Skeptical Invariantism Reconsidered 9780367370183, 9781032027463, 9780429353468

Table of contents :
Cover
Half Title
Series Page
Title Page
Copyright Page
Dedication
Table of Contents
Acknowledgements
1 Introduction
PART I: The Source of Skepticism
2 Sceptical Invariantism and the Source of Scepticism
3 Impersonal Epistemic Standards
PART II: Arguments for Infallibilist Skepticism
4 A Cumulative Case Argument for Infallibilism
5 Skeptical Invariantism, Considered
6 Moderate Pragmatic Skepticism, Moorean Invariantism and Attributions of Intellectual Virtue/Vice
PART III: Arguments for Fallibilist Skepticism
7 In Defense of a Moderate Skeptical Invariantism
8 A (Partial) Defense of Moderate Skeptical Invariantism
9 Skepticism, Fallibilism, and Rational Evaluation
10 Situationism, Implicit Bias, and Skepticism
PART IV: Wittgensteinian Anti-Skepticism
11 “I Know”, “I know”, “I know”. Hinge Epistemology, Invariantism, and Skepticism
12 ‘Logical’ and ‘Epistemic’ Uses of ‘to Know’ or ‘Hinges’ as Logical Enabling Conditions
PART V: Assertion and Knowledge Discourse
13 Assertion Compatibilism
14 Knowledge and Loose Talk
15 Knowledge Claims and the Context of Assessment
List of Contributors
Index

Citation preview

This volume represents a fresh wave of renewed reflection on the venerable topic of skeptical invariantism. The contributions to this volume provide a much-needed update to the discussion of this important topic, bringing skeptical invariantism into dialogue with recent developments in contemporary epistemology. – James R. Beebe, University at Buffalo, USA These days, most discussions of skepticism aim to explain away its plausibility or to blunt its effect: the usual message is that, while perhaps tempting, skepticism is false; or that, while skeptical claims might express truths, the implications of this aren’t as significant one might have thought. With contributions from a diverse group of outstanding scholars, this terrific collection breathes new life into the debate about skepticism’s merits. It invites renewed interest in skepticism as a family of substantive epistemological theories aimed at providing the best overall account of a variety of epistemologically significant data. – Patrick Rysiew, University of Victoria, Canada

Skeptical Invariantism Reconsidered

This collection of original essays explores the topic of skeptical invariantism in theory of knowledge. It eschews historical perspectives and focuses on this traditionally underexplored, semantic characterization of skepticism. The book provides a carefully structured, state-of-the-art overview of skeptical invariantism and offers up new questions and avenues for future research. It treats this semantic form of skepticism as a serious position rather than assuming that skepticism is false and attempting to diagnose where arguments for skepticism go wrong. The essays take up a wide range of different philosophical perspectives on three key questions in the debate about skeptical invariantism: (1) whether the standards for knowledge vary, (2) how demanding the standards for knowledge are, and (3) whether the kind of evidence, reasons, methods, processes, etc. that we can bring to bear are sufficient to meet those standards. Skeptical Invariantism Reconsidered will be of interest to scholars and advanced students in epistemology and the philosophy of language. Christos Kyriacou is a Lecturer at the University of Cyprus and received his PhD in Philosophy from the University of Edinburgh. His main interests lie in epistemology, metaethics, and their intersection. Kevin Wallbridge works on issues in epistemology, the philosophy of language, and the philosophy of mind. He has been a Lecturer at the University of Southampton and a postdoctoral research fellow at the University of Cyprus. His PhD in Philosophy is from the University of Edinburgh.

Routledge Studies in Epistemology

Edited by Kevin McCain, University of Alabama at Birmingham, USA and Scott Stapleford, St. Thomas University, Canada

The Ethics of Belief and Beyond Understanding Mental Normativity Edited by Sebastian Schmidt and Gerhard Ernst Ethno-Epistemology New Directions for Global Epistemology Edited by Masaharu Mizumoto, Jonardon Ganeri, and Cliff Goddard The Dispositional Architecture of Epistemic Reasons Hamid Vahid The Epistemology of Group Disagreement Edited by Fernando Broncano-Berrocal and J. Adam Carter The Philosophy of Group Polarization Epistemology, Metaphysics, Psychology Fernando Broncano-Berrocal and J. Adam Carter The Social Epistemology of Legal Trials Edited by Zachary Hoskins and Jon Robson Intellectual Dependability A Virtue Theory of the Epistemic and Educational Ideal T. Ryan Byerly Skeptical Invariantism Reconsidered Edited by Christos Kyriacou and Kevin Wallbridge

For more information about this series, please visit: https://www.routledge. com/Routledge-Studies-in-Epistemology/book-series/RSIE

Skeptical Invariantism Reconsidered

Edited by Christos Kyriacou and Kevin Wallbridge

First published 2021 by Routledge 52 Vanderbilt 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 © 2021 Taylor & Francis The right of Christos Kyriacou and Kevin Wallbridge to be identified as the authors of the editorial material, and of the authors for their individual chapters, has been asserted 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 A catalog record for this title has been requested ISBN: 978-0-367-37018-3 (hbk) ISBN: 978-1-032-02746-3 (pbk) ISBN: 978-0-429-35346-8 (ebk) Typeset in Sabon by codeMantra

Christos: To my parents, with gratitude.

Contents

Acknowledgements 1 Introduction

xi 1

C H R I S T O S K Y R I AC O U A N D K E V I N WA L L B R I D G E

PART I

The Source of Skepticism 2 Sceptical Invariantism and the Source of Scepticism

11 13

DU NCA N PR I TC H A R D

3 Impersonal Epistemic Standards

33

K R I S TA L AW L O R

PART II

Arguments for Infallibilist Skepticism 4 A Cumulative Case Argument for Infallibilism

55 57

N E V I N C L I M E N H AG A

5 Skeptical Invariantism, Considered

80

G R E G O RY S T O U T E N B U RG

6 Moderate Pragmatic Skepticism, Moorean Invariantism and Attributions of Intellectual Virtue/Vice C H R I S T O S K Y R I AC O U

102

x Contents PART III

Arguments for Fallibilist Skepticism

127

7 In Defense of a Moderate Skeptical Invariantism

129

DAV I D E FA S S I O

8 A (Partial) Defense of Moderate Skeptical Invariantism

154

RO B I N M c K E N N A

9 Skepticism, Fallibilism, and Rational Evaluation

172

M ICH A EL H A N NON

10 Situationism, Implicit Bias, and Skepticism

195

K E V I N WA L L B R I D G E

PART IV

Wittgensteinian Anti-Skepticism

211

11 “I Know”, “I know”, “I know”. Hinge Epistemology, Invariantism, and Skepticism

213

A N N A L I S A C O L I VA

12 ‘Logical’ and ‘Epistemic’ Uses of ‘to Know’ or ‘Hinges’ as Logical Enabling Conditions

235

G E N I A S C H Ö N B AU M S F E L D

PART V

Assertion and Knowledge Discourse

253

13 Assertion Compatibilism

255

MONA SI M ION

14 Knowledge and Loose Talk

272

A LEX A N DER DI NGE S

15 Knowledge Claims and the Context of Assessment

298

WAY N E A . DAV I S

List of Contributors Index

319 323

Acknowledgements

We would like to thank Andrew Weckenmann for all the editorial assistance he provided during the process of preparing the volume. We are also enormously grateful to the contributors for the papers they have provided as well as for serving as referees for other papers in the volume. A special mention is owed to Genia Schönbaumsfeld who, despite contracting Covid-19, not only submitted her own paper but also refereed another. Various other people have also reviewed papers and offered constructive comments, including Philippo Ferrari, Martin Montminy, James Bebee, Chris Ranalli, Allan Hazlett, Daniel Hoek, Guido Melchior, Christian Piller, Patrick Rysiew, and Martin Smith. Other referees preferred to remain anonymous. We are grateful to all for their kind assistance.

1

Introduction Christos Kyriacou and Kevin Wallbridge

1 Radical Skepticism Radical skepticism is, roughly, the thesis that we have little knowledge or even no knowledge at all.1 Radical skeptical positions have been at the fringes of mainstream epistemology because the majority of epistemologists consider skepticism to be false (often obviously, or even trivially, so). It is easy to understand why mainstream epistemology takes this view: radical skepticism seems to have disastrous implications for the cognitive and practical dimensions of our everyday life, in a way that is prima facie very counterintuitive. This is because skepticism denies the ordinary epistemic appearance that we have lots of knowledge and seems to imply that ordinary knowledge-based assertion and practical reasoning are impossible (or at least rarely possible).2 Accordingly, most epistemologists develop theories of knowledge that are anti-skeptical and purport to show that skepticism is mistaken (e.g. contextualism, Moorean invariantism and subject-sensitive invariantism).3 More recently, however, there have been some skepticism-friendly voices that have explored the possibility of various skeptical positions (some more moderate, some more radical) by engaging in explanatory and ameliorative work. That is, by exploring the possibility that skepticism (in some form or other) better explains various epistemic desiderata (e.g. epistemic modality, Gettier cases, the dogmatism paradox, the lottery paradox) and empirical data (e.g. biases in reasoning, situationist pressures on reasoning) than rivals and, on the other hand, exploring the possibility that the supposedly devastating problems of skepticism (e.g. for assertion and practical reasoning) can be lessened, or even avoided entirely. These recent skepticism-friendly voices do not constitute a homogeneous group. Some are more radical skeptics, and some are more moderate. Some of the radical skeptics are infallibilists and have understood skepticism in the traditional Cartesian infallibilist sense (e.g. Unger 1975; BonJour 2010; Dodd 2011; Stoutenburg 2017a, this volume; Kyriacou 2017, 2019, 2020, this volume; Climenhaga this volume), while

2  Christos Kyriacou and Kevin Wallbridge other radical skeptics are fallibilist and Pyrrhonist in character and have played out the argument from underdetermination (without any commitment to infallibility) (cf. Fogelin 1994; Rinard forthcoming). Others are more moderate skeptics and in Moorean style deny radical skepticism scenarios (e.g. evil demon, dreaming) and their presumably pernicious implications, but insist that in everyday life we often don’t know what we think we know (cf. Davis 2007). Some have argued for novel moderate forms of skepticism, such as Fassio (2020, this volume), who has argued that knowledge requires practical certainty, and although sometimes things that we might take ourselves to know are not practically certain for us, we can nonetheless be practically certain and know many ordinary propositions. A different form of moderate skepticism draws on recent empirical work in cognitive and social psychology indicating that our reasoning is susceptible to biases, implicit attitudes, vices, faulty heuristics, ideology, self and group interest. On the basis of these empirical results, it is argued that we should be skeptical about our ability to rationally evaluate evidence relating to certain ‘identity-constituting’ topics (politics, religion, morals, sports, history, race, gender, competences, etc.) (Carter and McKenna 2020; Hannon this volume; Wallbridge this volume).4 This volume aims to advance the current state of the debate about both radical and moderate skepticism and call for a reconsideration of skepticism as a family of theories in mainstream epistemology. Skepticism is a perennial theme in philosophy, but with this volume, we hope to rekindle a genuine debate about its plausibility.

2 The Structure and Content of the Volume The volume is divided into four parts grouping together papers on common or closely related themes: Part I Understanding the Source of Skepticism, Part II Infallibilist Skepticism, Part III Fallibilist Skepticism, Part IV Wittgensteinian Anti-Skepticism and Part V Assertion and Knowledge Discourse. Part I Understanding the Source of Skepticism includes papers by Duncan Pritchard and Krista Lawlor that seek, among other things, to offer diagnoses of the true source of skepticism. In ‘Sceptical Invariantism and the Source of Scepticism’, Pritchard argues that the true source of skeptical worries is not the supposedly demanding standard of knowledge, as infallibilist skeptics have claimed, but underdetermination. The underdetermination of the facts by the available evidence suffices to raise the radical skepticism paradox and the spectre of skepticism even if the standard of knowledge is construed in a relatively undemanding way. Thus, Pritchard calls for a reevaluation and correction of our understanding of the true source of skepticism.

Introduction  3 Krista Lawlor in ‘Impersonal Epistemic Standards’ argues that the debate about knowledge attributions has been misconceived in terms of what she calls ‘personalist’ contextualism and invariantism, both of which face difficulties in explaining the cross-contextual usefulness of knowledge attributions. When invariantism is taken too seriously, we are misled into thinking that infallibilist standards apply. We should instead understand that the true standards of knowledge are ‘impersonal’, namely, independent of our interests, stakes, attitudes, etc. Thus, Lawlor calls for a reconsideration of the standard semantic conceptualization of the debate (i.e. contextualism vs invariantism) about the theory of knowledge while offering a diagnosis of the source of the allure of skepticism. Part II Infallibilist Skepticism includes essays by Nevin Climenhaga, Christos Kyriacou and Gregory Stoutenburg that argue for, or explore, infallibilist skeptical positions. In ‘A Cumulative Case Argument for Infallibilism’, Climenhaga argues that infallibilist skeptical invariantism can better explain various epistemic desiderata (such as epistemic modality, epistemic value and Gettier cases) than the main fallibilist alternatives (attributor contextualism, interest-relativity and Moorean invariantism). In ‘Moderate Pragmatic Skepticism, Moorean Invariantism, and Attributions of Intellectual Virtue/Vice’, Kyriacou argues that a moderate version of infallibilism, ‘moderate pragmatic skepticism’, can better explain attributions of intellectual virtues/vices (such as open-mindedness and complacency) than Moorean invariantism. His version of skeptical invariantism is ‘moderate’ because it accepts that there is some infallible knowledge (e.g. modal truths, phenomenal states), and it is ‘pragmatic’ because it suggests that, all other things equal, in everyday life we can speak as if we have lots of ordinary (fallible) knowledge due to the usefulness of this kind of talk. Finally, in ‘Skeptical Invariantism, Reconsidered’, Stoutenburg explores how skeptical invariantism should best understand the appropriateness of everyday knowledge talk if it is largely false. He criticizes Unger’s (1975) entailment theory, Davis’ (2007) loose talk theory, Schaffer’s (2004) exaggeration theory and Everett’s (2006) disguised conditional theory, and offers instead a theory that he calls ‘pragmatic error skepticism’ which claims that ‘S knows that p’ is regularly and appropriately used because it communicates a proposition that is practically relevant (even though the knowledge attribution is false). Part III Fallibilist Skepticism includes essays by Davide Fassio, Robin McKenna, Michael Hannon and Kevin Wallbridge that argue for, or explore, versions of fallibilist skepticism. Fassio’s and McKenna’s papers deal with what Fassio calls ‘moderate practical skeptical invariantism’, while Hannon and Wallbridge explore the potential skeptical impact of

4  Christos Kyriacou and Kevin Wallbridge empirical-psychological work relating to our rational evaluation and inferential practices. In ‘In Defense of Moderate Skeptical Invariantism’, Fassio expands on previous work (cf. Fassio (2020)) developing a version of moderate, practical skeptical invariantism. The view is moderate because it does not lead to some version of radical skepticism (where we have little or no knowledge), but it does deny a considerable part of our ordinary knowledge assertions and ascriptions. According to this view, the threshold of evidential support required for knowledge should be partially fixed by a subject’s practical circumstances: someone is in a position to know a proposition p only if she would be rational to act as if p even if the practical stakes on p were maximally high. In the paper, he considers and addresses what he takes to be the most pressing problem for the view, namely, that it cannot avoid collapsing into a more radical form of skepticism. In ‘A (Partial) Defence of Moderate Skeptical Invariantism’, McKenna argues that a plausible theory of knowledge must account for the fact that we know a lot. By definition, radically skeptical positions do not respect this fact, and this explains why such positions are not popular. But a moderate version of skepticism, McKenna continues, such as Fassio’s (2020, this volume) practical skeptical invariantism promises to save much of our considered ‘core’ knowledge, such as perceptual, memorial and testimonial knowledge, knowledge of necessary and conceptual truths, introspective knowledge as well as scientific knowledge (e.g. about climate change or the reality of covid-19). McKenna explains how practical skeptical invariantism could account for our ‘core’ knowledge, even if we know less than we tend to presume that we know, and suggests that practical skeptical invariantism is a skeptical view we should take seriously (insofar as it respects the fact of having much knowledge). In ‘Skepticism, Fallibilism, and Rational Evaluation’, Hannon argues for a fallibilist kind of skepticism that is inspired by a plethora of empirical results in social and cognitive psychology indicating widespread cognitive shortcomings (biases, implicit attitudes, irrelevant cultural influences, etc.) in the evaluation of evidence and rational inference. He argues that the existence of these cognitive shortcomings casts doubt on the possibility of rational evaluation of evidence, especially about ‘identity-constituting’ matters such as politics, religion, morals, sports, and tradition. What is more, our introspective ability to identify that we are in such a condition is not reliable because we are often subject to an ‘illusion of objectivity’ about our patterns of reasoning. We do not realize from within that we are subject to such cognition-inhibiting shortcomings, something sometimes called ‘the bias bias’.5 Kevin Wallbridge in ‘Situationism, Implicit Bias, and Skepticism’ explores the same empirical-psychological skeptical vein as Hannon. He first argues that a recent (skepticism implying) empirical challenge to

Introduction  5 virtue epistemology suffers from the same problem as standard a priori arguments for radical skepticism: these skeptical arguments work better as a reductio for one of the premises than as a reason to adopt their radically skeptical conclusions. He then proposes an improved skeptical argument based on the involvement of implicit bias in our reasoning (especially when we form beliefs about people’s qualities and competences). He argues that, regardless of one’s particular theory of knowledge, beliefs formed under such conditions cannot amount to knowledge. He then goes on to explore how this model of argument can be expanded to undermine perceptual, philosophical and testimonial knowledge. Part IV Wittgensteinian Anti-Skepticism includes papers by Annalisa Coliva and Genia Schönbaumsfeld that present anti-skeptical strategies based on Wittgenstein’s writings (in particular in On Certainty). In ‘“I Know”, “I know”, “I know”. “Hinge Epistemology, Invariantism, and Skepticism”’, Annalisa Coliva presents Wittgenstein’s radical form of contextualism about knowledge ascriptions and his antiskeptical strategy. She then considers the relationship between her own version of hinge epistemology, contemporary contextualism and skepticism. While she does not follow Wittgenstein in the assertion that skeptical doubts are nonsensical, she argues that skeptical doubts do not pose a threat to our ordinary knowledge. She argues that although we do not have knowledge of hinges (since reasons for them would be circular), hinges are rationally assumed because they are constitutive of epistemic rationality. In ‘“Logical” and “Epistemic” Uses of “to Know” or “Hinges” as “Logical Enabling Conditions”’, Genia Schönbaumsfeld also treads the Wittgensteinian anti-skeptical path. She proposes that although hinge propositions ‘stand fast’, they cannot be known and they are not ‘certain’ in the ordinary sense (since doubt with respect to them is logically excluded). She distinguishes between a ‘logical’ and an epistemic sense of ‘know’. When the expression of uncertainty is senseless, but no further grounds can be given, we are dealing with a purely logical use of ‘to know’. When it is possible to be wrong or uncertain, but reasons for correctness can be given, we are dealing with the ordinary, epistemic sense. She then employs the distinction of logical and epistemic senses of ‘know’ in order to account for three desiderata in the theory of knowledge, namely, the joint avoidance of epistemic contextualism, denial of closure and radical skepticism. Finally, the last Part V Assertion and Knowledge Discourse includes essays by Mona Simion, Alexander Dinges and Wayne Davis that explore the relation between assertion and knowledge discourse. In ‘Assertion Compatibilism’, Simion argues for a function-first account of assertion according to which the sensitivity of assertion to practical stakes is compatible with invariantism, where this sensitivity is sourced in the prudential norms active at the context of discourse. This

6  Christos Kyriacou and Kevin Wallbridge position is a defense of invariantism as such, and can be applied to skeptical invariantism in particular. According to the view she defends, what varies with practical considerations is only the all-things-considered propriety of assertion. She contends that her view is superior to ‘pragmatic compatibilism’, namely, the view that invariantist assertion is rendered compatible with practical stakes merely due to pragmatics, because pragmatic compatibilism cannot account for knowledge-attribution-free assertion. In ‘Knowledge and Loose Talk’, Dinges assesses a prominent strategy among skeptical invariantists, whereby one appeals to ‘loose talk’ in order to account for ordinary knowledge discourse (see, for instance, Kyriacou and Stoutenburg in this volume). Based on recent developments in the theory of loose talk, Dinges argues that such appeals to loose talk fail. He goes on to present a closely related strategy which appeals to conversational exculpature, a phenomenon recently studied by Hoek (2018, 2019). The account combines the virtues of invariantism and versions of contextualism and posits a unique knowledge relation that makes precise predictions as to when knowledge ascriptions ordinarily count as appropriate. Finally, in ‘Knowledge Claims and the Context of Assessment’, Davis grapples with an alternative to skeptical invariantism: John MacFarlane’s assessment-sensitive relativism about knowledge. Davis argues that the assessment-relative semantics has little explanatory value, is subject to some of the same objections to contextualism and does not provide monadic truth conditions even when conjoined with an assessment-relative semantics for truth. He suggests that what is relative to contexts of assessment are conditions of appropriateness and that (like many other terms with strict and invariant meanings) loose uses of ‘know’ are appropriate in some contexts but not others.

Notes 1 For various radically skeptical positions, see Unger (1975), Fogelin (1994), Cappelen (2005), Conee (2005), Klein (2008), Frances (2005), BonJour (2010), Dodd (2011), Dinges (2016), Kyriacou (2017, 2019, 2020, this volume), Stoutenburg (2017a, 2017b, this volume), Rinard (forthcoming) and Climenhaga (this volume). Of course, these positions are far from constituting a homogeneous group, as there are many important differences among the various skeptical positions. For a recent discussion of some of the varieties of skeptical invariantism, see Kyriacou forthcoming-a & forthcoming-b. For a historical discussion of skepticism in philosophy, see Lagerlund (2020). 2 For a recent exposition of this line of thought, see Hannon (2019), and for a recent defense of the compatibility of some version of skepticism with assertion and practical reasoning, see Kyriacou (2020). The papers of Stoutenburg, Dinges and Simion in this volume are also related to the same issue. 3 See Dretske (1970), Nozick (1981), DeRose (1992, 1995), Williamson (2000), Sosa (1999, 2007), Hawthorne (2004), Stanley (2005), Pritchard (2005, 2012) and Greco (2011).

Introduction 7 4 See Ballantyne (2015), Allen and Lynch (forthcoming), Dorst (2020) and Kyriacou forthcoming-a for some optimism about the possibility of ‘debiasing’ and rescuing the possibility of rational evaluation of evidence. See Kyriacou (2021) for a defense of indispensable for epistemic evaluation basic epistemic rationality norms. 5 See Ballantyne (2015) for some discussion of ‘the bias bias’.

References Allen, Teresa and Lynch, Michael. (forthcoming). ‘Can We Be Reasonable? Bias, Skepticism and Public Discourse’. In N. Ballantyne and D. Dunning (eds.), Reason, Bias and Inquiry. Oxford, Oxford University Press. Ballantyne, Nathan. (2015). ‘Debunking Biased Thinkers (Including Ourselves)’. Journal of the American Philosophical Association 1(1):141–162. BonJour, Lawrence. (2010). ‘The Myth of Knowledge’. Philosophical Perspectives 24(1):57–83. Cappelen, Herman. (2005). ‘Pluralistic Skepticism: Advertisement for Speech Act Pluralism’. Philosophical Perspectives 19, Epistemology, 15–39. Carter, Adam and McKenna, Robin. (2020). ‘Skepticism Motivated: On the Skeptical Import of Motivated Reasoning’. Canadian Journal of Philosophy. Online: https://doi.org/10.1017/can.2020.16 Climenhaga, Nevin. (this volume). ‘A Cumulative Case Argument for Infallibilism’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Conee, Earl. (2005). ‘Contextualism Contested’. In E. Sosa and M. Steup (eds.), Contemporary Debates in Epistemology. Blackwell. pp. 47–56. Davis, Wayne. (2007). ‘Knowledge Claims and Context: Loose Use’. Philosophical Studies 137(3):395–438. DeRose, Keith. (1992). ‘Contextualism and Knowledge Attributions’. Philosophy and Phenomenological Research 52(4):913–929. ———. (1995). ‘Solving the Skeptical Problem’. Philosophical Review 104(1):1–52. Dinges, Alexander. (2016). ‘Skeptical Invariantism: Good, but Not Good Enough’. Synthese 193(8):2577–2593. ———. (this volume). ‘Knowledge and Loose Talk’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Dodd, Dylan. (2011). ‘Against Fallibilism’. Australasian Journal of Philosophy 89(4):665–685. Dorst, Kevin. (2020). ‘The Rational Question’. The Oxonian Review. March 14. Link: http://www.oxonianreview.org/wp/the-rational-question/ Dretske, Fred. (1970). ‘Epistemic Operators’. The Journal of Philosophy 67(24):1007–1023. Everett, Theodore. (2006). ‘Antiskeptical Conditionals’. Philosophy and Phenomenological Research 73(2):505–536. Fassio, David. (2020). ‘Moderate Skeptical Invariantism’. Erkenntnis. Online: https://link.springer.com/article/10.1007/s10670-018-0053-1 ———. (this volume). ‘In Defense of a Moderate Skeptical Invariantism’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge.

8  Christos Kyriacou and Kevin Wallbridge Fogelin, Robert. (1994). Pyrrhonian Reflections on Knowledge and Justification. Oxford, Oxford University Press. Frances, Bryan. (2005). Scepticism Comes Alive. Oxford, Oxford University Press. Greco, John. (2011). Achieving Knowledge. Oxford, Oxford University Press. Hannon, Michael. (2019). ‘Skepticism: Impractical, Therefore Implausible’. Philosophical Issues 29(1):143–158. ———. (this volume). ‘Skepticism, Fallibilism and Rational Evaluation’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Hawthorne, John. (2004). Knowledge and Lotteries. Oxford, Oxford University Press. Hoek, Daniel. (2018). ‘Conversational Exculpature’. The Philosophical Review 127(2):151–196. ———. (2019). ‘Loose Talk, Scale Presuppositions and QUD’. Proceedings of the 22nd Amsterdam Colloquium, 171–180. Klein, Peter. (2008). ‘How a Pyrrhonian Skeptic Might Respond to Academic Skepticism’. In E. Sosa et al. (eds.), Epistemology. An Anthology. Blackwell Publishing. pp. 35–50. Kyriacou, Christos. (2017). ‘Bifurcated Skeptical Invariantism: Between Gettier Cases and Saving Epistemic Appearances’. Journal of Philosophical Research 42:27–44. ———. (2018). ‘From Moral Fixed Points to Epistemic Fixed Points’. In C. Kyriacou and R. McKenna (eds.), Metaepistemology. Palgrave Macmillan. pp. 71–96. ———. (2019). ‘Semantic Awareness for Skeptical Pragmatic Invariantism’. Episteme Online First: https://link.springer.com/article/10.1007% 2Fs12136-019-00414-z ———. (2020). ‘Assertion and Practical Reasoning, Fallibilism and Pragmatic Skepticism’. Acta Analytica 1–19. Online First: https://link.springer.com/ article/10.1007%2Fs12136-019-00414-z ———. (2021). ‘Moderate Pragmatic Skepticism, Moorean Invariantism and Attributions of Intellectual Virtue/Vice’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. ———. (2021). ‘Debunking, Theoretical Indispensability and Irreducible Epistemic Rationality’. In Diego Machuca (ed.), Evolutionary Debunking Arguments, Routledge. ———. (forthcoming a) ‘Varieties of Skeptical Invariantism I’. Philosophy Compass. ———. (forthcoming b) ‘Varieties of Skeptical Invariantism II’. Philosophy Compass. Lagerlund, Henrik. (2020). Skepticism in Philosophy. Routledge. Nozick, Robert. (1981). Philosophical Explanations. Harvard University Press. Pritchard, Duncan. (2005). Epistemic Luck. Oxford, Oxford University Press. ———. (2012). ‘Anti-Luck Virtue Epistemology’. The Journal of Philosophy 109(3):247–279. Rinard, Susanna. (forthcoming). ‘Pragmatic Skepticism’. Philosophy and Phenomenological Research. Schaffer, Jonathan. (2004). ‘Skepticism, Contextualism, and Discrimination’. Philosophy and Phenomenological Research 69(1):138–155.

Introduction  9 Stanley, Jason. (2005). Knowledge and Practical Interests. Oxford, Oxford University Press. Stoutenburg, Gregory. (2017a). ‘Strict Moderate Invariantism and Knowledge Denials’. Philosophical Studies 174(8):2029–2044. ———. (2017b). ‘Unger’s Argument from Absolute Terms’. Philosophical Papers 46(3):444–461. ———. (this volume). ‘Skeptical Invariantism, Reconsidered’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Sosa, Ernest. (1999). ‘How to Defeat Opposition to Moore’. Philosophical Perspectives 13:141–153. ———. (2007). A Virtue Epistemology. Oxford, Oxford University Press. Unger, Peter. (1975). Ignorance. Oxford, Oxford University Press. Williamson, Timothy. (2000). Knowledge and Its Limits. Oxford, Oxford University Press.

Part I

The Source of Skepticism

2

Sceptical Invariantism and the Source of Scepticism Duncan Pritchard

1 Introductory Remarks Isn’t it obvious that the problem of radical scepticism essentially trades on an appeal to an austere epistemic standard for knowledge? As we will see, this is certainly a conception of radical scepticism that drives sceptical invariantism, the status of which is our primary concern here. I will be arguing that this way of thinking about radical scepticism is fundamentally mistaken. In particular, I will be claiming that the sceptical problem, in its most compelling form anyway, doesn’t trade on an appeal to an austere epistemic standard for knowledge at all. Indeed, the problem it identifies, if genuine, would apply at even the lowest of epistemic standards for knowledge. It would thus follow that sceptical invariantism should be rejected because it fails to understand the true source of the sceptical problem. This conclusion has wider relevance, which is particularly important given that sceptical invariantism is such a niche view in the literature (which is unsurprising, given that it is itself a broadly sceptical proposal). In particular, I will show how a range of anti-sceptical proposals also effectively treat the problem of radical scepticism as essentially trading on an appeal to an austere epistemic standard for knowledge. It follows that understanding why this is the wrong way to think about radical scepticism has implications not only for sceptical invariantism, but also for the wider debate about radical scepticism. In §2, I describe sceptical invariantism. In §3, I explain how sceptical invariantism presupposes a conception of radical scepticism such that it essentially involves an appeal to an austere epistemic standard. In the process, I describe two anti-sceptical proposals that are naturally associated with sceptical invariantism (low-standards invariantism and contextualism), and show that they share this background picture of what is driving the problem of radical scepticism. In §4, I say more about the nature of the radical sceptical challenge, in particular, the idea that it poses a putative paradox, and how this is relevant to understanding sceptical invariantism. In §5, I argue that radical scepticism, properly understood, does not essentially appeal to an austere epistemic standard for knowledge. In §6, I further argue that this point also undermines the

14  Duncan Pritchard explanatory story that sceptical invariantism tells regarding the practical purpose served by our everyday practices of making (false) knowledge ascriptions. Finally, in §7, I offer some concluding remarks, in particular, regarding where the true source of scepticism lies.

2 Sceptical Invariantism According to sceptical invariantism, ‘knows’ has an invariant meaning across contexts of assessment, such that the truth of a knowledge ascription does not change as one switches between such contexts. This is the invariantist element of the view. Where the position departs from standard invariantist accounts of ‘knows’, however—and here we come to the sceptical element of the view—is in the claim that the epistemic standards for knowledge ascriptions are so high that they can hardly ever, if ever, be truthfully asserted. We thus get the idea that ‘knows’ picks out an invariant epistemic standard that is rarely, if ever, satisfied, such that knowledge ascriptions are rarely, if ever, truthfully asserted.1 Consider, for example, the canonical version of sceptical invariantism found in the work of Peter Unger (1971, 1975). 2 Unger argues that some of our most familiar terms, like ‘empty’ or ‘flat’, are absolute terms, in the sense that they pick out an absolute standard. Nothing is ever really flat, since that would mean that it had no imperfections on its surface at all, and yet there is no such thing in nature as a frictionless plane. Similarly, nothing is ever really empty. A fridge might be empty of food, or shelves, but it always contains something, such as air (as there is no such thing in nature as a vacuum). Unger maintains that ‘knows’ functions in a similar way. More precisely, he holds that knowledge requires certainty and ‘certain’ is an absolute term, which means that ‘knows’ derivatively picks out an absolute epistemic standard. In particular, to know is to be certain, which in turn is to be absolutely certain, such that one, quite rightly, holds that there is no possibility of error with regard to the target belief. But since it is never appropriate to be absolutely certain about anything, as that would demand infallibility (which, some potential rare cares aside, we never possess), so nothing is ever really known. I doubt that it will be as straightforward to this audience, as it seemingly was to Unger, that knowledge demands certainty, which rather creates a lacuna in his argument. Nonetheless, what is important to sceptical invariantism is that ‘knows’ invariantly picks out an absolute standard, such that it demands infallible access to the truth. Arguing that knowledge entails certainty, and that ‘certainty’ is an absolute term, is one route to that conclusion. But for our purposes, we can simply take it that ‘knows’ is an absolute term directly—or at least that it behaves sufficiently like one—and so side-step the contentious issue of the relationship between knowledge and certainty. What is important is rather

Sceptical Invariantism  15 that ‘knows’, being an absolute term, demands infallibility, a demand that is rarely, if ever, satisfied. According to sceptical invariantism, assertions of knowledge ascriptions always express falsehoods. This is what it shares with radical scepticism as it is usually understood.3 The difference, however, is that sceptical invariantism doesn’t leave the matter there, but rather goes on to explain how such a use of ‘knows’ might cohere with our everyday practices. This is the sense in which sceptical invariantism is offering a kind of solution to the sceptical puzzle, albeit a sceptical one. After all, there is clearly a coherent practice of employing terms like ‘empty’ and ‘flat’ in everyday life even on the supposition that they are absolute terms that never actually apply to anything. Accordingly, if ‘knows’ is held to behave in relevantly similar ways, then we ought to be able to similarly make sense of our quotidian practices of knowledge ascriptions, even on the supposition that they systematically express falsehoods. So, for example, one could maintain when we say that ‘the fridge is empty’ or that ‘the table is flat’, we are implicitly setting aside the absolute standard—we mean only that the fridge is empty of (say) food, or that the table is flat enough for practical purposes. In this way one can account for how the everyday assertions using these terms, while strictly speaking false, are nonetheless appropriate in the conversational contexts in which they are uttered, in that they serve a practical purpose that is cut free from the truth of what is asserted. Proponents of sceptical invariantism, like Unger, can make a parallel claim about our everyday assertions of knowledge ascriptions. While they are also strictly speaking false, they can nonetheless be conversationally appropriate, but only because, as with our everyday usage of other absolute terms like ‘flat’ and ‘empty’, we are implicitly setting aside the applicable absolute standard.4 (Whether this proposal is entirely plausible is an issue that we will return to.) Sceptical invariantism is naturally thought of as part of a trio of views about radical scepticism, where the other two nodes in the trio are also solutions to the sceptical puzzle, albeit of an anti-sceptical nature. Sceptical invariantism is a high-standards version of invariantism, and so the immediate contrast is with a low-standards invariantism. According to this proposal, ‘knows’ picks out an invariant epistemic standard, but it is one that is low enough that we are able to satisfy it. Hence, contra the radical sceptic, we have much of the knowledge that we standardly attribute to ourselves. We will consider what such a proposal might look like in a moment. But one key feature of it is that, as with sceptical invariantism, low-standards invariantism maintains that appropriate reflection on our ordinary epistemic practices supports this view, by highlighting how the epistemic standards for knowledge ascriptions are much lower than that employed by the radical sceptic (and low enough to be comfortably satisfied). We can complete our trio of views by considering a third proposal which maintains that ‘knows’ is not invariant at all, but is rather employed in

16  Duncan Pritchard a context-sensitive way such that it picks out both a high epistemic standard and a low epistemic standard in different epistemic contexts. This is attributor contextualism—henceforth just contextualism— here explicitly understood as a response to radical scepticism.5 According to this view, the sceptical problem can be resolved by recognizing that the epistemic standards for knowledge ascriptions are not invariant but rather variant. In particular, they are sensitive to the epistemic standards of the context of the attributor. In everyday attributor contexts where the epistemic standards demanded are relatively low, then such knowledge ascriptions will express truths, contra the radical sceptic. In contrast, in the demanding contexts where scepticism is under consideration, the epistemic standards will be high, and hence the very same knowledge ascriptions will express falsehoods. There is thus a sense in which the radical sceptic speaks truly when she denies that we have knowledge, but this is only a truth that can be expressed relative to her context of attribution. Contextualism thus claims that ‘knows’ picks out a variant epistemic standard across contexts, one that can sometimes be low enough to ensure that knowledge ascriptions express truths, while sceptical invariantism claims that ‘knows’ picks out an invariant epistemic standard across contexts, one that is so demanding that knowledge ascriptions never express truths. There are, of course, lots of details regarding how these three proposals might be unpacked. But I want to focus on what is common to all three proposals, which is the way in which they each treat radical scepticism as being essentially concerned with our putative inability to satisfy an austere epistemic standard. Sceptical invariantism endorses this aspect of scepticism, and so claims that our ascriptions of knowledge express falsehoods. In contrast, while agreeing that radical scepticism is concerned with the question of whether we can satisfy an austere epistemic standard for knowledge, low-standards invariantism and contextualism both resist radical scepticism by disputing that we need to be able to satisfy this standard. In particular, low-standards invariantism maintains that although we can’t satisfy the austere epistemic standard demanded by the radical sceptic, we can nonetheless satisfy a lower epistemic standard, and that it is the latter which is relevant for knowledge ascriptions. Contextualism takes a more nuanced line and concedes that the radical sceptic is right that our knowledge ascriptions express falsehoods in sceptical contexts of evaluation. Nonetheless, it maintains that in everyday contexts of evaluation, such knowledge ascriptions express truths. On all three views, then, it is accepted that radical scepticism essentially concerns our inability to satisfy an austere epistemic standard, such that avoiding the sceptical conclusion that our knowledge ascriptions express falsehoods entails showing that we are not subject to these austere epistemic standards (at least in everyday contexts of evaluation). It is this claim that I will be contesting.6

Sceptical Invariantism  17

3 Radical Scepticism and Epistemic Standards I noted above that it can seem obvious that radical scepticism essentially trades on the idea that we are unable to satisfy an austere epistemic standard for knowledge. Doesn’t Descartes motivate the radical scepticism he ultimately rejects by appealing to certainty? More generally, doesn’t it seem right to suppose that our ordinary epistemic practices, and hence the epistemic standards that are embedded within them, are much less demanding than the epistemic practices that the radical sceptic is urging upon us? The case of Descartes is instructive in this regard. For although the appeal to certainty is no longer foremost in contemporary presentations of radical scepticism, these presentations do follow Descartes in appealing to radical sceptical hypotheses (albeit of a less metaphysically extravagant kind—so brains-in-vats rather than evil demons).7 These are errorpossibilities which call one’s beliefs into question en masse but which are also by their nature such that we cannot, it seems, know that they don’t obtain, since they are indistinguishable from normal veridical experience. But we never consider such error-possibilities in our everyday lives when we attribute knowledge to ourselves and others. Instead, we only consider rather more prosaic error-possibilities, usually of a kind where it is clear what it would take to exclude them. Doesn’t that suggest that there is something about the sceptical appeal to these scenarios that indicates a raising of epistemic standards, such that a more demanding range of error-possibilities is being required for knowledge by the radical sceptic than would be required in an everyday context of epistemic evaluation? To take a familiar example, what does it normally take to know that what one is looking at is a goldfinch? One can imagine a range of error-possibilities that might be thought relevant here, such as that the creature in question is a different kind of bird that one would ordinarily find in one’s garden. This is reflected in what we would standardly take to be a suitable response to the question of how one knows that it is a goldfinch—e.g., that this creature has certain markings (which are distinctive of goldfinches, compared with other birds that might be in the vicinity). Ordinarily, then, to know that what one is looking at is a goldfinch would only demand that one can rule out a very restricted range of error-possibilities. In particular, it would not normally demand that one can rule out the kind of seemingly outlandish error-possibilities that the radical sceptic offers, such as that what one is looking at is not a hologram of a goldfinch. Here is how J. L. Austin famously put the point: Enough is enough: it doesn’t mean everything. Enough means enough to show that (within reason, and for present intents and purposes) it ‘can’t’ be anything else, there is no room for an alternative, competing, description of it. It does not mean, e.g., enough to show it isn’t a stuffed goldfinch. (Austin 1946, 156)

18  Duncan Pritchard So expressed, the idea that radical scepticism functions by raising the epistemic standards can look irresistible. Our ordinary epistemic practices only require that for knowledge we should be able to exclude local possibilities of error. The radical sceptic, in contrast, demands that we should be able to exclude radical sceptical hypotheses. Doesn’t this mean that the radical sceptical challenge works by raising the epistemic standard for knowledge well above our everyday epistemic standard by dramatically extending the scope of error-possibilities that one must exclude in order to know? We can recast our trio of sceptical proposals explicitly in terms of this point about radical sceptical hypotheses. According to sceptical invariantism, ‘knows’ picks out an austere epistemic standard that demands that one is able to know the falsity of radical sceptical hypotheses. On this view, knowledge demands infallibility, and if we cannot exclude these error-possibilities then there is still the possibility that our beliefs are false. Accordingly, since we cannot know the denials of radical sceptical hypotheses, it follows that our knowledge ascriptions express falsehoods. Interestingly, sceptical invariantism, while agreeing that the epistemic standard for ‘knows’ is austere, rejects the narrative we just offered that this is a more demanding epistemic standard than that required by our ordinary epistemic practices. We will return to this point. Low-standards invariantism, in contrast, argues that our everyday epistemic practices in fact reveal that ‘knows’ picks out a low epistemic standard that we can satisfy. In particular, as we saw Austin arguing a moment ago, our everyday conception of knowledge only demands that one is able to exclude mundane possibilities of error, and not radical sceptical hypotheses, which is why our knowledge ascriptions ordinarily express truths. We thus seemingly have a basis on which to resist the radical sceptical challenge by insisting on the adequacy of our everyday epistemic practices and the non-austere epistemic standard for knowledge that they involve. Contextualism occupies an intermediate position between these two proposals. In sceptical contexts of evaluation, ‘knows’ picks out an austere epistemic standard whereby one needs to be able to exclude radical sceptical hypotheses, which is why knowledge ascriptions express falsehoods relative to this context. In everyday contexts of evaluation, however, ‘knows’ picks out a low epistemic standard whereby one doesn’t need to be able to exclude radical sceptical hypotheses, but only mundane error-possibilities (which one can normally exclude), and hence knowledge ascriptions express truths relative to this context. Contextualism can thus capture a sense in which radical scepticism expresses a truth, in that there is a context of epistemic evaluation where the austere epistemic standard that it insists upon applies. But radical scepticism is nonetheless stripped of its teeth, since relative to normal contexts of

Sceptical Invariantism  19 epistemic evaluation a much lower epistemic standard applies, one that we can satisfy. Framing the problem of radical scepticism in terms of the putative need to eliminate radical sceptical hypotheses thus provides us with a way of understanding how this problem trades on demanding an austere and unsatisfiable, epistemic standard. In particular, we can now make sense of how our trio of views each embody this way of thinking about radical scepticism in terms of how they respond to the sceptical challenge.

4 The Radical Sceptical Paradox With the foregoing in mind, let’s examine the problem of radical scepticism a little more closely. In its strongest form, radical scepticism purports to be a paradox, where this means that it is supposed to be identifying deep tensions within our own epistemological concepts. This means that we can formulate the problem in terms of claims that we would naturally endorse when taken in isolation, but which can be shown to be collectively inconsistent. In contrast to radical scepticism qua position, where the radical sceptic is committed to denying that we possess the everyday knowledge that we ascribe to ourselves, radical scepticism qua paradox doesn’t take any stance at all about which of the claims in the inconsistent set should go. We can bring this point into sharper relief by distinguishing between the radical sceptical paradox, as sceptical invariantism conceives of it, and the particular proposal regarding how we should respond to that alleged paradox that sceptical invariantism offers. In terms of the former, the paradox essentially consists of the following two incompatible claims: The Infallibilism-Based Radical Sceptical Paradox (I-I) Knowledge demands an absolute epistemic standard that we cannot satisfy. (I-II) One has widespread everyday knowledge. Our ordinary epistemic practices seem to license (I-II), and yet sceptical invariantism maintains that those same practices, properly understood, also license (I-I). So both claims look pre-theoretically compelling when taken individually, and yet they clearly cannot both be true. We thus have a putative paradox on our hands. Sceptical invariantism as a position offers a sceptical solution to this paradox. This is to argue that we should embrace (I-I) and thereby reject (I-II). What makes this a solution to the paradox, rather than the mere endorsement of radical scepticism as a position, is that it offers us an explanation of how (I-II) could be false. Indeed, this is an account not only

20  Duncan Pritchard of why it is false, but also of why we were mistaken in thinking that this claim is rooted in our everyday epistemic practices in the first place. We only believe that (I-II) is true because we are misled by our systematic practices of making knowledge ascriptions into thinking that we must be committed to the widespread truth of these ascriptions, but the sceptical invariantist has a story to tell about what is driving these knowledge ascriptions that has nothing to do with their truth. Sceptical invariantism is thus offering us an undercutting response to the putative radical sceptical paradox, albeit of a sceptical rather than an anti-sceptical kind. Undercutting responses to alleged paradoxes aim to show that the paradox is illusory (as opposed to overriding responses, which grant that the paradox is genuine, but maintain nonetheless that at least one of the claims that makes up the paradox should be rejected).8 In the radical sceptical case, this will involve arguing that it doesn’t arise out of our own epistemological concepts at all, but rather appeals to contentious theoretical claims that we should reject. In the case of sceptical invariantism, the claim is that we have the wrong theoretical account of our practice of making knowledge ascriptions, one that is not supported by a closer reflection on what those practices actually involve. This undercutting element of the proposal is crucial to its plausibility, given the overwhelming implausibility of denying (I-II); without it, the proposal is stripped of much of what makes it interesting. I want to flag one aspect of this approach that I think is particularly interesting. It has previously been noted in the debate about radical scepticism that it is not enough to resist such scepticism that one can show that our actual epistemic practices seem to license a claim like (I-II), such that one can straightforwardly respond to the puzzle by denying whichever claims in the puzzle conflict with (I-II), in this case (I-I). Consider the remarks that we saw Austin making above, whereby he argues that our everyday epistemic practices treat knowledge as a fallible notion, the possession of which is compatible with the knowing individual being unable to eliminate certain error-possibilities, including radical sceptical hypotheses. Austin clearly thinks that observations of this kind have a decisive anti-sceptical import for the debate about radical scepticism, such that his case for (I-II) should suffice to give us grounds to reject a claim like (I-I) and hence adopt a low-standards invariantism. But others are not so sure, and this is because it is entirely compatible with radical scepticism being a genuine paradox that Austin’s descriptions of our actual epistemic practices are accurate. The reason for this is that radical scepticism can be rooted in our actual epistemic practices even if it is not immediately manifest in them. In particular, it is natural to think that what the sceptic is doing is taking our ordinary epistemic practices and applying them in a purified fashion, one that is not tainted, as are our everyday epistemic practices, by practical limitations of time, imagination, thoroughness,

Sceptical Invariantism  21 and so on. Accordingly, one could argue that Austin is right that our actual epistemic practices reveal a fallibilist conception of knowledge and yet that nonetheless there is a genuine radical sceptical paradox to be contended with. The thought is that when we employ our everyday epistemic practices in a thoroughgoing way, unaffected by supposedly irrelevant practical constraints, then we are also led to treating knowledge along infallibilist lines, or at least as demanding that in order to know one should be able to rule out radical sceptical hypotheses that one is aware of. Hence it is far too quick to suppose that because our everyday practices firmly support a claim like (I-II) that they don’t also provide support for (I-I) as well.9 The line taken by sceptical invariantism in this regard is much more radical. For what it is claiming is not that the sceptical epistemic practices of treating knowledge as infallible are purified versions of our everyday epistemic practices that are merely implicit in our actual practices. Instead, it is arguing that this is directly manifested in our everyday epistemic practices. That is, properly understood, there is no everyday epistemic practice of widespread knowledge ascription that needs to be explained away, or which can be in tension with sceptical claims, as our ordinary knowledge ascriptions are simply false. The point of this practice, according to sceptical invariantism, is not to convey truths at all, but rather to serve a practical purpose that is divorced from the truth of the assertions in play.10 The supposed support for (I-II) from our everyday epistemic practices is thus illusory. Sceptical invariantism is thus distinctive both in terms of the conception of the putative radical sceptical paradox that it is engaging with, and in terms of the particular way that it responds to this paradox. On the former front, it treats the putative paradox as arising out of a tension between a supposed everyday commitment to widespread knowledge in conjunction with the fact that knowledge demands an unsatisfiable, infallible, epistemic basis. An appeal to austere epistemic standards is thus at the heart of the sceptical challenge. On the latter front, sceptical invariantism takes the radical line of claiming that our everyday epistemic practices, properly understood, do not imply any commitment to our possessing widespread knowledge. The paradox is thus undercut, albeit in a fashion whereby it is the sceptical horn of the dilemma, as opposed to the anti-sceptical horn, that is embraced.

5 Motivating Radical Scepticism Notice that the formulation of the radical sceptical paradox just offered to capture sceptical invariantism doesn’t make any essential appeal to radical sceptical hypotheses. Given that we already have the idea that knowledge needs to satisfy an absolute, and thus infallible, epistemic

22  Duncan Pritchard standard in play, there is really no need. For while it is true that an infallible conception of knowledge would demand the elimination of radical sceptical error-possibilities—since it would demand the elimination of all error-possibilities—one doesn’t need to appeal to this particular sub-set of error-possibilities in order to motivate the thought that we are unable to satisfy an absolute epistemic standard for knowledge. Interestingly, however, contemporary formulations of radical scepticism do appeal to radical sceptical hypotheses. I think it is important to understand why. The contemporary treatment of radical scepticism understands it as essentially a clash between, on the one hand, the idea that knowledge possession is widespread, and, on the other hand, our inability to know the denials of radical sceptical hypotheses. In particular, there is no appeal to an infallibilist conception of knowledge in play in the claim that we are unable to know the denials of radical sceptical hypotheses. It’s not that we can’t know these propositions because we fail to meet an austere infallibilist epistemic standard, but rather that we do not meet any epistemic standard at all, even a low one, as we simply have no rational basis for excluding them, given that they are, ex hypothesi, indistinguishable from ordinary veridical experience. This has an important consequence, which is that one needs to bring in a further thesis, supposedly rooted like the others in our ordinary epistemic practices, in order to explain why these two claims are in tension with one another. In particular, independently of an infallibilist conception of knowledge, why can’t it be the case that one knows many of the things that one takes oneself to know—about goldfinches and such like—even while being unable to know the denials of radical sceptical hypotheses? In the contemporary literature this theoretical lacuna is filled by appealing to the following ‘closure’ principle which maintains that knowledge is closed under competent deduction. The Closure Principle If S knows that p, and S competently deduces from p that q, thereby forming a belief that q on this basis while retaining her knowledge that p, then S knows that q. One can see the appeal of the principle. Competent deduction is a paradigm form of reasoning. Accordingly, if the input to that competent deduction is knowledge, and the output is a belief based on that competent deduction, then how can one fail to know the proposition that one believes on this basis? The principle thus looks entirely innocuous, and certainly in keeping with our ordinary epistemic practices.11 With this principle in play one can straightforwardly explain why one cannot consistently endorse the idea that knowledge is widespread even though one is unable to know the denials of radical sceptical hypotheses.

Sceptical Invariantism  23 This is because most of what one takes oneself to know is obviously inconsistent with some radical sceptical hypothesis or other. Accordingly, if one’s everyday knowledge is bona fide then one must have a way of coming to know, via closure, the denials of radical sceptical hypotheses. Conversely, given that one is independently aware that knowledge of the denials of radical sceptical hypotheses is impossible, it follows that one must lack the everyday knowledge that one typically ascribes to oneself and others. Indeed, this becomes very clear once one reflects on what it would mean to arrive at knowledge of the denials of sceptical hypotheses by simply deducing this from the everyday ‘knowledge’ that one thinks one has. One’s rational basis for believing that, for example, one is presently sitting at home in one’s living room doesn’t seem to be any rational basis at all for believing that one is not presently a brain-in-a-vat artificially stimulated to believe that one is sitting at home in one’s living room. We thus get the following formulation of the putative paradox posed by closure-based radical scepticism: The Closure-Based Radical Sceptical Paradox (C-I) One is unable to know the denials of radical sceptical hypotheses. (C-II) The closure principle. (C-III) One has widespread everyday knowledge. Each of these claims seems independently plausible and rooted in our everyday epistemic practices. And yet they appear to be an inconsistent triad. We thus have a plausible rendering of a paradox. Notice, however, that there is no mention of the idea that knowledge demands infallibility here. All that matters is that knowledge requires the exclusion of the denials of radical sceptical hypotheses. Moreover, and in contrast to sceptical invariantism, this demand is not arising out of the nature of knowledge itself, but rather from our everyday concept of knowledge coupled with the closure principle. We now have two formulations of the radical sceptical paradox, one that is cast in terms of knowledge requiring infallibility, and a second in terms of radical sceptical hypotheses and the closure principle. Despite the simplicity of the former formulation, I think it is relatively clear that the latter formulation is the more compelling. For there is now no need to appeal to an infallible conception of knowledge, much less to the idea that that such a conception is rooted in our everyday epistemic practices. Instead, by appealing to closure, one needs only an ordinary fallibilist conception of knowledge to extract the necessary sceptical tension, at least once radical sceptical hypotheses are in play. There is a further feature of the closure-based formulation of the radical sceptical paradox that is particularly important for our purposes. This is that it highlights why the radical sceptical problem isn’t really about epistemic standards at all. In particular, one can lower the

24  Duncan Pritchard epistemic standards as much as one likes, and it seems that the puzzle will still be with us. In order to see this, remember why it is that we cannot use closure to come to know the denials of radical sceptical hypotheses on the basis of a competent deduction from our putative everyday knowledge. The reason is that one’s rational basis for the everyday claim doesn’t seem to have any bearing at all on the radical sceptical scenario; indeed, it only seems to be a rational basis for that claim because one is already disregarding the possibility that one is radically in error. That’s why the fact that it seems to one as if one is presently sitting in one’s living room is no rational basis at all for excluding the radical sceptical hypothesis that one is a brain-in-a-vat. With this point in mind, it should be clear that lowering the epistemic standards doesn’t make the problem posed by closure-based radical scepticism any more tractable. According to either low-standards invariantism or contextualism, one ought to count as having knowledge (i.e., the relevant knowledge ascriptions will express truths) relative to everyday contexts of epistemic evaluation. For that to be the case, however, one’s belief needs to meet at least some significant epistemic standard. But what would that be, given the closure-based sceptical argument? For notice that the sceptic isn’t arguing that one’s failure to exclude radical sceptical hypotheses means that while one’s belief fails to meet the standards for knowledge it nonetheless enjoys some lesser epistemic pedigree. The point is rather that it has no epistemic pedigree at all. It’s not that one has some reason to think that one is presently seated in one’s living room, and hence some reason to think that one is not a deceived brain-in-a-vat, but that the reason in either case doesn’t suffice for knowledge. Rather, since one has no reason for excluding the brain-in-a-vat scenario, so one has no reason at all for thinking that one is presently seated in one’s living room. Lower the epistemic standards as much as one likes, one still lacks knowledge, which just goes to show that the issue is not really about epistemic standards at all.12 We can further bring out the point that radical scepticism does not essentially trade on an appeal to austere epistemic standards for knowledge by considering a further formulation of the paradox that is widely held to be broadly equivalent to the closure-based formulation.13 This version of the radical sceptical paradox employs not the closure principle but rather this underdetermination principle: The Underdetermination Principle If S knows that p and q describe incompatible scenarios, and yet S lacks a rational basis that favours p over q, then S lacks knowledge that p. This principle seems compelling. What would it mean for one to know that one’s car is parked in the garage if one knows that this entails that

Sceptical Invariantism  25 it isn’t parked on the drive, and one has no rational basis that favours the first scenario over the second? Note, in particular, that the underdetermination principle is in fact very undemanding. One needs to have a better reason for thinking that one’s car is in the garage rather than on the drive, but it needn’t be decisive. So, for example, that your son told you earlier that it was in the garage should suffice, ceteris paribus, even if his testimony about these matters is not especially reliable. The crux is that if one really takes oneself to have no more reason for thinking that the car is in the garage than for thinking that it is parked on the drive, then one doesn’t know that it is in the garage. What could be more uncontroversial? The problem emerges, however, once we bring in radical sceptical hypotheses. After all, it seems that we have no rational basis that favours our everyday beliefs over radical sceptical scenarios, even though we recognize the general incompatibility between the two. I don’t have a rational basis that favours my belief that I’m sitting in my living room over the known to be incompatible sceptical scenario that I’m a brain-in-a-vat. We can thus employ the underdetermination principle to problematize our everyday knowledge. Here is the formulation of the paradox that we end up with: Underdetermination-Based Radical Scepticism (U-I) One cannot have rational support that favours one’s belief in an everyday proposition over an incompatible radical sceptical hypothesis. (U-II) The underdetermination principle. (U-III) One has widespread everyday knowledge. As before, we have, it seems, an inconsistent triad, where each of these claims appears to arise out of our everyday epistemic practices. As with closure-based radical scepticism, the question of epistemic standards doesn’t seem to arise here at all. There is no appeal to an infallibilist thesis about knowledge, for example. Moreover, just like earlier, it’s also true that lowering the epistemic standards won’t by itself offer a way out of this puzzle. The sceptical contention, after all, is not that one has some rational support for one’s everyday beliefs, but that it doesn’t rise to the level of knowledge. The claim is rather that one has no good reason at all for one’s everyday beliefs. Think about the reasons that one can offer for thinking that one is presently sitting in one’s living room. If they don’t favour this scenario over the known to be incompatible brainin-a-vat scenario, then in what sense does one have any reason at all for this belief? Accordingly, lowering the epistemic standard for knowledge doesn’t make the underdetermination-based sceptical problem anymore tractable, and that’s because this problem doesn’t essentially trade on an appeal to an austere epistemic standard for knowledge in the first place. This point becomes especially obvious when we consider the most dramatic way of responding to underdetermination-based scepticism, as

26  Duncan Pritchard presented by epistemological disjunctivism. According to this proposal, we should reject (U-I) on the grounds that the rational support our perceptual beliefs enjoy in paradigm cases is factive, in that it entails the target proposition. It follows that (U-I) must be rejected, since in such cases one not only has a rational basis for one’s belief that favours it over radical sceptical alternatives, but one which actually entails that the latter are false.14 Crucially, however, epistemological disjunctivism is allied to a fallibilist conception of knowledge. Just as knowledge can be both fallible (i.e., acquired via a process that could have resulted in a false belief) and factive, so the rational support that one’s belief enjoys can be both fallible and factive. The point is that one doesn’t need to argue that knowledge can satisfy an infallibilist epistemic standard in order to block underdetermination-based radical scepticism. Indeed, although epistemological disjunctivism opts for factive rational support in response to this problem, one could in fact motivate the rejection of (U-I) with something much weaker, since all that is really required is favouring epistemic support. We are thus reminded that what is lacking, according to the radical sceptic, is not epistemic support that satisfies an austere epistemic standard, but rather epistemic support of a most minimal kind.

6 Sceptical Invariantism Unravels This brings me to my final point about sceptical invariantism, which concerns the story it tells about our everyday knowledge ascriptions. Recall that although these express falsehoods, they are also meant to serve a practical purpose, just as (on this view) ascriptions of emptiness and flatness can serve a practical purpose even while being false. As we noted above, the plausibility of this story is vital to the undercutting credentials of the proposal as a response to the radical sceptical paradox. Thus far we have accepted this story at face value, but it is now time to examine it more closely. In particular, notice that the explanation of why our everyday ascriptions of emptiness and flatness can be practically useful while strictly speaking expressing falsehoods is that they approximate to the absolute standard. So while nothing on this view is really empty, we can make sense of how a fridge can be said to be ‘empty’ of something specific like food. That is, the fridge, while not empty of anything, is empty of a particular thing, food. The everyday usage of the term thus captures a qualified sense of emptiness that can serve a particular practical purpose (e.g., in terms of knowing when one needs to do some shopping). Similarly, while nothing on this view is really flat, we can nonetheless make sense of how a country like Holland can be said to be ‘flat’. That is, while Holland is not absolutely flat (since nothing is), it is flat for a country (e.g., when compared with Switzerland). As before, the everyday usage of the term thus captures a qualified sense of flatness

Sceptical Invariantism  27 that can serve a particular practical purpose (e.g., in terms of knowing which country would be the best choice for a cycling holiday). There is nothing especially mysterious about this, even granted the putatively absolute nature of the terms involved. As philosophers of science have noted, the ideal gas law is never instantiated in nature, but this doesn’t prevent it from being of tremendous practical use in relevant scientific contexts.15 We can thus make sense of a term satisfying a non-absolute standard in a way that is a suitable approximation of an absolute standard. In order to account for the practical utility of these everyday assertions of absolute terms like ‘flat’ and ‘empty’, however, it is important that the everyday assertions do approximate in relevant ways to the relevant absolute standard. The fridge really needs to be empty of foodstuffs, and the country really needs to be flat for a country. If nothing even approximates to the standards in question, however, then the practical utility disappears. If emptiness or flatness never applies to anything, even in a qualified sense, then what practical point would it serve to describe anything as ‘flat’ or ‘empty’? With this in mind, let’s consider the kind of claim that sceptical invariantism makes about our use of knowledge ascriptions in everyday contexts. The idea is that, just as with our everyday use of ‘empty’ and ‘flat’, these ascriptions, while false because we cannot satisfy the absolute epistemic standard, are practically useful because we are able to approximate to it. In the epistemic context this can only mean that we satisfy some fallible standard for knowledge, such that we have an epistemic basis for our beliefs which, although insufficient to rule out radical sceptical hypotheses, is nonetheless sufficient to rule out local errorpossibilities. Given what we have argued in the previous section about how radical scepticism does not essentially trade on an appeal to an austere epistemic standard, however, this claim is evidently dubious. In particular, the sceptical invariantist is clearly presupposing that our beliefs satisfy a weak epistemic standard, one that suffices for the practical utility of our knowledge ascriptions even though it cannot make them true. But we have seen that the radical sceptical paradox, properly understood, calls even that weaker epistemic standing into question. Since sceptical invariantism doesn’t offer any basis for rejecting the claims that make up this paradox (beyond insisting that we do lack the widespread everyday knowledge in question), it follows that it simply cannot help itself to the story that it tells about the propriety of our everyday use of false knowledge ascriptions. This severely undermines the plausibility of sceptical invariantism. We noted earlier that a core part of the appeal of the proposal lies in the undercutting response it offers to the radical sceptical paradox. The undercutting credentials of the view in turn rest on the account it offers of what our everyday practices of knowledge ascriptions are doing if they

28  Duncan Pritchard aren’t in the business of expressing truths. That account has shown to be of dubious pedigree. But without this undercutting story then sceptical invariantism no longer has a way of explaining, consistent with our everyday epistemic practices, how the claim that we have widespread knowledge could be false. When we couple this fact with the previous point that sceptical invariantism also misdiagnoses the source of radical scepticism—by mistakenly thinking that it essentially trades on an appeal to an austere epistemic standard—then it is hard to see what is left to recommend the view.

7 Concluding Remarks We have argued that sceptical invariantism fails because it misunderstands the nature of the radical sceptical challenge. In particular, this challenge does not rest in any essential way on an appeal to an austere epistemic standard for knowledge as sceptical invariantism presupposes. As we have seen, this point undermines not only the credibility of sceptical invariantism as a response to the radical sceptical paradox, but also the supporting story that sceptical invariantism offers concerning the propriety of our false everyday knowledge ascriptions. If the problem of radical scepticism does not essentially turn on an appeal to an austere epistemic standard for knowledge, then what is motivating it? I think we have witnessed enough to offer the essentials of an answer to this question. For the underlying issue is whether, even in the very best epistemic conditions, the rational standing that our beliefs enjoy is compatible with those beliefs being, nonetheless, mostly false. Elsewhere I have referred to this claim as the insularity of reasons thesis.16 With this claim in play, radical scepticism immediately gets a grip, since it follows that one can always generate scenarios where one’s beliefs are mostly false but where they nonetheless enjoy the very same level of rational standing. On this way of thinking about radical scepticism, radical sceptical hypotheses are not a way of raising the epistemic standards, but rather simply devices that make explicit what seems to be implicit in our epistemic practices—viz., our commitment to the insularity of reasons thesis.17 If the foregoing is right, then while our critical focus has been sceptical invariantism, the critique that we have offered will be applicable, mutatis mutandis, to any treatment of radical scepticism which similarly treats the sceptical problem as essentially trading on an appeal to an austere epistemic standard for knowledge. In particular, both the low-standards invariantism and the contextualism that we noted above will be subject to this critique, given that they aim to resolve the sceptical problem by, in essence, making the case for a low epistemic standard for knowledge that we can satisfy (at least within a particular context of epistemic evaluation). Our critique of sceptical invariantism thus has a wider application to the broader contemporary debate about radical scepticism.18

Sceptical Invariantism  29

Notes 1 Or, at least, very few are ever truthfully asserted. See BonJour (2010) for a version of sceptical invariantism that allows for there to be knowledge of a restricted class of propositions, of a kind where we are held to be able to meet the austere epistemic standard in play (e.g., propositions regarding the apparent content of one’s immediate experiences). Since it doesn’t really matter to the radical nature of this sceptical proposal whether all or only most knowledge ascriptions are never truthfully asserted, I will set this complication to one side in what follows and express the view in its most straightforward, and thus strongest, form. 2 See also Unger (1984), which argues for an agnostic stance between sceptical invariantism and a version of contextualism (broadly of the kind that we are about to consider). 3 At least where radical scepticism is understood as a position. The importance of this point will become apparent below. 4 For some key recent discussions of this proposal, in addition to the original presentation by Unger (1971, 1975, 1984), see Hawthorne (2004, ch. 3), Dinges (2016), Kyriacou (2017, 2019; 2020), and Fassio (2018). See also BonJour (2010). 5 For some of the key defences of (attributer) contextualism which are explicitly advanced to deal with the problem of radical scepticism, see DeRose (1995), Lewis (1996), and Cohen (2000). Interestingly, in a later work, Unger (1984) explores a similar proposal, as a competitor to the sceptical invariantism that he previously developed. 6 As such, my critique of these three views would also apply to any other proposal with regard to radical scepticism that effectively understands the sceptical problem in terms of the putative difficulty of satisfying an austere epistemic standard. For example, it would apply to the subject-sensitive invariantism defended by Fantl and McGrath (2002, 2007, 2009), Hawthorne (2004), and Stanley (2005). It would also apply to non-standard versions of attributor contextualism, such as the contrastivism defended by Schaffer (2004, 2005), the evidential contextualism defended by Neta (2002, 2003), and the inferential contextualism defended by Williams (1991). I offer specific critiques of all these positions elsewhere—see especially Pritchard (2015, passim), but also Pritchard (2008a, 2018a, 2018c). 7 Interestingly, as noted above, Unger (1971, 1975), the foremost exponent of sceptical invariantism, is unusual in that he does treat the sceptical problem as involving a demand for certainty. As we also noted above, however, it doesn’t seem to be essential to sceptical invariantism that it makes this claim (and it’s arguably a much stronger position without it). Besides, Unger also appeals to radical sceptical hypotheses and the fact that we cannot exclude them as a reason why we can’t be absolutely certain, so such error-possibilities are still essentially in play even on his presentation of sceptical invariantism. 8 Overriding responses to paradoxes are thus essentially revisionary. As Schiffer (1996, 330) memorably puts the point, if radical scepticism poses a genuine paradox then there is no ‘happy face’ solution to this problem, only a ‘sad face’ solution, because it would mean that there is indeed a ‘deepseated incoherence’ within our pre-theoretical epistemological commitments. For further discussion of undercutting and overriding responses to radical scepticism, and their significance, see Pritchard (2014, 2015, part one). In this regard, see also the therapeutic/theoretical distinction presented by Williams (1991, ch. 1), and the obstacle-removing/obstacle-overcoming distinction described by Cassam (2007).

30  Duncan Pritchard







Sceptical Invariantism  31 ——— (2005a). ‘The Case Against Closure’, Contemporary Debates in Epistemology, (eds.) E. Sosa & M. Steup, 13–26, Oxford: Blackwell. ——— (2005b). ‘Reply to Hawthorne’, Contemporary Debates in Epistemology, (eds.) E. Sosa & M. Steup, 43–46, Oxford: Blackwell. Dinges, A. (2016). ‘Skeptical Pragmatic Invariantism: Good, but not Good Enough’, Synthese 193, 2577–2593. Elgin, C. (2017). True Enough, Cambridge, MA: MIT Press. Fantl, J., & McGrath, M. (2002). ‘Evidence, Pragmatics, and Justification’, Philosophical Review 111, 67–94. ——— (2007). ‘On Pragmatic Encroachment in Epistemology’, Philosophy and Phenomenological Research 75, 558–589. ——— (2009). Knowledge in an Uncertain World, New York: Oxford University Press. Fassio, D. (2018). ‘Moderate Skeptical Invariantism’, Erkenntnis [Online First, DOI: https://doi.org/10.1007/s10670-018-0053-1]. Hawthorne, J. (2004). Knowledge and Lotteries, Oxford: Oxford University Press. ——— (2005). ‘The Case for Closure’, Contemporary Debates in Epistemology, (eds.) E. Sosa & M. Steup, 26–43, Oxford: Blackwell. Kyriacou, C. (2017). ‘Bifurcated Sceptical Invariantism: Between Gettier Cases and Saving Epistemic Appearances’, Journal of Philosophical Research 42, 27–44. ——— (2019). ‘Semantic Awareness for Skeptical Pragmatic Invariantism’, Episteme [Online First, DOI: https://doi.org/10.1017/epi.2019.7]. ——— (2020). ‘Assertion and Practical Reasoning, Fallibilism and Pragmatic Skepticism’, Acta Analytica 35, 543–561. Lewis, D. (1996). ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549–567. McDowell, J. (1995). ‘Knowledge and the Internal’, Philosophy and Phenomenological Research 55, 877–893. Neta, R. (2002). ‘S Knows that P’, Noûs 36, 663–689. ——— (2003). ‘Contextualism and the Problem of the External World’, Philosophy and Phenomenological Research 66, 1–31. Nozick, R. (1981). Philosophical Explanations, Oxford: Oxford University Press. Pritchard, D. H. (2001). ‘Contextualism, Scepticism, and the Problem of Epistemic Descent’, Dialectica 55, 327–49. ——— (2005a). ‘Contextualism, Scepticism and Warranted Assertibility Manœuvres’, Knowledge and Skepticism, (eds.) J. Keim-Campbell, M. O’Rourke & H. Silverstein, 85–104, Cambridge, MA: MIT Press. ——— (2005b). Epistemic Luck, Oxford: Oxford University Press. ——— (2005c). ‘The Structure of Sceptical Arguments’, Philosophical Quarterly 55, 37–52. ——— (2008a). ‘Contrastivism, Evidence, and Scepticism’, Social Epistemology 22, 305–323. ——— (2008b). ‘McDowellian Neo-Mooreanism’, Disjunctivism: Perception, Action, Knowledge, (eds.) A. Haddock & F. Macpherson, 283–310, Oxford: Oxford University Press. ——— (2012). Epistemological Disjunctivism, Oxford: Oxford University Press.

