenters can presume that the subjects have the second-order beliefs they express. Again, regardless of whether that kind of presumption is correct, antagonism would be warranted, as per the antagonist’s “clever argument” above. In reply, I grant that the antagonist’s clever argument proves something uncongenial to infallibilism. It demonstrates that at least some OTJ selfattributions are false. Even so, it does not follow that it is possible for any OTJ self-attribution, whatsoever, to be false. And for that reason, it also does not follow that the evidence shows the fallibility of other OTJ selfattributions. In particular, the antagonist’s clever argument does nothing to discredit that experimental subjects are infallible about their occurrent, second-order beliefs, as reflected in their verbal expressions. Again, the experiments themselves assumed that a subject’s grip on those beliefs was accurate. One might even go a bit further. The experiments could be interpreted as even supporting the high reliability of (parts of) the subject’s selfunderstanding. For the psychological studies are cases where intelligent researchers unearthed evidence against her self-conception, presuming all the while that she successfully expresses her own second-order judgments. And they presumed this as a matter of course. They not only trusted the subject here, they did so without independent evidence, and without flagging it as part of their methodology. Researchers apparently believed it goes without saying that we are justified in taking second-order belief-expressions as indicating the subject’s actual second-order beliefs. That antagonistically inclined specialists adopted this trusting stance is a striking fact, one which is best explained. Perhaps, despite their credentials, they were making a methodological error. But naturally, another explanation is that our researchers are justified in trusting the subject to express her genuine second-order beliefs. Conceivably, their trust in the subject owes to a habituated linguistic practice, a practice that works precisely because

74 Preliminaries speakers are (at least) highly reliable in this respect. Naturally, such an explanatory suggestion is contentious. But it is worth remarking that, oddly, the suggestion might gain support from our researchers’ behavior, despite their antagonistic intent. At the least, this illustrates that the exception-clause proposed by the clever argument is enough to relinquish “evidence-based antagonism” as defined here. Again, given the background presumptions, the experiments do not debunk infallibilism about all OTJ-attributions. To the contrary, the infallibilist hypothesis might even help rationalize the trust given to subjects. More broadly, if any exceptions are made to (NP), that suffices to undercut the general claim at (A). For suppose it is sometimes rational (in an SAtesting context) to trust the subject’s grip on her second-order judgments. Then, a test that trusts the subject in this way will not discredit every infallibility hypothesis. After all: Either the test will discredit the very trust it employs—or not. If so, then the test is self-undermining. In which case, it could not demonstrate that any infallibilist suggestion is false (though, per the clever argument, it may well show that some OTJ self-attributions are false). On the other hand, suppose the test does not discredit the very trust it employs. Then, it is doubtful that the test discredits any infallibilist suggestion. For if it is rationally permissible for some OTJ self-attributions to be presumed, and the test itself does not undercut that trust, nothing about the test tells against the infallibility of those self-attributions. (Besides, for practical purposes, the test treats those self-attributions as if they were infallible.) Thus, the test in this case does not discredit any infallibilist hypotheses about OTJ self-attributions. And so, an evidence-based antagonist apparently cannot allow exceptions to (NP), i.e., allow that trust in the subject is sometimes rationally permissible in an SA-testing context. Which is to say, if (A) is to be upheld, (NP) must be accepted in its full generality. This, by the by, is the earlier promised argument showing that (A) entails (NP). (For a more rigorous presentation of the argument, see the appendix to this chapter.)

2.5  Onward to Part II The falsity of (NP) can come as a surprise, since again, it is odd that an experimenter must simply trust the subject in some instances, in order to judge whether she is wrong in other instances. That is so, even though there is good reason to say the subject is wrong in other instances. Even so, I suspect that psychology can do no better than this, at least at the present time. I cannot fully argue the point here; but briefly, if we do not trust the subject at all, then no utterance can be interpreted as expressing a thought that she actually has—a fortiori, expressing a mistaken second-order thought that she has. Perhaps some future experiment might detect such thoughts without depending on the subject’s own verbal expressions. Perhaps neuroscience will advance to the point where we can just “read off” the subjects’

The Empirical Case against Infallibility  75 second-order thoughts from an fMRI. But the evidence from psychology is de facto not gleaned in these ways. As things presently stand, psychologists presume the subject is right in her self-understanding, re: the second-order judgments she expresses as if her own. And the rational permissibility of this, I claim, shows that the evidence fails to discredit every hypothesis suggesting some amount of infallibility. However, it has likely occurred to the reader that there is a darker direction all this might take. Possibly, the unflattering results of the SA-tests suggest that it is not rationally permissible for psychologists to trust subjects’ self-understanding. If so, then much of psychology would be bankrupt— soliciting verbal expressions from subjects would be rationally self-effacing. Moreover, it is unclear what the prospects for psychology would be, if such soliciting is debarred from its practice. In general, then, perhaps the real lesson should be that there is a crisis of methodology in experimental psychology. In the foregoing, I have simply assumed that psychological experimentation is well-founded. And if a skeptic were to challenge this at this stage, I would have no quick rejoinder. One strategy, however, would be to argue for the infallibility of the subject’s self-understanding, as reflected in some of her verbal expressions. If such a view were correct, then experimenters would not fall into error in trusting the subject’s self-understanding. (See especially chapter 10, where this kind of idea is pursued further.) In the next chapter, we shall take an initial step in this direction, where an infallibilist hypothesis is developed about how a person might achieve knowledge of her own occurrent thoughts.

Appendix: Argument for the Entailment

We want to show that if evidence-based antagonism is true, then (NP) follows. To this end we will show the contrapositive: If ~(NP), then ~(A). Moreover, instead of talking about “the psychological evidence,” let us talk of “the results from T,” where T is the set of the tests from Nisbett and Wilson, Haidt, and the like (see n. 1). The argument then proceeds as follows: 1 (NP) is false: There is a class P of self-attributions, whose members are presumed in at least one experimental test t ∈ T—and these presumptions are not irrational in their respective testing-contexts, despite the results from T. [Assume for conditional proof] 2 The results from T are either evidence against some member of P or not. [From 1, LEM] 3 If the results from T are not evidence against some member of P, then the results of T do not show that members of P are fallible. [Premise] 4 If the results from T are evidence against some member of P, then T is self-incriminating. [By the current usage of ‘self-incriminating’] 5 If T is self-incriminating, then the results from T do not show that all OTJ self-attributions are fallible. [Premise] 6 So either way, the results from T do not show that all OTJ selfattributions are fallible, i.e. (A) is false. [From 2–5] Remarks: Premise 3 is based on the thought that there must be evidence from T against a self-attribution, if its infallibility is to be discredited by the evidence. So, since T yields no evidence against members of P in this case, the infallibility of these members is not discredited by the evidence.24

24 Some have objected that the experimenters are presuming the truth only of a subject’s token self-attribution. And that does not conflict with (A), since it disparages hypotheses about a self-attribution type. But at 3, I have collected together the tokens into the set P, to suggest

The Empirical Case against Infallibility  77 Regarding Premise 5, the background assumption is that self-incriminating experiments are limited in what they can show. After all, the experiments themselves indicate that something is faulty about the experiments. In which case, it is unlikely that they could show the fallibility of every OTJ self-attribution. Thus, if (NP) is false, then by the above proof, (A) is also false. In addition: (NP) is indeed false, or so I have argued. Therefore, (A) is false. This, in a brief summary, is the case advanced in this chapter.

References Anscombe, G.E.M. (1957/1963). Intention, 2nd edition. Harvard UP. Bargh J., Chen, M. & Burrows, L. (1996). Automaticity of social behavior: Direct effects of trait construct and stereotype activation on action. Journal of Personality and Social Psychology, 71, 230–244. Bartlett, G. (ms.). Occurrent states, unpublished draft. Bem, D. (1967). Self-perception: An alternative interpretation of cognitive dissonance phenomena. Psychological Review, 74, 183–200. Bortolotti, L. (2010). Delusions and other irrational beliefs. Oxford UP. Bortolotti, L. & Cox, R. (2009). “Faultless” ignorance: Strengths and limitations of epistemic definitions of confabulation. Consciousness and Cognition, 18(4), 952–965. Bortolotti, L., Cox, R., Broome, M. & Mameli, M. (2012). Rationality and selfknowledge in delusions and confabulations: Implications for autonomy as selfgovernance. In L. Radoilska (Ed.), Autonomy and mental illness (pp. 100–122). Oxford UP. Briñol, P. & Petty, R. (2003). Overt head movements and persuasion: A self-validation analysis. Journal of Personality and Social Psychology, 84(6), 1123–1139. Carruthers, P. (2011). The opacity of mind: An integrative theory of self-knowledge. Oxford UP. ———. (2014). On central cognition. Philosophical Studies, 170, 143–162. ———. (2015). The centered mind: What the science of working memory shows us about the nature of human thought. Oxford UP. Chalmers, D. (2003). The content and epistemology of phenomenal belief. In Q. Smith & A. Jokic (Eds.), Consciousness: New philosophical perspectives (pp. 220–272). Oxford UP. Davidson, D. (1963). Actions, reasons, and causes. Journal of Philosophy, 60(23), 685–700. ———. (1983). A coherence theory of truth and knowledge. In D. Henrich (Ed.), Kant oder Hegel? (pp. 423–438). Stuttgart: Klett-Cotta.

that the experiments do not debunk any infallibility hypothesis about tokens of the P-type. Granted, it has not been shown that P is a natural kind. But there is no reason to insist that the infallible self-attributions, if any, must form a single natural kind. There may be multiple kinds of self-attributions that turn out to be infallible.

78 Preliminaries Dennett, D. C. (1975). True believers: The intentional strategy and why it works. In A. F. Heath (Ed.), Scientific explanations: Papers based on Herbert Spencer lectures given in the University of Oxford (pp. 53–75). Oxford UP. Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56, 5–18. Drestke, F. (1981). Knowledge and the flow of information. MIT Press. Dutton, D. & Aron, A. (1974). Some evidence for heightened sexual attraction under conditions of high anxiety. Journal of Personality and Social Psychology, 30, 510–517. Evans, G. (1982). The varieties of reference. Oxford UP. Festinger, L. (1957). A theory of cognitive dissonance. Stanford UP. Garrett, J. & Brooks, C. (1987). Effect of ballot color, sex of candidate, and sex of college students of voting age on their voting behavior. Psychological Reports, 60, 39–44. Gazzaniga, M. (1985). The social brain: Discovering the networks of the mind. New York: BasicBooks. ———. (1995). Consciousness and the cerebral hemispheres. In M. Gazzaniga (Ed.), The cognitive neurosciences (pp. 1391–1400). MIT Press. Gopnik, A. (1993). How we can know our minds: The illusion of first person knowledge of intentionality. Brain and Behavioral Science, 16, 1–14. Gopnik, A. & Wellman, H., (1992).Why the child’s theory of mind really is a theory. Mind and Language, 7, 145–171.Gordon, R. (1986). Folk psychology as simulation. Mind and Language, 1, 158–171. Grice, H. P. (1975). Logic and conversation. In P. Cole & L. Morgan (Eds.), Syntax and semantics 3: Speech acts (pp. 41–58). New York: Academic Press. Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment. Psychological Review, 108, 814–834. Hall, L., Johansson, P., Tärning, B., Sikström, S. & Deutgen, T. (2010). Magic at the marketplace: Choice blindness for the taste of jam and the smell of tea. Cognition, 117, 54–61. Hall, L., Strandberg, T., Pärnamets, P., Lind, A., Tärning, B. & Johansson, P. (2013). How the polls can be both spot on and dead wrong: Using choice blindness to shift political attitudes and voter intentions. PLoS ONE, 8(4), e60554. Heal, J. (1986). Replication and functionalism. In J. Butterfield (Ed.), Language, mind, and logic (pp. 135-150). Cambridge UP. Horgan, T. & Kriegel, U. (2007). Phenomenal epistemology: What is consciousness that we may know it so well? Philosophical Issues, 17, 123–144. Johansson, P., Hall, L. & Sikstrom, S. (2008). From change blindness to choice blindness. Psychologia, 51(2), 142–155. Johansson, P., Hall, L., Sikström, S. & Olsson, A. (2005). Failure to detect mismatches between intention and outcome in a simple decision task. Science, 310(5745), 116–119. Johansson, P., Hall, L., Sikström, S., Tärning, B. & Lind, A. (2006). How something can be said about telling more than we can know: On choice blindness and introspection. Consciousness and Cognition, 15(4), 673–692. Kornblith, H. (1999). Distrusting reason. Midwest Studies in Philosophy, 23, 181–196. ———. (2002). Knowledge and its place in nature. Oxford UP.

The Empirical Case against Infallibility  79 ———. (2010). What reflective endorsement cannot do. Philosophy and Phenomenological Research, 80(1), 1–19. ———. (2012). On reflection. Oxford UP. ———. (2013). Naturalism versus the first-person perspective. Proceedings and Addresses of the American Philosophical Association, 87, 122–142. Langland-Hassan, P. (2014). Unwitting self-awareness? Philosophy and Phenomenological Research, 89(3), 719–726. Lucas, E. & Ball, L. (2005). Think-aloud protocols and the selection task: Evidence for relevance effects and rationalization processes. Thinking and Reasoning, 11, 35–66. Lycan, W. G. (1996). Consciousness and experience. MIT Press. Merikle, P. M. (1992). Perception without awareness: Critical issues. American Psychologist, 47, 792–795.Nisbett, R. & Ross, L. (1980). Human inference: Strategies and shortcomings of social judgment. Prentice-Hall. Nisbett, R. & Wilson, T. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 8, 231–259. Pronin, E. (2008). How we see ourselves and how we see others. Science, 320, 1177–1180. Pronin, E., Lin, D. Y. & Ross, L. (2002). The bias blindspot: Perceptions of bias in self and others. Personality and Social Psychology Bulletin, 28, 369–381. Proust, J. (2013). The philosophy of metacognition: Mental agency and self-awareness. Oxford UP. Rosenthal, D. (2005). Consciousness and mind. Oxford UP. Rubinoff, M. & Marsh, D. (1980). Candidates and color: An investigation. Perceptual and Motor Skills, 50, 868–870. Schwitzgebel, E. (2011). Perplexities of consciousness. MIT Press. Shoemaker, S. (1968). Self-reference and self-awareness. Journal of Philosophy, 65(19), 555–567. Smith, E. & Miller, F. (1978). Limits on perception of cognitive processes: A reply to Nisbett & Wilson. Psychological Review, 85(4), 355–362. Stanovich, K. & West, R. (2000). Individual differences in reasoning: Implications for the rationality debate? Behavioral and Brain Sciences, 23, 645–726. Tversky, A. & Kahneman, D. (1974). Judgement under uncertainty: Heuristics and biases. Sciences, 185, 1124–1131. ———. (1996). On the reality of cognitive illusions. Psychological Review, 103(3), 582–591. Wason, P. & Evans, J. (1975). Dual processing in reasoning. Cognition, 3, 141–154. Wegner, D. & Wheatley, T. (1999). Apparent mental causation: Sources of the experience of the will. American Psychologist, 54, 480–491. White, P. (1980). Limitations on verbal reports of internal events: A refutation of Nisbett & Wilson and of Bem. Psychological Review, 87(1), 105–112. Wilson, T. (2002). Strangers to ourselves. Harvard UP. Wright, C. (1998). Self-knowledge: The Wittgensteinian legacy. In C. Wright, B. Smith & C. Macdonald (Eds.), Knowing our own minds (pp. 13–46). Oxford UP. Vimal, R. (2009). Meanings attributed to the word “consciousness”: An overview. Journal of Consciousness Studies, 16, 9–27.

Part II

Knowledge of Thought

3 Infallibility in Knowing What One Thinks

The aim of this chapter is to defend the Cartesian view that a subject is infallible with respect to the contents of her own occurrent thoughts. These days, such a Cartesian claim is often met with stunned disbelief. But this is likely due to evidence-based antagonism, and to the absence of an adequate, positive defense. Yet whereas antagonism was mitigated in the last chapter, it is my aim here to supply the positive defense.

3.1  Ground Clearing The claim to be defended is not the claim that whenever a subject thinks that p, the subject knows that she thinks p. Clearly, Freudian repression and the like make this implausible. On the other hand, the converse of this claim is trivial: If a subject knows that she thinks p, then she thinks p—simply because, in general, knowledge that Φ entails Φ. However, a modified version of the converse claim has the potential to be both true and non-trivial (call it the “Cartesian Claim”): (CC) If a subject judges that she thinks that p, she indeed thinks that p. Since belief that Φ does not entail Φ, the threat of triviality is subverted. Whether (CC) is true is another matter. But if it is true, this would amount to our having inevitably true judgments about our own thoughts. Does infallibly true judgment count as knowledge? Actually, I think more is required, but I shall delay this discussion until chapter 5. Nevertheless, for convenience, I shall continue to call (CC) a claim about “self-knowledge,” even though for now, this talk of “knowledge” hangs on a promissory note. A related Cartesian view about self-knowledge is that a subject may possess indubitable judgments about the contents of our thoughts. I do not wish to defend this claim; I suspect that the range of what can be doubted is boundless. Even so, if my argument for infallibility is sound, this might ensure that self-knowledge is rationally indubitable, in that there may be no good reason for doubting it. Actually, I think that inference would be too quick, but I shall delay this discussion until section 3.4.

84  Knowledge of Thought Even if it is dubitable, infallible self-knowledge would still be quite remarkable in itself; one would like an account of it for its own sake. Moreover, there is additional pressure to account for such knowledge, since arguably it is impossible in light of certain semantic considerations by Putnam (1975) and Burge (1979). However, I will not address these externalist controversies in this chapter—I shall save them for chapters 4–6. Nevertheless, the present chapter’s argument for (CC), I think, can succeed regardless of whether content is individuated individualistically or not. A reminder is in order concerning the term ‘thinking’ and its cognates. As in other chapters, thinking that p does not entail judging that p. Further, thinking that p might be confused with “entertaining” that p—but note that thinking that p is always compatible with judging that p. (Whereas on some occasions of use, “entertaining” that p is not, e.g. entertaining a hypothesis.) More broadly, to think p is to have some propositional attitude or other with the content that p, but it does not entail having any particular attitude toward that content. Accordingly, with respect to (CC), we are concerned only with knowledge of the contents of propositional attitudes, and not of the attitudes one might have toward these contents. Further, our focus shall be on occurrent mental states rather than standing or dispositional states. My use of ‘thinking’, ‘judging’, etc. below should be read accordingly. Consider, then, a more general claim, where talk of “judging” in (CC) is replaced with talk of “thinking”: (CC') If a subject thinks that she thinks that p, she indeed thinks that p. It is this more general claim that is considered in most of what follows. But it is germane to (CC), since (CC') entails (CC): Judging that p implies thinking that p; thus, when the antecedent of (CC) is fulfilled, so is the antecedent of (CC'). Accordingly, if the latter antecedent entails that the subject has the first-order thought that p, then the former antecedent does too.

3.2  Burge’s Way of Self-Knowing One attempt to defend something like (CC') is given in Burge (1988), who argues that a second-order thought, a.k.a. a self-attribution of a thought, is guaranteed to be true. The crucial part of Burge’s account here is the “reflexive” or “self-referential” character of a second-order thought, which necessitates that such a thought have the first-order thought as a proper part. On this, Burge writes: By its reflexive, self-referential character, the content of the secondorder judgment is logically locked (self-referentially) onto the first-order content which it both contains and takes as its subject matter. (pp. 659–60)

Infallibility about What One Thinks  85 Burge does not say explicitly what it is for thoughts to be “logically locked,” but the natural reading would be that thinking the one thought logically necessitates thinking the other. Because second-order thoughts are logically locked to first-order thoughts, via self-reference, the upshot for Burge is that second-order thoughts are self-verifying. [Second-order] thoughts are self-referential and self-verifying. An error based on a gap between one’s thoughts and the subject matter is simply not possible in these cases. When I judge: I am thinking that writing requires concentration, the cognitive content that I am making a judgment about is self-referentially fixed by the judgment itself; and the judgment is self-verifying. (p. 658) In virtue of the self-referential mechanism in second-order thoughts, to have a second-order thought is always to create conditions which make the content of the thought true. But what exactly is the model here? A second-order thought, it seems, is typically not self-referential in the sense that such a thought refers to itself. Second-order thoughts may be self-referential on some occasions, e.g. “I am thinking this thought,” but they are not characteristically self-referential, as in “I am thinking that water is wet.” Perhaps Burge’s idea is to understand the relevant second-order thoughts as always having a self-referential device. For example, the thought “I am thinking that water is wet” might be reconstrued as “I am thinking, with this very thought, that water is wet” (cf. Burge 1988, p. 649). However, it unclear whether this would be a satisfying account of these self-attributions. My second-order thoughts, after all, are supposed to be thoughts of firstorder thoughts I have. But if second-order thoughts are infallible because of a self-referential device, then it seems such an account describes how the thought is infallibly “locked” to the second-order thought. It would not reveal how such a judgment is logically locked to the first-order thought. Hence, this would show how the second-order thought is logically locked to itself, but we wanted an explanation of how such thoughts are locked to first-order thoughts. A natural response here would be to say that a second-order thought is locked to the first-order thought by the fact that the second-order thought has a mechanism for referring to the first-order thought. However, Burge (1996) resists construing the first-order thought only as an object of reference. Suppose that I think that I am engaging in a thought that there are physical objects. In thinking this, I have to engage the very thought that I am referring to and ascribing to myself. The reference to the

86  Knowledge of Thought content—expressed in the that-clause—cannot be carried out unless I actually engage in the thought. The intentional content mentioned in the that-clause is not merely an object of reference or cognition; it is part of the cognition itself. (p. 96) The second-order thought does refer to the first-order thought, but the first-order thought is not merely an object of reference. Rather, it is also proper part of the cognition which constitutes the second-order thought. Thus, the idea is that if my first-order thought is a proper part of my second-order thought, then for me to think ‘I think that p’ is, in the very same act, for me to think p, since p is a proper part of the second-order thought.1 Yet it is not clear why we are compelled to draw this conclusion, for it seems to exemplify the fallacy of division, i.e. reasoning from the properties of the whole to the properties of the parts. As a matter of logic, my thinking that I think that p does not strictly entail that I do think p. (After all, in general, a subject thinking that Φ does not entail Φ.) Such an entailment may sound plausible in this instance, so the property of “being thought by me” is perhaps a special case. Perhaps my thinking ‘I think p’ necessitates my thinking all the parts of ‘I think p’, including p itself. But if so, this would be a nontrivial fact—a fact that would need to be explained rather than simply assumed. In the remainder of this chapter, the aim is to account for this fact.

3.3  Infallibilism Derived The claim in question, recall, is as follows: (CC') If a subject thinks that she thinks that p, she indeed thinks that p. My contention is that, given certain minimal assumptions about the language of thought, (CC') is a logical consequence. The language of thought (“LOT”) hypothesis, defended by Fodor (1975; 2009), is the hypothesis that thoughts are composed of concepts according to specific formation and transformation rules, i.e., a “grammar.” LOT is typically used as a device for modeling cognitive processes—but what interests me is its epistemological import for second-order thoughts.

1 This also seems to be the view in Heil (1988), where Heil talks of second-order thoughts including the content of first-order thoughts (see, e.g., p. 224). Presumably, this is just another way of saying that first-order thoughts are part of second-order thoughts; accordingly, what I say about Burge’s view may be applied to Heil’s view as well.

Infallibility about What One Thinks  87 Suppose I token the second-order LOT expression i think that water is In order to token such an expression, according to the LOT hypothesis, I must token water is wet, since this stands as the complement clause of the expression. Thus if I token i think that water is wet, I token water is wet. This is just to say that if I second-order think that I think water is wet, I do have the thought that water is wet. Hence any second-order thought that I am thinking some proposition p is guaranteed to be true; such a thought is infallible. Let us introduce the two-place Thinking predicate Txy, the index ‘i’ which picks out a subject S who tokens sentences in the language of thought,3 and variables ‘p’ and ‘q’, which are replaced by sentences. (Let us also acknowledge the LOT analogues to these devices: Txy, i, p, and q.) The derivation of (CC') proceeds as follows: wet.2

(1) S thinks Tip. (2) S thinks that q iff S tokens q.

[Assume for conditional proof]

[Assumption from the LOT hypothesis]4 (3) S tokens Tip. [From (2), (1)] (4) If q is of the form Fap, where a is a name for an individual a, and p is a complement clause, then if S tokens q, then S tokens p. [Minimal Compositionality Assumption] (5) Tip satisfies the antecedent of (4). [Assumption] (6) So, if S tokens Tip, then S tokens p. [From (4), (5)] (7) So, S tokens p. [From (6), (3)] (8) So, S thinks that p. [From (2), (7)] (9) So, if S thinks that Tip, then S thinks p. [By conditional proof, (1)-(8)] As should be clear, (9) is equivalent to (CC'). The most contentious premise in the foregoing is (4), a.k.a. the Minimal Compositionality Assumption (MCA). This premise reflects the compositional nature of the language of thought; however, it is minimal in the sense that one could accept (MCA) without accepting that thoughts are thoroughly compositional. In more detail, one might maintain that secondorder thoughts have first-order thoughts as complement clauses, without maintaining that thoughts are complexes of multiple constituents, with a full-blown grammar governing every such constituent.5

2 Following established conventions, English expressions in caps are the names for lexemes in LOT with the same content as the English expression. 3 Here, the index ‘i’ is what Perry (1979) would call an essential indexical, since in what follows ‘i’ is not substitutable salva veritate for a name of the person for which it indexes on a given occasion. 4 N.B., since ‘thought’ here denotes occurrent thoughts, it follows that all “thoughts” require a tokening of an LOT sentence. 5 Thanks to William Lycan for suggesting this minimal sort of compositionality for LOT.

88  Knowledge of Thought However, it is worth remarking that if one did accept a further compositionality thesis, then we could show not only that a subject is infallible about her own thoughts, but also that she is about the concepts she possesses. Consider the following Cartesian Claim about concepts: (CCc) If a subject thinks she has a concept of b, then she does have a b-concept. This claim can be shown by reasoning similar to the line above, by exploiting the compositionality assumption below in place of (MCA): (MCAc) If a thought q is of the form Fab, where a is a name for an individual a, and b is a concept of b, then if S tokens q, then S tokens a concept of b. It should be noted that although (MCAc) is (arguably) stronger than (MCA), it is still a relatively minimal assumption about compositionality. According to (MCAc), thoughts are just two-place relations, nothing more. Without going into great detail, the argument for (CCc) starts from a subject’s thought “I have a concept that represents b.” By (MCAc), then, it follows that having such a thought requires a tokening of a concept that represents b. But if the subject tokens such a concept, then she possesses such a concept. So whenever a subject thinks she possesses a b-concept, the thought is true.6 Note that one can accept compositionality principles without accepting other, controversial assumptions about LOT. For instance, it is possible to accept (MCA) without supposing that LOT is innate, or that it is universal among creatures capable of thought. For that matter, we might jettison talk

6 The following has been suggested to me as a counterexample to (CCc): Suppose an undergraduate hears the word ‘supervenience’, and starts using the word himself, thinking that he has the concept of supervenience, when in fact he is merely miming the use of the word. Then, contra (CCc), he thinks that he has the concept of supervenience, but doesn’t actually have it.   One might argue that the student does have a concept of supervenience, albeit an impoverished one. But supposing this is not so, we might ask what concept would be used in his second-order judgment if not the concept of supervenience? It may be a concept ‘sloopervenience’, and/or a concept of “whatever competent speakers designate by the word ‘supervenience.’ ” Regardless, if we call this (possibly complex) concept c, then his second-order judgment will be the judgment that he has the concept c. Yet in such a judgment, c is used; thus his judgment about having c is infallibly true. His mistake would be in thinking that his concept c is identical to the concept competent speakers express by the word ‘supervenience’. (It is important to note that, assuming that concepts are “hyperintentional,” the concept of supervenience and the concept c may be coreferring, yet nevertheless be nonidentical, as when c is a concept of “whatever competent speakers designate by the word ‘supervenience.’ ”)

Infallibility about What One Thinks  89 of LOT entirely, and simply note that as a surface phenomenon, secondorder thoughts appear compositional. For instance, such thoughts exhibit systematicity. One capable of thinking “Mary thinks that John likes Mary” may also think “John thinks that John likes Mary” by ordering the same concepts in different ways. But not just any thought will result from a particular ordering, nor will every ordering result in a thought, for example, “John thinks Mary John.” Thus it seems particular second-order thoughts result from particular orderings in line with certain rules. This is just to say that such thoughts display compositionality. Even someone like Dennett (1975) would seem to agree to this much. Compositionality would be, in the language of Dennett (1991), a “real pattern.” What Dennett questions is whether this compositionality is due to a parallel compositionality of a physically realized code in the brain. But if the compositionality of thought is nevertheless real, then in particular the composition of second-order thoughts seems real, by the same kinds of systematicity considerations. So we may acknowledge that some thoughts are composed of certain first-order thoughts and concepts, without suggesting there is literally “brain writing.”7 Nevertheless, it is worth emphasizing that the infallibility of such thoughts is a direct consequence of (arguably) our best theory of cognition available. It is also worth noting that the arguments above do not make use of any premise concerning whether mental content is “in the head.” All that the arguments rest upon is a structural feature of thoughts, to wit, that an occurrence of a thought about a thought (or concept) requires an occurrence of that thought (or concept).8 Beyond that, I have left open how mental content is individuated; thus, the argument is congenial to both externalists and internalists about content. Some may protest that this leaves a few important issues about selfknowledge unresolved. If, as Burge (1979) would have it, mental content is individuated in part by the social and physical environment, then a thought like water is wet will be a thought that is de H2O. But if so, then there is a sense that the corresponding second-order thought will not provide the subject knowledge that she is thinking about H2O. For in many cases the subject may lack all ability to distinguish her H2O-thoughts from thoughts about any qualitatively similar substance, hypothetical or real.

7 Accordingly, the argument at (1)–(9) could be purged of the commitment to a language of thought, by replacing the Fodorian small caps with, e.g., expressions enclosed in Sellars’ (1974) dot-quotation marks. 8 I do not mean to suggest that an infallible judgment can supervene on meaningless syntactic forms. (Thanks to Dorit Bar-On for discussion here.) I am assuming that a compositional syntactic structure mirrors a compositional semantics, and that a compositional semantics is also allowed by both externalism and internalism.

90  Knowledge of Thought If the subject cannot distinguish her H2O-thoughts from thoughts about qualitatively similar substances, it is not clear she ipso facto lacks knowledge of her H2O-thoughts—especially if there are no qualitatively similar substances to be reckoned with (cf. Brown 2004, ch. 2). Regardless, the point would hold that whatever substance a first-order thought is about, the second-order thought will infallibly concern a first-order thought of that substance. Whether these infallible second-order thoughts constitute knowledge of first-order thoughts is a further matter—and here, as I mentioned, the expression ‘knowledge’ is used provisionally. In the present chapter, although (CC') might be called a claim about “self-knowledge,” I am interested in defending (CC') itself, and not the additional thesis that (CC') can account for bona fide knowledge of one’s own thoughts. Nevertheless, even (CC') per se has interest to epistemology. Since thinking is required for any propositional attitude, we saw that (CC') has the following important consequence: (CC) If a subject judges that she thinks that p, she indeed thinks that p. Second-order judgments about one’s thoughts are infallible on this view. This is a striking fact about our position as epistemic agents. Accordingly, (CC') can function to explain (at least in part) why such judgments are epistemically privileged, as compared to the subject’s fallible store of empirical judgments. What’s more, the arguments here do not depend on the subject engaging in any cognitive activity beyond second-order judging. Thus, (CC') might explain how second-order judgments are epistemically privileged, even though the subject has not undertaken any of the usual actions to secure this epistemic status (e.g., empirical observation, inference, etc.). Although I call the infallibility of self-knowledge a “Cartesian” thesis, the defense of the thesis is distinctly unCartesian in some respects. For one, there is no invocation of a mysterious “Cartesian ego” which makes the infallibility of second-order judgments possible. Given the compositionality of thought, we can incorporate (CC) into a purely naturalistic theory. Relatedly, unlike Descartes, we argued (CC) on empirical grounds instead of apriori ones. After all, compositionality is an empirical thesis about thought, at least in the sense that it is falsifiable by experience. Even so, I leave open the possibility of introspecting on the compositionality of one’s thoughts, from which one might come to know (CC) just from the armchair (cf. chapter 5). Regardless, I take the empirical defense to lend further legitimacy to (CC) beyond what mere apriori reflection could bestow. And further, unlike Descartes, I do not aspire to make second-order judgments part of an epistemic “foundation” by which we may come to justify other beliefs. Indeed, it is possible that the argument for (CC) precludes second-order judgments from being foundational, since foundational beliefs would seem to be beliefs with no argument backing them (Bonjour 1978).