32  Duncan Pritchard ——— (2014). ‘Sceptical Intuitions’, Intuitions, (eds.) D. Rowbottom & T. Booth, 213–231, Oxford: Oxford University Press. ——— (2015). Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing, Princeton, NJ, Princeton University Press. ——— (2018a). ‘Contextualism and Radical Scepticism’, Synthese 195, 4733–4750. ——— (2018b). ‘Epistemic Angst’, Philosophy and Phenomenological Research 96, 70–90. ——— (2018c). ‘Unnatural Doubts’, Skepticism: Historical and Contemporary Inquiries, (eds.) G.A. Bruno & A. Rutherford, 223–247, London: Routledge. Rysiew, P. (2011). ‘Contextualism’, Routledge Companion to Epistemology, (eds.) S. Bernecker & D. H. Pritchard, 523–35, London: Routledge. Schaffer, J. (2004). ‘From Contextualism to Contrastivism’, Philosophical Studies 119, 73–104. ——— (2005). ‘Contrastive Knowledge’, Oxford Studies in Epistemology, (eds.) T. Gendler & J. Hawthorne, 235–272, Oxford: Oxford University Press. Schiffer, S. (1996). ‘Contextualist Solutions to Skepticism’, Proceedings of the Aristotelian Society (new series) 96, 317–333. Stanley, J. (2005). Knowledge and Practical Interests, Oxford: Clarendon Press. Stroud, B. (1984). The Significance of Philosophical Scepticism, Oxford: Clarendon. Unger, P. (1971). ‘A Defence of Skepticism’, Philosophical Review 80, 198–219. ——— (1975). Ignorance: A Case for Scepticism, Oxford: Clarendon. ——— (1984). Philosophical Relativity, Oxford: Blackwell. Williams, M. (1991). Unnatural Doubts: Epistemological Realism and the Basis of Scepticism, Oxford: Blackwell. Williamson, T. (2000). Knowledge and Its Limits, Oxford: Oxford University Press.

3

Impersonal Epistemic Standards1 Krista Lawlor

A knowledge claim is true or false depending in part on the epistemic standards that “knows” puts into play. Meg says, “Jim knows his friend took the car…” and her claim is true (false) depending in part on whether Jim’s epistemic position meets (fails to meet) the epistemic standards expressed by Meg’s use of “knows.” Does he have enough evidence to know? How demanding are the epistemic standards expressed by the use of “knows”?2 If “knows” puts into play exceedingly demanding standards, as some skeptics would have it, then we count as knowing very little, and our knowledge claims are largely false. If “knows” puts into play more forgiving standards, then we count as knowing more. Besides the question of how demanding epistemic standards are, there is the question of whether standards are invariant. An Invariantist argues that a knowledge claim “S knows p” made of a subject S and proposition p, is true (false) depending in part on whether S’s epistemic position with respect to p meets (fails to meet) an unvarying standard—that is, a standard that delivers the same judgment about a single knowledge claim made in different contexts of utterance. According to the traditional insensitive Invariantist, “Jim knows his friend took the car” is true only if Jim’s epistemic position with respect to the proposition my friend took the car is sufficiently strong to satisfy a single unvarying epistemic standard. The same standard applies regardless of who makes the claim or of whom the claim is made. What is thought to be valuable about invariant standards? Some argue that invariant standards are intuitively correct; others argue that invariant standards are what make possible the cross-contextual usefulness of knowledge claims: we can acquire epistemically useful information from knowledge claims made in contexts different from our own.3 Invariant standards are thought to explain this fact. I believe that the goods philosophers seek in explaining crosscontextual use are better delivered by impersonal standards. My aim here is to argue for this claim. By impersonal standards I mean standards that are insensitive to the idiosyncratic needs, interests or expectations of individual persons. Impersonal standards may be invariant or they may be variable—they may be sensitive to differences in common

34  Krista Lawlor or generic practical needs or interests of particular individuals. If I am right, the debate about cross-contextual usefulness that pits defenders of invariant standards against defenders of variable standards is mistaken. Impersonal standards are what deliver the sought-after goods.4 Here’s my plan for the paper. Many Invariantists argue that one advantage of their semantic theory is that it explains our ability to extract useful information from knowledge claims made in different contexts. In Section 1, I’ll recap the problem that cross-contextual use presents for opponents of Invariantism. Then, I’ll discuss how traditional Invariantists have their own, equally thorny, difficulty explaining the crosscontextual usefulness of knowledge claims. I’ll offer my diagnosis of why both Invariantists and their opponents (i.e. broadly Contextualist semantic theories) share this difficulty. In brief, both sides of the debate share a “personalist assumption” about the nature of the standards that are relevant (semantically or pragmatically) to the communication of important epistemic information: these standards are assumed to be fixed by the personal needs, interests, or expectations of individuals. I suggest that this widely shared assumption is mistaken, and it dooms us to fail when we try to build an account of how epistemically useful information flows across different contexts, no matter which semantic theory we adopt. The question then is, how are the relevant standards fixed so as to make the flow of information possible? Impersonal standards are the answer. But what sort of impersonal standard? In Sections 2 and 3, I begin answering this question by examining Edward Craig’s ideas about “objectivizing pressure” on the concept of knowledge. Objectivizing pressure seems to push us toward an epistemic standard that is invariant and skeptic-friendly. However, I argue that further consideration of Craig’s argument in fact leads us to an impersonal and anti- skeptical standard. Whether the impersonal anti-skeptical standard is also invariant, or whether it is variable, is an open question, deserving more inquiry. In Section 4, I flesh out the picture of what such an impersonal anti-skeptical standard might look like, illustrating it with a reasonable person standard. In Section 5, I consider objections. Let’s start by reviewing the problem that cross-contextual use presents for opponents of Invariantism.

1 Cross-Contextual Usefulness of Knowledge Claims We routinely rely on the reports of others, where those others are at some remove from us in time and place. Testimony is a vital source of knowledge for human beings. Often, we also garner epistemic information from speakers in circumstances very different than our own. For example, some years ago, a group of scientists published a paper in the journal Nature, reporting their finding that significant numbers of monarch larva die after eating genetically modified pollen. Lay people read this

Epistemic Standards  35 news and came to know that monarch butterflies were in danger. What they read gave them information not only about the danger to monarch butterflies, but also about the epistemic position of the scientists—that the scientists knew this fact. Knowledge claims are useful across contexts. (By “context”, I mean the context of utterance broadly understood to include both standard features such as time, location, addressee, and other variables such as the speaker’s practical interests or what is salient to her.) We extract information from knowledge claims about the state of the world, and also about “who knows what when.”5 It is important that there are natural limits on cross-contextual usefulness. A knowledge claim made in a context about which one knows nothing, and can assume nothing, is not so readily used to gain epistemic information. Walking on a foreign street one hears from behind a wall, “Now we know the Queen is abdicating”; even if one resolves the reference, it is not at all clear one can take this bit of information about the Queen on board, let alone suppose that it is known by the speaker. The idea of cross-contextual usefulness is not the idea that every knowledge claim provides knowledge to the consumer of the claim. Rather, the idea is that we can extract epistemic information from knowledge claims made in contexts different than our own, even in the absence of further communication about our respective differences. The challenge for any semantic account of knowledge claims is to explain how this truism holds, or at the least, not to create a mystery about how it might hold. A common complaint is that broadly Contextualist semantic theories fail this challenge.6 By “broadly Contextualist”, I mean any theory that has it that a single sentence attributing knowledge (“Jim knows that his car window is open”) can have different truth values in different contexts of utterance, owing to the semantic behavior of “knows.”7 Here is the complaint, using Indexical Contextualism as an illustration. The Indexical Contextualist holds that the truth-conditional content (or proposition) expressed with “Jim knows…” varies with the speaker’s context of utterance, elements of which determine the operative epistemic standards. Furthermore, Indexical Contextualists typically assume that these epistemic standards are fixed by the interests, aims, or expectations of the speaker.8 This is an important assumption—let’s call it the “personalist assumption.” Since epistemic standards help to determine the proposition expressed by a knowledge claim, our consumer needs to know about the speaker’s personal interests, aims, or expectations in order to learn the proposition expressed by her knowledge claim. But, the complaint continues, routinely a consumer in a different context does not know about such features of the speaker. So, the Indexical Contextualist cannot explain how such a consumer could gain epistemic information from a knowledge claim made in a context different than his own.

36  Krista Lawlor Patrick Rysiew notes that other “non-traditional,” broadly Contextualist, semantic theories also face difficulties meeting the challenge of cross-contextual use.9 For reasons of space, I will not recount all these theories—readers can consult Rysiew for details—and will just mention a second example from Rysiew: According to [Subject Sensitive Invariantism] (e.g., Stanley 2005; Fantl and McGrath 2009), the truth-value of a knowledge report depends on facts about the subject’s interests. But then, if I don’t know what those interests are (how high or low, etc.), I won’t know how far the putative fact of their knowing is owing to the (perceived) strength of various traditional, truth relevant properties (their having excellent evidence, etc.), and how far it’s owing to their having only a weak (/perceived) interest in the proposition in question. (288) For our purposes, the point is that as with other broadly Contextualist theories, what gets the defender of SSI into trouble is the personalist assumption. The epistemic standards that determine the truth of a knowledge claim are fixed by personal interests, aims, and expectations—this time those of the subject. Once again, it is only by having access to these that the consumer can extract epistemic information from the knowledge claim. Any view that makes the personalist assumption is in the same boat as the Indexical Contextualist when it comes to explaining how a consumer can extract useful epistemic information from a knowledge claim. The consumer has to know something about a particular person’s practical interests, aims, or expectations. But routinely enough, consumers do not have such information. At this point, it might seem that traditional Invariantism—what Rysiew calls the “ho-hum” view—has an advantage. Traditional Invariantists hold that the meaning of “knows” is constant across contexts of utterance. “Knows” expresses an invariant epistemic standard that does not vary with the attributor or subject. The consumer of a knowledge claim made in a practical context different than his own need not access anything about the practical interests of the speaker or subject, in order to know what proposition the speaker expressed. Invariantists thus seem to have an edge over Contextualists in accounting for the crosscontextual usefulness of knowledge claims. However, this appearance is deceiving—the Invariantist is not spared difficulty in explaining cross-contextual usefulness.10 To see why, first consider how on an Invariantist view a knowledge claim conveys information about the epistemic position of the subject. What might be called “standard Invariantism” distinguishes the truth- conditional content expressed by a knowledge claim (that the subject is in a

Epistemic Standards  37 knowledge- conferring epistemic position) from what is pragmatically imparted (that the subject is in a good enough epistemic position in light of the interests of the speaker).11 Rysiew writes: … the proposition literally expressed by the sentence ‘S knows that p’ is (unsurprisingly) that S knows that p; and, minimally, that S knows that p entails that S believes that p, p, and that S is in a good epistemic position with respect to p. It is a familiar idea, though, that what a given sentence means need not be what the speaker means in uttering it—i.e., what he intends to communicate. To capture this, I say that an utterance of ‘S knows that p’ pragmatically imparts (or conveys) a proposition of the form, ‘S’s epistemic position with respect to p is good enough…’, where the ellipsis is completed according to the interests, purposes, assumptions, etc., of the speaker. (My emphasis [Grazer “contesting contextualism”, 46].) Many traditional Invariantists acknowledge that what is pragmatically imparted is often the most important information the knowledge claim carries. A consumer of the claim often wants access to this information more than she wants access to the information conveyed by the truth-conditional content of the claim. That this is so is especially obvious about Skeptical Invariantism, since the truth-conditional content of a knowledge claim is rarely if ever true on such a view. It also holds for Modest Invariantism. Consider for example cases where the speaker denies a knowledge claim (“Jim doesn’t know…”). According to the traditional Modest Invariantist, this sort of denial is often used to pragmatically impart that the subject fails to meet the speaker’s standards, which are in some cases more exacting than the modest standards that fix the truth-conditional content of a knowledge claim.12 In such cases, the truth-conditional content of the denial is false, according to Modest Invariantism, but the consumer doesn’t focus on this content and instead focuses on what is pragmatically imparted. How does the consumer extract information about what is pragmatically imparted? Here many traditional Invariantists also make a personalist assumption. As Rysiew says, what is pragmatically imparted depends on “the interests, purposes, assumptions, etc., of the speaker.” In making this assumption, the Invariantist does not escape the problem of explaining how consumers make epistemic use of knowledge claims across contexts. The problem is relocated in the pragmatics of knowledge claims.13 Let me digress briefly to note that the personalist assumption creeps in to other discussions of knowledge claims. For instance, it rears its head in discussions of the so-called “stability problem.” As Stephen Grimm puts it, the problem is to avoid the manifest “absurdity of the idea … that the thresholds relevant to knowledge would vary wildly as the practical interests and concerns of the subject or the evaluator varied.”14

38  Krista Lawlor The stability problem is related, but not identical to, the problem of cross-contextual usefulness: it is conceivable that epistemic thresholds or standards vary quite a lot with the practical interests of relevant parties, but consumers of knowledge claims somehow manage to anticipate these variations and successfully glean information from one another’s knowledge claims. Nonetheless, the two problems are related: both arise with the assumption that practical interests of subjects or attributors control the standards attending knowledge claims. The problem of crosscontextual use arises when we add that people in different contexts do not routinely have access to these controlling factors. I want to emphasize that any solution to the stability problem that incorporates the personalist assumption will run into trouble with cross-contextual usefulness. To see how, consider Grimm’s own solution to the stability problem. Grimm’s solution is to fix epistemic standards by taking a set of individuals—the subject, the knowledge attributor, and a third party—and asking who in this set has the most demanding informational needs. On Grimm’s view, the stabilizing third party is a person the attributor thinks “might actually use the information” encoded in the knowledge claim. Now the worry is that this would seem to re-introduce the problem of cross-contextual usefulness of knowledge claims, inasmuch as knowledge attributors and consumers may have different thoughts about who might actually use the claim. If the consumer of the knowledge claim does not have access to the attributor’s thoughts, then it will prove difficult for the consumer to learn the proposition expressed by his knowledge claim. Other Craig-inspired functionalist views run into similar problems in explaining cross-contextual usefulness. For instance, Robin McKenna favors a “constrained local interest relativity” of epistemic standards, with standards depending on the practical interests of third parties whom people in the context of utterance (attributors and audience) have reason to consider.15 While allowing a psychological criterion to determine epistemic standards may help to stabilize standards, it threatens the cross-contextual usefulness of knowledge claims. Knowledge attributors and consumers in different contexts may have reasons to consider different third parties, and consumers in different contexts may not have access to the attributor’s reasons. Let’s recap. Existing semantic views create a real puzzle about how people manage to consume knowledge claims made in different contexts. There are possible solutions to this puzzle that both Invariantists and Contextualists can explore. But whatever semantic account of knowledge claims we favor, we need to reject the personalist assumption in all its forms. The assumption forces consumers to have information about the speaker’s or subject’s personal interests or expectations (be they personal stakes in the question at hand, or personal thoughts or expectations about the needs of third parties). But as consumers of knowledge claims we often don’t have this sort of information. The personalist

Epistemic Standards  39 assumption makes it difficult to explain how epistemically useful information flows across contexts. How do we replace the personalist assumption? It seems we need an impersonal epistemic standard. What does it mean for an epistemic standard to be impersonal? To answer this question, I turn to Edward Craig’s work on the concept of knowledge.16

2 Impersonal Standards and Common Practical Needs Start with Craig’s claim that concepts are as-if-designed by an “objectivizing pressure.” The main idea behind Craig’s claim is simple: concepts are tools with which to coordinate our actions, and consequently, a concept’s application conditions are shaped by this coordination.17 (Application conditions are conditions under which it is correct to apply the concept.18) Objectivizing pressure is pressure to make one’s use of a concept as helpful to others as possible. Objectivizing pressure forces the application conditions of a concept to be less subjective or idiosyncratic and to be more suitable to more people than just oneself, or one’s immediate interlocutor. Here is an illustration of how objectivizing pressure works on the concept chair. I might need something I can sit on now, but this need of mine does not determine the application conditions of the concept chair. As Craig writes, … in due course, I shall be interested, since I anticipate wanting to sit down at future times, in objects which I could sit on if I wanted to, or in whether there will be something which I can sit on when I want to (at the end of the walk). This interest will naturally lead to an interest in hearing the opinions of others as to where there are objects which I can sit on if I want to, irrespective of whether they want to sit on them or not; so I shall want them to operate an objectivised concept too. And if I grow a little more altruistic in my outlook, I may even be interested in whether there is something which Fred can sit on if he wants to, irrespective of whether I shall want to sit on anything or not. Hence the concept of something which is, in abstraction from what any particular person wants at any particular time or place, of even from whether anyone ever wants to sit down, simply suitable for sitting on. (84) Objectivizing pressure is a natural force, exerted as we coordinate with each other over the satisfaction of our needs. Concepts are coordinating tools that serve the practical needs of others as well as ourselves. We can call these “common” needs, to contrast them to the idiosyncratic needs of the particular person wielding the concept. By “common,” we might mean any of several different things: widely shared, routine, average, typical, or customary. Common needs may not be shared by absolutely

40  Krista Lawlor every human being. Some people can’t use chairs, for instance. How to characterize commonality is an interesting further question, and not one I can address fully here. It is likely that “common” is a context-sensitive term, with different extensions depending on the concept in question. On some ways of further specifying the idea, common needs may be tied to natural functions of human bodies—humans have a natural need to not stand all day. On other ways of specifying the idea of commonality, normative functions of humans as they pursue the good life may be what define common needs. I will use the term “common” throughout, but I caution the reader that this term is used primarily to contrast it with “idiosyncratic,” “personalized,” or “individualized.” Now consider the concept of knowledge. The application conditions for knows encode epistemic standards for what counts as being in a sufficiently strong epistemic position with respect to a target proposition p. (If a subject S falls into the category of knowers of some proposition p, then the application conditions of the concept tell us that is because S is in a sufficiently strong epistemic position with respect to p.) Craig’s insight is that objectivizing pressure works on these standards, so they are not fixed by the idiosyncratic or personal needs of the concept wielder. Rather, they are responsive to common practical needs. Thus, one person’s knowledge claim can be useful to a consumer with very different practical interests, aims, and expectations. Craig’s work points the way to an answer to our question about what it means for an epistemic standard to be impersonal. Impersonal standards are not fixed by the idiosyncratic or personal practical needs or expectations of individuals. Impersonal standards are fixed by common needs or expectations of human beings. In the case of knowledge, we have a need for reliable information widely shareable with others about whom we have little specific knowledge. This is our common need. In light of it, the appropriate standard for the truth of a knowledge claim is fixed.

3 Impersonal Standards: Skeptical or Anti-skeptical? We now know a bit about what it means to be an impersonal standard. If the foregoing line of thought is right, then the application conditions for the concept knows encode impersonal epistemic standards that are shaped by common our needs and expectations. What else can we say about the standard? In Craig’s text, the reader first finds an argument that suggests the correct impersonal standard is a skeptical invariantist one. It seems, after all, that objectivizing pressure on the concept of knowledge tends to raise epistemic standards to skeptical heights and to keep them there: All this is going to edge us towards the idea of [the knower as] someone who is a good informant as to whether p whatever the particular

Epistemic Standards  41 circumstances of the inquirer, whatever rewards and penalties hang over him and whatever his attitude to them. That means someone with a very high degree of reliability, someone who is very likely to be right—for he must be acceptable even to a very demanding inquirer. (91) In saying that someone knows whether p we are certifying him as an informant on that question, and we have no idea of the practical needs of the many people who may want to take him up on it; hence a practice develops of setting the standard very high, so that whatever turns, for them, on getting the truth about p, we need not fear reproach if they follow our recommendation. … If I don’t know what you want it for or what fate awaits you if you get it wrong, I had better make sure that the level of reliability [of my knowledge claim] is very high indeed. (94) … high enough to satisfy all, or all practical, purposes. (98) How high is high enough? Craig answers: Is there anything more we can say as to just what level this is? The above line of argument would suggest a fairly stark answer: the probability that the belief is true needs to be 1, no less. (98) So, on this line of thought, at the end of objectivizing pressure lies skeptical invariantism—unvarying epistemic standards so demanding that one knows p only if one’s evidence makes it certain that p is true. This line of interpretation is challenged by Contextualists.19 These philosophers point to evidence that Craig’s view is that epistemic standards are neither invariant nor set at a demanding, skeptic-friendly level: Real situations sometimes force us to set [standards] very high, but not actually at [probability] 1. … In everyday practice we happily bandy the word ‘know’ about without having to feel that our chances of being wrong are literally zero. (102, my emphasis) But, …even when we do set the likelihood at 1 that does not mean the type of absolute certainty, invulnerability to literally any theoretical possibility, which the sceptic characteristically demands… Nothing… need push us any further than the demand that the actual or ‘real’ chance

42  Krista Lawlor of being wrong should be zero—and this is in good accord with my genetic account of the concept and pragmatic account of its point. (102) Some Contextualists would argue that Craig here acknowledges a context sensitivity in our epistemic standards: sometimes real situations force our standards very high, other times … even higher. I do not wish to adjudicate whether Craig was an Invariantist or a Contextualist, or whether Invariantism or Contextualism is the correct semantic account for knowledge claims. What I want to note is that Craig offers an important argument against skeptic-friendly standards (“Nothing need push us further” toward requiring absolute certainty). Craig argues that objectivizing pressure works to keep the concept of knowledge practically useful. A concept with skeptic-friendly application conditions would find little use. 20 Recall our example of the concept chair. A person who is first ready to deploy the concept with application conditions that are geared to her specific personal needs eventually comes to deploy a concept geared to a much wider range of needs—the need to support a wide range of body sizes, and so on. With objectivization, the concept’s application conditions become less idiosyncratic and more impersonal. At the same time, application conditions do not lose touch with the practical needs of human beings—a chair is something for sitting on. So objectivizing pressure on a concept produces application conditions that are responsive to common human practical needs, not personal, or idiosyncratic ones. 21 Objectivizing pressure on the concept of knowledge pushes against skeptic-friendly epistemic standards and simultaneously pushes toward impersonal ones. There are further pressing questions to ask about the nature of the impersonal epistemic standard. How on Craig’s view do the more forgiving, anti-skeptical standards for knowledge get settled? What counts as a “real” chance of being wrong? Craig says little to guide us. It seems that on his view, the standards for knowing—the application conditions for the concept—are held in a state suspended between the competing pressures to be useful in a wide range of practical settings, yet also sufficiently stringent to satisfy “a very demanding inquirer.” Of course, the question what it takes to be in a sufficiently strong epistemic position to count as knowing is one of the central questions of epistemology, so it’s not surprising that, having brought us this far, Craig’s answers run out. At least Craig helps us sharpen our search: his insight is that we need to identify an impersonal and anti-skeptical epistemic standard that is simultaneously sensitive to our common practical interests in reliable information. 22 Because the standard is impersonal, but sensitive to common practical interests, it will allow us to avoid skepticism, while also making sense of the cross-contextual usefulness of knowledge claims.

Epistemic Standards

43

Whether the right impersonal anti-skeptical standard is invariant or whether it is variable is an open question, deserving more inquiry. I will not try to settle this question. What I aim to do in the next section is to further flesh out what one such impersonal standard might look like, and use cases to illustrate how an impersonal standard is superior to a personal standard.

4 An Impersonal Standard Sensitive to Common Practical Interests We can imagine different possible accounts of an impersonal epistemic standard that is simultaneously sensitive to our common practical interests in reliable information. Some ways of fleshing out the idea produce a standard that is Invariantist-friendly and others produce a standard that is Contextualist-friendly. For instance, a crude sort of Invariantist-friendly account might go this way: once a topic of inquiry is fixed, our common human interests in the topic determine the strength of epistemic position one needs to be in, to count as knowledgeable about the topic. If the topic is finding food or other essentials, say, as opposed to finding entertainment, then greater common practical interests in reliable information about the former drive the epistemic standards higher. This standard is sensitive to common human interests but also impersonal, in that idiosyncratic interests of individual knowers do not affect what is demanded of knowers. I will suggest one way of fleshing out what an impersonal standard might look like. My suggestion is that a “reasonable person” standard is a good candidate for the role. My aim is to give an illustrative example of an impersonal standard that is sensitive to practical interests, and to demonstrate how an impersonal standard is superior to a personal standard in the way it delivers cross-contextual usefulness of knowledge claims. Before discussing cases, I should say how I’ll understand what reasonableness is. If a person is reasonable then she is sensitive to what’s important or valuable in a situation, keeps in view her own ends and aims as well as those of others, takes into consideration the facts of a situation, sizes up the reasons for and against various beliefs and courses of action, including the reasons others see, and acts for the good in accord with those reasons. Here is how I’ll deploy the reasonable person standard. For illustrative purposes, I will assume an attributor-Contextualist semantics for knowledge claims: Whether a subject S is in a sufficiently strong epistemic position with respect to a target proposition p to count as knowing p will depend on whether S’s epistemic position is strong enough to satisfy a reasonable person that no further inquiry into p is required. As I will understand the reasonable person standard, it introduces constrained sensitivity

44  Krista Lawlor to practical interests. Let’s stipulate that the reasonable person is apprised of the features of the subject’s situation and aware of the practical interests of the knowledge attributor, and since the reasonable person takes account of such interests, they can affect how strong a subject’s epistemic position has to be in order to count as knowing. But crucially, these interests do not determine or control how strong an epistemic position S must be in to count as knowing; they are taken account of by the reasonable person, but the reasonable person standard is impersonal, demanding that people consider the question of sufficiency of epistemic position with an impartial eye.23 A reminder: I do not mean to be arguing for attributor Contextualism in what follows—an Invariantist semantics could deploy the reasonable person standard I’ve just described in an account of what is pragmatically imparted by a knowledge claim. I am simply using this semantic view to illustrate how an impersonal standard might work. Let’s see how this view contrasts with a personalist view of epistemic standards. Start with Stewart Cohen’s “airport case” (1999, 58): Mary and John are at the L.A. airport contemplating taking a certain flight to New York. They want to know whether the flight has a layover in Chicago. They overhear someone ask a passenger Smith if he knows whether the flight stops in Chicago. Smith looks at the flight itinerary he got from the travel agent and responds, ‘Yes I know—it does stop in Chicago.’ Contextualists say that it is intuitively plausible that Smith’s claim is true: we routinely make and accept knowledge claims made on such grounds. Here is how our suggested standard might be applied by an attributor-Contextualist to explain this judgment about the case. Smith’s knowledge claim, “I know the flight stops in Chicago” is true, because Smith’s epistemic position is sufficiently strong as judged by a reasonable person in Smith’s context. A reasonable person apprised of Smith’s situation, and aware of Smith’s interest in the question judges that Smith’s evidence is sufficient for taking Smith’s claim to be true. The motivating case for Contextualism continues: It turns out that Mary and John have a very important business contact they have to make at the Chicago airport. Mary says, ‘How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute.’ Mary and John agree that Smith doesn’t really know that the plane will stop in Chicago. They decide to check with the airline agent. Contextualists, employing personalist epistemic standards, have argued that Mary’s claim (“Smith doesn’t really know”) is also true. What does our account, with its impersonal standard, say about the case?

Epistemic Standards  45 Applying the impersonal reasonable person standard to this case reveals it to be under-described. Mary’s claim, “Smith doesn’t know” is true only if Smith’s epistemic position is insufficiently strong as judged by a reasonable person apprised of Smith’s situation and with Mary’s practical interests in mind. Is it insufficiently strong as so judged? The case as described doesn’t give enough information to answer this question. Note that this question doesn’t arise for those who make the personalist assumption, according to which Mary’s elevated personal interests simply determine the governing standard for knowledge claims. But using the reasonable person standard, the question is not settled in this way. If a reasonable person, aware of Mary’s practical situation and of Smith’s situation would want more information than delivered by Smith’s casual glance, then Mary’s claim is true; and if a reasonable would find Smith’s inquiry sufficient, then Mary’s claim is false. Deploying the reasonable person standard, the Contextualist will have to argue further for their favored reading of this motivating case. Let’s see more about how an impersonal standard such as the reasonable person standard works. In the cases that follow, I will stipulate what is reasonable, so we can focus on the advantages of impersonal over personal epistemic standards. Consider the case of Jones, whose practical interest in the question of the plane’s journey is much greater than Smith’s. Jones has an enormous aversion to stopping in Chicago because of memories of bad trips he’s made there in the past. Let’s also stipulate that Jones’s personal desire to avoid Chicago is real but nothing about his distaste for stopping in Chicago would make a reasonable person seek more evidence before claiming to know the plane stops there. “Smith knows the plane stops in Chicago” is true in Jones’ context, despite Jones’ greater practical interest in the question. Further, Jones can make epistemic use of Smith’s claim, provided he can be reasonable himself. Of course the actual cross-contextual use of knowledge claims raises all the thorny questions regarding the epistemology of testimony, and it’s not my business here to give a full account of what Jones needs to know about Smith, how much he can take on trust, and so forth. The point we’re focusing on is that because Smith’s evidence meets the reasonable person standard, his knowledge claim can be useful to someone, such as Jones, in a different context. The reasonable person standard places constraints on how practical interests can affect the truth-conditions of knowledge claims, and allows reasonable people to interpret each other using the usual (fallible) tools of interpretation. Consider next the case of Johnson, the sort of person who is very hard to convince—he is strictly skeptical about anything he hasn’t verified for himself. Is this reasonable? Let’s stipulate it is not. (Philosophers may argue about this, but ordinary folk will quickly lose patience with

46  Krista Lawlor Johnson. Again, I won’t settle the dispute, but stipulate for sake of discussion.) Even if Johnson’s personal epistemic practices are skeptical, Smith’s claim presents him a genuine opportunity to gain knowledge about where the plane stops. Should Johnson decide not to take advantage of Smith’s claim, his personal decision tells us nothing about the epistemic information that is there to be extracted. Were he reasonable, he could learn something from Smith’s claim. Finally, let’s imagine a case where practical interests do make a difference to the truth of a knowledge claim. Imagine Wilson has an extremely important, time-sensitive mission to complete in Chicago, and she urgently needs the plane to make a stop there. Let’s stipulate that while Smith is satisfied by a casual look at his itinerary, a reasonable person with Wilson’s interests in mind would not be. So Wilson’s denial (“Smith doesn’t know”) is true. Now what are we to say about the cross-contextual usefulness of the information conveyed by Smith’s claim? Wilson can make some use of Smith’s claim, even with her elevated practical interests. Stories about the shape of this epistemic information will differ in their details: for instance, we might say that for Wilson more alternatives are relevant, but Smith’s claim helps to eliminate at least some of these alternatives, and Wilson can learn this, in part because she can make her interpretation of Smith guided by a reasonable person standard. A solution to the problem of crosscontextual usefulness, recall, recognizes that there are natural limits on cross-contextual usefulness. The idea of cross-contextual usefulness is not the idea that every knowledge claim provides knowledge to the consumer of the claim. Rather the idea is that a knowledge claim can be consumed by people in different practical contexts from the knowledge attributor, even in the absence of further communication about their respective differences. These cases illustrate how the reasonable person standard, as an impersonal standard, can deliver cross-contextual usefulness of knowledge claims. 24 Even if reasonable person standard is—as I have construed it here—sensitive to practical interests, it is sufficiently impersonal to ensure that purely idiosyncratic features of the speaker or subject do not determine what it takes to count as knowing. An impersonal standard, even one that is sensitive to practical interests, thus helps ensure that epistemically useful information can be gleaned across contexts. Whether the reasonable person standard is the correct impersonal standard or whether a superior alternative can be identified is certainly worth more investigation. For present purposes, I hope to have provided a good working example that illustrates the idea of an impersonal standard.

Epistemic Standards  47

5 Objections Let’s turn to some objections. The approach I have pursued has been to identify epistemic standards by considering what they would have to be in order for us to make cross-contextual use of knowledge claims. One objection targets this approach: (a) Application conditions for the concept of knowledge cannot be shaped by our practical needs. Specifically, epistemic standards— key elements in the application conditions for knows—cannot be shaped or affected by our needs for cross-contextual usefulness of knowledge claims. Our need for knowledge claims to be cross-context useful is one thing, and what we need communicated—namely knowledge—is another. We should investigate what makes knowledge knowledge by identifying the epistemic standards for knowledge, and only after we have identified these should we go on to consider how it might be that knowledge claims can be useful across contexts. Reply: Compare an objection that the application conditions for the concept chair cannot be shaped by our needs for sitting. Regarding the concept chair, the objection isn’t compelling. The objection suggests that the application conditions for the concept should somehow be independent of considerations about our practical need for something to sit on. 25 The fact that we theorists investigate the concept chair by thinking about the function of chairs in no way makes chairs themselves anything other than objective physical things. So too, when we theorists investigate the epistemic standards that figure in the application conditions of the concept of knowledge by thinking about the cross-contextual function of knowledge claims, this fact about our investigation in no way makes these epistemic standards any less objective, or any less epistemic. A different sort of objection concerns my specific suggestion about the form the impersonal epistemic standard might take: (b) Doesn’t the reasonable person standard invoke another context sensitive term, “reasonable”, and if so, how can that help with the problem of explaining cross-context uses?26 Reply: As I understand it, “reasonable person” means the same thing in different contexts of utterance: a reasonable person is sensitive to what’s important in a situation—what is of value, to herself and to others, and makes impartial judgments in light of these values. As I understand the reasonable person standard, because there are different values to register

48  Krista Lawlor in different situations, this may result in variation in the strength of epistemic position required for knowledge. So the reasonable person standard is to that extent context-sensitive. In the cases above, I hope to have shown how this sort of variation is tightly constrained, and so is consistent with widespread cross-contextual usefulness of knowledge claims. As I noted earlier, our task is to explain how cross-contextual usefulness is possible, not to show that in every case it is easy to glean epistemically useful information from knowledge claims. Another concern targets the idea of the reasonable person standard from a different angle: (c) Talk of the “reasonable person standard” is like talk of “ordinary” or “usual” standards and is far too vague to do any significant philosophical work in explaining how knowledge or knowledge claims work.27 Reply: First, I should say that I doubt that the term “reasonable” is vague. 28 Vagueness is typically understood in terms of borderline cases, or hard-to-define boundary conditions. And most likely the objector is not concerned that one might set up a sorites-style paradox for reasonableness. What the objector might be concerned about is rather that what is reasonable is left to the judgment (say, of the knowledge attributor in their own context of use). This worry, though, involves a misunderstanding about how the standard works: the attributor does not decide what is reasonable; rather, a reasonable person with the attributor’s practical interests in mind decides. This means that knowledge attributors may be wrong about the truth value of their own utterances. Alternatively, what the objector might be worried about is more like the following: (d) “what is reasonable” is too open to dispute to in fact settle anything—if epistemic standards for knowing involve the reasonable person standard, then there is no fact of the matter about who knows what. The topic of reasonable disagreement is a large one, and my response here will be brief. My first response to this objection is that what is reasonable is not whatever people say is reasonable. “Reasonable” is not like “tasty”—a property ascribed correctly in light of one’s subjective take on the application of the term. So one sort of dispute the objector might be worrying about is not in fact a worry at all. That said, I think we might grant that it is an open possibility that reasonable disagreements about what is reasonable may arise. If that’s so, then reasonable disagreements about the truth of a knowledge claim may also arise. But now we also have to go a bit slower in assessing where this possibility

Epistemic Standards  49 would leave us. One consequence of such reasonable disagreement might be that there are sometimes gaps in the truth values of knowledge claims. But disagreement might not have this consequence. For it also seems possible that where reasonable people have disagreements about what is reasonable, there are also reasonable ways to settle these disagreements. For instance, if the need is great enough for a definitive judgment about what the epistemic standard is to be, reasonable people can agree to stipulate an epistemic standard. I believe something like this happens in scientific settings, when scientists stipulate a standard for significant results. This is a reasonable way to set the standard for joint inquiry. This is but one illustration—it shows that there may be many ways to answer questions about how we resolve reasonable disputes about our epistemic standards, once we liberate ourselves from the personalist assumption.

6 Conclusion Concepts are cognitive tools that help us coordinate our activity. Consequently, objectivizing pressure shapes the application conditions for the concept of knowledge, and thus shapes the epistemic standards that “knows” puts into play. I have argued that as a result, epistemic standards must be impersonal. Whether epistemic standards are invariant or not is open to further inquiry; impersonal standards can be part of a Contextualist or an Invariantist account of the semantics of knowledge claims. I have illustrated that impersonal standards need not be invariant. Impersonal epistemic standards help ensure that knowledge claims are useful across contexts, even when speakers and hearers don’t know much about each other’s needs or circumstances. An impersonal standard helps ensure that knowledge claims will be useful to many people with different practical interests.

Notes 1 Thanks to Christos Kyriacou and Patrick Rysiew for very helpful comments. 2 Hereafter, I’ll abbreviate “the epistemic standards that attend our use of ‘knows’” as “epistemic standards” or just “standards.” These standards may characterize what it takes to know, or what information a knowledge claim pragmatically imparts. 3 Williamson, “Knowledge, Context, and the Agent’s Point of View”; Schaffer, “From Contextualism to Contrastivism”; Rysiew, “Epistemic Scorekeeping.” 4 Not all Invariantists claim that cross-contextual use speaks for Invariantism. Some Invariantists have seen that the problem of cross-contextual usefulness does not tell clearly in favor of their view (see for instance Rysiew, ibid.). Also, it is consistent with traditional Invariantist views that epistemic standards governing the truth of knowledge claims are impersonal while pragmatic standards governing information conveyed pragmatically by knowledge claims are personal. The point I am concerned to make here is that for the communication of cross-contextual epistemic information, we need impersonal standards.

50

Krista Lawlor

5 Williamson, ibid., 101. 6 Rysiew presses this concern and notes that others have pressed this point against Contextualism. This is one place where non-traditional views struggle: being geared towards, indeed often designed specifically to handle, variable contexts, such theories must explain how knowledge reports come to be useful to and fit for ‘consumption’ by the inquiring public at large. (Ibid., 286)

7

8

9 10 11 12

13

14 15

It should be noted that in pointing out that the non-traditionalists struggle with cross-context usefulness, Rysiew’s aim is to argue that there are challenges for both traditional and non-traditional semantic views. Schaffer and Szabo, “Epistemic Comparativism: A Contextualist Semantics for Knowledge Ascriptions,” 492–493 use this formulation to describe Contextualist semantic theories. However, as I understand the stated formulation, other views such as Subject Sensitive Invariantism and Relativism meet the description. These are “non-traditional views” in Rysiew’s sense. See for instance DeRose, “Solving the Skeptical Problem”; Cohen, “Contextualism, Skepticism, and the Structure of Reasons”; Lewis, “Elusive Knowledge.” Lewis also emphasizes what is salient to the speaker, and the discussion below to this sort of view as well. By “broadly Contextualist” views I mean to include SSI and Relativism (see note 5 above). This point is made by Moeller, “Consuming Knowledge Claims across Contexts.” Traditional Invariantists include Rysiew (ibid.), Brown, “Contextualism and Warranted Assertibility Manoeuvres”; Black, “Classic Invariantism, Relevance and Warranted Assertability Manœvres.” Rysiew notes that different kinds of epistemic information can be pragmatically imparted. For example: In a given case I might say such a thing not to remark upon my knowing (/not knowing) per se, but to get across the information that, for instance, my epistemic position is good enough that you can take my word on the matter… Rysiew, “Contesting Contextualism.” This sort of epistemic information often is of greater interest to speakers and hearers than the truth-conditional content of the knowledge claim. Standard Invariantists might reply that there may be further information to be gleaned across contexts, if knowledge claims pack additional conventional information. For instance, Rysiew (ibid.) suggests that knowledge claims conventionally certify information as true, and competent speakers know as much; so the possibility opens up of giving an account of how consumers can retrieve some useful information from a knowledge claim without their having to know the specific communicative intentions of the speaker. One question is how to give such an account while handling the variability data made salient by Contextualists. Grimm, “Knowledge, Practical Interests and Rising Tides,” 135. See McKenna, “Knowledge Ascriptions, Social Roles and Semantics” section 2.2. Another functionalist, Michael Hannon (“Stabilizing Knowledge”) argues for a “default and exception” style view, where “epistemic standards are partly set by the practical needs and interests of a given inquirer but they are also constrained by a much wider set of interests” (130), should these interests prove more exigent. I believe there are questions for Hannon’s view

Epistemic Standards  51

16

17

18 19

20

21

22

23

about how knowledge claims are interpretable within a context, which also makes interpretation across contexts difficult. See Lawlor (forthcoming). Craig, Knowledge and the State of Nature. Craig’s work has focused our attention on the function(s) of the concept of knowledge and the bearing function has on the semantics of knowledge claims. My interest here, however, is not in the function(s) of the concept of knowledge. What interests me is Craig’s ideas about the cross-contextual usefulness of knowledge claims, and how this usefulness affects epistemic standards. In this, I follow other functionalists Grimm, McKenna and Hannon (see note 13). However, the moral I draw from Craig is different—I believe Craig’s work points us toward impersonal epistemic standards. We need not completely share judgments about application conditions in order to communicate. Craig’s point is that there is pressure on the application conditions themselves to work in a wide range of settings for people with differing personal interests. Traditional views of concepts have it that application conditions (or intensions) take the form of necessary and sufficient conditions for an item to fall under a concept. Nothing I say here requires this traditional view. Several philosophers read Craig as advocating skeptical Invariantism: Gerken, On Folk Epistemology. How We Think and Talk about Knowledge, 199; Kelp, “What’s the Point of ‘Knowledge’ Anyway?” n2. Others read Craig as advocating a Contextualist semantics: in addition to the functionalists mentioned above, see Greco, “What’s Wrong with Contextualism?” and Henderson, “Motivated Contextualism.” My concern is not to take sides on this interpretive question here, but to focus on Craig’s suggestions about impersonal standards. I assume here that such a concept would see little use, because it would require of its users that they keep a double set of books about the application conditions for the concept and the warranted assertability of the term that expresses the concept. Though this double set of books would be challenging enough that it would behoove people to coordinate around a concept with more forgiving application conditions. Henderson, “Gate-Keeping Contextualism,” makes the important point that it is not just individual hearers, but communities that have needs. Knowledge claims made to meet community needs result in context sensitivity. For instance a knowledge claim such as “we know aspirin is safe” made by an “applied practical communities” (i.e. ordinary people looking for pain relief) are to be evaluated with respect to different standards than claims made by “general-purpose source communities” (i.e. scientists studying NSAIDS), since the latter aim to provide information to a wide range of applied communities. I overlook these important observations about social aspects of knowledge claims here. Note that I wish to remain neutral here between Contextualist and Invariantist semantic accounts (see supra note 18 above). To say the standard is sensitive to common practical interests is not to say that Contextualism follows. An invariant standard might be sensitive to common practical interests, as for instance a standard that requires attention to the common need of the greatest number of people. I have argued that a reasonable person standard is the correct epistemic standard for knowledge. Lawlor, Assurance: An Austinian View of Knowledge and Knowledge Claims. There is much to be said about this standard: what gives it content, what sort of objectivity it exhibits, whether reasonableness is practical or theoretical, and so whether pragmatic encroachment attends its use. I say more about these issues in “Knowledge and Reasonableness.”

52  Krista Lawlor 24 I want to note that if one is an Invariantist, then one can also put the reasonable person standard as I have construed it to work explaining the standards that govern the pragmatics of knowledge claims. 25 Craig considers a related objection to his approach, ibid., 94–96. 26 The objection is similar to one made by Williamson. See Rysiew, “Epistemic Scorekeeping,” 290. 27 This objection is similar to one found in Williamson, cited by Rysiew, 289. 28 Although it may like many other terms be “unspecified” in the sense of Kompa, “The Context Sensitivity of Knowledge Ascriptions.”

Works Cited Black, Tim. “Classic Invariantism, Relevance and Warranted Assertability Manœvres.” Philosophical Quarterly 55, no. 219 (2005): 328–336. Brown, Jessica. “Contextualism and Warranted Assertibility Manoeuvres.” Philosophical Studies 130, no. 3 (2006): 407–435. Cohen, Stewart. “Contextualism, Skepticism, and the Structure of Reasons.” Noûs 33 (1999): 57–89. Craig, Edward. Knowledge and the State of Nature: An Essay in Conceptual Synthesis. Oxford University Press, 1999. DeRose, Keith. “Solving the Skeptical Problem.” Philosophical Review 104 (1995): 1–52. Gerken, Mikkel. On Folk Epistemology. How We Think and Talk about Knowledge. Oxford University Press, 2017. Greco, John. “What’s Wrong with Contextualism?” Philosophical Quarterly 58, no. 232 (2008): 416–436. Grimm, Stephen. “Knowledge, Practical Interests and Rising Tides.” In Epistemic Evaluation: Purposeful Epistemology, edited by David Henderson and John Greco, 117–137. Oxford University press, 2015. Hannon, Michael. “Stabilizing Knowledge.” Pacific Philosophical Quarterly 96: 116–139. Henderson, David. “Gate-Keeping Contextualism.” Episteme 8, no. 1 (2011): 83–98. ———. “Motivated Contextualism.” Philosophical Studies 142, no. 1 (2009): 119–131. Kelp, Christoph. “What’s the Point of ‘Knowledge’ Anyway?” Episteme 8, no. 1 (2011): 53–66. Kompa, Nikola. “The Context Sensitivity of Knowledge Ascriptions.” Grazer Philosophiche Studien 64 (2002): 1–18. Lawlor, Krista. Assurance: An Austinian View of Knowledge and Knowledge Claims. Oxford University Press, 2013. ———. “Comments on Michael Hannon’s What Is the Point of Knowledge?” forthcoming in Analysis. ———. “Knowledge and Reasonableness.” Synthese 2020, https://doi. org/10.1007/s11229-020-02803-z. Lewis, David. “Elusive Knowledge.” Australian Journal of Philosophy 74 (1996): 549–567. McKenna, Robin. “Knowledge Ascriptions, Social Roles and Semantics.” Episteme 10, no. 4 (2013): 336–350

Epistemic Standards  53 Moeller, Emil Frederik Lundbjerg. “Consuming Knowledge Claims across Contexts.” Synthese 192, no. 12 (2015): 4057–4070. Rysiew, Patrick. “Contesting Contextualism.” Grazer Philosophische Studien 69, no. 1 (2005): 51–70. ———. “Epistemic Scorekeeping.” In Knowledge Ascriptions, edited by Jessica Brown and Mikkel Gerken. Oxford University Press, 2012: 270–294. Schaffer, Jonathan. “From Contextualism to Contrastivism.” Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition 119, no. 1/2 (2004): 73–103. Schaffer, Jonathan, and Zoltan Gendler Szabo. “Epistemic Comparativism: A Contextualist Semantics for Knowledge Ascriptions.” Philosophical Studies 168, no. 2 (2013): 491–543. Williamson, Timothy. “Knowledge, Context, and the Agent’s Point of View.” In Contextualism in Philosophy: Knowledge, Meaning, and Truth, edited by Gerhard Preyer and Georg Peter, 91–114. Oxford University Press, 2005.

Part II

Arguments for Infallibilist Skepticism

4

A Cumulative Case Argument for Infallibilism Nevin Climenhaga

This essay is a defense of the following thesis about knowledge: Infallibilism

S knows that P only if P is certain for S.1

The certainty here is epistemic certainty, as distinct from psychological certainty. In ordinary language, we use the locution “I am certain that P” to say that I am psychologically certain that P, and the locution “P is certain” to say that P is epistemically certain for me (Audi 2003: 224; Stanley 2008: 36–37). The former locution ascribes a psychological property to me (relative to P), while the latter ascribes an epistemic property to P (relative to me). The two are connected, though: if P is epistemically certain for S, then S ought to be psychologically certain that P. 2 In the next eight sections, I present eight intuitive claims about knowledge. Each claim is plausible given infallibilism, but in tension with fallibilism. This essay is “big-picture,” and I do not try to demonstrate that, in each case, there is no way for the fallibilist to accept the intuitive claim in question. Rather, I aim to show that each of these claims is easier to reconcile with infallibilism than fallibilism, and that consequently, each intuition is a prima facie consideration in favor of infallibilism.3 While individually, these considerations might not be conclusive, together they provide strong support for infallibilism. Explanatory arguments like that advanced here are fundamentally comparative: my claim is that infallibilism explains these intuitions better than rival fallibilist theories. For the sake of space, I am unable to discuss all rival theories here. In particular, I limit my targets to fallibilist theories on which knowledge does not require epistemic probability 1. Williamson’s “knowledge-first epistemology,” on which knowledge implies epistemic probability 1 but not certainty (see Williamson 2000: chapter 10.2),4 deserves separate discussion. I suspect that Williamson’s theory can account for some of the intuitive data I go on to adduce and not others, but the issue is too complicated to get into here. Instead, I only argue here that infallibilism is superior to theories on which knowledge is compatible with epistemic probability less than 1; and when I refer to ‘fallibilism’ in what follows, this is the view I have in mind.

58  Nevin Climenhaga I do, however, devote special attention to two non-knowledge-first theories specially tailored to account for some of the intuitive data I present: contextualism and interest-relativism. In presenting my eight claims in Sections 1–8, I initially assume that if infallibilism is false, traditional (non-contextualist, non-interest-relativist) fallibilism is true. Then, in Section 9, I consider what resources contextualists and interest-relativists have for explaining the intuitions I put forward. A complete defense of infallibilism must respond to the main objection contemporary epistemologists level against the view: that it has unduly skeptical consequences that conflict with our ordinary intuitions and ascriptions of knowledge (see, e.g., DeRose 1999: 202–203; Hawthorne 2004: 126–131). After all, there seems to be very little we should be maximally confident in. Even if it’s highly probable that I have hands, it seems epistemically possible that I am a handless brain in a vat. One way to respond to this objection is to deny that infallibilism has such restrictive consequences about the scope of our knowledge (see, e.g., Dutant 2016). For the sake of argument, however, I assume in this essay that the critics are right that if infallibilism is true, we know very little. In particular, I assume that the only facts certain for us are introspectable facts about our mental lives and a priori facts that we can know immediately through intuition.5 I argue in Section 10 that, while the skeptical consequences of infallibilism are then a cost of the theory, this cost is outweighed by infallibilism’s capacity to explain such a large variety of intuitive data about the nature and theoretical roles of knowledge. On to the eight claims. The claims I will present in Sections 1–3 are as follows: 1 2 3

There is a qualitative difference between knowledge and nonknowledge. Knowledge is valuable in a way non-knowledge is not. Subjects in Gettier cases do not have knowledge.

The claims in Sections 4–8 have to do with theoretical roles that knowledge apparently plays—the ways in which it appears to be related to rational inference, epistemic modality, rational action, rational inquiry, and deduction: 4 5 6 7 8

If S knows that P, P is part of S’s evidence. If S knows that P, ~P is epistemically impossible for S. If S knows that P, S can rationally act as if P. If S knows that P, S can rationally stop inquiring whether P. If S knows each of {P1, P2 , … Pn}, and competently deduces Q from these propositions, S knows that Q.

I will argue that each of these claims is plausible if infallibilism is true, but inconsistent with or in strong tension with fallibilism.

A Cumulative Case Argument  59

1 Knowledge Is Qualitatively Different from Non-knowledge The most influential theories of knowledge advanced in the past century defend some condition or conditions as, together with belief and truth, necessary and sufficient for knowledge. Often these conditions are presented as analyses or varieties of “epistemic justification.” Here are several proposed necessary conditions on S’s knowing that P6: i ii iii iv v

S’s belief that P is sufficiently probable on S’s evidence. If P were false, S would not believe that P. If S were to believe that P, P would not be false. S’s belief is produced by a reliable cognitive process. S can rule out all relevant alternatives to P.

Most of these conditions contain threshold terms: “sufficiently probable,” “reliable,” “relevant alternatives,” etc. The others, while not containing threshold terms, are still open to more or less strict interpretations: for example, one could interpret sensitivity as requiring that in all worlds in which P is false, S does not believe P, or as simply requiring this for some particular worlds. It appears that there is a qualitative difference between knowing a proposition and having a belief that is highly justified, but not highly enough for knowledge. Infallibilism explains this, because if infallibilism is true, then I know P only when P is certain for me. In this case I can rule out all alternatives to P; and my belief is maximally probable, formed by a maximally reliable process, etc. (Climenhaga forthcoming). This is a qualitatively different situation from having a belief that is highly, but not maximally, justified in one of the above ways. So, infallibilism makes it plausible that 1

There is a qualitative difference between knowledge and nonknowledge.

By contrast, fallibilist theories based around conditions that come in degrees have difficulty explaining how knowledge could be qualitatively different from almost-but-not-quite knowledge. For the move from 0.89 to 0.9 probability, or truth in the five nearest possible worlds to truth in the six nearest worlds, and so on, is a quantitative one, and not a qualitative one.

2 Knowledge Is Uniquely Valuable BonJour (2010) observes that the apparent qualitative difference between knowledge and non-knowledge accompanies an apparent qualitative

60  Nevin Climenhaga difference in value. Knowledge appears to be valuable in a way that non-knowledge is not. Infallibilism explains this, because it is uniquely valuable to have epistemic certainty, in that this completely protects one’s belief from error (cf. Brown 2018: 7). Hence, infallibilism makes it plausible that 2

Knowledge is valuable in a way non-knowledge is not.

By contrast, fallibilism has a hard time explaining (2). Most of the necessary conditions for knowledge proposed by epistemologists are valuable in some way. So a theory of knowledge based on them can arguably explain why knowledge is valuable. However, the conditions mentioned above involve possession of some quantity that comes in degrees. And slight increases in this quantity, so long as it remains sub-maximal, constitute only quantitative differences in value. As BonJour puts it, finding “conclusive justification” for a proposition would be the best situation of all. But the claim of the weak [fallibilist] conception is that there is some specific level of justification that is less than conclusive but nonetheless transforms your cognitive situation in a much more radical way than did increases in justification up to that point (or further increases above it). Before this level is attained, you merely have a belief that is more and more likely or probable, but at that point you suddenly have knowledge. But why does achieving this specific level of justification make such a difference and what exactly is this difference supposed to amount to? … It is hard to see why such further increases are not valuable in exactly the same way, to precisely the same extent, as those that came earlier, before the supposed “magic” level was reached. (BonJour 2010: 61) Hence, fallibilism cannot easily explain why the value of knowledge is qualitatively different from the value of non-knowledge. One explanation for why fallible knowledge is qualitatively more valuable than near-knowledge is that knowing has extrinsic benefits that almost-knowing does not. If you pass your class only if you score 80% on the final exam, scoring 80% is better than scoring 79% in a way that scoring 81% is not better than scoring 80%, even though intrinsically both are mere quantitative increases. Hannon (2014) responds to BonJour’s argument by appeal to benefits of this sort. According to Hannon, we see S as knowing that P at the point when we think that S can legitimately close inquiry as to whether P: “This level of justification is cognitively valuable because it satisfies one of the platitudes about the functional role of knowledge ascription: it signals the point of legitimate inquiry closure” (Hannon 2014: 1132; cf. Kappel 2010: 191–192).

A Cumulative Case Argument  61 This response, however, is only plausible if coupled with interestrelativism. For as I argue in Section 7, if knowledge is not interest-relative, then we can only always close inquiry on whether P upon coming to know P if knowledge implies probability 1. I make similar arguments for other roles that knowledge plays, including rationalizing action and serving as evidence. If these arguments are right, then the only way for a non-interest-relativist to appeal to these roles of knowledge to explain its unique value is to adopt infallibilism. If traditional fallibilism is true, then the arbitrariness of whatever the threshold for knowledge is means that it does not have the extrinsic benefits, such as allowing us to rationally act or close inquiry, required by this response. So, the traditional fallibilist remains unable to easily explain the unique value of knowledge.

3 Gettiered Subjects Do Not Know BonJour mentions two other problems for fallibilism: the Gettier problem and the lottery paradox (BonJour 2010: 63–70; see also Reed 2012: 588–591). I will discuss the former in this section; I discuss the latter later, under the heading of closure. ‘The Gettier problem’ is a term of art, but contemporary epistemologists think of it primarily as a problem of ‘lucky’ knowledge (e.g., Zagzebski 1994; Howard-Snyder et al. 2003, Ichikawa and Steup 2014: Section 3): you give me a theory of the conditions under which S knows that P, and I’ll give you a case in which those conditions are met but S’s belief is only luckily true, and so intuitively not knowledge. But infallibilism completely eliminates the relevant kind of luck (Kirkham 1984; Kyriacou 2017: 29–30). If it is certain for you that P, it is not at all lucky that your belief turns out to be true. Indeed, in his original essay, Gettier (1963: 121) is explicit that his counterexamples to the justified true belief theory of knowledge rely on the assumption that it is possible for a person to be justified in believing a false proposition. Infallibilism thus entails 3

Subjects in Gettier cases do not have knowledge.

The last 50 years of epistemology, by contrast, show that it is difficult for fallibilists to give a theory which other epistemologists are not able to come up with “Gettier-style” counterexamples to. Fallibilists have a harder time than infallibilists allowing for (3).

4 Knowledge Is Evidence The term ‘evidence’ can be used in two related ways. First, we can say that P is evidence for Q: call this the “evidence-for” sense of ‘evidence.’ P is evidence for Q relative to some background iff P raises the probability of

62  Nevin Climenhaga Q relative to that background. The second sense is the “having-evidence” sense: we can say that S has P as evidence, or that P is part of S’s evidence. If S has P as evidence, then P, together with the other propositions in S’s evidence, determines the epistemic probability of any proposition for S. We are concerned here with the “having-evidence” sense of ‘evidence.’ Here is a simple argument that knowledge implies probability 1: 4 9 10 11

If S knows that P, P is part of S’s evidence.7 The epistemic probability of P for S is n iff n is the probability of P conditional on S’s evidence. For any evidence K apart from P, P(P|P&K) = 1. If S knows that P, the epistemic probability of P for S is 1. [from (4), (9), (10)]

(10) is analytically true for standard probability axioms. Thus, if the fallibilist wishes to accept (4), he must deny (9). While I lack the space to consider all the alternatives to (9) that might be proposed, I will make two points in defense of this premise. First, (9) is the simplest, most natural way to spell out the relation between probability and evidence. Other theories which seek to avoid the consequence that one’s evidence has probability 1 will generally be ad hoc. Second, the simplest theories that avoid this consequence will render obviously incorrect verdicts.8 For example, one might say that the epistemic probability of P for S is the probability of P conditional on all of S’s other evidence (besides P). This will wrongly deliver the result that if S has no strong evidence for P besides P itself, and P is antecedently unlikely, the epistemic probability of P for S is very low. For example, suppose that I suddenly feel a shooting pain in my foot. Relative to everything else I know at this moment, it is quite unlikely that I feel pain in my foot. Nevertheless, the epistemic probability that I feel pain in my foot is obviously not low for me. These remarks suffice to show that it is at least difficult for fallibilists to deny (9), and so difficult to accept (4) but deny (11). Now, fallibilists who deny (4) may be inclined to doubt that this claim is intuitive. But when we look at philosophers, scientists, and statisticians writing about probability in other contexts, the language they use suggests that they tacitly presuppose that knowledge is evidence. For example, Bayesians use the phrase “background knowledge” to refer to that part of an agent’s evidence which was not just now learned, or is not under explicit consideration right now (e.g., Eells and Fitelson 2000: 667–669). Here are some more specific examples (emphases mine): Probf(h) is to be a number representing the person’s personal probability for h, when he knows f; for short, his probability given f. (Hacking 1967: 313)

A Cumulative Case Argument  63 [T]o speak of the probability of an event tout court, without any qualification, does not have any concrete meaning. Rather, it must be kept in mind that probability is always relative to the state of knowledge of the person who is making the judgement. (de Finetti 1979/2008: 36) When you ask yourself how much support e gives h, you are plausibly asking how much a knowledge of e would increase the credibility of h, which is the same thing as asking how much e boosts the credibility of h relative to what else you currently know. (Howson 1991: 54) In principle perhaps, non-demonstrative inference should be based on ‘total evidence’… In practice, however, investigators must think about which bits of what they know really bear on their question. (Lipton 2001: 113) If P(A | B, C) = P(A | C), we say that A and B are conditionally independent given C; that is, once we know C, learning B would not change our belief in A. (Pearl 2000: 3) The [ideally reasoning] robot always takes into account all of the evidence it has relevant to a question. It does not arbitrarily ignore some of the information, basing its conclusions only on what remains. … The robot always represents equivalent states of knowledge by equivalent plausibility assignments. That is, if in two problems the robot’s state of knowledge is the same…, then it must assign the same plausibilities in both. (Jaynes 2003: 19) These quotes suggest that, in contexts in which philosophical issues about the nature of knowledge are not at stake, both philosophers (Hacking, de Finetti, Howson, Lipton) and scientists (Jaynes, Pearl) take for granted that if S knows that P, P is part of S’s evidence. To the extent that philosophers are inclined to deny that knowledge is evidence, then, it is most likely for theoretical reasons. In particular, inasmuch as philosophers recognize the truth of (9) and (10), they may realize that knowledge being evidence would imply that we know only what has probability 1 for us, and so reject this view because they reject this consequence of it. The infallibilist is able to explain the intuitive appeal of (4) more easily than the fallibilist. For the infallibilist can accept that (4) is true, and so hold that our ordinary thought and language which presupposes (4) reflects a tacit recognition of (4). By contrast, the traditional fallibilist can only accept (4) by denying (9), and (9) is very plausible. The fallibilist could deny (4)

64  Nevin Climenhaga and offer some error theory for why (4) is presupposed in so much ordinary thought and language. But it is not obvious what shape such an error theory would take, and I know of no fallibilist attempts to formulate one.9

5 Not-P Is Possible for S Only If S Does Not Know That P Dodd (2011) presents the following argument for the claim that if S knows that P, the epistemic probability of P for S is 1: 5 12 13 11

If S knows that P, ~P is epistemically impossible for S. ~P is epistemically impossible for S only if the epistemic probability of ~P for S is 0. If the epistemic probability of ~P for S is 0, the epistemic probability of P for S is 1. If S knows that P, the epistemic probability of P for S is 1. [from (5), (12), (13)]

(13), like (10), is analytic. Fallibilists looking to reject (11) must deny one of the other two premises. (12) is an extremely plausible claim about epistemic possibility. As Dodd (2011: 668) observes, the analog of (12) is obviously true for physical possibility and probability: P is physically impossible only if there is zero physical probability that it occurs. Dodd thinks, and I agree, that the most plausible fallibilist response to the argument is to deny (5). However, although it is not as obvious as (12), (5) is pretheoretically very plausible. In the last section, I noted that in contexts in which philosophers and scientists are not preoccupied with the nature of knowledge—and, consequently, not worried about the threat of skepticism—they tend to presuppose that knowledge is evidence. Dodd (2011: 669) similarly observes that “virtually all” writers on epistemic modals accept (5). When the threat of skepticism is not salient, the intuitive theoretical roles of knowledge come to the fore, and philosophers freely use it to analyze other concepts, such as epistemic possibility. (5) is further supported by the impropriety of concessive knowledge attributions (CKAs), that is, sentences of the form “I know that P, but maybe ~P.” CKAs are typically infelicitous to assert. For example, it is infelicitous to say “I know the Red Sox won’t win, but they might.” (5) gives a simple explanation of why CKAs are infelicitous: they are always false. If you know that P, then ~P is epistemically impossible for you.10 The infallibilist can easily accept (5). The fallibilist can only do so by denying (12). But (12) is extremely plausible. So, the fallibilist again has a harder time accepting an intuitive datum about knowledge than the infallibilist.

A Cumulative Case Argument  65

6 Knowledge Lets Us Rationally Act Here is another argument for (11): 6 14 11

If S knows that P, S can rationally act as if P. If knowledge does not imply epistemic probability 1, then it is possible that S knows that P but cannot rationally act as if P. If S knows that P, the epistemic probability of P for S is 1. [from (6), (14)]

(6) should be read as a necessary claim. Thus, if (14) is true, its antecedent is inconsistent with (6), and (11) follows. (6) is plausible because there is an apparent conceptual connection between knowledge and rational action, just as there is an apparent conceptual connection between knowledge and evidence. Principles similar to (6) are endorsed by many epistemologists, often in the context of arguing for interest-relativism about knowledge. For example, Hawthorne (2004: 30) writes that “it is acceptable to use the premise that p in one’s [practical] deliberations if one knows it and (at least in very many cases) unacceptable to use the premise that p in one’s practical reasoning if one doesn’t know it.” Similarly, Stanley (2005: 10) says that A standard use of knowledge attributions is to justify action. … To say that an action is only based on a belief is to criticize that action for not living up to an expected norm; to say that an action is based on knowledge is to declare that the action has met the expected norm.11 Interest-relativists are in a position to deny (14). This is because sometimes P is uncertain for S, but it is rational for S to act as if P because S’s evidence that P makes P sufficiently probable relative to the costs of S’s being wrong—and according to interest-relativism, this is just part of what it is to know that P. But the traditional (non-interest-relativist, non-contextualist) fallibilist is not in a position to deny (14). So (6), while usually presented as evidence for interest-relativism, is equally good evidence for infallibilism. For example, suppose that S has a choice between φ-ing and doing nothing. If S does nothing, nothing happens. If S φ-s and P is true, S gains $1. If S φ-s and P is false, then S loses some amount of money. S knows all this (with certainty). In this case φ-ing constitutes acting as if P: this is the course of action that has higher utility if P is true, and S knows this (with certainty). Now suppose that S’s belief that P has epistemic probability 0.9, that this exceeds any non-maximal probabilistic threshold for knowledge,

66  Nevin Climenhaga and that any other necessary conditions for fallibilist knowledge are met. Then S (fallibilistically) knows that P. However, suppose that we make the loss S will incur if P is false $10. Then the expected utility of φ-ing is (0.9)($1) + (0.1)(−$10) = $0.90 – $1 = −$0.10. Since the expected utility of not φ-ing is $0, S ought not φ. If the threshold for knowledge is above 0.9, then we can simply increase S’s loss if P is false. The only way to ensure that acting as if P maximizes expected utility is to give P an epistemic probability of 1. Hence, (14) is true. Thus, the infallibilist, but not the (traditional) fallibilist, can accept (6).

7 Knowledge Lets Us Rationally Close Inquiry We saw that one fallibilist response to BonJour’s argument that only infallibilism can explain the value of knowledge is that knowledge is valuable because it allows us to close inquiry. The fact that knowledge lets us close inquiry can itself be used to support infallibilism, via an argument structurally identical to the one in the last section: 7 15 11

If S knows that P, S can rationally stop inquiring whether P. If knowledge does not imply epistemic probability 1, it is possible that S knows that P but cannot rationally stop inquiring whether P. If S knows that P, the epistemic probability of P for S is 1. [from (7), (15)]

(7) is plausible because there is an apparent conceptual connection between knowledge and inquiry closure. If you come to know that P, then this settles the question of whether P for you. There is no more need for you to inquire whether P (cf. Kyriacou 2017: 30–31; Kappel 2010). Together with (15), however, (7) implies that what we know has epistemic probability 1. And the traditional fallibilist is not in a position to deny (15). For suppose that S’s belief that P has epistemic probability 0.9, that this exceeds the threshold for knowledge, and that any other necessary conditions for fallibilist knowledge are met. Then, S (fallibilistically) knows that P. However, if the costs of being wrong about P are high enough, then S ought not close inquiry with respect to P; instead, she ought to keep inquiring whether P. By continually increasing the stakes, we can run the same argument for any constant threshold that falls short of probability 1. Hence, (15) is true. Thus, the infallibilist, but not the (traditional) fallibilist, can accept (7).

8 Knowledge Is Closed under Competent Deduction Epistemologists have traditionally found plausible various closure principles about knowledge. Closure principles capture the intuitive idea that

A Cumulative Case Argument  67 we can extend our knowledge by deduction. In this section, I will argue that we can only accept what is sometimes called multi-premise closure if knowledge requires probability 1: 8 16 11

If S knows each of {P1, P2 , … Pn}, and competently deduces Q from these propositions, S knows that Q. If knowledge does not imply epistemic probability 1, then it is possible that S knows each of {P1, P2 , … Pn}, and competently deduces Q from these propositions, but does not know that Q. If S knows that P, the epistemic probability of P for S is 1. [from (8), (16)]

The proper formulation of closure principles is controversial (see Hawthorne 2005). For example, perhaps (8) fails in cases where S loses knowledge of the premises while performing the deduction. However, my argument that fallibilists cannot accept (8) also shows that they cannot accept principles that add conditions to the antecedent to avoid these kinds of counterexamples. For if we amended (8) to avoid these kinds of counterexamples, we could similarly amend (16), and it would remain plausible for the reasons adduced below. So I will stick with (8) here for simplicity’s sake. All my argument requires is that a multi-premise closure principle along these lines is plausible, even if the antecedent needs some chisholming. One common argument for (16) is from lottery cases (e.g., Hawthorne 2004: 6–7, 182; BonJour 2010: 66–70; Reed 2012: 588–590). However, these arguments require the controversial premise that, if fallibilism is true, one can know of a losing lottery ticket that it is a loser. Some fallibilists deny this, holding that, e.g., we cannot know on the basis of purely statistical evidence (Nelkin 2000; for criticism, see Christensen 2004: 62–64; BonJour 2010: 68–69). I thus prefer to rely on a different argument for (16), based on the preface case. Suppose you write a meticulously researched book consisting of 1000 claims: C1, C2 , …, C999, and C1000. Each of these claims has a high but non-maximal epistemic probability for you. Suppose that the probability of each claim is 0.99, and that conditional on your evidence, each claim is probabilistically independent of every other claim or conjunction of claims.12 This means that the probability of any conjunction of them is equal to the product of the probability of the conjuncts—e.g., P(C1&C2|K) = 0.992 = 0.9801 (where K is your evidence). It follows from this that the probability that all the claims in your book are true is equal to 0.991000 ≈ 0.00004. As such, you acknowledge in your preface that, in all probability, there are some false claims in this book, and say that they are solely the result of your own error. Since in this case the claims in your book can be about anything at all, we can assume that any conditions on knowledge of the individual

68  Nevin Climenhaga Ci are met: they are all true, your beliefs in them are reliably formed, your evidence for them is not purely statistical, and so on. It follows that you ought to be able to deduce, and come to know, that all the claims in your book are true. But clearly, you cannot know this. This follows immediately if we assume that there is any probabilistic requirement on knowledge. For example, if knowledge requires epistemic probability above 0.5, you cannot know that all the claims in your book are true. If the probabilistic threshold is lower than this, then we can always add more claims to the book to make the probability of their conjunction lower than that threshold. Likewise, if we raise the probabilistic threshold above 0.99 so as to preclude knowledge of the individual claims of the book, we can simply increase your evidence for them so that their individual probability now exceeds that threshold. The only way to avoid this result is to set the threshold to 1—i.e., to accept (11). One might hold that there is no probabilistic requirement on knowledge. In this case, one can coherently accept that you are in a position to know that all the claims in your book are true. But this is still an absurd result. It is an absurd result even if one rejects mathematical probability as an adequate formalization of plausibility.13 It just doesn’t seem that you can know a proposition like “all the claims in my book are true,” when this proposition is one you should be extremely confident is not true.14 Infallibilists can allow that knowledge can always be extended by deduction. But, I have argued, fallibilists cannot. Like the theoretical claims about knowledge considered in the previous several sections, the claim that knowledge is closed under competent deduction is not indubitable. If the costs of rejecting fallibilism are high enough, then we should deny it. But, just like these other theoretical claims about knowledge, multi-premise closure is very plausible. That a theory of knowledge entails that it is false is a cost of that theory.

9 Contextualism and Interest-Relativism In Sections 1–8 of this chapter, I presented eight intuitive claims, arguing that each is easier to reconcile with infallibilism than traditional fallibilism. In this section I will consider how well contextualism and interest-relativism can explain these intuitions. Contextualists about knowledge hold that ‘knows’ picks out different epistemic states in different contexts, so that the truth-conditions of ‘S knows that P’ depend on the context in which that sentence is uttered. Contextualists will typically hold that in contexts in which skeptical hypotheses are taken more seriously, or speakers think that getting it right is very important, ‘knows’ will pick out a harder-to-reach epistemic state than in more “ordinary” contexts. Interest-relativists hold that whether S knows that P depends partly on S’s practical interests vis-à-vis P: how much is at stake for S with regard

A Cumulative Case Argument  69 to being right about P? If it is very important for S to be right about P—if the costs of being wrong are high—then interest-relativists will typically hold that it is harder for S to know that P. Interest-relativists can offer elegant resolutions to several of the problems facing traditional fallibilists. For example, they can say that S’s belief meets the probabilistic threshold for knowledge just in case the expected utility of acting as if P is greater than the expected utility of not acting as if P. Although this threshold will vary depending on an agent’s utilities, it is not arbitrary—and interest-relativists can thus explain why (1) knowledge is qualitatively different from non-knowledge. In connecting rational belief to rational action in this way, interest-relativism can explain why (6) knowledge that P lets us act on the assumption that P. In addition, if we can close inquiry about whether P when P is probable enough on our evidence for us to act on P, interest-relativism can also explain why (7) knowledge that P lets us close inquiry on whether P. Finally, a belief that one can rationally act on and close inquiry regarding is valuable in a way that other belief is not; hence, interest-relativism can also explain (2) the unique value of knowledge. Contextualists can offer an analogous explanation of (1) that appeals to the speaker’s context.15 It is less obvious that this will explain our other intuitive data, but perhaps the contextualist could argue that the threshold for knowledge-attributions in a context varies with the threshold for what the speaker takes to be high enough (e.g.) probability to act upon/stop inquiring, and that in the speaker’s context belief that meets this threshold is conceived as valuable in a way that belief that fails to meet it is not. For the sake of argument, let us grant that the contextualist can in this way explain the unique value of knowledge and its conceptual connection to rational action and inquiry. We are granting, then, that interest-relativists and contextualists can explain (1), (2), (6), and (7). This leaves four other intuitive data: (3) subjects in Gettier cases lack knowledge, (4) knowledge is evidence, (5) knowledge that P makes ~P epistemically impossible, and (8) knowledge is closed under competent deduction. Interest-relativists and contextualists continue to face the Gettier problem. The same counterexamples that plague traditional fallibilist theories plague interest-relativist and contextualist theories, when we fix the speaker’s context and the knower’s practical stakes. So, interest-relativists and contextualists have a harder time than infallibilists explaining (3). Interest-relativists and contextualists cannot allow for multi-premise closure. For interest-relativists, this is because the property of being probable enough to rationally act upon is not closed under conjunction. For example, perhaps C1 (the first claim of your book) is probable enough for you to take it for granted in your deliberations. Even so, the conjunction C1&…&C1000 is not probable enough for you to take it for

70  Nevin Climenhaga granted in your deliberations. Suppose that your book is about to be scrutinized by a committee who will infallibly determine whether it contains errors. If it does not contain errors, you will be given a $1,000,000 cash prize. Even if each individual claim in the book is probable enough that you can act on the assumption that it is true, you cannot rationally act on the assumption that you will win $1,000,000 by, e.g., quitting your job and buying a ticket for a cruise around the world. However, if (8) were true, then, if you competently deduced that your book contains no errors, you would know that your book contains no errors, and so, according to (6), you could rationally quit your job and buy a ticket for a cruise around the world. Inasmuch as the interest-relativist is committed to (6), he must thus reject (8). The contextualist would like to save a metalinguistic version of closure, namely: 17 If in a context c, “S knows that P1,” “S knows that P2 ,” …, “S knows that Pn” all state true propositions, and S competently deduces Q from {P1, P2 , …, Pn}, then “S knows that Q” would state a true proposition in c. Suppose we just have two premises, P1 and P 2 , and let Q = P1&P2 . If the contextualist sets a certain threshold of probability for knowledge-attributions in a given context, that P1 and P2 meet that threshold does not imply that P1&P2 meets that threshold. For example, if the speaker’s threshold for knowledge-attributions is 0.8 probability, then P1 and P2 may each individually be at least this probable while their conjunction is less probable. Thus, their conjunction is not probable enough for the proposition that would be expressed by “S knows that P1&P2” to be true. The contextualist might try to rescue multi-premise closure by claiming that mention of a conjunction changes the context in such a way that the speaker can no longer truly say “S knows that Pi ” for the individual propositions Pi. If we are talking about your book, I might truly say, “You know that C1,” “You know that C2 ,” etc., but then when you ask me “And do I know C1&…&C1000?” you have shifted the context to a high-standards one in which I can no longer truly say “You know that C1,” “You know that C2 ,” etc.—nor, for that matter, “You know that C1&…&C1000.” Setting aside whether this claim about the change of contexts is plausible or not, it does not save multi-premise closure. (17) is about closure within a context, not closure across contexts. Even if your mentioning the conjunction of the claims in your book shifts the context, it is still the case that in the initial context, “You know that Ci ” would express a truth for each Ci, and that, if (17) is true, then “You know that C1&…&C1000” would express a truth in that context. But it is implausible that “You

A Cumulative Case Argument  71 know that C1&…&C1000” would express a truth in that context, even granting that it is a “low-standards” context. If standards are codified as probability, then this follows immediately given a sufficient number of claims and independence assumptions; the Ci could individually all be above the probability-threshold and the conjunction be below it. And even if standards are not codified in this way, it just is not plausible that there is any sense of ‘know’ (even one in which “You know that you are not a brain in a vat” expresses a truth!) such that the proposition expressed by “You know that C1&…&C1000” is true, when you should be extremely confident that C1&…&C1000 is false. We can see from the above conclusion that interest-relativists and contextualists cannot accept that knowledge is evidence either. For it follows from the claim that knowledge is evidence that knowledge has epistemic probability 1 (given (9), which says that the epistemic probability of P for S is n iff n is the probability of P conditional on S’s evidence). And if knowledge has epistemic probability 1, then the conjunction of everything one knows has epistemic probability 1. Hence, it ought to be knowable as well. But, since we just saw that an interest-relativist cannot allow that knowledge is closed under competent deduction, it follows that they cannot allow that the conjunction of everything one knows has epistemic probability 1. We need to be careful here. Interest-relativists will endorse (6), according to which S can know that P only if it is rational for S to act as if P. If we construe “acting as if P” as acting the way that would maximize expected utility conditional on P being true, it follows that, if the pragmatic condition on knowledge is met, one can “pretend” that P is part of one’s evidence and it will not make a difference for rational action— what it is rational for one to do will be the same whether one assigns P its actual probability on one’s evidence or one conditionalizes on it, assigning it probability 1. One can, as it were, hypothetically add P to one’s evidence for the purposes of action. It might initially seem, then, that the interest-relativist can endorse the rational permissibility (at least practically speaking) of assigning everything one knows probability 1, and hypothetically adding everything one knows to one’s evidence. This argument tacitly assumes a false closure principle of the following form: if adding P to one’s evidence does not change what actions it is rational to perform, and adding Q to one’s evidence does not change what actions it is rational to perform, then adding P&Q to one’s evidence does not change what actions it is rational to perform. Where E is your actual evidence, the first claim in your book, C1, may be probable enough for you that conditionalizing on C1&E rationalizes the same actions as conditionalizing on E. For example, if you are considering a bet on C1 that would have positive expected utility either way, you should take it either way, and if you are considering whether to quit your job and buy that cruise around the world, you should not do it either way,

72  Nevin Climenhaga because you almost certainly will not win the $1,000,000 prize even conditional on C1&E, and so that action has negative expected utility either way. The same may be true for each other Ci. But, conditionalizing on C1&…&C1000&E does not rationalize the same actions as just conditionalizing on E, for relative to C1&…&C1000&E, quitting your job and cruising around the world has (we can suppose) positive expected utility, whereas conditional just on E it still has negative expected utility. So interest-relativists cannot accept that knowledge is evidence, because interest-relativist conditions on knowledge are not closed. Contextualists cannot accept a metalinguistic analogue of the claim that knowledge is evidence for similar reasons. It follows from the claim that an utterance of “S knows that P” in a context expresses a truth only if an utterance of “P is part of S’s evidence” expresses a truth in that context that an utterance of “S knows that P” expresses a truth in that context only if an utterance of “P has epistemic probability 1 for S” also expresses a truth in that context. Contextualists must deny that the latter generally holds, because otherwise the truth of knowledge-ascriptions would (wrongly) be closed under competent deduction by the knower. Thus, they must deny that the former holds as well. If knowledge (or the truth of “S knows that P”) does not imply epistemic probability 1, then, as argued in Section 5 above, it is not the case that we are in a position to accept that S knows that P (or to truly say, “S knows that P”) only if ~P is epistemically impossible for S. So, just like traditional fallibilists, the contextualist and interest-relativist must deny the link between knowledge and epistemic modality posited by (5). Interest-relativists and contextualists can arguably avoid the arbitrary threshold problem, explain the value of knowledge, and preserve the conceptual links between knowledge, action, and inquiry. However, they cannot preserve the conceptual links between knowledge, evidence, and epistemic possibility, they still face the Gettier problem, and they cannot allow that knowledge is closed under competent deduction. So four of our eight intuitive data provide evidence against interest-relativist and contextualist versions of fallibilism in addition to traditional forms of fallibilism. One might hold that this is not as bad a result for the contextualist as the interest-relativist. This is because contextualists can accept the non-metalinguistic statement of infallibilism, that S knows that P iff P is certain. They can hold that, in the current context, ‘knows’ picks out “knowing with certainty”—even if in other contexts it picks out something that does not require certainty. And contextualists can happily endorse the above connections between evidence, epistemic possibility, and knowledge with certainty, and they can accept that knowledge with certainty is closed under competent deduction. I think, however, that the contextualist’s inability to accept the relevant metalinguistic theses connecting ‘knows,’ ‘evidence,’ ‘may,’ and so on remains a serious cost. When we fix our attention on ordinary, non-philosophical contexts, it

A Cumulative Case Argument  73 remains implausible that the proposition expressed by “I know that the bank is open, but perhaps it’s not” could ever be true. Similarly, sentences like those used by authors writing about probability in Section 4, about the connection between knowledge and evidence, might easily be uttered in low-standards contexts—e.g., contexts in which we are engaging in mundane, low-stakes reasoning and are not at all concerned about Cartesian skepticism. As for closure, one frequently touted advantage of contextualism is that it can preserve a metalinguistic formulation of single-premise closure (see, e.g., DeRose 1995: 27–29; Lewis 1996: 563–564). If this thesis is plausible, a metalinguistic formulation of multi-premise closure should also be plausible.

10 Counting the Costs I have presented a cumulative case argument for infallibilism, citing eight advantages infallibilism enjoys over fallibilism. Infallibilism can explain why knowledge is (1) qualitatively different from non-knowledge, (2) uniquely valuable, and (3) not possessed by subjects in Gettier cases. Traditional versions of fallibilism have a harder time allowing for these claims. In addition, infallibilism allows us to accept that knowledge is (4) evidence, (5) a basis for epistemic modals, (6) an inquiry stopper, (7) a basis for action, and (8) closed under competent deduction. Traditional fallibilists cannot accept any of these claims, or can do so only by denying plausible claims about the relation between probability and evidence, the relation between probability and possibility, and so on. Interest-relativists and contextualists can easily accept (1), (2), (6), and (7), but cannot as easily accept (3), (4), (5), or (8). In spite of these advantages, infallibilism is widely rejected today as an unacceptably skeptical theory of knowledge. If—as I am granting here for the sake of argument—the only facts certain for us are a priori truths and facts about our mental lives, infallibilism implies that we cannot know anything about the external world, like what we ate for breakfast this morning or what we are wearing. Infallibilists can try to offer error theories for why it wrongly appears to us that we know more than we do. BonJour (2010: 71–72) suggests that we often wrongly attribute knowledge of propositions that are not really certain because we wrongly think that they are certain: [A]n ordinary person may reasonably regard the justification for a belief as conclusive even where deeper philosophical insight shows, or at least seems to show, that it is not. Consider, for example, a case where an ordinary person seems to himself or herself to be perceiving a standard sort of “medium-sized” physical object at close range and under good conditions, and believes on this basis that such an object is there. Even if the person’s justification in such a case is not in fact conclusive (because of subtle philosophical objections having to do

74  Nevin Climenhaga with, for example, Cartesian demons or the possibility of being a brain-in-a-vat), it is easy to see how it might nonetheless seem to him or her to be conclusive, leading to a self-attribution of knowledge (and to attributions of knowledge to others whose situation is similar). A second error theory, suggested by Davis (2007), Fumerton (2010: 251), BonJour (2010: 73), and Kyriacou (this volume: section 2), is that we attribute knowledge in cases where we recognize that it is not present because we are engaging in loose talk. Just as we may say “it’s 3:00” when in fact it’s 3:02, and the difference between 3:00 and 3:02 is not important for conversational purposes, we may say “I know that the bank is open” when we do not in fact know that the bank is open, if we come close enough to knowing for conversational purposes. While fully evaluating the plausibility of these error theories is beyond the scope of this essay, their possibility shows that infallibilism’s implying that most of our ordinary knowledge-attributions are false is not conclusive reason to reject the theory: it needs to be weighed against its other advantages and disadvantages. I summarize these in Table 4.1. Evaluating the cumulative force of multiple evidences is a difficult task. Some evidences may be weightier than others, and there may be subtle dependencies among the evidences that need to be taken into account.16 Still, it seems to me that the skeptical costs of infallibilism are worth paying to reap all the explanatory benefits listed in Table 4.1. Infallibilism lets us account for eight intuitions that traditional fallibilism cannot explain. As for interest-relativism and contextualism, there are many independent objections to these views based on other counterintuitive consequences they have (e.g., Anderson 2015). When we take these together with the four intuitive data that even they cannot explain, it seems to Table 4.1 Infallibilism Traditional Contextualism fallibilist and interesttheories relativism Knowledge qualitatively different Knowledge uniquely valuable No Gettier problem Knowledge is evidence Knowledge that P makes ~P epistemically impossible Knowledge lets us act Knowledge lets us end inquiry Knowledge is closed under competent deduction We know a lot

Yes

No

Yes

Yes Yes Yes Yes

No No No No

Yes No No No

Yes Yes Yes

No No No

Yes Yes No

No

Yes

Yes

A Cumulative Case Argument  75 me that the disadvantages faced by interest-relativism and contextualism also outweigh the advantage of counting more of our ordinary knowledge-ascriptions as true. Some philosophers maintain that skepticism about knowledge in ordinary cases is so implausible that one should never accept a theory of knowledge that has skeptical implications. In allegedly “Moorean” fashion, they assert that the premises of a skeptical argument are never more certain than the ordinary knowledge claims they seek to undermine (Lewis 1996: 549; Pryor 2000: 518). But even if this is true for particular arguments, several arguments together can have a cumulative force beyond that of each particular argument, and the disjunction of their premises may indeed be more plausible than the ordinary knowledge claims Mooreans cling to. Compare: I am at a trial for my best friend, who has been accused of murdering his wife. I am almost certain that he did not kill her: I have known him my whole life and this is simply not something he would do. This is a “Moorean fact” for me. When a witness comes forward to testify that she saw him do it, I am rational in judging that it is more likely that she is lying or confused than that my friend really did it. And I am rational in making similar judgments about the credibility of any individual piece of evidence, if that is the only evidence presented. But when four more witnesses come forward, the murder weapon is found to have my friend’s fingerprints on it, he is demonstrated to have motive, means, and opportunity, and he confesses to the murder, the cumulative force of all these evidences should lead me to abandon my faith in my friend. It is not more plausible that all these witnesses are lying or mistaken, that my friend’s fingerprints got on the weapon some other way, and that he was pressured into confessing to protect someone else, and so on, than that I was mistaken about what my friend is capable of. Likewise, it is not more plausible—or so it seems to me—that the Gettier problem is irresolvable, that the unique status and value of knowledge are inexplicable, and that knowledge is not evidence, a basis for epistemic modals, a basis for action, an inquiry-closer, or closed under competent deduction than it is that I do not know that I have hands. At a certain point the evidence, or the arguments, must overturn common sense. I think that the evidences for infallibilism presented in this essay take us to that point. But even if you are not persuaded of that, I hope to have at least convinced you that infallibilism should not be summarily dismissed on account of its skeptical consequences. Its plausibility needs to be evaluated in light of all the intuitive evidence we have. This includes not only our ordinary claims to know things but also our intuitions about what kind of a thing knowledge is, and the roles that it plays in epistemology more broadly.17

76  Nevin Climenhaga

Notes 1 In Climenhaga (2021, forthcoming), I develop a fuller theory of knowledge, which I there call (capital-I) ‘Infallibilism,’ that implies what I am here calling (lowercase-i) ‘infallibilism.’ My definition here is close to Dodd’s (2011: 665), who defines infallibilism as the thesis that knowledge requires epistemic probability 1. Several other philosophers (e.g., Audi 2003: 224; Stanley 2005: 127; BonJour 2010: 57; Brown 2018) define infallibilism as the thesis that S knows that P only if S has entailing justification/evidence/ grounds for P. This definition implies that S can know that P on the basis of evidence E even if E is not itself certain for P. Neta (2011: 668–669) sees this consequence as consistent with infallibilism, but I agree with Dougherty (2011: 140–141) that a theory (such as Neta’s) that allows knowledge wholly on the basis of uncertain grounds should be considered fallibilist. 2 For more on the nature of epistemic certainty, see Climenhaga (2021, forthcoming). 3 In Climenhaga (2021), I consider several additional proposals for reconciling fallibilism with these intuitions that I omit here because of space considerations. 4 Many philosophers deny that probability 1 is sufficient for certainty in cases involving infinities. Suppose I throw an infinitely fine dart at a square dartboard the sides of which range from 0 to 1. The probability that I will hit point (0.2, 0.4) is apparently 0. So, the probability that I will not hit point (0.2, 0.4) is 1. And yet, it is not certain for me that I will not hit this point. Williamson’s view is more radical than this: he thinks that probability 1 and certainty come apart even in cases that do not involve infinite sample spaces of this sort. I compare Williamson’s theory of knowledge with the form of infallibilism I defend in Climenhaga (2021). 5 Brown (2018) argues that combining infallibilism with a less skeptical view about the scope of our knowledge leads to a variety of counterintuitive results. She also argues that non-skeptical infallibilists have just as hard a time as fallibilists accepting several of the intuitive claims I go on to discuss, including (1), (6), and (8). For the most part, though, these arguments leave open the possibility that skeptical infallibilists can accept these claims. 6 For an overview of theories of knowledge based on these and other popular conditions, see Shope (1983) and Ichikawa and Steup (2014). 7 I sometimes speak loosely of (4) as the claim that knowledge is evidence, even though, strictly speaking, what it says is that what is known is evidence. In this I follow the language of Williamson and others who endorse (4). 8 I am setting aside Jeffrey conditionalization, because this approach implies radical subjectivism about probabilities: see Williamson (2000: 216). 9 Brown (2018: ch. 4) argues that (4) has counterintuitive consequences when combined with a non-skeptical view about the scope of our knowledge. But this does not explain the intuitive plausibility of (4) itself; and to the extent that this remains unexplained, it is evidence for a skeptical form of infallibilism over a non-skeptical form of fallibilism. 10 For a fallibilist error theory of the infelicity of CKAs that denies (5), see Dougherty and Rysiew (2009, 2011). 11 Stanley and Hawthorne’s discussions suggest that knowledge is not only sufficient but also necessary for rational action. However, while this necessity claim may be plausible for using a proposition as a premise in practical reasoning, it is not plausible for acting as if a proposition is true. As I am using

A Cumulative Case Argument  77

12

13

14

15

16

17

this phrase, φ-ing counts as “acting as if P” just in case φ-ing is the optimal course of action conditional on P being true. In cases where there are no deontological moral considerations at play, I assume that this is equivalent to φ-ing maximizing expected utility conditional on P being true. So understood, if one can rationally use P as a premise in practical reasoning, one can rationally act as if P—so if knowledge that P is sufficient for the former, it is sufficient for the latter. On the other hand, even if knowledge that P is necessary for rationally using P as a premise in practical reasoning, it is not necessary for rationally acting as if P. This is because it can be rational to act as if P when P is (as it turns out) false, and so not known. (In the example below, even if P were false, it would be rational for S to φ if the loss for P being false was reduced to $1, provided that the epistemic probability of P for S remains at 0.9.) This is an unlikely stipulation to hold true in ordinary cases. In ordinary books, the claims made will be positively relevant to each other. However, even with some positive relevance, given enough claims we can still make the probability of their conjunction arbitrarily low. In response to BonJour’s argument that lottery cases show that fallibilists must reject closure, Pritchard and Turri (2014) suggest that rejecting probability theory as measuring justification lets fallibilists keep closure, writing, “If we pair the [fallibilist] conception of knowledge with a qualitative [i.e., non-probabilistic] model of justification…, then it’s no longer clear that the resulting view will fail to respect the closure of knowledge under conjunction, because the rules of probability theory employed in BonJour’s argument don’t obviously apply to the qualitative categories.” For further counterintuitive consequences of the claim that one can know that all of the claims in one’s book are true, see Christensen (2004: ch. 3.3– 3.4). (Christensen’s focus is on rational belief, but his remarks are equally applicable to knowledge.) For responses to other common objections to the Preface Paradox, see Christensen (2004: ch. 3.1) and Easwaran and Fitelson (2015: 65–70). Strictly speaking, when I talk about whether contextualists can accept (1)– (8), I mean whether they can accept meta-linguistic analogs of these claims. For example, in saying that they can accept (1), what I mean is that they can accept that there is a qualitative difference between the state picked out by “S knows that P” in a context and the state picked out by “S justifiably and truly believes that P but the degree of S’s justification is just shy of that required for knowledge” in that same context. I make these meta-linguistic formulations explicit when needed below, and at the end of this section discuss the relevance of whether contextualists can accept the non-metalinguistic versions of (1)–(8). In Climenhaga (2021), I address these complexities by presenting a Bayesian version of the cumulative case presented informally here, in which the weight of and dependencies among different evidences are formally quantified. I also consider some additional objections to infallibilism, and argue that they don’t substantially add to the evidential weight of our ordinary intuitions and attributions of knowledge. I am grateful to Robert Audi, Blake Roeber, Daniel Immerman, Al Hájek, Mark Satta, Greg Stoutenburg, Christos Kyriacou, and an anonymous reviewer for feedback on earlier drafts of this paper, and to audiences at Western Michigan University, Deakin University, the Society for Exact Philosophy, and Pacific APA for feedback on oral presentations of this work.

78  Nevin Climenhaga

References Anderson, Charity (2015). “On the Intimate Relation of Knowledge and Action,” Episteme 12: 343–353. Audi, Robert (2003). Epistemology: A Contemporary Introduction to the Theory of Knowledge, Second Edition (Routledge). BonJour, Laurence (2010). “The Myth of Knowledge,” Philosophical Perspectives 24: 57–83. Brown, Jessica (2018). Fallibilism: Evidence and Knowledge (Oxford: Oxford University Press). Christensen, David (2004). Putting Logic in Its Place (Oxford: Oxford University Press). Climenhaga, Nevin (2021). Knowledge and Certainty: A Defense of Infallibilist Epistemology. Manuscript. Climenhaga, Nevin (forthcoming). “How Infallibilists Can Have it All,” The Monist. Davis, Wayne (2007). “Knowledge Claims and Context: Loose Use,” Philosophical Studies 132: 395–438. De Finetti, Bruno (1979/2008). “Lecture VI: Bayes’ Theorem,” in Philosophical Lectures on Probability: Collected, Edited, and Annotated by Alberto Mura (Springer), pp. 31–45. DeRose, Keith (1995). “Solving the Skeptical Problem,” The Philosophical Review 104: 1–52. DeRose, Keith (1999). “Contextualism: An Explanation and Defense,” in J. Greco and E. Sosa (eds.), The Blackwell Guide to Epistemology (Malden, MA: Blackwell), pp. 185–203. Dodd, Dylan (2011). “Against Fallibilism,” Australasian Journal of Philosophy 89: 665–685. Dougherty, Trent (2011). “Fallibilism,” in Duncan Pritchard and Sven Bernecker (eds.), The Routledge Companion to Epistemology. Dougherty, Trent, and Patrick Rysiew (2009). “Fallibilism, Epistemic Possibility, and Concessive Knowledge Attributions,” Philosophy and Phenomenological Research 78: 123–132. Dougherty, Trent, and Patrick Rysiew (2011). “Clarity about Concessive Knowledge Attributions: Reply to Dodd,” Synthese 181: 123–132. Dutant, Julien (2016). “How to Be an Infallibilist,” Philosophical Issues 26: 148–171. Easwaran, Kenny, and Branden Fitelson (2015). “Accuracy, Coherence and Evidence,” in T. Szabo Gendler and J. Hawthorne (eds.), Oxford Studies in Epistemology, vol. 5 (Oxford: Oxford University Press). Eells, Ellery, and Branden Fitelson (2000). “Measuring Confirmation and Evidence,” Journal of Philosophy 97: 663–172. Fumerton, Richard (2010). “Fencing out Pragmatic Encroachment,” Philosophical Perspectives 24: 243–253. Gettier, Edmund L. (1963). “Is Justified True Belief Knowledge?” Analysis 23: 121–123. Hacking, Ian (1967). “Slightly More Realistic Personal Probability,” Philosophy of Science 34: 311–325. Hannon, Michael (2014). “Fallibilism and the Value of Knowledge,” Synthese 191: 1119–1146.

A Cumulative Case Argument  79 Hawthorne, John (2004). Knowledge and Lotteries (Oxford: Oxford University Press). Hawthorne, John (2005). “The Case for Closure,” in Matthias Steup and Ernest Sosa (eds.), Contemporary Debates in Epistemology (Malden, MA: Blackwell), pp. 40–55. Howard-Snyder, Daniel, Frances Howard-Snyder, and Neil Feit (2003). “Infallibilism and Gettier’s Legacy,” Philosophy and Phenomenological Research 66: 304–327. Howson, Colin (1991). “The ‘Old Evidence’ Problem,” The British Journal for the Philosophy of Science 42: 547–555. Ichikawa, Jonathan Jenkins, and Matthias Steup (2014). “The Analysis of Knowledge,” in The Stanford Encyclopedia of Philosophy, . Jaynes, E.T. (2003). Probability Theory: The Logic of Science (Cambridge: Cambridge University Press). Kappel, Klemens (2010). “Expressivism about Knowledge and the Value of Knowledge,” Acta Analytica 25: 175–194. Kirkham, Richard (1984). “Does the Gettier Problem Rest on a Mistake?” Mind 93: 501–513. Kyriacou, Christos (2017). “Bifurcated Sceptical Invariantism: Between Gettier Cases and Saving Epistemic Appearances,” Journal of Philosophical Research 42: 27–44. Kyriacou, Christos (this volume). “Moderate Pragmatic Skepticism, Moorean Invariantism and Attributions of Intellectual Virtue/Vice,” in Christos Kyriacou and Kevin Wallbridge (eds.), Skeptical Invariantism Reconsidered (Routledge). Lewis, David (1996). “Elusive Knowledge,” Australasian Journal of Philosophy 74: 549–567. Lipton, Peter (2001). “Is Explanation a Guide to Inference? A Reply to Wesley Salmon,” in Giora Hon and Sam Rakover (eds.), Explanation: Theoretical Approaches and Applications (London: Kluwer), pp. 93–120. Nelkin, Dana (2000). “The Lottery Paradox, Knowledge, and Rationality,” The Philosophical Review 109: 373–409. Neta, Ram (2011). “A Refutation of Cartesian Fallibilism,” Noûs 45: 658–695. Pearl, Judea (2000). Causality: Models, Reasoning, and Inference (Cambridge: Cambridge University Press). Pritchard, Duncan, and John Turri (2014). “The Value of Knowledge,” in The Stanford Encyclopedia of Philosophy, . Pryor, James (2000). “The Skeptic and the Dogmatist,” Noûs 34: 517–549. Reed, Baron (2012). “Fallibilism,” Philosophy Compass 7: 585–596. Shope, Robert (1983). The Analysis of Knowing (Princeton: Princeton University Press). Stanley, Jason (2005). Knowledge and Practical Interests (Oxford: Oxford University Press). Stanley, Jason (2008). “Knowledge and Certainty,” Philosophical Issues 18: 35–57. Williamson, Timothy (2000). Knowledge and Its Limits (Oxford: Oxford University Press). Zagzebski, Linda (1994). “The Inescapability of Gettier Problems,” Philosophical Quarterly 44: 65–73.

5

Skeptical Invariantism, Considered Gregory Stoutenburg

1 Motivating Infallibilist Invariantism Is there a good argument for skeptical invariantism? Skimming through the last few decades of literature on knowledge leaves the distinct impression that the following are data that must be explained by any theory of knowledge, and being the simplest available explanation makes infallibilism look like the obvious choice. 1 2 3 4 5 6 7

Knowledge attributions can be used to assure others that a proposition is true1 Knowledge is closed under known entailment Knowledge cannot be gained through merely probabilistic inference Knowledge is incompatible with luckily true belief2 Concessive knowledge attributions sound contradictory3 Knowing that p entails that it is impossible that not-p It is very often appropriate to attribute knowledge4

The hypothesis that knowledge is infallible so easily explains (1)–(6). By “knowledge is infallible,” I mean that S knowing that p is the very same state as S believing that p with infallible justification. So, knowledge requires (is) belief with infallible justification.5 And by “infallible,” I mean not fallible: for S’s justification for p to be infallible just means that S’s justification is sufficient for the truth of p (on the definition I prefer).6 I will call this view infallibilist invariantism, for the sake of distinguishing it from alternative versions of skeptical invariantism. I will now more explicitly connect this kind of infallibilism with skepticism and invariantism. If knowledge is infallible, that explains (1) because if I tell you that I know that p, I have thereby attested that p is not possibly false. It explains (2) because if someone has justification that is sufficient for the truth of a belief, then there is no other true proposition that is known to be entailed by the former for which the person does not have (or cannot very easily acquire) infallible justification. It explains (3) because mere probability leaves open some possibility of error, and so precludes establishing a proposition’s truth. And so on. Just looking at (1)–(6), infallibilism at

Skeptical Invariantism, Considered  81 least initially looks like the simple and obvious choice for an explanatorily virtuous theory of knowledge. And then comes (7). Fallibilists think that failing to satisfactorily account for (7) is decisive against a theory of knowledge in a way that denying any of the others is not (cf. Dretske 1970; Nozick 1981; Roush 2012). For instance, in her recent monograph, Brown (2018) argues for fallibilism and thus explicitly against (2) closure, (3) that knowledge cannot be gained through merely probabilistic inference, and (5) the universal falsity of concessive knowledge attributions. Defending an account that favors (7) over all of (1)–(6) would make little sense if infallibilism had a plausible non-skeptical version, which many philosophers may otherwise favor (Dutant 2015). But when we entertain a concept of knowledge that actually applies to subjects in our world, that concept can only be called “infallible” by ignoring or disqualifying a significant number of ways that our beliefs may be false (Dutant 2016; Lewis 1996; Stoutenburg 2019; Williamson 2000). Do we really want to say with disjunctivists and safety-infallibilists that I can have infallible—not possibly false—knowledge that I have hands, if the experience of seeing my hands is a fundamentally different experience from the indistinguishably similar experience I would have if I were deceived? That when I am not deceived, the ‘possibility’ that I am now being deceived is no possibility at all? It is with good reason that philosophers think infallibilism implies skepticism. For similar reasons, I think the only version of infallibilism worthy of the name is invariantist. That a subject is not thinking about a particular possibility, or that the stakes of false belief are low, or that an attributor and subject do not share contexts of assessment, are irrelevant to whether some justification is sufficient for the truth of a proposition. Now it should be clear why infallibilism implies both skepticism and invariantism, and it is this kind of skeptical invariantism that is under discussion. In this chapter, I consider the prospects for a skeptical version of infallibilism. For the reasons given above, I think skeptical invariantism has a lot going for it. However, a satisfactory theory of knowledge must account for all of our desiderata, including (7) that our ordinary knowledge attributions are appropriate. This last part will not be easy for the infallibilist invariantist. Indeed, I will argue that it is much more difficult than those sympathetic to skepticism have acknowledged, as there are serious problems with regarding paradigmatic, typical knowledge attributions as loose talk, exaggerations, or otherwise practical uses of language. So, I do not think the pragmatic story that skeptical invariantism needs is one that works without a supplemental error theory of the sort left aside by purely pragmatic accounts of knowledge attributions (Butchvarov 1970; Davis 2007; Kyriacou n.d.; Schaffer 2004; Stoutenburg 2016). In its place, I will offer a

82  Gregory Stoutenburg compromise pragmatic and error view that I think delivers everything that skeptics can reasonably hope to get.

2 Methodological Concerns for Two Non-Skeptical Views Before considering how skeptical invariantism can account for the appropriateness of knowledge attributions, I want to briefly argue that the two most popular competitors to skeptical invariantism—epistemic contextualism and strict moderate invariantism—have a serious methodological problem, one that generalizes to any non-skeptical theory of knowledge attributions. The problem is that if either of those theories is true, we are not justified in believing that they are true.7 When we think about a concept of philosophical interest, like what knowledge is aside from what we happen to know, we rely on what we find intuitive, our language-use dispositions, and knowledge of meanings to locate the thing that we want to examine. But how is this possible? The traditional philosophical method of considering our responses to thought experiments and hypothetical scenarios more generally can work only if we are competent speakers with concept-use dispositions that are sensitive (if fallible) to facts about the concepts that we wish to examine (Stoutenburg 2016). For example, because philosophers are competent users of the concept knowledge, when Gettier’s counterexamples to the fallible justified-true-belief analysis were proposed, (almost) everyone recognized that their own concept of knowledge is not the justified-true-belief concept. If philosophers had not recognized the mismatch between that account and their own concept of knowledge, then it would be a mystery as to why learning of Gettier’s examples prompted any kind of revision. This observation implies a constraint on which philosophical theories can be justifiably believed. Constraint: I can justifiably believe my own judgment about a hypothetical scenario only if I have no reason to believe that I am likely to be mistaken in my thought and judgment about the hypothetical scenario. Consider: If I have reason to believe that I am pathetically innumerate, and we attempt to split a restaurant bill and calculate an appropriate tip, I am at least typically not justified in believing the results of my probably unreliable calculation. Now I will apply this modest constraint. The most popular accounts of knowledge attributions are probably epistemic contextualism and strict moderate invariantism, so I will focus on them. Cohen’s Airport Case (1999) can be used to introduce both. Airport Case: Two passengers, Mary and John, are at the airport wondering whether the flight they plan to board has a layover in Chicago. They observe another passenger, Smith, look at his own itinerary and claim to himself that he knows the flight stops in Chicago.

Skeptical Invariantism, Considered  83 Mary and John have an important meeting in Chicago. Mary says to John, “How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute.” Mary and John conclude that Smith does not know there is a layover in Chicago and decide to answer their own doubts by checking with a gate attendant for more information about the flight. Smith says, (1) “I know that the flight stops in Chicago.” Mary says, (2) “Smith does not know that the flight stops in Chicago.” Contextualism says (1) and (2) are both true. On the surface, (1) and (2) contradict one another. Contextualism resolves this by claiming that “S knows that p” expresses different contents in different contexts of use. So, (1) and (2) are not genuine contradictions. They appear to be— which explains why Mary (according to contextualism, falsely) believes that standards applicable in her context support a knowledge-denial that contradicts (1). It is strange that Mary would think this if contextualism were true. If it really were true that our knowledge-attributing utterances pick up some of their content from the context in which we speak or think, and we were unaware of this fact, then we speakers would typically be unaware of the contents of our knowledge attributions (cf. Abath 2012).8 But this violates the constraint given above. The usual methodology has it that we consider judgments about hypothetical scenarios and try to sort them into a systematic theory. But if contextualism is true, we rely upon damaged instruments: like Mary, we should not be confident about what our judgments are, let alone that those judgments (whatever they are) are true. So, if contextualism is true, the error theory it requires to explain why we think non-contradictory knowledge attributions are contradictory also implies that we should not believe contextualism is true. Strict moderate invariantism says (1) is true and (2) is false, and it may be even more obviously self-undermining than contextualism is, as I have argued elsewhere (2017). As Smith and Mary have equally good evidence, it is assumed that Mary knows the flight stops in Chicago, but falsely denies it for some reason. Mary’s knowledge-denial seems correct, so strict moderate invariantism must explain it. The standard explanation, given by Rysiew (2001), holds that “S knows that p” expresses that S truly and justifiably believes that p and S can eliminate all likely not-p possibilities; “S knows that p” conveys that S can eliminate all salient not-p possibilities (which may or may not be relevant). Applied to the airport case, after the raising of the misprint possibility, Mary says (2) because she does not want to convey that Smith can eliminate the misprint possibility. For whatever reason, on this account Mary asserts the falsehood (2) instead of a relevant truth (“Smith knows the flight stops in Chicago” or “the misprint possibility is irrelevant to whether

84  Gregory Stoutenburg Smith knows the flight stops in Chicago”), in order to avoid implicating an irrelevant falsehood (“Smith can eliminate the misprint possibility”). The cause of the confusion in the exchange should not be overlooked: it is that Mary is unaware that the misprint possibility is irrelevant to Smith’s knowledge and her own, and consequently unaware of the truth-values of ordinary knowledge attributions in ordinary contexts. I assume that philosophers are ‘ordinary’ speakers, too. We access truths about what counts as knowledge only indirectly, through considered judgments about hypothetical scenarios. So, there is a serious methodological problem with claiming, as strict moderate invariantism must, that we are fundamentally confused about what is and is not relevant to knowledge, and also that we can use our thoughts about what is and is not relevant to knowledge to discover that strict moderate invariantism is true.9 One might worry that a similar argument works against skeptical invariantism. After all, we often sincerely assert that we know that something is true, but skepticism implies that such claims are false. While any skeptical view surely must meet that demand, it is important to recognize that the arguments I just gave point to a general methodological worry that uniquely affects the theories of knowledge attributions highlighted above. Those theories appear to undermine whatever support they receive from our ordinary knowledge-attributing practices by calling into question the reliability of the intuitions that govern those same practices. The primary way that skeptical invariantists have attempted to explain the appropriateness of our knowledge thought and talk is by offering a pragmatic account of false knowledge attributions, which by design attempts to preserve semantic competence at the expense of making ordinary knowledge attributions come out true. If our willingness to say and think “S knows that p” is similar to our willingness to exaggerate or speak loosely, for example, then it would be a mistake to think that skeptical invariantism is methodologically questionable in the ways highlighted above.

3 Skeptical Invariantism and Knowledge-Talk: Error and Pragmatic Solutions However, I will now argue that pure error or pragmatic explanations of our tendency to make false knowledge attributions face serious problems.10 A pure error account explains why we fail to recognize that our knowledge attributions are false, but does nothing to account for how our reliance on “knowledge” can be appropriate. Purely pragmatic accounts imply that speakers do not sincerely attribute knowledge, which is plainly false. In this section, I evaluate a few accounts of knowledge attributions that are compatible with or explicitly motivated by skeptical concerns. In the end, I think the skeptical invariantist is forced to accept a compromise between a pragmatic account and an error theory.

Skeptical Invariantism, Considered  85 I propose a compromise view that fits well with an independently plausible conception of how we arrive at a descriptive theory of a habitual practice. We skeptical invariantists must accept less than what we hoped for, but we will have all that we need. 3.1 Entailment Error Theory Unger (1971, 1975) argued that we falsely believe that we know some things because our false unqualified knowledge claims entail true qualified knowledge claims (1975, pp. 51–52). Since “S knows that p” entails “S is close enough to knowing that p for the purposes at hand,” we fail to notice that our knowledge claims are false because they entail true claims about being close enough to knowing. In Unger’s words, [S]uppose that…you falsely believe you know that there are elephants. As before, there is a true thing which is entailed by what you falsely believe, and which we should notice. The true thing here, which you presumably do not believe, is this: That you are in an intellectual (or ‘epistemic’) position with respect to the matter of whether there are elephants which is such that, for practical purposes, it makes no difference whether you know there are elephants or whether you are in that intellectual position with respect to the matter that you actually are in. (1975, p. 52) What counts as ‘close enough to knowing for the purposes at hand’ varies with context. Since the standard for being close enough to knowing for practical purposes is usually low enough to be satisfied, we do not notice that “S knows that p” is strictly speaking false. A virtue of Unger’s account over some alternative skeptical invariantist views is that it accepts that speakers usually do intend for their knowledge attributions to be taken truly and literally. However, it faces two significant shortcomings. The first concerns the semantics of knowledge attributions that lead to skeptical invariantism. Traditionally, radical knowledge skeptics have argued that we lack knowledge because our epistemic justification is insufficient for knowledge. If such a skeptic claims that “S knows that p” expresses something like S’s belief that p is not possibly false given the way that S formed it and insists that the only legitimate basis for belief consists in awareness of the intrinsic character of one’s occurrent mental states, as a Cartesian internalist would, then it is doubtful that we are close enough to knowing anything for any purposes. So, it is false that “S is close enough to knowing that p for practical purposes.”11 The second objection is that the view does not even attempt to provide an account of how our knowledge attributing behavior could be

86  Gregory Stoutenburg appropriate: it only explains why we do not recognize that we lack knowledge. An account of appropriateness needs to show that, given a skeptical semantics for knowledge attributions, there is a pragmatic account of how our knowledge talk is ‘okay.’ The Error Theory does not even attempt this.12 3.2 The Loose Talk Theory Davis (2007, 2015) argues that for many ordinary purposes we are content to utter approximate truths without regard for whether or not what we say is literally true, and knowledge attributions are an example of this pragmatic phenomenon. An example from Davis: A. When the scoop comes up empty in the coffee jar, I yell to my wife, “The coffee is all gone.” B. When my son comes down for breakfast a few minutes later, he announces that he needs a few coffee grounds for his science project, and then asks, “Is the coffee really all gone?” I say with no embarrassment, “No, there may be enough for you.” What I say in A contradicts what I say in B. But what I mean in A is that the coffee is close enough to being all gone for the purpose at hand—making coffee. That is, the coffee is close enough to being all gone to count as all gone for the indicated purpose. This is not what the sentence I use means. But by saying that the coffee is all gone, I convey the less precise thought indicated. By ignoring irrelevant detail, I make my point more effectively. When my conversational purposes change in context B, I use the term more strictly. (Davis 2007, pp. 406–407) The picture Davis gives us is this: A speaker utters U loosely when A intends to communicate that P, A believes that the content of U is close to the content of P, A recognizes that A’s audience will understand P by A’s uttering of U, and A does not accept the semantic content of U.13 On this view, speakers are not committed to the truth of their knowledge claims, only to some relevant proposition in the semantic neighborhood. It is instructive to compare Davis’s remarks about loose use with what he says about sloppy use. One uses a term sloppily if it is used “strictly without taking proper care to verify that the term does strictly apply” (p. 410). It would be sloppy use if Davis had answered his son’s question with “yes, the coffee is all gone” without bothering to check that the coffee is indeed gone. So, sloppy use is speech with a disregard for truth. Given that loose use and sloppy use are distinct, we can conclude that a speaker uses U loosely only if the speaker is aware that U is false but is not committed to U’s truth (otherwise the speaker would just be guessing or lying). Paradigm knowledge attributions are ones that the speaker believes are true, however. We typically use knowledge attributions literally,

Skeptical Invariantism, Considered  87 intending to communicate (among other things) that so-and-so knows such-and-such. Accounts of knowledge attributions must accept that speakers usually are committed to the truth of their knowledge attributions, even if their communicative purposes are not focused precisely on epistemic concerns.14 When it comes to skepticism about privileged domains of belief, like our knowledge of arithmetic equalities, how things phenomenally appear, and so forth, entertaining skeptical possibilities and denying knowledge strikes us as a discovery, not something that we have been assuming all along (Dinges 2016; Hawthorne 2004, pp. 164–165). As Dinges argues, if speakers were typically speaking loosely with their attributions of knowledge, we would expect speakers to reply to challenges to their knowledge attributions with “You know what I meant!” or similar, to communicate that the speaker did not intend to be committed to the literal truth of the utterance (2016, pp. 2588–2592). As we will see, this problem affects other pragmatic accounts of knowledge attributions. Unless we are convinced that speakers typically make knowledge attributions while believing that they are saying something false and not committed to the truth of their utterance anyway, then a purely pragmatic account of the appropriateness of our knowledge attributions cannot help skeptical invariantists reconcile skepticism with our knowledge-attributing practices. 3.3 The Exaggeration Theory Some of what we say that we know is plausibly a kind of exaggeration. Take knowledge claims about the distant future—“We know that the population of Earth will be 11 billion by 2100”—or knowledge attributions made for exhortation—“I know that you’ll ace the test!” Butchvarov (1970) argued that knowledge attributions are exaggerations: [T]he word know (and related expressions) falls into a class of words that quite naturally lend themselves to habitual yet perfectly justifiable exaggerated uses. One subgroup of such words consists of words whose very utterance is likely to have considerable practical import. Another subgroup consists of words referring to standards. Know belongs to both. (Butchvarov 1970, p. 54) He adds that speakers make hyperbolic knowledge attributions for practical purposes, even while aware that literally knowing something is rare: [P]eople very often claim to know when they do not, although they are not ignorant of the meaning of know or of the circumstances in which they use the word. By using the word, even though

88  Gregory Stoutenburg illegitimately, one can encourage important actions, gain respect and admiration, cause attitudes one regards as desirable, and… To deplore and attempt to eliminate such exaggerated uses of know would be a practical mistake. To regard these uses as paradigms of a sense of know would be an intellectual mistake. (Butchvarov 1970, p. 55) Later, Schaffer (2004) supplemented the Exaggeration account with Grice’s theory of conversational implicature (Grice 1975).15 In Schaffer’s telling, knowledge attributions express that a subject meets a very high epistemic standard that is typically unmet. Hearers, accepting this same standard, hear a knowledge attribution as straightforwardly false. Expecting that the speaker intends to comply with the Cooperative Principle and its maxims, including Quality (“Try to make your contribution one that is true”), infer that the speaker intends to comply with the Cooperative Principle with what the speaker means. Hearers thus infer that the speaker intends to communicate something else: perhaps that the subject is close enough to knowing for the present purposes. But knowledge attributions usually do not strike us as deliberate overstatements, unlike uncontroversial examples of hyperbole. I say “it took me forever to get here” as I walk in late to the meeting, very obviously exaggerating with “forever.” “The swimming pool is a mile long” behaves similarly.16 When we hear an utterance used hyperbolically it sounds false because it sounds like an overstatement. Contrast that with ordinary knowledge attributions: a new faculty member needs help logging into the department’s scanner, and I say, “Come with me, I know the code.” A friend asks what time the teeball game starts, and after a reflective pause, I say, “Oh, I know: 6:30.” It is extremely hard to accept the idea that hearers would take these straightforward knowledge claims as deliberate overstatements and thus as sparking a search for a true implicature. But without hearers taking knowledge attributions as straightforward overstatements, no search for an implicature begins.17 Schaffer anticipated this objection and replied that expressions that are usually used hyperbolically often do not strike us as hyperbolic (2004, fn. 3). But that reply gets the argumentative burden exactly backward. A principled account would begin with an understanding of hyperbole and argue that knowledge attributions are an instance of the kind. Instead, Schaffer’s reply shows that knowledge attributions are not like uncontroversial examples of hyperbole at all. 3.4 The Disguised Conditionals Theory Everett (2006) proposed that our ordinary knowledge attributions are disguised conditional claims. We utter sentences like “I know that p,”

Skeptical Invariantism, Considered  89 but the semantic content of the utterance (and the speaker’s intention behind uttering it) is a disguised conditional of the form, “I know that if I am not undetectably deceived, then p.” Undoubtedly, we do sometimes offer conditional statements as revisions of our claims to know: The Airport Case, Conditional Version 1: SMITH: MARY: SMITH:

I know that the flight stops in Chicago. How do you know that the flight stops in Chicago? The itinerary could contain a misprint. I just meant that I know that if there is no misprint or other strange error, then the flight stops in Chicago.

But the Conditionalist goes further than this. It is not just that we sometimes back off from straightforward knowledge attributions and offer a conditional knowledge attribution in retreat. The claim is that straightforward, unqualified knowledge attributions semantically express disguised conditional propositions. The Conditionalist claims that “S knows that p” is equivalent to “S knows that (if S is not undetectably deceived, then p).” If we never assert that we have unconditional knowledge, then many skeptical worries are irrelevant to ordinary (conditional!) knowledge. If I claim to know that ‘if p then q’ and an interlocutor objects that perhaps not-p because not-q, those concerns have no bearing on my claimed conditional knowledge of ‘if p then q.’ Skeptical concerns are bracketed as ‘undetectable’ possibilities that do not threaten ordinary (conditional) knowledge. One might claim this ability to bypass skepticism as a feature of the view. There is a significant associated cost, however. In the Smith-Mary exchange above, it is hard to accept that Smith really asserted the conditional claim when he made his first utterance. It is more plausible that when challenges are made to our knowledge attributions and we cannot rebut the challenge, we hedge by offering a conditional replacement for our original claim. But if that is right, then speakers are not making conditional claims in the first place. If the theory were true, exchanges like the following would be perfectly acceptable: The Airport Case, Conditional Version 2: SMITH: MARY: SMITH:

I know that the flight stops in Chicago. How do you know that the flight stops in Chicago? The itinerary could contain a misprint. Like I said the first time, I know that if there is no misprint or other strange error, then the flight stops in Chicago.

If the Disguised Conditionals Theory is true, Smith should sound like he is repeating himself. Instead, it sounds like he does not want to acknowledge that Mary undermined his knowledge claim.18,19

90  Gregory Stoutenburg

4 Combining Pragmatics and Error Theory Each account just considered failed either by not explaining the appropriateness of knowledge attributions, by denying that speakers assert knowledge claims, or by implying that speakers take skepticism for granted as obviously true. I propose a new skeptical invariantist account of knowledge attributions that avoids these failures. The view incorporates some insights from the positions just criticized. There is something important about the idea that speakers often make false claims about what they know without realizing it, as the Error Theory suggests. There is also an important truth in the suggestion that the appropriateness of much knowledge-talk involves an important pragmatic element. The view I propose accommodates both. Call it Pragmatic Error Skepticism. Pragmatic Error Skepticism claims that “S knows that p” (invariantly) expresses that S has infallible justification for believing that p; that ordinary speakers implicitly accept this standard and that it implies some form of knowledge skepticism; and that “S knows that p” is regularly and appropriately used because it communicates a proposition that is practically relevant. In the rest of this section, I will show that pragmatic error skepticism avoids the problems of other accounts while plausibly explaining the appropriateness of our knowledge-attributing behavior. 4.1 Appropriateness in Ordinary Contexts A desideratum for any theory of knowledge, including skeptical invariantism, is to show that ordinary knowledge attributions are at least usually appropriate. Pragmatic Error Skepticism claims that speakers implicitly accept that “S knows that p” expresses demanding standards that are rarely met, and that speakers nevertheless use “S knows that p” for ordinary, practical purposes. Showing how this works is possible using Grice’s maxims in a straightforward way. Distinguishing Relevant Alternatives: Landon and Jill are looking at a map of the wildlife sanctuary, searching for the zebras. Thanks to a convoluted map key, visitors looking for the zebra exhibit often wind up at the tiger exhibit, and vice versa. Jill thinks they should go right—but she is not sure—and Landon thinks they should go left—but he is not sure, either. Fred overhears their conversation and says, “I know where the zebras are. They are [pointing] that way, to the right.”

Skeptical Invariantism, Considered  91 Fred’s knowledge attribution semantically expresses that he meets an extremely demanding epistemic standard, which at the level of what is said is false and thus violates Quality (“Try to make your contribution one that is true”).20 Assuming that Fred intends to satisfy the conversational maxims using implicature rather than with what he said, Landon and Jill infer that Fred means to communicate something that satisfies Quality, which is that he can distinguish the things they are looking for—zebras—from an alternative that is conversationally relevant—the tigers. Furthermore, Jill and Landon infer that by implicating that he can distinguish these alternatives, Fred intends to direct them to the exhibit that they are looking for.21 More schematically, in this example, Fred makes an ordinary knowledge attribution, saying “I know where the zebras are.” According to Pragmatic Error Skepticism, Fred’s utterance expresses the proposition that Fred has infallible justification for believing that the zebras are [demonstrating] that way. Landon and Jill hear this proposition as false as thus as violation of Quality. Thus, they search for a contextually relevant, true proposition that satisfies all of the maxims, including Quality. They arrive at the proposition that Fred can distinguish the zebras from the tigers, and further infer from this that by expressing himself in this way, Fred is providing directions. The general pattern here is that when a speaker utters “S knows that p” hearers will take the utterance to violate Quality. Hearers assume that speakers intend to comply with Quality and the other maxims at the level of what is meant, so they infer a true and relevant implicature. 22 This way of speaking allows for knowledge-talk to be very expressive and useful in ordinary contexts. In general, the Pragmatic Error Skeptic can tell a story using implicature to accommodate roles for knowledge attributions that non-skeptical epistemologists have identified. My example had to do with distinguishing relevant alternatives (Dretske 1971; Goldman 1976; Stein 1976), but in principle the same basic pattern could also be used to show that false knowledge attributions can communicate that a subject is a reliable informant on a topic (Craig 1990) or that a proposition can be assumed to be true for the purposes of inquiry (Kappel 2010; Kelp 2011; Rysiew 2012). In each case, hearers implicitly take a knowledge claim to be false and search for a relevant and true implicature that has something to do with their salient practical or theoretical interests. Pragmatic Error Skepticism is not a version of the loose talk view, as I respect the plausible assumption that speakers and hearers typically do not conceive of nor intend their knowledge attributions to be imprecise and therefore false. Rather, knowledge attributions serve as a seemingly accurate way of communicating something of practical relevance. But like the loose talk view and others that rely on pragmatic

92  Gregory Stoutenburg principles to explain how knowledge attributions work in a way consistent with skepticism, there must be something that accounts for why speakers and hearers implicitly take positive knowledge attributions to be false. 4.2 Implicit Skepticism That account of appropriateness says that subjects (at least often) implicitly take literal knowledge attributions to be false because they implicitly accept that knowing requires meeting the infallibilist standard. The assumption that speakers implicitly accept skeptical standards is necessary because on a Gricean account, assumptions held by hearers govern which contents are implicated and when. We must distinguish implicit belief from occurrent belief, dispositional belief, and having a disposition to form a belief (cf. Audi 1994). Speakers do not occurrently believe that knowledge is demanding because knowledge claims are not usually disingenuous. If a person claims to know and means it, and you ask: “Do you know whether p?” you will get an affirmative answer. Moreover, few dispositionally believe that the standards for knowledge are high because few have considered skeptical arguments and arrived at any belief about standards for knowledge. Perhaps few are even disposed to believe that knowledge is demanding. It could be that once confronted with skeptical opposition to their ordinary practice of knowledge attribution, speakers realize that many of their beliefs are incompatible with this new information, so they stop thinking about the matter further to reduce cognitive dissonance and thus never outright believe skepticism. My view is that speakers implicitly believe that knowing is demanding. By that, I mean that speakers have a set of beliefs (B) about what they and others know and when appropriately prompted by challenges to their knowledge attributions speakers recognize that B entails that knowing requires meeting very demanding standards that are rarely met, thus the beliefs in B do not amount to knowledge. Speakers implicitly believe that knowing is demanding in the sense that they are logically committed to it and can easily become aware that they are logically committed to it. I do not think it is always obvious to speakers that knowledge attributions are usually false. Pragmatic Error Skepticism involves error. There is precedent for thinking that speakers can regularly and sincerely use an expression that contains a word denoting a standard while being implicitly committed to the falsity of the utterance because they implicitly recognize that the standard is extremely demanding. Claims involving geometric properties and specific quantities are like this. Geometry Class: Geometry class has just started. The teacher poorly draws a four-sided figure on the whiteboard. The ‘lines’ are visibly

Skeptical Invariantism, Considered  93 crooked. Two of the ‘corners’ do not connect. The teacher puts small ‘boxes’ in the ‘corners’ and tick marks on the sides. The teacher asks the class, “What shape is this?” They reply in unison, “A square.” The teacher replies, “Yes, that’s a square.” The teacher could point out that the figure on the board has none of the properties of a rectangle and by doing so get the class to retract their identification of the figure as a square. The class implicitly recognizes that calling the thing on the board a square is a useful falsehood in the context of geometry class, even if they were not actively thinking about the falsity of “that is a square” while agreeing to the utterance. Geometric properties are precisely defined, and speakers know this. Speakers regularly claim that a figure has some geometric property but they implicitly believe that such attributions are usually false because the shape in question does not meet the standard.23 Quantity attributions are similar, as in “I will be ready in five minutes” to mean “it won’t be long,” “there were a million cars in front of me” offered as an excuse for being late, “I paid the bill last week” when it may have been eight days ago, and so on. In these cases, speakers are implicitly committed to standards for “five,” “one million,” and “last week,” even though while using such expressions they are not thinking about whether the standard is met. However, speakers can recognize the falsity of such utterances, whether or not they entertain occurrent beliefs about the standards in question. 24 In each of these cases, a speaker who is probably not thinking about the literal truth or falsity of utterances makes or hears an utterance that expresses that a standard is met. The speaker is also able to recognize an easy entailment from the expressed content to some other, obviously false statement about the standard in question. For example, a student in the class understands “That is a square,” applied to the shape on the board, entails “that figure has four (perfectly straight) sides of equal length that meet at ninety-degree angles,” which the student implicitly believes is false. It is usually not necessary to consider the inference and the entailed statement, but that is only because contextually relevant purposes usually do not call for it. The same pattern works for knowledge attributions. Speakers are generally quick to recognize that their knowledge claims entail other claims that they cannot know because of what knowing requires: Skeptical Bank Case: I drive a friend to the bank to initiate a wire transfer, which requires presenting identification in person. I ask, “Do you know what forms of ID the bank will accept?” He replies, “Yeah, I brought my license.” I say, “Are you sure that’s enough?” He says, “I hadn’t thought about it before. I’ll check.”25

94  Gregory Stoutenburg Here a subject recognizes that knowing would enable him to disregard various possibilities, including the possibility that his driver’s license is insufficient as identification for initiating a wire transfer. He further implicitly believes that his belief that his driver’s license is sufficient identification does not meet that standard, and concludes that he does not know after all. If he did not implicitly believe that knowing which forms of identification are permitted involved meeting such a demanding standard, then there would be no way to make sense of his subtle retraction of his (also subtle) claim to know what forms of identification are acceptable, without his acquiring any new information. He would then be giving up his knowledge claim for no reason. In general, it is often not difficult to prompt speakers to retract or diminish previously made knowledge claims by raising possibilities of error. The practice only makes sense, though, if speakers implicitly believe that knowing involves meeting a very demanding standard, one what requires them, if they truly know, to be able to eliminate a very wide class of possibilities of error or maintain true belief in distant possible worlds. 4.3 Semantic Ignorance? It is plainly much harder to get anyone to explicitly believe that standards for knowledge are in general unmet. The source of the difficulty is not that subjects fail to understand the entailment from demanding standards for knowledge to their lack of knowledge in general. The source is that it is very difficult to get subjects to accept the skeptic’s account of what the standards for knowledge are in the first place. Speakers do not have an explicit belief with the content ‘knowledge is infallible justification.’ That may seem like a major concession, but it is not, because philosophical accounts are discoveries. We cannot read the correct standard for “knows” off of speaker behavior any more than we can for “object,” “responsible action,” “person,” “morally right,” “rational,” “God,” “virtue,” and so on. So, when the Pragmatic Error Skeptic claims that “S knows that p” expresses that S has infallible justification for believing that p and that subjects implicitly believe this, it is no objection that subjects asked to define “knowledge” do not state the infallibilist answer. R.M. Hare provides a helpful analogy. He compared the act of trying to analyze a concept that one uses regularly with a group of veteran dancers attempting to formulate the rules that constitute dancing a particular dance. Suppose that we are sitting at dinner and discussing how a certain dance is danced. Let us suppose that the dance in question is one requiring the participation of a number of people – say one of the

Skeptical Invariantism, Considered  95 Scottish reels. And let us suppose that we have a dispute about what happens at a particular point in the dance; and that, in order to settle it, we decide to dance the dance after dinner and find out. We have to imagine that there is among us a sufficiency of people who know, or say they know, how to dance the dance – in the sense of ‘know’ in which one may know how to do something without being able to say how it is done. (Hare 1960, p. 208) Continuing: If they know already how the dance is danced, what can they be arguing about? But if they do not know already, how will they know, when they have danced the dance, whether they have danced it correctly? The solution to the paradox lies in distinguishing between knowing how to dance a dance and being able to say how it is danced. Before the enquiry begins, they are able to do the former, but not the latter; after the enquiry is over they can do the latter, and they know that they are right because all along they could do the former. And it is the same with the analysis of concepts. We know how to use a certain expression, but are unable to say how it is used. Then we try to do the latter; and we know we have succeeded when we have found an analysis which is in accordance with our hitherto unformulated knowledge of how to use the word. (p. 216) Hare makes two important points here. One is that competent users of a term may actively dispute whether some use of a term is correct. The other is that competent users may be unable to provide an account of a concept without trying very hard to do so, and they may fail even then. Speakers, including philosophers, struggle to identify the correct account of “knowledge” and a whole slew of other concepts. So what? That is no objection to the plausibility of a skeptical semantics of knowledge attributions, or any other philosophical account, or any other account of how we do anything. Perhaps there is a further explanation of why describing the rules of the dance and stating the standards for knowledge (and other concepts) is so difficult. It is plausible that some of our beliefs have propositional contents that are sensorily or kinesthetically encoded. 26 If so, we could say that the dancers do have beliefs about what the steps of the dance are and what counts as “knowing,” but because those beliefs are encoded as sights and sounds and the feelings of movement of specific steps and turns, rather than remembered in language, it is very difficult to focus on the encoded propositional content to the exclusion of the sensory content that encodes it. Likewise, perhaps ordinary speakers using

96  Gregory Stoutenburg knowledge-talk do believe that knowledge requires infallible justification, but because beliefs about that standard are encoded as scenarios of conceding ignorance when error possibilities arise, being reluctant to ascribe knowledge of explicitly probabilistic propositions, and expecting “knowers” to be able to discriminate true belief from false belief, it is difficult to separate the encoded propositional content that includes the infallibilist standard of knowledge from the distinct knowledgeattributing sights and sounds that make up the belief.

5 Pragmatic Error Skepticism and Epistemic Contexts The examples I used to demonstrate how false knowledge attributions can be appropriate work in everyday contexts, where whether someone knows is almost never the focus of inquiry. But what about a context where knowledge is the focus, such as in criminal proceedings or a debate? There are two options for Pragmatic Error Skepticism here, and which one the skeptic will accept depends on the extent of the skeptic’s skepticism. The first option is for less-extreme skeptics who think that although our epistemic justification rarely or never meets the infallibilist standard expressed by a claim to know, we nevertheless enjoy solid justification for many ordinary beliefs. Assuming such a view, speakers and hearers will calculate to a relevant implicature using the maxim of Quality, just as before. They will hear “S knows that p” as false and calculate to an implicature that concerns a more reasonable standard of epistemic justification, short of infallibility. The second option is for extreme skeptics who think that even our epistemic justification is questionable or non-existent. Assuming that kind of skepticism, there will be no reasonable epistemic standard to which speakers and hearers are held. Instead, the skeptic will claim that our utterances that directly concern whether someone knows must be replaced by claims that associate “knowing that p” in a particular context with certain ways of forming beliefs. For example, a witness testifying on the stand, asked directly “Do you know whether p?” (and assuming that everyone takes the point of the inquiry to be whether the witness knows p and not thinking of the overarching point of the proceedings, namely whether some particular person is guilty of some particular crime) will count in this context as knowing that p if the witness had a sensory experience that we would normally call “seeing” some facts that entail p when a wide range of possibilities are ignored. Using a strategy like this, the skeptic will claim that when the focus is on knowledge, the normal mechanisms that make our knowledge-talk appropriate fail, so knowledge-talk must be replaced by something more accurate. It is a revisionary view, which surely counts against it. But the overall merits of the view, and problems with its alternatives, should also be remembered when making a judgment. 27

Skeptical Invariantism, Considered  97

Notes 1 2 3 4

5

6

7

8

9

10 11 12 13

(Cf. Austin 1962; Lawlor 2013) (Cf. Gettier 1963; Pritchard 2005) (Cf. Lewis 1996) Why not add: knowledge attributions are often true? Because the motivation for such a desideratum presupposes that what appear to be straightforward factual claims are typically true, and that assumption is demonstrably false, as I argue in (Stoutenburg 2016). I would deny (now) even that we seem to know a lot, if that state is distinct from the state of it seeming, in many particular contexts, that a positive knowledge attribution is acceptable. Related concerns appear in Section 4.2. Separately, my strategy of arriving at a conception of knowledge by way of the functional role played by knowledge attributions is much like Nevin Climenhaga’s in this volume, which he deploys toward a similar conclusion. That one’s belief be based on one’s infallible justification is not necessary, at least if B’s being based on J is anything but a necessary relation discernible through introspection. Otherwise, it would be possible to have infallible justification for p while being unable to know if one’s belief is justified. This definition is preferable to one defined in terms of entailment, which makes the dubious propositionalist view of evidence inevitable, as entailment holds only between propositions (Brown 2018). The truth-sufficiency definition is also preferable to the 1.0 probability definition, which does not explain why known propositions are not possibly false and struggles to account for knowledge of necessary truths, all of which are probability 1 on any body of evidence (cf. Stoutenburg 2019). For any theory that has this property—that if it is true, one is not justified in believing that it is true—it does not follow that the theory is false. But it does follow, by definition, that one should not believe whatever arguments are offered to support those theories, because those arguments do not justify believing the theory. Hofweber (1999) has argued that this kind of unawareness is common, so not an issue for contextualism. But this in the situations he has in mind are ones where, when pressed, we easily recognize that the standards we presuppose are in place, and this does not hold for knowledge attributions. My objection is not that contextualism requires “semantic blindness” (Hawthorne 2004). It does, and that is a problem, but my objection is that the content-unawareness implied by contextualism undermines the basic strategy of arguing for contextualism by traditional philosophical means. Gerken (2017) provides the most comprehensive strict moderate invariantist account, and a large portion of his account concerns how attributors can mistakenly think a possibility is relevant to knowledge when it is not. But one must wonder how, if our judgments are so unreliable, we can possibly trust them when they favor a fallibilist view. I was too sanguine when I opted for a version of a loose talk account in (2016). This is especially clear in chapters four and five of Unger (1975). Also see Stoutenburg (2017). Indeed, showing the opposite was among Unger’s goals in Ignorance: he proposed that it is always wrong to assert anything. Davis spells out the details of his loose talk account on pp. 410–412 of his (2007). I said Davis requires that “A believes that the content of U is close to the content of P.” Davis’s examples consistently involve speakers who are

98  Gregory Stoutenburg

14

15

16 17 18

19

20 21

22

fully aware that they speak falsely while intending to communicate something else. Kyriacou (n.d.) argues that an externalist view of meaning is compatible with speakers being unaware of the content expressed by “knows.” Importantly, in describing how an externalist theory of meaning affects the pragmatics of knowledge attributions, Kyriacou consistently attributes to speakers the intention to assert a claim about what properties are approximated by some entity or state of affairs (cf. pp. 11–16). So, I think it is most natural to characterize Kyriacou’s view as a non-standard version of the loose talk theory. Butchvarov (1970) was published before Grice’s “Logic and Conversation,” before any of Unger’s work on skepticism, and before philosophers gave serious attention to pragmatics. His contributions to issues surrounding our topic deserve far more recognition than they receive from contemporary epistemologists. Schaffer’s paper does not cite Butchvarov, despite offering basically the same account. The example is from Hawthorne (2004, p. 120). It is widely assumed in the literature that implicatures are calculated only when the hearer takes a speaker to have flouted a maxim at the level of what is said. For a challenge to this view, see (Dinges 2015). The Conditionalist could claim semantic blindness. See Schiffer (1996). The Disguised Conditionals theory is probably committed to the existence of unarticulated constituents that are contextually determined (Hofweber 1999). If so, then the Disguised Conditionals view is perhaps an unusual version of contextualism. It depends on how “undetectably deceived” is analyzed. See Dinges (2016, pp. 2580–2581) for criticism of another version of a Conditionals view. Another important proposal, but which I lack space to discuss, is proposed in Chung (2017). Chung calls her view “epistemic fictionalism,” which she defines as follows: “[E]xpressions within the relevant region of discourse— namely, knowledge-discourse—are generally used to assert (or otherwise illocute) some propositional content distinct from (and instead of) their semantic contents” (Chung 2017, p. 9). Chung proposes that knowledge talk is metaphorical. This view is also open to the objection that speakers at least typically take themselves to be sincere when they attribute knowledge, which is incompatible with conceiving of knowledge attributions as metaphorical. I am following the standard interpretation of Grice, which holds that implicatures arise when maxims are violated at the level of what is said (cf. 1975, p. 31). See Dinges (2015) for a critique of the standard interpretation. A similar account can be given in the language of Relevance Theory (Sperber and Wilson 1986; Wilson and Sperber 2004). Landon and Jill hear Fred’s utterance and process it in the usual way, by inferring to whatever is most relevant. So, they hear this knowledge claim, in this context, as being about something other than what Fred knows. Unger (1984, p. 9) once hinted at a strategy like this: “By trading on suitable premises concerning the context, and understood by the conversational participants, the speaker gets his hearers to infer from an (obviously) irrelevant falsehood he expressed to a relevant (presumed) truth then attended.” But “irrelevant” and “falsehood” are different concepts. Here, I focus on the idea that the implicitly recognized falsehood of knowledge attributions is what kicks off the search for an implicature. Thanks to Alexander Dinges for helpful comments about Quality and Relevance.