Infallibility about What One Thinks  91

*3.4  Necessity and the Apriori Brian Weatherson has observed (in correspondence) that (CC) should not be seen as bare material conditional. That would not result in an infallibility thesis, since an “infallible” judgment is one that is not only true, but also true in nearby worlds (or the like). It is thus best to reword (CC) to the following, where ‘Necessarily’ expresses the right sort of modality: (INF1) Necessarily, if a subject judges that she thinks p, then she thinks p. ‘Necessarily’ in (INF1) expresses a relatively weak kind of modality (certainly weaker than logical necessity). It expresses a kind of nomological necessity, since it concerns truth in worlds where a specific psychological law holds.9 The relevant law was explicit in the argument for (CC); this was the Minimal Compositionality Assumption, according to which a second-order judgment is composed (inter alia) of the first-order thought. The upshot is that the first-order thought “water is wet” is, as a matter of psychological law, a literal part of the second-order judgment. (N.B., this means that unlike in Burge, the “locking” between first- and second-order thoughts here is not a logical locking, but rather a nomological one.) A series of objections to such infallibilism has been more recently raised by Hoffmann (2011).10 As one example, Hoffmann writes: If we are correct, in reflecting on the warrant for a judgment regarding one’s thoughts, the reflecting thought is distinct from the reflected thought. This gives rise to an epistemological lacuna that rules out the possibility of infallible a priori self justification. This is just to say, in a manner of speaking, that the logical locking thesis (or any thesis regarding the mechanics of the reference fixing of self referential judgments) is not . . . indefeasible. (p. 251) The idea seems to be that, since there is a distinction between the firstand second-order states, it is possible for one to occur without the other. And that suggests that second-order judgments can misfire. (To be fair, the

  9 Talk of a “law” is equivocal between a regularity in nature versus a representation of that regularity, e.g., in a psychological theory. It is the latter sense of ‘law’ that is intended throughout. 10 One of Hoffmann’s objections I shall note here only; it is the Wittgensteinian claim that assertion or judgment requires the possibility of being incorrect. But on the contrary, there are cases where this dictum seems violated, e.g. ‘I am now tokening an expression’, ‘Language is actual’, ‘This sentence is a sentence’, etc.

92  Knowledge of Thought point targets logical infallibility in the first instance, yet the parenthetical remark suggests it is also meant to apply more widely.) There is of course a difference between first- and second-order states. Nevertheless, it may just be impossible for a second-order judgment to occur sans the first-order thought. Indeed, that is what the compositionality law states (assuming ‘impossible’ expresses the relevant nomological modality). Even so, the passage from Hoffmann might not be challenging infallibilism as such, but rather apriori infallibilism. This is the view that “a wide range of both (i) analytic and (ii) synthetic apriori propositions can be infallibly or absolutely justified, i.e., justified to a degree that guarantees their truth and precludes their falsity” (Hoffmann 2011, p. 241).11 But if the focus is here, then as Hoffmann acknowledges, the present author’s view is not of concern. The compositionality argument does not justify judgments apriori. It rests on an empirically falsifiable, psychological hypothesis; it is thus uncongenial to a Cartesian, hoping to justify infallibilism purely apriori. Nevertheless, I allowed that a Cartesian might somehow introspect the compositionality of second-order judgments, and justify infallibilism “apriori” in that manner.12 Hoffmann notes this about the view, yet still concludes that “Parent’s position regarding infallible self justification obliquely reveals . . . that infallible self justification cannot be purchased a priori” (p. 251). Possibly, Hoffmann meant only that my view is a kind of testimonial to the difficulties of defending the apriori. (Thanks to an anonymous referee for raising this point.) But as formulated, his claim is unexpected: If apriori justification was not ruled out, how was it shown impossible? Perhaps Hoffmann here implicitly relies on a Quinean naturalized epistemology that we share. I did not assume such an epistemology above. Yet there is special intrigue if Hoffmann is pitting a Quinean naturalized epistemology against infallibilism—since I also find Quine’s epistemology appealing. And the apriori can look out of place in a naturalistic picture. So if we are comrades for Quine, does this compel rejection of all forms of the apriori? On this matter, two senses of ‘apriori’ should be distinguished. The “apriori” justification I allowed has infallibilism justified by introspection. In contrast, the apriori justification dismissed by Quine would render a belief immune to empirical revision; it is “strongly apriori” justification (Field’s 1996; 2005 term). (Quine: “Revision even of the logical law of the excluded middle has been proposed as a means of simplifying quantum mechanics”

11 N.B., I do not say that second-order judgments are infallibly justified. My infallibility thesis does not bear on the justificatory grounds of the second-order judgment; it instead pertains to the second-order judgment itself. See also n. 13. 12 This concession is not merely a courtesy to the Cartesian. In reply to Brown (2004), Goldberg (2006) argues there is an advantage if a self-verifying judgment is justified by an introspective awareness of its self-verifying nature (see pp. 310–311). In chapter 5, I develop Goldberg’s proposal further.

Infallibility about What One Thinks  93 [1951, p. 43].) The point, then, is that one can reject strongly apriori justification without rejecting introspective justification. One only need allow that introspection, too, is vulnerable to empirical correction. So even under naturalized epistemology, Hoffmann has not discredited every “apriori” infallibilism. But he raises a further query: Does it make sense to justify infallibilism by a premise that is open to revision (a premise that is apparently fallible)? Early in his paper Hoffmann pushes this in a broad way, explicitly citing Quine on revisability. For present purposes, the objection would be that since psychological laws are revisable, the compositionality argument cannot demonstrate that second-order judgments are infallible or true “come what may.”13 There are two replies available here (though they may amount to much the same thing). First, we can distinguish epistemic or doxastic modality from the nomological modality expressed in (INF1). Then, even if it is a psychological law that second-order judgments are true, their truth need not be epistemically or doxastically necessary, relative to one’s current knowledge or beliefs. This allows second-order judgments to be doubted and revised (even rationally so, if the available evidence warrants it). But naturally, if a revision ever negates an infallible judgment, it replaces a truth with a falsity. Yet revision per se is not incompatible with infallibility. What is incompatible is infallibility with the possibility of a correct revision. A second reply, rooted directly in Quine, takes its cue from a passage in “Intensions Revisited:” Relative to a particular inquiry, some predicates may play a more basic role than others, or may apply more fixedly; and these may be treated as essential. . . . [S]imilarly dependent on context [is] . . . the whole quantified modal logic of necessity; for it collapses if essence is withdrawn. For that matter, the very notion of necessity makes sense to me only relative to context. Typically it is applied to what is assumed in an inquiry, as against what has yet to transpire. (1977, p. 121) Quine’s point is that one can speak sensibly of what is “necessary,” relative to the assumptions held fixed in an inquiry. That is so, even though other inquiries can unfix those assumptions and revise them in response to new evidence.14

13 Hoffmann sometimes speaks of a second-order judgment being infallibly justified, other times, of justifying the infallibility of the judgment. The latter issue is what I address in the text above. If the former is at issue, then Hoffmann’s point can be granted: I agree that the compositionality argument is a fallible means of justification. 14 For further elaboration, see §3 of Parent (2008).

94  Knowledge of Thought What is necessary under one set of assumptions, of course, need not be absolutely necessary. Yet under standard psycholinguistic assumptions, it is not wrongheaded to say, “A learnable language must be compositional.” That is so, even knowing that linguistic compositionality is a revisable, empirical hypothesis. Similarly, the Quinean reply is that it makes sense to say, “An occurrent second-order judgment must be true,” even if this takes some empirical contingencies for granted. In both instances, it is a case where a statement follows by necessity under some premises, though the premises themselves may well be contingent.15 Under either reply, the infallibility of second-order judgments is not itself infallibly known.16 Since the infallibility is premised on a revisable psychological law, the view is defeasible in the characteristic Quinean way. This alone does not rule out other ways one might conceivably know the infallibility, for example, by introspective means. Though again, my preferred route is to cite compositionality as an empirical hypothesis, in coming to know (fallibly) the infallibility of second-order judgments.

3.5  Looking Ahead Even though the present infallibilism has significant epistemological interest, it is limited in other respects, since it does not entail other theses which also seem important in an account of infallible self-knowledge. In particular, it does not entail anything like: (INFJ) If a subject has a second-order judgment that she judges p, then she has the first-order judgment that p. (INFQ) If a subject judges she has a quale of kind q, then she has a quale of kind q. Now in fact, I will eventually defend a version of (INFJ) in chapter 7. But I shall not be defending (INFQ) or the like in this book. In this chapter, I have tried to emphasize differences between the present view and Cartesian infallibilism. But like Descartes, some of the motivations clearly are epistemological. I take the infallibility of second-order judgments to be of interest in its own right; however, I am interested in infallibilism as a remedy to the Problem of Wayward Reflection (cf. chapter 1). Again, the

15 Plausibly, this can also be understood in terms of Lewis’ (1986) view, where quantifiers are restricted by context to a proper subset of possible worlds. It also calls to mind Lewis (1996), who emphasizes that some possibilities of error can be properly ignored in certain contexts, even though these possibilities may be relevant in other contexts. 16 Though infallibilism is known infallibly if the relevant modality is nomological throughout. (If second-order judgments are infallible as a matter of law, then as a matter of law, infallibilism cannot be wrong.) But of course, above I just mean to be reiterating the epistemic non-necessity of nomological necessities.

Infallibility about What One Thinks  95 worry is that the process of self-reflective inquiry won’t make rational sense if we lack secure access to our own thoughts. Even so, must our access to our own judgments be infallible? That may seem like overkill. Yet I take infallibilism to be better than just the “high reliability” of second-order judgments, as an antidote to the Wayward Reflection Problem. Indeed, if some errors creep into the reflection process, as per fallibilism, then those errors could snowball, as it were, leading to an increasingly skewed conception of what one believes. Accordingly, if reflective evaluation means considering how well a judgment stands up to one’s other beliefs, a skewed conception here would lead to a skewed evaluation process. The reflective process might then take one farther, rather than nearer, to an adequate first-personal assessment of one’s own judgments.17 Nevertheless, (INF1) alone does not wholly insure us against Wayward Reflection. Something stronger is needed, along the lines of (INFJ). Even so, (INF1) guarantees second-order judgments of a certain kind, namely, about the content of thoughts. So even though we have not discussed how we can discern our judging that p (as opposed to doubting, denying, wishing that p), we have shown how we infallibly judge that it is p we are thinking of, as opposed to some other twin-proposition p' (e.g., thoughts about water versus twin-water). In this respect, we have a partial solution to the Wayward Reflection Problem. But I have thus far ignored objections concerning externalism regarding content. The defense of (INF1), again, is meant to be independent of how one individuates content. Regardless, it is best to address the objections that arise under externalism. For one, that is because I happen to be a content externalist. But secondly, since externalism is a serious player on the current scene, it is dialectically best to remove any externalist barricades, so that (INF) creates the least amount of controversy. Addressing such obstacles is the task of the next two chapters.

References Bonjour, L. (1978). Can empirical knowledge have a foundation? American Philosophical Quarterly, 15, 1–13. Brown, J. (2004). Anti-individualism and knowledge. MIT Press. Burge, T. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, 73–121. ———. (1988). Individualism and self-knowledge. Journal of Philosophy, 85, 649–663. ———. (1996). Our entitlement to self-knowledge. Proceedings of the Aristotelian Society, 96, 91–116.

17 As a related argument for a similar conclusion, see Sidelle (2001). His contention is basically that any epistemic internalist should be an infallibilist about self-knowledge.

96  Knowledge of Thought Dennett, D. C. (1975). True believers: The intentional strategy and why it works. In A. F. Heath (Ed.), Scientific explanations: Papers based on Herbert Spencer Lectures given in the University of Oxford (pp. 53–75). Oxford UP. ———. (1991). Real patterns. Journal of Philosophy, 88, 21–51. Field, H. (1996). The a prioricity of logic. Proceedings of the Aristotelian Society, 36, 359–379. ———. (2005). Recent debates about the a priori. In T. Gendler & J. Hawthorne (Eds.), Oxford studies in epistemology vol. 1 (pp. 69–88). Oxford UP. Fodor, J. A. (1975). The language of thought. Harvard UP. ———. (2009). LOT2: The language of thought revisited. Oxford UP. Goldberg, S. (2006). Brown on self-knowledge and discriminability. Pacific Philosophical Quarterly, 87, 301–314. Heil, J. (1988). Privileged access. Mind, 98, 238–251. Hoffmann, G. (2011). Two kinds of a priori infallibility. Synthese, 181, 241–253. Lewis, D. K. (1986). On the plurality of worlds. Basil Blackwell. ———. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74(4), 549–567. Parent, T. (2008). Quine and logical truth. Erkenntnis, 68(1), 103–112. Perry, J. (1979). The problem of the essential indexical. Nous, 13, 3–21. Putnam, H. (1975). The meaning of “meaning”. In his Mind, language, and reality: Philosophical papers vol. 2 (pp. 215–271). Cambridge UP. Quine, W.V.O. (1951). Two dogmas of empiricism. Philosophical Review, 60(1), 20–43. ———. (1977). Intensions revisited. Midwest Studies in Philosophy, 2(1), 5–11. Sellars, W. (1974). Meaning as functional classification. Synthese, 27, 417–437. Sidelle, A. (2001). An argument that internalism requires infallibility. Philosophy and Phenomenological Research, 63(1), 163–179.

4 Objection 1: It’s Apriori that Water Exists

In the previous chapter, I argued that as a matter of nomological necessity, if I judge that I am thinking that water is wet, then I am thinking that water is wet. Still, I hesitated from saying that such infallibility suffices for knowledge of what I think. But as the reader might have guessed, it is my eventual aim to propound this, once other conditions are added in the next chapter. But potentially, the pursuit of such knowledge stops before it even gets started. At the end of the last chapter I endorsed content externalism: At minimum, this is the view that the content of a concept1 is determined partly by the social and/or physical environment. But it is a prominent worry that externalism is incompatible with knowing apriori the content of one’s own thoughts. Or more perspicuously, the worry is that externalism is incompatible with knowing, without any empirical investigation of the environment, what one is thinking. (For short, such knowing is termed “armchair selfknowledge,” following Davies 1998, 2000, etc.2) In the next chapter, I shall address the “slow switch” argument for this incompatibilist conclusion. But the aim in the present chapter is to oppose an incompatibilist argument originating in McKinsey (1991). In response to McKinsey, I shall be resisting his premise that the externalist thought experiments have armchair-knowable implications for the

1 A concept is assumed to be a type of mental particular which has a content. (This need not imply that such entities constitute a “language of thought.”) In contrast, Frege held that a concept just is a content (= an intension). (Thanks to Dorit Bar-On for stressing this to me.) Yet the anti-individulism literature has largely assumed the former view, and the paper follows this lead. (But to be clear, I assume that the water-concept has a water-content essentially, i.e., a content that determines an extension containing all and only H2O.) 2 The term ‘apriori’ is a contentious one in this context. For discussion, see Davies (1998), Nuccetelli (1999) and McLaughlin (2000; 2003). N.B., Macdonald (1995) helpfully separates two questions regarding SK: (q1) How is armchair knowledge of content possible, given externalism, and (q2) How is it that my knowledge of content is authoritative, given that it is gained from the armchair? Both this chapter and the next are focused on (q1); however, since the two chapters are defending an infallibilist version of armchair self-knowledge, an answer to (q2) is obviously implied.

98  Knowledge of Thought actual environment. I resist this by defending externalism about empty concepts from criticisms in Brown (1995), Boghossian (1997; 1998), and developed further more recently by Besson (2012).3 Interestingly, emptyconcept externalism has been advocated before (Stoneham 1999, Sawyer 2003, Goldberg 2006), but none of these advocates have confronted the Brown-Boghossian-Besson objections. The present chapter is a remedy to this situation.

4.1  The Objection First, a word of clarification. Externalism is often associated with, inter alia, Kripke’s (1972/1980) causal theory of reference. But externalists need not be causal theorists. One might instead be, for example, a Sellarsian (1954) inferential role semanticist—where content is determined by the inferences that feature the concept. (Apparently, “language-entry” and “languageexit” moves are partly individuated by the environmental objects in which the inferences originate or terminate.) Regardless, I shall not be concerned with Kripkean or Sellarsian externalism as such. Instead, “externalism” here shall concern Twin Earth externalism exclusively, i.e., the view supported by the usual Twin Earth thought experiments. (The experiments are here presumed familiar; see, e.g., Putnam 1975; Burge 1979.4) Suppose now that we have armchair self-knowledge. McKinsey’s (1991, 1994; 2002; 2003) objection is that if the usual arguments for externalism are sound, then one can reason from within the armchair as follows: (1) I have the concept water.5 (2) If I have the concept water, then water exists.

3 References to Brown and Boghossian will be to Brown (1995) and Boghossian (1997) unless otherwise stated. One might have predicted (falsely) that I would defend some version of “transmission failure” against McKinsey, given my interest in armchair reasoning. (Cf. Davies op. cits; Wright 2000, 2003, 2011, etc.) However, I do not think what I argue here is inconsistent with that line. Regardless, I would claim that the Davies-Wright response makes sense only if externalism about empty concepts is possible. Briefly, if externally individuated concepts have to be non-empty, then it is unfair to charge McKinsey with “presupposing” that such concepts are non-empty. (That would be like accusing McKinsey of “presupposing” that all thoughts are mental—of course they are!) So the Davies-Wright line, I think, already requires the possibility of externalism about empty concepts. Yet that is already enough to solve the McKinsey problem (or so I will argue); hence, there will be no need to venture into Davies-Wright territory. 4 Yet I will not be assuming Putnam’s (1975) attendant view that natural kind terms/concepts are indexical. For a forceful critique of that view, see Burge (1982). 5 Per usual, small caps are used to denote the corresponding mental representations. As in the previous chapter, this need not require Fodor’s (1975) language of thought hypothesis; it is merely to assume there are mental representations with the relevant contents.

It’s Apriori that Water Exists  99 (3) Water exists.6 Consider that (1) seems knowable from the armchair if the externalist has armchair self-knowledge generally.7 Further, since thought experimentation seems like an armchair activity, the Twin Earth thought experiments purportedly bestow armchair knowledge of (2).8 However, if both (1) and (2) are armchair-knowable, the externalist is then able to infer (3) immediately by modus ponens. Yet (3) should not be knowable just from the armchair.9 So apparently if the externalist thought experiments are cogent, armchair self-knowledge would result in absurdity. Twin Earth externalism thus appears incompatible with such knowledge.

4.2  Externalism about Empty Concepts However, I reject the claim that (2) is armchair knowable. But the most natural argument for this is not the one I press. The natural argument would be that externalism is a thesis about only non-empty concepts, i.e., concepts that refer (unlike phlogiston or unicorn). And whether a concept is empty is not knowable just from the armchair. If so, then according to this line, the externalist thought experiments simply presume that (3) is true—they do not demonstrate apriori that (3) is true (cf. McLaughlin & Tye 1998b). Yet Boghossian anticipates this objection and argues that externalism cannot be so limited to non-empty concepts. Briefly, that’s because the same sort of thought experiments work just as well on empty concepts.10 Thus 6 This is the reasoning as it appears in Boghossian. McKinsey’s version is mostly a notational variant of Brown’s, and Brown’s shall be addressed separately in §4.5. As all parties are aware, (3) and the consequent of (2) ideally would add “ . . . or existed at one time.” It is also notable that (in Boghossian’s version) water “aims” to refer to a natural kind. Yet these details will not bear on our discussion; so they are mentioned here only.   7 Naturally, some reject armchair self-knowledge, e.g., Carruthers (2011). Yet the present issue is whether externalism is incompatible with such knowledge. Whether we actually have it is a different matter (taken up in other chapters).   8 The argument is likely improved if this talk of armchair knowledge is replaced with talk of armchair warrant. (This in fact is how Davies, op. cit., and Wright 2000, 2003, etc., frame the issue. Though it deviates from what McKinsey-Brown-Boghossian say.) After all, it should still be a shock to externalism if the existence of water could be armchair warranted via Twin Earth experiments. And as the argument currently stands, it is a bit quixotic to suggest that philosophical thought experiments could bestow knowledge of anything. (The pessimistic metainduction may be too pessimistic in the case of science, but with philosophy, pessimism should probably be the default.)   9 Sawyer (1998) and Warfield (1998) are sometimes read as demuring; however, I think this is a bit misleading. See section 2.5 of Parent (2013) for discussion. Still, Putnam (1981) seems to be a genuine challenger to the claim. But unfortunately, I cannot delve into this dispute here. 10 Boghossian also notes that the externalist might say that an empty concept has no content. But I agree with his negative assessment of this option, so I pass it by.

100  Knowledge of Thought imagine “Dry Earth,” a planet exactly like Earth except, in place of water, there are only watery-looking holograms. (Though suppose Dry Earthians mistake them for a real kind.) Consider then Oscar on Earth (a normal, adult speaker) who is in the same narrow state as his doppelgänger “Doscar” on Dry Earth.11 Crucially, however, Doscar uses ‘water’ (and the concept it expresses) not in relation to water. Instead he apparently uses it to represent something nonexistent. Further, externalists often assume the following (see esp. Putnam 1975): (4) Content determines reference. Or in the Frege-Carnap terminology, this is the claim that intension determines extension. Given (4), then, if Doscar’s concept does not refer to water, it must have a different content. So even for empty concepts, content seems to be determined partly by environmental facts—including facts about what is not in the environment.12 My contention here is that this reasoning is perfectly cogent. Externalism about empty concepts may seem wildly unintuitive at first. Yet if (4) is assumed, sameness of content implies sameness of referent. So, given the difference in reference, it follows that the Earthian and Dry Earthian concepts have different contents. And thus even the content of the empty concept is not wholly determined “in the head.” Empty-concept externalism is possible since externalism is a claim only about content-determination; it does not imply that an actual referent is needed for a contentful concept.13 For the thought experiments show just that certain differences in the environment determine a different content. They do not illustrate that content comes and goes, depending on whether the object exists. Interestingly, Boghossian confirms this: All that Twin Earth externalism is committed to, strictly speaking, is the claim that, if the referent of a given word were different, the concept it would then express would be different, too. And that is consistent with the claim that the word would express a concept in a case where it fails

11 As usual in the literature, we ignore that people are mostly H2O; that’s an accident of the example. 12 Since Dry Earthians are fictitious, one may wonder what all this shows about actual concepts. But Boghossian notes the same kind of thought experiment can be run on our empty concepts (pp. 169, 170). 13 Burge (2006) construes externalism as the stronger claim that a water-thought is not just determined but constituted in part by water. Yet this “constitution” externalism does not strictly follow from the thought experiments. Constitution could explain why relevant changes in social/physical facts lead to changes in content, yet it is unclear whether this inference is to the best explanation.

It’s Apriori that Water Exists  101 to refer, provided that the concept it would there express is different from any it would express in a case where it does refer. (p. 171) It should be clear, then, that externalism does not require every contentful concept to have a referent. However, this seems to falsify Boghossian’s premise that (2) is armchairknowable. If the thought experiments work on both empty and non-empty concepts, then they would not reveal that water is non-empty. Granted, the Putnam-Burge thought experiments speak freely about water as something in our environment. But if the thought experiments apply regardless of whether water is empty, then this feature of the experiments is inessential. Again, the existence of water starts to look like merely an aposteriori assumption of the Putnam-Burge experiments. If so, then even if these thought experiments entail that water exists, they would not prove its existence.

*4.3  A Problem with Form? However, Boghossian recognizes this threat from empty-concept externalism, and offers two different objections to the view. The first takes the following as premises: (5) The concept water is atomic. (6) An empty concept is descriptive (hence, non-atomic). I grant that (5) is true; moreover, (6) is a natural view to have about empty concepts. The idea is that an unsatisfied linguistic (or mental) descriptor must wholly determine the content of an empty concept, given that the concept has no referent.14 As it happens, Segal (2000) and now Besson (2012) say that a descriptivist view like (6) already departs from externalism, since the content-fixing descriptor is invariant across environments. But to the contrary: Though the descriptor is constant, its content is not—if the describing-predicates (e.g., liquid, wet, etc.) also have wide contents. (Such “externalist descriptivism” has been noted before by Lepore and Loewer 1986, Devitt & Sterenly 1987/1999, and Taylor 1989.) To illustrate, suppose that ‘liquid’ is defined by Twin Earth experts as denoting liquids located specifically on Twin Earth. (N.B., Twin Earth is here imagined to be a planet in our universe, rather than an alternate possible world.) Then, since the extension excludes H2O, the concept expressed by ‘liquid’ on Twin Earth will have a different extension than on Earth. Hence, the concept expressed by ‘colorless, odorless liquid’

14 For the record, I myself do not accept descriptivism about empty concepts. (I suffer instead from a kind of Meinongianism; see Parent ms.). Yet I shall not contest (6) in what follows.

102  Knowledge of Thought will apply to H2O in one case but not in the other. Yet by (4), this difference in extension implies a difference in intension or content. That is so, even if the twins’ narrow states are the same. Indeed, Twin Oscar may be ignorant that his term ‘liquid’ is restricted to Twin Earth—and so the twins could be in the same narrow psychological state, despite the difference in content. At any rate, Boghossian thinks that if both (5) and (6) are adopted, emptyconcept externalism then becomes untenable: [T]he externalist [must] say that one and the same word, with one and the same functional role, may express an atomic concept under one set of external circumstances and a compound decompositional concept under another set of external circumstances. But it is hard to see how the compositionality of a concept could be a function of its external circumstances in this way. Compositionality, as I understand it, can only be a function of the internal syntax of a concept; it can’t supervene on external circumstances in the way that the compound proposal would require. (This is especially clear on a ‘language of thought’ picture of mental representation, but is independent of it.) (p. 172) Specifically, the form of the Dry Earthian concept will be descriptive and the Earthian one will be atomic. But a concept’s form should not be sensitive to the environment, since form is a purely syntactic affair.15 In reply, an externalist can turn the argument on its head, arguing that a concept’s form is partly determined by the environment. For logical form, at least, is not just a syntactic matter. After all, a formula f is the logical form of a concept only if f is equivalent to the concept. And equivalence is a semantic matter. Thus, if the semantics of the concept is determined partly by the environment, then whether the concept is equivalent to f is also determined partly by the environment. And thus, whether f is the logical form of the concept also depends partly on the environment. (See Korman 2006 for a similar view.16) In particular, whether ‘the colorless, odorless liquid . . . ’

15 McLaughlin and Tye (1998c) contend that this argument should prompt Boghossian to reject (5), whence he would see both empty and non-empty concepts as descriptive (p. 310). McLaughlin and Tye then go on to reject such descriptivism on the usual Kripkean grounds. Whether or not this is so, my claim is that Boghossian’s descriptivist premise at (6) can be left standing, and his case against externalist self-knowledge can be shown nondemonstrative regardless. 16 In Korman’s terms, the point is that the composition of a concept does not always supervene on one’s intrinsic properties. But unlike myself, Korman allows Boghossian’s assumption to stand that compositionality is merely syntactic. So although Korman and I both reject a type of supervenience, we may be talking about slightly different things. (Potentially, Korman’s rejection amounts to Ludlow’s odd externalism about syntax; see the next note.) Also, I would not necessarily embrace Korman’s counterexamples to the supervenience. Even if they show that the supervenience does not always obtain, they leave unanswered

It’s Apriori that Water Exists  103 is the logical form of a concept is determined partly by whether there is an environmental referent to provide for an alternate, non-descriptive content. Be that as it may, perhaps Boghossian’s concern is not really about logical form (where the concern is which formula in logical notation best expresses the concept). The real worry may be that under (5) and (6), the environment would partly determine the physical form or physical manifestation of the concept, for example, in a language of thought. (This fits better with his claim that only syntax is at issue.) The objection, then, would be that the physical realizer of a concept should not be type-distinct between the twins, given only the environmental differences. Yet assuming (5) and (6), that may be what empty-concept externalism suggests. Now one ineffective reply here would stress that the Dry Earth experiment stipulates sameness of narrow states. The two concepts then would have the same physical realizer ex hypothesi. Yet this reply fails since Boghossian could respond that the stipulation thus conceived is incoherent: Given the difference between empty and non-empty concepts, he might say it is impossible from the start for the “twins” to be in identical narrow states. However, a better reply is available. Recall that the rationale behind (6) was not that the physical manifestation of a concept must be such-and-such if the concept is empty. Rather, it was that an empty concept must have its content fixed by some unsatisfied linguistic or mental descriptor (roughly in the way that a stipulative definition does). Yet to say that some descriptive formula fixes the content in no way implies that the concept is physically realized in the brain as a descriptor, e.g., in a putative language of thought. In fact, externalists instead think of content as largely determined extra-cranially, by the linguistic community or by the experts therein. If so, then the twins’ concepts may well have their contents defined by different communal terms, without there being any physical difference between them. Granted, it is common in Chomskian linguistics to identify “logical form” with something realized in the brain.17 This likely owes to Chomsky’s internalism, where content supervenes on what’s in the head. If this were assumed, then a difference in content would imply a difference in internal state. But of course, the externalist denies the assumption; her view is precisely that a difference in content does not imply an internal difference. Before moving on, let us admit that since externalists are typically atomists about water, the switch to descriptivism in Doscar’s case can seem ad hoc (cf. Besson 2012, p. 413). However, once we recognize that logical form is partly a semantic matter, the switch has a semantic rationale. For if Oscar Boghossian’s allegation that in Doscar’s case the environmental difference should not result in a different logical form. 17 The Chomskian view apparently drives Ludlow (2003; 2011) to flirt with externalism about cognitive syntax, where the narrow states of the twins are different. Yet it is admitted that this view is awkward. Fortunately, as explained above, the Chomskian internalism in this is question-begging; thus the issue can be dispensed with more readily.

104  Knowledge of Thought and Doscar have concepts with different contents, one would hardly expect the concepts to have the same definition or logical form. True, a difference in form may not always be, specifically, a difference in atomic versus descriptive form. But what is ad hoc is to assume that the form of the concept never switches from atomic to descriptive. Since form is partly semantic, there would be no antecedent reason for the externalist to say that. Indeed, Doscar’s empty concept is exactly the sort of thing that would put pressure on that expectation.

*4.4  Concept Possession Notably, when responding to Corbí (1998), Boghossian (1998) offers a slightly different objection to empty-concept externalism. Still assuming (5) and (6), Boghossian writes: Suppose ‘water’ expresses an atomic concept A on Earth, and that on [Dry18] Earth it expresses the compound concept B and C and D and E. Then we would have to say external circumstances determine not only which concept a given word expresses, but also how many concepts a person has. Without any change in one’s cognitive or functional capacities, and just by being in one environment as opposed to another, one might have either one concept or five. (p. 259) The problem is that the number of concepts will differ between the twins. The Dry Earth concept is a composition, and the Earthian concept is a single conceptual atom. Potentially, however, this assumes that the Dry Earth concept expressed by ‘water’ just is the compound concept. But let us be clear: Even if a mental descriptor can fix the content of an empty concept, per (6), this does not mean the two are identical. Thus, pegasus may well be defined by the winged horse captured by bellerophon—yet children often have the pegasus concept without the descriptor concept. (Many children have never heard of Bellerophon.) Boghossian might respond that, even if pegasus is not identical to that descriptor, it still could be identical to some other descriptor. However, it is implausible that all children who possess the pegasus concept have the same description of Pegasus. So speaking quite generally, it seems that pegasus is not identical to any descriptor. As a result, the Dry Earthian does not automatically have more concepts than the Earthian. Still, even if all pegasus-users do not possess the same mental descriptor, perhaps the concept requires some descriptive representation.

18 Boghossian speaks of Twin Earth here, yet his point is less contentious if we consider the Dry Earth concept as a non-atomic twin concept.

It’s Apriori that Water Exists  105 That is (borrowing a term from Fodor and Lepore 1990), perhaps some version of “concept anatomism” is true. If so, then possessing the Dry Earthian concept would still demand more concepts than the Earthian one. And that would vindicate the basic worry—that the number of concepts possessed would depend on the environment. Regardless, even if the Dry Earth concept requires more concepts, it does not follow that the Earthian and Dry Earthian must have a different number of concepts. If possessing the Dry Earth concept requires possessing concepts like colourless, odorless, liquid, etc., the thought experiment can stipulate that the Earthian possesses those concepts anyway. So the environmental differences do not force us to say that the twins have a different number of concepts.

4.5  The “Revenge” Argument The final objection to empty-concept externalism, adapted from both Boghossian and Brown, claims that instead of (2), externalism would enable armchair knowledge of something weaker: (2') If I have the concept water, then either water exists, or I am part of a community which uses the concept water. Yet if (2') is known from the armchair along with (1), one can then deduce: (3') Either water exists, or I am part of a community which uses the concept water. But though (3') is weaker than (3), it too should not be known just from the armchair.19 Brown’s case for the armchair knowability of (2') starts by asking how the content of water could be fixed. Following Burge (1982), she finds it plausible that a chemist might fix its content by advancing the hypothesis ‘water = H2O’, even if the concept were empty. But what if the subject is agnostic about chemical essences? Then, it seems that (2') is indeed armchair knowable. For if it is armchair knowable that she has the concept water and that she is essence-agnostic, then she can infer immediately that its content is fixed by something else. In particular, it seems the content could only be

19 This type of argument was initially noted in Brueckner (1992), yet it was Brown who first defended it in earnest. But Brown (2001) expresses doubts about it, and then abandons it in her (2004). N.B., Brown’s version concerns Burge’s (1982) externalist view only (where further claims about content-fixing are explicit). Also, the argument comes with similar qualifications as those mentioned in n. 6; but again, such details are not important to the issues raised here.

106  Knowledge of Thought fixed by a communal definition or environmental kind. And that would be to conclude (2') within the armchair. However, McLaughlin and Tye (1998a, b, c) reply that nothing here shows that under externalism, a socially isolated, essence-agnostic definitely cannot possess a contentful empty concept. In fact, they make the stronger claim that no such argument is available: [T]here is no reason whatsoever to think that one cannot have a nonnatural kind concept, be agnostic about the necessary and sufficient conditions for its application, and yet not be a member of any linguistic community. (1998c, p. 213) But potentially, McLaughlin and Tye incur the burden of explaining how else the essence-agnostic “Lonely Dry Earthian” could learn the concept. They might retort, however, that the burden is on Brown: She must show that our Lonely Dry Earthian cannot possess the concept, so to complete her case against externalist self-knowledge. Yet instead of shifting the burden of proof, we can meet the explanatory challenge incurred by McLaughlin and Tye. For nothing rules out that the subject’s own descriptors fix the content of her empty concept, for example, ‘the colorless, odorless liquid . . . ’. Such “private language” may offend Wittgensteinians—though if the revenge argument were to require some kind of apriori private language argument, it would be much more contentious. (Wittgenstein certainly never said that his argument was apriori.20) Naturally, since the subject is indifferent about chemical essences, her descriptors do not fix exact necessary and sufficient conditions. But at least with empty concepts, normally there aren’t precise necessary and sufficient conditions.21 pegasus, for example, is not defined by a DNA sequence that constitutes Pegasus’s biological essence. Rather, its content seems fixed by Russellian descriptors like ‘the winged horse captured by Bellerophon’. Perhaps that means the content of pegasus is somewhat “vague.” But a vague content is still a content (McLaughlin & Tye 1998a, b), and nothing yet prevents the individual’s own descriptors from affixing such a content. And recall from earlier that a descriptive content need not be narrow. Again, the contents of the describing-predicates (e.g., ‘liquid’) might be wide, and then different environments would result in different descriptive contents.

20 It is worth pointing out that Wittgenstein was no Twin Earth externalist. He not only rejects the scientific realism implicit in the view, but also denies that a term like ‘water’ has a context-invariant meaning (Glock & Preston 1995). 21 Brown (2001) recognizes this as well. In reply, she stipulates that one is “essence agnostic” only if there is a real essence to be known. Thus the subject fails to qualify as ‘essence agnostic’ in the case of pegasus. However, as Brown (2004) admirably concedes, it is then dubious whether one can know apriori that one is “essence agnostic.”

It’s Apriori that Water Exists  107 Some may insist, however, that the externalist cannot appeal just to the subject’s own descriptors to fix content. “Look, externalism says that the contents of her concepts depend on the social and/or physical environment— and that means they can’t just depend on the individual. After all, if such environmental dependence is the rule, then there must be a social/physical environment for the contents to depend on.” And that leads us back to (2'). This I suspect is what drives a lot of intuitions against externalist armchair self-knowledge. But the thinking equivocates on how content “depends on” the environment. True, the Twin Earth experiments suggest that content depends on external factors in influencing which content a concept has. But they do not show that whether a concept has any content depends on extra-cranial factors. The experiments suggest merely that an environmental condition is part of the identity condition on the content of the concept expressed by ‘water’. Nothing is shown about its existence condition, the condition on which that concept has a content at all (pace McGinn 1989). If so, then it is consistent with externalism that empty concepts can have contents absent a social environment. The claim parallels the point from §4.2. There, we saw Boghossian admit that the externalist does not need a worldly referent for a contentful concept. But in the same way, the experiments do not reveal that a linguistic community is necessary for a contentful empty concept. For they do not describe cases where social isolation robs a concept of its content. Rather, they just illustrate how a different sociolinguistics can determine a different content. But for all that has been said, some externalist views may well imply (2'). Such views are often based on an independently motivated semantics, for example, teleosemantics, Davidsonian interpretivism, or what have you. And these semantic frameworks may imply that a concept has a content only if it has a worldly referent or a communal use. But if so, that would be due to the additional theory brought to bear on the matter, some of which may be empirical. So even if many externalist theories entail (2'), that does not mean the Twin Earth reasoning demonstrates (2') apriori or the like. Twin Earth reasoning as such neither assumes nor otherwise implies (2').

4.6  Transition to Chapter 5 In the foregoing, we have seen that (1)–(3) poses no problem for externalist armchair self-knowledge, once externalism is extended to empty concepts. In fact, Boghossian’s Dry Earth thought experiment provides a serious case in its favor, and it is virtually demanded by the Frege-Carnap claim at (4). Moreover, it was argued that externalist thought experiments do not guarantee the truth of (2) or (2'), much less in armchair-knowledge of these things. So it seems the externalist is unable to know just from the armchair any contingent, empirical facts—as it should be.

108  Knowledge of Thought But in fact, there remains one possibility for rescuing the incompatibilist’s argument. On this tack, she first concedes that (2) is not armchair-­knowable. She then argues, however, that armchair-knowledge of (1) becomes impossible (Gertler 2004). After all, if (2) is not armchair-knowable, then Oscar apparently cannot discriminate from the armchair between his water-­concept and the Dry Earthian concept. Hence, for all the subject can tell within the armchair, she may have the concept water or the concept dwater. So it seems incorrect to say that (1) is something she knows. In fact, this raises the extremely important question of whether such discriminations are necessary for self-knowledge. And it turns out that such a “discrimination requirement” is the central issue of slow switching arguments. (Indeed, I would argue that Gertler’s angle on the McKinsey argument renders it into a special case of a slow switch argument.22) It is to these issues that I shall now turn.

References Besson, C. (2012). Externalism and empty natural kind terms. Erkenntnis, 76, 403–425. Boghossian, P. (1997/1998). What the externalist can know a priori. Proceedings of the Aristotelian Society, 97, 161–175. Reprinted in Philosophical Issues, 9, 197–211. ———. (1998). Reply to commentators. Philosophical Issues, 9, 253–260. Brown, J. (1995). The incompatibility of anti-individualism and privileged access. Analysis, 55(3), 149–156. ———. (2001). Anti-individualism and agnosticism. Analysis, 61(3), 213–224. ———. (2004). Anti-individualism and knowledge. MIT Press. Brueckner, A. (1992). What an anti-individualist knows a priori. Analysis, 52, 111–118. Burge, T. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, 73–121. ———. (1982). Other bodies. In A. Woodfield (Ed.), Thought and object (pp. 97– 120). Oxford UP. ———. (2006). Introduction. In his Foundations of mind: Philosophical papers volume 2 (pp. 1–31). Oxford UP. Carruthers, P. (2011). The opacity of mind: An integrative theory of self-knowledge. Oxford UP. Corbí, J. (1998). A challenge to Boghossian’s incompatibilist argument. Philosophical Issues, 9, 231–242. Davies, M. (1998). Externalism, architecturalism, and epistemic warrant. In C. Wright, B. Smith & C. Macdonald (Eds.), Knowing our own minds (pp. 321– 362). Oxford UP.

22 Thus, if Gertler’s is the only viable version of McKinsey’s reductio, as I am inclined to think, then contra much of the literature, it turns out that the McKinsey argument ultimately is a slow switch argument! There wouldn’t be two distinct incompatibilist arguments after all. But I need not insist on this for our purposes.

It’s Apriori that Water Exists  109 ———. (2000). Externalism and armchair knowledge. In P. Boghossian & C. Peacocke (Eds.), New essays on the a priori (pp. 384–414). Oxford UP. Devitt, M. & Sterelny, K. (1987/1999). Language and reality: An introduction to the philosophy of language, 2nd edition. MIT Press. Fodor, J. A. (1975). The language of thought. New York: Thomas Y. Crowell Co. Fodor, J. A. & Lepore, E. (1990). Holism: A shopper’s guide. Blackwell. Gertler, B. (2004). We can’t know a priori that H2O exists, but can we know that water does? Analysis, 64(1), 44–47. Glock, H. & Preston, J. M. (1995). Externalism and first-person authority. The Monist, 78, 515–533. Goldberg, S. (2006). An anti-individualist semantics for “empty” kind terms. Grazer Philosophische Studien, 70(1), 147–168. Korman, D. (2006). What externalists should say about Dry Earth. Journal of Philosophy, 103, 503–520. Kripke, S. A. (1972/1980). Naming and necessity. Harvard UP. Lepore, E. & Loewer, B. (1986). Solipsist semantics. Midwest Studies in Philosophy, 10, 595–614. Ludlow, P. (2003). Externalism, logical form, and linguistic intentions. In A. Barber (Ed.), Epistemology of language (pp. 399–414). Oxford UP. ———. (2011). Philosophy of generative linguistics. Oxford UP. Macdonald, C. (1995). Externalism and first-person authority. Synthese, 104(1), 99–122. McGinn, C. (1989). Mental content. Blackwell. McKinsey, M. (1991). Anti-individualism and privileged access. Analysis, 51, 9–16. ———. (1994). Accepting the consequences of anti-individualism. Analysis, 54(2), 124–128. ———. (2002). Forms of externalism and privileged access. Philosophical Perspectives, 16, 199–224. ———. (2003). Transmission of warrant and the closure of apriority. In S. Nuccetelli (Ed.), (pp. 97–116). McLaughlin, B. (2000). Self-knowledge, externalism, and skepticism. Proceedings of the Aristotelian Society, 74(74), 93–118. ———. (2003). McKinsey’s challenge, warrant transmission, and skepticism. In S. Nuccetelli (Ed.), (pp. 79–96). McLaughlin, B. & Tye, M. (1998a). Externalism, twin earth, and self-knowledge. In C. Wright, B. Smith & C. Macdonald (Eds.), Knowing our own minds (pp. 285– 320). Oxford UP. ———. (1998b). Is content-externalism compatible with privileged access? Philosophical Review, 107(3), 349–380. ———. (1998c). The Brown-McKinsey charge of inconsistency. In P. Ludlow & N. Martin (Eds.), Externalism and self-knowledge (pp. 207–214). Stanford: CSLI Press. Nuccetelli, S. (1999). What anti-individualists cannot know a priori. Analysis, 59, 48–51. ———. (Ed.) (2003). New essays on semantic externalism and self-knowledge. MIT Press. Parent, T. (ms.). Conservative Meinongianism, unpublished draft. Available at http:// tparent.net/conservativeMeinong.pdf. ———. (2013). Externalism and self-knowledge. In E. Zalta (Ed.), The Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/entries/self-knowledgeexternalism/

110  Knowledge of Thought Putnam, H. (1975). The meaning of “meaning”. In his Mind, language, and reality: Philosophical papers vol. 2 (pp. 215–271). Cambridge UP. ———. (1981). Brains in a vat. In his Reason, truth, and history (pp. 1–21). Cambridge UP. Sawyer, S. (1998). Privileged access to the world. Australasian Journal of Philosophy, 76, 523–533. ———. (2003). Sufficient absences. Analysis, 63, 202–208. Segal, G. (2000). A slim book on narrow content. MIT Press. Sellars, W. (1954). Some reflections on language games. Philosophy of Science, 21, 204–222 Stoneham, T. (1999). Boghossian on empty natural kind concepts. Proceedings of the Aristotelian Society, 99, 119–122. Taylor, K. (1989). Narrow content functionalism and the mind-body problem. Noûs, 23, 355–372. Warfield, T. (1998). A priori knowledge of the world: Knowing the world by knowing our minds. Philosophical Studies, 92, 127–147. Wright, C. (2000). Cogency and question-begging: Some reflections on McKinsey’s paradox and Putnam’s proof. Philosophical Issues, 10, 140–163. ———. (2003). Some reflections on the acquisition of warrant by inference. In S. Nuccetelli (Ed.), (pp. 57–77). ———. (2011). McKinsey one more time. In A. Hatzimoysis (Ed.), Self-knowledge (pp. 80–104), Oxford UP.

5 Objection 2: Thought Switching

This chapter continues the defense of compatibilism, the view that externalism is compatible with knowing, while confined to the armchair, what one thinks. Here, I do so by defending an externalist’s introspective knowledge of thought content (a.k.a. introspective “self-knowledge”) from an especially prominent incompatibilist argument.1 This is the infamous slow switch argument. Following Brown (2004) and others, many believe that the slow switch argument is no threat to an externalist’s ordinary introspective selfknowledge. That’s because ordinarily, “slow switching” is an idle skeptical possibility, an irrelevant alternative. But it is often implicit that the argument is a threat if skeptical possibilities are ever taken seriously. I shall argue, however, that even this gives the argument too much credit. We will see that slow switches do not rule out all of an externalist’s introspective self-knowledge, even in a skeptical context.

5.1  The Objection Consider Oscar (a normal, adult English speaker) who is unwittingly switched from Earth to Twin Earth (perhaps while asleep). The intuition

1 I vary between talk of “knowing what content a thought has” and “knowing what one thinks.” The former describes most clearly what is at issue, but it is sometimes awkward rhetorically. When that is so, the latter is used in place of the former. Also: Talk of “introspection” may impute a kind of recognition to a self-knowing act. But a self-verifying judgment that one thinks that p does not reflect a “cognitive achievement.” Even so, the judgment will not count as knowledge, I think, unless there is an “introspective awareness” of its selfverifying status (see §5.2). So even though self-verifying judgments are central, what follows remains in that respect a defense of “introspective” self-knowledge.   Talk of “introspective knowledge” is unfortunate in a further way, since it suggests that the individual can achieve justification independently of her epistemic community. But as a social epistemologist (cf. §1.3), I think “peer review” or the like is required for the type of “justification” we care about. However, these points shall be ignored in what follows. The main question is whether externalist introspective self-knowledge is possible in a skeptical context. That can be investigated to a great extent without pronouncing on whether justification is social.

112  Knowledge of Thought is that, upon first arriving, Oscar will continue to refer to H2O when using ‘water’ to express his thoughts. But after a time, it seems he will come to refer to XYZ instead, even if he remains ignorant of the switch (Burge 1988, Boghossian 1989). Given externalism, however, a different referent suffices for a different content. Yet in these conditions, Oscar seems unable to introspectively know what thought-content he has. The reasoning here can be reconstructed as follows:2 (1) If externalism is true, then Oscar introspectively knows what content his thought has only if he can introspectively know whether his thought is de H2O versus XYZ.3 [Assumption] (2) Oscar cannot introspectively know whether his thought is de H2O vs. XYZ. [Assumption] (3) So, if externalism is true, Oscar does not introspectively know what content his thought has. [From (1), (2)] Arguably, moreover, the possibility of “switched contents” is not limited to the realm of fiction. As per Ludlow (1995), unnoticed shifts in thought actually seem to occur with some regularity, e.g., when a speaker moves between different sub-groups of the linguistic community. Different groups of English speakers might construe the word ‘chicory’ or ‘pragmatism’ in different ways; thus, what you express by the word can slightly differ on different occasions without notice. (Some may not constitute cases of slow switches, but the point is that thought switches would actually occur.) Ludlow’s examples may be contentious—yet given the vast range of possibilities for human psychology, I would not rule out apriori that thought switches really happen. In this respect, it is not only content externalism, but also just the possibilities for human psychology that animate the Problem of Wayward Reflection

2 There is probably no one argument that counts as the slow switch argument. Below is the argument as it occurs in Burge (1988, p. 653), where the notion of “knowing-what” is featured rather than “knowing-that.” (This contrasts with the argument as reconstructed in, e.g., Ludlow 1995.) Here I focus only on the original “knowing-what” version, but I think my remarks also carry over to the “knowing-that” version. Briefly, that’s because the latter justifies the premise that Oscar does not know that p by Oscar’s not knowing what he thinks. Per section 5.4 below, my claim would be that this is cogent on some precisifications of ‘knowing-what’, yet not on others. 3 A thought is “de H2O” when its object is H2O; and it is such even if it is opaquely ascribed as a “water” thought. Thus, ‘Bob thinks that water is wet’ does not entail ‘Bob thinks that H2O is wet’, yet both sentences still ascribe thoughts that are de water/H2O. Also, n.b., ‘knowing-whether’ is not straightforward either (see Lewis 1982), yet it is typically not purpose-relative like ‘knowing-what’ (cf. §5.4). Here, ‘knowing-whether’ can be seen as shorthand for: Either Oscar knows that his thought is de H2O and not XYZ, or he knows that his thought is de XYZ and not H2O. (Compare with Kallestrup 2009 and Stanley 2011 on Schaffer’s 2007 “convergent knowledge” problem.)

Thought Switching  113 via Thought Switching (cf. chapter 1). Either way, however, stealthy thought switches would threaten the rationality of reflection. If a switch occurs during reflection, it would amount to an unnoticed change in topic. A person might start by reflecting on the “morality” of abortion—but due to a switch, ends up drawing conclusions about “twin morality,” without her realizing it. This type of equivocation is explored in more detail in the next chapter. For now, the point is that thought switches seem cut into the rationality of reflective inquiry, whether they are made possible by content externalism or just by the wealth of unknown possibilities for human psychology. However, for concision, I shall henceforth focus just on switches owing content externalism (although my remarks would apply to switches made relevant by any other means). In response to (1)–(3), Burge (1988; 1998, etc.) argues that a variety of self-knowledge indicated in chapter 3 is unaffected by thought switches. Consider a case where I produce the judgment: (SV) I am thinking forthwith that water is wet.4 As per chapter 3, the judgment is distinctive in that it is Self-Verifying: If I judge “I am thinking forthwith that water is wet,” then it seems I really am thinking that.5 The judgment seems self-verifying, moreover, regardless of any shift in thought-content. Nevertheless, Brown (2004) objects that an (SV)-judgment is not enough for knowledge. Her reason is that per Alvin Goldman (1976; 1986), knowledge requires discrimination among alternatives, and Oscar cannot introspectively discriminate between H2O- and XYZ-contents. Yet as mentioned earlier, Brown claims that Oscar need only discriminate among relevant alternatives, and skeptical “twin” alternatives are rarely relevant.6 So normally, the externalist can claim introspective self-knowledge. In reply, however, Goldberg (2006) deems this to be too concessive to the skeptic. For Goldberg, there is a persistent Cartesian intuition that some self-knowledge should not be affected by skeptical possibilities concerning the environment (p. 311).7 Or at least, it seems an (SV)-judgment should be resistant to such skeptical challenges. (There is evidence that Burge 1988

4 Burge uses the self-referring ‘with this very thought’ in lieu of ‘forthwith’. However, selfreference is potentially problematic here (see §3.2). In contrast, ‘forthwith’ does not selfrefer, but rather refers only to the first-order thought that follows it. (Also, ‘I’ in (SV) should be read as an essential indexical; see Perry 1979. Though I doubt that anything genuinely indexical explains Perry’s examples; see Millikan 2001, 2012.) 5 Here, ‘think’ does not mean ‘believe’. Rather, to ‘think’ that p is to have some propositional attitude with the content p, but it does not imply any particular attitude. 6 Brown perhaps makes the point most forcefully, though there is some precedent in Warfield (1992). 7 A similar intuition is expressed in Gertler (2000).

114  Knowledge of Thought would concur as well.8) Yet Brown’s view apparently does not respect that. My aim, accordingly, is to resist the slow switch argument without surrendering (SV)-judgments as a type of “skeptic-immune” self-knowledge. Still, we must respect that the subject cannot introspect the answer to the following: (Q) Is your thought de H2O or de XYZ? Brueckner (1999) notes that this inability is unsurprising, for it would require empirical knowledge of chemistry.9 Regardless, if we accept Goldberg’s intuition that the (SV)-judgment should be skeptic-immune, then the inability creates a concern. It suggests that, for all the subject can introspect in a skeptical context, her thought may host one of two different contents (assuming externalism). In which case, apparently, the subject lacks skepticimmune knowledge of thought-content (or just: Knowledge of content, with a capital ‘K’). The issue can be summarized in a more refined type of slow switch argument: (1') If externalism is true, then a subject introspectively Knows what content her thought has only if she can introspectively Know the answer to (Q). [Assumption] (2') A subject cannot introspectively Know the answer to (Q). [Assumption] (3') So, if externalism is true, the subject does not introspectively Know what content her thought has. [From (1'), (2')] However, my goal here is to argue that an externalist can reasonably resist (1')—thus allowing the possibility of Burge-style, introspective self-knowledge.10

  8 E.g., Burge (1988) likens the “directness and certainty” of (SV) to the cogito (p. 649).   9 Perhaps the subject can answer (Q), if she knows that “XYZ” is just a fiction made up by Putnam in the ’70s. But ‘XYZ’ is here acting as a placeholder for any conceivable “twin” of H2O. Brueckner (op. cit.) also rightly notes that (Q) is not the same question as the more simplistic “Is your thought de water or de some kind of ‘fake water’?” After all, Oscar can know immediately that the answer to the latter has to be ‘water’ (regardless of whether he speaks English or Twin-English). Still, this does not mean he is in a position to answer (Q). 10 An anonymous referee poignantly asks why (1') is any less trouble for a content internalist (if ‘externalism’ in the antecedent is replaced by ‘internalism’). This is an important question; yet we need not pursue it here. Granted, if the trouble arises on both sides, then it does not provide an advantage to internalism. Yet this does not mean the slow switch argument loses all interest. Even if the problem is not unique to the externalist, it is still a problem that ideally she would want resolved. Besides, there is dispute among externalists on whether a conclusion like (3') must be accepted. (E.g., Millikan’s 1993 externalism claims that it must.) In this connection too, the present paper argues that an externalist need not acquiesce to (3') or related skeptical conclusions about introspective self-knowledge.

Thought Switching  115

5.2  Introspective Awareness and Skeptic-Immunity Goldberg (2006) resists (1') as well, for he holds that under externalism, the (SV)-judgment can count as self-Knowledge if: (a) the judgment is guaranteed to be true, (b) the subject is aware of the basis of this guarantee, and (c) the subject is aware of what this guarantee is a guarantee of, namely, the truth of the (SV)-judgment. (paraphrasing p. 308) Crucially, this position does not require an ability to introspect the answer to (Q), even under externalism. (Accordingly, such Knowledge is individuated intensionally—one can Know that one has a water-thought without Knowing that one has an H2O-thought.) Yet per Falvey and Owens (1994), Goldberg (1999) concedes that this ability would be necessary for discriminatory Knowledge between a content and a twin. Still, he maintains that the (SV)-judgment can qualify as some sort of self-Knowledge, even if one is unable to answer (Q) via introspection. Let us call such a position the “Introspective Awareness View.” The core claim is that a self-verifying judgment counts as self-Knowledge, provided a certain “introspective awareness” of the judgment. (This awareness can act as the key justificatory ingredient to such Knowledge.) The View is dialectically important, since it does not insist on the requirement at (1'). Hence, if viable, it would mean that premise (1') of the refined slow switch argument is undermotivated. Yet wouldn’t the Introspective Awareness View simply beg the question? The claim is not that the View shows slow switch arguments to be unsound. Rather, it is that (1') is defended insufficiently (regardless of whether it is true).11 If so, then contra much of the literature, slow switch arguments fail to debunk externalist self-knowledge—even in a skeptical context. Some version of the View strikes me as defensible. As it stands, however, there are three objections which must be addressed.12 First, even

11 Does a skeptic need to defend (1')? He may claim that the burden is on me to rule it out. For if its truth is epistemically possible, then it is similarly possible, per (1')–(3'), that I don’t Know what I think. And that suggests I don’t really Know. This last bit may assume a KK-thesis, but perhaps that too is my burden to rule out. However, I cannot settle disputes about KK here. But I would argue that they turn the skeptic into a mere “simple skeptic,” in DeRose’s (1995) sense, i.e., a skeptic who dogmatically (!) says that we don’t know. For KK alone suffices for a skeptical regress. So to presume KK without argument is really just to prejudge that we don’t know. 12 A fourth objection is that the View requires too much of the ordinary person to self-know. But the View gives only a sufficient condition for one type of self-knowledge; it is not stating a necessary condition on all self-knowledge. Also, as a fifth objection, some argue that if we know what we think, then we know that we think. I address this in chapter 9; briefly, the reply is that our talk of “thinking” may only be instrumental (cf. Dennett 1975).

116  Knowledge of Thought if conditions (a)–(c) are met, it seems hasty to conclude that an (SV)judgment is immune to skeptical doubt. After all, if Descartes doubts his ability to add two and two, then he clearly can doubt his introspective awareness of the (SV)-judgment. Yet if we bypass these skeptical doubts about introspection, then it is not implausible that the judgment is otherwise skeptic-immune. Admittedly, this means “skeptic-immunity” is limited to a proper subset of skeptics. But such limited immunity is still relevant, since Goldberg’s intuition concerns external world skepticism, specifically. It does not concern skepticism about the reliability of introspection. Nonetheless, since my ultimate concern is with the Problem of Wayward Reflection, it may seem too convenient for me to bracket such doubts. But I am putting them aside only temporarily; we shall return to them in later chapters. After all, those doubts are fortified when the Problem is motivated by the actual empirical evidence for widespread introspective error. Yet the slow switch argument does not challenge whether one can introspect the self-verifying status of the (SV)-judgment. For the main question concerns whether discriminating among twin contents is needed for a self-verifying judgment to count as Knowledge. So within the slow switch dispute, it begs no questions to assume the introspective awareness mentioned in (b) and (c). (Put another way: This chapter concerns only Wayward Reflection via Thought Switching. Chapters 10 and 11 shall bear on introspective error when discussing Wayward Reflection via Attitude Switching. See chapter 1 for this distinction between the two versions of Waywardness.)

5.3  Knowing-That versus Knowing-What In any event, there is a second objection which forces us to revise the Introspective Awareness View in significant ways. Yet it is an objection to any Burge-style view on slow switch arguments, which seems thus far unnoticed in the literature.13 Observe that such views contend that the subject knows that she is thinking that water is wet. (Ignore for now the distinction between “knowing” and “Knowing.”) In the argument at (1')–(3'), however, the conclusion is not that the subject fails to know that she is thinking that water is wet. Rather, it is that she fails to know what content her thought has.

13 I have since discovered that the knowing-what vs. knowing-that distinction has been noted (albeit briefly) by Goldberg (2003) and by Burge (1998). Yet Goldberg claims that slow switching debunks “knowing what” one thinks (although not “knowing that”) (p. 253). In contrast, I argue that some “knowing-what” remains viable. Burge concurs with this, but supports it via his theory of preservative memory (p. 362). Whereas, I argue the point via the purpose-relativity of “knowing-what.”

Thought Switching  117 5.3.1  Knowing-What and Informativeness The difference between knowing-that and knowing-what is not merely in the surface grammar. Consider that if a subject knows that (say) roentgenium = roentgenium, it may remain unclear whether she “knows what” roentgenium is. After all, besides such truisms, she may know of roentgenium only that it has some atomic number over 100. In which case, although the subject knows something about the stuff, it is tendentious to conclude she knows what it is. Similarly, if the subject knows that she is thinking that water is wet, this does not mean that she knows what content her thought has. She may well know something about a “water-content” (e.g., that her current thought has it). Yet she still might lack sufficient background to know what a watercontent is in the first place. Especially when talking with a skeptic, the subject’s knowledge of water may be as “thin” as the neophyte’s knowledge of roentgenium. In reply, the externalist should restrict the Introspective Awareness View to (SV)-judgments where the subject is already substantially informed about water. Thus, the View would be that an (SV)-judgment can suffice for knowing what one thinks, but only if substantive facts are already known about water, for example, that it is the colorless, odorless liquid found in lakes and rivers in the environment. (In this, I am assuming that knowing what a water-content is requires substantive knowledge of water, the stuff. Yet that is just because in the present context, ‘water-content’ is just shorthand for ‘content of a thought that is de water’.) So plausibly, in the case of the (SV)-judgment, an externalist knows what content the first-order thought has only if she knows what water is. And, like knowing what roentgenium is, knowing what water is seems to require knowing something nontrivial about the stuff. We thus arrive at the requirement that for the (SV)-judgment to qualify the subject as “knowing-what,” she must antecedently know something informative about water. (My thanks to an anonymous referee for pressing me to clarify this.) However, this requirement might leave one’s knowing-what vulnerable to skepticism. After all, it may require knowing that water is the colorless, odorless liquid found in lakes and rivers in the environment. But a subject might “bracket” the ontologically loaded portion of her background knowledge, and still plausibly know enough to know what. After all, she could still know that water is a possible colorless, odorless, liquid in lakes and rivers (where ‘possible’ is some kind of epistemic modal).14 Such

14 If there is concern that the description here is neither definite nor rigid, we can instead talk of “the possible liquid that appears to me as if it is in lakes, etc.;” cf. Searle (1983), ch. 8. Yet unlike Searle, I am neutral on whether this fixes a narrow content. (The (SV)-judgment is self-verifying regardless.) N.B., the quantifier in ‘the liquid’ must also be read as ontologically netural; see Azzouni (2007) for more on neutral quantifiers. Besides eluding skeptical

118  Knowledge of Thought background, along with the (SV)-judgment, may well suffice for knowing what she thinks. *5.3.2  Supplementary Remarks Fortuitously, buttressing the (SV)-judgment with such background also helps counter a further objection. (My thanks to the same referee for pointing this out.) The objection is that an (SV)-judgment is “cognitively insubstantial,” in the manner of ‘I am here now’ (Boghossian 1989, p. 18; Gertler 2000; Farkas 2008, ch. 6; Wikforss 2008). When I think, “I am here now,” the thought is invariably true, of course. But that is because it is designed to represent where I am at the time, wherever that may be. Taken by itself, then, it is uninformative of the particular locale I occupy. Similarly, the complaint against an (SV)-judgment has been that, taken in isolation, it is hardly informative of what I think. It represents by design whatever thought occurs after “I am thinking forthwith that . . . ”. Yet to know what she thinks, the subject is now required to possess topically relevant, substantive background knowledge. Intuitively, such background knowledge makes the (SV)-judgment “cognitively substantial;” the concept of water is not simply a “cognitive placeholder” for whatever it is I am thinking about. Some might add that the subject’s background knowledge defines a kind of narrow content, a content that is individuated independently of the environment. After all, descriptors like ‘the colorless, odorless liquid . . . ’ are often seen as characterizing narrow contents. (Witness ‘primary intensions’ in Chalmers 1996; 2002.15) But here I adopt neutrality on the matter. We are not forced to assume that the descriptors characterize a narrow content—for regardless, her descriptive knowledge still partly determines whether there is a specific epistemic (vs. semantic) property. In particular, it partly fixes whether the subject has a type of epistemic qualification, viz., whether she knows enough about x in order to “know what” it is. For if she has enough background knowledge about water (“water is a colorless, odorless liquid . . . ”), then plausibly, this along with the (SV)-judgment (in conditions (a)–(c)) qualifies her as “knowing what” she thinks. That seems defensible, whether or not her descriptors define a narrow content.

5.4  Purpose-Relativity and Knowing-What The third and final objection to the Introspective Awareness View is that it still fails to secure skeptic-immune knowledge (i.e., Knowledge). Again, the

doubts, neutral quantification is needed to avoid reviving McKinsey-type problems discussed in chapter 4. 15 Yet to be clear, primary intensions are not identified with descriptive contents; they are instead defined on worlds “considered as actual.”

Thought Switching  119 view does not require an ability to answer (Q) in order to Know what. But her inability indicates that her thought could harbor one of two different contents, for all she Knows (assuming externalism). After all, her inability reflects that she fails to Know that her thought is de H2O rather than XYZ. But given externalism, her thought would have a different content between the two cases—and that indicates a clear sense in which she does not Know what she thinks. Surprisingly, however, Knowing-what might not always require an ability to discriminate between the two contents thus described, even in a skeptical context. But to appreciate this, we first need to be clear that the semantics of ‘know what’ attributions are highly purpose-relative (Boër & Lycan 1986; Ginzburg 1995; 2011; Ginzburg & Sag 2000; Hookway 2008; see also DeRose’s 2009 ch. 2 appendix). This is to say that the truth-condition of a ‘knowing-what’ ascription is greatly determined by the purposes set in a given context.16 Consider, for instance, that we would ordinarily say a five-year-old “knows what” water is. After all, the child knows enough to fetch a glass of water, and so forth. Yet if asked what water is on a chemistry exam, the same child will not count as “knowing-what.” For he could only guess at the correct chemical formula among the answer choices. So intuitively, the child knows what water is for ordinary purposes, though not for the purposes of the exam. Note that the purpose-relativity view is importantly different from more familiar contextualist views in epistemology, as defended by Cohen (1986; 1998), DeRose (1995; 2009), and Lewis (1996). Such contextualists maintain, roughly, that the truth-condition on a knowledge ascription is affected by how high the evidential “standards” are in a context. Yet evidential standards are different from one’s purposes in a context. A child may “know what” for the purposes of fetching some water, though not with sufficient evidence to satisfy the skeptic. Naturally, there is some similarity between contextualism and the purpose-relativity view. Both help determine which possibilities are relevant to achieving knowledge. But the rule that determines relevance is different. The possibility that water = XYZ may be relevant because of the skeptic’s standards for evidence. Yet it may also be relevant simply because ‘XYZ’ is one of three answer choices on the chemistry exam. So if the purpose is to succeed on the exam, this requires us to eliminate “XYZ” as the answer, even if doing so does not demand much of any evidence.

16 Against Boër and Lycan, Braun (2006) argues that the purpose-relativity of ‘knowing-what’ attributions are not in the semantics, but rather in the pragmatics or speech-act content. This is not the place to adjudicate this dispute, yet one might note that purpose-relativity seems characteristic of “knowing-what” attributions in thought as well, where the notion of a speech-act has no application. (Cf. Heal 1994, Green 2005 on Moore’s paradox.) For more on the Boër-Lycan v. Braun dispute, see §3 of Parent (2014).