Skeptical Invariantism, Considered  99 23 When I have presented this example to philosophers, some resist it by claiming that it is too obvious that we are here using false descriptions, and that knowledge attributions are more challenging to retract. But when I have drawn such a figure in discussions with undergraduate students and then pointed out that its imperfections mean that the figure is not a square, many students are slow to accept this. The parallel between falsely attributed geometric properties and falsely attributed knowledge may be quite close. 24 For this reason, the ‘impermanence’ of skeptical attitudes is a weak objection to skepticism (Hawthorne 2004; Hume 1739 (2000), I.IV.1). Our interests rarely concern exact truth, so it is unsurprising that we would continue to speak in imprecise ways, even if we take our utterances seriously. 25 Inspired by DeRose (2002). If the answer to the knowledge question had been affirmative, my friend would say “yeah,” “yes,” or something similar. Outside of epistemology examples, people do not signal their ignorance explicitly, saying things like “I guess I don’t know, so I will check.” The knowledge-denial is implicit. 26 The concept empiricism of Prinz (2002) is akin to what I am suggesting here. 27 I thank David Alexander, Alexander Bow, Garret Caudle, Alexander Dinges, Landon Elkind, and Christos Kyriacou for comments on ancestors of this chapter. I also thank Christos Kyriacou and Kevin Wallbridge for inviting me to contribute this chapter to the present volume.

References Abath, A. J. (2012). Epistemic Contextualism, Semantic Blindness and Content Unawareness. Australasian Journal of Philosophy, 90(3), 593–597. https:// doi.org/10.1080/00048402.2011.631021 Audi, R. (1994). Dispositional Beliefs and Dispositions to Believe. Noûs, 28(4), 419–434. Austin, J. L. (1962). How to Do Things with Words. (J. O. Urmson & M. Sbisa, Eds.) (2nd ed.). Oxford: Oxford University Press. Brown, J. (2018). Fallibilism: Evidence and Knowledge. Oxford: Oxford University Press. https://doi.org/10.1093/oso/9780198801771.001.0001 Butchvarov, P. (1970). The Concept of Knowledge. Evanston: Northwestern University Press. Chung, J. (2017). Could Knowledge-Talk Be Largely Non-Literal? Episteme, 1–29. https://doi.org/10.1017/epi.2017.7 Cohen, S. (1999). Contextualism, Skepticism, and the Structure of Reasons. Noûs, 33, 57–89. https://doi.org/10.2307/2676096 Craig, E. (1990). Knowledge and the State of Nature. Oxford: Clarendon Press. Davis, W. A. (2007). Knowledge Claims and Context: Loose Use. Philosophical Studies, 132(3), 395–438. https://doi.org/10.1007/s11098-006-9035-2 Davis, W. A. (2015). Knowledge claims and context: belief. Philosophical Studies 172 (2): 399–432. DeRose, K. (2002). Assertion, Knowledge, and Context. Philosophical Review, 111(2), 167–203. https://doi.org/10.1215/00318108-111-2-167 Dinges, A. (2015). Innocent Implicatures. Journal of Pragmatics, 87, 54–63. Dinges, A. (2016). Skeptical Pragmatic Invariantism: Good, but Not Good Enough. Synthese, 193(8), 2577–2593. https://doi.org/10.1007/ s11229-015-0867-1

100  Gregory Stoutenburg Dretske, F. I. (1970). Epistemic Operators. The Journal of Philosophy, 67(24), 1007–1023. https://doi.org/10.2307/2024710 Dutant, J. (2015). The Legend of the Justified True Belief Analysis. Philosophical Perspectives, 29(1), 95–145. Dutant, J. (2016). How to Be an Infallibilist. Philosophical Issues, 26(1), 148– 171. https://doi.org/10.1111/phis.12085 Everett, T. (2006). Antiskeptical Conditionals. Philosophy and Phenomenological Research, 73(2), 505–536. Gerken, M. (2017). On Folk Epistemology: How We Think and Talk about Knowledge. New York: Oxford University Press. Gettier, E. L. (1963). Is Justified True Belief Knowledge? Analysis, 23(6), 121– 123. https://doi.org/10.2307/3326922 Goldman, A. I. (1976). Discrimination and Perceptual Knowledge. The Journal of Philosophy, 73(20), 771–791. https://doi.org/10.2307/2025679 Grice, H. P. (1975). Logic and Conversation. In Peter Cole & Jerry Morgan (Eds.), Syntax and Semantics, vol. 3: Speech Acts, 41–58. New York: Academic Press. Hare, R. M. (1960). Philosophical Discoveries. Mind: A Quarterly Review of Philosophy, LXIX(274), 145–162. https://doi.org/10.1093/mind/ LXIX.274.145 Hawthorne, J. (2004). Knowledge and Lotteries. New York: Oxford University Press. Hofweber, T. (1999). Contextualism and the Meaning-Intention Problem. In K. Korta, E. Sosa, & X. Arrazola (Eds.), Cognition, Agency and Rationality: Proceedings of the Fifth International Colloquium on Cognitive Science (pp. 93–104). Dordrecht: Springer Netherlands. Hume, D. (1739). A Treatise of Human Nature. (D. F. Norton & M. Norton, Eds.). New York: Oxford University Press. Kappel, K. (2010). On Saying that Someone Knows: Themes from Craig. In A. Haddock, A. Millar, & D. Pritchard (Eds.), Social Epistemology (pp. 69–88). New York: Oxford University Press. Kelp, C. (2011). What Is the Point of “Knowledge” Anyway? Episteme, 8(1), 53–66. Kyriacou, C. (2019). Semantic Awareness for Skeptical Pragmatic Invariantism. Episteme, 1–19. https://doi.org/10.1017/epi.2019.7 Lawlor, K. (2013). Assurance: An Austinian View of Knowledge and Knowledge Claims. New York: Oxford University Press. Lewis, D. (1996). Elusive Knowledge. Australasian Journal of Philosophy, 74(4), 549–567. https://doi.org/10.1080/00048409612347521 Nozick, R. (1981). Philosophical Explanations. Cambridge: Harvard University Press. Prinz, Jesse J. (2002). Furnishing the Mind: Concepts and Their Perceptual Basis. MIT Press. Pritchard, D. (2005). Epistemic Luck. New York: Oxford University Press. Roush, S. (2012). Sensitivity and Closure. In K. Becker & T. Black (Eds.), The Sensitivity Principle in Epistemology (pp. 242–268). New York: Cambridge University Press. Rysiew, P. (2001). The Context-Sensitivity of Knowledge Attributions. Noûs, 35(4), 477–514. https://doi.org/10.1111/0029-4624.00349

Skeptical Invariantism, Considered  101 Rysiew, P. (2012). Epistemic Scorekeeping. In J. G. Brown (Ed.), Knowledge Ascriptions (pp. 270–293). Oxford: Oxford University Press. Schaffer, J. (2004). Skepticism, Contextualism, and Discrimination. Philosophy and Phenomenological Research, 69(1), 138–155. https://doi. org/10.1111/j.1933-1592.2004.tb00387.x Schiffer, S. (1996), The hidden-indexical theory’s logical-form problem: a rejoinder. Analysis, 56: 92–97. Sperber, D., & Wilson, D. (1986). Relevance: Communication and Cognition. Cambridge: Harvard University Press. Stein, G. C. (1976). Skepticism, Relevant Alternatives, and Deductive Closure. Philosophical Studies, 29(4), 249–261. Stoutenburg, G. (2016). Principles of Interpretive Charity and the Semantics of Knowledge Attributions. Acta Analytica, 31(2), 153–168. https://doi. org/10.1007/s12136-015-0267-7 Stoutenburg, G. (2017). Strict Moderate Invariantism and KnowledgeDenials. Philosophical Studies, 174(8), 2029–2044. https://doi.org/10.1007/ s11098-016-0786-0 Stoutenburg, G. (2019). The Limits of Antiskeptical Infallibilism. In T. V. Rodrigues (Ed.), Epistemologia Analytica (Vol. 1, pp. 55–72). Editora Fi. Unger, P. (1971). A Defense of Skepticism. The Philosophical Review, 80(2), 198–219. Unger, Peter (1975). Ignorance: A Case for Scepticism. Oxford University Press. Unger, P. (1984). Philosophical Relativity. Minneapolis: University of Minnesota Press. Williamson, T. (2000). Knowledge and Its Limits. New York: Oxford University Press. Wilson, D., & Sperber, D. (2004). Relevance Theory. In L. Horn & G. Ward (Eds.), The Handbook of Pragmatics (pp. 607–633). Malden: Blackwell.

6

Moderate Pragmatic Skepticism, Moorean Invariantism and Attributions of Intellectual Virtue/Vice Christos Kyriacou ‘Nothing, therefore, can be more contrary than such a[n Academic or Sceptical] philosophy to the supine indolence of the mind, its rash arrogance, its lofty pretensions, and its superstitious credulity.’ Hume, Enquiry Concerning Human Understanding, Section V, Part I, p. 41.

1 Introduction Skepticism about knowledge is rarely taken seriously as a philosophical position in mainstream epistemology because it is typically considered to be false.1 Partly, the reason is that it would be almost insane to take skepticism seriously because it flatly denies a fundamental appearance (many would say fact, e.g. McKenna this volume) of our cognitive lives, namely, that we can and do have lots of knowledge. The post hoc task for epistemologists is then considered to be one of rationalization: how best to understand the challenge of skepticism, its seductive intuitive appeal and how best to respond to it in a way that saves the epistemic appearance that we know a lot. But although skepticism faces formidable problems (many would say fatal), a moderate version of skepticism can, arguably, be explanatorily fruitful and resourceful in both addressing various problems in the theory of knowledge and responding to objections to skepticism itself.2 To bolster this claim, in this paper I explore the possibility that, were it to be true, a refined moderate version of skeptical invariantism about knowledge—what I shall call moderate pragmatic skepticism—could, in principle, adequately explain an important and integral part of epistemic discourse: attributions of basic intellectual virtue(s)/vice(s). As I argue, it could even explain such attributions better than the contender of Moorean invariantism. My overall dialectical goal is to indicate that moderate pragmatic skepticism can not only be explanatorily promising with regard to various epistemic problems (e.g. Gettier cases), semantically-linguistically sophisticated and coherent with the everyday practice of assertion and practical reasoning (as I have already argued elsewhere), but also could, in principle, make good sense of another important facet of ordinary epistemic discourse and practice, namely, attributions of basic intellectual virtue(s)/vice(s).3

Moderate Pragmatic Skepticism  103 I proceed as follows. In Section 2, I stipulate and briefly motivate moderate pragmatic skepticism, in Section 3, I present four intuitive cases of attribution of intellectual virtue(s)/vice(s), and in Section 4, I argue that moderate pragmatic skepticism can adequately explain the attribution of basic intellectual virtue(s)/vice(s) in the canvassed cases. In Section 5, I examine how Moorean invariantism fares with regard to the presented cases and argue that it runs into difficulties that reveal three inherent epistemological problems with the condition of safety; in Section 6, I very briefly reply to three objections; and finally, in Section 7, I sum up with a conclusion.

2 Moderate Pragmatic Skepticism: Commitments and Motivation A standard version of skeptical invariantism is committed to three basic theses: a b

c

Invariantism: the standard of knowledge is invariant (and purely intellectual). Stringency of Standard: the standard of knowledge is stringent enough to be far less satisfied than we ordinarily think it is satisfied.4 The conjunction of (a) Invariantism and (b) Stringency of Standard entails: Radical Skepticism: We know far less than we ordinarily think we know (if we know anything at all) and, hence, radical skepticism is true.5

Typically, the stringent standard of knowledge is taken to be deductive justification that entails truth and, hence, skeptical invariantism is naturally conjoined with traditional infallibilism as well.6 But no matter how exactly stringent the standard of knowledge is, given that it is not satisfied as often as we ordinarily think it is satisfied, the implication of the conjunction of (a) invariantism and (b) stringency of standard is: (c) radical skepticism. Thus, the conjunction of (a), (b) and (c) seems to offer us a succinct characterization of the basic commitments of skeptical invariantism. Skeptical invariantism is considered to be false—perhaps even trivially false—by almost all epistemologists because it incurs obvious and serious problems. But more recently, some epistemologists have argued that some form or other of skeptical invariantism can offer interesting responses to various problems for a theory of knowledge, such as the Gettier problem, the lottery problem, the value problem, the dogmatism paradox, concessive knowledge attributions, assertion and practical reasoning, retraction and more.7 If this assertion bears some prima facie plausibility, skeptical invariantism should, presumably, be considered more seriously as an option in the theory of knowledge before the customary quick dismissal. In this spirit, I provide below some motivation for a specific skeptical invariantist view. I sketch four obvious problems for skeptical

104  Christos Kyriacou invariantism and quickly indicate that when refined to be moderate pragmatic skeptical invariantism (or, for short, moderate pragmatic skepticism) these problems can, in principle, be addressed rather easily. First, skeptical invariantism seems to fly in the face of everyday experience and, hence, to be empirically (and extensionally) inadequate. That is, in everyday life we tend to think—tacitly and explicitly—we know a lot and rely, by and large, on what we think we know for asserting and practical reasoning.8 Yet, if there were little or even no knowledge, then it would appear that we are under the illusion that we know a lot and, therefore, that we should refrain from knowledge-based assertion and practical reasoning. Thus, were it to be the case that skeptical invariantism is true, it would seem to demand the radical revision of our everyday epistemic lives. As a result, assertion and practical reasoning should not be knowledge-based, if there is little or no knowledge. This so-called argument from the knowledge norm for assertion and practical reasoning seems to provide a reductio against skeptical invariantism. However, the skeptical invariantist can be shrewdly pragmatic and deny that we need to radically revise our ordinary knowledge-based practice of assertion and practical reasoning (were skeptical invariantism to be true).9 Skeptical invariantism can take a Humean, pragmatic turn and ‘deflate’ our ordinary practice of assertion and practical reasoning without conceding the need for radical revision. That is, it can accept that although knowledge proper is infallibilist, in everyday life we can proceed, by and large, on broadly fallibilist terms because it is reasonably useful for practical purposes.10 So, skeptical invariantism can be refined to be skeptical pragmatic invariantism. Second, another important concern about skeptical invariantism is that it seems epistemically self-defeating if understood as Unger (1975)-style global skepticism, that is, as universally denying all knowledge whatsoever. For, if there is no knowledge at all, it seems incoherent to suggest that you can know that there is no knowledge. After all, how can you know that if there is no knowledge?11 Before we respond to the second problem let us introduce a third related problem. We ordinarily seem to have at least some minimal basic knowledge, such as introspective and modal knowledge. It is exceedingly hard, if not impossible, to deny that we have some Cartesian introspective knowledge, such as of conscious existence and raw phenomenal sensations (‘I am conscious’ or ‘I am in pain’).12 Equally hard it is to deny that there is some a priori knowledge of necessity, such as mathematical and logical necessity and even epistemic necessity, such as the factivity condition, namely, that knowledge entails truth (pace Quinean skepticism about modality).13 Fortunately, skeptical invariantism can be further refined in a way that speaks to these two reasonable concerns as well. That is, it can be relatively moderate and accept that there is at least some minimal basic knowledge—although not as much knowledge as we tend to assume in

Moderate Pragmatic Skepticism  105 everyday life. In other words, skeptical invariantism can assume that there is a robust (infallibilist) knowledge relation that we do not satisfy as often we tend to think we do. This would revise: (c) Radical Skepticism: We know far less than we ordinarily think we know (if we know anything at all) and, hence, radical skepticism is true, to the more lax: (c*) Moderate Skepticism: We know far less than we ordinarily think we know, but we have at least some minimal basic knowledge. The revision would alleviate concerns about skeptical invariantism with regard to both the second and third identified problems: epistemic self-defeat and minimal basic knowledge (introspective and modal). On the one hand, we can, in principle, (infallibly) know the truth of skeptical invariantism although, in practice, this is very unlikely (given the difficulty of the matter). It is more promising to look for an abductive justification of skeptical invariantism.14 On the other hand, we can, in principle, have at least some introspective and modal knowledge because the theory allows that we can reliably have some (infallibilist) knowledge of some truths (e.g. ‘I have a headache’, 1 + 1 = 2, Kp → p). Fourth, skeptical invariantism seems to misconstrue how we tend to understand the meaning of ‘know’ because in everyday life we tend to assume that we (fallibly) know many things. Thus, were skeptical invariantism to be true, it would attribute to ordinary speakers widespread semantic blindness about ‘know’ and it seems questionable whether such blindness could ever be justifiable. In reply, moderate pragmatic invariantism can offer a sophisticated story about the semantic/pragmatic workings of the theory that helps make sense of such widespread semantic blindness. A rather natural way to understand the semantic/pragmatic workings of skeptical pragmatic invariantism is by reference to linguistic loose use (cf. Davis (2007) and Dinges (2016)), the usage of modifiers (adjectives and adverbs, such as ‘real’ and ‘really’) and semantic externalism (cf. Kyriacou (2019)). According to this account, knowledge attributions of the form ‘S knows that p’, all other things equal, literally say one (infallibilist) thing but due to loose use mean another (fallibilist) that is close enough to what is said. The semantically internalist intuitions of agents about ‘know’ tend to be fallibilist, but the true meaning of ‘know’ need not be internalist and could be fixed by semantically externalist considerations. We can then use modifiers (adjectives and adverbs, as in ‘real knowledge’ and ‘really known’) in order to disambiguate between fallibilist and infallibilist usage of ‘know’.15 Loose use in knowledge attributions operates via substitutional implicatures: that is, “implicatures where the speaker doesn’t convey what is said but something else instead” (cf. Dinges (2016:2579)). What is literally said is the semantics component and what is loosely meant (via substitutional implicature) is the pragmatics component. Semantic closeness

106  Christos Kyriacou (or similarity) between infallibilist semantic meaning and fallibilist pragmatic meaning suggests that the two share all the basic conditions of knowledge (justification, truth, belief) but disagree on the stringency of the standard of justification required for real or proper knowledge. Infallibilists suggest that knowledge requires infallible justification while fallibilists suggest that it does not.16 To sum up, if we sharpen some of the crude edges of skeptical invariantism, we can have the more elegant version of moderate pragmatic skepticism that alleviates some of the obvious concerns about the plausibility of skeptical invariantism (e.g. epistemic self-defeat, semantic blindness, minimal basic knowledge, assertion and practical reasoning). No doubt, this conceptual sharpening does not suffice on its own to render moderate pragmatic skepticism plausible, but it at least showcases the resourcefulness and flexibility of skepticism and thereby leaves it somewhat more respectable as a theoretical option. With the introduction of moderate pragmatic invariantism now at our disposal, let us turn to introducing attributions of basic intellectual virtue(s)/virtue(s).

3 Attributions of Basic Intellectual Virtue(s)/Vice(s) Debates in epistemology have witnessed the blossoming of virtue epistemology in both responsibilist and reliabilist guises.17 Responsibilist virtue epistemology typically understands intellectual virtues in broadly Aristotelian terms: as deep-seated character traits that dispose for corresponding action (practical or theoretical/cognitive). For instance, being intellectually just disposes one to treat a theoretical question, problem or line of inquiry impartially and fairly. Being intellectually courageous disposes one to treat a theoretical question, problem or line of inquiry courageously and to persevere in the face of risk and adversity and so on. The recent focus on intellectual virtues and vices has long been overdue. Aretaic thought and talk is an integral part of our intellectual lives, just as it is of our moral lives, and deserves as much as attention. As part of epistemic folk psychology, we attribute intellectual virtues and vices to persons and collective agents in order to understand, explain and predict behavior. For example, if I know that no possible plausible argument could ever convince Philip about p, then I may judge him as dogmatic and narrow-minded. This attribution may help me rationalize, understand and explain his intellectual conduct about p. It may even help me predict that in future occasions, unless his character changes in some fashion, he is likely to conduct himself in a similar way about p. To illustrate in more detail how ascription of basic intellectual virtue and vice is an integral part of both everyday life as well as of scientific practice, I introduce below four intuitive cases. The cases are loosely inspired from the history of modern science (NEWTON, DARWIN), classic literature (SHERLOCK) and ordinary life (TAXI DRIVER) and

Moderate Pragmatic Skepticism  107 involve attribution of paradigmatic intellectual virtues and vices (virtues/vices exhibited in each case are in brackets).18 NEWTON [Intellectual Complacency and Incautiousness] Newtonian mechanics was an established orthodoxy in the 18th–19th centuries. The impressive ability of the theory to adequately explain diverse physical phenomena in a unifying way, its predictive power and mathematical elegance seemed to foreclose the question of whether it is the truth about physical reality. The thought of the physicists of the day was that ‘We know that Newtonian mechanics is the truth about the physical world’ because of the high epistemic probability of the thesis, given evidence at the time. In retrospect, there were anomalies within the Newtonian picture. Famously, Newtonian physics failed to predict the perihelion of Mercury with accuracy. But physicists of the day were positive that this was merely insignificant ‘noise’ or a minor measurement problem, or at any rate a negligible problem to be solved at some point when more information would be gathered about the details of the problem. As we know today, they were shown to be wrong by Einstein’s general relativity theory in the early 20th century. Physicists did not really know that Newtonian mechanics is the truth about the physical world because this was not and is not the truth. In being presumptuous and overconfident that they did know, they seemed to have committed a cognitive mistake, a mistake that betrays respective intellectual vices, such as intellectual complacency and incautiousness.19 Had physicists of the day, in light of the evidence from explanatory anomalies, been more intellectually cautious and humble, they could have avoided the mistake by humbly asserting only that Newtonian mechanics is currently our best theory, but not necessarily the truth about physical reality. DARWIN [Intellectual Courage, Open-mindedness, Autonomy] Before Darwin, there was relatively little doubt that the world had been directly designed, as it were, by an intelligent supreme being, God. Biological evidence at the time rendered highly epistemically probable that there is intelligent design. But Darwin meticulously studied through observation organisms and came to the conclusion that there was an overlooked alternative explanation to be given about the biological world. He famously discerned that there is a pattern in nature that could explain biological diversity and the appearance of Aristotelian teleology mechanically without a gratuitous appeal to any intelligent designer. That pattern was evolution by means of natural selection.

108  Christos Kyriacou In proposing the theory of natural selection, Darwin exhibited distinctive intellectual courage, open-mindedness and autonomy. Darwin showed courage because he stepped forth to question the Aristotelian teleological paradigm accepted by almost every biologist of the day and did not shrink of the task in spite of the corrosive consequences the theory could have about his social relations and the current worldview. He showed open-mindedness because he retained an open and non-dogmatic perspective on things that allowed him to consider seriously alternative theoretical understandings of the biological world. He showed intellectual autonomy because he sought an understanding of his own accord in spite of the Aristotelian biological orthodoxy. Had Darwin been less courageous, he might have not gone all the way to collect evidence and arrive to the conclusion he did. Had he been less open-minded and absolutely sure, or relatively sure that the biological world is intelligently designed, he either wouldn’t have looked for an alternative explanation in the first place, or have looked but likely have failed to take seriously such an alternative explanation. Had he been more conformist and less independent in thinking, he might not have searched for more evidence and pursued it to its rational conclusion. SHERLOCK HOLMES [Intellectual Imagination and Patience] In The Silver Blaze, the renowned detective Sherlock Holmes undertakes to investigate yet another murder case: the tragic murder of the trainer of the racing horse Silver Blaze and the disappearance of the horse itself. “Inspector Gregory, to whom the case has been committed on his arrival…arrested the man who suspicion naturally rested [given strong evidence at the time, rendering the belief highly epistemically probable]  …Fitzroy Simpson” (2008:9). Holmes carefully establishes the “essential facts of the case” and although acknowledges the prima facie plausibility of the police’s hypothesis, he refrains from endorsing ‘the theory of the police’. With his remarkable patience in collecting evidence, exceptional imagination and characteristic sensitivity to miniscule evidence and diligence in drawing inferential connections, he conceives a different ‘working hypothesis’ (2008:19) that eventually, in light of further accumulated evidence, justifies who is the true murderer and robber: Mr Strakke. Twice in the story Holmes himself extols the ‘value of imagination’ (2008:9, 20) and quips that although Inspector Gregory is overall ‘extremely competent’, he is lacking in that very quality. As Holmes himself indicates, had inspector Gregory been more gifted in imagination—and we may add, patience,—‘he might [have] rise[n] to great heights in his profession’ (2008:9).

Moderate Pragmatic Skepticism  109 TAXI DRIVER [Intellectual Conscientiousness] Johnny is a New York taxi driver. He works in the hustle and bustle of the center of the metropolis and everyday serves dozens of people. He has been in his sociable job for more than 30 years now, and receives lots of information every day from his customers, ranging from mundane facts to extraordinary conspiracy theories and other fairy tales of little credibility. He understands from past experience that some customers are know-it-all, others pathological liars and shameless ‘bullshitters’, others wishful and self-conceited, others so gullible and lazy that spread unsubstantiated rumors without responsible fact-checking, yet others like to confabulate and brag about their life saga of heroism and achievement and so on. Thus, given inductive evidence from past experience, random customer information is of low epistemic probability. In light of this, he sifts information carefully as he realizes that people are often unreliable, irresponsible and know-it-all. When some piece of information is of particular interest to him, he double-checks its credibility against independent reliable sources. In so doing, he is intellectually conscientious. Had Johnny been less conscientious, he would probably have formed many more unjustified (and false) beliefs on the basis of unreliable customer testimony. These cases indicate that aretaic folk psychology is an integral and valuable part of epistemic discourse and practice. We attribute basic intellectual virtues and vices to scientists, detectives and laymen in order to understand their (im)proper intellectual conduct. Inevitably, our theory of knowledge should account for our intellectual aretaic discourse and practice and, indeed, epistemologists have followed suit the tradition of virtue ethics and proposed virtue epistemological theories of knowledge and justification (of responsibilist and reliabilist ilk). 20 However, to the best of my knowledge, not much of systematic attention has been directed to the attribution of intellectual virtue(s)/vice(s) and its relation to the semantics/pragmatics of knowledge attribution. More attention, I think, has been paid to the nature, individuation and truth/falsityconducive role of intellectual virtues/vices. In the next section, we pay more attention to this neglected aspect of things.

4 Moderate Pragmatic Skepticism and the Attribution of Basic Intellectual Virtue(s)/Vice(s) Moderate pragmatic skepticism can provide us with a prima facie natural (if not ultimately plausible) explanation of the intuitive attribution of basic intellectual virtues and vices in the presented cases.

110  Christos Kyriacou Take NEWTON first. In the NEWTON case, we attributed to Newtonian physicists of the day presumption and overconfidence biases as well as the vices of intellectual complacency and incautiousness. Moderate pragmatic skepticism can explain the intellectual bias and vice attribution. The account suggests that they were biased by presumption and overconfidence and induced to commit a cognitive mistake because they were intellectually complacent and incautious. They were complacent and incautious because they should have refrained from committing to really (infallibly) knowing that Newtonian mechanics is the truth about physical reality, in light of the anomaly of the perihelion of Mercury. If there is some evidence against a belief, even if this evidence is not strong enough to seriously undermine the high epistemic probability of the belief, we should refrain from claiming to know because we might be shown mistaken. Otherwise, we are being intellectually complacent and incautious in ignoring as irrelevant the low epistemic probability possibility of error and in thinking that we know. Had Newtonian physicists been more intellectually humble and cautious, they could have avoided the mistake. Take now the DARWIN case. In the DARWIN case, we attributed to Darwin intellectual virtues such as autonomy, courage and open-mindedness. Moderate pragmatic skepticism can make sense of the intellectual virtue attribution. Darwin showed the required intellectual autonomy, courage and open-mindedness to seek an alternative understanding of his own accord of the biological phenomena because he questioned that we really (infallibly) knew that the established biological paradigm of Aristotelian teleology is true and beyond doubt. He treated the theory as one possible hypothesis, even a prima facie plausible hypothesis of high epistemic probability, but refused to accept that the theory is certain and beyond doubt. Had Darwin been more conformist, timid and narrow-minded and assumed that Aristotelian teleology is already really (infallibly) known, he would have never bothered to seriously question the theory and look for an alternative. SHERLOCK is next. In the SHERLOCK case we attributed to the detective the virtues of intellectual imagination and patience. Moderate pragmatic skepticism can explain the attribution. Sherlock had intellectual imagination and patience because he patiently refused to conclude that we really (infallibly) know that Fitzroy was the murderer in spite of the strong evidence supporting the hypothesis. Instead, he imaginatively counterfactually conceived an alternative working hypothesis about the course of events that had led to the murder. Initially, Sherlock’s imagined hypothesis was coherent with the evidence, but of sufficiently low epistemic probability. Eventually, as more evidence was patiently gathered, the conceived alternative hypothesis better explained the total accumulated evidence. In being patient, Sherlock resisted our natural tendency to jump

Moderate Pragmatic Skepticism  111 to conclusions on the basis of the psychological mechanism of—what Kahneman (2011) calls—WYSIATI (What You See Is All There Is) and, in tandem with the total evidence requirement, waited to diligently collect more evidence and explore inferential connections between pieces of evidence before he comes to a rational verdict. Had Sherlock been less imaginative and patient, he could have mistakenly concluded with Inspector Gregory that Fitzroy was the murderer. Last the TAXI DRIVER case. In the TAXI DRIVER case we attributed to Johnny the intellectual virtue of conscientiousness. Moderate pragmatic skepticism can explain this virtue attribution as well. Johnny is conscientious and responsibly sifts the information flow stemming from customers because, on the basis of his past experience, he is aware that all that passes for knowledge in the back seat of his cab is not really (infallible) knowledge. Very often the information is not even true or justified, and even when it is adequately justified and of high epistemic probability, given evidence at the time, he understands that it might not be really knowledge because it could still turn out to be false. Besides, time and again things he has considered to be knowledge, given strong evidence, have turned out eventually not to be knowledge, in light of more evidence. Now, we can observe that in the NEWTON, DARWIN and SHERLOCK cases there is an emerging common pattern. In the DARWIN and SHERLOCK cases, the attribution of basic intellectual virtue(s)/ vice(s) appears appropriate because an intellectually virtuous agent questions presumed (fallible) knowledge and desists to claim real (infallible) knowledge. Conversely, in the NEWTON case, some insufficiently intellectually virtuous (or even vicious) group of people fails to question presumed (fallible) knowledge and do not desist to claim real (infallible) knowledge. Presumed (fallible) knowledge is true belief sufficiently justified to be of high epistemic probability, given evidence at the time, but falls short of infallible epistemic probability 1. The TAXI DRIVER case is slightly different in that the driver realizes, in virtue of past experience, that random customer information is statistically unreliable and, therefore, of low epistemic probability and conscientiously desists to claim knowledge. But he is exactly conscientious because he desists from claiming knowledge. The observed pattern is consonant with moderate pragmatic skepticism because the theory suggests, in broadly Cartesian spirit, that we really know only when we have infallible justification. Of course, we might speak loosely of (fallible) knowledge in more relaxed everyday epistemic situations, and it generally works for us, practically speaking, but this might not be real (infallible) knowledge, properly speaking. 21 The corollary to be drawn from the presented cases is that the attribution of basic intellectual virtue(s)/vice(s) is prima facie explainable

112  Christos Kyriacou within a broadly skeptical pragmatic framework. The result is encouraging for a position that stands at the fringes of mainstream theory of knowledge. But it is sobering to have in mind that it does little in the way of supporting moderate pragmatic skepticism because other theoretical contenders might also explain the cases (and even better at that). In the next section, I briefly introduce the rudiments of the fallibilist framework of Moorean invariantism, as I understand it, and suggest that it runs into difficulties that reveal three inherent epistemological problems with the modal condition of safety. Given these difficulties, it appears unclear whether Moorean invariantism can offer as natural and unequivocal an explanation of the cases.

5 Moorean Invariantism and the Attribution of Intellectual Virtue(s)/Vice(s) Moorean accounts of knowledge are prominent in the theory of knowledge. Usually, Moorean accounts of knowledge explicate knowledge in terms of the modal condition of safety. 22 Roughly, they suggest that: (K) Knowledge is safe true belief23 Safety, which is understood to be a necessary condition of knowledge, is then, at first approximation, understood as: (S) Adequate justification of true belief that could not easily have been false (e.g. Pritchard 2012:253) On a natural reading, adequate justification of true belief that could not easily have been false is understood as: (AJ) Adequate justification (or warrant, or what have you) of true belief entitles us to rule out the possibility of error in all relevant nearby possible worlds (Williamson 2000:147). 24 Again, on a natural reading, entitlement to rule out the possibility of error in all relevant nearby possible worlds, may be understood, I think, as: (E) Entitlement to rule out the possibility of error in all relevant nearby possible worlds is grounded in adequate justification in virtue of the high epistemic probability of the belief, given evidence. 25 Safety-based accounts of knowledge claim to bear a number of theoretical attractions, such as that they save Moorean epistemic appearances about knowledge and address radical skepticism, account for ordinary (fallibilist) assertion and practical reasoning, empirical and inductive

Moderate Pragmatic Skepticism  113 knowledge, account for Gettier cases and more. 26 But their attractions notwithstanding, Moorean approaches do not seem to be able to offer as natural an interpretation of the attribution of basic intellectual virtue(s)\ vice(s) in the four presented cases without revealing three inherent epistemological problems with the condition of safety—or so I will argue. Take NEWTON first. In the NEWTON case, we attributed to physicists of the day the biases of presumption and overconfidence as well as the vices of intellectual complacency and incautiousness. According to moderate pragmatic skepticism, Newtonian physicists of the day did not really know what they presumed they knew and this is the rather natural interpretation of their epistemic condition. But by Moorean standards, the physicists’ belief in Newtonian mechanics would seem to qualify as knowledge because the belief was prima facie safe (and true) at the time. Or, more precisely, the belief was presumed at the time to be safe (and true) because, given the enormous evidence in favor of Newtonian mechanics, it couldn’t have easily been false. That is, given the overwhelming evidence at the time, in all relevant nearby possible worlds, the belief was deemed safe (and true) at the time (and the possibility of error was of low epistemic probability to be of any relevance). What is important is that if this is to the right direction, Moorean approaches do not seem in position to naturally attribute intellectual complacency and incautiousness to Newtonian physicists of the day because, according to the Moorean picture, Newtonian physicists did conduct themselves virtuously, as they should, and rightly came to what was deemed at the time to be knowledge. Newtonian physicists believed they knew because they considered their belief to be safe (and true), given the enormous supporting body of evidence at their disposal. Yet this seems to be the wrong prediction by the Moorean picture because intuitively we do attribute intellectual vice to Newtonian physicists of the day. The rather obvious Moorean rejoinder to the argument would be that the physicists’ belief was not even safe because it simply wasn’t true. Moorean safety, properly understood, implies that the belief is adequately justified to be modally stable and, of course, true. Hence, the safety theorist need not fear of the objection because it misunderstands Moorean safety: by its own lights, Moorean safety would not misattribute knowledge to the Newtonian physicists of the 18th–19th centuries. This is true as far as it goes, but the Moorean rejoinder points to a significant epistemological trio of problems for the safety condition. A first epistemological problem lies at the heart of the notion of safety: how are we to tell which prima facie safe and true beliefs are really safe and true, if even the beliefs that seem beyond reasonable doubt safe and true, given strong evidence at a time, can be shown in due course to be not really safe and true? Call this ‘the problem of elusive safety’.27

114  Christos Kyriacou The problem of elusive safety is potent because we cannot simply assume that some beliefs are safe and true at a time because, given strong evidence, they are of high epistemic probability and appear beyond reasonable doubt safe and true. If we proceed on the basis of such a facile assumption, we simply pave the way to think as knowledge of whatever is presumed knowledge because it is of sufficiently high epistemic probability, given adequate evidence at a time. But surely, whatever is considered knowledge and it is of sufficiently high epistemic probability, given adequate evidence at a time, need not, and often is not, really knowledge (cases abound, e.g. flat earth belief in medieval years).28 Thus, the Moorean does not have as natural an interpretation to offer for the intuitive attribution of intellectual complacency and incautiousness. To sum up, the Moorean faces a safety underdetermination problem: we have no obvious way to tell which beliefs are really safe and true and which merely appear to be safe and true, given evidence at a time, and if this is the case, then the Moorean would be at a loss about how to understand the intuitive attribution of the intellectual vices to Newtonian physicists. For, by Moorean lights, Newtonian physicists’ intellectual conduct was as it should be because they simply ignored the epistemically improbable, given strong evidence at the time, and thereby irrelevant possibility of error. Therefore, we should not ascribe the intellectual vices of complacency and incautiousness to them. Yet this conflicts with the firm intuition that Newtonian physicists did miss something important and that we should think them overconfident, complacent and incautious. It is this intuition that moderate pragmatic skepticism vindicates and Moorean invariantism does not. Next the DARWIN case. In the DARWIN case, we attributed to Darwin intellectual virtues such as autonomy, courage and open-mindedness and suggested that moderate pragmatic skepticism can make good sense of the intellectual virtue attribution. Darwin did not think we really knew the truth of Aristotelian biology, in spite of its high epistemic probability, given strong evidence at the time, and this helps explain his autonomy, courage and open-mindedness. The explanation is natural, simple and straightforward. In contrast, safety-based Mooreanism does not seem to be able to offer as natural an interpretation of the attribution of intellectual virtue in DARWIN. Mooreanism would instead seem to attribute knowledge to the biologists of the day that followed the Aristotelian paradigm. This is because their belief again was presumed to be safe and true. The Aristotelian biologists deemed their belief to be rather difficult to be shown false, or in the possible worlds parlance, that in all relevant nearby possible worlds they could conceive it was true, given strong evidence at the time. The possibility of error was of a sufficiently low epistemic likelihood to be considered as relevant, or worthy of serious consideration.

Moderate Pragmatic Skepticism  115 But if this is the case, then we cannot attribute autonomy, courage and open-mindedness to Darwin. These are virtues we attribute to Darwin in the case because he boldly questioned whether we really had knowledge. Otherwise, we would have to accept ‘the abominable conjunction’ that ‘Darwin knew that Aristotelian biology was true, but accepted it could be false’, which appears paradoxical because if we really know we do not, and it seems rationally should not, tend to seriously allow for error possibilities. The obvious reason for this attitude being that, all other things equal, knowledge seems to imply termination of inquiry about the truth of p.29 Besides, why further inquire for knowledge of something we already know the answer to? It seems pointless. The SHERLOCK case is no different for the safety theorist. In the SHERLOCK case, we attributed to the ingenious detective the virtues of intellectual imagination and patience and argued that moderate pragmatic skepticism can help explain and make sense of the attribution. Again, it is not clear that Moorean accounts can explain the attribution as naturally. By Moorean lights, inspector Gregory would have thought that Fitzroy is the murderer because strong evidence against him at the time suggested that the belief that he is the murderer is safe (and true). No more patience and imagination would be needed given that the evidence at the time was strongly supporting the belief—all other remote, low epistemic probability possibilities of error would appear irrelevant and unworthy of serious consideration. However, we may reasonably worry that in this way safety-based accounts of knowledge help instill complacence (and not humility and imagination) and impatience (and not patience) because they suggest that we know quite a lot, thereby saving Moorean epistemic appearances. They would also tend to be dogmatically silencing doubts and discouraging pursuing alternative unorthodox lines of inquiry that appear at the time to be irrelevant, low epistemic probability possibilities of error, given available evidence. Finally, the TAXI DRIVER case. In the TAXI DRIVER case, we attributed to Johnny the intellectual virtue of conscientiousness and argued that moderate pragmatic skepticism can offer a natural explanation of this virtue attribution as well. Johnny is intellectually conscientious because he responsibly sifts the customer information flow and denies accepting it as real knowledge at first instance. The Moorean safety theorist could offer the obvious reply that Johnny knows only in the cases that the belief is safe (and true) and to the extent that he is diligent about whether the information circulated is safe (and true) he is being conscientious. But we need not deny, the Moorean may continue, that he can have (fallible) knowledge. Johnny knows when his belief is safe (and true) and can properly exercise intellectual conscientiousness in sifting information for safe (and true) beliefs.

116  Christos Kyriacou But then, again, the epistemological challenge of the problem of elusive safety emerges: how does Johnny know which beliefs are really safe and true and which appear to be safe and true, if even beliefs that are beyond reasonable doubt safe and true can in due course turn out not really safe and true? If we assume that the beliefs we presume safe and true are indeed safe and true, then it will come to pass that whatever is deemed safe and true is knowledge and we know that often beliefs that were deemed safe and true, given strong evidence at a time, were afterwards proven unsafe and untrue. The problem is important because if we do not have a clear response to this question, we cannot really decide which possibilities of error should be counted as close-by and relevant for knowledge and which are remote and irrelevant for knowledge.30 This is the case because we decide whether a belief is safe and true on the basis of given evidence (incl. background beliefs and assumptions) at a time and this very evidence guides us to take some possibilities of error as relevant and others as irrelevant. As a result, safety is conceptually connected to (ir)relevance of error possibilities.31 But given that relevance is indispensable for knowledge attribution because we first need to decide which possibilities of error are relevant for elimination in order to appropriately attribute knowledge, it follows that the concept of safety is of little help to Mooreans unless we further clarify relevance. That is, unless we clarify what conception of relevance should be relevant in the calculation of which error possibilities are to be eliminated in order to have knowledge and which error possibilities are to be ‘properly ignored’ (in Lewis’ (1996) words).32 Call this second epistemological problem ‘the meta-relevance problem’ for safety accounts. In reality, unless we address the meta-relevance problem we will not be in position to address ‘the problem of elusive safety’. Moreover, unless we address this duo of problems, we may fear that the concept of safety is a merely useful cognitive heuristic to decide which possible worlds are relevant nearby worlds in a way that is question-begging and circular with what we already facilely presume to be knowledge (often beliefs we find commonsensical, platitudinous and intuitive).33 Surely, we do not want our theory of knowledge to be circular and recycle as knowledge whatever is presumed to be knowledge at a time. Such epistemic bootstrapping would have been perhaps psychologically convenient, but is not epistemically advisable because often what is presumed knowledge and it is of high epistemic probability, given evidence at a time, turns out to be unsafe and not knowledge. Call this third problem ‘the circularity of safety knowledge problem’.34 In the next section, I briefly consider three objections to the general argument put forth in favor of moderate pragmatic skepticism and sketch very brief replies. The list of objections is non-exhaustive and the very brief replies are far from definitive, but are meant to assuage some

Moderate Pragmatic Skepticism  117 obvious concerns and add more descriptive detail to the positive argument for moderate pragmatic skepticism.

6 Three Objections and Brief Replies Objection 1: All four cases rely on a simple misunderstanding of safety. In all cases the belief in question did not amount to knowledge because it was clearly not safe; at least it was not safe when we are careful not to misunderstand what safety is. In the NEWTON, DARWIN, SHERLOCK and TAXI DRIVER cases, the beliefs in question were not really safe because there was relevant counterevidence that should not have been ignored (the perihelion of mercury, evidence for natural selection, evidence that Strakke was the culprit, evidence that customers are often unreliable, etc.). That is, in a relevant nearby world the belief in question was false and, therefore, the belief was unsafe and not knowledge. Thus, the safety theorist has an easy reply. Reply: In all four cases the epistemic probability of the belief in question, given evidence at the time, was sufficiently low to render it epistemically improbable. If this is so, then by safety’s own lights (at least as stipulated here) such possibilities of error should be considered irrelevant and should be properly ignored. The problem of elusive safety and the related meta-relevance problem then emerge because the question what is really safe and which notion of relevance is relevant become potent (and the concomitant problem of the circularity of safety knowledge). To show that in the four cases the corresponding beliefs were not safe and therefore were not cases of knowledge—which is the desirable intuitive result—the Moorean will have to argue that high epistemic probability beliefs might not be safe because low epistemic probability error possibilities might still be relevant for consideration and elimination if we are to correctly attribute knowledge. In response, Mooreans might distinguish between subjective epistemic probability and objective epistemic probability. Subjective epistemic probability is the credence an agent has for a belief, given the agent’s possessed evidence. Objective epistemic probability is the credence an agent ought to have for a belief, given the agent’s possible accessible evidence. Mooreans might then point out that safety requires high objective epistemic probability and, therefore, in the four cases there was relevant counterevidence that should not have been ignored (the perihelion of mercury, evidence for natural selection, evidence that Strakke was the culprit, evidence that customers are often unreliable, etc.) because this evidence was, in principle, possible and accessible. Given this clarification about the kind of epistemic probability required for safety, in all four cases the corresponding beliefs do not amount to knowledge because they come out unsafe. They were of high

118  Christos Kyriacou subjective epistemic probability but not of high objective epistemic probability, given the ignored and possibly accessible counterevidence (and it is the latter epistemic probability that is required by safety). There are at least two inter-related problems with this Moorean rejoinder. First, we should not reasonably expect fallible, less-thanideal-agents to be able to access all possible accessible evidence. 35 In the DARWIN case, there was evidence of natural selection, but it took a genius to notice. In the NEWTON case and the perihelion of Mercury, it took another genius, Einstein, to construct a theory that properly understands and evaluates the evidential weight of the anomaly. In the SHERLOCK case, it takes an astute detective to resist jumping to conclusions, form another hypothesis, uncover more evidence and properly evaluate it and not many of us can emulate him intellectually. In the TAXI DRIVER case, things are not as cognitively demanding for ordinary agents as in the other cases, but Johnny can realize that there is evidence against random customer information only by exhibiting virtues such as conscientiousness and definitely many of us lack such virtues. Thus, the worry is that we may be stipulating safety in an over-demanding way for a distinctively Moorean position that does not want to raise the standards of knowledge too high. By implication, this would imply that we have less knowledge than we ordinarily think we have. Yet, this seems the wrong prediction for a Moorean. Second, objective epistemic probability requires ‘the agent’s possible accessible evidence’ and we may worry that the standards of knowledge are now set so high that are not only unMoorean in character but also verging on some form of high-standards skepticism, or even infallibilism. For if the standards of knowledge are pushed high enough, then we are all but in the hands of some form of skepticism. Thus, in trying to rescue safety from the canvassed problems we might have stepped into a slippery slope that lands us into some form of skepticism. Of course, much more could be said about this first objection. But at any rate, it remains to be shown whether the Mooreans can address the objection in a way that better explains the attribution of intellectual virtues/vices in the canvassed cases. Objection 2: There are already replies in the Moorean literature to the meta-relevance problem you pose that would help address the elusive safety problem (and by implication avoid the circularity of safety knowledge problem). For instance, take two proposals that appeal to the notion of normality to understand relevance. Rysiew (2001:488) has suggested that ‘the relevant alternatives are fixed by what we (normal) humans

Moderate Pragmatic Skepticism  119 take to be the likely counter-possibilities to what the subject is said to know’. More recently, Beddor and Pavese (2020:69–70) suggest that ‘the worlds that are relevant for assessing…include the nearest worlds where conditions are normal’ (2020:70; their emphasis). Normality (of agents or worlds) is a fairly intuitive notion that helps us explicate relevance in a natural and non-ad-hoc way. In principle, we can then understand safety in terms of normality and resolve difficulties that spring from the notion of high epistemic probability. Reply: Relevance, at least as intended in accounts of safety, is a normative concept and notoriously normative concepts are in need of elucidation. Trivially, relevant error possibility is what we ought to consider and eliminate as a possibility of error. But what is it that we ought to consider and eliminate as a possibility of error? Obviously, salience of error possibilities will not do as an answer to the question because as almost everybody accepts salience and relevance should not be conflated. Relevant error possibilities could not be salient and salient error possibilities could be irrelevant (again, easy examples come to mind).36 After all, salience is a psychological-descriptive concept and relevance is a normative concept and we know that normative concepts tend to resist reduction to descriptive ones (recall Moore’s (1903) ‘open question argument’ and the literature that has followed). However, the appeal to normality does not help because the word ‘normality’ is ambiguous between two different meanings (statisticaldescriptive and normative) and neither of the two helps.37 First, if we understand normality in the statistical sense of what most people do or find natural or proper to do, this is of little help because just because most people find something normal it does not follow that we ought to do it (recall Hume’s ‘law’ of the ‘is/ought gap’). Counterexamples come easy to mind (e.g. cannibalistic practices in tribes). Second, if we understand normality in the normative sense, which is the sense that we ought to opt for in order to hope for progress on the matter, then it is viciously circular and vacuous because it means that we ought to consider what we ought to consider. Yet this is entirely uninformative and still does not tell us what we ought to consider. Thus, the problem is important, and a response to it is far from trivial. Objection 3: It seems that there are cases of ordinary knowledge attribution that naturally comport with Moorean invariantism and would be intellectually vicious to deny as true knowledge attributions by the lights of moderate pragmatic skepticism. Take Moore’s famous hands case: perceiving we have hands seems to license knowing that we have hands. Resisting such profound knowledge attribution would, indeed, take special intellectual vices such as being intellectually prickly, over-hesitant and overcautious. Accepting such knowledge attribution would seem to comport with common sense, if not with being sane at all.

120  Christos Kyriacou Reply: This is true as far as it goes but the presented four cases also suggest that Moorean invariantism does not naturally address them. If there is conflicting evidence coming from the direction of attribution of intellectual virtues/vices from different cases, then it is at least open who is right and who is wrong in the theory of knowledge. Something will have to give way and it is not clear that this would be moderate pragmatic skepticism. So, the jury is out. Moreover, from the perspective of moderate pragmatic skepticism, we might provide a debunking explanation of the appearance of intellectual vices involved in ordinary Moorean knowledge. That is, we may grant that ordinary perceptual experience, memorial experience, etc. is of high epistemic probability, given evidence, but yet not real knowledge and this could help explain (away) the appearance of intellectual vices involved in denying that we have commonsensical knowledge.

7 Conclusion I have quickly sketched and motivated a specific version of skeptical invariantism, namely, moderate pragmatic skepticism and argued that it can explain paradigmatic cases of attribution of intellectual virtue/vice. I then examined how the contender of Moorean invariantism fares with regard to the presented cases and argued that it runs into difficulties that reveal three inherent epistemological problems with the modal condition of safety. I identified these problems as the problem of elusive safety, the meta-relevance problem and the problem of the circularity of safety knowledge. I concluded with very brief replies to three objections.38

Notes 1 More recently, skeptical positions of various kinds have been enjoying something of a renaissance in the literature. See Kyriacou (2017a, 2019, 2020, forthcoming-a, forthcoming-b), Stoutenburg (2017, this volume), Rinard (forthcoming), Fassio (forthcoming, this volume), Climenhaga (this volume), Hannon (this volume) and Wallbridge (this volume). 2 Skeptical positions promise to address the Gettier problem, the lottery puzzle, the value problem, concessive knowledge attributions, the preface paradox, the dogmatism paradox, assertion and practical reasoning and more. They can also have fairly sophisticated things to say about semantics/pragmatics and related linguistic issues (cf. Kyriacou 2019). For various sorts of skepticisms, see Unger (1975), Fogelin (1994), Schaffer (2004), Conee (2005), Cappelen (2005), Frances (2005), Davis (2007), BonJour (2010), Dodd (2011), Kyriacou (2017a, 2019, 2020), Stoutenburg (2017) and Climenhaga (this volume) for discussion. 3 See Kyriacou (2017a, 2017b, 2019, 2020). 4 See Conee (2005:52) and BonJour (2010:59–60) for a similar stipulation. Dinges (2016:6) is also kind of sympathetic to such a construal of skeptical invariantism. This stipulation of skeptical invariantism is modest because it is permissive enough to allow for non-global forms of skepticism.

Moderate Pragmatic Skepticism

121

5 Of course, not all skeptics think that the source of skepticism is the demandingness of standards. See Pritchard (2021), Rinard (forthcoming) and Hannon (this volume) for some discussion. 6 This is the case because, according to infallibilism, knowledge requires epistemic probability 1 of the belief on the given justification and, as we very rarely satisfy such a stringent constraint of epistemic probability, it follows that we rarely have knowledge. Thus, skeptical invariantism is naturally conjoined to infallibilism. See Dodd (2011) for an explication of infallibilism along such probabilistic lines. 7 See note 2 for references. 8 See Unger (1975), Williamson (2000), Hawthorne (2004) and Stanley (2005). 9 For detailed discussion, see Kyriacou (2020). 10 Pace Hannon (2019), who argues against this assertion. 11 We may add that it is also ad hoc to claim that we can know only that, namely, that there is no knowledge. Why only that and not more? It seems unjustified. 12 See Nagel (1997), Rowland (2013) and Kyriacou (2021). 13 For discussion of epistemic necessities (or ‘fixed points’), such as the factivity condition, see Kyriacou (2018, 2021). 14 Of course, this implies that skepticism does not penetrate at the level of justification (pace Unger (1975) and Fogelin (1994)). There are justified (or rational) beliefs, although there may be little or no knowledge. See Kyriacou (2020, 2021) for discussion. For an explication and defense of the abductive methodology in philosophy, see Williamson (2007, 2018). 15 See for discussion Kyriacou (2019). 16 Dinges (2016:4) proposes that we understand what is to be close enough (via substitutional implicature) to what is said in the following way: in order to be close enough to knowing that p, one must satisfy all the conditions for knowledge…except the justification condition…for example… S is close enough (for the purposes of the low standards case) to knowing that p iff p, S believes that p and S can rule out all likely alternatives to p. (His emphasis) For a safety-based understanding, see Kyriacou (2017a). 17 See, for example, Zagzebski (1996), Roberts and Wood (2007), Sosa (2007) and Baehr (2011). See also Turri (2017) for an overview and discussion. 18 The cases are not meant to be historically accurate. 19 See Kahneman (2011) for the overconfidence bias. 20 See Zagzebski (1996), Sosa (2007) and Baehr (2011). 21 See Kyriacou (2020) for detailed discussion. 22 See Williamson (2000), Sosa (1999, 2007), Greco (2011) and Pritchard (2005, 2012). 23 For Williamson (2000), in particular, safety is a structural feature of knowledge and inter-definable with knowledge. Unlike Pritchard (2012), Williamson (2000) does not take his account of knowledge to be a traditional—in necessary and sufficient conditions—analysis of the concept of knowledge. 24 The exact understanding of the possible worlds to be ruled out in order to have knowledge (e.g. all relevant nearby worlds or most of them) depends on the exact details of the position of each advocate. Here I stick to Williamson’s (2000:147) position. For the fine-grained differences between some of the prominent safety theorists, see Rabinowitz (2011).

122  Christos Kyriacou 25 Of note, is that the notion of evidence here is understood in broadly externalist contours, as for instance in Greco (2011). Prominent safety theorists are all of externalist sympathies. 26 See Pritchard (2012) and Beddor and Pavese (2018) for discussion. 27 Various Mooreans acknowledge that safety is inherently vague, such as Williamson (2000) and Beddor and Pavese (2018:69). 28 In a moment, I will stipulate this problem as ‘the problem of the circularity of safety knowledge’. 29 See Kappel (2010) and Climenhaga (this volume) for a similar point and discussion. For some discussion of concessive knowledge attributions from an infallibilist, skeptical perspective, see Dodd (2011) and Kyriacou (2017a). 30 Stoutenburg (2017) argues for what he calls ‘epistemic blindness of error possibilities’, namely, that agents seem to be blind as to which error possibilities are relevant for elimination. If he is right, this lends further support to ‘the problem of elusive safety’ because it makes it harder to discern which error possibilities should be taken into consideration for elimination. 31 See also Pritchard (2002) for the same point. 32 As Baumann (2008) has also noted, little work has been done on how exactly we rank similarity/proximity of possible worlds to the actual world. This is a very important question because it relates to the question of how we demarcate between relevant and irrelevant worlds as error possibilities. 33 Relatedly, Piller (2019) has argued that safety fails to capture our epistemic concerns and, therefore, is irrelevant for epistemological purposes. He also provides a diagnosis why many epistemologists have fallen for safety. 34 See Vogel (2000) for a similar bootstrapping problem with regard to reliability. Given that safety and reliability are closely inter-connected concepts, it is not particularly surprising that similar worries emerge for safety as well. 35 For some discussion of the problem of relevant evidence, see Cuneo and Kyriacou (2018) and Kyriacou (2019). 36 See Stoutenburg (2017) for some discussion. 37 See Plantinga (1993) for the distinction between the statistical-descriptive and the normative senses of normality. 38 I would like to thank Christian Piller, Nevin Climenhaga, Ioannis Trisokkas, Christos Panagides, William Wells and the audience of a talk at Complutense University, Madrid for helpful comments.

References Baehr, Jason. (2011). The Inquiring Mind. Oxford, Oxford University Press. Baumann, Peter. (2008). ‘Is Knowledge Safe?’ American Philosophical Quarterly 45(1): 19–30. Beddor, Bob and Pavese, Carlotta. (2020). ‘Modal Virtue Epistemology’. Philosophy and Phenomenological Research CI(1): 61–79. BonJour, Lawrence. (2010). ‘The Myth of Knowledge’. Philosophical Perspectives 24: 57–83. Cappelen, Herman. (2005). ‘Pluralistic Skepticism’. Philosophical Perspectives, 19: 15–39. Climenhaga, Nevin. (in this volume). ‘A Cumulative Case Argument for Infallibilism’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge.

Moderate Pragmatic Skepticism  123 Conan Doyle, Arthur. (2008). Sherlock Holmes. Selected Stories. Oxford, Oxford University Press. Conee, Earl. (2005). ‘Contextualism Contested’. In M. Steup and E. Sosa (eds.), Contemporary Debates in Epistemology. Oxford: Blackwell. pp. 60–83. Cuneo, Terence and Kyriacou, Christos. (2018). ‘Defending the Moral/Epistemic Parity’. In C. McHugh et al. (eds.), Metaepistemology. Oxford, Oxford University Press, pp. 27–45. Davis, Wayne. (2007). ‘Knowledge Claims and Context: Loose Use’. Philosophical Studies 132(3): 395–438. Dinges, Alexander. (2016). ‘Skeptical Pragmatic Invariantism: Good, but Not Good Enough’. Synthese 193(8): 2577–2593. Dodd, Julian. (2011). ‘Against Fallibilism’. Australasian Journal of Philosophy 89(4): 665–685. Fassio, Davide. (Forthcoming). ‘Moderate Skeptical Invariantism’. Erkenntnis 1–30. Fogelin, Robert. (1994). Pyrrhonian Reflections on Knowledge and Justification. Oxford, Oxford University Press. Frances, Brian. (2005). Scepticism Comes Alive. Oxford, Oxford University Press. Greco, John. (2011). Achieving Knowledge. Cambridge, Cambridge University Press. Hannon, Michael. (2019). ‘Skepticism: Impractical, Therefore Implausible’. Philosophical Issues 29(1): 143–158. ———. (this volume). ‘Skepticism, Fallibilism and Rational Evaluation’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Hawthorne, John. (2004). Knowledge and Lotteries. Oxford, Oxford University Press. Kahneman, David. (2011). Thinking, Fast and Slow. New York, Random House. Kappel, Klemens. (2010). ‘Expressivism about Knowledge and the Value of Knowledge’. Acta Analytica 25(2): 175–194. Kyriacou, Christos. (2017a). ‘Bifurcated Sceptical Invariantism: Between Gettier Cases and Saving Epistemic Appearances’. Journal of Philosophical Research 42: 27–44. ———. (2017b). ‘Assessment Sensitivity: Relative Truth and Its Applications’. John MacFarlane. Oxford, Oxford University Press. 2014. Dialectica 71(2): 322–332. ‘Review of J. MacFarlane’s Assessment Sensitivity. ———. (2018). ‘From Moral Fixed Points to Epistemic Fixed Points’. In C. Kyriacou and R. McKenna (eds.), Metaepistemology: Realism and Antirealism. London, Palgrave Macmillan, pp. 71–96. ———. (2019). ‘Semantic Awareness for Skeptical Pragmatic Invariantism’. Episteme Online First: https://link.springer.com/article/10.1007%2Fs 12136-019-00414-z ———. (2019). ‘Evolutionary Debunking: The Milvian Bridge Destabilized’. Synthese. 196(7): 2695–2713. ———. (2020). ‘Assertion and Practical Reasoning, Fallibilism and Pragmatic Skepticism’. Acta Analytica 1–19. Online First: https://link.springer.com/ article/10.1007%2Fs12136-019-00414-z

124  Christos Kyriacou ———. (2021). ‘Debunking, Theoretical Indispensability and Irreducible Epistemic Rationality’. In Diego Machuca (ed.), Evolutionary Debunking Arguments, Routledge. ———. (forthcoming-a). ‘Varieties of Skeptical Invariantism I’, Philosophy Compass. ———. (forthcoming-b). ‘Varieties of Skeptical Invariantism II’, Philosophy Compass. Lewis, David. (1996). ‘Elusive Knowledge’. Australasian Journal of Philosophy 74(4): 549–567. McKenna, Robin. (this volume). ‘A (Partial) Defence of Moderate Skeptical Invariantism’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Moore, G. E. (1903). Principia Ethica. Dover Publications. Nagel, Thomas. (1997). The Last Word. New York, Oxford University Press. Piller, Christian. (2019). ‘Beware of Safety’. Analytic Philosophy 60(4): 1–29. Plantinga, Alvin. (1993). Warrant and Proper Function. Oxford, Oxford University Press. Pritchard, Duncan. (2002). ‘Contemporary Skepticism’. Internet Encyclopedia of Philosophy. https://www.iep.utm.edu/skepcont/ ———. (2005). Epistemic Luck. Oxford, Oxford University Press. ———. (2012). ‘Anti-Luck Virtue Epistemology’. The Journal of Philosophy 109(3): 247–279. ———. (2021). ‘Skeptical Invariantism and the Source of Skepticism’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Rabinowitz, Dani. (2011). ‘The Safety Condition of Knowledge’. Internet Encyclopedia of Philosophy. https://www.iep.utm.edu/safety-c/ Rinard, Susanna. (Forthcoming). ‘Pragmatic Skepticism’. Philosophy and Phenomenological Research. Roberts, Robert and Wood, Jay. (2007). Intellectual Virtues. Oxford, Oxford University Press. Rowland, Richard. (2013). ‘Moral Error Theory and the Argument from Epistemic Reasons’. Journal of Ethics and Social Philosophy 7(1): 1–24. Rysiew, Patrick. (2001). ‘The Context-Sensitivity of Knowledge Attributions’. Nous 35(4): 477–514. Schaffer, Jonathan. (2004). ‘Skepticism, Contextualism and Discrimination’. Philosophy and Phenomenological Research 69(1): 138–155. Sosa, Ernest. (1999). ‘How to Defeat Opposition to Moore’. Philosophical Perspectives 13: 141–153. ———. (2007). A Virtue Epistemology. Oxford, Oxford University Press. Stanley, Jason. (2005). Knowledge and Practical Interests. Oxford, Oxford University Press. Stoutenburg, Gregory. (2017). ‘Strict Moderate Invariantism and Knowledge-Denials’. Philosophical Studies 174(8): 2029–2044. ———. (this volume). ‘Skeptical Invariantism, Reconsidered’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Turri, John, Alfano, Mark and Greco, John. (2017). ‘Virtue Epistemology’. In E. N. Zalta (ed.), Stanford Encyclopedia of Philosophy. https://plato.stanford. edu/entries/epistemology-virtue/ Unger, Peter. (1975). Ignorance. Oxford, Oxford University Press.

Moderate Pragmatic Skepticism  125 Vogel, Jonathan. (2000). ‘Reliabilism Leveled’. Journal of Philosophy 97(2): 607–623. Wallbridge, Kevin. (this volume). ‘Situationism, Implicit Bias and Skepticism’. In C. Kyriacou and K. Wallbridge (eds.), Skeptical Invariantism Reconsidered. Routledge. Williamson, Timothy. (2000). Knowledge and Its Limits. Oxford, Oxford University Press. ———. (2007). The Philosophy of Philosophy. Oxford, Oxford University Press. ———. (2018). Doing Philosophy. Oxford, Oxford University Press. Zagzebski, Linda. (1996). Virtues of the Mind. Cambridge, Cambridge University Press.

Part III

Arguments for Fallibilist Skepticism

7

In Defense of a Moderate Skeptical Invariantism Davide Fassio

Skeptical invariantism (hereafter, SI) is a form of epistemic invariantism. It holds that the semantic values of knowledge attributions do not vary across contexts of ascription or assessment. The distinctive feature of SI is that, according to this view, the epistemic standards for true knowledge attributions, and thus for knowledge, are particularly demanding and difficult to meet. The consequence is a form of skepticism: people know less than they ordinarily presume or claim to know. We can distinguish different versions of SI on the basis of how they set the standards for knowledge. Different standards determine distinct kinds and degrees of skepticism. The most radical version of SI holds that knowledge requires the highest level of justification. Only conclusive evidence is enough for knowledge.1 Radical versions of SI have several attractive features. For example, they can provide straightforward solutions to traditional epistemological issues such as the Gettier problem and the Lottery paradox.2 However, in recent times these views have not been very popular and are often dismissed as implausible. The main reason is that they seem to make knowledge too hard to attain, leading to an extreme form of skepticism. Since we can have conclusive justification only for very few or no claims, radical SI implies that it is extremely difficult, or even impossible, to know anything. The view clashes with our commonsensical intuition that we actually know a fair number of things, and implies that most of our ordinary knowledge attributions are systematically mistaken. Maybe we know less than we think, but radical SI seems to set standards for knowledge too high. A moderate version of SI could avoid this problem by setting the threshold on the level of justification required for knowledge still quite high, but below the maximum.3 Moderate SI claims that we know less than what we ordinarily presume, yet it maintains that we can have knowledge in a wide range of cases. This position doesn’t sound particularly counterintuitive. The thought that ordinary people do not know things such as that two plus two is four, that we have hands, or that Paris is in France sounds absurd to many.4 However, it doesn’t seem equally implausible that we do not really know several things we ordinarily claim to know, such as that our bike is still in the

130  Davide Fassio parking lot where we left it this morning, or that our bank will be open tomorrow.5 Moderate SI can preserve the idea that knowledge attributions are true in many ordinary conversations, thus avoiding extreme skepticism, while also allowing that a good portion of these attributions is false – the exact proportion depending on how high the threshold for knowledge is set. Versions of moderate SI differ depending on (i) the specific degree of justification at which they set the threshold for knowledge, and (ii) the rationale they provide for setting the threshold in that way. The degree of epistemic support necessary for knowledge should be set in a principled, non-arbitrary way, on pain of incurring familiar Threshold problems.6 Any plausible theory should explain why a degree of justification above that threshold has such special significance, and why it yields to specific dispositions attached to knowledge such as entitling flat-out assertion and terminating deliberation and inquiry. Some philosophers have recently entertained the hypothesis that this threshold should be partially set by the practical circumstances of the subject.7 The basic thought is that meeting the knowledge threshold requires that it be reasonable to ascribe knowledge to the subject across practical contexts, including when stakes on being wrong are particularly high. The hypothesis is usually illustrated by appeal to pairs of scenarios familiar in the contemporary literature, in which the evidential support for a certain belief is held fixed but in one scenario it’s much more important that the belief be right.8 If it seems reasonable to ascribe knowledge that p to the subject when stakes are low, but not when stakes are high, then the subject doesn’t know that p – neither when stakes are high, nor when they are low. Elsewhere I have defended a specific version of this view – which here I shall call Practical Skeptical Invariantism (PSI, for short). Like other pragmatic theories of knowledge such as interest-relative invariantism,9 this view assumes constitutive connections between knowledge and rational action. A rough formulation of the view is as follows10: (PSI) A subject S knows a proposition p only if it would be rational for S to act as if p even if the practical stakes on p were maximally high. (PSI) states a necessary condition on knowledge. This is a modal condition: a subject knows a proposition only if it is rational for her to act on it in all similar possible worlds in which stakes go extremely high, as high as they could, but all else is kept fixed or as little changed as possible. Features kept fixed include in particular truth-relevant factors such as evidential support for the relevant proposition and the reliability of belief-forming processes. So, for example, in order to answer the question whether the Pope is in a position to know that the Vatican is

Moderate Skeptical Invariantism  131 in Rome, we should look at possibilities in which all the Pope’s evidence for that claim remains fixed, but it becomes extremely important for him to be right about whether the Vatican is in Rome – say, his life and that of many other people depended on being right about that. If in these scenarios the Pope were reasonable to act as if the Vatican were in Rome, then PSI would answer positively to the question. According to this view, it’s pretty clear that the extent of our knowledge is more restricted than what we commonly presume. Intuitively, if stakes were particularly high many of us would not be rational to act as if our bike were still in the parking lot where we left it this morning, or as if the next summer we will go on holiday in Florence as we planned. Therefore, according to PSI, we don’t know such things. In general, the view will predict skeptical conclusions about many ordinary knowledge claims concerning the future and several ones about unobserved events. Many opinions based on weak testimony or distant memories will not pass the test either. However, arguably, PSI is compatible with the truth of a quite wide range of ordinary knowledge ascriptions.11 In spite of the high stakes, I wouldn’t consider myself irrational for acting as if there were not a bomb in my car that would explode when I turn the keys. Similarly, it seems perfectly reasonable to take for granted what your name is even when you have to enter that information into an important document, and a mistake would result in a catastrophic financial loss. And it seems reasonable for many of us to act as if Paris were the capital of France, or as if spring followed winter, even if a lot were at stake on being right about such things. This indicates that, according to PSI, we are in a position to know things such as our name; that there is not a bomb in our car; that spring follows winter; and so on. The theory provides a principled rationale for setting the knowledge threshold on justification quite high but below absolute certainty: the threshold lies at the level of epistemic support at which it is rational to act on a proposition no matter how high the practical stakes on that proposition are. The threshold would mark the level of justification at which an agent could safely rely on a proposition irrespective of her practical situation, use it in any deliberation and be entitled to assert it in any practical context. This way of setting the threshold would also explain why knowledge seems to have a special significance and value, being the attitude one could safely trust in any sort of practical circumstance, no matter how high the stakes are. In previous work I have considered several advantages of this view, and defended it from some objections.12 The aim of this paper is to further strengthen my case for PSI. I shall focus my discussion on what I take to be the most serious problem and the best argument for the view. The problem in short is that, as some philosophers have argued, it might not be rational to act on any proposition whatsoever given absurdly high

132  Davide Fassio costs of error. If this is right, then PSI collapses into a radical form of skepticism. In §1, I introduce the problem and show how it could be resisted. My response will also pave the way to some further interesting considerations about the view. In §2, I consider my positive argument for PSI. The argument relies on the claim that judgments in high stakes are generally more reliable than in low stakes. I argue that high stakes knowledge attributions constitute the best test for whether we know something, and that such attributions fit particularly well with PSI’s predictions. In §3 I conclude by summarizing the paper’s upshots and suggesting further directions of research.

1 PSI and Radical Skepticism PSI says that we know something only if we would be rational to act on it even if stakes were maximally high. This may look like a quite demanding condition. Several authors have remarked that it would not be reasonable to rely on any proposition whatsoever if stakes were absurdly high, not even on claims such as that I have hands or that two plus two is four.13 If these philosophers are right, then PSI sets standards for knowledge too high. The view is doomed to collapse into a radical form of skepticism.14 The objection comes in two variants: a naïve intuition-based version and a deeper decision-theoretic one. The naïve version is the most popular in the literature, but also the weakest. It appeals to intuitions about extreme betting scenarios, cases in which some person you don’t know approaches you in the street and proposes a bet on the truth of a proposition where the winning reward is miserable – one penny, one candy – and the costs of losing are extremely high – you will lose everything you have, you and all your dear friends and family will be tortured terribly and enslaved forever.15 Arguably, it seems pretty unwise and unreasonable to accept this sort of gamble, even when the bets bear on propositions of whose truth you are nearly certain.16 This sort of scenario is supposed to show that it is irrational to rely on any proposition whatsoever if stakes are sufficiently high. If this is right, also according to PSI, knowledge is impossible. As several authors have convincingly argued, the naïve version of the objection is weak. We could agree that we shouldn’t take extreme bets on any proposition whatsoever. However, this is not because it is irrational to rely on any proposition if stakes are very high. The reasons why we should refrain from taking extreme bets have to do with peculiar features of betting scenarios. Some such reasons are prudential, having to do with the unusual and odd nature of scenarios involving unknown people proposing very weird bets with big sums for small returns. Any reasonable person would judge people offering such sorts of bets as

Moderate Skeptical Invariantism  133 suspect, unpredictable and untrustworthy.17 Moreover, those who take such kinds of bets will likely develop bad habits and vicious traits of character leading to troubling consequences in future circumstances. A proper lesson we should draw from these cases is that we shouldn’t trade the risk of developing a bad habit for a candy.18 However, the most important reason why the naïve objection fails is strictly epistemic. Facing extreme bets on a proposition almost inevitably affects one’s evidence for that proposition.19 As I have argued elsewhere, the best way of representing such scenarios is within a game-theoretical framework, as a ‘zero-sum’ game: two agents play against each other, one wins what the other loses. In a ‘zero-sum’ game, a rational player should consider the information available to other players as well as the information available to herself. Similarly, when we evaluate whether to accept an extreme bet, we should consider the potential asymmetrical state of information between the other player (the bookmaker) and us. The best explanation why the bookie is willing to enter into such a weird bet is that she has some relevant information that we lack – judging from the bet, information that in the bookie’s lights counts as strong evidence against the proposition. This insight constitutes a defeater of our initial justification for that proposition, which modifies the evidential support we had for that proposition before being offered the bet. But remember that in order to see whether someone meets the PSI condition, we should consider possible high stakes scenarios in which she has the same evidential support for the relevant proposition – according to PSI, the possible circumstances relevant to assess whether a subject is in a position to know are possible worlds in which stakes go extremely high, but all truth-relevant factors are kept fixed. Since extreme betting modifies our initial epistemic position toward a proposition p, it is not a good test of how resilient our actual epistemic position is to high stakes – and hence, according to PSI, of whether we are in a position to know p. 20 In assessing PSI we should set aside intuitions about weird bets and focus on cases in which only the possible costs of error are high. All other factors, both practical and epistemic, should be kept fixed and not interfere with our intuitive judgments. In this respect standard cases in the literature such as DeRose’s Bank cases seem to be much better tests: these are situations in which only the possible costs change (e.g., financial losses), while the evidence is assumed to be the same and no special prudential consideration is involved. One may argue that the intuition-based version of the objection applies even if we focus only on the latter type of case. Some philosophers think that it’s not hard to imagine situations that do not involve explicit bets, but in which the consequences of being wrong are so terrible that it would be better not to rely on any proposition whatsoever, no matter how obvious that is for us, or how much it is supported by the

134  Davide Fassio evidence.21 In response, I must admit that I do not share this intuition. I agree that PSI will likely deliver skeptical verdicts for several propositions we ordinarily claim to know. However, for many other propositions it is not obvious to me that it would be unreasonable to act on them even if stakes were extremely high, but all other factors were kept fixed. On the contrary, I would contend that it seems perfectly reasonable to act on many propositions even if stakes were very high. When a mistake means a huge financial loss, it’s unreasonable for Hannah to act as if the bank were open on Saturday on the basis of a flimsy memory; but in the same circumstances I wouldn’t criticize her for acting as if the next day is Saturday, as if her name is Hannah, or as if she is not in Uzbekistan now, even if stakes are equally high on these propositions. 22 I am aware that some philosophers will not share my intuitions. However, at this stage my aim is not to ascertain which intuitive judgments are right. My more modest claim is that, when it comes to judgments about cases in which only stakes change, intuitions alone are inconclusive and insufficient to settle the matter. They do not unquestionably suggest that when stakes are extremely high it is unreasonable to act on any proposition whatsoever. Later in this section I shall provide a principled argument to the effect that, given a few plausible assumptions about the way we frame values, it is rational to act on propositions for which we only have a fallible (though quite strong) epistemic support even when stakes on these propositions are maximal (see point 4 below). Intuitive judgments about cases do not vindicate the claim that it is unreasonable to act on any proposition in super-high stakes. The naïve intuition-based version of the objection fails. But the objection comes in a second version, much more serious and difficult to dodge, which doesn’t rely on intuitions but on principled theoretical considerations. The objection can be easily framed within a decision-theoretic framework. In a nutshell, the problem is that, for any proposition p for which there is a non-zero chance of error, if stakes on p can be raised indefinitely (e.g., by raising the disvalue of being wrong), there will necessarily be a stakes level at which acting as if p will not maximize expected utility, and thus, it will not be rational. According to standard decision theory an action is rational when it maximizes expected utility. 23 The expected utility of an action is obtained by first multiplying the probability of a possible state obtaining for the utility of performing that action given the occurrence of that state, repeating this process for each possible state, and summing the results. The rational action is the one with the highest expected utility. Within this framework, we can represent changes in stakes on a proposition p by keeping the alternative actions, possible states, and probability assignments fixed, and manipulating the utilities of the possible outcomes. The greater the disutility of acting as if p and being wrong

Moderate Skeptical Invariantism  135 (not-p), and the lower the utility of acting as if p and being right (p), the higher will be the stakes on acting as if p. Consider a representation of a simple two-option and two-state scenario:

Acting as if p Not acting as if p

p

not-p

m+ m−

M− M+

The two available actions are acting as if p and refraining from doing it; m+, m−, M+, and M − are the utilities of the possible outcomes; p and not-p are the possible states. Let’s also assume that given our evidence the probability of p occurring is x – and correspondingly the probability of not-p is 1 − x. In this scenario, acting as if p will maximize expected utility if and only if the following equation is validated: + − − + (*) (x × m ) + [(1 − x) × M ] > (x × m ) + [(1 −   x) × M ] In this type of scenario, stakes for acting on p will be higher when the utility for acting as if p and being right is very small (e.g., one candy), but the disutility for acting as if p and being wrong is very high (bankruptcy, death…). In other terms, we have high stakes when: • •

the utility of acting as if p and being right exceeds the alternative choice only by a small difference: |m+ − m−| is very small; but the disutility of acting as if p and being wrong exceeds the utility of the alternative choice by a big difference: |M+ − M −| is very big.

According to the objection, we can arbitrarily increase the objective disvalue of a certain outcome (e.g., of acting as if p and being wrong), thereby infinitely increasing the disutility of that outcome. If this is the case, as long as there is a non-zero chance that an action could lead to that outcome, there will be utility assignments for which that action will not maximize expected utility. 24 In the case above, assume that there is a non-zero chance that not-p (i.e., x < 1). Then M − could be made arbitrarily low by adding disvalue – and the difference between M − and M+ arbitrarily big. The result is an assignment of utilities such that the equation (*) is falsified. In that circumstance, it will not be rational to act as if p. This result can be avoided only if the chance that not-p obtains is zero. In that case, and only in that case, it will be rational to act as if p no matter how high the stakes are. Thus, assuming that for any proposition we cannot completely exclude even the smallest chance of error, PSI collapses into the most radical form of skepticism.

136  Davide Fassio How could an upholder of PSI address this objection? Rejecting standard decision theory doesn’t seem to be an option. This is by far the most successful and powerful theory of rational decision-making we have, whose successes extend far beyond philosophical theorizing.25 The solution I favor rejects a hidden premise in the argument, namely, that by arbitrarily increasing the objective disvalue of a certain outcome, we thereby infinitely increase the disutility of that outcome. Perhaps we can imagine indefinitely adding objective disvalue to a possible outcome, but the result is not an infinite increase of disutility. The reason has to do with the way reasonable human beings evaluate things. In general, the utility of a type of thing monotonically decreases the more we have of it, until a point at which we become completely indifferent to acquiring more of it – the utility of $1,000 for a poor person is much higher than for a rich man; Bill Gates may well be completely insensitive to the value of $1,000. Similarly, disutility monotonically decreases the bigger the objective loss is. According to a prominent view in decision theory, the utility function of reasonable persons is a nonlinear bounded function, concave in the positive region and convex in the negative one. These properties reflect the decreasing marginal utility and disutility of valuable things such as wealth, commodities and health. If utility functions are bounded, then there are specific maximal and minimal boundaries at which utility and disutility will converge. Many decision theorists assume the existence of such boundaries for important theoretical reasons. The assumption of finite upper and lower bounds on utilities can provide simple and straightforward solutions to traditional problems of the theory such as the St Petersburg paradox and the Pasadena and Altadena games.26 Furthermore, unbounded utilities would violate the von Neumann-Morgenstern continuity axiom, and would thus be incompatible with well-known axiomatizations of the theory. 27 Moreover, a wide body of psychological studies in descriptive decision theory confirms that people’s choices conform to a decreasing marginal utility function with bounded margins.28 While such studies model real-life choices, it would be odd if rationality standards systematically recommended choices that radically diverge from what normal intelligent people do in normal circumstances. Such divergence would constitute a straightforward violation of ‘ought’ implies ‘can’ principles. If utility functions are bounded, it is not possible to infinitely increase the disutility of an outcome. This is an important step toward solving the problem, but it is not enough. Infinite increase in stakes does not depend only on maximal and minimal boundaries on utilities, but also on the existence of a minimum discrete difference in utility. To see why this is the case, we have to look again at (*). The validation of this equation for some x < 1 doesn’t depend exclusively on discrete values for M+ and M −, but also on the existence of a finite value assignment for |m+ − m−|.

Moderate Skeptical Invariantism  137 This is clearer if in (*) we take x as a variable and solve the equation. We obtain the following theorem 29:

(T *)

x>

M + − M − M + − M −   +  m + − m −

(T*) states the specific probabilistic value that x should have for acting as if p to maximize expected utility, and thus be the rational option no matter how high the stakes are. It is easy to verify that this probability can be below 1 only provided that (i) there are bounded maximal and minimal utility values M+ and M −, and (ii) there is a minimal discrete difference in utility values |m+ − m−|. If this minimal difference can be arbitrarily low, no probability value for x below 1 will be enough to validate (*). We would face a new version of the problem. However, it is reasonable to assume that condition (ii) also is satisfied, at least for normal human beings. Indefinitely small amounts of utility require infinitely fine-grained discriminatory capacities that no human being with limited powers of discrimination could possess. As a matter of fact, normal human beings are insensitive to super-small differences in objective value: I don’t think many would care if at the end of the year their revenues were 1$ higher or lower.30 In conclusion, my tentative response to the challenge is that, assuming quite plausible constraints on utilities, for any proposition p it could be rational to act as if p at any stake even if the probability of p on our evidence is below 1. We do not need a maximal degree of epistemic support to be rational in relying on a proposition no matter how high the stakes are. (T*) indicates the probability threshold for rationally acting as if p no matter the stakes – precisely what is necessary to meet the (PSI) condition on knowledge. Given the above constraints on utilities, this threshold doesn’t require a maximal level of justification. The collapse of PSI into a radical form of skepticism is avoided. How skeptical or moderate PSI is directly depends on the value of x in (T*), which in turn depends on specific maximal and minimal limits on utilities, but this threshold may well be much lower than probability 1.31 While my response isn’t a conclusive refutation of the objection, it shows a reasonable way in which the challenge can be resisted. Moreover, the present hypothesis also has several interesting consequences: 1

An important result is that the problem for PSI discussed in this section doesn’t ultimately depend on epistemological considerations such as the subject’s degree of uncertainty or other weaknesses in her epistemic position, but on specific conditions about the way in which we reasonably frame and discriminate our utilities, such as how risk-averse we are and how fine-grained our utility scales are.

138  Davide Fassio 2

3

4

If my hypothesis is correct, the above discussion also shows a way in which PSI could avoid another familiar problem affecting fallibilist accounts of knowledge – views for which the threshold on justification required for knowledge falls below the maximal level. According to the Threshold problem, there is no principled, non-arbitrary way to fix this threshold.32 (T*) provides a straightforward answer to the challenge: the threshold on p is fixed by the degree of evidential support necessary for rationally acting as if p even if stakes were maximally high.33 This threshold is a factor of specific boundaries and limits on rational utility scales. If we assume that normal people’s attitudes can be indicative of reasonability conditions, this threshold could in principle be measured empirically, by testing people’s risk-aversion and utility discrimination. Such empirical work could provide further evidence relevant to assessing this view. Another interesting upshot of my hypothesis is that, if it could be rational for different people to have different utility boundaries and minimal utility gaps, they would also have different thresholds for knowledge. People have different attitudes toward risk. Some are more scrupulous and cautious, others more prone to risk. These different attitudes could naturally modify utility boundaries, thus affecting personal thresholds for knowledge. The result is that how much a person would be in a position to know would also partially depend on her character traits, personalities, and attitudes toward risk.34 This feature seems to be commonly reflected in our ordinary knowledge attributions: cautious people are much less prone to self-ascribe knowledge than risk-seeking ones. Given a few plausible assumptions about rational utility scales, the probability threshold (T*) falls below absolute certainty. This means that a fallible (though quite strong) epistemic support is sufficient for rationally acting on p even when stakes on p are maximal. This provides further reasons to think that the intuition-based version of the objection discussed at the beginning of this section is invalid. In particular, my result provides a further response to those who think it is intuitive that stakes could be raised so high that it would be irrational to rely on any proposition whatsoever. These intuitions depend on contentious assumptions about the way we frame utilities. If our utility scales are bounded in the way I argued above, these (alleged) intuitions are not just questionable, but mistaken.

2 An Argument for PSI In this section I advance a new argument for PSI. The argument is primarily directed to other forms of invariantism – in particular moderate and standard interest-relative invariantism, but also radical SI. I am

Moderate Skeptical Invariantism  139 aware that the current formulation of the argument will be less persuasive against variantist theories such as contextualism and relativism, which however I deem less plausible for independent reasons which have been already widely discussed in the literature. The argument moves from the premise that judgments in high stakes circumstances are more reliable than in low stakes ones. Judgments about knowledge do not seem to constitute an exception in this regard. However, high stakes knowledge judgments fit well with PSI’s predictions about whether someone knows something or not. The conclusion is that PSI provides more reliable verdicts, and is thus to be preferred to alternative views which deliver incompatible predictions. Here is a more precise formulation of the argument: 1 2 3 4

C

Judgments about whether a condition C obtains are normally more reliable when a lot is at stake on being wrong. Judgments about whether an epistemic condition C (e.g., knowledge, ignorance) obtains are normally more reliable when a lot is at stake on being wrong (from 1). High stakes judgments about whether someone knows/knew something are more reliable than low stakes judgments (from 2). PSI is the view whose verdicts about knowledge better fit with high stakes judgments about whether someone knows/knew something. PSI provides more reliable verdicts than alternative views, and thus, it is more plausible (from 3 and 4).

(1) is the main premise of the argument. (2) is a natural extension of the general claim in (1) to specific judgments about epistemic conditions. I shall argue that (2), besides receiving support from (1), is also independently plausible. (3) is a straightforward consequence of (2). (C) follows from (3) and (4). In the next subsections, I shall defend the key premises and steps in the argument. 2.1 Premise 1 According to (1), normal human beings’ judgments tend to be more reliable in high stakes circumstances, when the perceived costs of being wrong are high and it is particularly important to avoid errors. Mistakes are possible also in high stakes, but are less frequent. It seems eminently plausible that we reach more accurate judgments when we have much more to lose. For instance, when we proofread a draft we tend to be more sensitive to mistakes and typos just before an important submission deadline, when stakes are higher, than in our initial readings when the costs of a mistake are still quite low. This

140  Davide Fassio commonsensical consideration is also reflected in our attitudes toward trusting the testimony of others: other things being equal, we consider more trustworthy about p someone for whom a lot is at stake on being right than one to whom it is utterly unimportant whether p. Claim (1) is not particularly surprising either. In general, we should expect people to be particularly conscientious and careful in minimizing the risk of error when the costs of being wrong are particularly serious, and judgments based on careful ponderation are generally regarded as more accurate and reliable. Contemporary psychological literature provides further overwhelming support for (1). A wide range of studies indicates clear interconnections between perceived high stakes and higher degrees of accuracy. Part of these empirical studies is already familiar in the epistemological debate from the work of Jennifer Nagel.35 These studies show that in general, other things being equal, high stakes subjects – subjects for which there is greater reward in accurate judgments or greater cost in being inaccurate – tend to invest more cognitive efforts, seeking for more evidence and processing possessed information more thoroughly than their low stakes counterparts. These efforts normally produce increased accuracy in their judgments. Several cognitive biases are attenuated when subjects are in high stakes contexts. Moreover, high stakes subjects tend to adopt more systematic and controlled cognitive strategies, which result in increased reliability of the epistemic outputs.36 2.2 Claim 2 The previous considerations in support of (1) can be naturally extended to judgments about epistemic conditions such as knowledge and ignorance.37 We tend to be more reliable about whether we know something when the costs of error are high and it is particularly important not to be wrong. Mistakes about our epistemic states are always possible, but they are less frequent when stakes are high.38 If high stakes subjects tend to invest more cognitive efforts, adopt more systematic and controlled cognitive strategies, be less affected by biases, and process their information more thoroughly, their judgments also will typically be more accurate in the specific circumstances in which such judgments bear on their own epistemic states. Moreover, we should expect people to be particularly careful about knowledge judgments when stakes are high, since what we take to know is often related to what we are disposed to take for granted in our deliberation. So the more important it is to minimize the risk of error in our decisions, the more careful and accurate we should expect our knowledge judgments to be.39 (2) is independently supported by a further range of considerations. Several authors have remarked that high stakes knowledge judgments carry with them a feeling of enlightenment and a sense of discovery. For

Moderate Skeptical Invariantism  141 example, according to Cohen, there is no doubt that when one is in a high standards context, “one has a feeling of enlightenment regarding the correct application of the predicate in question. We feel as if we are seeing the truth of the matter that has, up until that point, eluded us” (2005: 58). Similarly, according to Cappelen, such contexts are “accompanied by a sense of discovery – by a sense of having understood something new about our epistemic condition” (2005: 19).40 Feelings of enlightenment and sense of discovery are often indicative of the acquisition of new, better perspectives about the truth of a certain matter. Of course, in some circumstances such reactive attitudes may be misleading. Sometimes a feeling of enlightenment may accompany a state of delusion, and a sense of discovery a false conclusion. Nonetheless, it seems reasonable to take such responses as prima facie clues that we discovered something new and reached a better epistemic position about a certain matter. Indeed, a sense of discovery is the standard emotional response to a discovery, and a feeling of enlightenment is normally a perception of an enlightened state. This is precisely why we name these responses in this way. The hypothesis that subjects in high stakes contexts tend to be more reliable in assessing whether someone knows or knew something is further corroborated by the following data: •

•

Easy epistemic ascent – difficult epistemic descent. Several philosophers have remarked that it is relatively easy to lose the disposition to self-ascribe knowledge when the stakes rise, but it is more difficult to regain such disposition when stakes go down to normal.41 More generally, dispositions to self-ascribe knowledge or lack of it in high stakes subjects seem to be much more stable and resilient than those of their low stakes counterparts. These features are indicative of a deeper and better-grounded awareness of the truth.42 In contrast, a disposition to ascribe properties only in limited contexts in which mistakes are unimportant seems to be a clear indication of insecurity about the obtaining of such properties. Backward self-judgments. When a subject loses a disposition to self-ascribe knowledge when she moves from low to high stakes, she tends to judge herself, not just as not knowing in her present situation, but also as not having known in her past low stakes circumstances. Normally she retracts her earlier knowledge attributions and concedes that they were incorrect, admitting that she didn’t know all along. Moreover, when stakes go down, she typically continues to judge that she doesn’t know and didn’t know in the past.43 These dispositions suggest that the subject has realized that the attitudes she held in low stakes circumstances were mistaken. Such dispositions are indicative of discovery and deeper awareness of one’s epistemic conditions in high stakes contexts.

142  Davide Fassio •

•

Hypothetical self-judgments. Even when high stakes subjects don’t retract a previous knowledge attribution, normally they are disposed to say that maybe in an ordinary context they would have claimed to know, while also recognizing that such attributions would have been mistaken. These seem to be overt recognitions of now being in a better epistemic position to judge whether they know. In contrast, it seems implausible that a low stakes subject, imagining herself in a very high stakes context, would be disposed to claim that in that context she would take herself not to know while also judging that attribution as mistaken. Even when considered as merely hypothetical, high stakes circumstances are recognized as providing more stable and conclusive assessments of one’s epistemic position, both from high and low stakes perspectives. Again, this is an indication of a higher reliability of high stakes knowledge judgments. Inter-subjective judgments. The above considerations are not limited to self-ascriptions. They extend to judgments about other people’s knowledge. When in high stakes we judge that we don’t know a certain proposition, we also judge as wrong other people with the same sort of evidence who claim to know that same proposition (or who claim that we know it).44 Conversely, Fantl and McGrath observe a general “tendency of ordinary people to deny knowledge to themselves when they think about people in high-stakes situations who do not know” (2009: 211).45 The latter also deny that they ever did know in the past. This indicates that, even at an inter-subjective level, we always adjust our reflective knowledge judgments in the direction of high stakes assessments, never in the opposite direction, as if high stakes positions were revelatory of our real epistemic conditions.

These data suggest that experience of and reflection about high stakes situations provide us with a better perspective to judge our own and others’ epistemic positions. In high stakes circumstances our knowledge judgments reach equilibrium points relatively stable across times, subjects and stakes levels, which are indicative of clearer and more reliable assessments of one’s own and other people’s knowledge.46 Some philosophers have suggested that at least some of the above data could be explained away by psychological error theories. For example, Nagel (2010a: §IV, esp. p. 304) appeals to egocentric biases to explain the tendency of people in high stakes to ascribe lack of knowledge to low stakes subjects with similar evidence. High stakes people would mistakenly project on the subjects their own concerns with error, which would lead to misjudging the subject’s real doxastic and epistemic positions.47 Nagel (2008: 291–292) also argues that the tendency of ordinary people to ascribe lack of knowledge to persons in high stakes could be put down to a natural perception of these subjects as having unfounded confidence.

Moderate Skeptical Invariantism  143 In response, Nagel’s psychological error theories have recently been the target of several criticisms.48 Moreover, the variety of effects on knowledge attribution discussed above is so wide and diversified as to resist an explanation in terms of a restricted range of cognitive biases. For instance, it’s not clear that egocentric biases could explain all instances of backward self-judgments, including those in which we have a clear memory of our previous epistemic condition. Considerations of simplicity and unity should lead us to prefer a common and straightforward explanation of such a wide, systematic and coherent pattern of phenomena to a fragmented and seemingly ad hoc collection of error theories. 2.3 Premise 4 According to premise (4), PSI is the view whose verdicts about knowledge better fit with our ordinary high stakes knowledge judgments. These judgments suggest that we know less than what we ordinarily think, both in low and in high stakes contexts. They indicate that we are not in a position to know a range of things we ordinarily claim to know, such as that our bike is still where we left it this morning, that a certain bank will be open tomorrow based on a memory that it was open two weeks ago on the same day of the week, or where we will go on holiday this summer. From a high stakes perspective we wouldn’t judge ourselves as knowing such things, nor we would attribute knowledge to other people in ordinary contexts with roughly the same evidence. However, even when the costs of error are very high, barring weird betting scenarios, most of us would be disposed to ascribe to ourselves knowledge of a wide range of facts such as that we have hands, that the building in front of us is a bank, what our name and job is, that spring follows winter, or that we are not in Uzbekistan now.49 High stakes knowledge judgments do not fit well with other views in the literature predicting that in low stakes contexts we can know all the things we ordinarily claim to know. The standard verdicts in high stakes judgments are also incompatible with radical forms of SI according to which we cannot know even the most obvious truths – that we have hands, what our name is, that two plus two is four, or where we are now. As many philosophers have remarked, normal people react to radical skeptical scenarios with a sense of puzzlement and surprise, but they rarely concede ignorance and retract their previous knowledge attributions about this sort of facts, even after careful reflection.50 This is true also in high stakes situations: normal people do not lose their conviction that they have hands even when it is particularly important to be right about that. On the other hand, high stakes knowledge judgments are exactly the verdicts we should expect from a moderate form of SI. More precisely, our knowledge judgments in high stakes track the propositions we would

144  Davide Fassio be disposed to take for granted and rely on even if the practical stakes were very high. Sense of discovery and retraction of previous knowledge claims seem to be a peculiar mark of contexts in which practical stakes are too high for us to reasonably rely on a proposition in our deliberation.51 These judgments perfectly align with PSI’s predictions. Assume that, as argued above, high stakes knowledge judgments are more reliable than low stakes ones. Their verdicts are often right, or at least more often than corresponding low stakes judgments. Since PSI is the view whose verdicts about knowledge better fit high stakes knowledge judgments, and these are our most reliable judgments on the matter, it follows that this view provides more accurate predictions compared to competing views. Thus, PSI is more plausible than its alternatives.

3 Conclusion In this paper I have provided a partial defense of a moderate form of SI. I focused on a specific version of the view, which I labeled PSI, according to which knowledge of a proposition requires a level of justification sufficient to make it rational for the subject to rely on that proposition even if the practical costs of error were extremely high. We could think of PSI as claiming a ‘courage’ condition on knowledge: in order to know p, a subject should be brave enough to act as if p even if the consequences of a mistake would be catastrophic.52 PSI predicts that we know only a restricted range of the things we ordinarily think and say we know. My case for PSI was centered around two specific points: I defended the view from what I take to be its most impressive problem, and I articulated a specific argument for PSI that I find particularly compelling. PSI characterizes knowledge as a stable and firm achievement and a guide we can always trust and rely on even in the worst adversities, when it is particularly important to avoid mistakes.53 In my opinion, this view strikes a perfect balance between the extremes of a too demanding radical skepticism and too lax forms of moderate invariantism making knowledge too easily available and oftentimes practically irrelevant. Admittedly, the considerations and arguments in this paper constitute only a partial case for PSI. The short space of an essay didn’t allow me to elaborate the view in greater detail and discuss further problems, alleged advantages and implications. I am convinced that this view could provide new insights into several epistemological debates such as the relation between knowledge and action, the fallibilism-infallibilism debate, and the stability and value of knowledge. A discussion of these issues must await further occasions. Acknowledgments. I would like to thank Julien Dutant, Jie Gao, Allan Hazlett, Christos Kyriacou, Guido Melchior, and Melanie Sarzano for helpful comments on earlier drafts of this paper. Earlier versions of this paper were presented at the 2018 Workshop “Kinds of Knowledge”

Moderate Skeptical Invariantism  145 in Hangzhou, the Frontiers of Epistemology conference at Yonsei University, and the inaugural conference of the Institute of Mind and Cognition at Shandong University. Thanks to the audiences for their helpful feedback.

Notes 1 For radical versions of SI see Unger (1975), Cappelen (2005), BonJour (2010), Kyriacou (2017). 2 See in particular BonJour (2010), Kyriacou (2017) for recent insightful discussions. See also Schaffer (2004), Conee (2005), Dodd (2011), Hawthorne (2004: Ch.3) for further advantages. 3 Moderate SI also goes under the name of ‘semi-skepticism’. See Davis (2007: 427, ff., 436; 2015: §4), Douven (2007: 331), Levin (2008: 382), Vogel (1990: 20–21). How moderate an SI is will depend on how the view sets the threshold for knowledge. 4 Some empirical studies indicate that this thought is particularly common among non-philosophers. See DeRose (2017: appendix B). 5 Yet moderate SI is compatible with knowledge that the latter claims are highly likely given other things we know, and knowledge that such things are the case unless something very unlikely occurs (Douven 2007: §3). 6 See, e.g., BonJour (2002: 43, 2010), Brown (2014), Fantl and McGrath (2009: 26). 7 For versions of this view see Dinges (2016), Douven (2007: 330–331), Levin (2008), Fassio (2020). See also Robin McKenna’s contribution to this volume. This sort of view was anticipated, but not endorsed, by Roderick Chisholm: “Suppose, then, we say that a man knows h to be true, provided that no matter what he may do, he has the right to rely upon h—that is to say, no matter what he may do, he does not have the duty to take precautions against the possibility that h is false” (1966: 13). 8 Examples of such cases are DeRose (1992), Stanley’s (2005) bank case, Cohen’s (1999) airport case, Fantl and McGrath’s (2009) train case, Ross and Schroeder’s (2014) peanut allergy case. 9 E.g., Hawthorne (2004), Stanley (2005), Fantl and McGrath (2009). 10 This is a simplified formulation of the view I introduce and defend in Fassio (2020). There I add several further clauses and conditions, which however are not essential for the present discussion. 11 I say ‘arguably’ because, as we will see in §1, the main potential problem for this view is precisely the risk of a collapse into a radical form of skepticism. 12 More precisely, in Fassio (2020) I argue that (i) this view provides the best explanation of several intuitions about specific cases; (ii) the view can preserve the intuition that there is a necessary normative connection between knowledge and rational action; and (iii) the view provides better explanations of intuitions about concessive knowledge attributions and skeptical scenarios. I defend the view from the objection that it delivers wrong predictions about low stakes cases. My diagnosis is that the appropriateness of low standard false knowledge ascriptions should be accounted for in terms of a combination of pragmatics (loose use and exaggeration) and error theory. See BonJour (2010: §II.2) for a similar mixed diagnosis. I also anticipate a ‘naïve’ version of the main objection I will discuss in §1. 13 An incomplete list includes Brown (2008: 175–176; 2014: 185–186), Christensen (2004: 21), Clarke (2013: 9), Greco (2013, 2015: 186), Fantl and

146  Davide Fassio

14

15 16

17 18

19 20

21 22

23

McGrath (2009: 134), Hawthorne (2004), Hawthorne and Stanley (2008), Leitgeb (2014: 132), Maher (1993: 133), Reed (2010), Wedgwood (2012), Williamson (2000: 213), Worsnip (2016: 554). For a short statement of this objection see Goldman and McGrath (2014: 115). Another upholder of moderate SI, Janet Levin (2008), agrees that the greatest challenge for this kind of view is to prevent a slide into a radical Cartesian skepticism. For some examples see Brown (2008: 175–176), Hawthorne (2004: 29), Maher (1993: 133), Reed (2010: 228–229), Williamson (2000: 213), Worsnip (2016: 554). See Salas (2019) for an overview of the literature and discussion. Hawthorne claims: “I wouldn’t even bet on the law of noncontradiction at any odds, and I think myself rational on that score” (2004: 29, fn. 72). Similarly, according to Hawthorne and Stanley “…it’s clear that for any ordinary empirical claim, one should not bet on that claim at any odds and no matter how high the stakes” (2008: 587). According to Daniel Greco “…it would be irrational to bet on anything at arbitrarily unfavorable odds” (2015: 192). For similar remarks see also other authors mentioned in endnote 13. This consideration is further developed in Salas (2019). Several other authors have recently discussed assessments relative to prudential traits of character: see, e.g., Williamson (2005), Hawthorne and Stanley (2008: 588–589). The prudential nature of the intuitions in extreme bets is sometimes apparent in the way in which some philosophers state the objection. For example, Hawthorne and Stanley claim that “it’s clear that for any ordinary empirical claim, one should not bet on that claim at any odds and no matter how high the stakes’’ (2008: 587). Notice that in this formulation there is no reference to epistemic standing of the agent. The judgment seems to rely exclusively on moral and prudential pressures not to engage in vicious activities. Dodd (2017), Eriksson and Rabinowicz (2013), Hacking (1965: 206–207), Salas (2019), Fassio (2020). What if we amend betting scenarios in such a way that bets do not modify truth-relevant features such as evidential support? Imagine, for example, that a machine proposes bets to customers about questions selected at random (Fassio 2020: 855). This example doesn’t seem to involve any factor affecting the epistemic position of the subject. While taking such bets could still be considered problematic for prudential reasons mentioned above, my own intuition is that in such cases it is much more reasonable to take the bet. If the reader shares my intuition, she could take this as a confirmation that additional epistemic factors interfere with our assessment of PSI in radical betting scenarios. Thanks to Allan Hazlett for encouraging me to address this further worry. Several philosophers share my intuition. For instance, Davis (2015: 408) claims that in the high stakes Bank case Hannah has enough evidence to justifiedly believe (and thus act on) propositions such as “I am speaking” and “It is Friday”. Fantl and McGrath (2009: 193) claim that it is hard to imagine someone who would concede that he doesn’t know that George W. Bush wasn’t the first President even if stakes were very high. For similar considerations see Davis (2007: 436), Dinges (2016: 2586–2587), Douven (2007: §2), Fantl and McGrath (2009: 188–191), McKenna (this volume). More precisely, Maximizing Expected Utility is the main rule in decisionmaking under risk, where we assume probability assignments of possible states. In our present discussion, such probabilities represent the degree of epistemic support one has for a proposition.

Moderate Skeptical Invariantism  147 24 As Fantl and McGrath observe, so long as a subject’s epistemic position regarding p is non-maximal with respect to justification, her stakes in whether p can make a difference to whether she is rational to act as if p, independently of her strength of epistemic position regarding p and of the other traditional conditions on knowledge (e.g. belief, truth, proper basing, etc.). (2007: 560) 25 Nonetheless, it is worth noting that the decision theoretic framework presupposes a consequentialist theory of rational choice. A non-consequentialist (e.g., deontologist or virtue theorist) could evade the problem by denying that a rational agent should always maximize expected utility. Thanks to Ren Huimin and Christos Kyriacou for bringing this further option to my attention. 26 Traditional solutions to the St Petersburg paradox appealing to bounded utilities have been advocated by, for example, Bernoulli and Pareto, and more recently Arrow. See Peterson (2019: §4) and Briggs (2019: §3.2.4) for overviews and references. For other defenses of bounded utilities see, for example, Savage (1972: §5.4) and Jeffrey (1983: ch.10). For a solution of the Pasadena game appealing to bounded utilities see Sprenger and Heesen (2011). 27 Peterson observes that “The well-known axiomatizations proposed by Ramsey (1926), von Neumann and Morgenstern (1947), Savage (1972) do, for instance, all entail that the decision maker’s utility function is bounded”. See also Colyvan et al. (2010: see esp. fn.19), Peterson (2019: §4). 28 See, for example, Tversky & Kanheman’s Prospect Theory. For an overview and relevant literature see Buchak (2013: §§1.2–1.4). 29 I leave the derivation to the reader as an exercise. To my knowledge, Lin (2013) has been the first to discuss this theorem in print, though he uses the result for different purposes. See in particular his Theorem 1, p. 839 and appendix A. The theorem was first brought to my attention by Julien Dutant, whom I thank. More recently, Dutant and Fassio (manuscript) discuss further important implications of the theorem for other issues in epistemology and decision theory. 30 See Dutant and Fassio (manuscript) for further arguments for the two conditions. 31 More precisely, this threshold ultimately depends on the ratio between the maximal utility boundaries and the minimal utility difference. Suppose that the gap between minimal and maximal utility values is k times the minimal utility difference: |M+ − M −| = |m+ − m−| × k. Then the probability threshold (T*) is equivalent to the ratio k/(k + 1). Again, I leave the derivation to the reader (see also Dutant & Fassio manuscript). By way of example, suppose your utility scale can be framed in 10,000 discriminable degrees. This is another way of saying that your k ratio is 10,000. It follows that your threshold for knowledge is 10,000/10,000 + 1 = 0.9999. This is not a super-high probability. Your evidential support for many propositions, such as what your name is or that you are not in Uzbekistan now, is certainly much higher. People are mistaken about such things far less often than once every 10,000 times. 32 BonJour claims that “it is simply unclear what sort of basis or rationale there might be for fixing this level of justification in a non-arbitrary way” (2002: 43). See also, e.g., BonJour (2010), Brown (2014), Hannon (2017), Hetherington (2006), Stalnaker (1984: 91), Weatherson (2005: §2).

148 Davide Fassio 33 The present solution resembles the one defended by Fantl and McGrath. They write: “How probable must p be for p to be known? It must be probable enough to be properly put to work as a basis of belief and action” (2009: 26). I substantially agree with their claim, except that I would add ‘no matter how high the stakes’. 34 See Foley (1993: 199–200) for a similar view. While I take this to be a quite intuitive and reasonable feature of knowledge, others may find it a bit puzzling. One may argue that there seems to be something problematic about saying that cautious people really do not know and are thus not warranted to act as if p, but risk-seeking ones are. For instance, assuming men are more prone to take risks than women, this would have the undesirable consequence that men would be in a position to know more than women. Those who dislike sensitivity of knowledge to personal attitudes toward risk could instead assume that there is a unique rational set of attitudes that human beings should have toward risk, which in turn would determine a unique stable threshold for knowledge. Thanks to Melanie Sarzano for helpful comments on this point. 35 See in particular Nagel (2008: §1; 2010a: 299, fn.20; 2010b: §1) for discussion of the features listed below and references to the psychology literature. For examples of such studies see Broder and Newell (2008), Johnson et al. (1993), Johnson and Raye (2000), Lerner and Tetlock (1999), Payne et al. (1993). 36 Premise 1 seems to presuppose that we can know that some judgments are more reliable than others. It may be argued that this presupposition begs the question against the radical skeptic, who would deny the possibility of such knowledge (thanks to Christos Kyriacou for directing my attention to this potential worry). A possible reply is that since the argument is supposed to provide mere defeasible support for its conclusion, it doesn’t require anything as strong as knowledge of its premises. Weaker epistemic standings such as rational confidence would also do the work. 37 Notice that since premise (1) is an empirical generalization, its support to (2) is merely inductive. Still, in the absence of evidence that epistemic conditions constitute an exception to the general rule in (1), the general claim provides at least some defeasible support to (1). Below I shall back (2) with some further independent considerations. 38 Thus, (2) is compatible with occasional lack of transparency about our own mental attitudes. 39 Some philosophers explicitly recognize this point. For example, according to Worsnip, “a change in stakes causes me to revisit my original assessment, to attend to some possibilities that I had been overlooking, and to adjust an unreasonably overconfident credence” (2016: 556). It is worth observing that not all philosophers find the above claims intuitive as I do. Fortunately, my defense of (2) doesn’t merely rely on intuitive judgments. It also relies on a defeasible inference from the general claim in (1), and it is independently supported by phenomenological data and considerations about our practices of knowledge ascription (see below). While I agree that these considerations are not conclusive, I think they are sufficient to shift the burden of proof onto my opponent to provide evidence that judgments about epistemic conditions constitute an exception to the general rule that higher stakes yield greater accuracy and reliable judgments. 40 For similar remarks see also, e.g., Hawthorne (2004: 164–165), Cappelen (2005: 28, 30), Dinges (2016: 2591–2592), MacFarlane (2005: 213). 41 For examples see Lewis (1979: 355), Pritchard (2001), Davis (2007: 420–421), Levin (2008: 381), McKenna (2011), Schroeder (2018).

Moderate Skeptical Invariantism 149 42 Again, there are exceptions. The stability and resilience of opinions may sometimes be characteristic of delusions, ideologies, and prejudices. However, in rational subjects the most resilient and stable beliefs are those based on safer epistemic grounds. Several philosophers have argued that stability is a characteristic mark of knowledge and justified belief (e.g., Williamson 2000; Leitgeb 2017). 43 See, for example, DeRose (2000), Hawthorne (2004: 6, 124), Cappelen (2005: 19, 31), Davis (2007: 398–399, 420–421), Levin (2008: 381), Fantl and McGrath (2009: 210–211), MacFarlane (2005). 44 See, e.g., Cappelen (2005: 19, 30–31), Davis (2007: 431–432), Stanley (2005: 97–98). 45 See also Davis (2007: 433), DeRose (2009). 46 In personal correspondence, Christos Kyriacou observes that high stakes situations could reveal to us the laxity of the everyday standard of knowledge and that the correct standard is a high one. Such situations enlighten us about the fact that the true theory of knowledge is high standard. I fully agree with Christos’ remarks. 47 See also Nagel (2008: 292–293; 2010b: 413, 425), Fantl and McGrath (2009: 56–57). See DeRose (2000, 2009), Hawthorne (2004: 162–165), Gerken (2017), Stanley (2005: 100–102), Williamson (2005: 234) for alternative psychological error theories of the relevant cases. 48 Dimmock (2019), Dinges (2018), Stoutenburg (2017a). 49 In listing these propositions I’m here following my own intuitions about which propositions I would judge to know in high stakes and which I wouldn’t. For instance, I would judge that I know I’ve hands even if a lot were at stake on this proposition (say, if I were a surgeon practicing a risky surgery on a person I truly care about and I should proceed with the surgery only if I had hands). Other people could have different intuitions. They are free to change the above lists as they deem more fitting. See McKenna (this volume) for a more principled and exhaustive list. 50 As Davis pointed out, “[c]areful reflection does not ordinarily prompt a retraction or correction of the claim to know one has a hand, the way it prompts Hannah to retract her claim to know the bank will be open on Saturday” (2007: 436). According to Bach “just imagining yourself in such a scenario is not to take seriously the possibility that you’re actually in it. So-called skeptical “hypotheses” are really just fantasies” (2010: Sect. 5). Many philosophers share the same thought. See, for example, Adler (2012: 264–265), Davis (2015: 421), Dinges (2016: 2586–2587), Fantl and McGrath (2009: 193), Douven (2007: §2), MacFarlane (2005), Pritchard (2001). There are, of course, some exceptions; see Stoutenburg (2017b). Studies in experimental philosophy confirm that ordinary people do not generally find radical skeptical considerations particularly persuasive. See DeRose (2017: appendix B). As Hookway (2008: §5) observes, the above considerations have been familiar in the history of epistemology, especially in the pragmatist tradition. Peirce urges us to “not pretend to doubt in philosophy what we do not doubt in our hearts” (1992: 28–29). See also Peirce (1992: 115) on the distinction between Cartesian doubt and ‘real and living doubt’. 51 A reviewer wonders whether we also don’t feel a sense of discovery when as undergraduates we first learn of Descartes’ evil genius. My own experience as an undergraduate is similar to that of my undergraduate students:  we didn’t feel any sense of discovery that we actually didn’t know whether we  had hands, that we were in a classroom, and so on; we didn’t recognize we had been wrong in ascribing knowledge of these obvious truths; and we didn’t retract such knowledge claims. We rather experienced puzzlement

150  Davide Fassio due to the difficulty of providing rationally articulated responses to Descartes’ challenge. 52 On knowledge as a kind of bravery see Fassio (2020). For recent discussions of the relation between knowledge and emotional attitudes such as courage and fear, see Hookway (2008), Schönbaumsfeld (2019). Relations between cognitive states and feelings have been widely discussed in the history of epistemology, both in the Pragmatist and in the Humean tradition. See, e.g., James (1889), Hume (1993). 53 In this respect, PSI could provide a specific account of the stability of knowledge and knowledge-level beliefs, in terms of reasonability to rely on a fact no matter the stakes. For recent discussions of the stability of knowledge and justified belief see, e.g., Williamson (2000), Leitgeb (2017). On knowledge as an ability to be guided by the facts, see Hyman (1999, 2015).

References Bach, K. (2010). Knowledge in and out of context. In J. K. Campbell, M. O’Rourke, & H. S. Silverstein (Eds.), Knowledge and Skepticism (pp. 105–136). Cambridge: MIT Press. BonJour, L. (2002). Epistemology: Classic Problems and Contemporary Responses. Lanham, MD: Roman and Littlefield. BonJour, L. (2010). The myth of knowledge. Philosophical Perspectives 24(1): 57–83. Briggs, R. A. (Fall 2019). Normative theories of rational choice: Expected utility. In Edward N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy. https:// plato.stanford.edu/archives/fall2019/entries/rationality-normative-utility/ Broder, A., & Newell, B. (2008). Challenging some common beliefs: Empirical work within the adaptive toolbox metaphor. Judgment and Decision Making 3(3): 205–214. Brown, J. (2008). Subject-sensitive invariantism and the knowledge norm for practical reasoning. Noûs 42(2): 167–189. Brown, J. (2014). Impurism, practical reasoning, and the threshold problem. Noûs 48(1): 179–192. Buchak, L. (2013). Risk and Rationality. Oxford: Oxford University Press. Cappelen, H. (2005). Pluralistic skepticism: Advertisement for speech act pluralism. Philosophical Perspectives 19: 15–39. Chisholm, R. (1966). Theory of Knowledge. Englewood Cliffs, NJ: Prentice-Hall. Christensen, D. (2004). Putting Logic in Its Place: Formal Constraints on Rational Belief. Oxford: Oxford University Press. Clarke, R. (2013). Belief is credence one (in context). Philosopher’s Imprint 13: 1–18. Cohen, S. (1999). Contextualism, skepticism, and the structure of reasons. Philosophical Perspectives 13: 57–89. Cohen, S. (2005). Contextualism defended. In M. Steup & E. Sosa (Eds.), Contemporary Debates in Epistemology (pp. 56–62). Malden, MA: Blackwell. Colyvan, M., Cox, D., & Steele, K. (2010). Modelling the moral dimension of decisions. Noûs 44(3): 503–529. Conee, E. (2005). Contextualism contested. In M. Steup & E. Sosa (Eds.), Contemporary Debates in Epistemology (pp. 47–56). Malden, MA: Blackwell.

Moderate Skeptical Invariantism  151 Davis, W. A. (2007). Knowledge claims and context: Loose use. Philosophical Studies 132(3): 395–438. Davis, W. A. (2015). Knowledge claims and context: Belief. Philosophical Studies 172(2): 399–432. DeRose, K. (1992). Contextualism and knowledge attributions. Philosophy and Phenomenological Research 52(4): 913–929. DeRose, K. (2000). Now you know it, now you don’t. Proceedings of the Twentieth World Congress of Philosophy (Bowling Green, OH: Philosophy Documentation Center); Vol. V, Epistemology: 91–106. DeRose, K. (2009). The Case for Contextualism: Knowledge, Skepticism, and Context, Vol. 1. Oxford: Clarendon Press. DeRose, K. (2017). The Appearance of Ignorance: Knowledge, Skepticism, and Context. Oxford: Oxford University Press. Dimmock, P. (2019). Knowledge, belief, and egocentric bias. Synthese 196(8): 3409–3432. Dinges, A. (2016). Skeptical pragmatic invariantism: Good, but not good enough. Synthese 193(8): 2577–2593. Dinges, A. (2018). Anti-intellectualism, egocentrism and bank case intuitions. Philosophical Studies 175(11): 2841–2857. Dodd, D. (2011). Against fallibilism. Australasian Journal of Philosophy 89(4): 665–685. Dodd, D. (2017). Belief and certainty. Synthese 194(11): 4597–4621. Douven, I. (2007). A pragmatic dissolution of Harman’s paradox. Philosophy and Phenomenological Research 74(2): 326–345. Dutant, J. & Fassio, D. (unpublished manuscript). On the distinction between practical and theoretical certainty. Eriksson, L., & Rabinowicz, W. (2013). The interference problem for the betting interpretation of degrees of belief. Synthese 190(5): 809–830. Fantl, J., & McGrath, M. (2007). On pragmatic encroachment in epistemology. Philosophy and Phenomenological Research 75(3): 558–589. Fantl, J., & McGrath, M. (2009). Knowledge in an Uncertain World. Oxford: Oxford University Press. Fassio, D. (2020). Moderate skeptical invariantism. Erkenntnis 85: 841–870. Foley, R. (1993). Working without a Net: A Study of Egocentric Epistemology. Oxford: Oxford University Press. Gerken, M. (2017). On Folk Epistemology: How We Think and Talk about Knowledge. Oxford: Oxford University Press. Goldman, A. I., & McGrath, M. (2014). Epistemology: A Contemporary Introduction. Oxford: Oxford University Press. Greco, D. (2013). Probability and prodigality. Oxford Studies in Epistemology 4: 82–107. Greco, D. (2015). How I learned to stop worrying and love probability 1. Philosophical Perspectives 29(1): 179–201. Hacking, I. (1965). Logic of Statistical Inference. Cambridge: Cambridge University Press. Hannon, M. (2017). A solution to knowledge’s threshold problem. Philosophical Studies 174(3): 607–629. Hawthorne, J. (2004). Knowledge and Lotteries. Oxford: Oxford University Press. Hawthorne, J., & Stanley, J. (2008). Knowledge and action. The Journal of Philosophy 105(10): 571–590.

152  Davide Fassio Hetherington, S. (2006). Knowledge’s boundary problem. Synthese 150: 41–56. Hookway, C. (2008). Epistemic immediacy, doubt and anxiety: On a role for affective states in epistemic evaluation. In G. Brun, U. Doguoglu, & D. Kuenzle (Eds.), Epistemology and Emotions (pp. 51–65). Farnham: Ashgate. Hume, D. (1993). An Enquiry Concerning Human Understanding (2nd ed.). Indianapolis, IN: Hackett. Hyman, J. (1999). How knowledge works. Philosophical Quarterly 49(197): 433–451. Hyman, J. (2015). Action, Knowledge, and Will. Oxford: Oxford University Press. James, W. (1889). The psychology of belief. Mind 14(55): 321–352. Jeffrey, R. C. (1983). The Logic of Decision (2nd ed.). Chicago, IL: University of Chicago Press. Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin 114: 3–28. Johnson, M. K., & Raye, C. (2000). Cognitive and brain mechanisms of false memories and beliefs. In D. Schacter & E. Scarry (Eds.), Memory, Brain and Belief (pp. 35–86). Cambridge, MA: Harvard University Press. Kyriacou, C. (2017). Bifurcated sceptical invariantism. Journal of Philosophical Research 42: 27–44. Leitgeb, H. (2014). The stability theory of belief. Philosophical Review 123(2): 131–171. Leitgeb, H. (2017). The Stability of Belief: How Rational Belief Coheres with Probability. Oxford: Oxford University Press. Lerner, J., & Tetlock, P. (1999). Accounting for the effects of accountability. Psychological Bulletin 125: 255–275. Levin, J. (2008). Assertion, practical reason, and pragmatic theories of knowledge. Philosophy and Phenomenological Research 76(2): 359–384. Lewis, D. (1979). Scorekeeping in a language game. Journal of Philosophical Logic 8(1): 339–359. Lin, H. (2013). Foundations of everyday practical reasoning. Journal of Philosophical Logic 42: 831–862. MacFarlane, J. (2005). The assessment sensitivity of knowledge attributions. In T. Szabo Gendler & J. Hawthorne (Eds.), Oxford Studies in Epistemology (pp. 197–234). Oxford: Oxford University Press. McKenna, R. (2011), Interest contextualism. Philosophia 39(4): 741–750. McKenna, R. (this volume). A (Partial) Defense of Moderate Skeptical Invariantism. In Christos Kyriacou & Kevin Wallbridge (Eds.), Skeptical Invariantism Reconsidered. Routledge. Maher, P. (1993). Betting on Theories. Cambridge: Cambridge University Press. Nagel, J. (2008). Knowledge ascriptions and the psychological consequences of changing stakes. Australasian Journal of Philosophy 86(2): 279–294. Nagel, J. (2010a). Knowledge ascriptions and the psychological consequences of thinking about rrror. The Philosophical Quarterly 60(239): 286–306. Nagel, J. (2010b). Epistemic anxiety and adaptive invariantism. Philosophical Perspectives 24(1): 407–435. Payne, J., Bettman, J., & Johnson, E. (1993). The Adaptive Decision Maker. New York: Cambridge University Press. Peirce, C. S. 1992. The Essential Peirce, Vol. 1. N. Houser & C. Kloesel (Eds.). Indianapolis: Indiana University Press.

Moderate Skeptical Invariantism  153 Peterson, M. (2019). The St. Petersburg paradox. In Edward N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/archives/ fall2019/entries/paradox-stpetersburg/. Pritchard, D. (2001). Contextualism, scepticism, and the problem of epistemic descent. Dialectica 55(4): 327–349. Ramsey, F. P. (1926). Truth and probability. In R. B. Braithwaite (Ed.), The Foundations of Mathematics and Other Logical Essays (pp. 156–198). London: Kegan Paul, Trench, Trubner & Co. Reed, B. (2010). A defense of stable invariantism. Nous 44(2): 224–244. Ross, J., & Schroeder, M. (2014). Belief, credence, and pragmatic encroachment1. Philosophy and Phenomenological Research 88(2): 259–288. Salas, Javier González de Prado (2019). Extreme betting. Ratio 32(1): 32–41. Savage, L. (1972). The Foundations of Statistics (2nd ed.). New York: Wiley. Schaffer, J. (2004). Skepticism, contextualism, and discrimination. Philosophy and Phenomenological Research 69(1): 138–155. Schönbaumsfeld, G. (2019). Epistemic angst, intellectual courage and radical skepticism. International Journal for the Study of Skepticism, 9(3): 206–222. Schroeder, M. (2018). Rational stability under pragmatic encroachment. Episteme 15(3): 297–312. Sprenger, J., & Heesen, R. (2011). The bounded strength of weak expectations. Mind 120(479): 819–832. Stalnaker, R. (1984). Inquiry. Cambridge: MIT Press. Stanley, J. (2005). Knowledge and Practical Interests. Oxford: Oxford University Press. Stoutenburg, G. (2017a). Strict moderate invariantism and knowledge-denials. Philosophical Studies 174(8): 2029–2044. Stoutenburg, G. (2017b). Unger’s argument from absolute terms. Philosophical Papers 46(3): 443–461. Unger, P. (1975). Ignorance: A Case for Scepticism. Oxford: Oxford University Press. Vogel, J. (1990). Are there counterexamples to the closure principle? In M.D. Roth & G. Ross (Eds.), Doubting: Contemporary Perspectives on Skepiticism. Dordrecht: Kluwer Academic Publishers. pp. 13–29. von Neumann, J., & Morgenstern, O. (1947). Theory of Games and Economic Behavior (2nd ed.). Princeton, NJ: Princeton University Press. Weatherson, B. (2005). Can we do without pragmatic encroachment. Philosophical Perspectives 19(1): 417–443. Wedgwood, R. (2012). Outright belief. Dialectica 66(3): 309–329. Williamson, T. (2000). Knowledge and Its Limits. Oxford: Oxford University Press. Williamson, T. (2005). Contextualism, subject-sensitive invariantism and knowledge of knowledge. Philosophical Quarterly 55(219): 213–235. Worsnip, A. (2016). Belief, credence, and the preface paradox. Australasian Journal of Philosophy 94(3): 549–562.

8

A (Partial) Defense of Moderate Skeptical Invariantism Robin McKenna

1 Introductory Remarks Skepticism is not a popular view in epistemology. Most epistemologists reject it.1 Why? Some reject it because they view the arguments for it as based on a mistake (Sosa 2009; Williams 1991; Wright 1991). Others reject it for the simple reason that they think we know lots of things (Moore 1939; Pritchard 2002; Sosa 1999). These epistemologists take it as a key desideratum on an account of knowledge that it fits with the (pre-theoretic?) intuition that we know a lot of things. Because skepticism, by definition, cannot accommodate this desideratum, it is to be rejected. I think that something like this desideratum is right. We do know certain things, and any view of knowledge which says we don’t is therefore to be rejected. This might not sound like a promising starting point for an exploration of skeptical invariantism. This chapter looks like it is over before it has started. But this would be too hasty for three reasons. The first is that the desideratum is vague. Which things do we know? It makes no sense to require an account of knowledge to vindicate every intuition someone might have about the extension of the concept of knowledge. We need to come up with, and defend, a list of things we should want an account of knowledge to say we know. Second, I am open to revisionary accounts of knowledge (Craig 1990; Fassio and McKenna 2015; Haslanger 1999; Kusch and McKenna forthcoming). An account of knowledge that is theoretically attractive may also revise certain aspects of our ordinary understanding of knowledge, such as how much of it we actually have. But there has to be a trade-off here: the more revisionary an account is, the greater the theoretical benefits must be (Fassio and McKenna 2015). 2 Third, the label “skeptical invariantism” denotes a family of views (Cappelen 2005; Davis 2007; Fassio forthcoming; Kyriacou 2017; Levin 2008). Some of these views are “radical” in that they hold that we know very little, if anything. But others are more “moderate” in that they merely hold that we know significantly less than we ordinarily take

Moderate Skeptical Invariantism  155 ourselves to know. You might hold that we know a lot less than we think we do, yet still hold that we know a fair bit. Putting all this together, it should be clear that, properly understood, the key desideratum need not rule out skeptical invariantist accounts of knowledge from the outset. But is there a version of skeptical invariantism that can actually accommodate it? This is my question in this chapter. Here is the plan. In the first part, I distinguish between “radical” and “moderate” skeptical invariantism, and identify specific versions of each view that I focus on in the rest of the chapter. In the second part, I say more about what I take the core cases of knowledge to be. Over and above the sorts of cases you might expect—e.g. simple perceptual beliefs formed in normal conditions, knowledge of a priori, analytic or necessary truths—I argue we should follow the naturalized epistemologist and look to science to identify core cases of knowledge. I focus on one item of scientific knowledge in particular: we know that global temperatures are rising. In the third part, I investigate whether skeptical invariantism can accommodate the core cases identified in the second part. I focus on Davide Fassio’s (forthcoming) moderate skeptical invariantism and investigate how it fairs with the key desideratum. While this investigation raises several issues, I argue that his view can go some way towards accommodating the cases, and so satisfying the desideratum. This by no means constitutes a full defense of moderate skeptical invariantism. But it does show that it is a view well worth taking seriously. Maybe we really do know a lot less than we ordinarily take ourselves to know.

2 Skeptical Invariantism Since the late 1980s, there has been much debate in epistemology about the semantics of knowledge ascriptions—utterances of sentences including the word “knows” and its cognates. The crucial question has been: is the word “knows” (and its cognates) context-sensitive, in the same way that indexicals (“I”, “here”) or gradable adjectives (“tall”, “heavy”) are context-sensitive? That is, do uses of the word “knows” refer to different relations in different contexts of utterance? The “invariantist” holds that they refer to the same relation in all contexts of utterance: the knowledge relation (Brown 2006; Hazlett 2009; Rysiew 2001). In contrast, the “contextualist” holds that they refer to different relations in different contexts of utterance (Cohen 1986; DeRose 1992; Lewis 1996). While more could be said about the differences between invariantism and contextualism, we can put contextualism to one side and focus on invariantism. There are two important distinctions to draw between different forms of invariantism. The first is between “traditional” and “interest-relative” invariantism. The traditional invariantist and the

156  Robin McKenna interest-relative invariantist agree that the word “knows” and its cognates is context-insensitive: uses of it express the same relation irrespective of context. They differ over the nature of that relation. For the traditional invariantist, whether someone knows something depends on purely epistemic or truth-connected factors, such as the strength of their evidence (Brown 2008; Gerken 2017; Reed 2010). For the interestrelative invariantist, whether someone knows depends on a combination of epistemic and “practical” factors, such as the costs of being wrong or “stakes” (Fantl and McGrath 2009; Hawthorne 2004; Stanley 2005). Imagine Catriona and Laurie both have the same evidence that the next stop is South Parkway, and they both believe that the next stop is South Parkway on the basis of this evidence. For Catriona it doesn’t matter if she is wrong: she doesn’t need to get off at South Parkway, as she can equally well get off at the next stop. In contrast, for Laurie, it is absolutely imperative that she not get this wrong. For the traditional invariantist, this difference in what is “at stake” is irrelevant: either the evidence is sufficient for them both to know, or it isn’t. For the interest-relative invariantist, the difference in stakes matters. It may be that the evidence is sufficient for Catriona to know given how little it would matter if she were wrong, but it is insufficient for Laurie to know, given how much it would matter if she were wrong. The second distinction is between “skeptical” and “moderate” invariantism. The invariantist holds that the word “knows” is univocal: it expresses the same relation in all contexts of utterance. But how (to put it roughly) stringent is this relation? That is, how high are the standards for knowledge? The moderate invariantist says that the standards are relatively modest.3 They are high enough to draw a distinction between, say, merely truly believing that it is raining and knowing that it is raining. But they are sufficiently low that we count as knowing an awful lot. The skeptical invariantist says that the standards are considerably higher—high enough that we count as knowing a lot less than the moderate invariantist thinks we do (Cappelen 2005; Davis 2007; Fassio forthcoming; Kyriacou 2017; Levin 2008). With these distinctions in hand, we can now focus on skeptical invariantism. I start with two preliminaries. First, as I use the label, skeptical invariantism is neutral between traditional and interest-relative invariantism.4 That is, the skeptical invariantist might hold that knowledge depends purely on epistemic factors, but she might also allow that knowledge depends partly on pragmatic factors. Second, skeptical invariantism denotes a family of positions, ranging from a view on which the standards for knowledge are very high indeed, so high that we rarely if ever meet them, to views on which the standards for knowledge are more stringent than the moderate invariantist would allow, but not so high that we can’t sometimes meet them. We can distinguish between “radical” and “moderate” skeptical invariantism. But these

Moderate Skeptical Invariantism  157 are perhaps best seen as labels for two contrasting tendencies, rather than for distinct views: the radical skeptical invariantist tends to insist that the standards for knowledge are about as high as they could be, whereas the moderate skeptical invariantist tends to insist that the standards are very high, but not so high as the radical skeptical invariantist insists they are. In the rest of the chapter, it will be helpful to have concrete examples of radical and moderate skeptical invariantist positions to hand. I focus on the radical skeptical invariantism defended by Christos Kyriacou (2017) and the moderate skeptical invariantism defended by Davide Fassio (2020), though for reasons that will become clear below I focus more on Fassio’s moderate skeptical invariantism than Kyriacou’s radical skeptical invariantism. Kyriacou defends what he calls “bifurcated skeptical invariantism”. His skeptical invariantism is “bifurcated” because he holds that there are two related but distinct concepts of knowledge, both of which are operative, but in different contexts. These are (put roughly) a concept of knowledge as fallible and a concept of knowledge as infallible. These concepts are related in that they hold that knowledge requires justification. But they are distinct in that they take different views of the nature of justification. In order to have fallible knowledge, one must merely believe on the basis of strong but defeasible justification. In contrast, in order to have infallible knowledge, one must believe on the basis of a justification that entails the truth of one’s belief.5 Kyriacou’s view is that, so far as the semantics of “knowledge” and knowledge ascriptions is concerned, to say that someone “knows” something is to say that they have infallible knowledge. Thus, many of our knowledge ascriptions—indeed, all of the ones where the subject lacks infallible justification—are, strictly speaking, false.6 But, so far as the pragmatics of “knowledge” and knowledge ascriptions is concerned, we often use them to communicate that subjects have fallible knowledge. Because subjects often do have fallible knowledge—they often believe on the basis of fallible justification—our knowledge ascriptions often pragmatically convey truths, even though they are largely, strictly speaking, false. Now turning to Fassio, there are two crucial components of his view: [A] subject knows p only if she is practically certain that p, where practical certainty can be defined as the confidence a rational subject would have to have for her to believe that p and act on p no matter the stakes on being right about whether p. (2020, 843) S is practically certain that p if and only if [if S were rationally responsive to stakes, S would maintain a degree of confidence sufficient to believe that p and act on p no matter how much turned on p]. (2020, 845)

158  Robin McKenna The thought is that you know that p only if you are practically certain that p, and you are practically certain that p iff you are sufficiently confident that p such that, no matter the stakes, the rational response to them would be to still believe that p and act on p. Note that Fassio does not “build it in” to his view that we know more than the radical skeptical invariantist would allow, but less than the moderate invariantist would allow.7 Rather, he provides us with a condition on knowledge (practical certainty), which he argues is easier to meet than the sort of condition a radical skeptical invariantist would endorse, but harder to meet than the sort of condition a moderate invariantist would endorse. To see how his “practical certainty condition” works, consider one of the cases above. Recall that Catriona and Laurie had the same evidence that the next stop is South Parkway, and they both believed that the next stop is South Parkway on the basis of this evidence. To determine whether either of them have knowledge we need to ask if they are practically certain that the next stop is South Parkway, and they are practically certain that the next stop is South Parkway if and only if, if they rationally responded to the stakes, they would be confident enough to believe and act on the proposition that the next stop is South Parkway no matter the stakes.8 Whether this condition is met is going to depend on the strength of the evidence. We can distinguish three cases. First, let’s imagine it is relatively flimsy: they both glanced at the list of stops on the platform before they boarded the train, and they seem to recall that South Parkway was on it. In this case, it is clear that the condition will not be met. Imagine a high stakes situation. In such a situation, if they rationally responded to the stakes, they would be insufficiently confident to believe that the train stops at South Parkway, or to act on this basis. So they aren’t practically certain. Second, let’s imagine it is considerably stronger, but still far from conclusive: they have taken this train before, and it usually stops at South Parkway. In this case, it is (or at least Fassio thinks it is) clear that the condition will also not be met. That is, we can imagine a situation where the stakes are high enough that, if they were to rationally respond to them, they would be insufficiently confident to believe that the train stops at South Parkway, or to act on this basis. So they aren’t practically certain here either. Finally, let’s imagine it is very strong: they just heard an announcement from the driver saying the next stop is South Parkway, all the signs say the next stop is South Parkway, they recognize that they are approaching South Parkway, the train is slowing down and so on. Fassio’s view is that, in this case, the condition is met. That is, he thinks that, no matter how high we imagine the stakes being, it remains the case that, if they were to rationally respond to them, they would be confident enough to believe that the train stops at South Parkway, and to act on this basis. So they are practically certain.

Moderate Skeptical Invariantism  159 These three cases not only illustrate how the practical certainty condition works; they also illustrate the “moderate” and “skeptical” aspects of Fassio’s view. The view is moderate in that, in the third case, there plausibly is knowledge. The view is skeptical in that, in the second case, there isn’t. Thus, they distinguish Fassio’s view from moderate invariantism (which would hold there is knowledge in the second case), and radical skeptical invariantism (which would deny there is knowledge in the third case). This completes my overview of skeptical invariantism. I am now going to consider the extent of our knowledge, and whether there is a skeptical invariantist position that can plausibly accommodate the things we should take ourselves to know.