120  Knowledge of Thought Philosophers in the analytic tradition often react negatively to any mention of interest-relativity, and the purpose-relativity view has met such resistance (see, e.g., Sterelny 1988, Kripke 2011, n. 21). Yet on the present view, the truth of a response to (Q) is not relative to interests or purposes. Truths and the corresponding facts about water obtain regardless of our interests, and ditto with whether a subject knows the truth. Instead, the basic observation is that when one’s purpose is to answer (Q), not just any information about water is of interest. Yet which information is of interest is, trivially, relative to one’s interests—e.g., relative to the goal of answering (Q). So when we have this goal in mind, we “flag” specific people as possessing the information we want; we say that they “know what” relative to this purpose. This is the only way, I submit, in which knowing-what depends on interests or purposes. Certainly, nothing suggests that knowing-what depends only on our interests. If the purpose-relative view is right, then whether a subject “knows what” she thinks will similarly depend on her purposes. Accordingly, if the goal is to answer (Q) via introspection, then the subject apparently cannot “know what” she thinks. For she is unable to answer without further empirical knowledge whether her thought is de H2O or something that merely looks like H2O. So for the purposes of answering (Q), the externalist does not introspectively know (much less Know) what she thinks. Yet here is the vital point: There are other anti-skeptical purposes which do not require answering (Q). As the key example, think back to our old friend Descartes.17 Recall that skeptic-immune knowledge or “certainty” is precisely the aim of the Meditations. But after the skepticism of Meditation One, Descartes begins Meditation Two with some trepidation about whether anything is certain. Soon enough, however, the cogito is identified— and it is also here where Descartes claims to Know what he thinks. It may be unclear whether such matters are certain. Again, if Descartes doubts his ability to add two and two, then it seems fair to doubt his inner awareness of these things. But to repeat, if we ignore those kinds of doubts, it is not obviously wrong to say that an (SV)-judgment is skeptic-immune in conditions (a)–(c). If so, then an (SV)-judgment might qualify Descartes as “knowing what” he thinks—at least for the purposes of settling one kind of skeptical question, namely: (Q2) Is anything known with certainty (= Known)? After all, when Descartes judges (SV) in conditions (a)–(c), he has an introspective awareness that his judgment is self-verifying. Hence, assuming doubts about introspection are bracketed, the truth of the judgment is impossible to doubt, or at least, not in a consistent way (where an inevitably

17 What follows is not intended as serious Descartes scholarship; the point is just to consider whether externalism allows the sort of self-knowledge claimed in the Meditations.

Thought Switching  121 true judgment might be false).18 So for the purposes of meeting the challenge posed at (Q2), Descartes seems within his rights to reply, “Well, I at least Know what I’m thinking.” Granted, Descartes would not Know exactly what he thinks, among all possible twin thoughts. So the skeptical question he answers would not be the characteristic question of the slow switch skeptic. Regardless, thanks to his introspective awareness of his (SV)-judgment, he plausibly Knows what he thinks for the purpose of answering a kind of Cartesian skeptic. That is so, even if slow switching is a live possibility. For regardless of any slow switches, the truth of the (SV)-judgment in conditions (a)–(c) remains (rationally) impossible to doubt. It thus seems that Descartes retains the ability to answer (Q2) affirmatively, hence, retains Knowledge of what he thinks for that purpose.19 Yet it may still sound odd to say that Descartes can Know what he thinks—even though for all introspection shows, his thought may have one of two contents (assuming externalism). But again, this just reflects that for some purposes, a subject isn’t required to distinguish water from a twin in order to “know what.” Oddly, then, knowing-what for antiskeptical purposes is sometimes like knowing-what for ordinary purposes. (Though of course there are also stark differences.) For ordinary purposes, a child indeed knows what water is despite ignorance of chemistry. Similarly, Descartes may indeed Know what he thinks, for the purposes of meeting the challenge at (Q2), even if introspection yields no answer to (Q).

5.5  Transition to Chapter 6 To recap: The present Introspective Awareness View says that an (SV)judgment in conditions (a)–(c) suffices for introspective Knowledge of thought-content. In this, the intensionally individuated (SV)-judgment is ontologically neutral, and sufficient to know what one thinks, at least relative to some anti-skeptical purposes. Dialectically, it means that premise (1') of the refined slow switch argument is undermotivated, even for Oscar’s skeptic-immune self-knowledge. For the Introspective Awareness View is a viable position which implies that: (4) If externalism is true, one can still introspectively Know what content one’s thought has for some purposes, even if one cannot Know via introspection whether it is de H2O or de XYZ.

18 This indicates that the skeptic here is also not challenging the more obvious judgments of consistency. But that too is consistent with the skeptic being an external world skeptic. 19 For convenience, I am glossing over the distinction between knowing for certain that the (SV)-judgment is true, versus knowing for certain (SV). The distinction doesn’t seem to affect the present situation.

122  Knowledge of Thought Still, it should be conceded: (5) It cannot be introspectively Known what one thinks, at least for the purposes of answering (Q). But insofar as the Introspective Awareness View is tenable against (1'): (6) The slow switch argument does not establish that given externalism, it is impossible to introspectively Know what one thinks. That is, the slow switch argument fails to show an incompatibility between externalism and introspective, skeptic-immune, knowledge of thought. It is also worth remarking that: (7) (5) is no cause for philosophical anxiety. It should be no surprise if Knowledge of thought is impossible in relation to answering (Q). Who would have believed we could introspect the actual chemical kind that a thought picks out? Certainly not Descartes. Not only was he writing before the advent of chemistry, he admitted all along that his thoughts may be of non-actual objects, for all he can introspect. Yet given the Introspective Awareness View, it remains defensible that under externalism, Descartes Knows introspectively what he thinks, for the anti-skeptical purpose of answering (Q2) affirmatively. Unfortunately, however, it does not follow that Descartes Knows enough to thwart the Problem of Wayward Reflection. In particular, if stealthy thought switches occur, it is dubious whether introspection affords enough Knowledge for Descartes to avoid simple equivocations in his reasoning, owing to a shifty ‘water’-concept or the like. Yet if introspection is not up to this task, that seems to enliven the Wayward Reflection Problem: Such equivocation could undermine the legitimacy of Descartes’ reasoning, even if he is being thoroughly epistemically responsible in the process. Thus: Even if Descartes Knows what he thinks for the purposes of answering (Q2) affirmatively, it remains unclear whether he Knows what he thinks for the purposes of self-reflective inquiry. That is the issue taken up in the next chapter.

Appendix: Quasi-Intellectualism about Knowing-Wh

The semantics and metaphysics concerning “knowing-wh” is receiving more attention than ever these days.20 Because of that, I wish to spell out in a bit more detail the purpose-relative semantics, and what it says about Descartes and his “knowing what” he thinks. Further details on such a semantics are also found in Boër and Lycan (1986); I refer the reader to their impressive discussion in their first two chapters. A “purpose” is a state of successful action, such as “fetching water” or “answering (Q).” Assume, then, that S is a subject, μ is a variable, ψ is a name, demonstrative, or definite description. And suppose that Φ is an “important predicate” for purposes P (cf. Boër & Lycan, pp. 34–35). Then given a ψ, define the (possibly empty) set Kst of predicates that S satisfies at time t, where each predicate has the normal form ┌ μ knows-true that Φ(ψ) ┐.21 The conditions on “knowing-what” are then roughly as follows: (KW) S satisfies ┌ μ knows what ψ is┐ at t for the purpose of P iff: Kst requires no further predicates for an ability22 in S to achieve P at t.

20 Knowing-wh includes knowing-what, knowing-who, knowing-when, etc. (They correspond to “knowing-wh” constructions in English, but only when the wh-clause is an embedded question, vs. a free relative or scalar clause.) 21 Boër and Lycan use ‘knows-true’ rather than ‘knows’; this is because ┌ Φ(ψ) ┐ is a representation, and it is a bit misleading to talk of S “knowing” a representation. It suggests a kind of de re knowledge of the representational vehicle as such. Even so, to say that S ‘knows-true’ a representation is not to attribute meta-representational knowledge of the truth of the representation. Instead, “It is for the token to perform the sort of job . . . which distinguishes beliefs from desires, intentions, and other attitudes” (p. 185, n. 4). 22 What counts as an “ability” is not obvious; I would refer the reader to Millikan (2004, ch. 4). Boër and Lycan do not explicitly mention abilities in their official account. But this may owe to restricting their account to theoretical purposes, viz., information-gathering and such. When speaking more broadly, they hint at the ability-entailing idea: S is said to know who N is relative to such-and-such goal iff S knows an answer with “the information about N that the questioner needs in order to carry out his project” (p. 26, emphasis mine). Still, I do not wish to commit them to such a view; this is only to illustrate that the abilityentailing idea seems to be a natural way to extend their remarks.

124  Knowledge of Thought More intuitively, S knows what iff she has enough knowledge-that about what ψ denotes, for an ability to achieve P. Substitutions for ‘P’ here can include ‘fetching a glass of water’, ‘answering correctly the question on the chemistry exam’, ‘answering (Q)’, etc. Whereas, important predicates may include ‘is H2O’, ‘is that’ [attended by a suitable ostension], ‘is possibly a liquid in lakes and rivers’, etc. Thus, if ψ is replaced by ‘water’, ‘P’ is replaced with ‘fetching a glass of water’ and Kst = {‘x knows-true that water flows out of taps’}, then (KW) implies that at t: (KW1) S knows what water is for the purpose of fetching a glass of water iff: Kst requires no predicate beyond ‘x knows-true that water flows out of taps’ for an ability in S to achieve fetching a glass of water at t. Less formally, S knows what water is for this purpose iff her knowing that water flows out of taps is all she needs to know to fetch a glass of water. Or better, it is all the knowledge-that is needed about water.23 Note that (KW1) does not require S to actually possess the ability. A differently-abled person can “know what” just as much as anyone else. The idea is rather that S “knows what” if the relevant ability does not demand any further knowledge-that about water, beyond what S already has.24 Naturally, the ability might require other things as well. But as long as she has the requisite knowledge-that, she then counts as “knowing-what,” even if other requisites for the ability are absent. In the literature on “knowing-wh,” competing views are sometimes classified as either “intellectualist” (e.g., Stanley & Williamson 2001, Stanley 2011, Brogaard 2011), or as “anti-intellectualist” (e.g., Hetherington 2011). The distinction turns on whether knowing-wh is seen as wholly reducible to knowing-that, or as wholly reducible to an ability (which in turn, does

23 Schaffer (2007; 2009) would add that the child “knows what water is” only if s/he knows that the relevant propositional knowledge is an answer to the embedded question, ‘What is water’? I myself doubt that this is a wholly general requirement; my guess is that it is required for some purposes but not for others. (Thus, I would explain away Schaffer’s examples via the purpose-relativity of knowing-wh.) 24 Farkas (2016) has argued that sometimes the mere ability to gain the relevant propositional knowledge (vs. actually possessing it) is sufficient to “know what.” Thus, there is a sense in which I know what Ram’s phone number is, since I can look it up anytime in my cell phone. As Farkas is aware, the key objection to her is that such an attribution is really shorthand for “I know how to find out Ram’s phone number.” Farkas’ reply is that the ability to find the relevant info is not always sufficient for knowing-what. Yet this confuses the issue. The question is whether Farkas’ examples are really disguised cases of knowing-how rather than knowing-what. The issue is not whether in general know-how is sufficient for a correlative knowing-what.

Thought Switching  125 not reduce to knowing-that). But the present account represents a third way (“quasi-intellectualism”). Like intellectualism, knowing-wh on the present view is a matter of having the relevant sorts of knowing-that. Yet unlike intellectualism, ability is also central to determining facts about “knowingwh:” Given a proposition p, knowing that p is relevant only if S’s knowing that p is needed for an ability in S to achieve P, where P is some contextually salient purpose. It is not assumed, moreover, that the ability reduces wholly to the relevant body of knowing-that (though it would seem to do so in part). Indeed, the ability is not even strictly necessary for knowing-wh, as shown by the differently-abled person.25 But this last feature implies a limitation on the account. Knowing-wh standardly includes knowing-how—yet I would grant that there is a type of “know-how” where the relevant ability is required. Consider Kalpesh, who has read voraciously on perfecting the 3-point shot. In spite of this, however, he remains unable to execute on the court. Then, as we watch him miss shot after shot, it would be fair to say that Kalpesh does not “know how” to make a 3-pointer, precisely because of his inability (even if we appreciate that Kalpesh can recite by heart all the necessary techniques). Yet (KW) would not require the ability for “knowing how;” thus, the account cannot capture this sense in which Kalpesh lacks know-how.26 Still, there is a different sense in which Kalpesh “knows how” to make a 3-point shot, which (KW) indeed captures. Kalpesh may not know how to make a 3-pointer for the purposes of reliably scoring points himself (since that requires the ability). But he indeed knows how for (e.g.) coaching purposes. After all, he has all the knowing-that needed for answering his players’ questions and whatnot. (He exemplifies the contentious adage “those who can’t, teach.”27)

25 In Parent (2014, n. 19), I misworded my view as knowing-wh “partly reducing” to an ability. But due to fully knowing differently-abled persons, even partial reduction fails (yet not with the ability-entailing readings of “knowing-wh,” discussed in the main body after this note). Instead, I should say that (non-ability-entailing) knowing-wh wholly reduces to knowing-that—yet the ability to gain a contextually salient goal fixes which propositional knowledge it reduces to. For the nature of the ability fixes whether the ability demands more knowing-that than what one already has. (But again, knowing-that is not the whole of what’s needed to possess the ability.) 26 Thanks to Stanley and Williamson (2001) and Stanley (2011), there is a highly active debate on Ryle’s argument for anti-intellectualism about know-how. I side with Ryle, yet I do not mean to take a stand on his regress argument. 27 The ability-entailing vs. non-ability-entailing readings are not unique to ‘know how’ attributions. Consider: ‘Kalpesh knows what to do to make a 3-point shot’, and ‘Kalpesh knows when to hold ’em, and knows when to fold ’em’. (My guess is that normally the two readings are possible when the wh-word is followed by an infinitive.) On the ability-entailing readings, my view is that “knowing-wh” wholly reduces to the ability, although I cannot

126  Knowledge of Thought What of Descartes’ “knowing what” he thinks? Where ‘x’ is replaceable only with an essential indexical (see n. 4), let S = Descartes and Kst = {‘x knows-true that water is a possible, colorless, odorless liquid in lakes and rivers’, ‘x knows-true that x is thinking forthwith that water is wet’, ‘x knows-true it is certain that x is thinking forthwith that water is wet’}. Then, (KW) implies that: (KWD) Descartes satisfies ┌ μ knows what μ thinks┐ at t for the purpose of answering (Q2) affirmatively iff: for an ability in Descartes to answer (Q2) affirmatively, Kst requires no predicates beyond ‘x knows-true that water is a possible, colorless, odorless liquid in lakes and rivers’, ‘x knows-true that x is thinking forthwith that water is wet’, and ‘x knows-true it is certain that x is thinking forthwith that water is wet’. Furthermore, the Introspective Awareness View will suggest that: (IAD) Descartes satisfies at t ‘x knows-true that x is thinking forthwith that water is wet’ and ‘x knows-true it is certain that x is thinking forthwith that water is wet’ if Descartes judges (SV) at t in conditions (a)–(c). Thus, if Descartes judges (SV) while introspecting on the judgment so as to meet (a)–(c), the Introspective Awareness View implies that (with the right background knowledge) Descartes indeed “knows what” he thinks, for the purposes of answering (Q2) affirmatively.

References Azzouni, J. (2007). Ontological commitment in the vernacular. Noûs, 41(2), 204–226. Boer, S. & Lycan, W. G. (1986). Knowing who. MIT Press. Boghossian, P. (1989). Content and self-knowledge. Philosophical Topics, 17, 5–26. Braun, D. (2006). Now you know who Hong Oak Yung is. Philosophical Issues, 16, 24–42. Brogaard, B. (2011). Knowledge-how: A unified account. In J. Bengson & M. ­Moffett (Eds.), Knowing how: Essays on knowledge, mind, and action (pp. 136– 160). Oxford UP. Brown, J. (1995). The incompatibility of anti-individualism and privileged access. Analysis, 55(3), 149–156. Brown, J. (2004). Anti-individualism and knowledge. MIT Press.

delve into this here. But taken together, this means I am an anti-intellectualist about abilityentailing knowing-wh, and an intellectualist about all other cases of knowing-wh. Here lies the rationale for the moniker ‘quasi-intellectualism’.

Thought Switching  127 Brueckner, A. (1999). Difficulties in generating skepticism about knowledge of content. Analysis, 59(1), 59–62. Burge, T. (1988). Individualism and self-knowledge. Journal of Philosophy, 85(1), 649–663. ———. (1998). Memory and self-knowledge. In P. Ludlow & N. Martin (Eds.), Externalism and self-knowledge (pp. 351–370). Stanford: CSLI Press. Chalmers, D. (1996). The conscious mind. Oxford UP. ———. (2002). The foundations of two-dimensional semantics. In M. GarcíaCapintero & J. Macìa (Eds.), Two-dimensional semantics: Foundations and applications (pp. 55–140). Oxford UP. Cohen, S. (1986). Knowledge and context. Journal of Philosophy, 83, 574–583. ———. (1998). Contextualist solutions to epistemological problems: Skepticism, Gettier, and the lottery. Australasian Journal of Philosophy, 76, 289–306. Dennett, D. C. (1975). True believers: The intentional strategy and why it works. In A. F. Heath (Ed.), Scientific explanation: Papers based on Herbert Spencer lectures given in the University of Oxford (pp. 53–75). Oxford UP. DeRose, K. (1995). Solving the skeptical problem. Philosophical Review, 104(1), 1–52. ———. (2009). The case for contextualism: Knowledge, skepticism, and context, vol. 1. Oxford UP. Falvey, K. & Owens, J. (1994). Externalism, self-knowledge, and skepticism. Philosophical Review, 103, 107–137. Farkas, K. (2008). The subject’s point of view. Oxford UP. ———. (2016). Know-wh does not reduce to know-that. American Philosophical Quarterly, 53(2), 109–122. Gertler, B. (2000). The mechanics of self-knowledge. Philosophical Topics, 28(2), 125–146. Ginzburg, J. (1995). Resolving questions: I. Linguistics and Philosophy, 18, 459–527. ———. (2011). How to resolve how to. In J. Bengson & M. Moffett (Eds.), Knowing how: Essays on knowledge, mind, and action (pp. 215–243). Oxford UP. Ginzburg, J. & Sag, I. (2000). Interrogative investigations: The form, meaning, and use of English interrogatives. Stanford: CLSI Publications. Goldberg, S. (1999). The relevance of discriminatory knowledge of content. Pacific Philosophical Quarterly, 80, 136–156. ———. (2003). What do you know when you know your own thoughts? In S. Nuccetelli (Ed.), New essays on externalism and self-knowledge (pp. 241–256). MIT Press. ———. (2006). Brown on self-knowledge and discriminability. Pacific Philosophical Quarterly, 87, 301–314. Goldman, A. I. (1976). Discrimination and perceptual knowledge. Journal of Philosophy, 64, 355–372. ———. (1986). Epistemology and cognition. Harvard UP. Green, M. (2005). Moorean absurdity and seeing what’s within. In M. Green & J. Williams (Eds.), Moore’s paradox: New essays on belief, rationality, and the first person (pp. 189–216). Oxford UP. Heal, J. (1994). Moore’s paradox: A Wittgensteinian approach. Mind, 103, 5–24. Hetherington, S. (2011). How to know: A practicalist’s conception of knowledge. Wiley-Blackwell.

128  Knowledge of Thought Hookway, C. (2008). Questions, epistemology, and inquiry. Grazer Philosophische Studien, 77, 1–22. Kallestrup, J. (2009). Knowledge-wh and the problem of convergent knowledge. Philosophy and Phenomenological Research, 78, 469–476. Kripke, S. A. (2011). Frege’s theory of sense and reference: Some exegetical notes. In his Philosophical troubles: Collected papers vol.1 (pp. 254–291). Oxford UP. Lewis, D. K. (1982). Whether report. In T. Pauli et al. (Eds.), Philosophical essays dedicated to Lennart Åqvist on his fiftieth birthday (pp. 194–206). Uppsala: Filosofi ska Studier. ———. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74(4), 549–567. Ludlow, P. (1995). Externalism, self-knowledge, and the prevalence of slow-switching. Analysis, 55, 45–49. Millikan, R. G. (1993). White queen psychology; or, the last myth of the given. In her White queen psychology and other essays for Alice (pp. 279–364). MIT Press. ———. (2001). The myth of mental indexicals. In A. Brook & R. Devidi (Eds.), Self-reference and self-awareness: Advances in consciousness research vol. 11 (pp. 167–181). John Benjamins. ———. (2004). On clear and confused ideas. Cambridge UP. ———. (2012). Are there mental indexicals and demonstratives? Philosophical Perspectives, 26, 217–234. Parent, T. (2014). Knowing-wh and embedded questions. Philosophy Compass, 9(2), 81–95. Perry, J. (1979). The problem of the essential indexical. Noûs, 13, 3–21. Schaffer, J. (2007). Knowing the answer. Philosophy and Phenomenological Research, 75(2), 383–403. ———. (2009). Knowing the answer redux: Replies to Brogaard and Kallestrup. Philosophy and Phenomenological Research, 78(2), 477–500. Searle, J. (1983). Intentionality: An essay in the philosophy of mind. Cambridge UP. Stanley, J. (2011). Know how. Oxford UP. Stanley, J. & Williamson, T. (2001). Knowing how. Journal of Philosophy, 98(8), 411–444. Sterelny, K. (1988). Review of Knowing who by Stephen E. Boer and William G. Lycan. Philosophy of Science, 55, 654–656. Warfield, T. (1992). Privileged self-knowledge and externalism are compatible. Analysis, 52(4), 232–237. Wikforss, Å. (2008). Self-knowledge and knowledge of content. Canadian Journal of Philosophy, 38, 399–424.

6 Content Externalism Does Not Imply Wayward Reflection

The infallibilist account of self-knowledge, developed thus far, is meant to be compatible with content externalism. However, Boghossian (1992a, b, 1994, 2010, 2014) has argued that such Burge-inspired accounts face (what I call) the Problem of Wayward Reflection via Thought Switching. Briefly, Boghossian highlights that in a “slow switch” scenario, it is still possible for a thinker to unwittingly equivocate when reasoning in the armchair. Such non-introspectable equivocation, moreover, is seen as cutting into the subject’s rationality. The worry, then, is that externalism remains incompatible with the kind of self-knowledge needed for rational reflection. In reply, I argue two points. First, I suggest that if non-introspectable equivocation is a problem, it is a problem that is not unique to externalism. Second, I show that non-introspectable equivocation is ultimately not a problem—that rational reflection is not threatened by it, even in a skeptical context. Very briefly, that’s because the logical flaw in Boghossian’s examples is due not so much to a shift in content as a shift in reference. And one should not expect changes in reference to be introspectively detectable.

6.1  Boghossian on Wayward Reflection In the last chapter, we first raised the possibility of slow switching as a way to doubt the externalist’s introspective, skeptic-immune self-knowledge. And like the last chapter, the present chapter will assume that slow switches are epistemically relevant alternatives. Even if this is fanciful, there is interest in seeing whether externalism still allows introspective self-knowledge. Again, there is an intuition that some introspective self-knowledge should not be affected by skeptical possibilities. Hence, if one’s account of selfknowledge fails to respect that, it would seem to be a defect of the account. The “skeptic-immune” nature of some such knowledge is an interesting and unique feature—one’s account thus should not ignore it, even if it hardly ever matters in our epistemic endeavors.

130  Knowledge of Thought In the last chapter, we saw that under a specified set of conditions, a selfverifying judgment like (SV) may well constitute such “skeptic-immune” knowledge: (SV) I am thinking that water is wet. Whether or not this is correct, Boghossian’s objection is that such judgments do not enable one to avoid simple logical errors in reasoning, errors that we would expect a rational person to avoid. If that is correct, then self-verifying judgments at best represent a “hollow” kind of self-knowledge (Boghossian 1992a, p. 15). In more detail, the worry is that such judgments do not seem to reveal whether one concept-token has the same content as an earlier token. And if one cannot introspect “same content,” then one cannot introspect whether a piece of reasoning equivocates between (e.g.) water versus twin-water. Boghossian illustrates this by considering Peter, a fan of Pavarotti, who is slow switched unawares to Twin Earth. After the switch, Peter continues to pursue his interests—yet now he is unwittingly reading interviews of, and buying new albums by, Twin Pavarotti. The problem then arises when he draws upon memories of Pavarotti in his reasoning. For instance, suppose he correctly remembers seeing Pavarotti (on Earth) swim in Lake Taupo. After the slow switch, then, suppose this memory in an inference that he would formulate in his language as follows: (1) Pavarotti once swam in Lake Taupo. (2) The singer I heard yesterday was Pavarotti. (3) Hence, the singer I heard yesterday once swam in Lake Taupo. From Peter’s armchair, the reasoning looks perfectly valid. Yet in fact, ‘Pavarotti’ in (1) refers to Pavarotti on Earth, but ‘Pavarotti’ in (2) refers to Twin Pavarotti. After all, (1) expresses a memory of Pavarotti—whereas (2) expresses a judgment having a causal origin with Twin Pavarotti. If so, then his reasoning expressed at (1)–(3) blatantly equivocates on ‘Pavarotti’, even though Peter is unable to discern this from the armchair—no matter what self-verifying judgments are available at the time.1

1 The issue is reminiscent of Kripke’s (1979) puzzle about belief. Recall that Kripke considers a bilingual, Pierre, who is unaware that ‘Londres’ co-refers with ‘London’ (though is otherwise competent with the terms). Due to happenstance, Pierre ends up asserting both ‘Londres est jolie’ and ‘London is not pretty’; the question is then whether the speaker is guilty of inconsistency. Regardless, Peter’s case is notably different. Whereas Pierre is ignorant of coreference, Peter is unaware that ‘Pavarotti’ in (1) vs. (2) has different referents. Consequently, only Peter’s case shows that, assuming externalism, one may equivocate without being able to know it from the armchair.

Externalism and Wayward Reflection  131 Accordingly, the incompatibilist holds that the Pavarotti case falsifies the following, “transparent difference” thesis: (TD) If two mental tokens have different contents at time t, then the subject can introspectively know that the tokens have different contents at t. However, (TD) is said to be important to the subject’s status as a rational agent. So if (TD) is false under externalism, it seems externalism has a serious problem on her hands. Even so, take heed that nothing here touches the similar “transparent sameness” thesis: (TS) If two mental tokens have the same content at time t, then the subject can introspectively know that they have the same content at t. After all, the Pavarotti-case is ex hypothesi one where different contents are involved. So the antecedent of (TS) is false here, meaning the case is trivially not a counterexample to (TS). But if (TD) is falsified by the case, then even with self-verifying judgments, externalism does not afford enough self-knowledge to avoid equivocation. It is controversial, however, whether a “thicker” kind of introspective selfknowledge should be allowed. Especially since we are talking of skepticimmune self-knowledge, an externalist might instead dismiss “thicker” self-knowledge as Cartesian wishful thinking.2 Regardless, I shall argue that Boghossian’s particular objection does not mandate a decision here, for Peter’s reasoning need not be seen as equivocal. One tactic along these lines is to insist that Peter refers to the same person in (1) versus (2) (Schiffer 1992; Burge 1998, Goldberg 2007a). The most natural interpretation might have ‘Pavarotti’ denoting Twin Pavarotti in both instances, in which case, Peter’s argument is valid. Of course, this would mean that (1) expresses a false memory. But at least there is no unnoticed equivocation from the armchair. It is contentious, however, to see Peter’s memory as “shifty” between environments—where (1) comes to express a false memory about Twin Pavarotti, despite the memory being caused in the normal way by Earthian Pavarotti. Such a view is nonetheless defended by Ludlow (1995), Butler (1997), and Tye (1998). Yet Heal’s outstanding and underappreciated (1998) paper reveals a multitude of complications for the view, re: which thoughts have which contents at which times.3 The matter is especially convoluted, since different interpretations arise at this point for the thought

2 See, e.g., Millikan (1984, 1993), Goldberg (1999, 2007b), Williamson (2000, ch. 4). 3 For further criticisms of “shifty memory,” see also Brueckner (1997) and Burge (1998.)

132  Knowledge of Thought experiment. Thus far, we have assumed that Peter possesses both a concept of Twin Pavarotti and a concept of Earth Pavarotti—but there are several ways to see things otherwise. (See Gerken 2013 and Bernecker 2010, ch. 6, for an overview of such options; Brueckner 2000 also is helpful on this.) Yet here, let us simply grant Boghossian that Peter has two separate concepts, and that he might equivocate in the suggested manner. But my eventual aim is to show this is unproblematic. Though it may behoove us to show first that, if there is a problem, it is not a problem that is unique to externalism.

*6.2  The Owens-Brown Argument Owens (1989) has also suggested that the problem is not unique to externalism, and his line has been further developed by Brown (2004). According to these authors, (TD) faces counterexamples regardless of whether externalism is true. But while I agree with this, the particular counterexamples offered do not strike me as especially compelling. Brown’s preferred counterexample (borrowed from Owens) is a case drawn from philosophy itself. This concerns the debate in philosophy of language on whether the following two sentences express the same thought (where ‘doctor’ and ‘physician’ are co-extensional): (a) No one doubts that whoever believes that Mary is a physician believes that Mary is a physician. (b) No one doubts that whoever believes that Mary is a physician believes that Mary is a doctor. Thus, Mates (1952) and Burge (1976) hold that (a) and (b) express different contents, while Church (1954) and Kripke (1979) maintain that they express the same content. Naturally, only one side of this dispute can be right. Yet that means one party is misidentifying the content of (a): Someone is either misidentifying it as the same as (b)’s content, or misidentifying it as different from (b)’s content. That seems undeniable, independently of externalism— even though (presumably) both sides have thoroughly introspected the contents expressed by these sentences. So it seems that transparency of content fails, regardless of externalism. But thus far, this leaves open whether (TD) or (TS) has failed. Consider: If Church and Kripke are right, then the other side has misidentified the contents as different. In which case, the dispute provides a counterexample to (TS) only. But if Mates and Burge are right, then the others have misidentified the contents as the same—it is in this instance alone that we have a counterexample to (TD). Unfortunately, this means that the Owens-Brown case can be only as convincing as the Mates-Burge view, and the latter is of course contentious. In fact, as Brown recognizes, some of Burge’s arguments on the matter seem externalist in kind. So it is doubtful whether we have a

Externalism and Wayward Reflection  133 clear counterexample to (TD), which is also clearly independent of externalism. (For the record, I myself happen to side with Mates and Burge, yet I grant that this is not the most dialectically effective starting-point.) More significantly, even if Mates and Burge are right so that the example falsifies (TD), Boghossian’s objection still seems compelling if he instead appeals to the weaker (TD'): (TD') If two mental tokens have different contents at time t, such that the difference in content is sufficient for a difference in reference, then the subject can introspectively know that they have different contents at t. Consider that the Peter case falsifies (TD'), yet the Owens-Brown counterexample does not. That’s because unlike Peter’s ‘Pavarotti’-tokens, the concepts expressed by ‘doctor’ and ‘physician’ are co-referring even if they have different contents. Because of that, moreover, (TD') seems defensible in a way that (TD) is not. Re: ‘doctor’ vs. ‘physician’, Boghossian could concede that the difference in content is too “subtle” to be introspected. (The difference is not even sufficient for a difference in reference.) Yet he could still insist that when contents are different enough to create a referential difference, as is true of Peter’s ‘Pavarotti’-tokens, then the difference in content should be introspectively noticeable. The problem, then, is that externalism permits cases like Peter’s, where two contents cannot be thus differentiated, despite the different contents creating a difference in reference. So the Owens-Brown counterexample is less than effective, not just because it requires the truth of the Mates-Burge view, but also because it is not a counterexample to the weaker (TD'). Yet assuming externalism, (TD') is counterexampled by Peter’s ‘Pavarotti’-tokens—and that may seem worrisome enough. With (TD'), the intuition is that a difference in content should be introspectively detectable, at least if the difference is dramatic enough for co-reference to fail. Yet Peter cannot introspect the difference, and this is why he falls prey to non-introspectable equivocation. And since nothing analogous happens with ‘doctor’ vs. ‘physician’, non-introspectable equivocation thus far remains unique to externalism.

*6.3  Non-Introspectable Equivocation Sans Externalism It turns out, however, that (TD') has other counterexamples that are more plausibly independent of externalism. The counterexample developed here features a magic trick which creates the illusion of teleportation. In the trick, a person (let’s call her “Alejandra”) enters a wooden box at stage left. The magician closes the box and then, presto!, Alejandra suddenly exits from a different wooden box, at stage right. The first box is then opened to show it

134  Knowledge of Thought is empty. Teleportation! (The reader may recall this from the 2006 film The Prestige, though the film concerns a disturbing variant of the trick.) The illusion owes to a body double. At stage right, the box contains a look-alike, who steps out when cued by the magic word. Meanwhile, Alejandra at stage left escapes through a trapdoor at the bottom, thus leaving an empty box for the magician to reveal. Now to pre-empt some objections, suppose that by sheer co-incidence, Alejandra’s look-alike has the name ‘Alejandra’ as well. Further, assume that an audience member, Sanjay, met both women before the performance, each of whom was introduced as “Alejandra.” However, suppose Sanjay was introduced to them separately, and failed to realize he was meeting two different people. He thus thinks he was introduced to the same person twice by mistake. Now when the magic trick is performed, assume Sanjay is unaware of how the trick works. The illusion thus leads him through steps of reasoning that he would express as follows: (4) Alejandra entered the box on stage left. (5) Alejandra exited the box on stage right. (6) So, a unique person both entered one box and exited the other. Premise (4) concerns a different person than premise (5)—yet Sanjay is unable to detect the equivocation just via introspection. He abductively infers (post-hoc) that something is amiss, but the point is that introspection on his mental contents cannot reveal the equivocation. Thus, we have a counterexample to (TD') which does not assume externalism. To be honest, I might be persuaded that some type of externalism follows from the case (though others may find that dubious). But externalism is at least not overtly assumed. In Sanjay’s case, moreover, we can easily imagine he is prompted to conclude (6), if only for a split-second, after observation leads him to believe (4) and (5). Sanjay’s case may be somewhat contrived, but there are more commonplace examples as well. Whenever one is dealing with two people with the same name, there is potential for such equivocation. Once in grad school, I overheard a fellow student say that “Ted” was giving a symposium at the Eastern APA. It turned out she was talking about Ted Sider, but for a moment, I had to ask myself “Am I giving a symposium at the APA?” Naturally, I soon recognized that my friend was speaking of a different “Ted.” But that was something I had to infer using memory and background knowledge. If I had just introspected on the content of my friend’s use of ‘Ted’, this hardly would have revealed that it named someone other than me. As a reply, however, perhaps only an externalist would agree that ‘Alejandra’ in (4) vs. in (5) have different contents. At least, one could easily imagine some internalists denying that. If so, then the Sanjay counterexample is not obviously free of externalist assumptions—and so externalists might not have escaped blame for non-introspectable equivocation.