3 Articulating the Key Desideratum How much do we know? And, more importantly for our purposes, how much does a theory of knowledge need to say we know in order to be plausible? In this section, I want to make two suggestions. The first—which I won’t spend much time on—is that the debate about invariantism and the semantics of knowledge ascriptions tends to implicitly assume an implausible answer to the second of these questions. The second—which I will spend a little more time on—is that we can draw up a list of several classes of beliefs that we should expect a theory of knowledge to vindicate as cases of knowledge. I should say from the outset that I make no claim to the effect that my list is complete. It may be there are classes of beliefs that a theory of knowledge should vindicate that are not on my list, and I won’t discuss the extent to which the skeptical invariantist can accommodate them. But my aim in this chapter is to develop a partial, not a full, defense of (moderate) skeptical invariantism.9 Let’s start with the first suggestion. In the debate about knowledge ascriptions it often seems to be taken as obvious that the subjects in the “low stakes” version of DeRose’s bank cases have knowledge. This case goes like this: Low: Hannah and her wife Sarah are driving home on a Friday afternoon. They plan to stop at the bank on the way home to deposit a check. It’s not important that they do so, as they have no impending bills. But, as they drive past the bank, they notice that the lines inside are very long. Realising that it isn’t very important that the check is deposited right away, Hannah says, “I know that the bank will be open tomorrow. I was there just two weeks ago on Saturday morning. So we can deposit our check tomorrow morning”.10

160  Robin McKenna The general assumption seems to be that, clearly, Hannah speaks truly when she says that she knows the bank will be open on Saturday. Some views seek to vindicate this assumption by complicating it: for the contextualist, the subjects in these cases can truly be said to “know” in some contexts but not others, and for the interest-relative invariantist, they know in part because the stakes are low. Other views seek to vindicate it in a more straightforward manner (e.g. traditional moderate invariantism). The problem for the skeptical invariantist (whether radical or moderate) is meant to be that they deny that the subjects in these cases have knowledge irrespective of the stakes, and no matter in what context one ascribes knowledge to them. But, for this to be a problem, we need to be justified in treating low stakes bank cases as core or non-negotiable cases of knowledge, and this strikes me as highly implausible. Note: my claim is not that Hannah does not know in this case. Maybe she does. It may well be that our best account of knowledge rules that the subjects in bank cases have knowledge. That doesn’t mean that we should build it into our theorizing about knowledge that an acceptable account of knowledge has to have this result (cf. Dinges 2016) Turning to the second suggestion, we can start with what Fassio says we can know according to his moderate skeptical invariantism. Fassio says: MSI [moderate skeptical invariantism] allows that knowledge is compatible with absence of absolute certainty. One may be practically certain, and therefore in a position to know that p, even if one is not in a position to exclude some abstract error possibilities. While MSI excludes knowledge in Bank-like cases (a subject that is rational and aware of the high stakes will lack the confidence necessary to believe and act on the proposition that the bank is open on Saturday), all beliefs based on a rational degree of confidence that is ‘robust’ enough to survive any significant increase in the stakes are good candidates for knowledge. (2020, 849) But what beliefs are based on a rational degree of confidence robust enough to survive any increase in the stakes? Fassio is reluctant to list concrete examples, but he ventures the following: • • • • • •

His belief that the capital of Italy is Rome. His belief (formed on a Monday) that tomorrow is Tuesday. His belief that he ate spaghetti for lunch. His belief that in December he bought presents for his family. His belief that there are two oranges on the table. Certain testimonial beliefs (e.g. Hannah’s belief that the next stop is South Parkway, based on being told by a ticket inspector a few minutes before).

Moderate Skeptical Invariantism  161 Rather than focus on concrete examples, I think a better approach is to focus on broad classes of beliefs which, on a plausible account of knowledge, we should be said to know. The claim is not that any particular individual has some piece of knowledge. Rather, the claim is that many subjects in certain sorts of situations know certain things. For instance: 1 2 3

Beliefs about our immediate surroundings formed on the basis of perception in normal conditions (“there are two oranges on the table”). Very well supported memorial and testimonial beliefs (“I ate spaghetti for lunch”, “the next stop is South Parkway”). Beliefs in matters of fact that we generally take for granted, including facts about the future (“the capital of Italy is Rome”, “tomorrow is Tuesday”, “the sun will rise tomorrow”).

In addition to these three classes of beliefs suggested by Fassio, I would like to add some more. Consider: 4 5

Some introspective beliefs (e.g. “I am in pain right now”). Beliefs in a priori/necessary/analytic truths (e.g. “all bachelors are unmarried”, “2 + 2 = 4”, “water is H 2O”).11

Finally, I want to add certain scientific beliefs. Now, I don’t think that an account of knowledge should show that we know all and everything that science takes us to know. Scientists may differ as to what we know, and it may be that some scientists take us to know that some hypothesis H is true whereas others take us to know that H is false. Rather, I think that, when it comes to hypotheses that are established with as much certainty as science can provide us with, an account of knowledge should allow that we do indeed know that these hypotheses are true. I want to focus on a particular example here: science, I take it, has shown with as much certainty as we could hope for that global temperatures are rising due to human activities. Of course, science hasn’t established that global warming is happening due to human activities with absolute certainty. But science doesn’t deal in absolute certainties.12 I offer two reasons for including certain items of scientific knowledge in our key desideratum, the first of which is less theoretically loaded than the second. First, it would be more than a little odd to accept testimonial beliefs, like Catriona’s belief that the train stops at South Parkway, as core cases of knowledge, yet not to accept our belief that global warming is happening due to human activities as a core case. I don’t know about you, but insofar as the comparison makes sense, we have far more and far better evidence that human activities are causing global temperatures to rise than Catriona does that the next stop is South Parkway.13 Similar points can, I think, be made about memorial beliefs, and perhaps some perceptual beliefs.

162  Robin McKenna One might object that what science has established is not that climate change is happening due to human activities, but rather that it is (very) likely that climate change is happening due to human activities, and so what we can be said to know is not that climate change is happening, but that it is (very) likely that it is.14 Of course, science does establish the likelihood claim. But why not also take it to establish the unqualified claim? If you agree that we have more and better evidence for some scientific hypotheses than we do for testimonial (and memorial and some perceptual) beliefs, then it would be odd to take science to only establish claims about probabilities, and yet to take it that we can come to know what the next stop is on the basis of testimony, rather than merely that it is (very) likely that the next stop is such-and-such. Second, a key insight from naturalized epistemology is that, when it comes to the traditional epistemological question of the extent of our knowledge, we should look to science to answer it (cf. Quine 1969). Now, you might object that there are things we clearly seem to know that have nothing to do with science. Take, for instance, simple perceptual knowledge or memorial knowledge. These forms of knowledge can be scientific, but they need not be. This point is fair so far as it goes but, first, you might respond that science can explain the mechanisms involved in producing perceptual and memorial knowledge and, second, this would only show that we can’t look to science to identify the full extent of our knowledge. I have identified six classes of things we want an account of knowledge to say we know: 1 2 3 4 5 6

Beliefs about our immediate surroundings formed on the basis of perception in normal conditions. Very well supported memorial and testimonial beliefs. Beliefs in matters of fact that we generally take for granted. Some introspective beliefs. A priori/necessary/analytic truths. Some items of scientific knowledge (e.g. “global temperatures are rising due to human activities”).

I now turn to seeing how skeptical invariantism, particularly moderate forms of it, does with respect to this desideratum.

4 Assessing Skeptical Invariantism I have argued that it is a key desideratum on an account of knowledge that it accommodates certain core cases of knowledge. While for obvious reasons I will focus on moderate skeptical invariantism, I want to—albeit very briefly—say something about the basis on which I am

Moderate Skeptical Invariantism  163 dismissing radical forms of skeptical invariantism, and particularly Kyriacou’s bifurcated skeptical invariantism. I myself have no problem in principle with revisionary accounts, whether of knowledge or of anything else. A revisionary account of knowledge may be more theoretically attractive than its rivals, and this may provide ample reason for revising our ordinary understanding of knowledge, or of its extent (see Fassio and McKenna 2015). But I do think that there are certain constraints on any revisionary project. In Fassio and McKenna (2015), we argue for a “moderately conservative” revisionary methodology, according to which, the more revisionary an account of knowledge is, the higher the barrier to accepting it. Put simply, an account that radically revises our ordinary understanding of knowledge—for instance, by claiming that we know an awful lot less than we ordinarily take ourselves to know—would not just have to secure several theoretical benefits. It would have to secure far more theoretical benefits than any rival account.15 While Kyriacou claims some theoretical benefits for his radical skeptical invariantist account of knowledge (e.g. it can (dis)solve the Gettier problem, and it addresses Kripke’s dogmatism paradox), I don’t think he makes (or attempts to make) the case that it secures far more theoretical benefits than any rival account. While more could be said here, I now turn to my main question in this section: can Fassio’s moderate skeptical invariantism accommodate the core cases of knowledge identified in the previous section? While I won’t give an unqualified affirmative answer to this question, it is at least initially plausible that it can. We can start with perceptual, memorial and testimonial beliefs. Fassio spends some time arguing that we can be practically certain when it comes to some of our perceptual, memorial and testimonial beliefs. We can split his argument into two parts. First, he considers an objection which, if right, would show that none of our perceptual, memorial or testimonial beliefs could ever be practically certain.16 Recall Catriona. Can she be practically certain that the next stop is South Parkway, on the basis of being told that it is by the ticket inspector a few moments before? Consider a context where Catriona is offered an extremely high stakes bet with the following pay-off structure. If Catriona is right that the next stop is South Parkway, she gets a penny. If Catriona is wrong about this, something terrible will happen to her—make this thing as terrible as you can imagine. Many think that, in such a context, if Catriona were to rationally respond to the stakes, she would not be confident enough that the next stop is South Parkway to take the bet. Indeed, the rational thing to do would be to decline to take the bet. Fassio’s response to this strikes me as right. He agrees that, in this sort of case, it would not be rational for Catriona to take the bet. But he denies that this poses a problem for his account of knowledge. Now, “bet

164  Robin McKenna cases” would be a problem if the reason why it would not be rational for Catriona to take the bet is that the stakes are so high. But, thinks Fassio, the reason why it would not be rational for Catriona to take the bet has nothing to do with the stakes. Put roughly, the thought is that bet cases raise certain factors to salience, and it is these factors that explain why it would not be rational for Catriona to take the bet. To make this point more persuasive, he considers two cases: Great Importance (GI): Mark is on a train going to Sterling. It is very important that Mark be at the station soon because he is bringing a medicine to his sister Jane. If Mark cannot meet her at the station and give her the medicine, she is very likely to die. If he doesn’t get off the train at right station (e.g., he gets off at a different station), he will not be able to meet his sister on time. According to the route plan the next stop is Sterling. A ticket inspector passed in the coach a few minutes ago and announced that the next stop is Sterling. The train is approaching a station. Mark forms the outright belief that the train is approaching Sterling station, stands up and goes to the exit in order to get out High Bet (HB): Mark is on a train going to Sterling. As in GI, according to the route plan the next stop is Sterling. A ticket inspector passed in the coach a few minutes ago and announced that the next stop is Sterling. A man approaches him. He proposes a bet about whether the next station is Sterling: if it is, he will give Mark £10000, otherwise Mark will lose £10000. The train is approaching a station. Mark has the same evidence as in GI, but doesn’t feel confident enough to take the bet. (2020, 853) The thought is that it would be rational for Mark to believe that the next stop is Sterling, and to act on the basis of this belief, in GI, but it would not be rational for Mark to take the bet in HB. But the stakes are actually higher in GI than in HB (and, as Fassio says, if you don’t think they are, just make the stakes even higher in GI). So it can’t be stakes that are driving intuitions in bet cases. Now, this does not establish that our perceptual, memorial or testimonial beliefs are ever practically certain. This leads to the second part of Fassio’s argument. His basic strategy is to argue that there are examples of perceptual, memorial and testimonial beliefs of which we are practically certain. Take some of the examples considered above: the belief that there are two oranges on the table, formed in normal perceptual conditions; the belief that Rome is the capital of Italy; Catriona’s belief that the next stop is South Parkway. I agree with Fassio that he can accommodate these sorts of cases. But how does this fare as an attempt to satisfy the key desideratum?

Moderate Skeptical Invariantism  165 The most Fassio can claim is that his account can accommodate some core cases of perceptual, memorial and testimonial knowledge (the cases discussed above). This is not to say that, on his account, we have as much perceptual, memorial and testimonial knowledge as we would ordinarily assume. But recall my objection to Kyriacou’s radical skeptical invariantism: the more revisionary of our ordinary understanding a view is, the greater its theoretical benefits need to be. Because Kyriacou’s view is extremely revisionary of our ordinary understanding of the extent of knowledge, it needs to have far greater theoretical benefits than competitors. Because Fassio’s view is far less revisionary, it doesn’t need to have as many theoretical benefits. This of course is not to say that it has sufficient theoretical benefits. But the barrier to acceptance is considerably lower. How does Fassio’s view fare with introspective beliefs, or beliefs in a priori, analytic or necessary truths? Things are far simpler here. While he doesn’t consider such beliefs in much depth, one might think it is very plausible that we can be absolutely certain when it comes to beliefs in either of these classes. For Fassio, absolute certainty entails, but is not entailed by, practical certainty. So Fassio’s view does well here. Finally, we can consider the sixth and final class of beliefs I said an account of knowledge should accommodate: certain scientific beliefs, like the belief that climate change is caused by human activities. I am going to argue that it is plausible this belief is practically certain in Fassio’s sense, at least for some epistemic agents. Let’s imagine an epistemic agent, Ailsa, who has a keen amateur interest in climate science. For a layperson, she is very well informed about the scientific evidence that human activities are causing global warming. She has read some journal articles, knows all the most common climate skeptical arguments, and can explain, in some detail, where these arguments go wrong. She is also a passionate environmentalist. But her acceptance of the science on climate change isn’t dogmatic: she recognizes that, while the basic hypothesis that humans are causing global warming is as established as any hypothesis could be, there are still a lot of unanswered questions and uncertainties. I would submit that Ailsa knows that human activities are causing global warming. So, the question is, is her belief practically certain? For Ailsa’s belief to be practically certain it would have to be that, no matter what the stakes were, if she were rationally responsive to the stakes, she would remain confident enough to believe that human activities are causing global warming, and to act on this basis. Is this the case? At least initially, it might seem like the answer is clear: yes! After all, you might think, the stakes are incredibly high: the survival of humankind may well depend on dealing with global warming. So, if we think Ailsa is rational in believing that human activities are causing global warming, and acting on this basis, in the scenario described above, then we presumably would think the same no matter the stakes. After all, how could they be higher?

166  Robin McKenna There are three issues. The first issue concerns what the stakes actually are here. Of course, in some sense, the stakes are clearly incredibly high: the survival of humankind may well depend on dealing with global warming. But, in the literature on interest-relative invariantism, when people talk about the stakes relating to a proposition, they generally mean the costs of being wrong in believing that proposition.17 Applying this here, what would matter would be the costs (for Ailsa) of being wrong in believing that human activities are causing global warming. It isn’t so clear that the costs (at least for Ailsa) of being wrong in believing this are particularly high. Perhaps this issue can be dealt with by slightly changing how we think about stakes. Instead of just considering the costs (for a subject) of being wrong in believing some proposition, we can also consider the costs (for that subject) of not being right.18 That is, we can also consider the costs of not believing something that is true. If we put things this way, it may become easier to see why the stakes are high for Ailsa: she presumably would think that the costs of not believing that human activities are causing global warming, when they in fact are doing so, are very high. The second issue is that, even if we think of the stakes in terms of the costs of not being right, it isn’t so clear that the stakes are high for Ailsa here. What bad consequences would follow if Ailsa didn’t believe that human activities were causing global warming? Of course, if lots of people don’t believe this, that might be bad for society, and detrimental to the long-term survival of the human race. But that’s different from it being bad for Ailsa. To address this issue, we need to further consider the way in which stakes are usually thought of in the literature. We can distinguish between the “perceived” and “actual” stakes for a subject. The perceived stakes are, roughly, what they take the costs of being wrong (and, we might add, of not being right) to be. The actual stakes are, roughly, what the consequences of being wrong will actually be for the subject. Some (e.g. Fassio) interpret stakes in the first way, others (Stanley 2005) in the second. Applying this to Ailsa, it seems clear that the stakes are high in the first sense, but it is less clear that they are high in the second sense. Does this matter? It is important to note that, even though the notion of stakes may be difficult to apply in Ailsa’s case, it is not clear it makes a difference so far as practical certainty is concerned. Imagine you thought that, contrary to first impressions, Ailsa is actually not in a high stakes situation. Does that mean she isn’t practically certain? Well, she would only not be practically certain if there were a high stakes situation in which, if she were rationally responsive to the stakes, she would not remain confident enough to believe and act. While I have no argument that there couldn’t be such a situation, that is not to say that there actually is one.

Moderate Skeptical Invariantism  167 The third issue is whether what I have said about Ailsa is going to generalize. It clearly isn’t going to generalize in the sense that everyone can be said to know that human activities are causing global warming. But we wouldn’t want it to generalize in this way, because some people don’t know this. Some people don’t know it because they don’t believe it. Some people don’t know it because, even though they believe it, they believe it on the basis of weak or possibly no evidence.19 There would only be a problem if there were subjects who believe it on the basis of evidence we would regard as strong enough for knowledge, but aren’t practically certain. Are there such subjects? I’m not sure. In any case, if it turns out that, on Fassio’s account, some subjects don’t know that human activities are causing global warming because of how they would (rationally) respond to the stakes in counterfactual scenarios, then this could be the basis of an objection to Fassio’s account of knowledge. But it isn’t clear that the objection would have anything to do with the skeptical aspect of his account. The objection would rather be that it just doesn’t seem like how you would respond to the stakes in counterfactual scenarios should make a difference to whether you know these things.

5 Concluding Remarks This chapter started in a rather unpromising way. I noted that it is a desideratum on any account of knowledge that it accommodates the fact—and I think it is a fact—that we know quite a lot of things. But, while this rules out radical forms of skeptical invariantism, it doesn’t automatically rule out more moderate forms. I have proposed a list of things I think any account of knowledge should say that we know, and considered whether Fassio’s moderate skeptical invariantist account can accommodate them. I have argued that the answer is a qualified “yes”. While Fassio focuses on showing that his view can accommodate some perceptual, memorial and testimonial beliefs, I have argued that his account accommodates certain introspective beliefs, beliefs in a priori, analytic or necessary truths and, finally, certain scientific beliefs. This is in no way a “complete” defense of moderate skeptical invariantism, or Fassio’s specific version of it. But I hope I have succeeded in showing that it is plausible to hold that, even though we know a fair bit, we know quite a lot less than we ordinarily take ourselves to know. 20

Notes 1 Exceptions include Fumerton (1995), Stroud (1984) and Unger (1975). 2 Michael Hannon suggested that it is unclear whether radical skepticism really is revisionary, because the radical skeptic generally does not think we should change our ordinary practice of knowledge ascriptions (see, e.g., Unger 1975). This is a fair point, but it does seem to require a workable pragmatic account of knowledge ascriptions. For discussion of skeptical

168  Robin McKenna

3 4

5

6 7 8

9

10 11 12

13

invariantism and the pragmatics of knowledge ascriptions, see Cappelen (2005), Davis (2007), DeRose (2017), Dinges (2016) and Kyriacou (2017). The label “moderate invariantism” is from Hawthorne (2004). All of the invariantists cited above are moderate in Hawthorne’s sense. I make this stipulation because I don’t want to get embroiled in discussion about whether the moderate skeptical invariantist view I go on to consider (Fassio 2020) is really a version of traditional invariantism. It agrees with the traditional invariantist that, if a subject knows in a low stakes case, then she knows in the corresponding high stakes case. But it characterizes knowledge in terms of a notion (practical certainty) that looks a little pragmatic. Part of the issue here is that the terms “epistemic” and “pragmatic” are not usually clearly defined and distinguished. For discussion, see Hannon (2020). In contemporary epistemology infallibilism is enjoying a renaissance, in large part due to Williamson (2000). Williamson identifies one’s evidence with one’s knowledge so, when you know that p, you have evidence which entails p (i.e. your knowledge that p). Kyriacou thinks his view entails that we know little, if anything, so he clearly must reject Williamsonian infallibilism. More generally, he must reject any view on which our evidence goes beyond our experiences and includes claims about the external world. For discussion of Williamsonian infallibilism, see Brown (2018). Recall from fn. 5 that I am ignoring Williamsonian infallibilism. Compare: a safety theorist like Sosa (2007) does not build it into his account that we know anything, but rather proposes a particular condition on knowledge (safety), then argues that many of our beliefs are safe in his sense. On Fassio’s view, what matters is what the rational response to the stakes would be, not what some subject’s actual response would be. You might ask: what is it to rationally respond to the stakes? For Fassio’s discussion of this see Section 1 of his paper. But the rationale for the rationality requirement is that Fassio wants to avoid situations where a subject would remain confident enough to believe something, or act on the basis of it, no matter the stakes because they are habitually over-confident. His thought is that the over-confident subject would not be rationally responding to the stakes, In personal correspondence Fassio told me that he thinks his account can also accommodate the following classes of knowledge: inferential knowledge (where inference need not be deductive), knowledge about what is probable or will be the case conditional on other events, modal knowledge, aesthetic knowledge, historical knowledge, medical knowledge and technical knowledge. If he is right about this, then perhaps a full defense of moderate skeptical invariantism can be mounted. But I lack the space to deal with all these classes of knowledge here. This version of the cases is from Stanley (2005, 3). For the original case see DeRose (1992). On some views we can distinguish between a priori, necessary or analytic truths and on some views we can’t. For my purposes it doesn’t much matter where you stand on these issues. While it is hard to quantify the degree of certainty, we do know the degree of consensus among climate scientists that human activities are causing global warming, which is 97% (see Cook et al. 2016). Degree of consensus is, of course, a poor proxy for degree of certainty, but it is worth noting that such a high degree of consensus is unusual. Of course, I’m comparing a case of individual testimonial knowledge with group scientific knowledge, so there are some differences. If you like, replace

Moderate Skeptical Invariantism  169

14 15

16 17 18 19 20

talk of “us” knowing that global warming is happening due to human activities with talk of individual climate scientists knowing this. I still think it is right to say that an individual climate scientist has better evidence that climate change is happening due to human activities than Catriona has from testimony in the case we’re imagining. Thanks to Christos Kyriacou and Michael Hannon for pushing me to say something about this objection. This is perhaps where the defense of the justified true belief account of knowledge considered (though not endorsed) in Weatherson (2003) goes wrong. There are some theoretical benefits to holding on to the justified true belief account, but they are surely not sufficient to adequately defend it. For this sort of objection see Weatherson (2012). For discussion of the lack of clarity about what stakes are in this literature see Anderson (2015) and Anderson and Hawthorne (2019). For similar points see Crewe and Jenkins Ichikawa (forthcoming), Mueller (2017), Pace (2011) and Worsnip (2015). For an empirically informed discussion of this see McKenna (2019). Thanks to Davide Fassio, Michael Hannon and Christos Kyriacou for comments on an earlier version of this paper.

References Anderson, Charity. 2015. “On the Intimate Relationship of Knowledge and Action.” Episteme 12 (3): 343–353. Anderson, Charity, and John Hawthorne. 2019. “Knowledge, Practical Adequacy, and Stakes.” Oxford Studies in Epistemology 6: 234–257. Brown, Jessica. 2006. “Contextualism and Warranted Assertibility Manoeuvres.” Philosophical Studies 130 (3): 407–435. ———. 2008. “Subject-Sensitive Invariantism and the Knowledge Norm for Practical Reasoning.” Noûs 42 (2): 167–189. ———. 2018. Fallibilism: Evidence and Knowledge. Oxford: Oxford University Press. Cappelen, Herman. 2005. “Pluralistic Skepticism: Advertisement for Speech Act Pluralism.” Philosophical Perspectives 19 (1): 15–39. Cohen, Stewart. 1986. “Knowledge and Context.” Journal of Philosophy 83 (10): 574–583. Cook, John, Naomi Oreskes, Peter T. Doran, William R. L. Anderegg, Bart Verheggen, Edward H. Maibach, J. Stuart Carlton, Stephan Lewandowsky, Andrew G. Skuce, and Sarah A. Green. 2016. “Consensus on Consensus: A Synthesis of Consensus Estimates on Human-Caused Global Warming.” Environmental Research Letters 11 (4): 048002. Craig, Edward. 1990. Knowledge and the State of Nature: An Essay in Conceptual Synthesis. Oxford: Oxford University Press. Crewe, Bianca, and Jonathan Jenkins Ichikawa. Forthcoming. “Rape Culture and Epistemology.” In Applied Epistemology, edited by Jennifer Lackey. Oxford: Oxford University Press. Davis, Wayne A. 2007. “Knowledge Claims and Context: Loose Use.” Philosophical Studies 132 (3): 395–438. DeRose, Keith. 1992. “Contextualism and Knowledge Attributions.” Philosophy and Phenomenological Research 52 (4): 913–929.

170  Robin McKenna ———. 2017. The Appearance of Ignorance: Knowledge, Skepticism, and Context, Vol. 2. Oxford: Oxford University Press. Dinges, Alexander. 2016. “Skeptical Pragmatic Invariantism: Good, but Not Good Enough.” Synthese 193 (8): 2577–2593. Fantl, Jeremy, and Matthew McGrath. 2009. Knowledge in an Uncertain World. Oxford: Oxford University Press. Fassio,Davide. 2020, “Moderate Skeptical Invariantism,” Erkenntnis 85: 841–870 Fassio, Davide, and Robin McKenna. 2015. “Revisionary Epistemology.” Inquiry: An Interdisciplinary Journal of Philosophy 58 (7–8): 755–779. Fumerton, Richard A. 1995. Metaepistemology and Skepticism. Lanham: Rowman & Littlefield. Gerken, Mikkel. 2017. On Folk Epistemology. How We Think and Talk about Knowledge. Oxford: Oxford University Press. Hannon, Michael. 2020. “Why Purists Should Be Infallibilists.” Philosophical Studies 177 (3): 689–704. Haslanger, Sally. 1999. “What Knowledge Is and What It Ought to Be: Feminist Values and Normative Epistemology.” Philosophical Perspectives 13 (s13): 459–480. Hawthorne, John. 2004. Knowledge and Lotteries. Oxford: Oxford University Press. Hazlett, Allan. 2009. “Knowledge and Conversation.” Philosophy and Phenomenological Research 78 (3): 591–620. Kusch, Martin, and Robin McKenna. (2020). “The Genealogical Method in Epistemology.” Synthese, 197: 1057–1076. Kyriacou, Christos. 2017. “Bifurcated Sceptical Invariantism.” Journal of Philosophical Research 42: 27–44. Levin, Janet. 2008. “Assertion, Practical Reason, and Pragmatic Theories of Knowledge.” Philosophy and Phenomenological Research 76 (2): 359–384. Lewis, David. 1996. “Elusive Knowledge.” Australasian Journal of Philosophy 74 (4): 549–567. McKenna, Robin. 2019. “Irrelevant Cultural Influences on Belief.” Journal of Applied Philosophy 36 (5): 755–768. Moore, G. E. 1939. “Proof of an External World.” Proceedings of the British Academy 25: 273–300. Mueller, Andy. 2017. “Pragmatic or Pascalian Encroachment?” Logos and Episteme 8 (2): 235–241. Pace, Michael. 2011. “The Epistemic Value of Moral Considerations: Justification, Moral Encroachment, and James’ ‘Will To Believe.’” Noûs 45 (2): 239–268. Pritchard, Duncan. 2002. “Resurrecting the Moorean Response to the Sceptic.” International Journal of Philosophical Studies 10 (3): 283–307. Quine, W. V. O. 1969. “Epistemology Naturalized.” In Ontological Relativity and Other Essays, 69–90. New York: Columbia University Press. Reed, Baron. 2010. “A Defense of Stable Invariantism.” Noûs 44 (2): 224–244. Rysiew, Patrick. 2001. “The Context-Sensitivity of Knowledge Attributions.” Noûs 35 (4): 477–514. Sosa, Ernest. 1999. “How to Defeat Opposition to Moore.” Philosophical Perspectives 13 (s13): 137–149.

Moderate Skeptical Invariantism  171 ———. 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, Volume I. Oxford: Oxford University Press. ———. 2009. Reflective Knowledge: Apt Belief and Reflective Knowledge, Volume II. Oxford: Oxford University Press. Stanley, Jason. 2005. Knowledge and Practical Interests. Oxford: Oxford University Press. Stroud, Barry. 1984. The Significance of Philosophical Scepticism. Oxford: Oxford University Press. Unger, Peter. 1975. Ignorance: A Case for Scepticism. Oxford: Oxford University Press. Weatherson, Brian. 2003. “What Good Are Counterexamples?” Philosophical Studies 115 (1): 1–31. ———. 2012. “Knowledge, Bets, and Interests.” In Knowledge Ascriptions, edited by Jessica Brown and Mikkel Gerken, 75–103. Oxford University Press. Williams, Michael. 1991. Unnatural Doubts: Epistemological Realism and the Basis of Scepticism. Princeton University Press. Williamson, Timothy. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Worsnip, Alex. 2015. “Two Kinds of Stakes.” Pacific Philosophical Quarterly 96 (3): 307–324. Wright, Crispin. 1991. “Scepticism and Dreaming: Imploding the Demon.” Mind 100 (1): 87–116.

9

Skepticism, Fallibilism, and Rational Evaluation Michael Hannon

1 Introduction According to many historical philosophical figures, knowledge must be based on infallible foundations. These foundations have been characterized in different ways, e.g., as “cognitive impressions” by the ancient Stoics, as “clear and distinct perceptions” by Descartes, and as “the given” element in experience by C. I. Lewis and other twentieth-century philosophers (Reed 2012: 585). In each case, it has been assumed that these foundations are infallible in that they preclude error on the part of the knower. To have knowledge, in other words, we must have justification that guarantees that our belief is true.1 This is infallibilism. It is the view that knowledge demands the highest degree of justification. In contemporary epistemology, it is widely accepted that infallibilist theories of knowledge are “doomed to a skeptical conclusion” (Cohen 1988: 91). 2 We humans are fallible creatures that can rarely guarantee the truth of our beliefs; indeed, almost no belief can be rationally supported or justified in a way that removes all possible doubt. By demanding infallibility, we would prevent ourselves from knowing almost anything. This is an unwelcome skeptical result. To avoid skepticism, many contemporary epistemologists endorse a fallibilist view of knowledge. A fallibilist believes that we can know things on the basis of justification that is less than fully conclusive. As Jim Pryor says, “a fallibilist is someone who believes that we can have knowledge on the basis of defeasible justification, justification that does not guarantee that our beliefs are correct” (2000: 518). This view is attractive because it allegedly avoids the skeptical consequences of infallibilist conceptions of knowledge. It is for this reason that fallibilism is typically regarded as the only serious option in epistemology. As Harvey Siegel says, “we are all fallibilists now” (1997: 164).3 Although fallibilism is almost universally accepted in epistemology, the nature of fallibilist knowledge is still poorly understood. There are at least two reasons for this. First, it is unclear how to formulate fallibilism precisely.4 Second, it is surprisingly difficult to describe the level of fallible justification required for knowledge in a clear and non-arbitrary

Skepticism, Fallibilism, Evaluation  173 way.5 Despite this lack of precision, however, it is clear that contemporary fallibilists typically endorse the following two claims: first, knowledge is compatible with our cognitive fallibilities as inquirers; second, we typically meet the level of justification required for knowledge. In other words, fallibilists are not usually skeptics. As Stephen Hetherington writes in his encyclopedia entry on fallibilism, it is fallibilist epistemologists (which is to say, the majority of epistemologists) who tend not to be skeptics … Generally, those epistemologists see themselves as thinking about knowledge and justification in a comparatively realistic way — by recognizing the fallibilist realities of human cognitive capacities, even while accommodating those fallibilities within a theory that allows perpetually fallible people to have knowledge and justified beliefs. (Hetherington 2019) But fallibilism does not necessarily escape skepticism. A theory might be fallibilist while still espousing standards too demanding to be regularly met. In other words, it is coherent to be both a fallibilist and a skeptic. A ‘fallibilist skeptic’ is someone who endorses the following two theses: first, the level of justification required for knowledge is “less than fully conclusive”, so we need not guarantee the truth of our beliefs to have knowledge (i.e. fallibilism); second, many of our ordinary knowledge claims are nevertheless false or unjustified (i.e. skepticism). This might seem like a puzzling combination of ideas. In contemporary epistemology, fallibilism and skepticism are often depicted as opposing views: we embrace fallibilism to escape skepticism, and to deny fallibilism is to risk skepticism. However, fallibilism alone does not guarantee that most of our ordinary knowledge claims are true. In this chapter, I will defend a version of skepticism that is compatible with fallibilism and supported by recent work in psychology. In particular, I will argue that we often cannot properly trust our ability to rationally evaluate reasons, arguments, and evidence (a fundamental knowledgeseeking faculty). We humans are just too cognitively impaired to achieve even fallible knowledge, at least for many beliefs.

2 High-Standards Skepticism Fallibilists often complain that skeptics wrongly impose impossibly high standards for knowledge. The skeptic claims, for instance, that knowledge requires one to be absolutely certain and that absolute certainty is impossible or rare (see Unger 1975).6 This view demands that we justify our beliefs to the highest possible degree. I will therefore call it ‘high-standards skepticism’.

174  Michael Hannon Descartes seems to endorse a version of high-standards skepticism in the Meditations. He writes: Reason now leads me to think I should hold back my assent from opinions which are not completely certain and indubitable just as carefully as I do from those which are patently false. So, for the purpose of rejecting all my opinions, it will be enough if I find in each of them at least some reason for doubt. (Descartes 1998: 59) Scholars often interpret Descartes as endorsing the following idea: the bar for knowledge is set so high as to demand infallibility or absolute certainty. We also find this idea in Descartes’ Rules for the Direction of the Mind, where he writes: “All knowledge is certain and evident cognition”. Although Descartes was himself not a skeptic, he seems to imply that knowledge requires us to meet a very demanding standard.7 Likewise, the skeptic seems to presuppose an infallibilist principle like this one: if I know that p, then I can eliminate all grounds for doubting it. Put another way, Descartes and the skeptic seem to suggest that knowing p requires one to have evidence or reasons sufficient to rule out all the alternative possibilities to p. This is a version of ‘high-standards’ skepticism because it requires knowers to justify their belief to the highest possible degree.8 A common response to such skeptical arguments is to treat them as depending on too stringent a conception of knowledge (Reed 2012: 585). That is, we may agree with the skeptic that hardly any belief can be justified to the highest degree. This is because our cognitive faculties are too imperfect to establish the truth of a proposition with 100% certainty. But who cares? So what if nothing meets this incredibly high standard? After all, many things are probably true and it is reasonable for us to believe those things. Indeed, even granting that nothing is certain, why conclude that we have no knowledge? While the skeptic believes that we cannot know what we cannot confirm with 100% certainty, fallibilists have a more moderate view. Absolute certainty is not required for knowledge. As Baron Reed puts it, “our faculties are still very good … they allow us to achieve a more modest sort of cognitive success. Fallibilism, then, takes that modest success to be knowledge” (2012: 585). Duncan Pritchard expresses a similar thought: it doesn’t seem at all credible that the bar for knowledge should be set so high as to demand infallibility or absolute certainty (or, for that matter, indubitability). On the contrary, our everyday conception of knowledge seems to leave us perfectly happy with the idea that knowledge can be fallible and not absolutely certain (and thus to a degree dubitable) while being bona fide knowledge nonetheless. (2019: 36)

Skepticism, Fallibilism, Evaluation  175 I think this reaction is exactly right. If the skeptic sets the bar for knowledge too high, then we should reject that standard (see Hannon 2019a). The epistemic contextualists have adopted this line of reasoning.9 Their solution to skepticism involves two basic elements: first, in ordinary contexts we often meet the reasonable (fallible) epistemic standard for knowledge; second, in skeptical contexts the standards to know are much higher.10 As DeRose puts it, In some contexts, ‘‘S knows that p’’ requires for its truth that S have a true belief that p and also be in a very strong epistemic position with respect to p, while in other contexts, an assertion of the very same sentence may require for its truth, in addition to S’s having a true belief that p, only that S meet some lower epistemic standard. (2009: 3) A core feature of contextualism is that we need not meet the skeptic’s very high epistemic standard to have knowledge in daily life. This is fallibilism. The contextualist is a fallibilist who rejects (or at least confines) the infallible standard assumed by the skeptic. While this response to skepticism is prima facie plausible, it only gains purchase if we satisfy some reasonable epistemic standard that ordinarily suffices for knowledge. After all, the contextualist line is precisely that in everyday contexts our knowledge claims are true because the standards are not too demanding. However, the most challenging skeptical arguments do not simply claim that we fail to meet some extraordinary standard for knowledge. Rather, they claim that we do not meet even ordinary (fallible) standards. I will consider this view in the next section.

3 Hard-Hitting Skepticism The contextualist portrays the dispute between the skeptic and the non-skeptic as a difference between using stricter standards and more lax ones. On this interpretation, the skeptic may be accused of merely imposing abnormal requirements on our familiar concept of knowledge. As Hans-Johann Glock says, “the skeptic … is like someone who claims that there are no physicians in London, since by ‘physician’ he understands someone who can cure any disease within twenty minutes” (2010: 100). But this misportrays the skeptic’s view. As Bryan Frances writes, “the skeptic isn’t complaining that our knowledge doesn’t satisfy some super-duper high-octane condition that only a philosopher could love” (2008: 243). Rather, the skeptic is arguing that it is much more difficult than we realized for a belief to qualify as knowledge even by ordinary standards.11 In presenting her argument, the skeptic raises doubts about whether we actually satisfy the very same epistemic standards that we

176  Michael Hannon have always thought we satisfied, not some unattainably high standard. Thus, any solution that characterizes the skeptic as “raising the standards” or smuggling in abnormal requirements would mischaracterize the view. What about the skepticism of Descartes’ Meditations? As previously mentioned, several philosophers have interpreted the Cartesian skeptic as imposing high standards or presupposing infallibilism.12 But this connection between infallibilism and skepticism may be a red herring. As John Greco (2008: 116), Allan Hazlett (2014: 90), and others have pointed out, the Cartesian skeptical argument does not essentially depend on infallibilism. The Cartesian skeptical argument runs as follows: 1 2 3

I know that I have hands only if I can know that I’m not deceived by a demon (about whether I have hands). I can’t know that I’m not deceived by a demon (about whether I have hands). Therefore, I don’t know that I have hands.

None of the motivations for the first or second premise presuppose infallibilism. The Cartesian skeptic is not claiming that you cannot be certain that you’re not deceived by a demon, and therefore know very little. That would presuppose infallibilism. The idea behind the Cartesian skeptical argument is, as Hazlett says, “that you’ve got no way of knowing whether you’re deceived, and therefore know very little” (2014: 90). Nevertheless, many people are unwilling to grant the skeptical premise that we don’t (or can’t) know that we’re not deceived. Instead of allowing the skeptic to use claims like “I can’t know that I’m not being deceived” as premises in her reasoning, we should instead expect the skeptic to convince us that we can’t know this.13 As Greco writes, Is that premise [that I cannot know that I am not a brain in a vat] initially (or pretheoretically) plausible? It seems to me that it is not. In fact, it seems to me that it is initially obvious that I do know that I am not a brain in a vat. (2008: 111) Whatever we think of this response to the skeptic, there are two important lessons to draw out from the current discussion. First, to portray the skeptic as demanding that we justify our beliefs to the highest degree is to mischaracterize their view. The really worrying form of skepticism is not ‘high standards’ skepticism but rather what I call ‘hard-hitting skepticism’. The hard-hitting skeptic says that our beliefs fail to qualify as knowledge by ordinary standards.

Skepticism, Fallibilism, Evaluation  177 The second lesson is this: a common source of resistance to skepticism is the implausibility of far-fetched scenarios involving evil demons, brains in vats, and so forth. As Frances (2008) observes, this is one thing that commonly bothers undergraduates in philosophy. They object: why on earth do some philosophers take the BIV [brain in a vat] hypothesis to pose any threat at all to our beliefs, given that those very same philosophers think that there is no real chance that the BIV hypothesis is true? Sure, the BIV hypothesis is formally inconsistent with my belief that I have hands, so if the former is true, then my belief is false. But so what? Why should that bare inconsistency matter so much? The students would understand the fuss over the BIV hypothesis if there were some decent reason to think that the BIV hypothesis was really true. (Frances 2008: 225) A more threatening and significant type of skepticism, then, would meet two conditions. First, it would not characterize the skeptic as demanding infallibility or the highest possible degree of justification. Second, it would not be based on doubts that are “purely philosophical” or “merely academic threats” (Frances 2008: 228). Rather, it would involve what Charles Peirce calls a “real” doubt. (We don’t really doubt the existence of the external world.) In the next section, I will outline a skeptical argument that meets these two criteria and thereby generates a “real” skeptical threat.14

4 Skepticism and Rational Evaluation This section will outline a type of contingent real-world skepticism that has not received much attention. Unlike ‘high-standards’ skepticism, this view does not demand infallibility or extraordinarily high justification. Unlike Cartesian skepticism, it does not rely on remote possibilities, such as the possibility that one is a brain in a vat or radically deceived by an evil demon.15 Rather, I will present an empirically informed, scientifically respectable skeptical hypothesis that targets many of our most cherished beliefs. More specifically, I will outline a type of skepticism that targets many beliefs that are the product of reasoning. I am using ‘reasoning’ quite broadly to refer to cases in which we have evidence and draw conclusions on the basis of rationally evaluating the evidence. Put differently, I am thinking of cases in which we exercise our rational capacities to acquire knowledge. This contrasts with more basic and immediate cases of perceptual knowledge, which require little to no reflection or reasoning. Consider the following example.16 Irena suspects that the death penalty is not an effective tool to reduce the murder rate, but she doesn’t

178  Michael Hannon have much evidence either way. To become more informed, she decides to look for evidence about the deterrent effects of capital punishment. After searching several reputable websites and reading multiple academic articles, Irena concludes that the evidence strongly indicates that the death penalty does not reduce the murder rate. Irena is intelligent, articulate, and is now able to present reasons for her belief. Moreover, she now believes that the reasons for which she holds her belief are the reasons she is now able to present. This might seem like a reliable way to form beliefs, but we have abundant evidence that this process of reasoning is likely beset by a host of cognitive biases and reasoning errors. For example, the general human tendency to rationalize is incredibly common.17 This occurs even when people offer what in fact turn out to be sound arguments. Though they may have successfully hit on a good argument, they are still rationalizing because they would have held the same view with or without good reasons, i.e. on the basis of non-rational considerations. Suppose that Irena has views about the morality of the death penalty that she holds on grounds independently of her views about its deterrent effects. These moral commitments will likely drive her views about the deterrent effects of the death penalty. For instance, Irena will inflate the quality of the studies that present evidence in favor of her moral view and she will be far more critical of (and more likely to downplay the significance of) studies that provide evidence that cut against her moral view. In other words, her views about the death penalty’s effectiveness will not be the result of her understanding the relevant data. Quite the opposite: her understanding of the relevant data will be the product of her moral beliefs. Moreover, Irena will be completely unaware of this fact. She will sincerely believe that her reasons for belief are different from her actual reasons. Irena is not a unique case. A vast amount of work in cognitive psychology indicates that we all frequently interpret and filter evidence in ways that fit with our antecedent worldview.18 For example, we selectively expose ourselves to evidence that confirms our pre-existing beliefs and avoid information that conflicts with them. This is known as selective exposure (Nickerson 1998). We also tend to uncritically accept (and better remember) evidence that is favorable to our view, whereas we are far more critical (and forgetful) of counterevidence. We “routinely rationalize the facts, figures, and arguments that [we] cannot effortlessly discount, depreciate, denigrate, or deny” (Lodge and Taber 2013: 59). This general human tendency to accept confirming evidence without much scrutiny and subject disconfirming evidence to highly critical evaluation is known as biased assimilation or confirmation bias (Lord et al. 1979). When this occurs, two people can look at the exact same body of evidence and yet walk away with radically different conclusions about what the evidence shows, thereby drawing undue support for their initial positions.19

Skepticism, Fallibilism, Evaluation  179 In general, our belief-forming processes are often corrupted by an array of normal human cognitive and affective tendencies. Following Aaron Ancell (2019: 411), I will call these “sources of unreason”. Sources of unreason include: our prejudices and biases (both implicit and explicit); the tendency of self-interest and group-interest to distort our judgments; stubbornness and dogmatism; bad reasoning; the desire to reduce cognitive dissonance; and psychological comfort. Our reasoning is especially prone to error or bias when it comes to beliefs that matter to us; e.g., our moral, political, and personal beliefs that are partly constitutive of our identity (Haidt 2012). These ‘identity-constitutive beliefs’ include any belief that reflects one’s conception of “who they are, of what sort of people they are, and how they relate to others” (Abrams and Hogg 1988: 2). Moreover, these corrupting motivations are not transparent to us. We believe that the reasons we present are the reasons for which we hold our beliefs, but we are often wrong about this. We think we are motivated by the desire for truth, but our other motives are concealed from our view (Wilson 2002). This raises a general doubt about human reasoning. As Ancell writes, “Our reasoning capacities are beset by an array of built-in cognitive and affective biases that make it very difficult—often practically impossible—for us to think clearly and objectively about issues that affect our interests and arouse our passions” (2019: 418). This would not be so worrying if we were able to introspectively discern our own biases and cognitive shortcomings. However, we are typically introspectively blind to them (Pronin et al. 2002). 20 While it might seem to us that we are being impartial and unbiased, we are often “twisting the argument and evidence to make them fit the conclusions we ‘want’ to reach” (Ancell 2019: 419). We suffer from what psychologists call the illusion of objectivity (Kunda 1990). In addition to the general human tendencies to rationalize, assimilate information in biased ways, and selectively expose ourselves to favorable information, there are many other biases and cognitive errors to which we humans are prone. For instance, multiple studies have confirmed the existence of a hindsight bias, which occurs when people who know the outcome of an event judge it to be more probable than people who are ignorant of the outcome (Roese and Vohs 2012). This is known to affect judgments about topics as diverse as terrorist attacks, medical diagnoses, and accounting decisions. In addition, we all have implicit biases where we unconsciously and automatically associate concepts with one another. Such biases are especially pernicious when we associate certain traits (e.g., dangerous) with members of particular social groups.21 Further, there is considerable evidence that many of our most cherished beliefs are shaped by irrelevant influences that do not bear on the truth of what we believe (Vavova 2018). For instance, the fact that you were raised in one community rather than another seems epistemically irrelevant to what you ought to believe about God, morality, or politics. But

180  Michael Hannon factors like upbringing are known to guide our convictions on these and other, less charged, topics. 22 (While this is not itself a psychological bias, it does further illustrate the vast extent to which our beliefs are not the product of a reliable belief-forming process.) These are just a handful of our many cognitive shortcomings. It would be impossible to provide a comprehensive overview of all the epistemically irrational biases in human reasoning that should be of interest to epistemologists. However, these few examples illustrate a general worry about human reasoning. We fail on a variety of cognitive dimensions and, as a result, we have a distorted sense of the plausibility of our own beliefs. In other words, our epistemic situation is likely much worse than we think. Moreover, these cognitive shortcomings are not rare but rather are the norm. There is plenty of evidence that rationalization is extremely widespread, especially when it comes to beliefs that matter to us, e.g., our political beliefs, moral beliefs, religious beliefs, or any other issue that affects our interests or stirs our emotions (Haidt 2012). Astute observers of human nature anticipated these psychological findings. Francis Bacon wrote, “The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable in itself) draws all things else to support and agree with it” (1620: XLVI). Bertrand Russell said, “It is a law of our being that, whenever it is in any way possible, we adopt beliefs as will preserve our self-respect” (2004 [1928]: 51). British essayist William Hazlitt noted, “The narrowness of the heart warps the understanding, and makes us weigh objects in the scales of our self-love, instead of those of truth and justice” (1930: 34). And John Locke remarked on the deplorable state of the human mind in his posthumously published work, Of the Conduct of the Understanding: There are several weaknesses and defects in the understanding, either from the natural temper of the mind or ill habits taken up, which hinder it in its progress to knowledge. Of these there are as many possibly to be found, if the mind were thoroughly studied, as there are diseases of the body, each whereof clogs and disables the understanding to some degree, and therefore deserves to be looked at and cured. (1996 [1706], 187; §12) Although Locke was keenly aware of our intellectual imperfections, he was optimistic about our ability to overcome them. He prescribes that we impartially self-examine our own beliefs to root out “the prejudices imbibed from education, party, reverence, fashion, interest” (1996 [1706], 184; §10). In other words, he thinks that rigorous self-study will help us to expose our biases. But this seems overly optimistic. Self-judgment

Skepticism, Fallibilism, Evaluation  181 is often clouded by rationalization and prejudice, so attempts to watch for signs of bad reasoning will go undetected and leave us thinking that we’ve actually tuned them out. As Nathan Ballantyne writes, “The feeling that we’ve done our best to be unbiased will encourage us to think we are unbiased, but that feeling should not be trusted” (2019: 131–132).

5 A Skeptical Argument These biases and psychological shortcomings pose a significant epistemic threat to our beliefs. When our beliefs are shaped by sources of unreason, this casts doubt on the epistemic quality of those beliefs. This implies a form of skepticism. Consider the following skeptical argument, which I will call skepticism from unreason: 1 2 3

We have good evidence that many of our beliefs are shaped by sources of unreason. When one has good evidence that a belief is shaped by a source of unreason, we should significantly reduce confidence in it. Therefore, we should significantly reduce our confidence in many of our beliefs.

We can also derive a form of skepticism about knowledge: 4 5 6

We have a reason to significantly reduce our confidence in many of our beliefs. If we have a reason to significantly reduce our confidence in a belief, then that belief does not amount to knowledge. Therefore, many of our beliefs do not amount to knowledge.

This does not yet give us a very worrying skeptical result. While it does show that many of our beliefs do not constitute knowledge, it leaves open the possibility that we can tell when the process of rational evaluation generates justified (or known) beliefs and when it goes awry. To derive a worrying skeptical conclusion, we must not only show that some of our beliefs are shaped by sources of unreason. (We knew that already.) Rather, it must also be true that we are unable to distinguish the beliefs that are shaped by sources of unreason from the beliefs that have not been corrupted by our cognitive biases. (If we were able to determine which beliefs are the result of twisting the evidence, the skeptical conclusion would be much less threatening.) As mentioned, however, the biasing processes that lead us to rationalize, to selectively filter our evidence, and so forth, typically take place behind the scenes.23 Thus, we often cannot cancel out the threat posed by our psychological shortcomings, for we lack some epistemic feature (e.g., appropriate evidence,

182  Michael Hannon reliability, or what not) that would otherwise cancel out this threat. As Ancell writes, we lack reliable means of detecting, avoiding, and correcting for these biases, [so] we cannot plausibly expect that otherwise reasonable people will never be led astray by them. Indeed, we must expect the opposite; the … reasoning of sincere and conscientious people will often be warped by their biases, interests, partisan loyalties, and so on. Because such warped views are liable to be unreasonable, it follows that sincere and conscientious people will often hold unreasonable views. (2019: 411) All this should lead us to doubt that our reasoning faculties are reliable routes to knowledge. Much empirical research over the past 50 years reveals the disturbingly expansive range to which fallibility enters our cognitive lives. 24 This challenges the rational standing of our beliefs. One might try to safeguard much of our knowledge from these empirical findings by arguing as follows: these findings do not undermine the justification for our beliefs but rather they show that we are often more confident than we actually should be, and thus we have irrational credences. However (the objection continues), it is not obvious how irrational credences translate to the justification of beliefs, which are ultimately relevant for knowledge.25 To illustrate, let’s assume a simple-minded threshold view where to have a belief is to have a credence above a certain threshold. Given that, a belief should count as justified when it is rational to have a credence above the threshold. Now it could easily be that it is rational to have a credence above the threshold even when it is not rational to have the exact credence one has. Thus, even if we grant that our credences are rarely rational, it may still be that our beliefs are justified most of the time. The only cases where a bias affects the justificatory status of a belief are cases where the respective bias leads us to cross the threshold. For instance, let’s say the threshold is 0.8 and my rational credence is 0.75, but due to a bias, my actual credence is 0.85. Here, I end up with an unjustified belief, given the threshold view. The conditions described, however, seem relatively specific, which may lead one to conclude that they are rarely instantiated—or so the objection goes. This objection highlights an important difference between externalist vs. internalist conceptions of justification. According to the above objection, an agent may have an irrational credence that results from an inability to properly evaluate the strength of their evidence; yet the agent may nonetheless achieve a level of justification that suffices for their belief to be knowledge. However, many epistemologists will argue that these various biases make it impossible for one to determine how strong one’s

Skepticism, Fallibilism, Evaluation  183 justification actually is, and thus one cannot tell whether their justification is good enough for knowledge. For a hardline externalist, what matters might be whether our actual justification is good enough for knowledge. But those with internalist leanings, like myself, will claim that our inability to tell how good our justification actually is provides us with a relevant defeater for knowledge. After all, biases like selective exposure and biased assimilation lead us to draw undue support for our views by filtering and processing the relevant evidence in epistemically problematic ways. As a result, the justification for our beliefs is often the result of twisting the evidence and arguments to make them fit with conclusions we want to reach. Thus, our evidence is likely worse than we think. Moreover, we cannot tell whether it is good enough to qualify as knowledge. This provides us with a reason to doubt whether our beliefs do qualify as knowledge, in addition to doubting that our credences are rational. While I have argued that our beliefs often are shaped by sources of unreason, my skeptical argument does not actually require our beliefs to have been shaped by any such source. All that is required is that the relevant belief is one that reasonably could have been influenced by a source of unreason and we cannot tell whether this has occurred in the relevant case. When this happens, we have a defeater even for those beliefs that are, in fact, based on good reasons or evidence. To illustrate, recall my example of Irena and the death penalty. In this case, I admitted the possibility that she in fact hit upon good arguments for her resulting belief. Still, the worry is that humans have a general tendency to rationalize and Irena lacks the ability to tell whether she holds her belief about the death penalty’s deterrent effects on the basis of good reasons or instead due to her views about the morality of the death penalty. 26 These pessimistic facts about reasoning and rationalization go some distance toward making sense of “fallibilist skepticism”. This view is skeptical because it provides a real, live hypothesis that targets reasoning as a source of knowledge. It is also fallibilist because this new skeptical hypothesis does not merely deny certainty or ‘high standards’ knowledge. Rather, it generates a reason to deny the entirely modest amount of epistemic warrant we ordinarily expect to know something. Even though many of our beliefs may be true, the justification we have for these beliefs is much less than any of us have supposed in our antiskeptical moments. In short, the epistemic quality of our position is much worse than we thought.

6 The Scope and Force of Skepticism This view has a narrower scope than some traditional forms of skepticism. The scope of skepticism is determined by the set of propositions it targets, where these propositions are said to be unknowable, unjustified, or those about which we should suspend judgment. Some skeptics

184  Michael Hannon target all claims about the external world. Others target our knowledge of the mental lives of others. Still others target only propositions about the future, the past, or religious matters. The type of skepticism I am outlining targets the propositional contents of many beliefs yielded by rational evaluation.27 These beliefs are targeted on the grounds that they are likely corrupted by error and bias, especially when these beliefs are partly constitutive of our identity. Not all beliefs that are the product of reasoning will be subject to the concerns I have raised. Suppose, for example, that I come to believe that we have enough milk for the next week on the grounds that we have four liters and we tend to go through about a liter every two days. A belief like this is not (obviously) likely to be shaped by “sources of unreason”. 29 Many of our beliefs formed through reasoning will be of this form. Still, the literature in psychology makes clear that the extent to which our beliefs are shaped by sources of unreason is indeed quite broad. In addition to pretty much all of our moral, political, philosophical, and religious beliefs, it will also concern other “cherished beliefs” (e.g., related to family members, favorite sports teams, professional employment), beliefs that are influenced by our implicit biases and prejudices (see Saul 2013 for the extent to which this is troubling), beliefs influenced by ‘ingroup’ and ‘outgroup’ bias (the research on ‘minimal group paradigms’ reveals that even arbitrary distinctions between groups, such as preferences for certain paintings, beliefs about whether hotdogs are sandwiches, or the color of their shirts can trigger a tendency to favor one’s own group), or other issues that affect our interests, stir our emotions, or challenge our worldview. Perhaps more disturbingly, we also have a reason to doubt any information via testimony that may have been shaped by someone else’s cognitive bias. When acquiring information from the local newspaper, for instance, we have reason to significantly reduce our confidence if it is plausible that the journalist may have been partial in their gathering and reporting of news. As Walter Lippmann warned us back in the early 1900s, there is an abundance of biased news reportage that is not free of irrational, unexamined, and unacknowledged prejudgments in the observing, understanding, and reporting of news. This worry extends far beyond journalism: it concerns any form of testimony that may have been the product of sources of unreason. As the literature on epistemic injustice make clear, people routinely give too little (or too much) credibility to certain sources of testimony (see Fricker 2007). So, it’s not just our own biases that we must worry about. It is any information that has been filtered and slanted by the biases of the testifier. 28 Although this version of skepticism has narrower scope than traditional (i.e. radical) skepticism, I believe it packs more punch. Instead of relying on the far-fetched possibility that an evil demon is controlling

Skepticism, Fallibilism, Evaluation  185 our minds (or that we are brains in vats, etc.), we have the much more live (and empirically supported) hypothesis that our cognitive biases are leading us astray. It is not just the possibility of cognitive error or bias that is raised; the research in psychology and cognitive science suggests that it is very likely that we are doing these things quite often.30 As a result, this type of skepticism is more likely to engender doubt and to inspire behavior. This contrasts with the fairly ineffective change in belief and behavior brought about by traditional forms of skepticism. As Hume observed, the activities of ordinary life were sufficient to dispel traditional skeptical doubts: Most fortunately it happens, that since reason is incapable of dispelling these clouds, nature herself suffices for that purpose, and cures me of this philosophical melancholy and delirium … I dine, I play a game of back-gammon, I converse, and am merry with my friends; and when after three or four hour’s amusement, I wou’d return to these speculations, they appear so cold, and strain’d, and ridiculous, that I cannot find in my heart to enter into them any farther. (2003: bk.1, pt.4, sec.7) Similarly, Peirce wrote, “Let us not pretend to doubt in philosophy what we do not doubt in our hearts” (1935: CP 5.265). The most forceful and interesting skeptical challenges provide us with doubts that are genuinely compelling—i.e. they present us with good reasons to think we are very likely making errors and they challenge our ability to inquire responsibly. 31 Skepticism about rational evaluation does precisely this.

7 Final Thoughts Is this type of skepticism really all that troubling? You might think the doubt cast on our beliefs is fairly localized. Saul considers this objection in her article on skepticism and implicit bias. She writes, It seems, at first, to be like the sort of doubt we experience when we discover how poor we are at probabilistic reasoning. We have extremely good reason to think we’re making errors when we make judgments of likelihood. But this sort of doubt doesn’t trouble us all that much because we know exactly when we should worry and what we should do about it: if we find ourselves estimating likelihood, we should mistrust our instincts and either follow mechanical procedures we’ve learned or consult an expert (if not in person, then on the internet). This kind of worry is one that everyone can accept without feeling drawn into anything like skepticism. (Saul 2012: 250)

186  Michael Hannon It may seem as though skepticism about rational evaluation is like this. One might argue, for example, that we should not worry about this type of skepticism because it only concerns a small set of our total beliefs. Moreover, one might argue that we are aware of these various biases and cognitive shortcomings (after all, they are well documented in psychology), so we can guard against them. There are at least two problems with this reply. First, we should very often be worried about these biases influencing our judgments. As the literature in psychology makes clear, these biases tend to be triggered whenever an issue affects our interests or stirs our emotions. Thus, we should be worried about the rationality of our beliefs in a variety of situations. Second, even though we are aware of the existence of these biases, it is unlikely that we can overcome them. Although a variety of debiasing strategies have been proposed (see Jolls and Sunstein 2006; Larrick 2004; Lilienfeld et al. 2009), Kristoffer Ahlstrom-Vij (2013) identifies two obstacles with the attempt to debias ourselves. First, we are not motivated to engage in debiasing because we do not view ourselves as biased (which is very common). Second, people who are persuaded to engage in debiasing efforts run the risk of both overcorrection and undercorrection (Wilson 2002). Some may prefer to interpret the evidence from psychology in a less pessimistic way. According to an alternative view, many of our so-called biases are not really irrational at all. Rather, they reflect “bounded rationality” (e.g., Gigerenzer et al. 2001) or “instrumental rationality” (e.g., Kolodny and Brunero 2018). We act in ways that are boundedly rational when we employ cognitive shortcuts and rules that give rise to biases. As fallible beings with limited time and cognitive resources, it is rational for us to use heuristics and shortcuts in order to make the best decisions we can, given our limitations, even though this type of rationality gives rise to biases and errors in judgment. 32 Additionally, many of our seemingly irrational beliefs may be the result of instrumental rationality, since these beliefs play an important role in achieving our goals. For example, it may be instrumentally rational for an individual to dogmatically hold on to certain political beliefs in the face of counterevidence because changing one’s mind is a psychologically difficult process that could potentially alienate the believer from their community and sense of self. This line of reasoning is perfectly sensible, but it does little to blunt the force of the skeptical argument I have outlined. The skeptic may simply argue that many of our boundedly rational beliefs do not amount to knowledge. Bounded rationality is achieved when we rely on fast and frugal heuristics to help us make sense of the world. In general, this is a good thing because the world is complex and we don’t always have the time to make a well-thought-out rational choice about a decision. Nevertheless, the skeptic will insist that she is targeting beliefs that are not sufficiently epistemically justified, even though such beliefs may be the result of a boundedly rational process. A similar point can be made about

Skepticism, Fallibilism, Evaluation  187 instrumentally rational beliefs. While it may be instrumentally rational to dogmatically hold on to certain beliefs, it is not epistemically rational to do so. The psychological costs associated with giving up such beliefs may be brutal, but this does not make the belief rational in an epistemic sense. Thus, this line of response does not escape the skeptic’s clutches. A final worry about my argument is that it is self-defeating. After all, haven’t I attempted to convince you to endorse my skeptical conclusion through rational argument? If so, then shouldn’t we doubt whether we know this conclusion? As a skeptic, I am willing to admit that we do not know the conclusion of my argument. It is likely that philosophical argumentation is subject to various epistemic vices and cognitive biases, just as political reasoning, moral thinking, and religious belief are subject to these biases. However, to say that we do not know my conclusion is not to say that we do know that it is false. Rather, we are simply left uncertain (or at least lacking knowledge) as to whether the beliefs that are the product of rational evaluation are known or adequately justified. This is still a skeptical conclusion.33 Moreover, this claim is compatible with the idea that we have some epistemic justification to believe the conclusion of my argument. I need not argue that beliefs entirely lack justification when they are the product of rational evaluation. That is an especially strong form of skepticism that would run into the self-defeat problem, for the following reason: If rational evaluation is unjustified or unreliable, then the rational evaluation that ‘rational evaluation is unjustified or unreliable’ would itself be unjustified. But if it were unjustified, then we have good reason not to believe it or trust it. Still, we may throw the epistemic status of these beliefs into doubt without undermining their justification entirely. Thus, we may have some epistemic justification for believing that my conclusion is true, but without knowing that it is true. Another way to avoid the self-defeat worry, which I do not pursue, is to suggest that my skepticism about rational evaluation does not apply to itself.34 This would allow me to escape the self-defeat worry by claiming that my rational evaluation about the unreliability of rational evaluation is not itself unjustified. However, going this route would leave me with the challenge of convincing you that my rational evaluation is justified. In other words, my skeptical argument would provide a defeater for most beliefs that are the product of rational evaluation, but I would provide a defeater for why this argument applies to the rational evaluation of my argument. This may not be an insurmountable task, for I do not claim that rational evaluation is always or universally an unreliable process. Nevertheless, I have argued that we should very often be worried about these biases influencing our judgments, that we do not know exactly when we should worry, and that attempting to overcome these biases is often unhelpful.35 In summary, I have argued that one of the main lessons from the literature on human psychology is that we should not trust ourselves as inquirers. Many beliefs that we take to be the product of the careful

188  Michael Hannon exercise of reason are likely biased and wrong. Put differently, an ordinary but fundamental belief-forming method is much less reliable than we thought. This gives rise to a form of skepticism about our rational capacities. This skeptical argument challenges what Chris Hookway calls “our cognitive instruments”, since it is generated by doubts about our ability to engage in rational, unbiased evaluation. Moreover, it is unclear how to develop techniques that will drive out what John Stuart Mill called “the fogs which hide from us our own ignorance” (1984 [1867]: 239). 36 I have not attempted to lift the fog; I have simply attempted to illuminate it.

Acknowledgements Thanks to Alexander Dinges, Xingming Hu, Nick Hughes, Christos Kyriacou, Martin Smith, and Mark Walker for valuable comments on a draft of this chapter.

Notes 1 I will use ‘justification’ broadly to include the related notions of ‘evidence’, ‘probability’, ‘warrant’, and ‘reliability’. 2 Some have denied that infallibilism must lead to skepticism. For example, Fred Dretske (1981), Timothy Williamson (2000), Wayne Davis (2007), and Ram Neta (2011) each defend a version of infallibilism that allegedly has non-skeptical results. If these authors are right, then the step from infallibilism to skepticism is not an inevitable one. In contrast, Brown (2018) argues that infallibilists can avoid skepticism only at the cost of problematic commitments concerning evidence and evidential support. However, my focus in this chapter is not on whether infallibilism leads to skepticism, so I will set this issue aside. 3 Michael Williams agrees: “We are all fallibilists nowadays” (2001: 5). And Stewart Cohen says, “the acceptance of fallibilism in epistemology is virtually universal” (1988: 91). 4 See Reed (2002) and Brown (2018) for a discussion. 5 BonJour (2010) makes this objection. See Hetherington (2006) and Hannon (2017) for replies. 6 ‘Certainty’ is ambiguous (see Reed 2008). A belief is psychologically certain when one is supremely convinced of its truth. A belief is epistemically certain when it has the highest possible epistemic status. Some infallibilists say that psychological certainty is necessary for knowledge, while others say that epistemic certainty is necessary. 7 I do not actually agree with this interpretation of Descartes, but I set this issue aside. See Pasnau (2017) for a discussion of this point. 8 Stroud (1984) seems to interpret Descartes as endorsing a ‘high standards’ view. 9 See Cohen (1988), DeRose (1995), and Lewis (1996) for early statements of this view. 10 I have not indicated semantic ascent by putting quotation marks around ‘know’ and ‘knowledge’. I state my discussion in the object language for ease of exposition.

Skepticism, Fallibilism, Evaluation

189

11 Feldman (1999; 2001), Klein (2000), Kornblith (2000), Bach (2005), Hazlett (2014), and Pritchard (2019) raise versions of this objection. 12 Unger (1975) and BonJour (2010) also seem to defend versions of highstandards skepticism. 13 Pryor (2000) attempts to formulate a version of skepticism that does not require this controversial premise. 14 Pyrrhonian skepticism in no way depends on the problematic assumptions outlined above. The pyrrhonist did not claim the standards for knowledge were incredibly high, nor did they appeal to far-fetched skeptical scenarios. 15 Here I follow Frances’s (2005) strategy to outline what he calls a “live” skeptical hypothesis. Frances says that a hypothesis is live when it satisfies five conditions: (i) it has been through a significant evaluation in the community by experts over many years; (ii) it is judged actually true or as likely to be true as any relevant possibility by a significant number of experts; (iii) the judgment of those experts has been reached in an epistemically responsible way; (iv) those experts consider there to be several independent sources of evidence for the hypothesis; and (v) many of those experts consider the hypothesis to be a live possibility (see Frances 2005: 18–19). 16 This example is adapted from Kornblith (1999: 181–182). 17 For a defense of this view, see Kornblith (1999), Tavris and Aronson (2008), Haidt (2012), and Sperber and Mercier (2017). 18 See Gilovich (1993) for an overview. 19 These biases are sometimes collectively referred to as “motivated reasoning”, which is “the tendency to seek out, interpret, evaluate, and weigh evidence and arguments in ways that are systematically biased toward conclusions that we ‘want’ to reach for reasons independent of their truth or warrant” (Ancell 2019: 418). See also Kunda (1990), Ditto et al. (2009), and Lodge and Taber (2013). 20 As Ballantyne (2019: 131) writes, Acentral idea in psychology is that most biases are not reliably detected by introspection (Nisbett and Wilson 1977; Wilson and Brekke 1994; Kahneman 2003). We typically can’t figure out whether we are biased by merely gazing into our minds. Biases normally ‘leave no trace’ in consciousness. As Timothy Wilson and Nancy Brekke quip, ‘Human judgments—even very bad ones—do not smell’ (1994: 121). From the inside, biased attitudes seem just like unbiased ones. 21 Jennifer Saul (2013) has argued, compellingly, that implicit biases present a skeptical challenge to the ordinary ways that we assess reasons, arguments, and evidence. While my argument is similar to Saul’s, it has wider scope. Saul presents a challenge to beliefs that may be influenced by implicit biases, whereas I raise doubts about our rational capacities more generally. That said, we both agree that (a) the rational evaluation of evidence and arguments is often corrupted by problematic psychological tendencies, (b) this influence operates below the level of consciousness, and (c) this provides a reason for skepticism about such beliefs. 22 Sometimes factors like upbringing will affect our convictions in perfectly rational ways, since they expose us to different experiences and thus different bodies of evidence. However, there are also many cases in which factors like upbringing are epistemically irrelevant. 23 Externalists (e.g., reliabilists) might take this as a point in their favor, since they will claim that we needn’t be aware of the reliability of our beliefforming processes. However, externalists are not clearly at an advantage here because the relevant point is that our beliefs are often the result of an unreliable process—not that we lack introspective awareness of when we are rationalizing.