Externalism and Wayward Reflection  135 However, most internalists (I think) will agree that Sanjay’s observational beliefs at (4) and (5) reference different people. Sanjay of course does not immediately recognize that. Still, the two observational beliefs are in fact referring to two different persons in the world. This is at least consistent with internalism as such—since conceivably, to say that the tokens have different referents is not to say they have different contents. (Our internalist might deny that content wholly determines reference, i.e., that intension wholly determines extension.) Nevertheless, if the internalist grants that there are different referents, that suffices to concede the equivocation regardless of what one says about content. For it is a shift in reference that makes for the equivocation fallacy. After all, even if the concepts expressed by ‘doctor’ and ‘physician’ have different contents, their co-reference ensures that the following remains valid: (7) Mary is a doctor. (8) Any physician has an M.D. (9) So, Mary has an M.D. The reasoning is valid, assuming the concepts are co-referring, and in spite of any difference in content. So again, if the internalist allows the difference in reference in Peter’s ‘Pavarotti’-tokens (which seems consistent with her thesis about content), then that suffices for Peter’s non-introspectable equivocation. We may summarize matters as follows (where an ‘Alejandra’-token is a conceptual token expressed by a use of ‘Alejandra’): (i) Sanjay’s ‘Alejandra’-tokens have the same internally individuated content, yet are not co-referring. [Assumed by the present internalist] (ii) If his ‘Alejandra’-tokens are not co-referring, then the argument in (4)– (6) is invalid by equivocating. [Assume] (iii) So, his ‘Alejandra’-tokens have the same internally individuated content, and the argument in (4)–(6) is invalid by equivocating. [From (i), (ii)] Thus, since Sanjay cannot detect the shift in reference introspectively, (TD') is falsified, even under the internalist view. Non-introspectable equivocation can occur regardless of externalism.4

4 Pérez-Otero (2014) makes a similar argument against (what he calls) “Same-Indexical-Meaning” internalists. However, as he is aware, there are well-known problems with “indexical” accounts of names/kind terms (see especially Burge 1982). So in the context of his paper, his argument might seem to reflect only the problems with the indexical account, and not with internalism as such. (And arguably, the indexical account is indeed orthogonal to internalism. Putnam 1975 famously adopted it as part of his externalism.) In contrast, my version of

136  Knowledge of Thought Stepping back, I do not think there is anything unintuitive or surprising about this result. Plausibly, even Descartes would admit that the evil demon could switch the objects he refers to, right under his nose. (Indeed, Descartes comes surprisingly close to entertaining a “Pavarotti possibility,” when remarking that the people he observes might be automatons.) Moreover, since equivocation is concerned with reference rather than content, admitting this possibility is sufficient to allow non-introspectable equivocation. That remains the case, even if one reads Descartes as an internalist about content.

*6.4  Internalist Replies Yet there is a series of objections to consider. I said that internalists can grant that the ‘Alejandra’-tokens have different referents, since internalism is not a thesis about reference. But some internalists may balk. They may see it as part of “internalism” that the two ‘Alejandra’-tokens do not differ in reference. Alternatively, one might say these internalists deny “Referential Difference,” even knowing that other so-called internalists affirm it. Yet if the truly consistent variety of internalism does not accept Referential Difference, then the internalist may well avoid the Alejandra-counterexample to (TD'). Such counterexamples would thus remain a properly externalist affliction. As far as I can tell, there are four ways that an internalist might deny Referential Difference.5 We can list these as follows: (One) Sanjay’s ‘Alejandra’-tokens co-refer; they both refer to (let’s assume) the first woman named ‘Alejandra’ that Sanjay was introduced to. (Two) Sanjay’s ‘Alejandra’-tokens co-refer, since they are each defined as denoting “anyone who is Alejandra1 or Alejandra2.” (Three) Sanjay’s ‘Alejandra’-tokens co-refer, yet neither woman on stage is the referent of either token. (The tokens end up co-denoting some third thing.) (Four) Sanjay’s ‘Alejandra’-tokens are indeterminate in reference between the two women. Hence, they do not (determinately) differ in reference.6

the argument does not saddle the internalist with the indexical account of meaning. Yet it is shown that the problem of non-introspectable equivocation arises regardless. 5 A fifth way would be to say that the ‘Alejandra’-tokens are referentially empty (so, vacuously, they do not differ in reference). But I assume this would be a last resort. It at least seems preferable to say that Sanjay’s observational beliefs are referentially indeterminate, rather than completely referentially empty. 6 Pérez-Otero (2014) also considers internalist rejoinders within this sort of dialectic, some of which resemble those above. But he does not consider anything like (Four), and this (we shall see) may well be the internalist’s best bet.

Externalism and Wayward Reflection  137 However, we can safely dismiss (One). Suppose for concreteness that Sanjay was first introduced to Alejandra at stage left. Then, (One) would imply that the belief at (5) refers to Alejandra at stage left. This is so, even though Sanjay believes (5) on the basis of his perceptual acquaintance with Alejandra on the right! The second option, moreover, is not really an option. Given the proposed definition, each token would refer to both “Alejandras,” meaning that, for example, (4) would be true only if two people entered the box at stage left. The third option I shall bracket. Unlike (One) and (Two), I regard (Three) as a viable choice (though I also think it ends up being somewhat externalist). Yet to avoid a thicket of complexities, we shall bypass all this. We may rest content with the fact that most everyone would reject (Three) as formulated. After all, it is quite radical to suggest that both of Sanjay’s ‘Alejandra’-tokens refer to the same thing, even though each token denotes neither woman on stage.7 The most compelling option for the internalist thus seems to be (Four)— where each ‘Alejandra’-token is indeterminate in which person it refers to (meaning that the truth-values of (4) and (5) are also indeterminate). This still sits uncomfortably with the fact that when Sanjay is prompted to believe (4), it is via his perceptual acquaintance with only one Alejandra. (And ditto when he is prompted to believe (5).) But because of his misimpression when meeting the two Alejandras, some amount of indeterminacy may well be tenable. So (Four) thwarts Referential Difference, claiming it is indeterminate whether the two ‘Alejandra’-tokens differ in reference, since both tokens would indeterminately refer in the first place. The issue for this tactic, however, is that it is unclear why internalism is needed for any of this. It seems that an externalist could agree that, in Sanjay’s case, the reference of each ‘Alejandra’-token is indeterminate—and that there is no determinate referential switch in (4)–(6). In which case, non-introspectable equivocation need not be a problem for either the internalist or externalist. The upshot would be that the internalist and externalist versions of (Four) stand or fall together—thus, there would be no reason to think that externalism faces a special problem.

7 The internalist I mostly have in mind would be some kind of neo-Kantian, who thinks that the ‘Alejandra’-tokens refer only to “Alejandra as she exists in the understanding,” where this is distinct from Alejandra as she is “in itself.” This is a radical view, since (I would argue) even Kant believed that we can refer to things in themselves, as when he speaks of “things in themselves.” Kant instead denied that we can know the intrinsic properties of things-inthemselves. (Here I am following Langton’s 1998, 2004 interpretation.) But I still admit to taking seriously the radical Kantian, especially if (Three) concerns only the “speaker’s referent” and not the “semantic referent” of the tokens (Cf. Kripke 1977). (I am currently drafting a piece where I engage this Kantian more earnestly.)

138  Knowledge of Thought Obviously, if our externalist antecedently endorsed Referential Difference, then she cannot go along with (Four). But some externalists can conceivably reject Referential Difference on the putatively “internalist” grounds at (Four). To be clear, this would not be the most familiar sort of externalist. After all, a typical externalist says that an observational belief like (4) is referring determinately to the object perceived on that occasion. (Perceptual acquaintance is often the externalist’s causal-cum-semantic relation par excellence.) But even if that is normally how an externalist treats observational belief, it need not be that way, especially when Sanjay is confused about how many “Alejandras” he has met. And like internalism, externalism is a view about content in the first instance rather than reference. So it is not inconsistent with externalism as such to concur that in Sanjay’s case, the ‘Alejandra’-tokens are indeterminate in reference. This may be surprising. Externalism is often glossed as the view that conceptual content is determined partly by the environmental referent. Thus, an externalist may seem committed to a determinate referent if the content is determinate. Yet for one, an externalist could simply allow that the ‘Alejandra’-tokens are indeterminate in content. An indeterminate content is still a content (as with vague concepts); so it is not as if such an externalist thereby renders the tokens contentless (McLaughlin & Tye 1998a, b make a similar point in a different context). Besides, as long as the content remains partly determined by the speaker’s environment, then externalism as such is respected. And such partial determination can be argued by an externalist in Sanjay’s case. For, plausibly, there being two “Alejandras” in Sanjay’s environment determines that his ‘Alejandra’-tokens have an atypical content. But this already means that facts about the environment partly determine the content of the tokens. That is especially clear if the externalist adds that the ‘Alejandra’-tokens would have determinately referred if there had been only one Alejandra— even if Sanjay’s experiences remained subjectively indistinguishable. True, in the usual Twin Earth thought experiments, the referent of ‘water’ is assumed to be determinate. So if an externalist now wanted to say that the ‘Alejandra’-tokens were indeterminate in either content or reference, this could not be argued via a standard sort of Twin Earth experiment. Still, the present point is just that it is not inconsistent with externalism to say that content and/or reference can be indeterminate. Whether the externalist has a good argument for this sometime-indeterminacy is another matter. Yet consistency is all that’s needed to show that (Four) does not require anything internalist to avoid non-introspectable equivocation. So if desired, an externalist could happily agree to this so-called “internalist” solution.8

  8 There is a parallel here with the debate on whether externalism allows contentful empty concepts. Empty concepts seem precluded if externalism is glossed as “content is partly fixed by an environmental referent” (implying a need for a referent). Also, the gloss can

Externalism and Wayward Reflection  139 However, in contrast to Sanjay’s case, referential indeterminacy may be less plausible in Peter’s case. After all, Peter has long-standing knowledge of Earthian Pavarotti; so it is more plausible that his memory at (1) is determinately about that Pavarotti. Yet one is inclined to say this, be my guest. For the point cuts against anyone who opts for referential indeterminacy— whether s/he be externalist or not. Accordingly, my claim is not that the indeterminacy solution is correct. It is rather that the solution is orthogonal to the internalism/externalism dispute. Again, insofar as there is a problem with non-introspectable equivocation, it seems to arise for both sides.

6.5  Tacit Belief and Identity Considering all that, how should the problem be solved? I allowed earlier that the two ‘Pavarotti’ tokens determinately refer to different people. So I do not wish to defend (Four) as a general solution. Indeed, Peter ex hypothesi has undergone a slow switch—so it is more or less stipulative that one token denotes Twin Pavarotti determinately. (Also, per the previous paragraph, it is likely the other token determinately refers to Pavarotti on Earth.) I thus propose to leave behind (Four). Instead, I shall plump for a solution that is almost universally rejected in the literature. If we take Peter as our example, the rough idea is that his reasoning includes a suppressed premise along the lines of: (†) Pavarotti, as referenced in (1) = Pavarotti, as referenced in (2).9 It is generally agreed that if Peter’s reasoning has (†) or the like as a premise, then it is valid.10 The reasoning would be unsound, given that (†) is false. But (†) is a heterophonic identity-statement about at least one object under

suggest that a contentful concept must have a determinate referent. But as argued in chapter 4, Twin Earth experiments do not imply that empty concepts lack content; they show only that an empty concept must eo ipso have a different content. Analogously, if only one ‘Alejandra’-token is referentially indeterminate, then externalism says just that it differs in content. It does not preclude its having content.   9 More exactly, I understand (†) as follows. When ‘Refers(x, y, S)’ means that x occurs in a sentential representation S, and x refers to y: ιx (x = Pavarotti & Refers(‘Pavarotti’, x, (1))) = ιy (y = Pavarotti & Refers(‘Pavarotti’, y, (2))). Yet I am not wedded to this particular identity-assumption being the operative one in Peter’s reasoning. It also seems enough if he assumes, e.g., “that guy I once saw in Lake Taupo = that singer I heard yesterday.” But I focus on (†), since the existing literature usually assumes that (†) is closer to what (if anything) Peter is tacitly assuming. 10 N.B., Brown (2004, p. 179) discusses the suppressed premise as if it states the identity of the Pavarotti-concepts, rather than of the referents. This seemingly owes to her presumption that content is centrally involved in non-introspectable equivocation. In contrast, I argued earlier that reference is the central thing. Accordingly I find the matter clearer if we focus on the identity of the referents.

140  Knowledge of Thought two different “guises.” And one would not expect Peter to know its truthvalue just by introspection. If this is right, then although Peter has made a mistake, it seems unproblematic as a mistake—it is not as if the possibility of such a mistake is absurd. So it looks like Boghossian’s worry is neutralized.11 However, barring some mixed strategy, we must say not just that Peter has a tacit identitypremise, but also that an analogous premise exists in any putative case of non-introspectable equivocation. Yet Brown (2004, pp. 179–180) finds this psychologically implausible—and I would agree that the possibilities for abnormal psychology falsify the general rule. But abnormal minds are not what concern Brown; she instead thinks that in anyone, there are inferences where the subject “merely relies on the putative identity of the concepts . . . rather than making an identity judgment” (p. 179). (Collins 2008 holds a similar view.) However, my claim is not that one always judges an identity-claim during such reasoning. Rather, it is that the identity-statement is tacitly assumed. Yet what does that mean exactly? Well, there are lots of different views in the area, but we can put forth a few modest hypotheses without causing offense. For instance, we may reasonably hypothesize that: (h1) If Peter were queried about (†), he would affirm its truth in standard communicative conditions. Assume here that “standard” conditions are ones where Peter can hear the question, when the question is put in terms he understands, when he has no conflicting interests, when he is not immobilized and thus unable to answer, and so forth. Further, we can suppose that: (h2) The disposition mentioned in (h1) is a law-like indicator that Peter presently believes, albeit perhaps just tacitly, what he would affirm.12

11 An objection: That the subject cannot introspect whether (†) is true is just the sort of failure of “transparency” which Boghossian presses against the externalist. Reply: It is obvious that such heterophonic identity claims are not knowable simply by introspection. So if Boghossian’s point indeed just reduces to the non-introspective status of (†), then I do not see why anyone should be worried. 12 Strictly speaking, the disposition ascribed to Peter is not sufficient for the tacit belief. At least, Peter must not be one of Lycan’s (1988, ch. 3) “opinionated people,” who is inclined to randomly affirm or deny any queried proposition (simply because they cannot bear the Peircean “irritation of doubt”). Also, there may be borderline cases. For instance, contra Plato, it may be indeterminate whether the slave-boy in the Meno all along tacitly believed the theorems that Socrates prompts him to “recollect.”

Externalism and Wayward Reflection  141 Third, we may add that: (h3) Peter is disposed to answer ‘yes’ (in standard conditions) to the query “Must (†) be true before (3) can be validly inferred via (1) and (2)?” Finally, suppose that: (h4) The disposition mentioned in (h3) is a nomological indicator that Peter’s reasoning has (†) as a (suppressed/) premise. Generalizing and summarizing, Peter’s reasoning has p as a (possibly suppressed) premise only if he would answer ‘yes’ in standard conditions to each of: (Q1) Is p true? (Q2) Is the truth of p required for the validity of the reasoning? The wording of the second question is a bit unfortunate, for it assumes Peter understands what the technical term ‘validity’ means. But this is just to make clear what he needs to be asked. The question need not be phrased exactly in those terms, as long as it communicates the same thing. (Again, it is part of “standard” communicative conditions that the questions are put in terms he grasps.)13 It is key that Peter is standardly disposed to answer ‘yes’ to a question like (Q2). For if Peter tacitly believes p, it does not follow that Peter’s reasoning utilizes p as a premise. Peter, after all, tacitly believes that there are no unicorns—but that does not figure into his reasoning at (1)–(3), not even tacitly. Yet because of that, he would presumably answer ‘no’ (in standard conditions) if asked ‘Must “there are no unicorns” be true in order for (1)– (3) to be valid?’14 Two asides. First, it does not matter if Peter’s answer to (Q2) is incorrect. Suppose Peter mistakenly answers ‘yes’ (in standard conditions) when asked ‘Must “there are no unicorns” be true before (1)–(3) is valid?’ Despite Peter’s mistake here, this is still a nomological indicator that ‘there are unicorns’

13 All this can suggest that one suppressed premise in Peter’s reasoning is: “If (1) and (2) are true, then so is (3).” If so, then Peter’s reasoning is inevitably valid (hence, not equivocating). This and related objections will indeed force us to revise (h4) a bit; see section 6.2.2. For now, I shall ignore such complications. 14 Relatedly, if p is a tautology or other obvious logical truth, Peter’s answer to (Q2) would standardly be ‘yes’, simply because he knows p must be true tout court. But plausibly, not every such p is a suppressed premise of Peter’s reasoning (nor is every such p even a “background assumption” in the sense discussed later). Thus, obvious logical truths must be bracketed off as exceptions. This does not seem like too much of a limitation on the account, since reasoning often does not depend crucially such truisms.

142  Knowledge of Thought is indeed a premise of his reasoning. That is so, even though the premise is not strictly necessary for validity. Yet there is nothing absurd in this; people often adopt premises that end up being superfluous in reasoning. Second, it helps to note what happens if Peter incorrectly answers ‘no’ (in standard conditions) to ‘Must (†) be true to infer (3) validly from (1) and (2)?’. His reasoning in this case would not have (†) as a premise, and he would thus be equivocating. Even so, Peter’s eschewing (†) would suffice for a logical error that he is able to detect from the armchair: He can recognize non-empirically that if (†) is false, then (1)–(3) is definitely equivocating. So if he equivocates as a result of rejecting (†), that should not constitute a problem. Accordingly, the present proposal should not be accused of making equivocation impossible (although this concern will be addressed again later). But in brief, even if Peter does not reason to (3) by explicitly judging or citing (†), his reasoning would normally have (†) as a (suppressed/) premise. In which case, again, his reasoning is valid. More, since the Peter-example is arbitrary, such hypotheses would extend to other putative cases of nonintrospectable equivocation. And it is not implausible in such cases; speakers are disposed to answer similar questions in similar ways (in standard conditions). So it is not unlikely that these putative equivocators are implicitly assuming some identity-premise, in the above sense.15

6.6. Further Dialectics Through the years, Boghossian’s argument has generated an extensive literature, and there are a number of other issues that must be addressed before the preceding account can be seen as acceptable. *6.6.1  Vicious Regress? Brown (2004) offers a further argument against the suppressed-premise gambit. The basic thought is that there cannot always be a suppressed

15 Relatedly, Brown’s case of “merely relying on the putative identity of the concepts” also is normally the case where an identity-statement is a suppressed premise. (Or at least, the statement is tacitly assumed in a way that absolves the subject of irrationality; cf. section 6.2.2 below.) That’s because, presumably, if the subject is so “relying,” she has a disposition to answer ‘yes’ (in standard conditions) to similar questions as above. After all, if she lacks such dispositions, it is no longer clear that the subject is relying on the putative identity of the concepts. For it is unclear what would distinguish this from a case where the subject is just being reckless in her reasoning—where she just infers (3) incautiously or randomly, without any tacit or explicit judgment about the identity of concepts or of the “Pavarottis.” (This last possibility illustrates, again, that equivocation remains possible on the present approach.)

Externalism and Wayward Reflection  143 identity-premise in an inference, on pain of vicious regress. Brown illustrates this in relation to the following, less contentious bit of reasoning (pp. 180–181): (10) Water is clear. (11) Water is good to drink. (12) So something is clear and good to drink. Suppose that as it stands, (10)–(12) does not equivocate. Even so, since it is possible for ‘water’ to switch referents, the present approach would apparently acknowledge a suppressed premise here, along the lines of: (10.1) Water, as referenced in (10) = water, as referenced in (11).16 Yet if so, matters would not end there. For again, it is possible for ‘water’ to shift reference between (10) and (10.1) in a way that makes for equivocation. So apparently, it must be further supposed that: (10.11) Water, as referenced in (10) = water, as referenced in (10.1). (10.111) Water, as referenced in (10.1) = water, as referenced in (10.11). (10.1111) Water, as referenced in (10.11) = water, as referenced in (10.111). . . . and so on. . . . (The regress here is presented in English, but naturally the objection is that such a regress would occur in thought.) The upshot is that in Peter’s case, the simple little argument at (10)–(12) ends up having infinitely many premises. But that is absurd. So we cannot suppose that all such inferences implicitly assume some identity-premise. Toward a reply, a distinction is needed. This is between a subject’s reasoning versus the argument, where the latter is a set of sentences (seen as an unchanging, abstract object). The former would be studied in psychology, whereas the latter is studied in logic. (The distinction should be a familiar one; it is what Frege insisted on when opposing “psychologism” in mathematics/logic.) Granted, there is a temptation to reject abstract objects, since they can seem ontologically gratuitous. But the danger is that we lose the difference between correct and incorrect reasoning. After all, the sense is that 80,000,000 Germans can be wrong about which inferences in German

16 Brown in fact formulates the identity-claim here as ‘water = water’ (p. 181). But my suggestion for (1)–(3) was not that it tacitly assumes ‘Pavarotti = Pavarotti’. Rather, it was that it implicitly assumes the non-trivial (†)—which describes the “Pavarottis” in relation to the different tokens. Accordingly, Brown’s regress argument has been re-construed to oppose (10.1) as an implicit assumption, rather than just ‘water = water’.

144  Knowledge of Thought are correct. But if there is no standard beyond the Germans’ own judgments, it is hard to see how that is possible. There is a whole lot to debate here, and we are certainly not at leisure to investigate it all. Suffice it to say that we need some normative/factual distinction in the area, and for convenience I shall continue calling it the distinction between reasonings (qua psychological occurrences) and arguments (qua sets of sentences). The thing to stress, then, is that the current approach does not claim that all arguments include identity-sentences. Like everyone else, I allow that any finite set of sentences constitutes an argument (where some sentence is a conclusion to which others stand as premises). So in particular, the set containing (10), (11), and (12) is an argument. That is so, whether or not it equivocates. Understood thusly, the argument patently contains no identitypremise, much less infinitely many of them. In contrast, however, Peter’s reasoning rests on some identity-assumption. N.B., I am now distinguishing a “premise” from an “assumption” as well: The former is part of an argument; the latter is part of someone’s reasoning. In more precise terms, then, the main thesis is that Peter’s reasoning tacitly assumes (†): In standard communicative conditions, he will answer ‘yes’ when queried about (†) along the lines of (Q1) and (Q2). Concurrently, Peter normally would not perform the inference to (3) unless (†) was assumed somehow. Our focus has been on the argument at (1)–(3)—yet this need not be a red herring. For the argument (or rather, a token thereof) functioned to communicate Peter’s reasoning. As far as his explicit reasoning is concerned, moreover, it may communicate it perfectly. But (1)–(3) may not articulate his reasoning in toto, given the reality of tacit assumptions. It is typical, moreover, for philosophers to recognize tacit assumptions in evaluating an argument. For example, one typically assumes interpretations of the terms/concepts being used, e.g., ‘Pavarotti’ in (1) is assumed to refer to Earthian Pavarotti, even though this was not made explicit within (1)–(3). Even so, must we say that Peter assumes infinitely many identity-statements, as the regress objection would suggest? I think we must. But this is not the burden it may seem to be. Remember that we are talking about tacit assumptions only; thus, there is no suggestion that Peter is consciously entertaining infinitely many propositions during the reasoning. It is pertinent, moreover, that everyone already tacitly assumes infinitely many propositions. For instance, up until this moment, I suspect you tacitly assumed that there is at most zero unicorns, that there is at most 1 unicorn, that there is at most 2 unicorns, and so on. (Similarly, you believe that p ⊃ p is a theorem, that ~~(p ⊃ p) is a theorem, etc.) Here too, the criterion for such beliefs is that one would answer ‘yes’ (in standard conditions) to key questions about unicorns or theoremhood.

Externalism and Wayward Reflection  145 In light of the foregoing, matters are improved if we now replace (h4) with the following revision: (h4') The disposition mentioned in (h3) is a nomological indicator that Peter’s reasoning has (†) as a (tacit/) assumption. According to (h4'), Peter’s reasoning tacitly assumes (†) if, in standard conditions, he has the disposition to answer affirmatively ‘Must (†) be true before (3) can be validly inferred from (1) and (2)?’. Even so, I mean (h4) to be neutral on whether (†) is a suppressed premise, in whichever argument might be used to communicate Peter’s reasoning. Neutrality is possible since, no matter which argument is used to represent his reasoning, the bearing of (h4') is not logical but psychological: The claim is that the reasoning itself, qua psychological occurrence, will have (†) as a (tacit/) assumption. That is so, regardless of whether (†) is best designated as a “suppressed premise” of his expressed argument. These points are directed at Peter’s case, but they also carry over to Brown’s reasoning expressed at (10)–(12). Naturally, the two cases differ in that only the former makes use of a term/concept that is actually equivocal. But each piece of reasoning normally would have infinitely many tacit assumptions. (Of course, one member of the infinite sequence is actually false in one case but not in the other.17) It is worth mentioning too that you, dear reader, already have the regress of tacit beliefs about (10) and (11). We all antecedently believed (10) and (11) tacitly. And we all already tacitly believed that those two tacit beliefs are de water. Further, each of us tacitly believed that that tacit belief is de water as well. And so on! Of course, no one had a regress of beliefs whose ascriptions would feature numerical terms like ‘(10)’, ‘(10.1)’, etc. So the claim more precisely is that each of us had a regress of extensionally equivalent tacit beliefs. But there is reason to posit such a regress, since each of us could have been asked questions that prompted expressions of such beliefs (in standard conditions). Nevertheless, let me acknowledge that tacit belief is a strange bird.18 I myself have little idea about what the phenomenon amounts to. But if

17 This explains why the regress may seem stranger in Brown’s case versus the Peter case. Interpretive charity may pressure us to attribute (†) to Peter as an assumption, and that starts the regress. But what (10)–(12) expresses does not equivocate as it stands, and so there is no such interpretive push. Still, reasoning is valid only if it is impossible to have true assumptions with a false conclusion. And to secure that, one must (tacitly/) assume a particular interpretation of one’s assumptions, whereby no equivocation can occur. So even for the reasoning expressed at (10)–(12), there remains an impetus for recognizing a regress of identity-assumptions. 18 Lycan (1988, ch. 3) is especially good at illustrating how odd tacit belief is. See also Bartlett (ms.).

146  Knowledge of Thought there are beliefs at all, then not all of our beliefs are occurrent at every moment. So if there are beliefs at all, then apparently there are tacit beliefs. And if there are tacit beliefs, then there are apparently infinitely many tacit beliefs. And in this vein, the present contention is that Brown’s regress is not vicious but rather exactly the right way to see cognitive reality. Before moving on, it might be said that we need only one tacit belief in Peter’s case, namely, a belief that “all conceptual tokens that would be expressed by ‘water’ are co-referring.” And the appearance of infinitely many beliefs may just reflect that this generalized tacit belief has a potential infinitely many of instances. Going by our question-and-answer tests, Peter’s reasoning indeed normally has the generalized belief as a (tacit/) assumption. And I would agree that this can already explain why Peter is not irrational, given that it has consequences like (†) which are not introspectively verifiable.19 But the same question-and-answer tests also suggest that normally Peter (tacitly/) assumes the particular instances of the generalization as well. That is a datum that one should not shove under the rug, but rather accept as part our cognitive lives. *6.6.2  Collins’ Objection and Related Matters As a different issue, Collins (2008) suggests that the question-and-answer tests, akin to those indicated by (Q1) and (Q2), would show that Peter does not implicitly assume the relevant identity-statement. Collins writes: Press Peter on the status of the suppressed premise, and, if he really has training in logic, he would insist that the premise was superfluous for the validity of the inference, and that it was not a constituent of his inference. This premise, after all, although its content is that water is twater, is expressed by the sentence ‘water is water’. . . . We ought to be reluctant to attribute to someone reliance on a premise when she explicitly denies that she is using it at all. (p. 569) It is not true, however, that the first member of the infinite series is expressed simply as ‘water is water’ (at least, not in the series generated by this author’s view; cf. n. 16). For this whole affair really concerns a referential shift between the ‘Pavarotti’-tokens. (10.1) thus expresses more precisely the tacit assumption that Peter should be attributed.

19 Brown rejects attributing Peter the generalized belief, on the grounds that one can make a “ground floor inference about water without invoking a premise that involves the concept of a concept” (2004, pp. 181–182). She may be right about that; even so, I can still appeal to the fact that Peter has infinitely many tacit beliefs, as per the regress. (And again, we apparently do have infinitely many tacit beliefs, strange as that may seem.)

Externalism and Wayward Reflection  147 Regardless, Collins’ point is that if we ask Peter about (†) or (10.1), he apparently would deny that such a premise is needed for validity. So again, the same question-and-answer tactic suggests that Peter’s reasoning does not rest on an identity-statement. But in fact, it is equivocal what Peter would mean in stating that “(†) is superfluous.” If he means that (1)–(3) is valid even if (†) is false, then he is obviously mistaken. More likely, Peter would mean that he need not include (†) among the premises in order for (1)–(3) to be a valid argument. But this would be rational for him to say only if he assumes (†). For again, the validity of (1)–(3) is not indifferent to the truth of (†).20 So again, even if (†) is not designated as an official “premise” in the argument used to reconstruct his reasoning, Peter is still assuming (†) during his reasoning. Again, in this sort of case, (†) can be seen as a background assumption, akin to the way that the interpretations of ‘Lake Taupo’, ‘singer’, ‘yesterday’, etc., are typically part of the background. Moreover, if (†) is assumed just as part of the background, this equally excuses Peter for inferring the false (3) from the truths at (1) and (2), at least while in the armchair.21 For even if (†) is not designated as a “premise,” it remains that Peter endorses (1)–(3) only because he assumes a false heterophonic identity-statement about a worldly object. And one cannot be expected to detect introspectively that such a thing is false. Granted, if (†) is merely a background assumption, then it still could be apt to say that Peter’s reasoning equivocates. For in such a case, the shift in reference between the ‘water’-tokens would remain, and there would be no official “premise” of the argument to gainsay this. But as long as (†) is tacitly assumed somehow, we can allow this without creating problems. For if (†) is at least assumed in the background, then Peter remains rationally blameless even if the reasoning “equivocates” in this respect. It is precisely because the shift in reference eludes introspection that Peter cannot be held rationally responsible for such equivocating. Or rather (to avoid begging any questions), Peter should not be seen as irrational even here. Blamelessly, Peter’s reasoning tacitly assumes (†), and he cannot detect that (†) is false in the slow switch scenario while confined to the armchair.

20 Later on, Collins (2008, p. 571) states that to assume (†) just is to commit a logical error. For this is a mistaken view about the logical form of the reasoning, viz., that (1) and (2) employ the same proper name. Yet to concede that Peter has misidentified the form of his reasoning is not to grant that he has equivocated (and it is the latter that is ultimately at issue). In fact, since Collins seems to grant at this stage that (†) is tacitly assumed, this already suggests that Peter’s reasoning is valid. That is so, even if Peter also has a false metatheoretic view about its logical form. 21 My thanks to Greg Novack for this observation.

148  Knowledge of Thought Just to be clear, many cases of reference-shift remain introspectively detectable, and accordingly, can make for irrationality. A classic example of rationally blameworthy equivocation is the following: (13) Evolutionary theory is a theory. (14) If x is a theory, then x is not scientifically established. (15) So, evolutionary theory is not scientifically established. In (13), ‘theory’ presumably refers to “any set of statements,” yet then denotes “a set of statements that is not scientifically established” in (14). Under the respective readings of ‘theory’, both premises are trivial, yet the conclusion does not follow. More, if someone were to draw the conclusion on this basis, they are rationally culpable. For normally the person could (and should) be aware that ‘theory’ is used equivocally, even while confined to the armchair.22 These reflections on Collins help us quell a further worry, which has been waiting in the wings. Earlier, the test-questions may have suggested that Peter’s reasoning standardly has as a suppressed premise “If (1) and (2) are true, then so is (3).” (Obviously, the truth of this is necessary for validity.) But then, Peter inevitably reasons validly. Yet it should be possible for Peter to equivocate (even if conditions are “standard” as defined here). Indeed, if his reasoning just consists in that extra premise plus (1)–(3), he intuitively still is equivocating. The extra premise merely articulates the mistake that the equivocal premises suffice for the conclusion. But at this stage, affirmative answers to questions like (Q1) and (Q2) no longer mean that the addition stands as an extra premise. It can plausibly stand instead as a background assumption in the following sense: Even if Peter expliclty denies the need for such a “premise,” he normally would not execute the inference unless he assumed the proposition, implicitly or explicitly. Indeed, if Peter reasons as per (1)–(3), the natural explanation would be that he (tacitly/) believes that (1) and (2) are a sufficient condition for (3). Nevertheless, this is the kind of thought that seems to overgeneralize— something analogous appears true for every case of reasoning. Prototypically, a reasoner (tacitly/) assumes that her premises suffice for her conclusion. But again, sometimes the equivocation-threat can and should be introspectively detected. Yet this is not true in Peter’s case. Peter may,

22 If this last point is not obvious, consider: Given a modicum of linguistic competence, an actual defender of (13)–(15) would likely answer ‘yes’ (in standard conditions) to “Does ‘theory’ in (14) refer to something not scientifically established?” Whereupon, we could point out that (13) begs the question if ‘theory’ remains constant in reference. However, assuming the person did not mean (13) in a question-begging way, s/he would normally have the ability to observe, “In (13), ‘theory’ was not used to refer only to unestablished theories.” That shows, however, she could detect the equivocation, just via armchaircontemplation. That can justify the charge of irrationality in this case, whereas such a justification is not available in Peter’s case.