190  Michael Hannon 24 This may reflect the malaise of our allegedly ‘post-truth’ era. It is not so much the idea of truth or the existence of various truths that has come under attack, but rather the notion that there can be any such thing as objective inquiry into it (Blackburn 2019). 25 Thanks to Alexander Dinges for raising this objection. 26 One might take this as a reason in favor of externalist. For what it’s worth, I don’t think we should adopt philosophical positions simply because we don’t like skeptical conclusions. In any case, adopting a version of externalism that allows us to call these beliefs ‘justified’ or ‘knowledge’ would still leave untouched my real concern, namely, that we care about being able to tell how good our epistemic position actually is. I find the externalist retreat of little practical value in responding to this concern, even though it may allow us to continue calling certain beliefs ‘justified’ or ‘knowledge’. 27 We can also put this point in terms of distrusting a putative source of knowledge. While the traditional skeptic targets beliefs yielded by sense perception, my skeptic targets those produced by reasoning or rational evaluation. 28 While I have focused on beliefs that are the product of reasoning (either one’s own or someone else’s), the concerns I have raised may also target some perceptual beliefs and beliefs based on memory. In the case of perceptual beliefs, it is well known that our ‘group identity’ can shape our perceptions of reality. This explains why sports fans who cheer for different teams will perceive games differently (see Hastorf and Cantril 1954). In the case of memory, we suffer from ‘rosy retrospection’ (Mitchell et al. 1997) and hindsight bias (Roese and Vohs 2002). 29 Thanks to Martin Smith for this point and example. 30 Saul (2013) makes a similar point. 31 Frances’s (2005) work on ‘live skeptical scenarios’ provides an instructive comparison. However, the reasons he offers for his brand of skepticism strike me as less compelling. As Saul (2013: 254) observes, The hypotheses in question are things like eliminativism about belief and error theory about colour. And the reasons for thinking that they are still live is that some sensible people who know a great deal endorse (or might endorse) these theories on the grounds of compelling scientific or philosophical reasons. But this falls short of the general doubts about human reasoning that I have outlined. My claim is that we all have very good reason to believe that we are frequently making errors that have their root in motivated reasoning. This is stronger than Frances’s claim that a hypothesis is “live”, by which he means (roughly) that sensible and knowledgeable people might endorse it on the basis of good reasons. I borrow this point from Saul (2013), who makes a similar observation in the context of her own skeptical argument that is based on implicit biases. 32 For this objection, see Nick Hughes’s forthcoming paper, “Evidence and Bias”. 33 Mark Walker suggests another version of the self-undermining objection. It runs as follows: If we are systemically subject to sources of unreason, doesn’t this worry infect cognitive psychology itself? So, if my argument is correct, then we should reduce our confidence after reviewing the cognitive psychology literature. But now it seems I have less to worry about from cognitive psychology, since my confidence in their results is reduced.

Skepticism, Fallibilism, Evaluation

191

In response to this worry, I am willing to admit that we should reduce our confidence in what we take the evidence about cognitive biases to show. However, this does not mean we should doubt whether such biases are operative. Rather, it suggests that the confidence with which we hold our beliefs about cognitive bias may not accurately reflect the evidence. In other words, our situation might be cognitively better than the evidence about bias suggests, but it might also be worse. We can’t tell which way our biases are pushing us. Are we weighing the evidence about bias too heavily or not heavily enough? Thus, we may have a reason to doubt that our confidence about what the evidence shows is adequately reflecting the evidence. But this doesn’t permit us to reduce our confidence in the existence or extent of cognitive bias. 34 Adam Elga (2010: 179–182) argues that self-defeat objections can be avoided because, in general, methods can be exempted from self-application. I will not outline the details of his argument here, but the rough idea is that self-exemption is not ad hoc or arbitrary because, unless we exempt our methods from self-application, they can’t be coherent—that is, they will give inconsistent recommendations in possible cases. As Elga puts it, “in order to be consistent, a fundamental policy, rule, or method must be dogmatic with respect to its own correctness” (2010: 185). 35 See Machuca (2017) for additional strategies to avoid the self-defeat objection. 36 I thank Nathan Ballantyne (2019: 64) for drawing my attention to this quote from Mill.

Bibliography Abrams, Dominic, and Michael Hogg (1988). Comments on the motivational status of self-esteem in social identity and intergroup discrimination. European Journal of Social Psychology 18 (4): 317–334. Ahlstrom-Vij, Kristoffer (2013). Why we cannot rely on ourselves for epistemic improvement. Philosophical Issues 23 (1): 276–296. Ancell, Aaron (2019). The fact of unreasonable pluralism. Journal of the American Philosophical Association 5 (4): 410–428. Bach, Kent (2005). The emperor’s new ‘knows’. In Gerhard Preyer and Georg Peter (eds.), Contextualism in Philosophy: Knowledge, Meaning, and Truth. Oxford University Press. pp. 51–89. Bacon, Francis (1620). The New Organon. Ballantyne, Nathan (2019). Knowing Our Limits. Oxford University Press. Blackburn, Simon (2019). How can we teach objectivity in a post-truth era? New Statesman. Accessed at: https://www.newstatesman.com/culture/2019/02/ how-can-we-teach-objectivity-post-truth-era BonJour, Laurence (2010). The myth of knowledge. Philosophical Perspectives 24 (1): 57–83. Brown, Jessica (2018). Fallibilism: Evidence and Knowledge. Oxford University Press. Kolodny, Niko and John Brunero, “Instrumental Rationality”, The Stanford Encyclopedia of Philosophy (Spring 2020 Edition), Edward N. Zalta (ed.), URL = . Cohen, Stewart (1988). How to be a fallibilist. Philosophical Perspectives 2: 91–123.

192  Michael Hannon Davis, Wayne (2007). Knowledge claims and context: Loose use. Philosophical Studies 132 (3): 395–438. DeRose, Keith (2009). The Case for Contextualism. Oxford University Press. DeRose, Keith (1995). Solving the skeptical problem. Philosophical Review 104 (1): 1–52. Descartes, Rene (1998). Discourse on Method and Meditations on First Philosophy, 4th Ed. Hackett. Ditto, Peter H., David A. Pizarro, and David Tannenbaum (2009). Motivated moral reasoning. Psychology of Learning and Motivation 50: 307–338. Dretske, Fred (1981). The pragmatic dimension of knowledge. Philosophical Studies 40 (3): 363–378. Feldman, Richard (2001). Skeptical problems, contextualist solutions. Philosophical Studies 103 (1): 61–85. Feldman, Richard (1999). Contextualism and skepticism. Philosophical Perspectives 13 (s13): 91–114. Frances, Bryan (2008). Live skeptical hypotheses. In John Greco (ed.), Oxford Handbook of Skepticism. Oxford University Press. pp. 225–245. Frances, Bryan (2005). Scepticism Comes Alive. Oxford University Press. Gigerenzer, G. et al. (2001). Simple Heuristics That Make Us Smart. Oxford University Press. Gilovich, Thomas (1993). How We Know What Isn’t So. Free Press. Glock, Hans-Johann (2010). The development of analytic philosophy: Wittgenstein and after. In D. Moran (ed.), The Routledge Companion to Twentieth Century Philosophy. Routledge. pp. 76–117 Greco, John (2008). Skepticism about the external world. In The Oxford Handbook of Skepticism. Oxford University Press. pp. 108–128. Haidt, Jonathan (2012). The Righteous Mind. Penguin. Hannon, Michael (2019a). What’s the Point of Knowledge? Oxford University Press. Hannon, Michael (2019b). Skepticism: Impractical, therefore implausible. Philosophical Issues 29 (1): 143–158. Hannon, Michael (2017). A solution to knowledge’s threshold problem. Philosophical Studies 174 (3): 607–629. Hastorf, Albert H., and Hadley Cantril (1954). They saw a game; A case study. The Journal of Abnormal and Social Psychology 49 (1): 129–134. Hazlett, Allan (2014). A Critical Introduction to Skepticism. Bloomsbury Academic. Hazlitt, William, P. P. Howe, A. R. Waller, Arnold Glover, and James Thornton. (1930). The complete works of William Hazlitt. London: J.M. Dent and Sons, Ltd. Hetherington, Stephen (2019). “Fallibilism”. Internet Encyclopedia of Philosophy, ISSN 2161-0002, https://www.iep.utm.edu/. Hetherington, Stephen (2006). Knowledge’s boundary problem. Synthese 150 (1): 41–56. Hughes, Nick. Forthcoming. Evidence and bias. Routledge Handbook of the Philosophy of Evidence. Routledge. Hume, David (2003). A Treatise of Human Nature. Courier Corporation. Jolls, Christine, and Cass R. Sunstein (2006). Debiasing through law. The Journal of Legal Studies 35 (1): 199–242.

Skepticism, Fallibilism, Evaluation  193 Kahneman, Daniel (2003). A perspective on judgment and choice: Mapping bounded rationality. American psychologist 58 (9): 697. Klein, Peter D. (2000). Contextualism and the real nature of academic skepticism. Philosophical Issues 10 (1): 108–116. Kornblith, Hilary (2000). The contextualist evasion of epistemology. Noûs 34 (s1): 24–32. Kornblith, Hilary (1999). Distrusting reason. Midwest Studies in Philosophy 23 (1): 181–196. Kunda, Ziva (1990). The case for motivated reasoning. Psychological bulletin 108 (3): 480. Larrick, Richard P. (2004). Debiasing. Blackwell Handbook of Judgment and Decision Making. Blackwell. pp. 316–338. Lewis, David K. (1996). Elusive knowledge. Australasian Journal of Philosophy 74 (4): 549–567. Lilienfeld, Scott O., Rachel Ammirati, and Kristin Landfield (2009). Giving debiasing away: Can psychological research on correcting cognitive errors promote human welfare? Perspectives on Psychological Science 4 (4): 390–398. Locke. John (1996) [1706]. Some Thoughts Concerning Education and Of the Conduct of the Understanding. Edited by Ruth Grant and Nathan Tarcov. Hackett. Lodge, Milton, and Charles S. Taber (2013). The Rationalizing Voter. Cambridge University Press. Lord, Charles G., Lee Ross, and Mark R. Lepper (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology 37 (11): 2098. Machuca, Diego (2017). A neo-Pyrrhonian response to the disagreeing about disagreement argument. Synthese 194 (5): 1663–1680. Mitchell, Terence R., et al. (1997). Temporal adjustments in the evaluation of events: The “rosy view”. Journal of Experimental Social Psychology 33 (4): 421–448. Neta, Ram (2011). A refutation of Cartesian fallibilism. Noûs 45 (4): 658–695. Nickerson, Raymond S (1998). “Confirmation Bias: A Ubiquitous Phenomenon in Many Guises.” Review of General Psychology 2(2): 175–220. Nisbett, Richard E., and Timothy D. Wilson (1977). Telling more than we can know: Verbal reports on mental processes. Psychological review 84 (3): 231. Pasnau, Robert (2017). After Certainty: A History of Our Epistemic Ideals and Illusions. Oxford University Press. Peirce, Charles S. (1935) Collected Papers. Volume 5. Charles Hartshorne and Paul Weiss (eds.). (Cambridge, MA: Harvard University Press). Peirce, Charles S. (1868). Some consequences of four incapacities. Journal of Speculative Philosophy 2: 140–157. Pritchard, Duncan (2019). Skepticism: A Very Short Introduction. Oxford University Press. Pronin, Emily, Daniel Y. Lin, and Lee Ross (2002). The bias blind spot: Perceptions of bias in self versus others. Personality and Social Psychology Bulletin 28 (3): 369–381. Pryor, James (2000). The skeptic and the dogmatist. Noûs 34 (4): 517–549. Reed, Baron (2012). Fallibilism. Philosophy Compass 7 (9): 585–596.

194  Michael Hannon Reed, Baron, “Certainty”, The Stanford Encyclopedia of Philosophy (Winter 2011 Edition), Edward N. Zalta (ed.), URL = . Reed, Baron (2002). How to think about fallibilism. Philosophical Studies 107 (2): 143–157. Roese, Neal J., and Kathleen D. Vohs (2012). Hindsight bias. Perspectives on Psychological Science 7(5): 411–426. Russell, Bertrand (2004) [1928]. Sceptical Essays. Routledge. Saul, Jennifer (2012). Ranking Exercises in Philosophy and Implicit Bias. Journal of Social Philosophy 43 (3): 256–273. Saul, Jennifer (2013). Scepticism and implicit bias. Disputatio 5 (37): 243–263. Siegel, Harvey (1997). Rationality Redeemed? Routledge. Sperber, Dan, and Hugo Mercier (2017). The Enigma of Reason. Harvard University Press. Stroud, Barry (1984). The Significance of Philosophical Scepticism. Oxford University Press. Tavris, Carol, and Elliot Aronson (2008). Mistakes Were Made (but Not by Me): Why We Justify Foolish Beliefs, Bad Decisions, and Hurtful Acts. Houghton Mifflin Harcourt. Unger, Peter (1975). Ignorance: A Case for Scepticism. Oxford University Press. Vavova, Katia (2018). Irrelevant influences. Philosophy and Phenomenological Research 96 (1): 134–152. Williams, Michael (2001). Contextualism, externalism and epistemic standards. Philosophical Studies 103: 1–23. Williamson, Timothy (2000). Knowledge and Its Limits. Oxford University Press. Wilson, Timothy (2002). Strangers to Ourselves. Harvard University Press. Wilson, Timothy, and Nancy Brekke (1994). Mental contamination and mental correction: Unwanted influences on judgments and evaluations. Psychological Bulletin 116 (1): 117.

10 Situationism, Implicit Bias, and Skepticism Kevin Wallbridge

1 Introduction Paradigmatic skeptical arguments argue on the basis of a priori considerations about the nature of knowledge and/or some inescapable aspect of our epistemic situation that knowledge (either tout court or of a certain kind) is impossible. However, these arguments fail to convince most philosophers to be skeptics (some of the present company excluded).1 My aim here is to diagnose why these arguments fail to be convincing and to present a different kind of skeptical argument that is much more convincing. My skeptical argument will be unusual in that it will appeal significantly to a posteriori empirical facts. The conclusion of this skeptical argument is modest (compared to traditional arguments) in two ways: firstly, because it concludes that we know a great deal less than we ordinarily suppose but certainly not nothing, and secondly, because it does not conclude that such knowledge is impossible but merely that, as a matter of contingent fact, we frequently lack it. (The boundary marking the exact extent of the things that we do not know is something that I can only specify vaguely, but well enough to make clear that it is a great deal.) Along the way, it will be informative to look at a recent skeptical challenge that also appeals to empirical evidence but, I argue, still falls into the same problem as traditional skeptical arguments and is, for the same reason, unconvincing.

2 Traditional Skeptical Arguments Traditional skeptical arguments (or at least the kind that is particularly prevalent in modern and contemporary philosophy)2 begin by considering some situation in which we are subject to misleading impressions and form false beliefs as a result. And they draw attention to the fact that we lack independent evidence that this situation is not our actual situation. Then, by applying some premise about the nature of knowledge, they conclude that we must lack knowledge. The most common way of putting this premise in contemporary epistemology has been in terms of a

196  Kevin Wallbridge knowledge closure principle. So the traditional skeptical argument takes the following form: Traditional Argument (TA1) I cannot know that I am not a BIV (brain-in-a-vat). (Or dreaming, or fooled by an evil demon, etc.) (TA2) If I know that I have hands then I can know that I am not a BIV, since I can just deduce the latter from the former.3 (TA3) But I cannot know that, therefore I do not know that I have hands. This kind of skeptical argument, despite being logically sound and having initially appealing premises, has generally failed to convince epistemologists of its conclusion. The problem with this kind of argument is that it is generally considered much more likely that one of the premises is false than that the conclusion is true. Moore (1939), for instance, thought that it was very certain that the conclusion was not true and this in fact meant one could prove the falsity of (TA1) by inverting the argument: Moore’s Proof of an External World (MP1) I know that I have hands. (MP2) If I know that I have hands then I can know that I am not a BIV, since I can just deduce the latter from the former. (MP3) I can know that I am not a BIV. But while Moore rejected (TA1), one might equally reject (TA2)4 or even remain uncommitted to which of the premises is false. The important point is that the simple claim that we have knowledge of everyday propositions (such as I have hands) is considered more certain than the philosophical premises that traditional skeptical arguments appeal to. (Or perhaps to the soundness of the argument.) So such arguments are bound to be found unconvincing. This problem is very general. So long as the common sense claim that we have knowledge of everyday propositions seems quite certain, then any philosophical argument that appeals to philosophical principles that are any less certain will suffer from the same problem (for instance the underdetermination principle that Pritchard identifies as a potential source of skepticism elsewhere in this volume). But nonetheless, I believe that there is a different kind of skeptical argument that does not fall into this trap.

3 Situationism More recently, a form of skeptical argument has been discussed which departs from traditional skeptical arguments (and shares several features with the kind of skeptical argument that I ultimately wish to endorse,

Situationism, Implicit Bias, Skepticism  197 namely it is empirically based and concludes that we lack knowledge as a matter of contingent fact, not that it is impossible). The argument goes roughly like this:5 The Situationist Challenge (SC1) Knowledge is true beliefs that are formed via the exercise of intellectually virtuous character traits. (SC2) Empirical fact: human beings don’t actually have character traits. (SC3) Therefore, human beings don’t actually know anything. This is ‘the situationist challenge’ to virtue responsibilism. However, although (SC1)–(SC3) takes the form of an argument for skepticism, the situationist challenge is actually intended to serve as a reductio ad absurdum of the first premise, i.e. the existence of this argument is taken to be a reason to reject virtue responsibilist accounts of knowledge, so as to avoid the skeptical conclusion. So it is not treated as a compelling argument for skepticism. Nonetheless this kind of argument provides an informative parallel to my own argument, so it is worth spending some time appreciating it. The middle premise (SC2) is argued to follow from ‘situationist’ accounts of cognitive psychology, according to which people lack any stable character traits, i.e. people do not really have characters that make them brave or cowardly or some mid-point between these two, nor are they selfish/unselfish/somewhat unselfish, nor compassionate, generous, sadistic, cold hearted, etc. According to this view, differential human behavior is not the result of robust character traits and motivations as we suppose in our folk psychology. Folk psychology is mistaken and has posited psychological categories that do not actually exist. Whether someone acts bravely, selfishly, or generously is not to be explained by appeal to stable character traits that people possess but instead is predicted and explained by external, situational factors. This view is supported by various studies showing that the way that people respond to a situation, say a moral situation, is surprisingly well predicted by morally irrelevant features of the situation, such as whether the environment smells of bread, whether it is noisy, how many bystanders there are, etc.6 Meanwhile their responses aren’t well predicted by the moral features of the situation in combination with the character traits that they and their peers assess them as having. In the light of results like these, the idea that people possess robust character traits and motivations which underlie their actions is taken to have been shown to be false—in actual fact people do not possess these kinds of virtues (people are not in fact brave, honest, caring, although they might act bravely, honestly, or with compassion if they are in a situation with certain features).

198  Kevin Wallbridge 3.1 Responsibilist Replies to Situationism Needless to say, virtue theorists haven’t taken the claim that there are no such things as character virtues lying down. The general line of response has been to clarify the virtue condition that they are working with in such a way that it is weak enough that the kinds of experimental results cited above do not show there to be a lack of virtues in the relevant sense. One way of doing so is to abandon the traditional broad virtues like courage and kindness and replace them with narrow (more empirically adequate) virtues, like ‘kind when in a sweet-smelling environment.’ See, e.g., Doris (2002) and Olin and Doris (2014). These are perhaps less impressive than the traditional virtues, but these are still character traits and still normatively potent (albeit less so than the traditional virtues). Similarly, Pritchard (2014) and Carter and Pritchard (2017) argue that the situationist challenge does not touch modest virtue epistemology, according to which features of the agent need not be the only thing playing a substantial role in explaining why one believes the right thing. Aspects of the environment can contribute to explanations of an agent’s actions, so long as there is still some role for an agent’s cognitive traits, then the agent can be credited with having done the right thing. Finally, King (2014) points out that the premise (SC1), that knowledge is true beliefs that are formed via the exercise of intellectually virtuous character traits, is not a feature of all virtue responsibilist views, for instance Zagzebski’s. According to Zagzebski, knowledge is true belief that arises out of virtuous action, but she carefully states that to perform a virtuous act, an agent need not actually possess the relevant virtue, just that “she must be virtuously motivated, she must act the way a virtuous person would characteristically act in the same circumstances” (1996, p. 279). So the claim that no one actually possesses any virtues does not present an obstacle to Zagzebski’s kind of virtue responsibilism, since it is compatible with this that people nonetheless often do act virtuously. 3.2 Summing Up the Situationist Skeptical Argument Although the situationist challenge presents an argument for skepticism, no one takes it to provide a good reason to endorse skepticism. Proponents of the challenge take it to be a reason to reject (SC1) the claim that knowledge requires intellectually virtuous character traits (although, like Zagzebski, they may still endorse some other claim linking knowledge and virtue). While the epistemologists targeted by the challenge instead reject the empirical claim (2) that people don’t have the required character traits (aka virtues), arguing that the empirical results are perfectly compatible with the claim that people often come to their beliefs as the result of exercising intellectual virtue.

Situationism, Implicit Bias, Skepticism  199 In other words, the situationist skeptical argument suffers from the same problem as traditional skeptical arguments: the fact that skepticism is false and we have everyday knowledge seems to be more certain than the fact that the premises of the argument are true since there is room to resist both premises.

4 A Challenge from Implicit Bias But even if virtue epistemologists are right about the falsity of (SC2) (so there is some relevant sense of virtuous character or virtuous action on which such things exist), there is a related skeptical worry that one might have. Even if there are such things as character traits (and intellectual character traits in particular), it might turn out that there are fewer cases of intellectual virtue, and more cases of intellectual vice, than we usually suppose. In which case, a different kind of skeptical argument could be made: An Alternative Challenge (AC1) Knowledge is true beliefs that are formed via the exercise of intellectual virtues. (AC2) A large amount of the time our beliefs are formed via the exercise of intellectual vices (not virtues). (AC3) Therefore a large amount of the time our beliefs fail to be knowledge. (This argument is a stepping stone, partway to what I take to be the most compelling kind of skeptical argument.) The question, of course, is how plausible is (AC2)? I think it is very plausible. The phenomenon of identity-based implicit bias shows that this is true for many beliefs that we form about people’s qualities and competences.7 In cases of identity-based implicit bias, even though a subject harbors no explicit prejudiced attitudes, i.e. they would not self-ascribe such attitudes even when being fully sincere, they are nonetheless systematically biased in the judgements that they make about people. Such biases are unconscious, involuntary, and highly pervasive (so that, for instance, even highly committed and engaged feminist activists can find themselves ‘guilty’ of forming judgements and expectations that are shaped in illicit ways by gender stereotypes). There is ample evidence demonstrating the widespread existence of this kind of bias, far more than I can outline here.8 But to give a brief example, in various CV studies employers are presented with the same CVs but with different names attached. Half of the names are associated with one particular gender, race, or nationality, while the other half are

200  Kevin Wallbridge associated with another. The results of these studies show clearly that the very same CV will be judged more or less favorably according to whether the name at the top is a typical female name or a typical male name, a typically White name or a typically Arab name, etc. (And all this without the gender/race/nationality of the applicant entering into the potential employer’s explicit reasoning.)9 The existence and prevalence of implicit bias may seem to some to be an initially surprising result. However the wealth of evidence (from controlled studies, analysis of existing data, and anecdotally from those whose lives are constantly impacted by this kind of discrimination) makes it highly certain than this phenomenon exists and is widespread. With the widespread existence of implicit bias as a premise, we can give an empirically based skeptical argument similar in structure to the situationist challenge10: An Implicit Bias Challenge (IB1) Knowledge is true beliefs that are formed via the exercise of intellectual virtues. (IB2) Empirical fact: frequently, when we form beliefs about people’s qualities and competences, our beliefs are formed via the exercise of epistemic vices.11 (IB3) Therefore, many of our beliefs about other people’s qualities and competences fail to be knowledge. Just as with the situationist skeptical argument, this argument has two premises: one philosophical claim about the virtue theoretic requirements of knowledge, and one empirical claim about the instantiation of the relevant virtues. In this case however, the empirical premise is not about the lack of character traits per se, but about the lack of the right character traits being exercised.12 (It appears that vices are playing a starring role.) This premise is backed up by the evidence for implicit bias, which demonstrates that in many cases we do not form our belief by conscientiously weighing up the relevant facts, but rather by resorting to stereotyping which is both epistemically and morally objectionable. There are three possible responses to the argument: reject (IB1), reject (IB2), or accept the argument’s skeptical conclusion (IB3). Let us consider the second option first. Rejecting the claim that these kinds of biases exist tout court is simply not plausible given the preponderance of evidence. But neither does it look possible to retreat to a weaker virtue condition (as some virtue theorists have suggested in response to the situationist challenge), some weaker condition with which these empirical results are compatible. The problem is that there is simply no plausible virtue condition on which people who form

Situationism, Implicit Bias, Skepticism  201 their beliefs as the result of implicit bias count as forming their beliefs virtuously. A much more plausible response to the argument then is to reject (IB1). (Virtue epistemology is after all a contentious theoretical claim that will seem to many to be far from certain.) However, although this may be a viable response to the particular argument given by (IB1)–(IB3), it can’t disarm the general strategy of appealing to implicit bias to motivate skepticism. A virtue theoretic conception of knowledge is unnecessary to make a more general form of this argument work. This is because the prevalence of implicit bias means that facts like the following are often true of our beliefs: you only think that he’s so smart because he’s a well-spoken white male. But this doesn’t only mean that we often fail to exercise appropriately virtuous character traits in forming our beliefs, it also means that there is often a straightforward lack of connection between the reasons for our beliefs and the truth. And this is something that is going to be incompatible with knowledge on any plausible account of what knowledge requires. For instance, no kind of anti-luck condition on knowledge is going to be met in this kind of case, since one could easily have formed a false belief by believing in this kind of way. Such beliefs fail to be non-lucky, safe, sensitive, reliable, based on good evidence, etc. Hence the argument can work while making much weaker presumptions about what is required for knowledge than virtue responsibilism: A Convincing Argument for Skepticism (CA1) Knowledge is true belief formed via the exercise of intellectual virtues, or formed for the right kinds for reasons, or that is non-luckily true.13 (CA2) Empirical fact: frequently when we form beliefs about people’s qualities and competences, our beliefs are not formed via the exercise of epistemic virtues, are not formed for the right kinds for reasons, and are not non-luckily true. (CA3) Therefore, many of our beliefs about other people’s qualities and competences fail to be knowledge. Rejecting the first premise of this argument just doesn’t seem to be a live option. Any plausible account of knowledge is going to be committed to (CA1). So, unlike the skeptical argument presented by the situationist challenge, there is no room to escape the conclusion by rejecting a controversial theoretical claim. Nor can the second premise be rejected. The phenomenon of implicit bias is well documented, robust, and stable. Moreover, there is no way to make these empirical results compatible with the claim that the conditions for knowledge (as outlined in the first premise) are met.

202  Kevin Wallbridge This leaves no option in this case but to accept the skeptical conclusion that many of our beliefs about other people’s qualities and competences (that we take to be knowledge) are in fact not knowledge. Of course it may be initially surprising to learn that a not-inconsiderable swathe of our beliefs fail to be knowledge. But this surprise simply tracks the initial surprise that can come from finding out that we are prone to these implicit biases. I believe that this conclusion should not be controversial. I take it to be highly intuitive that there is a lack of knowledge in cases of implicit bias, and all major views of the nature of knowledge imply that there isn’t knowledge in such cases. Therefore I take it that we have succeeded in finding a genuinely convincing skeptical argument. Although this argument is of limited scope (since it only applies to our beliefs about other people’s qualities and competences), I want to be sure not to understate the extent of implicit bias and so the skepticism implied by (CA1)–(CA3). We are basically all of us, basically all of the time, making judgements about people that are shaped by irrelevant facts about their gender, race, religious group, sexuality, institutional affiliation, etc.14 Implicit bias is also reflexive: so you are not even a good judge of your own qualities or competences, despite the privileged position that you have to observe yourself. So the degree of skepticism implied by (CA1)–(CA3) is substantial, even if it falls short (for the moment) of the kind of skepticism implied by traditional skeptical arguments.

5 Expanding the Skeptical Argument Although a convincing argument, the skeptical scope of (CA1)–(CA3) is limited to beliefs about people’s qualities and competences. Is there a way to expand on this argument so that it implies skepticism about a wider range of our beliefs? One way to do so is to broaden the scope of the second premise by considering a wider range of biases, concerning our reasoning about a wider range of subjects. It is relatively straightforward to see how expanding the argument in this way would work: one simply adds to the range of beliefs specified in the argument by appealing to further biases. So readers may add in whatever biases they consider there to be good evidence for. (Hannon’s contribution to this volume discusses some such biases, and indeed, he makes a closely related skeptical argument. Carter and McKenna (2020) also note the import of ‘motivated reasoning’ here.) Rather than trying to enumerate all of the possible kinds of bias that one might appeal to here however, I will focus on looking more closely at what skeptical results follow from identity-based implicit bias in particular.

Situationism, Implicit Bias, Skepticism  203 5.1 Perceptual Knowledge One easy way to expand the argument while still sticking with the case of implicit bias is to note that implicit bias itself has a wider reach than we have so far discussed. In particular, it extends to cases of perception. ‘Shooter bias’ experiments have demonstrated that an ambiguous object is more often perceived as a gun when in the hands of young black men, and more often as something innocuous in the hands of young white men.15 Implicit bias does not just lead to our making bad judgements about people’s qualities and competences then, it also undermines some of our visual perceptual knowledge of ‘medium sized dry goods’ (something that we would otherwise take to be a paradigm case of perceptual knowledge). Further studies have also shown that implicit bias affects other sense modalities, with the existence of auditory discrimination against women auditioning for orchestras. Female musicians were consistently judged to be less skilled than their male counterparts—an effect that vanishes with the use of blind auditions.16 This kind of case highlights the extent of implicit bias, since these cases involve experts being implicitly biased even when aiming to make careful and considered judgements in their very field of expertise. Furthermore, it could be that there is an even more skeptical conclusion to be drawn here (although I am not aware of sufficient evidence for this at the moment). It might be that in shooter bias experiments subjects are not just mistaken about whether or not the object in question is a gun, but mistaken about the nature of their visual perception: they believe that they have a visual experience in which it looks like there was a gun, whereas in fact they did not have a visual experience like that. (And likewise, for the orchestral interviewers mentioned, it could be that they are mistaken about the nature of their own auditory experiences.) If this turns out to be the case, then the skeptical argument from implicit bias would in one way go further than ordinary skeptical arguments, which do not typically question our knowledge of the nature of our own phenomenal experience. 5.2 A Priori and Philosophical Knowledge Perhaps especially worrying to readers is the thought (pressed by Saul 2013) that philosophers beliefs are not secure from implicit bias in their own realm of expertise, namely the assessment of reasons and arguments. The effects of prestige bias and other biases on publication suggest that our beliefs about whether an argument is good, or whether certain reasons are good, or whether they provide adequate support or evidence for a conclusion, are also influenced by social markers which are irrelevant to the argument at hand.

204  Kevin Wallbridge So even as experts in these matters, philosophers’ beliefs about whether an argument justifies its conclusion will often fail to be knowledge. And notably these kinds of a priori facts are traditionally thought to be especially epistemically secure, and therefore this is another way in which a skeptical argument from implicit bias goes further than traditional skeptical arguments, which typically only challenge our a posteriori knowledge of the external world. 5.3 Testimonial Knowledge In a traditional skeptical argument the skeptic first gets one to agree to some seemingly minor concession (that I don’t know that I am not dreaming) but then shows how giving up this much commits one to giving up much more (that I don’t know anything about the external world). I plan to do something similar. Having argued for the concession that a large proportion of our beliefs about other people’s qualities and competences fail to be knowledge, I will now try to show how this concession leads to a much more comprehensive skepticism, undermining a great deal of our testimonial knowledge. We rely on the testimony of others to a huge extent in finding out about the world. Crucially, this means of acquiring knowledge depends on our placing epistemic trust in other people. And I take that this means that it depends on our being able to make judgements about whom to trust and whom not, where that trust applies to both someone’s sincerity (are they lying) as well as their competence (do they know what they are talking about). This is at least a highly plausible claim about how testimonial knowledge works for typical adult humans. This is not the place to debate reductionism vs. anti-reductionism in the epistemology of testimony, so let me just state that I am relying on something along the lines of Fricker’s claim that “a hearer should always engage in some assessment of the speaker for trustworthiness. To believe what is asserted without doing so is to believe blindly, uncritically. This is gullibility” (1994, p. 145). The following argument can be considered conditional on this claim. An Argument for Testimonial Skepticism (TS1) Testimonial knowledge requires the ability to aptly judge whom to trust. (TS2) We are biased judges of competence and this comprises our ability to make apt judgements about whom to trust: we often do not know whom to trust, and to what degree, even when we think we do. (TS3) Therefore, our testimonial beliefs often fail to be knowledge. If we trust a testifier in virtue of features which are not genuine markers of their reliability as a source of information, such as their whiteness or

Situationism, Implicit Bias, Skepticism  205 their maleness, then we fail in our duty to assess a speaker for competence, and as a result we fail to know what we believe on the basis of their testimony. Or if we ignore or downgrade the contradictory testimony of others on the basis of irrelevant features such as their gender or race, then we have ignored valid counterevidence, and this may also prevent a belief from being knowledge. And there is ample evidence that these are both things that we often do. And when we do this we lack knowledge, because what we believe has been swayed by biases that track features that make no relevant difference. (Although, not all of our beliefs need to fall into this category, only those for which epistemic bias has actually affected what we believe.)17 In other words, beliefs that we have formed on the basis of testimony very often fail to be knowledge (even when we think they are).

6 Conclusion Traditional skeptical arguments are a priori and deny the very possibility of knowledge; however these arguments aren’t convincing because their premises seem less certain than the claim that we can and do know everyday propositions (like I have hands). A skeptical argument based on the situationist challenge to virtue epistemology is similarly unconvincing. In both of these cases, the argument seems to serve best as a reductio for some of the premises. In contrast to these arguments, I have presented a skeptical argument based on the empirical fact that our judgements are often subject to identity-based implicit bias (although other biases may also be added to expand the scope of the argument). Like the situationist argument, it relies on an empirical premise and it concludes that (as a contingent matter of fact) we lack much of the knowledge we take for granted. However, unlike that argument, this one relies only on very uncontentious claims about what exactly is required for knowledge (and empirical results that are also well verified). I have argued that this kind of skeptical argument is genuinely convincing: we should embrace its conclusion rather than reject its premises. Although it might appear that the trade-off for getting a genuinely convincing argument is that the scope of the ensuing skepticism is quite limited compared to the global skepticism implied by traditional skeptical arguments, this argument does actually imply quite a wide ranging skepticism. This argument implies that much of the knowledge of others characters and competences, some of the perceptual knowledge, some of the reasoning-based knowledge, lots of the testimonial knowledge, and potentially even some of the knowledge of the phenomenal character of

206  Kevin Wallbridge our own experiences that we take ourselves to have, all turns out not to really be known. (And that is just appealing to identity-based implicit bias but not to other cognitive biases that we could expand the argument to include.) So although the degree of skepticism implied is less all-encompassing than traditional external world skepticism, it is still highly disruptive. And this skeptical argument has a major advantage over the traditional kind: rather than motivating the rejection of some prior theoretical commitment, this skeptical argument is genuinely convincing, and in that respect it is a great deal more skeptical.

Notes 1 See Climenhaga, Kyriacou, and Stoutenburg in Part II of this volume. 2 Which is not to deny that there are other long standing arguments, such as Pyrrho’s, that do not take this form. 3 Because of the truth of something like Universal deductive closure: ‘If one knows some premises and competently deduces Q from those premises, thereby coming to believe Q, while retaining one’s knowledge of those premises throughout, one comes to know that Q’ (Hawthorne 2004). 4 This is my own view, as well as that of Nozick (1981), Dretske (2005), and Adams, Barker, and Figurelli (2011). 5 Alfano (2012, 2014). 6 Latané and Nida (1981), Schwartz, and Gottlieb (1980), Baron (1997), Baron and Thomley (1994), Boles and Hayward (1978), and Donnerstein and Wilson (1976). 7 Hannon, elsewhere in this volume, also notes the role that implicit bias can play in motivating skepticism. 8 For a concise overview of just some of this, see Jost et al. (2009). 9 There is a burgeoning literature on these CV studies. For some recent CV studies however, see Bertrand and Mullainathan (2003) and Carlsson and Rooth (2007). 10 Jennifer Saul (2013) also considers a skeptical threat raised by the discovery of implicit bias. Her route to skepticism is different to my own, however. She appeals to our awareness of bias and the rational doubt this generates, as compared to the kind of skeptical worries raised by Cartesian doubts that I may be a brain in a vat. My route does not go via the idea that having discovered implicit bias we are faced with a doubt which we cannot dispel. My argument is that just by being biased we fail to meet conditions on knowledge and that this is true even if one is unaware of implicit bias and can continue to believe as one does free from any doubts. 11 Similar but weaker versions of this argument are available if one prefers, e.g. ‘Probably I am one of the people who frequently form beliefs… therefore probably…).’ But I will stick to the straightforward version of this and other arguments throughout. 12 Even in cases where it appears that implicit bias has been reduced or eradicated, things are not always as rosy as they seem. Sinclair, Lowery, Hardin and Colangelo (2005) show how results on implicit racial bias tests can be reduced by having the experimenter wear a t-shirt conveying an anti-racist message/slogan. However, this effect depends on the subject finding the

Situationism, Implicit Bias, Skepticism  207

13 14 15 16 17

experimenter likable, i.e. the effect appears to be the result of the subject simply aping the beliefs of people they like. So in this case, even though the outward effects of implicit bias are reduced, it does not appear to be due to intellectual virtue on the part of the subject, so much as to the effect of a competing kind of intellectual vice. Or insert whatever other specific thing your preferred account of knowledge requires, e.g. ‘on the basis of good evidence or that one would not have believed were it false.’ Webb, Sheeran, and Pepper (2012), Olson and Fazio (2004), Amodio and Devine (2006), and Peters and Ceci (1982). For studies on shooter bias, see Correll et al. (2007) and Plant and Peruche (2005). For some work on the epistemological ramifications of this kind of effect, see Siegel (2013). Goldin and Rouse (1997). The problem may be more pronounced for credences than beliefs, since it is plausible that bias impacts the degree of belief one ends up holding in just about every case of testimony.

Bibliography Adams, F., Barker, J. A., & Figurelli, J. (2011). Towards closure on closure. Synthese, 188(2), 179–196. Alfano, M. (2012). Expanding the situationist challenge to responsibilist virtue epistemology. Philosophical Quarterly, 62(247), 223–249. Alfano, M. (2014). Extending the situationist challenge to reliabilism about inference. In A. Fairweather & O. Flanagan (Eds.), Virtue Epistemology Naturalized: Bridges between Virtue Epistemology and Philosophy of Science (pp. 103–122). Synthese Library. Amodio, D. M., & Devine, P. G. (2006). Stereotyping and evaluation in implicit race bias: Evidence for independent constructs and unique effects on behavior. Journal of Personality and Social Psychology, 91(4), 652–661. Baron, R. A. (1997). The sweet smell of… helping: Effects of pleasant ambient fragrance on prosocial behavior in shopping malls. Personality and Social Psychology Bulletin, 23(5), 498–503. Baron, R. A., & Thomley, J. (1994). A whiff of reality positive affect as a potential mediator of the effects of pleasant fragrances on task performance and helping. Environment and Behavior, 26(6), 766–784. Bertrand, M., & Mullainathan, S. (2003). Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination (Working Paper No. 9873). National Bureau of Economic Research. Boles, W. E., & Hayward, S. C. (1978). Effects of urban noise and sidewalk density upon pedestrian cooperation and tempo. The Journal of Social Psychology, 104(1), 29–35. Carlsson, M., & Rooth, D.-O. (2007). Evidence of ethnic discrimination in the Swedish labor market using experimental data. Labour Economics, 14(4), 716–729. Carter, A., & McKenna, R. (2020). Skepticism motivated: On the skeptical import of motivated reasoning. Canadian Journal of Philosophy, 6, 1–17.

208  Kevin Wallbridge Carter, A., & Pritchard, D. (2017). Cognitive bias, scepticism and understanding. In S. R. Grimm, C. Baumberger, & S. Ammon (Eds.), Explaining Understanding: New Perspectives from Epistemology and Philosophy of Science (pp. 272–292). Routledge. Correll, J., Park, B., Judd, C. M., Wittenbrink, B., Sadler, M. S., & Keesee, T. (2007). Across the thin blue line: Police officers and racial bias in the decision to shoot. Journal of Personality and Social Psychology, 92(6), 1006–1023. Donnerstein, E., & Wilson, D. W. (1976). Effects of noise and perceived control on ongoing and subsequent aggressive behavior. Journal of Personality and Social Psychology, 34(5), 774–781. Doris, J. M. (2002). Lack of Character: Personality and Moral Behavior. Cambridge University Press. Dretske, F. (2005). Is knowledge closed under known entailment? The case against closure. In M. Steup & E. Sosa (Eds.), Contemporary Debates in Epistemology (pp. 13–26). Blackwell. Fricker, Elizabeth (1994). Against Gullibility. In A. Chakrabarti & B. K. Matilal (Eds.), Knowing from Words. (pp. 125–161) Kluwer Academic Publishers. Goldin, C., & Rouse, C. (1997). Orchestrating Impartiality: The Impact of (Working Paper No. 5903). National Bureau of Economic Research. Hawthorne, J. (2004). Knowledge and Lotteries. Oxford University Press. Jost, J. T, Rudman, L. A., Blair, I. V., Carney, D. R., Dasgupta, N., Glaser, J., & Hardin, C. D. (2009). The existence of implicit bias is beyond reasonable doubt: A refutation of ideological and methodological objections and executive summary of ten studies that no manager should ignore. Research in Organizational Behavior, 29, 39–69. King, N. L. (2014). Responsibilist virtue epistemology: A reply to the situationist challenge. Philosophical Quarterly, 64(255), 243–253. Latané, B., & Nida, S. (1981). Ten years of research on group size and helping. Psychological Bulletin, 89(2), 308–324. Moore, G. E. (1939). Proof of an external world. Proceedings of the British Academy, 25(5), 273–300. Nozick, R. (1981). Philosophical Explanations. Harvard University Press. Olin, L., & Doris, J. M. (2014). Vicious minds. Philosophical Studies, 168(3), 665–692. Olson, M. A., & Fazio, R. H. (2004). Reducing the influence of extrapersonal associations on the implicit association test: Personalizing the IAT. Journal of Personality and Social Psychology, 86(5), 653–667. Peters, D. P., & Ceci, S. J. (1982). Peer-review practices of psychological journals: The fate of published articles, submitted again. Behavioral and Brain Sciences, 5(2), 187–195. Plant, E. A., & Peruche, B. M. (2005). The consequences of race for police officers’ responses to criminal suspects. Psychological Science, 16(3), 180–183. Pritchard, D. (2014). Re-evaluating the situationist challenge to virtue epistemology.In A. Fairweather, & O. Flanagan (Eds.), Naturalizing Epistemic Virtue. (pp. 143–154). Cambridge University Press. Saul, J. (2013). Scepticism and implicit bias. Disputatio, 5(37): 243–263. Schwartz, S. H., & Gottlieb, A. (1980). Bystander anonymity and reactions to emergencies. Journal of Personality and Social Psychology, 39(3), 418–430.

Situationism, Implicit Bias, Skepticism  209 Siegel, S. (2013). The epistemic impact of the etiology of experience. Philosophical Studies, 162(3), 697–722. Sinclair, S., Lowery, B. S., Hardin, C. D., & Colangelo, A. (2005). Social tuning of automatic racial attitudes: The role of affiliative motivation. Journal of Personality and Social Psychology, 89(4), 583–592. Webb, T. L., Sheeran, P., & Pepper, J. (2012). Gaining control over responses to implicit attitude tests: Implementation intentions engender fast responses on attitude-incongruent trials. British Journal of Social Psychology, 51(1), 13–32. Zagzebski, L. (1996). Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical Foundations of Knowledge. Cambridge University Press.

Part IV

Wittgensteinian Anti-Skepticism

11 “I Know”, “I know”, “I know”. Hinge Epistemology, Invariantism, and Skepticism1 Annalisa Coliva 1 Wittgenstein’s Contextualism In Coliva (2010), I have proposed an interpretation of Wittgenstein’s criticism of Moore’s use of “I know” in relation to his truisms and the premises of his proof, as well as a defense of Wittgenstein’s position against John Searle’s (1969) allegation that his remarks were based on an “assertion fallacy”, which rely on the decisive role of use in the determination of meaning. Contrary to Moore and Searle, Wittgenstein did not think that different occurrences of “I know” would always have the same meaning or, indeed, that they would always have a meaning. That is, the same words, by occurring in different contexts of use would change their meaning for him, up to the point of losing it altogether, in some deranged circumstances. By contrast, for Moore and Searle they would always retain the same meaning and could at most be used to make assertions that, depending on the case at hand, would be felicitous or infelicitous. Wittgenstein was therefore a contextualist of sorts. Yet, his form of contextualism was utterly different from contemporary variants of it. Indeed, as we will see, by contemporary lights, his position would come close to invariantism. Still, before venturing into that issue, let us see in more depth what his contextualism amounted to. 1.1 The Ordinary Use of “I know” As is well known, Wittgenstein accorded pride of place to ordinary language. By observing how we normally use “I know”, he identified several criteria that are operative in the ordinary (or empirical) use of “I know”, henceforth “I Know” with a capital ‘K’. According to him, only by conforming to these criteria would knowledge claims make sense, in the literal sense of having a meaning. Here they are: i ii

knowledge claims make sense only when they are based on reasons, and, in fact, on reasons that are either non-circular, 2 or at least stronger than what they are supposed to ground;

214

Annalisa Coliva

iii they make sense only when it is possible to make an inquiry to confirm that things are as one claims to know and yet the possibility that things may not be so is open; iv it is therefore possible to say “I don’t know”; v finally and surprisingly, claims to knowledge make sense only when they are relevant within the communicative exchange. To substantiate my interpretation, consider the following passages, in support of (i) and (ii):3 Someone with bad sight asks me: “do you believe that the thing we can see there is a tree?” I reply “I know it is; I can see it clearly and am familiar with it”.—A: “Is N. N. at home?”—I: “I believe he is.”—A: “Was he at home yesterday?”—I: “Yesterday he was—I know he was; I spoke to him.”—A: “Do you know or only believe that this part of the house was built on later than the rest?” —I: “I know it is; I got it from so and so”. (OC 483) In these cases, then, one says “I know” and mentions how one knows [der Grund], or at least one can do so. (OC 484) One says “I know” when one is ready to give compelling grounds. (…) But if what he believes is of such a kind that the grounds that he can give are no surer than his assertion, then he cannot say that he knows what he believes. (OC 243) (…) If I say “I know that I have two hands” (…) I must be able to satisfy myself that I am right. But I can’t do that, for my having two hands is not less certain before I have looked at them than afterwards (…). (OC 245) Here are some other passages that support (iii): Only in certain cases it is possible to make an investigation “is that really a hand?” (or “my hand”). For “I doubt whether that is really my (or a) hand” makes no sense without some more precise determination. One cannot tell from these words alone whether any doubt at all is meant—nor what kind of doubt. (OC 372)

“I Know”, “I know”, “I know”  215 If I don’t know whether someone has two hands (say, whether they have been amputated or not) I shall believe his assurance that he has two hands, if he is trustworthy. And if he says he knows it, that can only signify to me that he has been able to make sure, and hence that his arms are e.g. not still concealed by coverings and bandages, etc. etc. My believing the trustworthy man stems from my admitting that it is possible for him to make sure (…). (OC 23) Some well-known passages in OC speak in favor of (iv): “I know that I am a human being.” In order to see how unclear the sense of this proposition is, consider its negation. At most it might be taken to mean “I know I have the organs of a human”. (E.g. a brain which, after all, no one has ever yet seen.) But what about such a proposition as “I know I have a brain”? Can I doubt it? Grounds for doubt are lacking! Everything speaks in its favour, nothing against it.4 (OC 4) Finally, the following passages back (v): My difficulty can also be shewn like this: I am sitting talking to a friend. Suddenly he says: “I knew all along that you were soand-so.” Is that really just a superfluous, though true, remark? I feel as if these words were like “Good morning” said to someone in the middle of a conversation. (OC 464) Thus it seems to me that I have known something the whole time, and yet there is no meaning (my emphasis) in saying so, in uttering this truth. (OC 466) (…) Shouldn’t I be at liberty to assume that he doesn’t know what he is saying, if he is insane enough to want to give me this information? (OC 468) (…) I am astonished; but later I realize that these words connect up with his thoughts about me. And now they don’t strike me as meaningless (emphasis mine) anymore. (OC 469)

216  Annalisa Coliva 1.2 The Grammatical Use of “I know” Wittgenstein, however, recognized also a “grammatical” use of “I know”, henceforth “I know” with an italicized ‘k’. Contrary to “I Know”, and as surprising as it may seem, for Wittgenstein “I know” does not express the obtaining of an epistemic relation between a subject and a proposition, it isn’t based on grounds, and doesn’t express knowledge but objective certainty.5 Writes Wittgenstein: If “I know etc” is conceived as a grammatical proposition, of course the “I” cannot be important. And it properly means “There is no such thing as a doubt in this case” or “The expression ‘I do not know’ makes no sense in this case”. And of course it follows from this that “I know” makes no sense either. (OC58) “I know” is here a logical insight. Only realism can’t be proved by means of it. (OC 59) I know that this is my foot. I could not accept any experience as proof to the contrary.—That may be an exclamation; but what follows from it? At least that I shall act with a certainty that knows no doubt, in accordance with my belief. (OC 360) I should like to say: Moore does not know what he asserts he knows, but it stands fast for him, as also for me; regarding it as absolutely solid is part of our method of doubt and inquiry. (OC 151) Hence, to be more perspicuous, uses of “I know” could be replaced by “Here I can’t be wrong” or “Here a mistake/doubt is logically impossible”,6 or, even, “I couldn’t admit any experience as proof to the contrary”. Now, according to Wittgenstein, the fact that one can’t be wrong in these cases doesn’t depend on one’s privileged epistemic position, but on the role that the propositions one claims (grammatically) to know play in our language games.7 These propositions – like “Here is my hand”, “There are physical objects”, “I am a human being”, “The earth has existed for a very long time”, etc. – have a normative function, even when they are the content of a judgment and not the explicit statement of a rule. They determine what needs to stay put to be allowed to take various kinds of empirical evidence as evidence in favor of ordinary empirical propositions. For instance, only by taking for granted that there are physical objects can one take one’s sensory experience at face value as evidence in favor of “Here is a computer in front of me”. Analogously,

“I Know”, “I know”, “I know”  217 it is only by taking for granted that here is my hand, that I can take my sensory evidence as being produced by the reliable workings of my sensory system. If I doubted of the former, I should also doubt of the latter, which is presenting to me that portion of the world as occupied by such an object. Similarly, it is only by taking for granted that the earth has existed for a very long time that I can take a fossil as evidence relevant to the determination of the specific age of the earth. If I didn’t take that for granted, then that fossil by itself would not play any evidential role, since its existence would then become compatible with the hypothesis that the world had just popped into existence a few minutes ago, replete with everything we find in it (including ourselves, our memories, etc.). Since these propositions need to stay put in order for the door to turn, that is, in order for our epistemic practices to function as they do, Wittgenstein famously likens them to “hinges”. He writes: That is to say, the questions that we raise and our doubts depend on the fact that some propositions are exempt from doubt, are as it were like hinges on which those turn. (OC 341) That is to say, it belongs to the logic of our scientific investigations that certain things are in deed not doubted. (OC 342) But it isn’t that the situation is like this: We just can’t investigate everything, and for that reason we are forced to rest content with assumption. If I want the door to turn, the hinges must stay put. (OC 343) In context, however, hinges can also be used as explicit formulations of rules, e.g. when we use “This is a hand” as an ostensive definition of the meaning of the word “hand” in English; or “Here is a human being” said while pointing at myself to teach my three year-old kid the meaning of “human being”; or “This is a physical object”, said while pointing at a pen I hold in my hand, to explain to her the meaning of that expression. Here are some of Wittgenstein’s own examples: “A is a physical object” is a piece of instruction which we give only to someone who doesn’t yet understand either what “A” means, or what “physical object” means. Thus it is an instruction about the use of words, and “physical object” is a logical [categorial] concept. (Like colour, quantity,…) And that is why no such proposition as: “There are physical objects” can be formulated. Yet we encounter such unsuccessful shots at every turn. (OC 36)

218  Annalisa Coliva If I wanted to doubt whether this was my hand, how could I avoid doubting whether the word “hand” has any meaning? So that is something I seem to know after all. (OC 369) But more correctly: The fact that I use the word “hand” and all the other words in my sentence without a second thought, indeed that I should stand before the abyss if I wanted so much as to try doubting their meanings – shows that absence of doubt belongs to the essence of the language-game, that the question “How do I know…” drags out the language-game, or else does away with it. (OC 370) Hinges, that is, contribute to the determination of meaning in two ways. First, when we use paradigms to ostensively define the meaning of words. For the existence of the object must be taken for granted for this kind of definition to take place. Second, meaning is established not only by definitions, for Wittgenstein, but also by agreement in judgment. Thus, some of our judgements, such as “Here is my hand” (in Moore-like circumstances), must be taken for granted, in order for “hand” to mean what it does. That is, supposing we had ostensively defined “hand” as usual, in order for that word to have its ordinary meaning, its canonical applications should be agreed upon. That is why, we, as speakers of English, cannot sensibly revoke into doubt the judgement “Here is my hand”, made in a Moore-like context. If we did, it would no longer be clear that by “hand” we mean hand or anything else at all. Hinges, that is, play a normative role, while also being judgments, since they constitutively contribute to the determination both of meaning and of what would count as, for instance, normal conditions of perception, evidence for or against historical or geological specific judgments, normal conditions of human functioning, etc. For if those judgments were given up, one could no longer count on one’s perception, memories, or on apparent testimonies regarding the age of the Earth, to acquire evidence which could, in its turn, disprove those very propositions.8 Finally, for Wittgenstein, the way in which we acquire hinges depends on what we have been drilled and taught to say (or think), either implicitly or explicitly, in the process of acquiring language, and of learning to take part in our various epistemic practices. That is why “I know”, in connection with them, can also be glossed as “I have been drilled /taught to say/think thus-and-so”. Such an explanation, however, plays a causal role and it isn’t an epistemic reason which could sustain a genuine claim to knowledge.

“I Know”, “I know”, “I know”  219 1.3 The ‘Philosophical’ Use of “I know” In the ‘philosophical’ use of “I know”, henceforth “I know” with an underlined ‘k’, the grammatical function is conflated with the empirical one, according to Wittgenstein. Like “I Know”, “I know” is taken to express the obtaining of an epistemic relationship between a subject and a proposition or a fact. Yet, like “I know”, “I know” is taken to be indubitable. That is, “I know” straddles the ordinary and the grammatical use of “I know”, and it produces nonsense, for Wittgenstein, because no genuinely epistemic relation between a subject and a proposition (or a fact) is sure-fire, for him. Thus, when a philosopher like Moore says “I know here is my hand”, he is using “I know” in its ‘philosophical’ sense and is actually producing nonsense. This is not typically evident to philosophers, who, on the contrary, work on the assumption of semantic uniformity. That is, philosophers typically do not notice the difference between “I Know”, “I know”, and “I know”. Since they do not notice the difference, they are then confronted with the false problem of having to explain why one cannot go wrong epistemically speaking with respect to, for instance, there being one’s hand here. Since, in ordinary contexts that is certainly possible (in non-Moorelike situations), the skeptical thought that we might be wrong anyway, even now that we are seeing our hand in optimal environmental and cognitive conditions, becomes salient. Yet, for Wittgenstein, that doubt too would be nonsensical, since it would straddle contexts. That is, it would depend on noticing that since in some contexts we might have gone wrong in our judgement “Here is my hand”, for instance after a car accident, we could be wrong even now, when in fact we are not in that kind of context.9 Another way of putting the same point is to say that philosophers tend to project the meaning of “I Know”, to other uses of “I know”, which do not conform to those criteria. By running together contexts and criteria of use, they are in the grip of the illusion of producing sense and of being confronted with genuinely philosophical problems, when in fact they have just been deluded into the appearance of a problem brought about by a conceptual confusion, which ultimately resides in not seeing clearly the functioning of our language. This kind of typically philosophical attitude needs therapy, according to Wittgenstein, but the therapy does not consist in condoning the use of “I know” in that connection – e.g. “I know there is my hand here” (in a Moore-like context), while simply adding the proviso that one does not mean to be doing any philosophy by speaking that way.10 Rather, the therapy consists in unraveling a particular way in which “I know” is sometimes used – that is, to express the impossibility of a doubt (“I know”) – and by providing a philosophical account of why, in such a case, a doubt and a mistake are impossible. To repeat: not because we are in a

220  Annalisa Coliva privileged epistemic position, but because of the role those propositions play, at least in context, for all of us.11 Thus, according to Wittgenstein, the solution of many philosophical puzzles or paradoxes, including skepticism about the external world, about other minds, and the possibility of privileged self-knowledge, depends on developing a peculiar “philosophical” ear, which is capable of sensing the wrong note struck by the traditional philosopher. Furthermore, it requires a kind of “absolute ear” capable of noticing the differences between various uses of “I know”. Indeed, Wittgenstein thought of “I’ll teach you differences” from Shakespeare’s King Lear, as a possible motto for the Philosophical Investigations. The differences he was alluding to were those that conceal themselves under identical words and expressions (“I know”, “is”, and many more (cf. PI I, 116)). But you will also need to provide an explanation – in the sense of an elucidation, if you will – which does not fall into the same traps as the ones offered by traditional philosophers, as to why, in our leading example – “I know there is my hand here” – a doubt or a mistake are impossible. To conclude, as is apparent, Wittgenstein’s contextualism is radical for at least the following reasons. First, because it attributes different meanings (or indeed a lack of meaning) to different uses of “I know”. Thus it differs from contemporary forms of contextualism, where only the ordinary use of “I know” is recognized and a difference in standards is supposed to affect merely the semantic evaluation of occurrences of “I Know” or of “S Knows” in different contexts (of use or of assessment, depending on the form of contextualism at issue).12 Second, because his contextualism is meant to show that skepticism is actually nonsensical, rather than a genuine context with standards for knowledge that are impossible for us to satisfy. Let us dwell a bit more on Wittgenstein’s reasons to deem philosophical doubts nonsensical.

2 The Nonsense of Philosophical Skepticism There are two strands to Wittgenstein’s argument in On Certainty, targeting two different forms of skepticism – Cartesian and Humean skepticism – respectively. Against the former, Wittgenstein’s views are quite bold. For he claims that if, in a dream, we uttered or entertained the words “I may be dreaming”, those words would have no meaning. The argument “I may be dreaming” is senseless for this reason: if I am dreaming, this remark is being dreamed as well – and indeed it is also being dreamed that these words have any meaning. (OC 383) “But even if in such cases I can’t be mistaken, isn’t it possible that I am drugged?” If I am and if the drug has taken away my consciousness,

“I Know”, “I know”, “I know”  221 then I am not now really talking and thinking. I cannot seriously suppose that I am at this moment dreaming. Someone who, dreaming, says “I am dreaming”, even if he speaks audibly in doing so, is no more right than if he said in his dream “it is raining”, while it was in fact raining. Even if his dream were actually connected with the noise of the rain. (OC 676) Hence, for those words to have a meaning at all, they must be occurring outside a dream and, in their ordinary and only meaningful use, they could never be used to describe a possible state of affairs. Rather, they would be used to express something else, like, for instance, one’s surprise with respect to an unusual circumstance. Yet, if those words cannot be used to make a true assertion, their negation, for Wittgenstein, could not be used to that effect either. Thus, from the fact that there cannot be a genuinely descriptive use of “I am dreaming”, it does not follow that we can meaningfully assert the opposite. Rather, either we are using that phrase to emphasize that our experience, however unusual, is veridical; or else, we are hitting against a hinge-like use of it. The second strand to Wittgenstein’s argument is not targeting Cartesian skepticism, but Humean skepticism. The key aspect of this latter form of skepticism is that it demands an epistemic justification for our hinges, to acknowledge their being rationally held. Yet, for Wittgenstein, neither “Here is my hand” (in Moore-like circumstances), nor “There is an external world”, or “There are physical objects”, or any other presupposition of rational inquiry can actually be sustained by reasons. For it is only by taking them for granted that we could produce anything like epistemic reasons for or against ordinary empirical propositions. To repeat, I have to take it for granted that there is my hand here (in Moore-like circumstances), and that therefore there are physical objects, and that my senses are working reliably, to be within my rights in taking my sensory experience as justifying my belief that there is no glass of water on my desk at the moment; as well as to take that experience as a guide in my subsequent empirical inquiry aimed at determining where I might have left it. Yet, the fact that hinges lie beyond justification does not mean that they can be doubted. For doubts too should be based on reasons, but nothing we regard as a reason actually speaks against these hinges, nor could. Thus, their lying beyond justification makes them immune to sensible doubt as well.13

3 Some Comparisons 3.1 Therapeutic Readings As the connoisseurs will have noticed, I have come close to so-called “therapeutic” interpretations of Wittgenstein’s thought,14 according

222 Annalisa Coliva to which Moore with his use of “I know”, and more generally philosophers, often tend to fall prey to an illusion of meaning. For they think they are making sense, when they employ certain words, and that they are therefore discovering deep philosophical truths (or problems), while they are in fact speaking nonsense. Yet, I think therapeutic interpretations of Wittgenstein’s thought are too partial. For they fail to see that while use comes first and determines meaning, it does so, for Wittgenstein, because it fixes the rules for the correct employment of signs and for the actual deployment of our epistemic practices. Thus, there is no opposition, in my view, between stressing the importance and primacy of use in the later Wittgenstein and his insistence on rules and grammar. When philosophers speak nonsense, they do so because they run against the rules of language, as well as of evidential significance, established by use and by our actual language games and epistemic practices.15 Furthermore, in my view, therapists tend to discard the important insights, contained in On Certainty, on the grammatical role of “I know”, which clarify the normative status of our certainties, viz. of those “hinge” propositions with respect to which “I know” may be used grammatically. 3.2 Travis: Wittgenstein as a Contemporary Contextualist ante litteram Semantic contextualist readings of On Certainty and of Wittgenstein’s observations on Moore’s use of “I know”, aim to interpret them as examples of contemporary contextualism ante litteram. Supporters of such a reading join therapists and myself in stressing the fact that Wittgenstein’s insistence on use has deep philosophical consequences. In particular, it entails that most philosophical employments of “I know” would actually fail to make sense (Travis 1989, 153). This, however, may suggest that there are more similarities between these “contemporary” contextualist readings of Wittgenstein and mine, than it is in fact the case. Charles Travis, in his The Uses of Sense (1989), has argued that Wittgenstein held that “A knows that p” can express many different thoughts – viz. various truth-conditions – on different contexts of its use. However, he bases his interpretation on a passage where Wittgenstein is actually discussing a different example, i.e. “I’m here”. Just as the words “I am here” have a meaning only in certain contexts, and not when I say them to someone who is sitting in front of me and sees me clearly, —and not because they are superfluous, but because their meaning is not determined by the situation, yet stands in need of such a determination. (OC 348)

“I Know”, “I know”, “I know”  223 So too, the comparison goes, do the words “I know that that’s a tree” uttered when one is clearly in view of a tree (cf. OC 347). Travis’ idea is that, unless we look at context, the truth-conditions of a given utterance of those words are underdetermined. Hence, it is only by considering the context that we can find out which thought they express and establish whether they are used to express a truth or a falsity. I find this suggestion odd for a number of reasons. First, I think Wittgenstein’s view is more radical than what Travis makes of it. For in the passage quoted above, Wittgenstein is saying that outside a specific context of use, those words have no meaning. Using a piece of classical semanticist’s terminology, I take that to entail that they don’t even have a role or a character. Therefore, they naturally fail to determine a thought, i.e. specific truth-conditions. Travis, on the contrary, seems to think that they have a meaning, on Wittgenstein’s view of the matter, while context has to be invoked in order for them to express a determinate thought, i.e. determinate truth-conditions. This, however, wouldn’t be very surprising, since in OC 348 indexicals such as ‘I’ and ‘here’ occur. It is indeed commonplace that context has to be invoked to determine what thought is expressed in each occurrence of their use. Yet, those words are typically considered to have an invariant meaning or character, or role, which, together with a context (formed by a subject, a place, and a time), manages to express a determinate thought (that is, determinate truth-conditions), on each occasion of utterance. Yet, clearly, Wittgenstein’s remark is meant to make a stronger and surely more contentious claim. Second and connectedly, the only kind of “contextualism” envisaged by Wittgenstein concerns the fact that in different contexts “I/A know(s) that p” could actually mean different things, in particular, either an epistemic relationship between the subject and “p”; or else, a grammatical remark where no epistemic property is attributed to a subject. Alternatively, those words would fail to mean anything at all, when they are used by philosophers such as Moore, irrespective of all criteria that govern their employment in the ordinary and in the grammatical case. Furthermore, it is to be stressed that only in the first case could those words express a Fregean thought, i.e. something susceptible of being either true or false. For in the second case, the role of the ascription would be that of expressing a rule, hence nothing which could either be true or false, according to Wittgenstein;16 and, in the third, those words would fail to have a meaning altogether and would then express no Fregean thought at all. Lastly, contrary to Travis (1989, 156–166), I find no suggestion in Wittgenstein that, depending on various factors, such as our practical interests and world-involving conditions, an utterance of “I Know” could actually express different truth-conditions. Let us illustrate this idea with a classic example. Consider a subject who needs to deposit a check in the bank by Friday. Now, depending on whether it is a matter

224  Annalisa Coliva of urgency or not and, therefore, on whether certain defeaters are salient to the case, an ascription of knowledge to her (either made by the subject herself or by a third party) regarding whether the bank is going to be open on Fridays, could be either true or false. However, as remarked, given the textual evidence at our disposal, there are no hints of this sort in On Certainty. For Wittgenstein doesn’t discuss such a kind of case and isn’t so much interested in knowledge per se as in certainty. Thus, if it makes sense to describe Wittgenstein’s position in these anachronistic terms, it turns out that he was more of an invariantist with respect to knowledge ascriptions than a contemporary contextualist. That is, while a contextualist in his own way, he was not leaning towards anything resembling contemporary contextualism about knowledge ascriptions. I will return to Wittgenstein’s invariantism in Section 4. 3.3 Williams Following Michael Williams (2004a, b), one might think that though a semantic invariantist with respect to knowledge ascriptions, Wittgenstein was after all an epistemological contextualist. For, allegedly, he pointed out that “Here is my hand” could be the object of a genuine claim to knowledge in some cases and could thus be supported by grounds, while it would fail to be so in different ones. On such an epistemological contextualist reading, in the latter case the relationship between the subject and the proposition would still be epistemic, though non-evidential. Hence, the subject would actually know that there is a hand in front of her, for she would be entitled – i.e. non-evidentially justified – to take that truth for granted, though she couldn’t articulate reasons in its favor such as to ground an eventual claim to knowledge. She would be non-evidentially justified because no reasons to the contrary could be produced and because to take that much for granted would be necessary to gather any evidence in favor of other propositions. Writes Williams: Holding some particular propositions fast need not be a matter of credulity: to hold them fast is reasonable. So while they are (in one sense) ungrounded, we are justified in cleaving to them. In this way, they can be the objects of beliefs that are true (in a deflationary sense of ‘true’) and justified (though not derived from evidence). Thinking of ‘justified true belief’ in this second way, even basic certainties can amount to knowledge. (2004b, 280) Yet, I find Williams’ suggestion misleading for several reasons. First, Wittgenstein was not at all in favor of non-evidential justifications for hinges. Consider the following passage, which, ironically, is often appealed to by supporters of such an epistemic reading of Wittgenstein’s

“I Know”, “I know”, “I know”  225 remarks, in which he is actually denying that we are non-evidentially justified, by way of pragmatic considerations, to believe them: But it isn’t (my emphasis) that the situation is like this: we just can’t investigate everything, and for that reason we are forced to rest content with assumption. If I want the door to turn, the hinges must (my emphasis) stay put. (OC 343) Wittgenstein’s point, that is, is not that the impossibility of an investigation into the presuppositions of each inquiry would make inquiry impossible and that we thereby have an un-earned justification. Rather, his point is that it belongs to the logic of inquiry – and it is therefore a grammatical remark about “inquiry” – that its hinges be held fast, even if they are not justified and justifiable, evidentially or otherwise.17 Second, Williams’ suggestion is misleading because, on Wittgenstein’s view, in a Moore-like case “Here is my hand” would be a “hinge” and hence a rule.18 If so, it would simply fail to be in the business for epistemic justification and assessment.19 As Wittgenstein puts it, these propositions lie “beyond being justified or unjustified” (OC 359, my emphasis). Thus, I think Wittgenstein didn’t even argue that the nature of justification varies depending on context. That is, he did not argue for the view that in some cases our justification is earned through empirical or a priori work, while sometimes it is unearned. Rather, he argued for the view that, in certain cases, propositions that, in different circumstances, may really be subject to verification, would fail to be so, as their status and role would be different in those different contexts. To repeat, in some contexts they wouldn’t be genuinely empirical propositions but rules and would thus be unsuitable for epistemic support, no matter how non-evidential that might be. In their connection, the use of “I know” could therefore sensibly be taken only as grammatical, according to Wittgenstein, and, as we have seen, such as to fail to attribute any epistemic property to the subject at all. 20 Finally, there isn’t anything that suggests that knowledge, for Wittgenstein, is to be understood somewhat in an externalist spirit and that it can be attained even if one is in no position to produce reasons for one’s own true beliefs. In fact, the idea that Wittgenstein would allow for the possibility of knowledge even when no reasons can be produced in favor of one’s claim is based on disregarding the distinction between ordinary and grammatical uses of “I know” he draws in On Certainty. As we saw (in Section 1.2), for Wittgenstein, only grammatical uses of “I know” are not backed by reasons. Furthermore, for him, they actually don’t express an epistemic relation between a subject and a proposition. Rather, they express a kind of certainty, which would be more perspicuously

226  Annalisa Coliva expressed by saying “Here a doubt is (logically) impossible”. Thus, we do not have anything in On Certainty resembling the externalist idea that, even if we cannot produce reasons in favor of knowledge claims regarding hinges, we would know them nonetheless. Rather, we have the suggestion that the use of “I know” in their connection is a misleading expression of the kind of certainty that characterizes them. Namely, a certainty that has nothing subjective or psychological about it, but which depends on the role hinges play in the relevant epistemic practices. Hence, all we get from these and other key passages in On Certainty is the idea that justification and knowledge do not take place in a vacuum. They always depend on there being certain hinges, which, as such, cannot themselves be justified or known, yet allow us to acquire evidence for or against ordinary empirical propositions. They are therefore constitutive of the practice that in turn determines what being epistemically rational amounts to. Some of them can change in time – “Nobody has ever been on the Moon” is no longer a hinge for us – and according to context – sometimes “Here is my hand” is an empirical proposition we subject to verification. Yet, there are several of them we simply cannot revise, e.g. “The Earth has existed for a very long time” and “There are physical objects”, on pain of giving up all our system of beliefs. This is not to say that they are metaphysically necessary but only that they play such a fundamental role in our Weltbild that, from within it – as we in fact are – we can’t actually find any reason to doubt them.