Externalism and Wayward Reflection  149 naturally enough, assume that what (1) and (2) actually express are enough for the truth of (3). But in a slow switch case, this fails to hold up—and we should not expect the reason for this to be introspectively detectable. 6.6.3  Explicit Equivocation One final issue is that Peter’s reasoning may be disconcerting even if (†) is tacitly assumed. Since Peter’s explicit reasoning still equivocates, that can seem damning enough. After all, if he is explicitly aware of only (1)–(3), then it is as if he is unwittingly equivocating. Yet if the preceding sections are correct, then Peter is ultimately not equivocating. So why should the mere appearance of equivocation matter? Perhaps the thought is that some kind of “safety” condition is violated (cf. Sosa 1999, Pritchard 2005). Even if he is not actually equivocating, there is a nearby possible world where Peter’s phenomenology is the same, and where he really is equivocating. The upshot, then, is that his reasoning could have very easily been invalid, even though it is actually valid. Yet if that is so, then it may be unclear how legitimate the reasoning is. But the nearby possible world in question is, I suspect, not really that near. For if Peter is equivocating in that world, then assuming the previous section is on target, this means he does not believe (†) in that world. And if he is not assuming (†), why would he be inclined to (sincerely) draw the conclusion from his explicit premises? Again, it may be that Peter in this scenario is just inferring the conclusion on a whim, with a devil-may-care attitude. Yet that would make the relevant world significantly different from how the Peter-case actually unfolds. Peter is assumed to be reasoning in the ordinary way, where the intent is to reason well, not just to reason at whim. So it seems that the relevant “safety” condition, whatever it is exactly, should not be a concern. (Or rather, if it is a concern, more argument is needed to show that.) More broadly, there is a sense in which Peter’s reasoning is guided by his tacit belief in (†). For, given his intent to reason well, the following counterfactual holds: (CF) If Peter had not (implicitly/) assumed (†), he would not have reasoned as per (1)–(3). Actually, given the possibilities for psychological deviance, (CF) might also hold only ceteris paribus. But the exceptions would be cases of deviance, and there is no indication that Peter’s mind is aberrant in this respect.

6.7  Onward to Part III It is important for a further reason that Peter is (normally) guided by his assuming (†), in the sense indicated above. For the larger issue, recall, is that non-introspectable equivocation might thwart the rationality of

150  Knowledge of Thought armchair reasoning. Fortunately, however, it is plausible that Peter tacitly assumes (†) whenever he reasons along the lines of (1)–(3) (assuming he is not psychologically deviant, apart from having a slow-switched concept). So it is plausible that Peter is innocent of a rationally condemning kind of equivocation. The tacit assumption, moreover, is not something we attribute ad hoc or merely out of interpretive charity. (If we did, then it would remain open whether his reasoning really is problematic, barring some radical Davidsonian interpretivism; cf. Stalnaker 1990; 2008.) In the end, positing the implicit assumption is warranted, given that Peter’s reasoning seems guided by the assumption, in that he seems to satisfy (CF). In Peter’s case, the assumption of (†) is of course false. But again, that sort of mistake is unremarkable. Thus, the present view rules that non-introspectable equivocation ultimately should not be a concern. Peter avoids any rationally offensive equivocation, because he implicitly assumes (†), whose falsity is not detectable just by introspection.23 Nevertheless, there are other threats to the rationality of armchair reasoning besides non-introspectable equivocation. It is important that when Peter is engaged in reflection, he recognizes that the truth of (1) and (2) is something he presently believes, rather than denies, desires, hopes for, etc. After all, his believing (1) and (2) is partly why he is right to treat them as premises. In contrast, if he instead denied them, then their negations would be appropriate as premises, and (1) and (2) themselves would not. Thus, if he recruits (1) and (2) as premises, he must be able to know from the armchair that he presently believes in them. However, knowing what one believes is notoriously problematic in light of Freudian phenomena, and more commonplace examples of self-ignorance, as reviewed in chapter 2. The rationality of armchair reflection requires further defense in light of such things. Now we saw in chapter 2 that the relevant phenomena do not discredit the infallibility of self-attributions of judgments. But it certainly does not follow that we are infallible (nor even that we are reliable) about what we judge. In the next chapter, however, a story is given about how infallibility toward one’s judgments is possible. In the chapter after that, we then see how certain linguistic practices are evidence for such infallibility. Further, that chapter will illustrate how it is possible for one to recognize (albeit ­fallibly) when one executes an infallible self-attribution. Taken together, the

23 The point applies equally in my own case. In section 1.3.1, I considered the objection that the arguments of this very book might engage in Wayward Reflection, due to equivocating in ways that are non-introspectable. But here too, any putative equivocations of this sort are more plausibly seen as cases where an implicit (albeit perhaps false) identity-claim is assumed.

Externalism and Wayward Reflection  151 next two chapters thus provide a defense of infallibilism about some selfattributions of judgment, along with an explanation about how Peter can know (fallibly) when his self-attributions are infallible. Eventually, in chapters 10 and 11, the latter will help answer how Peter is able to recognize which propositions play the role of assumptions in his reasoning.

References Bartlett, G. (ms.). Occurrent states, unpublished draft. Bernecker, S. (2010). Memory: A philosophical study. Oxford UP. Boghossian, P. (1992a). Externalism and inference. Philosophical Issues, 2, 11–28. ———. (1992b). Reply to Schiffer. Philosophical Issues, 2, 39–42. ———. (1994). The transparency of mental content. Philosophical Perspectives, 8, 33–50. ———. (2010). The transparency of mental content revisited. Philosophical Studies, 155(3), 457–465. ———. (2014). Reply to Otero’s “Boghossian’s inference argument against content externalism reversed”. Philosophy and Phenomenological Research, 89(1), 182–184. Brown, J. (2004). Anti-individualism and knowledge. MIT Press. Brueckner, A. (1997). Externalism and memory. Pacific Philosophical Quarterly, 78(1), 1–12. ———. (2000). Ambiguity and knowledge of content. Analysis, 60(3), 257–260. Burge, T. (1976). Belief and synonymy. Journal of Philosophy, 75, 119–138. ———. (1982). Other bodies. In A. Woodfield (Ed.), Thought and object (pp. 97– 120). Oxford UP. ———. (1998). Memory and self-knowledge. In P. Ludlow & N. Martin (Eds.), Externalism and self-knowledge (pp. 351–370). Stanford: CSLI Press. Butler, K. (1997). Externalism, internalism, and knowledge of content. Philosophy and Phenomenological Research, 57(4), 773–800. Church, A. (1954). Intensional isomorphism and identity of belief. Philosophical Studies, 5, 65–73. Collins, J. M. (2008). Content externalism and brute logical error. Canadian Journal of Philosophy, 38(4), 549–574. Gerken, M. (2013). Epistemic reasoning and the mental. Baskingstoke, UK: Palgrave. Goldberg, S. (1999). The relevance of discriminatory knowledge of content. Pacific Philosophical Quarterly, 80, 136–156. ———. (2007a). Anti-individualism, content preservation, and discursive justification. Noûs, 41(2), 178–203. ———. (2007b). Semantic externalism and epistemic illusions. In S. Goldberg (Ed.), Internalism and externalism in semantics and epistemology (pp. 235–252). Oxford UP. Heal, J. (1998). Externalism and memory. Proceedings of the Aristotelian Society, 72, 95–109. Kripke, S. A. (1977). Speaker’s reference and semantic reference. Midwest Studies in Philosophy, 11, 255–296. ———. (1979). A puzzle about belief. In A. Margalit (Ed.), Meaning and use (pp. 239–283). Dordrecht: D. Reidel.

152  Knowledge of Thought Langton, R. (1998). Kantian humility. Oxford UP. ———. (2004). Elusive knowledge of things in themselves. Australasian Journal of Philosophy, 82(1), 129–136. Ludlow, P. (1995). Social externalism, self-knowledge, and memory. Analysis, 55(3), 157–159. Lycan, W. G. (1988). Judgement and justification. Cambridge UP. Mates, B. (1952). Synonymity. In L. Linsky (Ed.), Semantics and the philosophy of language (pp. 111–138). Urbana: University of Illinois Press. McLaughlin, B. & Tye, M. (1998a). Externalism, twin earth, and self-knowledge. In C. Wright, B. Smith & C. Macdonald (Eds.), Knowing our own minds (pp. 285– 320). Oxford UP. ———. (1998b). Is content-externalism compatible with privileged access? Philosophical Review, 107(3), 349–380. Millikan, R. G. (1984). Language, thought, and other biological categories: New foundations for realism. MIT Press. ———. (1993). White queen psychology; or, the last myth of the given. In her White queen psychology and other essays for Alice (pp. 279–364). MIT Press. Owens, J. (1989). Contradictory belief and cognitive access. Midwest Studies in Philosophy, 14, 289–316. Pérez-Otero, M. (2014). Boghossian’s inference argument against content externalism reversed. Philosophy and Phenomenological Research, 89(1), 159–181. Pritchard, D. (2005). Epistemic luck. Oxford UP. Putnam, H. (1975). The meaning of “meaning”. In his Mind, language, and reality: Philosophical papers vol. 2 (pp. 215–271). Cambridge UP. Schiffer, S. (1992). Boghossian on externalism and inference. Philosophical Issues, 2, 29–38. Sosa, E. (1999). How to defeat opposition to Moore. Philosophical Perspectives, 13, 141–154. Stalnaker, R. (1990). Narrow content. In C. Anthony Anderson & J. Owens (Eds.), Propositional attitudes: The role of content in logic, language, and mind (pp. 131–146). Stanford: CSLI. ———. (2008). Our knowledge of the internal world. Oxford UP. Tye, M. (1998). Externalism and memory. Proceedings of the Aristotelian Society, 72, 77–94. Williamson, T. (2000). Knowledge and its limits. Oxford UP.

Part III

Knowledge of Judging

7 Infallibility in Knowing What One Judges

The previous chapters stand as a defense of infallibilism about what one is occurrently thinking, so to address issues about Wayward Reflection. Yet throughout the book, “thinking” that p just means something like “entering some mental state or other with a p-content”—it does not require any specific attitude toward that content. Consequently, knowing that you think that p is not sufficient to know whether you believe that p, rather than deny that p or suspend judgment on whether p. Reflective inquiry, however, seems to require knowing which attitude you take toward a proposition. This was mentioned at the end of the previous chapter; however, to illustrate further: If q is incompatible with p, then in the reflective process q may be thereby disconfirmed, but only if I presently believe p. If I instead suspend judgment on p, then q is neither confirmed nor disconfirmed by p. Thus, deciding the evidential status of some propositions depends on my knowing (in some way) what my attitude is toward them. (See chapter 11, section 1, for more details on this kind of line.) Thus, it is not enough for rational reflection just to know what you think. Normally it is also required to know which attitude attends the thought. This chapter is a first step toward showing that, in a range of cases, we indeed know our attitudes. Its ambition is to justify that under some circumstances, we are in fact infallible about our attitude toward an occurrent thought. But: It takes more to explain how one can recognize an attitude-attribution as infallible. It also takes further argument (thanks to the complexities of knowing-wh) to show that you thereby know which attitude you have toward a thought. These two matters shall be taken up separately in chapters 8 and 10, after the core infallibility thesis is established in this one.

7.1 Chapter 3 Redux It will help to review first the infallibilism about thought from chapter 3. The view is attended by no less than six caveats, and these will carry over

156  Knowledge of Judging to infallibilism about attitudes.1 Most basically, infallibilism about thought was the claim that: (INF1) Necessarily, if a subject judges that she thinks p, then she thinks p. The idea is that a subject’s second-order judgment necessitates that she indeed tokens the relevant first-order thought. To illustrate, consider the judgment expressed as: (SV) I am thinking, hereby, that water is wet. Again, the judgment expressed by (SV) appears “self-verifying” in that its occurrence seems sufficient for its own truth. For the subject must ‘run through’ the thought in the process of executing2 the judgment. (N.B., “the judgment expressed by S” concerns what is expressed only when the sentence S is used assertively, competently, non-metaphorically, etc.) As the first caveat, I describe infallibilism as a view about “self-knowledge,” but this talk of knowledge should not be taken too seriously, at least not yet. (I wish to put aside debates about Gettier and whether “S knows that p.”) In Chapter 10, I shall eventually defend that some infallible judgments of judgments can count as knowledge, but for now, “knowledge” can be understood as just implying “true belief.”3 Second, note the difference between (INF1) and its converse. The latter is clearly falsified by Freudian repression and such. In contrast, (INF1) concerns only those (rather rare) cases where one takes the time and effort to execute a second-order judgment. A third clarification: (INF1) is restricted to occurrent judgings and thinkings; this side-steps further counterexamples involving Freudian repression and such. (Repressed mental states are usually standing states—not datable events like making a decision or levying a verdict.) Accordingly, my talk of ‘judging’ and ‘thinking’ should be read as limited to occurrent states. Nevertheless, the possibility of occurrent

1 Three additional, less important caveats shall be reviewed here only. First, if the subject judges that she thinks that p in an infallible way, I hold that she represents herself by means of an “essential indexical” (or the mental correlate thereof); cf. Perry (1979). Second, Burge’s selfverifying judgments were occasioned by an apparent conflict between his anti-individualism (Burge 1979, etc.) and introspective access to one’s own thoughts. But for the record, selfverifying judgments can be accepted by individualists and anti-individualists alike, or so I argued in chapter 3. I will not require this point above, however. Third, (INF1) is not affected by the data of Schwitzgebel’s excellent (2011) book. Not only is he mainly targeting phenomenal states, I would argue that the data at most suggests a conflict with the converse of (INF1), rather than (INF1) itself. 2 I occasionally speak of “executing the judgment that p,” but this is just a stylistic variant of “judging that p.” 3 For discussion of self-verifying judgments qua knowledge, see also Burge (1996; 1998).

Infallibility about What One Judges  157 repressed thoughts should be admitted. So as a fourth clarification, (INF1) should not imply that we are infallible about the entire contents of our mind at a given time. Rather, the infallibility of a second-order judgment is limited to a single first-order occurrent thought. Fifth, even if a judgment is infallible, it is not necessarily indubitable.4 Infallibility is a semantic property—an immunity to falsity—but this differs from the subject’s psycho-epistemic abilities for doubt. Accordingly, a judgment expressed by (SV) could be de facto infallible, even if the subject (perhaps irrationally) harbors doubts about it. The sixth qualification on (INF1) requires a bit more elaboration; it concerns the modality in the thesis expressed by ‘Necessarily’. The infallibilist thesis is not a bare material conditional; some kind of modality is intended, but which? Well, it may depend on how one explains the infallibility. In chapter 3, I argued the infallibility owes to the compositionality of second-order thoughts; the first-order thought is a literal (and ineliminable) constituent of the second-order thought. The occurrence of the whole thus necessitates the occurrence of the part, much like writing down ‘I am thinking that grass is green’ necessitates writing down the complement-clause ‘grass is green’. And in the case of thought, the second-order judgment is invariably true, thanks to its having the first-order thought as a necessary constituent. Assuming compositionality is some kind of psychological law, this portrays second-order thoughts as infallible as a matter of psychological necessity. Such would then be the modality in (INF1). Notably, Burge (1988) invokes a different modality; he writes, “By its reflexive, self-referential character, the content of the second-order judgment is logically locked (self-referentially) onto the first-order content” (pp. 659–660). Here the infallibility is explained by a self-referential device, expressed as ‘hereby’ in (SV), or as made explicit in “I am thinking with this very thought that writing requires concentration” (cf. p. 649). This talk of self-reference and of logical locking suggests that, on Burge’s view, the modality in (INF1) is not just a psychological modality. Rather, the necessity apparently stems from the second-order thought semantically entailing the first-order thought—akin to the truth of ‘Feron is a bachelor’ necessitating the truth of ‘Feron is a man’. Such then is Burge’s modality in (INF1). Yet if “I am thinking hereby that I think that p” semantically entails “I think that p,” it is not because of an analytic connection (as in the Feron example). Instead, the semantic entailment owes to some kind of self-referential device.

4 Whether infallibility implies incorrigibility depends on what is meant. Roughly, a judgment is incorrigible iff it is impossible to correct the judgment. And if ‘correcting’ means ‘revising a falsity to a truth’, then infallible judgments cannot be corrected, obviously. Yet if ‘correcting’ implies only that a justified revision is possible, then mere infallibility does not preclude this possibility. (See section 4 of ch. 3 for elaboration.)

158  Knowledge of Judging Burge’s explanation of the infallibility is not worry-free. Chapter 3 complained that self-reference at best explains why the second-order thought is logically locked to itself. It does not yet explain why the second-order thought is locked to the first-order thought. Indeed, once we are clear on that, self-reference starts to look like a red herring. Nonetheless, Burge’s account—and the self-reference element in particular—is onto something important. In this chapter, we will consider an alternative self-verifying mechanism along these lines, whence the modality in (INF1) rightly becomes viewed as a matter of semantic necessity, owing to the content of the secondorder judgment. (In that respect, this chapter also functions as a reply on Burge’s behalf to the objection from chapter 3.) In fact, the alternate mechanism will permit something stronger than what Burge himself allowed: It can support an infallibilism, roughly where ‘think’ in (INF1) and (SV) is interpreted to mean judge (understood as an occurrent state). Put more precisely, but still roughly, the new infallibility thesis is as follows (where ‘I’ is an essential indexical; cf. fn. 1): (INF2) ┌ I am judging that I am judging that p┐ semantically entails ┌I am judging that p┐. This is an approximation, in the sense that it is eventually restricted to a proper subset of second-order judgments. But to illustrate the basic idea, consider the following: (1) I am judging, hereby, that water is wet. Suppose a subject executes the second-order judgment expressed by (1). (INF2) would suggest that this necessitates, in the manner of semantic entailment, that she is also first-order judging that water is wet. As before, that may remain dubitable and the judgment might not suffice for knowledge, even if it is de facto infallible. Yet the claim concerns only the infallibility as such. Since I take (INF2) to be of greater interest, it shall henceforth be my focus. Yet the account of judgment that supports (INF2) can also vindicate (INF) when ‘think’ does not mean “judge.” So although the remainder addresses a thesis stronger than Burge’s, the discussion remains highly relevant to the defense of his view. There is at least one objection to (INF2) that is worth clearing away at the outset. Like (INF), (INF2) avoids many Freudian objections by focusing only on judgments of a single, occurrent, first-order state. However, there is a further issue about the unconscious. Consider the judgment expressed by ‘I judge that my mother loves me’. Suppose I judge this while hosting a repressed standing judgment (or even an occurrent one) that my mother never loved me. (The judgment is too painful for me to bring into

Infallibility about What One Judges  159 awareness.) Then contra (INF2), it seems I do not really judge my mother loves me—even though I earnestly believe that I do. There may be a sense in which that is right.5 However, a second-order judgment is strictly a commitment only to the occurrence of one specific first-order judgment. Thus, even with a repressed judgment that ~p, the second-order judgment is true as long as the judgment that p is also occurrent. This of course would be a case of inconsistent judgments. But such inconsistency is all too possible, especially when unconscious forces are at work. No quick refutation of (INF2) is to be found here. (For more distinctions of this sort, see chapter 11, section 1.) Before moving on, we must reckon with an unfortunate polysemy in terms like ‘thought’. In one sense, a “thought” is a propositional content. But in another sense, it is a syntactic vehicle for such a content—a sentence of Mentalese, perhaps, or a state filling some functional role. As yet a third reading, ‘thought’ can also denote the fusion of a vehicle and a content. For my part, I shall use ‘thought’ to denote the fusions exclusively. When a term for the semantic component or feature is needed, I shall speak of the “content,” and similarly when the syntactic element is the exclusive concern, I will speak of the “vehicle.” (Note that, since a judgment is a kind of thought, a “judgment” is thus a vehicle-content fusion as well. Further, since the same type of three-way equivocation occurs with ‘concept’, I speak selectively here of the vehicle versus its content as well, and use ‘concept’ for the fusion.) One further preliminary. I often speak of “propositions” in the same breath as thought-contents. But there is a distinction here too. In general, I assume a standard view where a proposition is an actual or nonactual state-of-affairs, e.g., the state-of-affairs where water is wet.6 Accordingly, a proposition is what is represented by a thought. (In a more customary idiom, I may also say a proposition is “expressed” by the vehicle, though I intend this to be much the same thing.) In contrast, a thought represents a proposition in virtue of its “content.” So on the present usage, a thought normally does not represent a “content.” Rather, the thought bears the representing-relation to a proposition (= an actual or nonactual state-of-affairs), in virtue of the thought having a “mental content.” If this usage is contentious, I’m willing to coin a novel terminology. But for readability’s sake, I will continue with the more familiar vocabulary.

5 In particular, if “really judging” p requires the absence of any contrary judgment, then I do not really judge that p in this case. Yet the inconsistency itself indicates there is still a clear sense in which I judge p. 6 I realize there are ontological scruples about nonactual states-of-affairs; unfortunately I must leave these aside.

160  Knowledge of Judging

7.2  Paratactic Judging? The aim again is to argue, roughly, that an occurrent judgment like that expressed by (1) semantically implies that there is an act of first-order judging which makes the former judgment true. Yet this will seem odd. How can the semantics of a second-order judgment necessitate an act of firstorder judging? How can mere semantics guarantee that sort of attitudinal event? A natural and useful answer to examine (though I ultimately reject it) is that the guarantee exists if the two judgments are related in a specific paratactic manner.7 Now when parataxis is mentioned, one typically thinks of Davidson (1968) on saying ‘that’. However, the aim here is not to explain opacity, much less explain it in the Davidsonian way. Also, unlike in Davidson, the parataxis would not be a natural language sentence with a demonstrative for a sentence-name. Rather, it would be a second-order thought that demonstrates an attendant act of first-order judging, i.e., adopting the “judging-attitude” toward the first-order proposition.8 Matters will be clearer if we adopt a certain expository device, which one may call the “Mentalish” hypothesis. This is meant to be the most innocuous version possible of Fodor’s (1975; 2009) Mentalese hypothesis—thus, the Mentalish hypothesis implies neither nativism, nor universalism about primitive concepts, nor computationalism about cognition, etc. (though it is compatible with such things). And again, we restrict ourselves to occurrent mental states. Thus the Mentalish hypothesis says only that: (M-ish) There are concepts (whatever those are) which compose in lawlike ways in realizing occurrent thoughts (whatever those are). Though here too, it is assumed that thoughts and concepts are vehiclecontent fusions. But except for this and compositionality, (M-ish) is neutral on all other issues concerning the mind.

7 Jacobsen (2009) also invokes parataxis to explain second-order judgments. Yet there are two differences with the present proposal. First, Jacobsen’s focus is on Davidsonian firstperson authority, which is not a kind of infallibility, but rather a ‘presumption of truth’. Second, Jacobsen’s paratactic representations are linguistic rather than mentalistic. But as a view of judgment, I would question whether it oversimplifies the relation between mind and language. (I also worry that the predicate ‘sincere’ on p. 262 is applied too freely. It seems applied to the second-order assertion at first, but then it is applied to the embedded thatclause, in a way that resembles a division fallacy.) 8 The idea that judgment is an act fits with Burge’s (2003; 2007/2011) claim that there is something “performative” about second-order judgments, and that this is key to their distinctive security. But ultimately, my own preference may be to talk of “events” rather than acts of judgment. Metzinger (2014: line 1): “Thinking is not something you do. Most of the time it is something that happens to you.”

Infallibility about What One Judges  161 Accordingly, I treat the Mentalish hypothesis as fairly uncontroversial. Even someone like Dennett (1975) might accept it. For there is no implication that thoughts/concepts are realized as a kind of “brain writing.” After all, talk of “syntactic vehicle” is just a picturesque way to describe a physical state with a certain kind of function, namely, of representing or “having a mental content” in the sense described above. (N.B., this is not to insist that every mental state is individuated by its function.) Granted, (M-ish) implies that thoughts are compositions of concepts. But saying this still allows that compositionality may be nothing but a “real pattern” (in the sense of Dennett 1991). Besides, in calling Mentalish an “expository device,” I am foreshadowing that it is ultimately unnecessary for defending the new infallibilism (and will be dropped in the penultimate section). Suppose, then, we let small caps name Mentalish expressions.9 Let us also suppose that a Mentalish self-attribution (of any propositional attitude) demonstratively refers to the proposition represented by the first-order thought. Thus, if we also let the arrow represent direct (i.e., non-descriptive) reference, a Mentalish self-attribution might be portrayed along the following lines (where ‘i’ names an essential indexical): (2) I AM JUDGING (/ HOPING/FEARING) THAT [ WATER IS WET]

Here, square brackets are used to denote the proposition expressed by the Mentalish sentence, whose name occurs inside the brackets. Thus, (2) depicts that as a demonstrative referring to the proposition expressed by the Mentalish sentence water is wet. Again, this is akin to Davidson on propositional attitude reports in natural language. But to reiterate, the key difference is that the demonstrative is referring to the proposition expressed by a complement-clause, rather than to the clause itself. Moreover, rather than trying to explain the opacity of thoughts, here we shall take for granted that propositions and contents are individuated intensionally. Perhaps a Davidsonian explanation of opacity could be added. But to avoid unnecessary complications, let intensionality be a given. Thus the present account sees the judgment expressed at (1) in a schematic way as per (2). However, the account posits an extra constituent in the judgment, viz., hereby, which is a second deictic element. Yet unlike that, hereby does not demonstrate a proposition but rather an act of judging that a certain proposition obtains.

9 N.B., small caps name mentalistic vehicles in the first instance, yet these co-refer with names for content-vehicle fusions, and I sometimes treat them as such.

162  Knowledge of Judging In an attempt to make this clearer, let ‘J’ indicate when judging occurs during the tokening of the parataxis. (And let the attending square brackets indicate the proposition being judged). Then, what is expressed by (1) can be pictured thusly:

(3) J[I AM JUDGING HEREBY THAT] J[WATER IS WET]

In this, hereby is indexed to the subsequent act of judging.10 Accordingly, (3) portrays a second-order judgment as attributing an act of first-order judging by demonstrating it as it “crosses the scene.” So on this view, a competent, non-metaphorical, etc., assertion of (1) expresses two judgments. Thus, the second-order judgment starts to look invariably true—for it selfattributes the first-order judgment by pointing it out as it passes by. Notably, the second-order thought would not have the first-order thought as a part. Syntactically speaking, the second-order vehicle terminates with the deictic term in the middle of (3), meaning that the subsequent first-order vehicle is not one of its syntactic parts. Nor do the two thoughts semantically compose. It is true that the second-order vehicle must denote the act of first-order judging, in order to express the right second-order proposition. But the second-order content is not composed with the first-order content. After all, water’s being wet is not part of the truth-condition of the secondorder thought. (This is as it should be; a second-order thought can be true, whether or not the first-order thought is.) Instead, the truth-condition of the second-order thought has the act of first-order judging as a part. But I prefer to be more neutral about the form of the second-order judgment—it is contentious to assume the existence of ‘mental deixis’ (cf. Millikan 2001; 2006; 2012; Cappelen & Dever 2013). And it would be no great concession to allow that the judgment expressed by (1) uses Mentalish names instead to denote the judging-act and the proposition judged:

(4) J[S IS JUDGING A BY DOING B] J[WATER IS WET]

10 The act of judging should not be confused with the act of assigning a referent to hereby. In Burge (1977/2008; 1991), the latter act is required in deictic thought—yet it is normally a different act from the first-order judging per se. (In the case of (3), assigning hereby a referent is one act; but it is a different act that is so assigned.)

Infallibility about What One Judges  163 This is a case where the first-order judging and the proposition would be named rather than ostended by the second-order judgment. (And that would create a further difference between Davidsonian parataxis and the present variety.) Regardless of the details, however, the paratactic account seems to falter, for the judgment expressed by (1) can still end up false. It is entirely possible to assert (1) competently, non-metaphorically, sincerely, etc., yet for the first-order judging-act to go missing.11 In such a case, we would have something like: (5) J[I AM JUDGING HEREBY THAT] [WATER IS WET]

In this, even though the first-order thought is occurrent, it goes unjudged, whence hereby is nonreferring. (Or if something like (4) is more apt, the name b ends up empty.) Now cases of empty reference need not be cases where the second-order judgment is false. Perhaps it instead has a non-truth-evaluable content (if such a thing makes sense). Nevertheless, the judgment expressed by (1) is supposed to be invariably true for the infallibilist—and not merely “non-false.” So the point remains that (3) and (4) are not adequate to model an infallible judgment, given the possibility of empty reference. Conceivably, an infallibilist could reply by stipulating that there is no empty reference in the judgments of concern. Yet infallibilism then seems trivialized. Alternatively, one might hold that there simply is no thought content if reference to the first-order act winds up empty. But to the contrary, it is natural to think that (5) would be a genuine thought even if hereby is referentially empty. (There may be ways to argue against this, but I shall not press the issue.) A different fix would be to appeal to the grammatical rules of Mentalish. On this proposal, if the first-order component goes missing, then there would remain only a syntactic fragment—hence, no genuine ­second-order judgment to be right or wrong. So on this view, when an assertion of (1) expresses a judgment in the first place, it would invariably include the first-order element. In which case, it invariably expresses a true judgment. One can flesh this out further by supposing that the second-order vehicle includes a complementizer (indicated below as ‘cc:’). This would be a syntactic marker indicating that a first-order element is needed (qua complementclause) for the second-order vehicle to be well-formed. Accordingly, if the

11 N.B., sincerity means that the second-order judgment is genuinely occurrent, but that alone does not suffice for the first-order judgment. Contrast with Jacobsen as discussed in n. 7.

164  Knowledge of Judging first-order component goes missing, what remains is a non-truth-evaluable syntactic fragment, owing to the dangling complementizer. A complete second-order judgment might therefore look like this:

(6) J[I AM JUDGING HEREBY THAT CC: J[WATER IS WET]]

Again, the complementizer here suggests that if J[water is wet] goes missing, what remains is a non-truth-evaluable fragment, thanks to a dangling complementizer. However, there are two problems with the proposal. First, the second-order judgment would be at best infallible as a matter of psychological necessity. For infallibilism about paratactic judgment would then rest on a hypothesized psychological law, viz., that a Mentalish vehicle for a second-order thought includes a complementizer. But second, and more seriously, even psychological necessity would be too strong. The judgment expressed by (1) could still end up false, even if the psychological law holds. For the complementizer demands only the first-order thought as a complement-clause. It does not also require the act of judging the thought. Hence, if the thought is tokened sans the act of judging, (6) could still be well-formed yet false. We can summarize the situation by an inconsistent triad. Our infallibilist has some inclination to assert each of the following: (a) The judgment expressed by (1) semantically entails its own truth, as per (INF2). (b) If (a) is true, then the judgment expressed by (1) must reference an act of first-order judging. (c) The judgment expressed by (1) need not reference an act of first-order judging. Toward a resolution, I propose to restrict infallibilism to a hypothesized subtype of second-order judgment. In such cases, the judgment expressed by (1) will not denote the first-order judging via the usual sort of naming or demonstrating. Rather, a different kind of referring act would occur, a kind where empty reference is simply not possible.

7.3  Lagadonian Reference We can get acquainted with this alternative kind of mental representation by considering first a similar phenomenon in ordinary linguistic practice. Millikan (2012) gives an example in the following:

Infallibility about What One Judges  165 Consider, for example, the convention of writing or pasting labels on things—a can with a label on it that says ‘spinach’. . . . The conventional signs that result are not composed just of words. The word ‘spinach’ obviously says nothing hanging by itself in the air. . . . Signs, in the first instance, always sign complete propositions. . . . In the case of ‘spinach’ as a label on a can, the rest of the sign is the filled can. One might think of the relation of the physical word ‘spinach’ to the contents of the can—the relation being-on-a-container-with-X-contents— as the syntactic form of the sign. (p. 220) In Millikan’s case, there is a proposition signaled, namely that the can contains spinach. But note that in the example, there is no English expression of that proposition. After all, the only English present is the kind term ‘spinach’. Yet since a proposition is indeed expressed, there must be something that expresses it—and this vehicle or “propositional sign” likely consists in something in addition to the term ‘spinach’. Plausibly, Millikan suggests the proposition is expressed by the fact that the ‘spinach’ label is affixed to the can. So what represents that the can contains spinach is the spatiotemporal arrangement of the ‘spinach’-label and the can. (The proposition is expressed by the label being on the can.) But if so, then as a relatum of this relation, the can itself is part of the representational vehicle. A proposition about the can is expressed by a vehicle which has the can itself as a constituent. We can think of things this way. In Millikan’s example, the can is being “exhibited” in a certain manner; it is indicated to the audience as the locale of some spinach. This is accomplished by the can bearing the ‘spinach’ label. This spatiotemporal arrangement thus has the function of a representational vehicle—it is what expresses that the can contains spinach. Hence the can itself is a part of the representational vehicle, by being part of that spatiotemporal arrangement. In particular, the can is being used to indicate itself as a container of spinach. The can thus “completes the sign” for the proposition by acting as a self-referring “term” for the thing that contains spinach. Broadly speaking, we might say this is a case where the language “goes Lagadonian.” In Lewis (1986), a “Lagadonian language” (borrowed from a Jonathan Swift parody) is one where reference to objects is accomplished by holding up the objects themselves to one’s audience, rather than uttering linguistic proxies for them. As in Millikan’s example, the objects are then functioning to indicate themselves as the speaker’s intended referent. Naturally, a Lagadonian language would generally be inconvenient to use (as Swift makes plain). But as with the spinach can, it is sometimes more convenient to go Lagadonian, at least momentarily. That is so, assuming the audience can cotton on to the Lagadonian usage.

166  Knowledge of Judging Further details about Lagadonian reference will be necessary. For one, note that if the can is a Lagadonian term, it need not be that each part of the can is so used. A small dent on the side is not used to self-name; after all, whether or not the dent exists, the same proposition is expressed. As a special case of this, sometimes a Lagadonian “term” has a proper part that also happens to be an English term with its own English meaning. What’s more, in some such cases, both the English and the Lagadonian term contribute their meanings to the expression of the proposition. Since this will be crucial in clarifying the new infallibility account later, let us review the point more slowly. Consider that a thing is sometimes used as a Lagadonian term, even though it has an English expression as a proper part. Suppose for instance that Millikan’s spinach can has an imprint on the bottom ‘Made in Taiwan’. This bit of English would contribute nothing to expressing that the can contains spinach. But in some cases, an English expression is not idle in this way. Even though an English token may be a proper part of a Lagadonian item, the English token is used compositionally with the Lagadonian term. The Lagadonian whole and the English proper part then contribute separate meanings, composed together to express the relevant proposition. If the token of ‘spinach’ is literally a part of the spinach can, that is true in the previous example. But since the antecedent is contentious, consider a different example, again from Millikan: [C]onsider a stop sign. The convention is that a stop sign is placed just [at] the place at which one is to stop. Where the stop sign is placed signifies where one is to stop. The complete sign consists not just in a sign saying ‘stop’ but in its standing in a certain place. . . . So it seems that not only can an object such as a can and its contents become incorporated into a sign and made to stand for itself, a place can be incorporated into a sign and made to stand for itself. (2012, p. 221) The place functions as a Lagadonian term, referring to itself as the place where drivers must stop. (The place is “exhibited” as the place where one stops.) And the English token of ‘stop’ is a part of this place (its location is at that place)—though ‘stop’ is not being used in the Lagadonian way. After all, the proposition expressed is not about the word ‘stop’. Rather, it is one about stopping. So apparently the token of ‘stop’ is used with its English meaning when the proposition is expressed, even though the token is part of the place that is used, simultaneously, with a Lagadonian meaning. Still, it is no accident that they are simultaneously used. For the meanings they

Infallibility about What One Judges  167 individually express are composed in expressing a proposition about stopping at that place.12 For brevity, I will describe this as a case where the Lagadonian term has a “contra-Lagadonian” part: The whole contributes a Lagadonian meaning, yet it has a proper part that simultaneously contributes a non-Lagadonian meaning when the proposition is expressed. That is so, despite that element being a proper part of a Lagadonian “expression.” Initially, this can sound like the Lagadonian term is somehow semantically at odds with itself. But the whole is expressing a meaning independently of the meaning expressed by one of its parts, although the two meanings are being composed. And again, it is by the composition of these meanings that the proposition is expressed.

7.4  Semi-Lagadonian Judging Recall now that the infallibilist’s task was to circumvent the problem with empty reference. Thus far, nothing rules out that (1) expresses a mental parataxis where the act of first-order judging is omitted—whence the secondorder judgment contains a nonreferring term. Yet let us now consider the possibility of a Lagadonian term for an act of first-order judging. Such would be a judging-act which also has the function of self-referring or representing itself. Here, then, is the infallibilist’s new gambit: If we suppose a mental parataxis can self-attribute the first-order act by means of its Lagadonian name, then in those cases there is no issue about failed reference. After all, when the act itself is used in self-attributing the act, it is impossible for the self-attribution to occur in the absence of the act. The term used to attribute the act just is that self-same act! So if the act goes missing, then the attribution cannot be made in the relevant Lagadonian way. Thus if infallibilism is limited to such “semi-Lagadonian” second-order judgments, then all the relevant self-attributions are invariably true. For visual learners, the relevant parataxis now looks like this:

(7) J[I AM JUDGING THAT] J[WATER IS WET]

12 This apparently contradicts the natural view that ‘Stop’ is being used as a one-word English sentence, to express the command Stop here now, where the relevant time and place are implicitly ostended somehow. Millikan (2001) is especially convincing, however, in explaining why this view is inadequate (and the obscure notion of “implicit ostension” is a clue as to why). Unfortunately, however, I cannot discuss the matter here, and so I must leave this controversial commitment undefended.

168  Knowledge of Judging In (7), the latter half portrays (i) an act of judging, (ii) a self-naming of that act, and (iii) an occurrent thought being judged. The content of the parataxis is thus akin to “I am hereby judging p,” where ‘hereby’ denotes the first-order judging act. But instead of using the Mentalish equivalent of ‘hereby’, the act of judging is instead denoted by its Lagadonian name, i.e., itself.13 Yet here is a key clarification: The Lagadonian term in (7), i.e., the act of first-order judgment, features a contra-Lagadonian part. After all, despite being part of the Lagadonian judging-act, the first-order vehicle does not represent itself. Rather, the vehicle represents the state-of-affairs where water is wet. Nonetheless, the judging act remains self-representing. That is so, even though the act is an event that has the occurrent first-order vehicle as a part. But the vehicle still contributes a contra-Lagadonian content in the execution of the second-order judgment. That is why the second-order judgment concerns, inter alia, the state-of-affairs where water is wet. It is comparable to the contribution of ‘stop’ in Millikan’s example—where the term contributes a non-Lagadonian meaning, while being part of the Lagadonian term representing the place where drivers should stop. One might suggest, however, that the second-order thought is indeed signaling a proposition about Mentalish, at least in the “de re” sense of ‘about’. It expresses a proposition about the first-order act of judgment, and the act of judgment has a Mentalish vehicle as a part. If Mentalish is cognitively real, this is fair enough. Still, it would be similar to how the Lagadonian place represents de re the place with a token of the word ‘stop’—even though de dicto nothing is said about the word. In general, I have taken the intensionality of mental representation for granted. And it remains that in the intensional sense of ‘about’, a second-order thought about a judging-act is not a thought about every part of the act.14 All this remains obviously speculative, however, and Lagadonian mental representation at best has the status of an unconfirmed posit. Though I take the hypothesis to have some viability. And at the least, if we understand Burge’s account as describing “semi-Lagadonian” second-order judgments, it avoids the chapter 3 objection, as I shall elaborate in the final section. 13 Gertler (2001) and Balog (2012) defend somewhat analogous accounts for self-attributions of phenomenal states. Yet Gertler’s account invokes demonstrative rather than Lagadonian reference—and unfortunately, the possibility of empty ostension is discounted without much comment. Balog’s account instead invokes reference by means of quotation marks (or rather, by the mental analog thereof). Yet the mental operation of “quotation” is left mostly unexplained, and it is unclear how reliable it is in producing true phenomenal judgments. Nevertheless, a genuinely Lagadonian account of phenomenal self-attributions is found in Kriegel (2009); see also McClelland (2013). 14 Perhaps Lagadonian mental representation makes intensionality especially odd. Still, many explain intensionality by the inferential role of a mental representation (e.g., Lycan 1994, ch. 7; Fodor 2009, ch. 3, Sainsbury & Tye 2012, ch. 7). And it is not clear why it would be problematic to apply that account to Lagadonian mental representation.

Infallibility about What One Judges  169 In this respect, the present aim need not be to render the semi-Lagadonian account plausible, but rather just to offer it as an alternative and more charitable interpretation of Burge. We should acknowledge, however, there are special doubts about Lagadonian reference in thought versus in public linguistic practice. After all, it is not as if I am able to use a spinach can as a mental representation of the can. The can is not part of my cognition; it is not even in my mind. Actually, this might be debated in light of the “extended mind” program, but waive that. For nothing in the present account requires spinach cans or other external objects to function as mental representations. It requires Lagadonain usage only of entities that are patently in the mind, viz., acts of judgment.15 This accords with Millikan (2012), when she remarks that “[mental] self signs, given that external objects and properties cannot literally be in the mind . . . could at most be thoughts about perceptions or thoughts” (p. 230, italics mine). So for Millikan as well, it seems Lagadonian mental representation is limited to mental phenomena.16 This raises the question, however: What exactly does it mean for a mental state or event to be “used” in the Lagadonian way? Well, Millikan’s view is that it means the mental state/event causally impacts the subject’s inferences or outer actions, in the same way that other mental representations do (i.e., in the way we would expect of an information-carrier). Specifically, Millikan holds that the Lagadonian mental representation must “control” the subject’s inferences and outer actions in a way that “suits” it (Millikan 2012). I take this to mean, roughly, that the inferences/actions must be rationally guided by the information carried by the Lagadonian vehicle. One’s actions and inferences must be informed by the information carried. At any rate, if Lagadonian mental representation is permitted, the earlier worries about empty reference are quelled. A Lagadonian term is a case where empty reference is simply not possible. The term is identical to the referent; so necessarily, if there is such a term at all, then there is a referent. One might raise an analogous worry, however, about that in (7) occurring as an empty demonstrative. However, I assume here that an act of judging is individuated partly by its content. Consequently, if we guarantee that the 15 Lagadonian acts of judging can be motivated in other ways. Consider what Shoemaker (1996, pp. 288–289) calls “self-conscious thinking:” One apparently can contemplate something first-order with an attendant awareness of what one is thinking (even though this awareness does not seem to be a case of introspection, at least not of the paradigmatic sort). Shoemaker offers a brief account of the phenomenon, but admits it is sketchy. However, it could be further developed by assigning first-order judgments the function of representing themselves. If this is at all plausible, it would be another reason to posit Lagadonian mental representation. 16 This, by the way, can explain why Lagadonian self-verifying attributions are possible only qua first-person attributions. A thought can be attributed via a Lagadonian token only by the first-person, given that the Lagadonian token exists in the mind of the first-person alone.

170  Knowledge of Judging judging-act exists, we thereby secure that there is a first-order content, and thus, a proposition that is available to demonstrate. The upshot, then, is that the second-order judgment as portrayed at (7) necessitates the very thing that makes it true. It necessitates the act of firstorder judging, since the latter is a self-referring constituent of the former. “Necessitation,” again, means “necessitation in the manner of semantic entailment.” For the second-order judgment necessitates the first-order judging-act, assuming the latter functions as a term that is interpreted in a specific way.

*7.5  Infallibility without Mentalish One problem, however, is that infallibility here seems to be a matter of psychological rather than semantic necessity. For the infallibility holds only under the psychological hypothesis that an act of first-order judging can be used as a Lagadonian term (with a contra-Lagadonian part). Yet while the account indeed assumes this, it owes not to a psychological premise but a semantic premise—roughly, that there can be Lagadonian terms for events. It is true that this hypothesis has consequences for psychology: If one can use any event to represent itself, then one can use psychological events to represent themselves, including one’s own acts of judging. For my part, that semantics should have consequences for psychology is only natural. Still, I grant that this psychological import is substantive, and is open to falsification by empirical evidence. So the semantic hypothesis is proposed merely as a hypothesis. Still, since it is a semantic hypothesis, it remains appropriate to say that the infallibility is explained semantically—by ┌ I am judging that I am judging that p ┐ semantically entailing ┌ I am judging that p ┐. True, the psychological import of this means the entailment holds only in worlds where certain psychological states-of-affairs obtain. So from that angle, the modality is a psychological modality. But again, if the psychological import results just from a semantic interpretation, it still seems right to say that the infallibility owes to a semantic entailment. At least, the paratactic account does not require adopting any special psychological hypotheses, beyond what is implied by the semantic hypothesis. This is clearest if we now jettison the Mentalish device. It also helps if we can derive the infallibility thesis in a regimented language, where it is obvious that the only supplemental premises are semantic. That is what I proposed to do in this section. But the basic reasoning is one I have already expressed informally: Since a Lagadonian term just is its referent, the selfattribution of a first-order judging is strictly infallible, if it is attributed via its Lagadoinian name. When using the act as a representation of itself, the attribution of the act is inevitably correct. Henceforth, let ┌ J(S, p) ┐ be a sentential form expressing that a subject S is judging p. (Thus, ‘J(I, water is wet)’ expresses that I am judging that

Infallibility about What One Judges  171 water is wet, and so on.) Further, let ‘X’ name a function that takes an actual object X as input, and outputs the Lagadonian name identical to X. (This is the inverse of the Lagadonese interpretation-function.) Thus, ‘Obama’ names the output of the function when Obama is input. That is to say, Obama = Obama’s Lagadonian name (i.e., Obama himself, interpreted qua Lagadonian name). Similarly, snow is white is the Lagadonian name for the state-of-affairs where snow is white (i.e., that self-same state-of-affairs, as a piece of Lagadonese). Consider now the sentence: ‘J(S, J(I, q))’ where q is a proposition. This is roughly equivalent to: “S judges that ‘I am judging that q’,” where the act of judging is represented by its Lagadonian name. In the language, ‘J(S, J( I, q))’ can then be shown to semantically entail ‘J(I, q)’, as per (INF2). However, the proof requires adding the following rule in one’s favorite intensional logic. (The logic is intensional since ‘judging’ is intensional.) Call the rule “Existential Expansion:” ∃ Expand Φ(p) ∴ (∃x)[Φ(x) & x = p]

This particular rule licenses existential generalization, but only from formulae using a Lagadonian term (here restricted further to terms for states-ofaffairs). Unrestricted existential generalization is invalid in intensional logic, yet ∃ Expand isolates one type of case where that sort of inference indeed preserves truth, one where the quantified variable replaces a Lagadonian term. The rule can be glossed as “if a Lagadonian term for a state-of-affairs is deployed, then the Lagadonian term exists.” Relatedly, the state-of-affairs p exists or obtains if its Lagadonian term exists. Let us recognize this in a further inference rule; call it “Lagadonian substitution:” LagSub (∃x)x = p ∴p

The rule works since any Lagadonian term is guaranteed to be non-empty; such a “term” is identical to its object; hence, its object must exist if the “term” does. With these two new rules, the entailment between the second- and firstorder judgments now follows in short order: 1. 2. 3. 4.

J(S, J(I, q)) (∃x)[J(S, x) & x = J(I, q)] (∃x) x = J(I, q) J(I, q)

[Assume] [1., ∃ Expand] [From 2.] [3., LagSub]

172  Knowledge of Judging In translation, this reads: If the Lagadonian term for the first-order judgment is used in self-attributing the judgment, then there is such a term (by ∃ Expand). And if there is such a term, then the object of the term exists. In this particular instance, this means there exists an act of my judging that q. Note well: This does not establish that the second-order, “semi-Lagadonian” judgment logically necessitates the first-order judging. Even though the necessitation has been formally represented, this is only because we are holding fixed the interpretation of a non-logical element, viz., the underline as applied to expressions. If its interpretation is fixed, one can articulate rules like ∃ Expand and LagSub, and perform derivations as above. Even so, this is to go beyond the logical constants in fixing what is constant in the language. That is why we are talking about a semantic entailment between 1. and 4., an entailment that holds under certain semantic assumptions. But in demonstrating this entailment, the derivation establishes a refined version of (INF2), which stands are our final statement of the new Burgestyle infallibilism: (INF2') ┌ I am judging that J(I, p)┐ semantically entails ┌ J(I, p) ┐. That is, when the act of first-order judging is self-attributed via its Lagadonian name, this necessitates (in a way explicable by semantic entailment) the act of first-order judging. Similarly, if applied to the more modest (INF1), the present account supposes that a first-order act of thinking can be attributed via its Lagadonian name. This too would necessitate the act in the same fashion. In this way, Burge’s own infallibilism can achieve more than the psychological necessity offered by the chapter 3 compositionality account. The self-naming use of the judging-act means that the second-order judgment necessitates the act as a matter of semantic entailment. Thus, Burge: “By its reflexive, self-referential character, the content of the second-order judgment is logically locked (self-referentially) onto the first-order content” (1988, pp. 659–660). At this point, I worry that the infallibility starts to look trivial again. The appearance is lessened if we see the left-hand side of (INF2') as a proposed analysis of the judgment expressed by (1). But more importantly, if a person tokens a “semi-Lagadonian” mental parataxis, it is a striking case where ‘thinking makes it so’, i.e., where judging that Φ is enough to make it true that Φ. In executing the second-order judgment, the person is effectively authoring or creating the fact which she judges to be a fact. It is thus a novel sort of “self-verification;” judging that a certain state-of-affairs obtains is sufficient for its actually obtaining. That we have this authorship of these facts, that it stems from Lagadonian reference, and that it results in an infallibility explicable by semantic entailment, are far from trivial. The infallibility might trivially follow once we recognize the situation. But that this

Infallibility about What One Judges  173 situation arises at all is far from trivial. (Indeed, this is why the infallibility of these judgments do not secure their indubitability.)

7.6  Link with the Next Chapter The chapter 3 objection to Burge was that, if a second-order judgment features a self-referring element, then the second-order judgment is “locked” to itself rather than to the first-order cognition. But in light of the foregoing, the criticism mistakenly assumes that the self-reference is a reference to the second-order judgment. However, in a semi-Lagadonian second-order judgment, the self-referring element does not denote the second-order whole, but rather a first-order part. Considering this, the chapter 3 objection rests on a false premise.17 Be that as it may, a similar complaint would remain—namely, that the account explains only how the first-order act is semantically locked to itself. It does not explain how the second-order judgment is locked to the firstorder bit. This forgets, however, that a semi-Lagadonian, second-order judgment is “locked” to the judging-act, owing to a constituent that refers to the judging-act. And since this constituent refers in the Lagadonian manner, it is a case where empty reference is impossible. In such a scenario, the first-order component is “thought and thought about in the same mental act” (Burge 1988, p. 654). Burge has suggested (in conversation) that the present type of infallibilism may have other sorts of significance. Specifically, it might underwrite a fundamental norm about judgment (and I pursue this topic in the appendix to this chapter). A different ramification is that, if a non-metaphorical, sincere, competent, etc., use of “I judge that p” semantically necessitates judging that p, then it is strictly inconsistent to assert “~p” in the same breath. So on their normal use, Moore-paradoxical sentences (‘I judge that it is raining and it is not raining’) would not just be inconsistent qua speech-act, but semantically inconsistent (in the way that Wittgenstein suspected18). The primary motive, however, has been to counteract the Problem of Wayward Reflection. The infallibility of semi-Lagadonian judgments may offer some hope; however, it is, thus far, a meager hope. For one, as I mentioned at the start of this chapter, the infallibility does automatically translate to knowing what one judges. The connection with knowing-wh shall be pursued in chapter 10. Second, even if this connection is made, it is not very satisfying against Wayward Reflection to conclude merely that

17 This similarly rebuts Warfield’s (2005) claim that Burge’s self-verifying judgments are in fact self-falsifying. (Warfield argues that (1) falsely attributes the second-order judgment at the time that the first-order thought occurs.) Here too, the mistake is in assuming that the self-referential element is second- rather than first-order. 18 See also Heal (1994), Green (2005).

174  Knowledge of Judging sometimes, I know not when, I am infallible about my occurrent thoughts and judgments. That may provide some generic reassurance that the practice of reflection ultimately is not too ill-behaved. Regardless, it does little to show that in a particular instance, my reflections are successfully directed at my own judgments. Yet while actually engaged in reflection, that would be the more pressing concern. I would hope for some indication that in that specific case, I am effectively evaluating what I actually believe. In the next chapter, my ambition is to argue that there are some recognizable circumstances under which second-order judgments are infallible. When I recognize that these circumstances obtain, then, I will have reassurances that the particular reflections I have in those circumstances successfully target my own judgments.

Appendix: Infallibility as an Ideal of Rationality

More than once, it has been asked why judgment has been my primary focus. There seems to be nothing essential to the judging-attitude, such that the semi-Lagadonain account is especially appropriate in that case. One might equally propose that there are semi-Lagadonian (and hence, infallible) self-attributions of fears, or of desires, etc. The acts of fearing, of desiring, etc., could equally be hypothesized to be Lagadonian terms in some attributions of those very states. I am willing to allow as much, insofar as we are just considering hypotheses. Yet I am also willing to acknowledge that there is something distinctive about judging. If we assume (as always) the nine caveats and clarifications from section 1, then there is a sense that the self-attribution of a judgment is especially secure—more so than a self-attribution of a co-occurrent fear, desire, etc. In this appendix, I want to gesture at an explanation of this. But though I myself am fond of the explanation, it introduces more controversies than can be dealt with here. Thus I myself do not adopt it even as a hypothesis, but rather identify it as a possibility that might be further explored later. Let me be clear that, if the proposal is correct, it does not follow that a given second-order judgment is true, much less infallible (unless the judgment happens to be semi-Lagadonian, of course). Nevertheless, the explanation is suggestive of why in the normal cases, a second-order belief may well be infallible. (In this appendix, let ‘belief’ denote occurrent or non-occurrent belief, whereas ‘judgment’ exclusively denotes occurrent belief.) That’s because, roughly, it provides a reasonable framework wherein S’s believing “I believe that p” entails (model-theoretically) that S believes that p. Yet the entailment rests on the idealizing assumption that if S believes “I believe that p,” then she also believes (perhaps tacitly) the obvious consequences of that. However, people sometimes fail to believe even the “obvious” entailments of their beliefs; so the idealizing assumption is false, strictly speaking. Yet the proposal is that in a subset of cases (the “normal” cases, whatever those are), having the second-order belief entails that the subject has the first-order belief.

176  Knowledge of Judging That, in turn, could explain why all second-order judgments appear to have a special security, compared to self-attributions of first-order desires, fears, etc. But given the possibilities for deviant psychology, nothing should suggest that second-order judgments are universally infallible. (Although, a tendency to gloss over the deviant cases may constitute an error theory of why it seems otherwise to, e.g., a Cartesian.) The task, then, is to illustrate the framework on which S’s believing “I believe that p” entails that S believes that p. Start by observing that a subject’s beliefs can be seen as (partly) defining a model where all those beliefs are true. (Cf. Quine 1951, p. 11.19) Next, observe that, necessarily, p is true in a model iff p obtains in the model. (This is just the proposition-version of the disquotational T-schema.20) From this, we can infer the following (if MS is a model in which all of a subject S’s second-order beliefs are true): (8) Necessarily, p is true in MS iff p ∈ MS (where p ∈ MS if S believes p). Equivalently, (9) Necessarily, p is true in MS iff S believes that p. Suppose now that ┌ B(S, p) ┐ is a sentential form expressing the proposition that S believes p (where both occurrent and non-occurrent beliefs are relevant). Then, the sentential form ┌ B(S, B(S, p)) ┐ expresses that S believes that she herself believes that p (where ‘S’ is to be replaced only by an essential indexical or a functional equivalent). One corollary of (9) is then: (10) Necessarily, B(S, p) is true in MS iff B(S, B(S, p)) That is to say: Necessarily, S believes that S believes p just in case it is true in MS that S believes that p—which is just what (10) says. But take heed that in general, if p is true in MS, it does not follow that p. (I.e., it does not follow that p is true in the actual world.) This just reflects that S can have false beliefs. Nonetheless, if B(S, B(S, p) holds in the actual world, we are now in a position to derive that p obtains MS. And given how we have defined MS, this suggests that, necessarily, if B(S, B(S, p)) holds in the actual world, then B(S, p) does too! The derivation is as follows (where “SubEQ” is the rule for the substitution of strict equivalents): 1  B(S, B(S, p)) 2  B(S, p) is true in MS

[Suppose for conditional proof] [From (10) and 1.]

19 This Quinean idea of defining a model where all one’s beliefs are true is contentious, inter alia, since human beings can believe contradictory things. And the idea of a contradictory model may, itself, be a contradiction. This is a key reason why I hesitate to endorse as a viable hypothesis the proposal developed in this appendix. 20 I am assuming that truth is a property of propositions rather than sentences, though sentential truth may well be more appropriate. See note 22.

Infallibility about What One Judges  177 3  p is true in MS is true in MS 4  p is true in MS 5  B(S, p) 6 If B(S, B(S, p)), then B(S, p) 7  Necessarily, if B(S, B(S, p)) then B(S, p)

[From 2., (9), SubEQ] [From 3. and the T-schema] [From (9) and 4.] [By conditional proof] [6., by modal necessitation]21

Moreover, since B(S, p) is the condition on which B(S, B(S, p)) holds, the conclusion amounts to infallibilism about second-order judgment. As earlier indicated, however, the derivation rests on the idealizing assumption that the subject’s beliefs conform to the logic of the derivation. More concretely, classical (modal) logic is assumed to be sound in MS. But from a psychological point of view, that is not the case. It is well known that SubEQ fails in belief-contexts; e.g., if S believes that Cicero is an orator, it does not follow that S believes that Tully is an ­orator—even though necessarily, Cicero is an orator iff Tully is an orator. Similarly, if S believes that she herself judges that p, it does not follow that S believes that “p is true” is true in the model defined by her beliefs. (After all, in many cases, subjects will not even have the concept of a “model.”)22 Nevertheless, the derivation may explain the normative pressure on the subject to believe the propositions that she believes she believes. The idea would be that her beliefs ideally are governed by classical (modal) logic—and under the current plausible assumptions, the claim that she has the secondorder belief entails the claim that she has the first-order belief, according to that logic. So if she does not have the first-order belief in that case, that is an indication of how she falls short of the ideal. This also removes one concern about the derivation. One might have noticed that, since the premises are biconditionals, one could equally run the derivation in reverse, to show that if B(S, p), then B(S, B(S, p)). (This converse thesis is sometimes called “transparency” in the literature.) Yet this new thesis seems much more contentious. Indeed, as a claim about the actual psychology of human beings, I take it to be patently false. Even so, the derivation of the thesis might also explain whatever “normative pressure” exists to believe that you believe those propositions which you in fact believe. Though there is less pressure here, plausibly, because we do not expect a person to take an exhaustive inventory of her beliefs. Though in contrast, if ex hypothesi S believes that she believes p, it is then odd if she ends up lacking that first-order belief. If the second-order belief is merely a

21 The necessitation rule applies here, since the conditional proof will be sound in an arbitrary (classical) world. 22 On a related note, the inference from 2. to 3. in the above derivation also fails if truth is predicated of sentences rather than propositions. For there would be a substitution within quotes: ‘B(I, p)’ would occur inside quotes, yet would be replaced with ‘p is true in MS’ (such that the latter would also appear inside quotes).

178  Knowledge of Judging tacit belief, then her inattention might explain the error. Yet such things also indicate why the oddity is more pronounced in the case of judgment. For if S is judging that she believes p, then S has a significant degree of self-opacity if she does not actually believe p. By the way, the inconsistency derived explains why (absent any defeating evidence) there is a norm of interpreting S as believing p if S self-attributes this belief. Accordingly, it explains principles of charity, such as rule 4 mentioned in the Preamble, repeated here: (R4) If S is rationally committed to p, then commit to S’s believing that p (unless there is sufficient evidence against her so believing). And in particular, when S = you, one is pressured by norms of rationality to interpret yourself charitably as believing p, if you bother to self-attribute that belief. Basically, the derivation at 1.–7. shows that charity toward higher-order judgment is motivated by a desire to avoid the lower-order inconsistency that otherwise results. The sense was that second-order judgments were somehow especially secure—more so than self-attributions of fears, hopes, etc. But in this case, our intuitions did not lead to an argument for infallibilism; nothing in this appendix rules out false second-order judgments. Rather, the idea was to explain our initial intuition as stemming from certain norms of rationality. Thus, the proposed explanation is that the intuition derives from the fact that having the second-order belief entails (in the normal cases) having the correlative first-order belief. In the case of judgment, however, it is also explanatorily relevant that S is not just dealing with a tacit second-order belief (where error might be explained by S’s inattention). If the subject is explicitly judging that she believes that p, then the absence of the first-order belief seems like a significant departure from ideal rationality.

References Balog, K. (2012). Acquaintance and the mind-body problem. In C. Hill & S. Gozzano (Eds.), New perspectives on type identity: The mental and the physical (pp. 16–42). Cambridge UP. Burge, T. (1977/2008). Belief de re. Journal of Philosophy, 74(6), 338–362. Reprinted with a New postscript in his Foundations of Mind (pp. 44–81). Oxford UP. ———. (1988). Individualism and self-knowledge. Journal of Philosophy, 85, 649–663. ———. (1991). Russell’s problem and intentional identity. In E. Lepore & R. Van Gulick (Eds.), John Searle and his critics. Basil Blackwell. ———. (1996). Our entitlement to self-knowledge. Proceedings of the Aristotelian Society, 96, 91–116. ———. (1998). Memory and self-knowledge. In P. Ludlow & N. Martin (Eds.), Externalism and self-knowledge (pp. 351–370). Stanford: CSLI Press.

Infallibility about What One Judges  179 ———. (2003). Reply to Kobes. In M. Hahn & B. Ramberg (Eds.), Reflections and replies: Essays on the philosophy of Tyler Burge (pp. 417–434). MIT Press. ———. (2007/2011). Self and self-understanding: The Dewey lectures. Journal of Philosophy, 108, 287–383. Cappelen, H. & Dever, J. (2013). The inessential indexical: On the philosophical insignificance of perspective and the first person. Oxford UP. Davidson, D. (1968). On saying “that”. Synthèse, 19, 130–146. Dennett, D. C. (1975). True believers: The intentional strategy and why it works. In A. F. Heath (Ed.), Scientific explanations: Papers based on Herbert Spencer lectures given in the University of Oxford (pp. 53–75). Oxford UP. ———. (1991). Real patterns. Journal of Philosophy, 88, 21–51. Fodor, J. A. (1975). The language of thought. Harvard UP. ———. (2009). LOT2: The language of thought revisited. Oxford UP. Gertler, B. (2001). Introspecting phenomenal states. Philosophy and Phenomenological Research, 63, 305–328. Green, M. (2005). Moorean absurdity and seeing what’s within. In M. Green & J. Williams (Eds.), Moore’s paradox: New essays on belief, rationality, and the first person (pp. 189–216). Oxford UP. Heal, J. (1994). Moore’s paradox: A Wittgensteinian approach. Mind, 103, 5–24. Jacobsen, R. (2009). Davidson and first-person authority: Parataxis and self-expression. Pacific Philosophical Quarterly, 90, 251–266. Kriegel, U. (2009). Subjective consciousness: A self-representational theory. Oxford UP. Lewis, D. K. (1986). On the plurality of worlds. Basil Blackwell. Lycan, W. G. (1994). Modality and meaning. Dordrecht: Kluwer. McClelland, T. (2013). Receptivity and phenomenal self-knowledge. Thought, 2, 293–302. Metzinger, T. (2014). Cognitive agency. Edge. Available at http://www.edge.org/ response-detail/25446. Millikan, R. G. (2001). The myth of mental indexicals. In A. Brook & R. Devidi (Eds.), Self-reference and self-awareness: Advances in consciousness research vol. 11 (pp. 167–181). John Benjamins. ———. (2006). The varieties of meaning: The 2002 Jean Nicod lectures. MIT Press. ———. (2012). Are there mental indexicals and demonstratives? Philosophical Perspectives, 26, 217–234. Perry, J. (1979). The problem of the essential indexical. Nous, 13, 3–21. Quine, W.V.O. (1951). Ontology and ideology. Philosophical Studies, 2, 11–15. Sainsbury, R. M. & Tye, M. (2012). Seven puzzles of thought. Oxford UP. Schwitzgebel, E. (2011). Perplexities of consciousness. MIT Press. Shoemaker, S. (1996). The first-person perspective and other essays. Cambridge UP. Warfield, T. (2005). Tyler Burge’s self-knowledge. Grazer Philosophische Studien, 70, 169–178.

8 Infallibility in Knowing What One Expresses

Suppose the Sartrean “Other” sees me looking through a keyhole. “What are you doing?” she stammers. Suppose I reply in all honesty, “My key wasn’t working—I’m thinking that the keyhole is jammed.” The response indicates that I was not spying (at least, not intentionally), but rather checking the hole for blockage. This is so, even if my bodily behavior is indistinguishable from that of a spy. It is a case where the act is individuated partly by the intent. This also holds of a second intentional act in this example, viz., the assertion given in reply to the Other’s question. (The intent of the second act was to communicate the intent of the first.) Moreover, part of this assertive act was of a special sort; it was an expression of the second-order belief that I believe that the keyhole is jammed. What is the nature of such an assertion? Like many acts, this importantly depends on the intent. For instance, if the intent is to introspectively report on a standing mental state, then the intent is purely descriptive, one might say. Yet what if the assertion concerns what I occurrently believe? In such a case, my assertion might be not so much descriptive as expressive of the judgment (BarOn & Long 2001, Bar-On 2004a, b, 2009, 2010a, b, 2011, 2012a, b, 2013, 2015). That is, rather than reporting on some persisting state, I am giving voice to or speaking from the first-order judgment while it occurs—in the manner of “thinking out loud” (Bar-On 2004a, pp. 217–18). That is so, even though semantically, the assertion looks purely descriptive. But again, intent is important to individuating acts, and that goes for assertion as well. Thus, in asserting, “I’m thinking the keyhole is jammed,” an expressive intent can mean the assertion is less reportive and more performative of the first-order judgment. Accordingly, one’s sincere expression of the second-order judgment “I am thinking that p” might be seen as simultaneously an expression of one’s first-order judgment that p. Question: In that case, is the assertion infallibly correct? No—the second-order assertion can be false “due to some kind of failure or irregularity on the part of the avowing subject” (Bar-On 2004a, p. 10). Or, following some remarks from the last chapter, it is possible to assert “I judge that p” in a conventional, competent, and sincere manner,

Infallibility about What One Expresses  181 while the first-order act of judgment goes missing. This is what was illustrated at (5) in the previous chapter, repeated here: (5R) J[I AM JUDGING HEREBY THAT]

[WATER IS WET]

Again, this depicts a false second-order judgment, for there is no act of judging toward the proposition expressed by the Mentalish complement-clause. In which case, what is expressed by “I am judging hereby that water is wet” is false. Regardless, what of the reverse situation? Suppose I answered the Other just by first-order asserting “The keyhole is jammed.” Yet suppose the intent was the same, to explain away my spy-like behavior by saying what I judge. Then plausibly, my first-order assertion also expresses the secondorder judgment “I am judging that the keyhole is jammed.” And in this sort of case (if the first-order assertion is sincere) is the second-order judgment infallible?

8.1  Introduction to the Criterion This chapter argues for a positive answer under certain provisos. In this, it forges a link between one type of linguistic act and self-verifying secondorder judgments (as defended in chapters 3 and 7). Bar-On (2004a) has fashioned this sort of link before. Yet on her view, self-verifying phenomena are to be explained by expressive acts—whereas I shall suggest that one type of expressive act is underwritten by a self-verifying judgment.1 (But I do not wish to quibble over whose order-of-explanation is better; it may depend on one’s theoretical interests at the time.2) Chapters 3 and 7 can be seen as showing, at the cognitive level, how positing self-verifying judgments is viable, whereas this chapter argues that one sort of linguistic act is self-verifying. The idea is not the simplistic one that if I assert, “I judge that p,” then it must be that I judge that p. Nor is the claim that whenever one asserts “p,” it is correct to self-attribute the judgment that p. Rather, the claim will be that in one specific sort of context (to be described), there is a specific type of asserting (to be described) which not only expresses the speaker’s judgment that p, but also a self-attribution of her judgment that p. And, since the occurrence of the former necessitates the truth of the latter, the

1 Burge (2003; 2007/2011) notes this kind of explanatory link as well. But as he admits, he does not pursue it in any detail. In part, the present work is an effort to remedy this. 2 Part of the difference with Bar-On may be only verbal, owing to different uses of the term ‘expression’. See n. 8.

182  Knowledge of Judging self-attribution is infallibly correct. However, as in chapters 3 and 7, none of the infallibilist’s arguments are themselves seen as infallible. The arguments below adopt a number of hypotheses about speech-acts, the nature of judgment, and so forth. And given the defeasibility of these hypotheses, the main contention is defeasible as well. Yet since the hypotheses can be defended, they are able to make rationally respectable the idea that one type of assertion is a “sure sign” of an infallible self-attribution.3 The motivation for all this, again, is the Problem of Wayward Reflection (cf. chapter 1). To repeat, the worry is that a person cannot critically reflect on her beliefs without knowing what those beliefs are. But the psychological evidence is overwhelming that we are often ignorant of our own beliefs. So potentially, the activity of reflection just shows naivety about psychology. (Or, since you and I already know about the evidence, our attempts at reflection may just show “epistemic bad faith,” as Sartre might say.) Yet if one type of speech-act is an infallible sign of a true self-attribution—then in recognizing the speech-act, one might know that the expressed self-attribution is correct, and that the self-attributed judgment is not simply confabulated. Granted, there’s no reason to think that any true self-attribution will be accompanied by the sign. But there can be a substantive class of self-attributions whose truth one can justify by appeal to the sign. Note well: the attentive reader may see a shift in the last paragraph. It is one thing to argue that there is an infallible sign of a true self-attribution, and another to argue for an infallible sign of an infallible self-attribution. For the most part, only the weaker claim will be of concern, since the stronger claim is unnecessary for opposing Wayward Reflection. (If there is a “sure sign” that my self-attributions are de facto correct, that suffices to know my reflections are non-Wayward.) Still, the relevant speech-act phenomenon makes a strong case for the existence of infallible, semiLagadonian judgments, as per the last chapter. So I eventually defend the speech-act as an infallible sign of such infallible judgments as well. (This discussion also helps illuminate the metaphysics of paradigmatic semiLagadonian judgments.) But the defense of this stronger claim is intended more to complete the case from the last chapter. It is thus regulated to an appendix. As a means to defending the weaker claim, the bulk of the chapter argues for the following “dual expression” thesis. (The term is adapted from BarOn 2004a, pp. 307–311.) Here and throughout, assume that “p” is a declarative sentence (or an ellipsis thereof), and p is a proposition that linguistic

3 The relevant speech-act will not be a “sure sign” in the sense of being an indubitable sign. A sign x can be an infallible indicator of y, even while we earnestly doubt it. What’s more, our fallibility in recognizing the sign should be conceded (though I will suggest that one can be reliable in recognizing it).

Infallibility about What One Expresses  183 convention assigns to the sentence on the occasion of use. Then, for some context c, to be specified, the dual expression thesis is roughly that: (DE) If a speaker asserts “p” in a standard fashion in c, then (a) she expresses her own judgment that p, and (b) she expresses a judgment that she judges that p. If (DE) or the like is true, then there is a kind of context where a standard use of “p” expresses both the speaker’s judgment that p and a self-attribution of that same judgment. And if such co-expression occurs, it necessitates the truth of the relevant attribution; the truth of the “I judge that p” is semantically entailed by the truth of (a) in the thesis. In that respect, (DE) leads immediately to an infallible sign of a true self-attribution. Namely, if c is the type of context in which (DE) is satisfied (and ‘Necessarily’ expresses the modality of semantic entailment): (INF3) Necessarily, a standard assertion of “p” in c expresses a true second-order judgment. If something like (DE) is shown, it may act as an infallible guide to some true self-attributions. (More on this later.) I thus often refer to (INF3) as a criterion for true self-attributions. Before moving on, note that such a criterion may seem more ready to hand than has been acknowledged. When a person expresses a belief by sincerely asserting “p,” she normally knows (at least implicitly) that this is what she is doing. (Assertion is an intentional action; and it is likely, following Anscombe 1957/1963, that intentional action implies knowing what one is doing, at least implicitly.) So at the time of the assertion, normally a speaker (implicitly/) knows what she occurrently believes, given her intentional expression of the belief. Hence, why not just treat all assertions as infallibly indicating an (implicit/) true second-order belief?4 I am not unfriendly to the idea. Even so, this is not adequate for my purposes. Since the higher-order beliefs would often be non-occurrent, they would often be irrelevant to the Problem of Wayward Reflection. For during reflection, the relevant higher-order beliefs are occurrent rather than merely dispositional. Concordantly, a sincere assertion of “it is wrong to eat meat” is not always an occurrent reflection on one’s belief about eating meat, even if there is a disposition to correctly self-attribute that belief.5 4 This also recalls Shoemaker (1996): “To the extent that a subject is rational, and possessed of the relevant concepts (most importantly, the concept of belief), believing that p brings with it the cognitive dispositions that an explicit belief that one has that belief would bring, and so brings with it the at least tacit belief that one has it” (p. 241). 5 Actually, we shall see that an assertion is sufficient for a reflective state about the belief in the right context. But very often, to assert what you believe is not to be in a state of reflection, even if the belief exists alongside a disposition for infallibly self-attributing the belief.

184  Knowledge of Judging Instead, my plan will be as follows. After introducing some terminology [Section 8.2], I will specify a context [Sections 8.3 and 8.4] where something like (DE) appears satisfied by any assertion if uttered in a standard way. It is then demonstrated that this appearance is not misleading [Section 8.5]. The idea is filled out further in light of some objections [Sections 8.6 through 8.8], and I close by sketching in more detail how later chapters use the account against the Problem of Wayward Reflection [Section 8.9].

8.2  Some Terminology Assume here and throughout that the meaning of a sentence “p” is neither ambiguous, nor metaphorical, nor overly vague, nor otherwise obscure to a problematic degree. Then, define the term ‘asserting’ as follows: (A) Necessarily, a speaker asserts “p” iff she utters “p” competently, with an intention of communicating a judgment that p (which may or may not be her own judgment). Asserting here is understood as a type of speech-act, per usual. However, (A) is somewhat nonstandard in that it talks of asserting a sentence, rather than a proposition expressed by the sentence. But if preferred, “asserting” can be seen as shorthand for “asserting the proposition conventionally assigned to the sentence (in the context of use).” Relatedly, note that (A) is restricted to declarative utterances used competently—which is to say (at least) that the utterance semantically expresses the proposition assigned to “p” by linguistic convention.6 Yet such restrictions are just meant to establish a focus. The case for (DE) and (INF3) will ultimately be less contentious under these limitations (although I think the arguments could be extended beyond them). The definition at (A) is about specific types of verbal acts. However, it can be readily used to characterize the products of these acts as well, i.e., “assertions.” Yet the term ‘assertion’ itself is, strictly speaking, equivocal between an act and the product of the act. However, in this discussion it shall denote the product exclusively. Whereas, when a noun for the act is needed, I shall

6 Unlike previous chapters, I use ‘proposition’ here rather than ‘thought’ to talk of content. That’s because my talk of “mental states” below does not cover thoughts qua contents. (The term ‘proposition’ helps keep this clear, yet in the end I think of contents as thoughts. Ultimately, “propositions” must be mental entities somehow . . .) ‘Competent’ is here applied to individual utterings rather than to speakers. Thus, I shall not talk of “competent speakers,” but rather to particular utterings as competent or incompetent. (So on the present usage, a native English speaker can still use an English term “incompetently,” due to a performance error or whatnot.)

Infallibility about What One Expresses  185 use ‘asserting’ as a gerund. This gerund-rule shall also extend to other act/ product equivocal terms, such as ‘judgment’ and ‘expression’. ‘Expression’ is a key term in what follows, and it remains equivocal in a further respect. An assertion expresses a proposition that p in a semantic sense; yet a speaker can also “express” a mental state such as a judgment, a desire, or what have you. In the first sense, expressing concerns a semantic relation between an assertion and a proposition (I call it “semantic expressing” or “s-expressing for short”). In the second sense, expressing concerns a relation between a speaker’s action and a mental state (“action-expressing” or “a-expressing”). In brief, an act expresses a speaker’s mental state iff communicating the mental state is an intention in the act.7 Our concern shall be primarily with the latter, although strictly speaking, “action-expressing” need not be a linguistic act. There are numerous kinds of acts, and numerous mental states, where communicating the state is an intent in the act.8 However, it will be best to limit ourselves to acts of asserting, performed with an intent of communicating a judgment in particular (i.e., an occurrent belief). This focus is again to make things less contentious in the end. To avoid further nomenclature, let us stipulate that ‘expressing’ hereafter shall denote this specific type of action-expressing. (I doubt this will cause confusion, as long as it is kept distinct from s-expressing.) Our definition of ‘expressing’ shall thus be as follows: (EX) Necessarily, a speaker expresses a judgment that p iff she utters a sentence competently with an intent of communicating a judgment that p (which may or may not be her own judgment). Moreover, I will assume that expressing a judgment suffices to communicate the judgment in standard communicative conditions (i.e., conditions where the audience can hear the speaker, where the assertion uses just terms that the audience understands, etc.). Observe further that (A) and (EX) immediately entail: (AEX) Necessarily, if a speaker asserts “p,” then she expresses a judgment that p (which may or may not be her own judgment). This clarifies that asserting “p” always comes with an act of expressing a judgment that p. But the reverse is not always true: I can express a judgment that p not by an assertion that has p as its conventional meaning, but by 7 I am here following Bar-On’s (op. cit.) distinctions, who is in turn following Sellars (1969). These authors further acknowledge a third kind of expressing, “causal expressing,” but I shall leave that aside. 8 Indeed, Bar-On even says that some events in thought can count as action-expressions. But I shall leave the idea of “thought expressing” aside.

186  Knowledge of Judging one that has a judgment that p just as a conversational implicature (cf. Grice 1975). Regardless, asserting is always the means by which one expresses a judgment. Relatedly, note that the patient of an assertive act is a sentence (or if preferred, the proposition s-expressed by the sentence). Whereas with expressing, the patient of the act is a judgment. In cases of asserting, I shall assume for simplicity’s sake that any expressed judgment that p is so expressed because either p is s-expressed by the asserted sentence, or the judgment that p is a conversational implicature of the asserting. If the former holds, I shall say that the judgment is expressed overtly; if only the latter, that it is expressed pragmatically. Note that in ordinary cases of assertion, a speaker can overtly express her judgment and pragmatically express other judgments simultaneously. Thus, in response to the question, ‘How is Bob doing these days?’, one might assert, ‘He hasn’t been to jail yet’. (The example is Grice’s.) In this, a speaker can overtly express her own judgment about Bob’s clean jail record, while also conversationally implicating her judgment that Bob is a troublemaker, the sort that is likely to be a jailbird. Thus, a speaker can use an assertion to express overtly one of her judgments and, at the same time, express pragmatically another such judgment. Per (EX), this means that one intention in her asserting is communicating the one judgment, but it also means there is a second intention in the act, viz., that of communicating the second judgment. A final point of terminology: The previous sentence uses the notion of “intention-in,” which I borrow from Anscombe (1957/1963). This must be distinguished from her notion of an “intention-to” (and from the general notion of intentional action).9 Anscombean intention-to always pertains to some future act—it can be thought of as a possibly unconscious (futuredirected) plan. But an intention-in is something different. For example, if a basketball is suddenly hurled at my head, covering my head with my arms is a spontaneous, automatic behavior. Preventing harm to my head is the intention in the behavior; though ordinarily, we would not say that I had a plan to protect my head (not even in the unconscious). The threat arose too quickly to design and implement a plan; instead, my behavior was just triggered in an automatic, mechanistic way.10 Speaking broadly, an intention-in concerns a telos or end which rationalizes an act qua means to an end. (N.B., I often use ‘intent’ as short for ‘intention’; and, following Anscombe, I use ‘intention with’ as notational variant

  9 Re: “intentional action,” if I twiddle my thumbs, we usually would say I do this intentionally, but not that I have an intention in so acting. (Nor would I have ordinarily planned to twiddle my thumbs.) A helpful gloss on Anscombe’s three-way distinction is: (a) plan to act, (b) telos of an act, and (c) act done knowingly and voluntarily. 10 Actually, English allows us to say there was a non-conscious “intention to” protect my head. (Yet it is still not as if I had a plan in the unconscious.) But in all probability, this really denotes an intention-in. Since the act was a mere reflex, talk of a plan seems like overintellectualizing, even though protecting my head was the intent in my act.

Infallibility about What One Expresses  187 on ‘intention in’, as seen in (A).) Now it is possible for the telos of act to derive from an agent’s plans. For instance, I might antecedently plan to keep my head covered throughout a game of dodgeball. There, if the telos of the plan is to protect my head, then protecting my head is the telos of the act as well. But intentions-in need not derive from plans, since again, some wholly spontaneous actions nonetheless have a telos. The distinction between intention-in and intention-to is important, since eventually we shall focus on unplanned and automatic assertings as a sign of an infallible self-attribution (in one specific sort of context). Qua unplanned, such “reflex-like” assertings do not result from an intention-to; nevertheless, there is still a communicative intention in a wholly spontaneous asserting.11 It is important, moreover, that this communicative telos is more readily recognizable than one’s true plans behind one’s asserting (if any). For while the former is constant, the latter are variable and all too easily hide from one’s conscious awareness. Thus, you might assert with any number of intentions-to (e.g., plans to deceive, to betray), which you may or may not be conscious of. Regardless, communicating a judgment that p is always one intention in asserting “p,” albeit perhaps one of several. Thus, we shall see that when our concern is with the communicative intention in a wholly unplanned assertive act, any hidden plans in the unconscious will not be relevant.12

8.3  The Criterion Again and a Sketch of Its Argument With the present terminology, the dual expression thesis can be understood more precisely as follows. There is a context c (to be specified) such that: (DE') If a speaker asserts “p” in a reflex-like way in c, (a) she expresses her judgment that p, (b) and she [pragmatically] expresses a judgment that she judges that p.

11 During a reflex-like uttering, what determines the intention-in, if not an intention-to? Per Millikan (1984; 2005, etc.), the communicative intention-in may then be fixed by the “Normal” use of the expression in the community. The Normal use need not be the use that most often occurs; it is rather the use that explains the continued proliferation of the expression in the community. (“Normalcy” is thus akin to normalcy in evolutionary biology: Sex normally results in pregnancy, yet most sex does not, thank god.) See also Green (2007) for a well-informed evolutionary view of expression. Yet nothing in what follows depends on the evolutionary gambit. I mention it only to give some idea on how one might understand communicative intentions-in with respect to reflex-like uttering. 12 By the way, while inconsistent plans are possible, I assume that the only possible telos is a consistent telos: An act has a telos in virtue of being a means to an end—and plausibly, a behavior cannot be a means to an end if the end is impossible. (As a result, when a person plans to attain a de facto impossibility, I would say her act either has no telos or its telos does not derive from her plan. Though she may falsely believe that her plan fixes the telos of her act.)

188  Knowledge of Judging Akin to (DE), if (DE') is true, then there is a context where a specific type of assertion suffices to express simultaneously the speaker’s judgment and a self-attribution of that self-same judgment—whence the former necessitates the truth of the latter. In this way, any reflex-like assertion in the relevant context will be vindicated as an infallible sign of a true second-order judgment. (DE') thus leads immediately to: (INF3') Necessarily, any reflex-like asserting in c [pragmatically] expresses a true second-order judgment. Again, (INF3') will be of use in identifying a “sure sign” for a true self-attribution. The task now is to specify the context c. The basic thought is that a reflexlike assertion of “p” satisfies (DE') if it occurs in the context of a certain type of question. After a few preliminaries, I shall identify the question in question. Subsequently, I will demonstrate under plausible assumptions why a reflex-like act of replying will satisfy (DE'). A key intuition behind the “question-theoretic” criterion was once articulated by Bar-On, when discussing “constitution” views of self-knowledge (from, e.g., Shoemaker 1996). Bar-On is somewhat hesitant about such views, yet allows that: If I take up a question about how I feel about something as an invitation to make up my mind how to feel about the matter, then it makes sense to suggest that the authority I have over the answer is of the ‘constitutive’ type. For presumably, in such a case, there is no antecedent fact of the matter . . . or if there is, it is not what I aim to pronounce on . . . [acts of expressing] would be seen as performative acts which bring into existence the relevant state of affairs . . . [e.g.] deciding what to want, believe . . . etc. (2004a, p. 141) So to illustrate, suppose a pollster asks you, ‘Who will be the next U.S. president?’. Then, if you never considered the topic before, the judgment expressed by a sincere reply of ‘Hillary’ would invariably be the judgment that the pollster is inquiring after. Of course, you remain free to change your reply upon further reflection. And the pollster might have more interest in the latter judgment than in your initial, knee-jerk reaction. Regardless, your most recent (sincere) reply would be an expression of a judgment which, necessarily, is the judgment of interest at the time. However, the view to be defended is more ambitious than what this can suggest. For an expressed judgment shall be the judgment of interest, even when one has previously considered the pollster’s question. That is so, as long as the pollster’s question concerns only what you (re-)occurrently

Infallibility about What One Expresses  189 believe.13 Relatedly, however, this means that the expression of the judgment might not “bring into existence” the relevant psychological facts. Instead, as I shall explain later, the view is that sincere replying determines which judgment is the judgment of interest. (The act of replying is not the existencecondition for the judgment, but rather an individuation-condition.) Before moving on, let me provide an intuitive presentation, albeit an approximation, of the basic argument for (INF3'). Suppose that someone asks you the oddball question, “What will you say next?” And suppose that you reply in a reflex-like way, “Mammals are animals.” Now, does this appropriately answer the question? Well, your reply is semantically offtopic; after all, it s-expresses nothing about what you’re saying and when. But normally, your act of replying would conversationally implicate something that is topically appropriate—viz, a judgment that What I am saying next is: “Mammals are animals.” In virtue of that implicature, then, we can see your asserting as replying to the question. Notice, moreover, that your reply is what you’re saying next. Accordingly, your reply is sufficient for the truth of that conversationally implicated judgment. Your replying is thus, in one sense, “self-verifying”—although not in the sense as from previous chapters. For it suffices for the truth of a contextually triggered conversational implicature, rather than the truth of the s-expressed proposition. And the point generalizes: Any reflex-like replying in the context of this question-type is invariably sufficient for (pragmatically) expressing a true judgment-attribution. That is basically what (INF3') says.

8.4  Evansian Autological Examination From another angle, the criterion comports with Bernard Williams’ (2002) remark that in the appropriate sorts of conditions, “I am confronted with my belief as what I would spontaneously assert” (p. 76).14 The trick, however, is to articulate when the appropriate conditions obtain (and when they suffice for explicit knowledge of judgment in particular)—as well as to 13 Of course, actual pollsters do not usually limit their inquiry in this way, for this version of the query regards as irrelevant your doxastic history, including what you usually believe on the matter. But this will be a helpful idealization. 14 Still, the chapter is not simply an elaboration of his claim. (I discovered the Williams quotation only after writing the bulk of it.) Indeed, Williams’ aim is to account not for self-knowing as much as norms of saying (a) what is true (“Accuracy”) and (b) what you believe to be true (“Sincerity”)—in reply to postmodern skepticism about truth. In the context of the quote, his observation is just that spontaneous assertion typically satisfies Sincerity, meaning that Sincerity does not demand a “special exercise . . . of discovering what it is that I believe” (ibid.). (Coincidentally, this fits the “anti-intellectualism” of section 8.7 below. But unlike myself, Williams does not explain this primitive ability to express our own beliefs . . . much less by means of neural network models.)

190  Knowledge of Judging articulate an argument for why my spontaneous assertion, in the relevant conditions, is indeed an accurate guide to my own belief. The questiontheoretic idea, mentioned above, will be crucial on these matters—and in developing it, we will draw form independently plausible considerations about the logic of questions and answers. 8.4.1  Autoexams and Replies to Autoexams Such considerations may be broached by first recruiting Huw Price’s (2006) distinction between “autological” and “heterological” examinations. Price explains the distinction thus: The first kind . . . is familiar from popular ‘know thyself ’ questionnaires, the point of which is simply to reveal the examinee’s own state of mind, in relevant respects. Thus if I’m the examinee, and the first question is ‘Was Aristotle a Belgian?’, I should answer ‘Yes’ if I believe that Aristotle was Belgian and ‘No’ if I believe he was not. The crucial point is that . . . as long as my answers accurately reflect my own beliefs, I won’t lose any marks. Call this the autological examination, or autoexam for short. . . . [In contrast, in another sort of context] someone who answers ‘Yes’ to the question ‘Was Aristotle a Belgian?’ loses a mark—gets the question wrong, as we’d say—because, as we’d put it, ‘Aristotle wasn’t a Belgian’. . . . [That is] heterological examination. (p. 607) What is striking is that in the two cases the same interrogative is uttered, and the same ‘Yes’-reply is tokened in response. But although this is an inappropriate reply in the heterological case, it may well be appropriate in the autological case. Oddly, then, an “appropriate reply” need not be true.15 In the autoexam, if a person sincerely asserts ‘Yes,’ she has affirmed a falsehood about Aristotle. But since she is sincere, the assertion is an appropriate reply. Whereas, if she insincerely asserts ‘No’, this is not appropriate—despite s-expressing a truth about Aristotle. What is going on exactly? In formal semantics, it is reasonably well-established that an appropriate reply to a question communicates (under standard communicative conditions) the information the inquisitor is seeking.16 (See, e.g., Boër & Lycan 1986, Ginzburg 1995, Ginzburg & Sag 2000, Hookway 2008.) Accordingly, in considering the heterological exam ‘What nationality is Aristotle?’, it may 15 I speak of “an appropriate reply” since usually there is more than one. E.g., the respondent might also appropriately reply with ‘Affirmative’, ‘Correct’, etc. 16 Again, “standard” communicative conditions are when the audience can hear the reply, when the reply is put in terms the audience understands, etc. For brevity’s sake, when discussing what communicates what, I henceforth drop ‘in standard communicative conditions’, although the clause should be seen as implicit.

Infallibility about What One Expresses  191 or may not be appropriate to reply with ‘He isn’t Belgian . . . ’. Whether it is appropriate would depend on the information the inquisitor wants. If she is concerned exclusively with who is Belgian, then the reply may be quite in order. But the same reply is inappropriate if the inquisitor wished to know Aristotle’s nationality by name. That is so, even though her assertion s-expresses a truth, and does so in response to a heterological exam. So even with heterological exams, truth is not sufficient for an appropriate reply. An assertion must also communicate the specific information the inquisitor seeks. Yet there is a further complication: Under present assumptions, an assertion can communicate information via its semantic content or via a conversational implicature. And sometimes it is only an implicature which makes the assertion into an appropriate reply. Price’s example is a case in point: A sincere assertion of ‘Yes’ is an appropriate reply if “Was Aristotle Belgian?” is posed as an autoexam. That’s because it communicates what the respondent judges on the matter. However, this occurs not in virtue of its semantic content, since the reply s-expresses nothing about judgment. Rather, it occurs in virtue of a conversational implicature, re: what the speaker judges. Despite s-expressing something false, then, a reply of ‘Yes’ is appropriate, and that is only because of the implicature. For expository reasons, the foregoing points about exams were illustrated by examples of appropriate replies. However, since the normative term ‘appropriate’ can raise suspicion (and since our concern shall be exclusively with autoexams), I shall often speak instead of “sincere” replies. It turns out that the term ‘sincere’ requires some caution as well (see below)—but a “reply” (appropriate or not) can now be adequately defined for our purposes as follows: (R) Necessarily, an assertion of “p” is a reply to a question Q iff: (i) the assertion is a response to the stimulus of Q being asked, (ii) the speaker utters “p” competently with an intent of communicating a judgment on Q (albeit not necessarily qua “judgment on Q”). One notable idea in (R) is that the communicative intent in a reply to a question Q is always “on topic:” By definition, communicating a judgment on Q is one intent in a “reply to Q.” So by stipulation, a verbal response where the communicative intent concerns a different question is not a “reply to Q.” (That is so, regardless of why the intent is “off topic,” whether it owes to an innocent misunderstanding of the question, or a desire to be uncooperative, or what have you.17)

17 One objection concerning (R) is that a “reply” should be an utterance intended to communicate a judgment on Q. The intuition is that there is a plan in a reply to communicate a judgment on Q. But to the contrary, a question-and-answer session can unfold too

192  Knowledge of Judging The definition (R) can be unexpected in another respect. Although communicating a judgment on Q is one intent in an act of replying, it need not be an intent of communicating a judgment on Q qua “judgment on Q.” Consider for instance that sincerely asserting ‘Aristotle is Belgian’ communicates a judgment on “What nationality is Aristotle?” But if it is a response to that question, a person could have in mind a different question. She may instead see herself as addressing “Is Aristotle Belgian?” (Perhaps she slightly misheard the query.) In any event, on the present usage it would count as a “reply” to both questions. Thus, ‘reply’ is here a quasi-technical term; it is not meant to capture all pre-theoretic intuitions about “replies.”18 The ambition, then, is to specify the question defining the context pertinent to (DE') and (INF3'). To this end, we must make some distinctions among autoexams. Using Evans’ (1982) example, suppose that a pollster utters the following interrogative: (?) Will there be a WWIII? If posed as an autoexam, there remain several possibilities for what is being asked. The first is: (Q1) What is your standing belief on whether there will be a WWIII?19 If (?) is used to ask (Q1), then the pollster is soliciting a pre-existing opinion one is presumed to have. If that presupposition is correct, then an appropriate reply communicates a judgment based in introspection and/or memory about one’s standing belief on the matter. (Such judgments are of course fallible.)

spontaneously and organically for a speaker to plan out each reply. (Though there can be a generic intent to reply to “whatever questions are asked.”). The essential thing, rather, is that there is a specific telos or communicative intent in “replying,” namely, communicating a judgment on the question. By the way, the main argument below requires only that (i) are (ii) are necessary conditions on a “reply.” But, since ‘reply’ is being used here as a quasitechnical term (see the next page), it avoids obscurity at no real cost to think of (i) and (ii) as sufficient conditions as well. 18 But if desired, we can always narrow our focus to replies uttered with an intent of addressing ‘What nationality is Aristotle?” specifically. (And we could further narrow our attention to appropriate replies to that question.) 19 It is not that (?) can be semantically interpreted as asking (Q1). After all, (?) poses a yes/ no question, and (Q1) is a wh-question. Rather, (?) says roughly what (Q1) puts more precisely. Aside: (Q1) can suggest (?) is really asking a heterological exam—where the facts at issue happen to be the speaker’s own judgment. Generally, one might think that autoexams are elliptical for such heterological exams, but I make no explicit appeal to this idea in what follows.

Infallibility about What One Expresses  193 A different case, however, is where (?) serves as a request for the examinee to consider anew the chances of WWIII, and to execute a judgment reflecting her present perspective. Then, the question is something like: (Q2) What do you judge, forthwith, on whether there will be a WWIII? Here, an appropriate reply does not result from introspecting/recalling a standing belief. Even though the query is autological (the goal is to know the respondent’s judgment of the facts, not the facts per se), an appropriate reply expresses a (re/)occurrent belief on WWIII’s chances. This normally means that one considers, however briefly, the evidence or worldly facts that bear on the matter. We thus might see this as Evans’ version of the autoexam. An appropriate reply illustrates the “transparency” of mind to the world, in the sense made famous by Evans (1982). *8.4.2  Some Complications However, (Q2) as formulated raises a puzzle. On a natural reading, it presupposes that the examinee has only one judgment, re: WWIII’s prospects. Yet we can easily imagine her vacillating while contemplating her evidence: Initially she favors one side, then the other, and back again. In virtue of these opposing judgments, the uniqueness presupposition of (Q2) then fails. But if so, then there is no appropriate reply to the question.20 Now in some cases, that may be the right verdict. Sometimes, the examinee should say “I can’t give a good reply; I’m of two minds.”21 But other times, an appropriate reply remains possible. For her vacillating might be only momentary, and she might eventually settle on the assertion that: (Yes) There will be a WWIII. If (Yes) is asserted sincerely, this may well be appropriate, re: (Q2). However, if her initial vacillation falsifies the uniqueness presupposition, how does an appropriate reply remain possible? Well, I suspect that the real question being asked is a bit different. After all, consider that the only judgment the pollster logs is the one articulated in speech. For all practical purposes, then, that is the only judgment he really cares about. In this light,

20 An appropriate reply requires that the presuppositions of the questions are met. Thus, most cannot appropriately reply to “Have you stopped beating your husband?” It is possible to reply inappropriately, of course. But note that inappropriate replying differs from rejecting the question—where the question is identified as wrong-headed in light of the facts. (Rejecting a query is still a way of responding to the question, but it is not an act of replying.) 21 N.B., she might appropriately respond in that way, but this does not mean she has given the appropriate reply to (Q2). The response is apt not qua reply to (Q2), but qua indication that there is no appropriate reply.

194  Knowledge of Judging it is perhaps best to think of the pollster’s query as: What judgment will you express, forthwith, on whether there will be a WWIII?22 Or in simpler terms: (Q3) What will you expressly judge, forthwith, on whether there will be a WWIII? Here it is not presupposed that only one topically apt judgment exists, yet it is presupposed that only one such judgment is expressed. But while this may help with the initial puzzle, it leads to a related puzzle. Namely, an appropriate reply still seems possible even when that presupposition fails—when the respondent vacillates in speech. Presumably, however, the uniqueness presupposition holds when the subject gives her initial reply. After all, the expressed judgment is momentarily unique; thus, if her replying is sincere, the reply is indeed appropriate at the time. If so, then following-up with a different reply does not change the fact that the initial question had an answer. That question has already been answered. Instead, her second reply could be seen as responding to a reiteration of (Q3). In this, the examinee would likely assume that the inquisitor has not lost interest in the autoexam—thus, in noticing her shifting opinion, she voices it again, as if she re-raised the question to herself on the inquisitor’s behalf. If so, then her initial assertion was an appropriate reply when asserted, assuming it was sincere. It’s just that she is “updating” her reply as she contemplates the matter further. In so updating it, it seems the earlier reply is no longer appropriate. But it would not follow that her earlier reply was never appropriate. We can keep all this clear if we think of the subject as replying not to (Q3), strictly speaking, but rather to: (Q4) What will you expressly judge on whether there will be a WWIII, in response to the present iteration of this question? The query now features a self-referential element. The point of interest, again, is that vacillation in one’s replies does not automatically debunk a presupposition of (Q4). Vacillation may just mean that different iterations of the question are being addressed.23

22 Before I spoke of “sincere asserting,” but “sincere expressing” should be understood (naturally enough) as the expressing of a judgment via a sincere asserting. 23 Some will prefer to say that if the speaker vacillates in speech, there remains only one iteration of the question, and it concerns what the respondent’s settled judgment is. Let me grant that this suggests a fifth way to construe (?). Regardless, as long as (Q4) is a cogent possibility as well, that suffices for our purposes. For we can then basically stipulate that our focus is on possible cases where (?) raises a question with the normal form of (Q4).

Infallibility about What One Expresses  195 Naturally, pollsters and respondents in the real world make no distinctions between the questions that (?) might be used to ask. So prototypically, if (?) is presented as an autoexam, a sincere reply would indiscriminately address any or all these questions. Usually, this is harmless enough. Yet it is worth observing that a reply can simultaneously be appropriate to one precisification of (?) yet not to another. Thus, suppose that (Yes) is asserted as a reply, where the reply unwittingly reflects a sudden change in view. Then, besides expressing her occurrent belief, the speaker may conversationally implicate falsely that (Yes) expresses a long-time belief of hers. This mistaken implicature would mean her reply is not appropriate as a reply to (Q1), even though it would be appropriate as a reply to (Q4). Having said all that, however, I shall henceforth refer to those autoexams unlike (Q1) as “Evansian autoexams.” As this suggests, one may use (Q2) as a rough guide to what type of examination is at issue. But for the record, ‘Evansian autoexam’ should really denote questions like (Q4) only, though often the differences between (Q2)–(Q4) can be glossed.

8.5  The Argument for the Criterion We shall now venture into the argument for (DE'). This will take place in two stages: First, I shall demonstrate a less interesting version of (DE'); then, I shall conjoin this with further principles to reach the more interesting one. 8.5.1  Sincerity as Criterial The less interesting version of (DE') is: (