4 Wittgenstein’s Invariantism What, on reflection, the preceding shows is that the ordinary use of “I know” is the only partial common ground between Wittgenstein’s position and contemporary contextualism, for the latter simply do not countenance the grammatical use and deem the philosophical use continuous with the ordinary one. Yet, with respect to it, it turns out that Wittgenstein was in fact an invariantist. Reasons for knowledge claims may be good or bad, but when they are bad, knowledge does not obtain and one’s claim to knowledge is simply false. Reasons can also be stronger or weaker and it may be an open issue how strong they need to be in order to make one’s true belief knowledgeable. Yet, even in this case, there is no play made with the idea that different contexts raise or lower standards such that one’s identical reasons may suffice or fail to suffice for knowledge. Thus, Wittgenstein’s and contemporary contextualism about knowledge are actually completely different and pass each other by. However, I would like to inquire a bit deeper into Wittgenstein’s account of “I Know” to forestall a possible misunderstanding of his views, which I find problematic and which, alas, is gaining currency in contemporary discussions of skepticism. Namely, that Wittgenstein’s views on

“I Know”, “I know”, “I know”  227 the matter if not already committed to it, are at least compatible with epistemic disjunctivism. 21 Accordingly, he would be committed to or at least in favor of the idea of the factivity of reasons (and not just of knowledge). That is, the idea that, in good case scenarios, when we do perceive objects in our surroundings, then the fact itself and not its perceptual representation would be our reason for holding “p” true. Consider the following passage in On Certainty, in which Wittgenstein seems, strikingly, to be discussing ante litteram disjunctivism: “I know” has a primitive meaning similar to and related to “I see” (“wissen”, “videre”). And “I knew he was in the room, but he wasn’t in the room” is like “I saw him in the room, but he wasn’t there”. “I know” is supposed to express a relation, not between me and the sense of a proposition (like “I believe”) but between me and a fact. So that the fact is taken into my consciousness. (Here is the reason why one wants to say that nothing that goes on in the outer world is really known, but only what happens in the domain of what are called sense-data.) This would give us a picture of knowing as the perception of an outer event through visual rays which project it as it is into the eye and the consciousness. Only then the question at once arises whether one can be certain of this projection. And this picture does indeed show how our imagination presents knowledge, but not what lies at the bottom of this presentation. (OC 90) Wittgenstein was struck by the fact that in German the verb for “to know” is “wissen”, which has the same root as the Latin “videre” that obviously means “to see”. He also recognized the factivity of “to know”. For instance, in OC 178, he writes the wrong use made by Moore of the proposition “I know…” lies in his regarding it as an utterance as little subject to doubt as “I am in pain”. And since from “I know it is so” there follows “It is so”, then the latter can’t be doubted either. Taking knowledge to be akin to seeing, together with its factivity, produces a captivating “picture”, as he calls it. Namely, a picture according to which knowledge would consist in taking outer facts into one’s consciousness. Yet this is just a picture, and a bad one at that, for Wittgenstein. For we do not take outer facts into consciousness; nor do we have knowledge only of inner facts, since the latter, being mental in nature, might be thought of being taken in consciousness directly. In fact, it is safe to say that for Wittgenstein there are no facts – outer or inner – prior to their being conceptualized in certain ways. Hence, it is only

228  Annalisa Coliva thanks to pertinent hinges that our perceptual experiences can be taken to bear onto physical objects and thus give us access to what we thereby categorize as outer facts. Furthermore, our knowledge, as fallible as it is, just consists in having true beliefs supported by defeasible reasons. Indeed, it is of the essence of knowledge, for Wittgenstein, that it be defeasible and this can be so only if knowledge does not depend on or entail having factive reasons for the propositions known. As he writes in OC 12, ‘I know’ seems to describe a state of affairs which guarantees what is known, guarantees it as a fact. One always forgets the expression ‘I thought I knew’. Disjunctivism, with its insistence on factive reasons, is forgetful of this grammatical fact, as Wittgenstein would consider it. For, saying that one does know that p, iff one has factive reasons for p – viz. because one has taken in the very fact that p – excludes the possibility of being wrong – that one’s reasons may not be conclusive – and that p isn’t the case. Yet Wittgenstein insists that it is only when that possibility remains open that we do have, and are allowed to claim, knowledge. When that possibility is not open, then we are actually stumbling on something categorically different. Namely, a hinge, and our use of “I know” in that connection would actually not be the ordinary but the grammatical one, in which no epistemic relation obtaining between a subject and a proposition is expressed. A disjunctivist might reply that her position is compatible with Wittgenstein’s because one might altogether lack factive reasons, contrary to one’s initial belief. Hence, one might say, “I thought I knew”. This, however, would simply mean that one didn’t have knowledge in the first place. It would not mean that one’s reasons, when one does know, are defeasible. For, according to disjunctivists, either we have perceptual reasons, and in that case, they reach all the way down to facts, and cannot be defeated by any increment in information; or else, we don’t have perceptual reasons at all. Yet, the gist of OC 90 is that when genuine knowledge occurs – that is, when “I know” expresses a genuine epistemic relation (and is thus a case of “I Know”) – then reasons must be defeasible. If they weren’t, given that for Wittgenstein knowledge is factive, knowledge and our knowledge claims would then become infallible. Yet, it they were so infallible, they could not be genuinely epistemic. Thus, either they would amount to nonsense or could only be reinterpreted as grammatical. For, as we saw in §1.3, straddling contexts and taking knowledge claims to be epistemic and infallible at once is what, for Wittgenstein, produces nonsense.

“I Know”, “I know”, “I know”  229

5 Hinge Epistemology vs. Contemporary Contextualism vis-à-vis the Skeptical Challenge In closing, let me turn to the relationship between invariantism about “I Know”, skepticism, and my own version of hinge epistemology. The key point is that while I do not follow Wittgenstein in his claim that skeptical doubts are nonsensical, I don’t think they pose a threat to our knowledge either. Contemporary contextualists hold that, when the stakes are low, we have Knowledge of the first premise and of the conclusion of Moore’s argument: 1 2 3

Here is my hand;22 If there is my hand here, I am not a BIV; I am not a BIV.

They also hold that, when produced in the philosophical context, where standards for Knowledge go up and we are required to exclude the negation of (3) in order to have Knowledge of (1), we end up having Knowledge of neither. By contrast, on my view, thanks to empirical evidence and to the relevant hinges such as “There is an external world”, “My sense organs are generally reliable”, and “I am not a BIV”, we can have Knowledge of ordinary empirical propositions. This is indeed the gist of what I call the moderate account of perceptual justification. Namely: Moderate conception of perceptual justification: in order to have a perceptual justification for P (e.g. “Here is my hand”) you need a hand-like experience, no defeaters, and to assume Q (“There is an external world”, etc). 23 Hence, if we possess perceptual justification and our belief is true, then we also have Knowledge of ordinary empirical propositions. Yet, on my account, we do not have Knowledge of hinges, since reasons for them would be circular. Hinges, rather, are assumed and are assumed rationally, albeit without any justification, because they themselves are constitutive of epistemic rationality. Elsewhere,24 I have developed the details of the argument. In short, just as the constitutive rules of a game are part of the game even if they are not moves within it, so hinges, while neither justified nor justifiable (and therefore knowable), are part of epistemic rationality because they make it possible to acquire justification and knowledge for ordinary empirical propositions, as well as to raise motivated doubts about them. Amongst hinges, besides “There is an external world”

230  Annalisa Coliva and “My sense organs work mostly reliably”, there are also anti-Cartesian ones such as “I am not a BIV” or “I am not a victim of a lucid and sustained dream”. Since the very possibility of any empirical inquiry hinges on – pun intended – taking for granted that we are acquiring information through our senses about our surroundings and that we are cognitively well-functioning, hinges are constitutive of epistemic rationality itself. A noteworthy consequence of my position is that rational justification does not transmit across known entailment, when the conclusion of the argument is a hinge. For it is only by presupposing the conclusion that one can have a perceptual justification for the premise. Yet, since the premise can be justified and known while the conclusion, if it is a hinge, never is, this also entails that the Closure Principle does not hold unconditionally. That is, when the premise is something like “I have hands” and the conclusion is “There is an external world”, or “I am not a BIV”, one may have justification and knowledge of the former, and know that the premise entails the conclusion, and still have no justification or knowledge of it.25 As is well known, DeRose has raised an objection against those who, like me, end up denying Closure. Namely that if we don’t have Knowledge of hinges, while retaining Knowledge of ordinary empirical propositions, we would have to condone abominable conjunctions such as “I Know there is a hand here, but I don’t Know I am not a BIV”. The prima facie force of this objection is, in my view, due to a failure of appreciating the role that hinges such as “I am not a BIV” actually play in the practice of providing reasons for, and Knowledge of ordinary empirical propositions. Surely, then, we do not Know them, yet they play a positive constitutive role with respect to epistemic rationality. As I have argued at length elsewhere, we do not Know them but we (skeptics and non-skeptics alike) are mandated by epistemic rationality to accept them, 26 since they are constitutive of epistemic rationality itself. Indeed, a skeptic’s mistake, in my view, depends on concluding that since they are not sustained by epistemic reasons, they fall outside the scope of epistemic rationality altogether, when in fact they are within it since they are its constitutive elements. To repeat, just as rules are part of a game, so those hinges the assumption of which is constitutive of epistemic rationality are part of epistemic rationality itself. Therefore, we are mandated to accept them by epistemic rationality itself, and not just in virtue of pragmatic considerations. Hence, in my view, the skeptic’s mistake depends on holding on to Epistemic RationalitySk: a proposition is epistemically rationally held if and only if there is an evidential justification for it, instead of embracing the correct, extended characterization of epistemic rationality, namely, Epistemic RationalityExt: a proposition is epistemically rationally held if and only if either there is an evidential justification for it, or,

“I Know”, “I know”, “I know”  231 while being unjustifiable, it needs to be assumed in order to have epistemic justifications. Now, as Harman and Sherman (2011) notice, prima facie abominable conjunctions can always be produced when one is dealing with neighboring notions, which only philosophical analysis can tease apart. The neighboring notions here at play are belief and assumption, on the one hand, and being justified (which I take to be a necessary condition for Knowledge) and being rationally mandated, on the other. Thus, armed with such an analysis, we can then counter the objection with the following, non-abominable conjunction “I Know I have a hand, and although I don’t Know I am not a BIV, I am rationally mandated to assume that I am not”. Thus, according to my own version of hinge epistemology, one can retain invariantism about Knowledge, as well as Knowledge of ordinary empirical propositions, without having to embrace infallibilism. By going constitutivist about hinges, one will not be making them the object of Knowledge. Yet, such a “skeptical” solution is in fact also a powerful “undercutting” solution to skepticism. For, ultimately, it shows that the skeptical paradox, in either of its forms, rests on a narrow and mistaken account of epistemic rationality.

Notes 1 I would like to thank Christos Kyriacou and Genia Schönbaumsfeld for very helpful comments on previous versions of this paper. While elements of disagreement may remain, I have learnt a lot from them. 2 One might say that given that foundationalism and coherentism involve some inherent circularity (and infinitism is implausible), some circularity in reasoning is inevitable. Maybe so but notice that the kind of circularity in question would be one in which the existence of physical objects needs to be presupposed in order to have a perceptual justification for “here is a hand”, say. Hence, using that justified premise to try and get a justification for “there are physical objects” with a Moore-style proof would be epistemically bad. 3 To avoid unnecessary repetitions, I will expand on the non-circularity requirement in the following. 4 Here Wittgenstein is noticing how puzzling it would be to hold “I don’t know I am a human being”. It could be taken to mean that I don’t know the meaning of “human being”, say, but failing that, it seems senseless to say that one does not know to be a human being, since everything speaks in favor of it and nothing against it. 5 For Wittgenstein ‘know’ is a family resemblance concept. Important remarks on this can be found both in On Certainty and in Philosophical Investigations (cf. §78, 246–247). He also distinguished between objective and subjective certainty. The former, which is the object of On Certainty, is not an epistemic category, for him, but a grammatical one; the latter, in contrast, is a psychological category and has no special philosophical relevance, for him (cf. especially OC 194, 270, 273; see also 15–16, 203). 6 In the later Wittgenstein, the adjective “logical” is often synonymous with “grammatical” and should not be conflated with the contemporary understanding of it.

232 Annalisa Coliva 7 Notice that a grammatical role or function in OC is neither epistemic nor pragmatic, but constitutive of a practice and/or meaning. 8 As remarked in Wright (1985) and in Moyal-Sharrock (2005), in On Certainty, the notions of a rule, as well as that of grammar, are extended to cover not only linguistic norms but also those which we may call “norms of evidential significance”. 9 I am not considering here a Cartesian skeptical scenario for that is dealt with differently by Wittgenstein. See Section 2. 10 This is the gist of therapeutic proposals. In this vein, Penelope Maddy (2017, 200) takes Wittgenstein to be proposing a “sadly crimped view” of philosophy, which would consist simply in bringing us back to common sense, and the ordinary use of “I know”, after catching the philosophical disease. In Coliva (Forthcoming), I contend that therapeutic readings have arguably misunderstood the aim and scope of “therapy” in Wittgenstein’s later writings. 11 Actually, in the Philosophical Investigations Wittgenstein had already developed several of these considerations regarding various uses of “I know”. One should read the passages on the possibility of knowing other minds and the impossibility of self-knowledge understood as a sure-fire way of epistemically knowing one’s own sensations and on-going propositional attitudes in light of this tripartite distinction. For a discussion, cf. Coliva (2010, chapter 2). 12 For an example of the former option, see DeRose (1995), and for an example of the latter, see MacFarlane (2014). 13 For a fuller treatment of Wittgenstein’s anti-skeptical strategies, see Coliva (2010, chapter 3). 14 Cf. Conant (1998). OC 347–360 is a sequence of passages that clearly brings out Wittgenstein’s resolute attitude towards philosophical uses of “I know” and towards philosophical forms of doubt. 15 Notice that this is entirely compatible with the fact that these practices may change due to social pressures of various nature (e.g. scientific, political). Wittgenstein’s descriptivism, therefore, need not be in the service of the defense of any status quo. It is just that these changes are not typically brought about by philosophy itself. 16 At least according to standard “framework” interpretations of Wittgenstein’s views, such as McGinn (1989), Moyal-Sharrock (2005), and Coliva (2010). Alternatively, grammatical uses would still be truth-apt, but their truth would neither depend on their correspondence with a certain state of affairs, nor on there being conclusive evidence in their favor. I have dealt at length with these issues in Coliva (2018a). 17 I have discussed Williams and Wright’s epistemic readings of On Certainty at length in Coliva (2019, 2020a). 18 Further reasons of discontent can be found in Coliva (2010, chapter 3). 19 Unsurprisingly Travis (1989, 234–235) and Williams (2004a, b) deny, citing no passage in On Certainty, that hinges are norms. 20 Williams simply ignores Wittgenstein’s suggestion that there are “grammatical” uses of “I know” and takes them to show that Wittgenstein favored an externalist view of knowledge, whereby one may have knowledge of P even if unable to articulate grounds in its favor, provided one were entitled to it. Another possible source of Williams’ confusion is the fact that sometimes Wittgenstein considers practical, non-propositional knowledge for which it is obviously unnecessary to be able to offer reasons, which would prove that one knows. 21 See Pritchard (2016) and Schönbaumsfeld (2017) and in conversation.

“I Know”, “I know”, “I know”  233 22 Notice that I allow for fewer hinges than Wittgenstein, and like contemporary epistemologists, I regard “Here is my hand” as an ordinary empirical proposition. For a defense of the moderate account as opposed to the liberal and conser23 vative ones, endorsed by Pryor (2004) and Wright (1985, 2004) respectively, see Coliva (2015, chapters 1 and 2). 24 The details of the argument are presented in Coliva (2015, chapter 4). 25 For the details of my position on Transmission Principles and Closure, see Coliva (2015, chapter 3). 26 Acceptance is preferable to belief in this context because acceptances, contrary to beliefs, can be in good standing even if they are not supported (nor are they supportable) by reasons. For a more thorough treatment of this aspect of my proposal, see Coliva (2015, chapter 2). Notice that, in my view, acceptance is a propositional attitude of holding true without epistemic reasons in favor of the proposition so accepted, and it is also fully compatible with strong commitment to the truth of what is accepted. In short, it is a categorical acceptance rather than a hypothetical one.

References Coliva, A. 2010 Moore and Wittgenstein. Scepticism, Certainty and Common Sense, London, Palgrave. Coliva, A. 2015 Extended Rationality. A Hinge Epistemology, London, Palgrave. Coliva, A. 2018a “What anti-realism about hinges could possibly be”, in C. Kyriacou and R. McKenna (eds.), Metaepistemology. Realism and Anti-Realism, London, Palgrave Macmillan, pp. 267–288. Coliva, A. 2018b “Strange bedfellows. On Pritchard’s disjunctivist hinge epistemology”, Synthese, https://doi.org/10.1007/s11229-018-02046-z. Coliva, A. 2019 “In quest of a Wittgensteinian hinge epistemology”, in N. Da Costa and S. Wuppuluri (eds.), Wittgensteinian. Looking at the World from the Viewpoint of Wittgenstein’s Philosophy, Cham, Springer, pp. 107–121. Coliva, A. 2020a “Against neo-Wittgensteinian entitlements”, in P. Graham and N. Pedersen (eds.), New Essays on Entitlements, Oxford, Oxford University Press, pp. 327–343. Coliva, A. Forthcoming. “Which philosophy after philosophy? Rorty, Wittgenstein and Goethe”. Conant, J. 1998 “Wittgenstein on meaning and use”, Philosophical Investigations 21, pp. 222–250. DeRose, K. 1995 “Solving the skeptical problem”, Philosophical Review 104/1, pp. 1–52. Harman, G. and Sherman, B. 2011 “Knowledge and assumptions”, Philosophical Studies 156/1, pp. 131–140. MacFarlane, J. (2014). Assessment Sensitivity: Relative Truth and its Applications. Oxford University Press. Maddy, P. 2017 What Do Philosophers Do? Skepticism and the Practice of Philosophy, Oxford, Oxford University Press. McGinn, M. 1989 Sense and Certainty: A Dissolution of Scepticism, Oxford, Blackwell.

234  Annalisa Coliva Moyal-Sharrock, D. 2005 Understanding Wittgenstein’s On Certainty, London, Palgrave. Pritchard, D. 2016 Epistemic Angst. Radical Skepticism and the Groundlessness of Our Believing, Princeton, Princeton University Press. Pryor, J. (2004). What’s wrong with Moore’s argument? Philosophical Issues 14 (1): 349–378. Schönbaumsfeld, G. 2017 The Illusion of Doubt, Oxford, Oxford University Press. Searle, J. 1969 Speech Acts: An Essay in the Philosophy of Language, Cambridge, Cambridge University Press. Travis, C. 1989 The Uses of Sense, Oxford, Oxford University Press. Williams, M. 2004a “Wittgenstein’s refutation of idealism”, in Denis MacManus (ed.), Wittgenstein and Scepticism, London-New York, Routledge, pp. 76–96. Williams, M. 2004b “Wittgenstein, truth and certainty”, in Max Kölbel and Bernhard Weiss (eds.), Wittgenstein’s Lasting Significance, London-New York, Routledge, pp. 249–284. Wittgenstein, L. 1953 Philosophical Investigations, Oxford, Blackwell. Wittgenstein, L. 1969 On Certainty, Oxford, Blackwell. Wright, C. 1985 “Facts and certainty”, Proceedings of the British Academy 71, pp. 429–472. Wright, C. 2004 “Warrant for nothing (and foundations for free?)”, Aristotelian Society Supplementary 78/1, pp. 167–212.

12 ‘Logical’ and ‘Epistemic’ Uses of ‘to Know’ or ‘Hinges’ as Logical Enabling Conditions Genia Schönbaumsfeld 1 Introduction In The Illusion of Doubt, I developed an account of Wittgensteinian ‘hinge propositions’ within the context of a wider diagnostic antisceptical strategy that showed that radical scepticism is an illusion generated by a prior endorsement of a Cartesian picture of our evidential situation that we have no good independent reason to accept. This strategy was bolstered by Wittgenstein’s insights that a ‘global’ doubt is not a doubt, and the very idea of a ‘global validation’ of all our epistemic practices taken together incoherent. My interpretation of those propositions that must ‘stand fast’ if epistemic enquiry is to be possible, however, is quite different from those currently available in the literature. For, according to most hinge epistemologists,1 ‘hinges’ are basically certain, but cannot be known. Conversely, proponents of what I call a ‘quasiepistemic’ reading 2 contend that hinges can be known, though knowledge claims about them are trivial. I, on the other hand, propose that ‘hinge propositions’, although they ‘stand fast’, can neither be known, nor are they certain in the ordinary sense, since, doubt, in respect to them, is ‘logically excluded’ (OC §194). In order to better understand what this means, I distinguish between a ‘logical’ and an epistemic sense of ‘to know’: where the expression of uncertainty is senseless, but no further grounds can be given, we are dealing with a purely ‘logical’ use of ‘to know’ (OC §58–59); where it is possible to be wrong or uncertain, but reasons for correctness can be given, we are dealing with the ordinary, ‘epistemic’ sense. As we shall see in this chapter, this distinction does important philosophical work, as it enables us to fulfil three desiderata usually thought impossible to fulfil together, namely, the joint avoidance of (1) epistemic contextualism,3 (2) denials of closure, and (3) concessions to radical scepticism. I will begin by drawing out what I take to be the key aspects of Wittgenstein’s conception, and will then go on to discuss what bearing they have on the aforementioned three points.

236  Genia Schönbaumsfeld

2 ‘Hinges’: Neither Certain nor Known The Cartesian idea that one could scrutinize and reject all of one’s epistemic beliefs taken together comes under sustained attack in Wittgenstein’s last collection of notes, On Certainty. In this work, Wittgenstein shows that doubt needs a context, a common human background that stands fast and against which specific epistemic claims can be assessed. A ‘global’ doubt is a paper doubt, as it tries to doubt everything at the same time and in this way fails to doubt anything at all: ‘if you tried to doubt everything you would not get as far as doubting anything. The game of doubting itself presupposes certainty’ (OC §115). Radical scepticism is the thought that all our perceptual beliefs taken together could be false – i.e. could fail to ‘match’ the way the world is (if, indeed, there is one). That is to say, even in paradigmatic cases of perceptual knowledge, such as when I’m looking at my hands and nothing out of the ordinary is occurring, I might be wrong. One way of countering the sceptic’s proposal would be to insist, with Moore, that in such cases I do know that I’ve got hands. But Wittgenstein thinks that this is actually not a good tactic: ‘Moore’s mistake lies in this – countering the assertion that one cannot know that, by saying “I do know it”’ (OC §521). For if the existence of my hands is in doubt, I cannot rely on any ordinary way of checking that a physical object is in front of me, say by looking more closely, as I cannot, in such radical circumstances, trust my eyes either. Nothing that we would ordinarily call ‘evidence’ could help us here. What is more, while an ordinary mistake, say such as can occur when adding together large numbers in one’s head, does not call into question the intelligibility of the entire practice in which one is participating (arithmetic, for example) – but only one’s own competence – the notion that one could be ‘wrong’ about the most fundamental things (such as the presence of one’s hands) undermines the coherence of the very practices on which one’s expression of doubt at the same time depends (since if there were no such practice, there would be no such doubt). Complete epistemic devastation can, therefore, only be avoided if we recognize that there are some propositions, call them ‘hinges’ (§341), where doubt is ‘logically excluded’ (§194). Consequently, it is impossible to doubt all our most fundamental commitments taken together, as some of these commitments – the ‘hinges’ – must stay in place ‘if I want the door to turn’ (§343). But if this is so, then neither a ‘global validation’ nor a ‘global indictment’ of all our epistemic practices taken together is possible. Rational assessment is necessarily ‘local’ – that is, it can only take place relative to a background of hinge commitments that stand fast.4 If this is right, it implies that we don’t stand in an epistemic relation to ‘hinge propositions’ at all. Rather, ‘their role is like that of rules of a

‘Logical’ and ‘Epistemic’ Uses  237 game; and the game can be learned purely practically, without learning any explicit rules’ (OC §95). In other words, propositions expressing these commitments may look like ordinary empirical claims, but that appearance is deceptive, for their role is ‘descriptive of the language-game’ (OC §56), and in this much ‘logical’ or ‘technique-constituting’. So one might say that ‘hinge propositions’ are an attempt to articulate the logical enabling conditions that allow our epistemic practices to operate, and without which even our words could not mean anything.5 As Wittgenstein puts it: If ‘I know etc.’ is conceived as a grammatical proposition, of course the ‘I’ cannot be important. And it properly means ‘There is no such thing as a doubt in this case’ or ‘The expression “I do not know” makes no sense in this case’. And of course it follows from this that ‘I know’ makes no sense either. ‘I know’ is here a logical insight. Only realism can’t be proved by means of it. (§58–59) In order not to get confused between what Wittgenstein calls a ‘grammatical’ or ‘logical’ sense of ‘to know’ and ordinary uses of these terms, I therefore propose to distinguish between a ‘logical’ and an ‘epistemic’ sense of ‘to know’. For example, an unproblematic – i.e. straightforwardly ‘epistemic’ – employment of ‘I know I have two hands’ would be in the following context. I visit someone in hospital who has been in an accident and his whole body is covered in bandages. In such a case, I might not know whether this person still has hands, for they might have been amputated. If he reassures me by saying ‘I know I have hands’, then I will take this to imply that he has been able to check (say, by having had the bandages removed by a doctor earlier in the day).6 Here the claim ‘I know I have hands’ makes sense, since we also have a clear idea of what it would be like not to know that one has hands. That is to say, in ordinary, what I’m going to call, ‘epistemic’, cases of ‘knowing’, there is ‘logical space’ both for knowing the proposition in question as well as for not doing so (depending on the way things happen to be). Furthermore, if one does know, one is able to offer justification for one’s knowledge claim (e.g. having made sure in the relevant way, say by looking). None of these conditions are met, however, in the case of what one might call ‘radical’ doubt. For example, if I assert in ordinary circumstances where nothing unusual (such as accidents, etc.) has occurred, that I know that I have hands (in order to counter a radical sceptical claim, say), it is unclear what sort of justification I could offer for this. That is to say, if, in such a context, I could nevertheless be ‘wrong’ about having hands, then this would be no ordinary ‘mistake’,

238  Genia Schönbaumsfeld but would rather constitute what Wittgenstein calls ‘an annihilation of all yardsticks’ (OC §492). So, one might begin to suspect that what the radical sceptic is really after is not a demonstration that my hands (or any other particular physical objects) exist, but rather a proof that the ‘external world’ as such exists – that all of our epistemic practices taken together manage to ‘latch on’ to something ‘outside’ of us. Contrary to what Moore is trying to do, however, namely, to play the same game as the sceptic, but to come down on the opposite side, Wittgenstein thinks that what we need to show is that this kind of ‘global validation’ demand is not a coherent idea. There is no such thing, nor could there be such a thing, as a global validation of all our epistemic practices taken together, and, consequently, its absence is neither lamentable nor an epistemic limitation.7 Since if we were to suppose, with the radical sceptic, that all of our ‘external world’ propositions taken together could be false, then the words we use to formulate this ‘doubt’, would lose their meaning too: ‘If you are not certain of any fact, you cannot be certain of the meaning of your words either’ (OC §114).8 That your words mean what they do is also a fact about them. So, if everything is uncertain, as the sceptic supposes, then ‘meaning’ – if indeed there could still be such a thing – would no longer be stable either. Consequently, it’s not just that scepticism is unstateable – but might, for all that, be true9 – it is rather that, if Wittgenstein is right, it is no longer clear what the sceptic’s ‘thought’ can even amount to (i.e. this is not merely a ‘semantic’ point). Hence, it seems that the kind of justification that the sceptic is after – a fully general evaluation of our entire rational or epistemic system – is impossible. But this is not because human powers are unequal to the task. Rather, Wittgenstein thinks, the very idea is misguided: ‘So is the hypothesis possible that all the things around us don’t exist? Would that not be like the hypothesis of our having miscalculated in all our calculations?’ (OC §55). For a ‘miscalculation’ in all our calculations is not, as it were, just an ‘aggregated’ mistake, but rather implies that we have never calculated at all – that nothing that we have ever done counts as an instance of calculating. But, if so, the sceptic owes us an explanation of what ‘calculation’ really means, and of what it is, that we allegedly cannot do (or know). In the absence of such an explanation, we need not accept the sceptic’s challenge that unless we can show in advance that we are not radically mistaken about everything, we aren’t entitled to our ordinary knowledge claims. Consequently, it’s not that we groundlessly need to accept that we know that we have hands, for example, it is rather that we have no clear idea of what it might mean not to ‘accept’ it. This is why Wittgenstein says that if, in ordinary circumstances, Moore were to pronounce the opposite of those propositions which he declares certain, ‘we should not just not share his opinion: we should regard him as demented’ (OC

‘Logical’ and ‘Epistemic’ Uses  239 §155).10 If this were an ‘ordinary’ case of not knowing, where one could cite evidence, one could just not share Moore’s opinion. Since this is not possible, however, Moore-type propositions can, at best, express a ‘logical’ (or ‘grammatical’) insight: ‘I know I have two hands’, asserted in ordinary circumstances, really means ‘there is no such thing as a doubt in this case’ – it does not make sense as an anti-sceptical (Moorean) claim. At this point, a sceptically minded philosopher might perhaps object that, contrary to what I have just argued, we do have a clear idea of what it would be like not to know that one has hands, for, surely, that’s just the BIV (or radical sceptical) scenario! That is to say, if I were a BIV, I could not know that I have hands, since, firstly, I could not know anything, and, secondly, I would not even have hands. Although the latter two claims are true, the objection does not constitute a genuine counter-example to the foregoing argument. Taking the second point first: if I don’t have hands (say, because I’m a BIV), and I could (miraculously) come to know this fact, then my claim ‘I know I don’t have hands’ (or ‘I doubt that I have hands’) would be perfectly meaningful, and, indeed, not relevantly different from the accident scenario discussed above. If, on the other hand, the BIV scenario is only a fancy way of fleshing out the thought that one might be globally wrong about everything (as it tends to be), then ‘I know I have two hands’ is no longer a proposition about hands (or other body parts), but rather means something like: ‘I know I cannot be globally wrong’ or ‘I know I’m not the victim of a “global” sceptical hypothesis’. But, if so, we have moved from an ‘epistemic’ use of ‘to know’ to what I have called a ‘logical’ or ‘grammatical’ sense. So, we are not concerned with an ordinary factual (empirical) claim, but rather with a logical enabling condition: if the obtaining of this condition is called into question, then the notions of ‘truth’ and ‘falsity’ that the sceptic needs in order to be able to formulate his doubt lose their meaning too. ‘I know I have two hands’ can, therefore, look like an ordinary epistemic (empirical) claim, but in the radical sceptical context (i.e. in ordinary circumstances where radical sceptical error-possibilities have been raised) either has no clear meaning or else constitutes an articulation of the ‘logical truth’ that ‘there is no such thing as a doubt in this case’ (OC §58).11 Ordinary knowledge attributions (epistemic senses of ‘to know’) are bipolar, in other words: it only makes sense to think one knows a proposition, if it also makes sense to think one might not have known it (had the world been different). This applies both to cases of claims to know as well as to knowledge possession itself. Where a doubt is unintelligible, on the other hand, ordinary knowledge attributions make no sense. OC §10 makes this clear: I know that a sick man is lying there? Nonsense! I am sitting at his bedside, I am looking attentively into his face. – So, I don’t know,

240  Genia Schönbaumsfeld then, that there is a sick man lying here? Neither the question nor the assertion makes sense. Any more than the assertion “I am here”, which I might yet use at any moment, if suitable occasion presented itself… Williams (forthcoming) has recently objected that this passage is only about the intelligibility of asserting a knowledge claim, not about the knowledge attribution itself, which, Williams thinks, Wittgenstein would endorse. But this seems false. For the question ‘I know that a sick man is lying here?’ is not about a knowledge claim, but about whether I know that a sick man is lying there. When Wittgenstein says ‘neither the question nor the assertion makes sense’, what he is saying is that neither the knowledge claim, nor the question whether I know makes sense, thus showing that nothing much hangs for him on distinguishing between the assertability of knowledge and knowledge attribution in this context. Where a doubt is unintelligible, neither knowledge attribution, nor claims to know make sense. Williams goes on to raise the following problem: ‘Reporting my experience, I say, “I knew right away that there was a sick man lying there.” I speak truly. How can this be if, as Schönbaumsfeld supposes, I didn’t know at the time?’ The ‘right away’ makes all the difference here. Seeing what looks like a human form huddled in blankets, I might not have known straight away that a sick man was lying there. I might have had to look more closely. This is what makes this case an ordinary case of knowing, where a doubt is perfectly intelligible and grounds can be given. I know that there is a sick man lying there, because I have taken a close enough look. But, from further away, I might have mistaken the man for a child. Of course, none of these grounds – taking a closer look, switching on the light, etc. – would be ‘robust’ enough to satisfy a sceptic. This is why Wittgenstein criticizes Moore’s refutation attempt by appealing to things one knows. But Wittgenstein also thinks that the sceptic’s doubt is an illusion of doubt – not because we know the things in question, but because such a doubt is itself unintelligible. Williams comes quite close to saying something similar: ‘But where there is no room for doubt, there is no room for assuaging it by giving compelling grounds. In such cases there is knowledge, but nothing would be meant by saying “I know.”’ I agree with the first sentence, but not with the second. ‘Objective certainty’ (where doubt is logically excluded) is not a kind of certainty. In cases where something is ‘objectively certain’, it is impossible to be wrong (to make a ‘mistake’), because ‘being wrong’ would constitute an ‘annihilation of all yardsticks’ (OC §492) – i.e., it would be a ‘move’ for which there is no room in the game (any game). Consequently, ‘objective certainty’ refers to the logical enabling conditions that make our practices possible, and Wittgenstein

‘Logical’ and ‘Epistemic’ Uses  241 specifically contrasts such conditions with knowledge (and ordinary certainty12). There are two reasons for this. If I know something, it must be possible to provide some grounds for how or why I know. This does not require attributing a JTB conception of knowledge to Wittgenstein, as this is anachronistic and at odds with Wittgenstein’s anti-theoretical bent, but it seems uncontroversial that Wittgenstein does think that knowledge requires justification of some kind: ‘Whether I know something depends on whether the evidence backs me up or contradicts me’ (OC §504). Second, and as already mentioned, knowledge attribution is bipolar for Wittgenstein. Since ‘objective certainties’ fail both of these conditions, they cannot be knowledge. The following passages make this clear: With the word “certain” we express complete conviction, the total absence of doubt, and thereby we seek to convince other people. That is subjective certainty. But when is something objectively certain? When a mistake is not possible. But what kind of possibility is that? Mustn’t mistake be logically excluded? (OC §194) In respect to ordinary contingent empirical propositions, doubt cannot be logically excluded. Since, as Wittgenstein says, ‘It is always by favour of Nature that one knows something’ (OC §505), perceptual knowledge cannot, pace Williams, be an ‘objective certainty’. Rather, once the radical sceptical problem has been exposed as an illusion,13 the possibility (and ordinary certainty) of perceptual knowledge will depend on whether I find myself in the good or the bad case. If I am in the good case, then I can know the thing in question (such as that there is a computer in front of me), whereas in the bad case (where I am confronted by a simulation, say), I do not.14 What stands fast, on the other hand, and is ‘objectively certain’, are the epistemic and rational principles that allow us to draw such distinctions in the first place. These are the logical enabling conditions, the ‘hinges’, that allow our epistemic practices to operate. Since such logical enabling conditions function like the constitutive rules of a game, they cannot be called into question without undermining the game as such. Does this imply that they are known or ‘certain’ in the ordinary sense? No. For this would imply that they could also be uncertain had the world been different. But this is impossible, as they constitute the background conditions that give meaning to our ordinary concepts. In this respect, the ‘objective certainty’ of the hinges is like what Wittgenstein says of the standard metre in Paris – just as the standard metre is neither 1 metre nor not 1 metre long, so hinges are neither certain nor uncertain: ‘But this is, of course, not to ascribe any remarkable property to it [them],

242  Genia Schönbaumsfeld but only to mark its [their] peculiar role in the game of measuring with a metre-rule’ (PI 50).15 In other words, just as the standard metre, hinges are instruments of the language (practice); they are not something that is represented, and, hence, known, but a ‘means of representation’ (ibid.). For this reason, in radical sceptical contexts, ‘I know I have two hands’, is not really a specific empirical claim about hands, but rather conveys the ‘logical insight’ (OC §59) that there is no such thing as a doubt in this case or that ‘“The expression ‘I do not know’ makes no sense in this case”.’ Since it follows from this, according to Wittgenstein, that ‘“I know” makes no sense either’ (OC 58), objective certainty cannot be a form of knowledge (but this, of course, is not a detraction16). Neither, however, is ‘objective certainty’ a way of being ‘basically certain’. If, in particular contexts, it is logically impossible to doubt, it is logically impossible to be certain too – even if, as Moyal-Sharrock (2004, forthcoming), for example, contends, this is supposed to be a non-reflective, ‘animal’ certainty.17 That is to say, Moyal-Sharrock endorses the notion that I reject, namely, that ‘objective certainty’ is a kind of certainty, which she takes to be a non-epistemic, non-propositional type of ‘basic belief’. I, on the other hand, propose that Wittgenstein rejects the whole idea that we need ‘basic beliefs’ at all. From the way in which Moyal-Sharrock (forthcoming) writes, one might think that she believes that there is a genuine regress problem and that Wittgenstein solved it by presenting an enactive theory of hinge certainty. But such a notion not only flies in the face of Wittgenstein’s anti-theoretical bent18; it also assumes that Wittgenstein is accepting the terms of the debate – for example, that he agrees that there is a need for ‘basic beliefs’ of some kind – and then offers a novel way of conceiving of them. This strikes me as wrong-headed: it is Wittgenstein’s aim to dissolve philosophical problems (PI §133) by showing that they arise because a picture (say the Cartesian one) has held us captive (PI §115); once the picture has been undermined, the house of cards (PI §118) collapses of its own accord. Applied to OC, what this means is that Wittgenstein wants to show that radical doubt is an illusion (OC §19). Since illusions or the problems generated by them cannot be refuted or solved, but only dissolved, no ‘positive’ response to scepticism is required. In other words, if radical doubt is an illusion, then so is the regress problem. Why? Because the regress problem is one ramification of radical doubt. If someone believes that there is a genuine radical sceptical problem, then one way of responding to that problem would be by appeal to something ‘indubitable’ (e.g. the Cogito, the ‘certainty’ of introspection, ‘hinge’ or ‘objective certainty’) that the sceptic cannot challenge. Conversely, if the radical sceptical problem is exposed as illusory, there is no need to appeal to ‘basic beliefs’ of whatever form, in order to solve it (or its ramifications). If, as Wittgenstein says, justification comes to an end (OC §192), then

‘Logical’ and ‘Epistemic’ Uses  243 there is no ‘final’ justification or ‘ground’,19 be that epistemic, or, as on Moyal-Sharrock’s conception, a non-epistemic, basic ‘animal certainty’. Wittgenstein rejects the Cartesian Picture root and branch. 20 If this is right, then ‘standing fast’ is not to be equated with ‘hinge certainty’, where that is to be construed as some kind of affective state. There is no ‘what it is like’ to be basically certain, just as there is no ‘what it is like’ to doubt all of one’s perceptual beliefs taken together. For what could such a ‘what it is like’ consist in? Does my taking hold of a towel without hesitation (OC §510), for instance, show that I am ‘basically certain’ that the towel exists? Of course not. In the course of ordinary life such a strange idea would not even occur to me. I just use the towel. There is no question of its existing (or not existing), or its being certain that it exists. The absence of doubt does not manifest the presence of some pro-attitude, not even a non-reflective one. 21 What is more, if this attitude is supposed to be like a faith or a trust, as Moyal-Sharrock also claims, then, surely, this must imply that hinges could, in principle, be up for grabs. Otherwise, what is the point of calling it a faith or a trust at all? One only trusts or has faith in contexts where that trust (or faith) could in principle be disappointed. If it is logically impossible for the trust to be disappointed, however, as in the case of ‘hinge certainty’, what is it that still makes this a trust? It doesn’t appear to have anything in common with ordinary cases of trust or faith where one is always exposing oneself to the contingencies (and vicissitudes) of life. So, it seems that it is only if Cartesian global doubt continues to worry us that we need basic certainty about what might otherwise be thought inherently doubtful. But once this has been exposed as an illusory concern, there is no longer room for this kind of anxiety, and, hence, nothing that we need to put in its place. 22

3 Wittgenstein and Epistemic Contextualism I’m now going to consider the question whether the foregoing account implies that Wittgenstein thinks that ‘to know’ is context-sensitive, as contemporary epistemic contextualists, for instance, maintain. 23 Epistemic contextualism is the view that the standards for whether one is able to attribute knowledge to someone (or oneself) vary with context, such that ‘S knows that P’ expresses a truth in ordinary contexts, but not in a ‘high standards’ context where sceptical error-possibilities have, for example, been raised. The first point to make here is that Wittgenstein would reject the whole idea that there is a genuine radical sceptical context, along with the concomitant notion that the radical sceptic’s standards are ‘higher’ than ordinary standards. For, as we have already seen, ‘global’ doubts, according to Wittgenstein, are not doubts at all: ‘A doubt that doubted

244  Genia Schönbaumsfeld everything would not be a doubt’ (OC §450). So, it’s not that the radical sceptic is somehow applying a higher than normal standard, it is rather that she is rejecting the very idea of one – she is, in effect, and as Wittgenstein says, annihilating all yardsticks. Consequently, Wittgenstein would not be in favour of ceding the high ground to radical scepticism in the way that epistemic contextualism does by granting to the sceptic that, by her standards, we know nothing. Neither would Wittgenstein agree that it is impossible to know ‘hinge propositions’ to be true, because, in a ‘high standards’ context, we are operating with standards of knowledge that outstrip the available evidence. Rather, we cannot know ‘hinge propositions’ to be true, because we can also not know them to be false. Hinges are not truth-assessable ordinary empirical propositions at all, but logical enabling conditions that make it possible to speak of ‘truth’ and ‘falsehood’ in the first place. 24 If it is indeed a ‘truth of logic’ (in Wittgenstein’s sense) that no ‘global’ evaluation of all of our practices taken together is possible, then the thought that one could be wrong about everything all of the time turns out not to be false, but to lack sense (since the opposite of a ‘logical truth’ is an incoherent notion). In other words, one cannot assert a logical enabling condition, one can only mention for clarificatory or heuristic purposes (e.g. in order to dispel illusions), that there is no such thing as a doubt in this case. But this is not to make any kind of concession to scepticism. That ‘hinges’ cannot, in an epistemic sense, either be known or not known, either be true or false, just marks them out as logical enabling conditions rather than ordinary empirical propositions. 25 The notion that this must somehow render them deficient probably stems from the incoherent idea that our relation to the world must be descriptive and epistemic all the way down. If Wittgenstein is right, however, this is wrong-headed. For just as a game, in order to be a game, must have rules distinct from the moves that they make possible, so our practices require an unquestioned background in order to function. It is for this reason that Wittgenstein uses the famous ‘riverbed’ metaphor: The mythology26 may change back into a state of flux, the river-bed of thoughts may shift. But I distinguish between the movement of the waters on the river-bed and the shift of the bed itself; though there is not a sharp division of the one from the other. (OC §97) The problem with radical scepticism and traditional attempts to refute it is that they don’t take on board this fundamental distinction, and so the temptation to think that one could, or should, have the movement of the waters on its own continues to persist.

‘Logical’ and ‘Epistemic’ Uses  245 This also explains why Wittgenstein thinks that the proposition ‘there are physical objects’ is nonsense (OC §35). For if we cannot ultimately make sense of the notion that there could be no physical objects – just ‘sensory experiences’ – then we cannot coherently ‘prove’ (or ‘assert’) that there are, either. 27 Rather, and as Wittgenstein says, “A is a physical object” is a piece of instruction which we give only to someone who doesn’t yet understand either what “A” means, or what “physical object” means. Thus it is instruction about the use of words, and “physical object” is a logical concept. (OC §36) That is to say, the concept ‘physical object’ belongs to our epistemic (or logico-grammatical) scaffolding. It is not an empirical concept that could be up for grabs, 28 as there are no means of finding out that the world consists only of experiential seemings, say. Consequently, one cannot assert that ‘there are physical objects’, as this would constitute an attempt to ‘ground’ the ‘background’ (an attempt to assert a logical or grammatical truth as if it were an ordinary empirical claim), and any such attempt will itself presuppose the very background in question:29 ‘If the true is what is grounded, then the ground is not true, nor yet false’ (OC §205). It follows from all this that the distinction between an epistemic and a logical sense of ‘to know’ has nothing to do with the context-sensitivity of standards for knowledge that epistemic contextualists invoke. Rather, Wittgenstein’s distinction is categorical: in its logical employment, ‘to know’ signals the unintelligibility of a doubt; in its ordinary, epistemic use ‘to know’ shows that we are certain. Regarding the latter, it is always possible that we could have made a mistake in ascribing knowledge to ourselves or others (had things been otherwise), in the former case, no ‘mistake’ is possible. So, one might say that epistemic uses of ‘to know’ rule out ordinary doubts, while the logical sense rules out ‘extraordinary’ ones.30

4 Implications for ‘Closure’ Most philosophers who have considered Wittgenstein’s On Certainty think that Wittgenstein must deny closure. Here is Williams: ‘To say that our relation to framework judgements is non-epistemic is to say that they neither amount nor fail to amount to knowledge. But surely they can be entailed by judgements to which we do have an epistemic relation, in which case knowledge fails to close under known entailment’ (Williams (1996: 187)).31 Since most epistemologists don’t exactly welcome this implication,32 however, some form of damage limitation seems required. Williams

246  Genia Schönbaumsfeld himself, for example, argues for the instability of knowledge under the conditions of reflection, 33 while Wright (2004) has famously defended a non-evidential warrant – or ‘entitlement’ – strategy for what he calls ‘cornerstones’.34 Pritchard (2014, 2015), on the other hand, believes that the way out of this predicament is to recognize that commitment to hinges is non-optional, and specifically not to be construed as ordinary belief (2014: 208). Whatever the relative merits of these strategies, this problem does not arise at all on my conception. For if I am right and ‘I know I have two hands’ does not, in ‘global’ sceptical contexts, have an epistemic sense, then ‘closure’-based sceptical arguments are impossible to construct altogether. Since if Moore’s insistence against the sceptic that he knows that he has two hands is really used to mean something like ‘I know I’m not radically mistaken’ (or ‘I know I can’t be wrong all the time’), and we substitute this into a ‘closure’-based sceptical argument, we get the following: (BIV1) If I know I have two hands (read ‘if I know that I’m not radically mistaken’), then I know I’m not a BIV. (BIV2) I don’t know I’m not a BIV. (BIV3) I don’t know I have two hands (read ‘I don’t know that I’m not radically mistaken’). But since, as everyone would agree, the BIV scenario is just colourful shorthand for the thought that I am radically mistaken, the closure-based sceptical argument now reads: (BIV1*) If I know I’m not radically mistaken, then I know I’m not radically mistaken. (BIV2*) I don’t know I’m not radically mistaken. (BIV3*) I don’t know I’m not radically mistaken. And this, of course, isn’t an argument, but a tautology followed by a repetition. So, it seems that, on my interpretation of Wittgenstein’s conception, there can be no such thing as a closure-based sceptical argument: the appearance that there might be one is generated purely by the confusion that ‘I know I have two hands’ functions like an ordinary empirical proposition in radical sceptical contexts (i.e. has an epistemic sense).35,36 If, for the reasons Wittgenstein gives, it doesn’t, however, then Pritchard’s attempt to preserve the closure principle by arguing that hinge commitments cannot be acquired through competent deduction is unnecessary (even if true). For without a closure-based sceptical argument, the closure principle simply has no application in radical (‘global’)

‘Logical’ and ‘Epistemic’ Uses  247 sceptical contexts.37,38 Given that Wittgenstein thinks that radical scepticism is an illusion, this conclusion should not surprise us.

5 Conclusion If what I have argued in this chapter is correct, then the distinction between epistemic and logical uses of ‘to know’ is a powerful tool that allows us to circumnavigate the Scylla of making concessions to radical scepticism as well as the Charybdis of denying closure. What is more, this proposal does not, contrary to competing proposals, 39 come at the price of turning hinges into a kind of assumption for which we lack evidence. For if, in ordinary circumstances where grounds for doubt are absent, I cannot coherently doubt that I have hands, then I cannot coherently assume (or ‘posit’) that I do either.40

Notes 1 See, for example, Moyal-Sharrock (2004) and Coliva (2010). 2 See, for example, Williams (1996) and Wright (2004). Pritchard’s view (2014, 2015) seems to sit in the middle of the two camps. 3 Since (1) and (3) are linked, I take this to be a good thing. 4 Also see Pritchard (2015). 5 Since there would be no practices in which they could play a role. 6 Cf. On Certainty §23. 7 Pace what Pritchard (2014) seemed to argue. Pritchard’s current view is that there is no limitation at work here, it is more that we cannot be returned to our original state of ‘epistemic innocence’ once we have engaged with the radical sceptical problem (see the final chapter of Epistemic Angst). In other words, Pritchard thinks that once we become aware of the role that hinges play in our epistemic life, we are made to feel vertigo, even after we realize there is no basis for scepticism. But if we feel vertigo that can only be because we continue to believe that there ought to be the kinds of epistemic grounds that it is logically impossible to have; otherwise we wouldn’t find their absence dizzying. Consequently, once we genuinely appreciate why such global validation is logically impossible, the feeling of vertigo should cease. 8 Compare: ‘If my name is not L.W., how can I rely on what is meant by “true” and “false”?’(OC §515). 9 A view that Stroud seemed to defend (in conversation). 10 This means that, on pain of irrationality, the sceptic cannot just bite the bullet here and ask why her demand would only be reasonable if our inability to meet it is a contingency. For one cannot coherently ‘want’ what is logically impossible (it only makes sense to want what one could, at least in principle, have). If, for example, there is no such thing as ‘round squares’, then I cannot coherently ‘want’ to find such things. If it makes no sense for people ‘to be’ prime numbers, then I cannot coherently want to ‘find’ a prime called ‘Caesar’. Hence, by parity of reasoning, if I grant that ‘global validation’ is logically impossible, then I cannot coherently continue to demand that there nevertheless be such a thing.

248 Genia Schönbaumsfeld 11 The only context where ‘I know I have two hands’ literally means I know I have two hands is in versions of the accident scenario discussed above. 12 That is to say, there is a categorical difference, for Wittgenstein, between ‘certainty’ – what we can have when we know propositions – and ‘objective certainty’ (where a doubt is ultimately unintelligible). 13 I do not take myself to have done enough to do that here, but see The Illusion of Doubt for the full picture. 14 For more on this, see my 2019 and forthcoming. 15 I take it to be a strength of my reading that the views that I attribute to Wittgenstein in OC square nicely with what he says in PI. 16 In other words, contrary to what it says in the Stanford Encyclopedia of Philosophy (see Comesaña and Klein’s entry on scepticism), hinges are not ‘posits’ – i.e. assumptions that we take to be true for pragmatic reasons. Neither is Wittgenstein advocating a form of relativism. For more on this, see The Illusion of Doubt, chapter 5. 17 For more on this, see The Illusion of Doubt, chapter 4, section III. 18 See, for example, PI §109, PI §124–128. 19 Why is no ‘final’ justification needed? Because, as Wittgenstein says, the ‘foundation- walls are carried by the whole house’ (OC §248). 20 For more on this, see my 2016 and 2019. 21 In some of the passages that Moyal-Sharrock appeals to, in order to back up her reading, Wittgenstein is talking about ordinary certainty (not objective or ‘hinge certainty’), and ordinary certainty is precisely what Wittgenstein reclaims from the sceptic (but ordinary certainty concerns matters that one could, in principle, also be uncertain about, if compelling grounds for this were given). So, for example, Moyal-Sharrock quotes: ‘If you tried to doubt everything you would not get as far as doubting anything. The game of doubting itself presupposes certainty’ (OC §115). Here Wittgenstein is talking about being certain of ordinary facts, and what he is saying is that you cannot coherently doubt all facts at the same time. The previous paragraph makes this clear: ‘If you are not certain of any fact, you cannot be certain of the meaning of your words either’ (OC §114). So, these passages do not support Moyal-Sharrock’s contention that the word ‘certainty’ in these contexts means ‘hinge certainty’. 22 This conception rules out ‘global’ doubt – the thought that we could be wrong about most things most (or all) of the time. It, of course, leaves ‘local’, ordinary doubts intact – e.g. doubts about where I left my keys, whether I locked my car, whether you love me etc. But ‘local’ doubts can, in principle (albeit not always in fact, due to various contingencies), be answered. For more on this, see The Illusion of Doubt. 23 See, for example, DeRose (1995) and Cohen (1999). 24 If hinges are not truth-assessable, empirical propositions (i.e. bipolar), then a fortiori they are not truth-evaluable either in the sense that they could be either true or false. Rather, they are logical or grammatical ‘truths’ and hence, at best, trivially true, because their negation is incoherent. 25 In this respect they are ‘propositions’ only in the sense that they express a truism – the content of the logical enabling condition, not something that could be either true or false. For more on this, see my forthcoming. 26 Wittgenstein uses the word ‘mythology’ here not to indicate that the background is ‘made up’ or ‘arbitrary’ (for then it might be false), but merely to get us away from thinking that the background must itself be a set of true or false propositions (i.e. something ‘factual’).

‘Logical’ and ‘Epistemic’ Uses

249

27 For more on this, see The Illusion of Doubt, chapter 3. I cannot do full justice to this point here. 28 Of course, we could decide to use another word instead of ‘physical object’, say, ‘sense-datum’. But if this word behaves in exactly the same way as ‘physical object’ does, then all we have done is change the sign, not the symbol (i.e. we are still operating with the same concept). See chapter 3 of The Illusion of Doubt for more on this. 29 This is why Moore’s attempt seems question-begging. 30 This, pace Williams (forthcoming) is the reason why it makes sense to use the word ‘to know’ in both cases. 31 Also compare Wright (2004) and Coliva (2014). 32 Some, of course, such as Dretske (1970, 2005) and Nozick (1981) do. For a discussion of Dretske’s strategy see chapter 1 of The Illusion of Doubt. 33 Although Williams holds that this perspective is an artefact of the Cartesian picture, and not, therefore, to be endorsed. 34 Coliva (2014, 2015) defends an ‘internal rationalism’ where assuming the truth of hinge propositions is thought to be constitutive of rationality. 35 And in this respect it is irrelevant what phrase you substitute after ‘I know’. I.e. if, in global sceptical contexts, ‘I know I have two hands’ does not function like an ordinary claim about hands, then neither would ‘I know my cat’s name’ or ‘I know that London is the capital of England’ function like ordinary claims about cats’ names or London. For the sceptic is not calling into question whether I know my cat’s name or can see my hands, which would be ordinary questions that one could respond to by saying ‘I know my cat’s name’, etc. If I say this, however, in response to a ‘global’ sceptical challenge, then this really constitutes, if it means anything, a roundabout way of insisting that I cannot be wrong about such things. 36 This is not a revisionary proposal – it does not attempt to reinterpret the ordinary sense of ‘to know’. Rather, it is because ‘to know’ means what it does that Wittgenstein believes that knowledge claims (or denials) are not fully intelligible in the radical sceptical context. In such a context, the only sense that can be made of them is therefore ‘logical’, not ‘epistemic’. I believe that this is also the significance of Wittgenstein’s notorious remark at PI §402: For this is what disputes between idealists, solipsists and realists look like. The one party attacks the normal form of expression as if they were attacking an assertion; the others defend it, as if they were stating facts recognized by every reasonable human being. On my reading, it is clear why Wittgenstein thinks one should do neither. 37 One might wonder, however, whether in contexts where it does make sense to claim that one knows that one has hands, such as in the accident scenarios described above, where one would, if one accepted disjunctivism, be in possession of a factive reason (such as seeing that one has hands) that would entail the truth of this proposition, one could not use this knowledge to refute radical scepticism. But this would be a mistake. One could not turn the fact that one knows that one has hands in such scenarios into an anti-sceptical claim à la Moore, for the attempt to do so would result in a radical change of context: one would be moving from an ordinary context, where mundane knowledge claims can be made, to a radical sceptical context, where we are trying to assert that we are not radically mistaken. If what I have argued in this chapter is right, however, then this is something we cannot do. For it implies that we have moved from an ‘epistemic’ use of ‘I know’ to a ‘logical’

250  Genia Schönbaumsfeld or grammatical one, which means that Moore would not be informing us about a true state of affairs (as he wants), but would rather be articulating a condition for the meaningfulness of our epistemic practices taken as a whole. Hence, it is not that closure fails, but rather that this principle has no application in this ‘global’ sceptical context. 38 For more on the closure principle and its relation to scepticism, see The Illusion of Doubt, chapter 1. 39 See, for example, Wright (2004) and Coliva (2014, 2015). 40 I would like to thank Annalisa Coliva, Christos Kyriacou and Duncan Pritchard for comments on an earlier draft.

References Cohen, S. (1999), ‘Contextualism, Scepticism, and the Structure of Reasons’ in J. Tomberlin (ed.), Philosophical Perspectives (Atascadero, CA: Ridgeview), 57–89. Coliva, A. (2010), Moore and Wittgenstein. Scepticism, Certainty, and Common Sense (Basingstoke: Macmillan). Coliva, A. (2014), ‘Moderatism, Transmission Failures, Closure and Humean Scepticism’ in D. Dodd and E. Zardini (eds.), Scepticism and Perceptual Justification (Oxford: Oxford University Press), 248–276. Coliva, A. (2015), Extended Rationality: A Hinge Epistemology (London: Palgrave Macmillan). Comesaña, J. and Klein, P. ‘Skepticism’ in Edward N. Zalta (ed.), The Stanford Encyclopedia of Philosophy (Winter 2019 Edition), https://plato.stanford. edu/archives/win2019/entries/skepticism/. DeRose, K. (1995), ‘Solving the Sceptical Problem’, Philosophical Review 104(1), 1–52. Dretske, F. (1970), ‘Epistemic Operators’, Journal of Philosophy 60, 1007–1023. Dretske, F. (2005), ‘The Case against Closure’ in Ernest Sosa and Matthias Steup (eds.), Contemporary Debates in Epistemology (Oxford: Blackwell), 13–25. Moyal-Sharrock. D. (2004), Understanding Wittgenstein’s On Certainty (London: Macmillan). Moyal-Sharrock, D. (forthcoming), ‘Restoring Certainty’, book symposium on The Illusion of Doubt, International Journal for the Study of Skepticism. Nozick, R. (1981), Philosophical Explanations (Oxford: Oxford University Press). Pritchard, D. (2014), ‘Entitlement and the Groundlessness of Our Believing’ in D. Dodd and E. Zardini (eds.), Scepticism and Perceptual Justification (Oxford: Oxford University Press), 190–212. Pritchard, D. (2015), Epistemic Angst (Princeton, NJ: Princeton University Press). Schönbaumsfeld, G. (2016), The Illusion of Doubt (Oxford: Oxford University Press). Schönbaumsfeld, G. (2019), ‘Was Wittgenstein an Epistemological Disjunctivist avant la lettre?’ in C. Doyle, J. Milburn and D. Pritchard (eds.), Epistemological Disjunctivism (London: Routledge), 113–130.

‘Logical’ and ‘Epistemic’ Uses  251 Schönbaumsfeld, G. (forthcoming), ‘Response to Critics’, book symposium on The Illusion of Doubt, International Journal for the Study of Skeptcism. Williams, M. (1996), Unnatural Doubts (Princeton, NJ: Princeton University Press). Williams, M. (forthcoming), ‘Knowledge without “Experience”’, book symposium on The Illusion of Doubt, International Journal for the Study of Skepticism. Wittgenstein, L. (1969), On Certainty, ed. G. E. M. Anscombe and G. H. von Wright, trans. Anscombe and D. Paul (Oxford: Blackwell). Wright, C. (2004), ‘On Epistemic Entitlement: Warrant for Nothing (and Foundations for Free?), Aristotelian Society, Supplementary Volume 78(1), 167–212.

Part V

Assertion and Knowledge Discourse

13 Assertion Compatibilism Mona Simion

1 Introduction Here are two attractive theses: The Knowledge Norm of Assertion (KNA): One’s assertion is epistemically permissible iff one knows that p. and Non-Shifty Invariantism (NSI): The truth value of knowledge claims is insensitive to practical matters. Now, theoretical considerations speak in favour of not quickly abandoning either of these two theses. First, friends and foes alike agree that NSI is the default epistemological position:1 we need to be argued out of it. Second, if knowledge is both necessary and sufficient for epistemically permissible assertion, 2 we have a very straightforward and elegant way of explaining quite a few otherwise puzzling linguistic data, such as: the paradoxical soundingness of Moorean statements of the form ‘p but I don’t know that p’; the unassertability of lottery propositions; ‘How do you know?’ challenges; (intuitively permissible) criticisms such as ‘Why didn’t you say so? You knew it all along!’ Alas, popular as they might be, for the most part, the epistemological literature of the last decade takes KNA and NSI to be incompatible. The culprit is the intuitive sensitivity of permissible assertion to practical stakes. Roughly, here is the thought behind the incompatibility claim: Since permissible assertion seems to require more warrant in high stakes than in low-stakes scenarios, we’ll have to choose between one of the following two options: We embrace KNA, but then we’ll have to adopt a view that takes knowledge/knowledge attribution to be sensitive to practical considerations. Alternatively, we stick to our non-shifty invariantist guns, but then we’ll have to give up the knowledge norm of assertion. Let’s call the view that KNA and NSI are incompatible ‘Assertion Incompatibilism.’

256  Mona Simion While the vast majority of philosophers accept Assertion Incompatibilism, there are also a few dissenting voices: some invariantists venture to explain the sensitivity data for proper assertion in a fashion that preserves both NSI and KNA. Let us dub this position ‘Assertion Compatibilism.’ One notable compatibilist strategy on the market explains why proper assertion varies with stakes in terms of what is pragmatically conveyed by the assertion in question rather than in terms of what is, strictly speaking, said (e.g. Rysiew (2001), Brown (2005), Hazlett (2009), Kyriacou (2019) and Unger (1975)). It is argued that neither knowledge (nor ‘knowledge’) nor proper assertion are sensitive to practical stakes: rather, the felt impropriety sometimes associated with knowledgeable assertions in high stakes is triggered by the generated implicatures. Let us call this view Pragmatic Compatibilism (henceforth also PC). This chapter argues that my preferred variety of Compatibilism – Functionalist Compatibilism – compares favourably to the PC competition. On Functionalist Compatibilism, what explains the intuition of shiftiness with stakes is the all-things-considered propriety rather than the epistemic propriety of assertion. Here is the game plan: Section 2 outlines the Incompatibilist challenge. In Section 3, I look at Pragmatic Compatibilism and argue that the view fails to generalize to knowledge-attribution-free assertions. Section 4 outlines my Functionalist Compatibilist response. In the last section I conclude.

2 Contextualist Cases and the Generality Objection Intuitively, proper assertion is epistemically more expensive when the stakes are higher in the sense that it appears to require more warrant. Let us dub this the Shiftiness Intuition.3 The Shiftiness Intuition is not exactly breaking news in epistemology.4 That said, it came into popularity when epistemic contextualists started using it in support of their view. Epistemic contextualism is a semantic thesis about attributions of knowledge: it holds that the meaning of ‘knows’ varies with context. It is typically supported by pairs of cases like the following ‘bank cases’ by Keith DeRose (1992: 913): BANK CASE A. My wife and I are driving home on a Friday afternoon. We plan to stop at the bank on the way home to deposit our paychecks. But as we drive past the bank, we notice that the lines inside are very long, as they often are on Friday afternoons. Although we generally like to deposit our paychecks as soon as possible, it is not especially important in this case that they be deposited right away, so I suggest that we drive straight home and deposit our paychecks on Saturday morning. My wife says, ‘Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays.’ I reply, ‘No,

Assertion Compatibilism  257 I know it’ll be open. I was just there two weeks ago on Saturday. It’s open until noon.’ BANK CASE B. My wife and I drive past the bank on a Friday afternoon, as in Case A, and notice the long lines. I again suggest that we deposit our paychecks on Saturday morning, explaining that I was at the bank on Saturday morning only two weeks ago and discovered that it was open until noon. But in this case, we have just written a very large and very important check. If our paychecks are not deposited into our checking account before Monday morning, the important check we wrote will bounce, leaving us in a very bad situation. And, of course, the bank is not open on Sunday. My wife reminds me of these facts. She then says, ‘Banks do change their hours. Do you know the bank will be open tomorrow?’ Remaining as confident as I was before that the bank will be open then, still, I reply, ‘Well, no, I don’t know. I’d better go in and make sure.’ Let’s assume that the bank will, in fact, be open on Saturday and so DeRose’s corresponding belief is true in both CASE A and CASE B. While DeRose’s attribution of knowledge that the bank is open tomorrow is intuitively permissible in CASE A, the same does not hold for CASE B. Contextualism has a straightforward explanation of these data: the truth conditions of knowledge attributions vary with context with the result that, in CASE A, DeRose’s attribution of knowledge is true, whereas in CASE B, it is false. Now, here is one important difficulty for epistemic contextualism, which DeRose himself recognizes and dubs The Generality Objection: the shiftiness intuition does not only arise for knowledge attributions. Rather, it is a much more general phenomenon. For instance, in CASE A, it is not only fine for DeRose to assert that he knows that the bank will be open tomorrow, it would also be fine for him to assert that the bank will be open tomorrow. In contrast, in CASE B, it would not only seem wrong for DeRose to assert that he knows that the bank will be open tomorrow, but also to assert that the bank will be open tomorrow (DeRose 2002: 177). To see this more clearly, consider the following cases: ASPIRIN1. You remember having bought aspirin last month. You are heading together with your sister towards your place for dinner, and she lets you know she has a minor headache. Your sister asks you: ‘Do you have aspirin at home, or should we go to the pharmacy?’ You flat out assert: ‘Don’t worry, I have aspirin at home, I remember having bought some.’ ASPIRIN2. You remember having bought aspirin last month. Your sister’s two years old baby is having a fever, and needs an aspirin

258  Mona Simion as soon as possible. Your sister asks you: ‘Do you have aspirin at home, or should we go to the pharmacy?’ You give the matter a bit of thought, and answer: ‘Well, let’s drop by the pharmacy, just in case.’ Intuitively, in ASPIRIN1, but not in ASPIRIN2, it is appropriate to flat out assert that you have aspirin at home. What these cases suggest is that the shiftiness intuition is not limited to cases of knowledge attributions. As a result, the worry is that epistemic contextualism, in virtue of being a view about the semantics of ‘know,’ won’t give us the right explanation of the shiftiness intuition after all. In order to address this worry, contextualists have appealed to the knowledge norm of assertion (KNA).5 Roughly, here is the thought: according to KNA, one’s assertion that p is epistemically proper if and only if one knows that p. If that is the case, however, it follows that the standards for knowledge go hand in hand with the standards for proper assertion. But now it looks as though epistemic contextualism is well positioned to avoid the generality objection. After all, the objection suggests that the standards for proper assertion vary in general (that is, not just in cases featuring knowledge attributions). But if KNA and contextualism are both true, this is just what we’d expect anyway. In fact, KNA and contextualism in conjunction promise to offer an appealing explanation of the relevant data. To see this, consider first the variation of CASE A in which DeRose asserts that the bank is open. Suppose, plausibly enough, that the bar for ‘knowledge’ is low here, low enough that DeRose ‘knows’ what he asserts and so his assertion comes out proper. In contrast, suppose DeRose were to assert that the bank is open in a variation of CASE B. Suppose, again plausibly, that the bar for ‘knowledge’ is high here, high enough that DeRose doesn’t ‘know’ what he asserts and so his assertion would come out improper. In this way, KNA and contextualism together can offer an appealing solution to the generality problem. In fact, there is reason to think that DeRose can now turn the tables on the generality objector and argue that KNA demands an account of knowledge according to which whether one knows is sensitive to practical stakes: What of the advocate of the knowledge account of assertion who does not accept contextualism? Such a character is in serious trouble. Given invariantism about knowledge, the knowledge account of assertion is an untenable attempt to rest a madly swaying distinction upon a stubbornly fixed foundation. […] The knowledge account of assertion demands a contextualist account of knowledge and is simply incredible without it. (2002: 182)

Assertion Compatibilism  259 If DeRose is right, KNA is incompatible with Non-Shifty Invariantism: since the standards for knowledge and permissible assertion co-vary, and since the latter are shifty with stakes, it seems to follow that the standards for knowledge will be shifty with stakes as well.

3 Pragmatic Compatibilism, Modest and Sceptical According to Pragmatic Compatibilists, the source of variability in contextualist cases pertains to what is pragmatically conveyed by the assertion in question rather than by what is, strictly speaking, said (e.g. Brown 2005; Hazlett 2009; Kyriacou 2019; Rysiew 2001; Unger 1975). Non-shifty invariantism comes in two different flavours: sceptical and moderate. The sceptical invariantist (e.g. Kyriacou 2019; Schaffer 2004; Unger 1975) claims that the semantic value of the word ‘know’ is such that all or nearly all ordinary positive knowledge ascriptions of the form ‘S knows that p’ are false. The moderate invariantist claims that the semantic value of ‘know’ is such that many of the positive knowledge ascriptions that we make in daily life are true. Both views are Non-Shifty varieties of invariantism. Correspondingly, Pragmatic Compatibilism comes in two different flavours as well. On Moderate Pragmatic Compatibilism (Brown 2005; Hazlett 2009; Rysiew 2001, henceforth also MPC), in both CASE A and CASE B, DeRose does, in fact, know that the bank will be open on Saturday. However, given the presumption of relevance and informativeness, in CASE B, saying that he knows pragmatically conveys that he is able to eliminate all contextually salient error possibilities – such as, for instance, the possibility that the bank changed its hours in the last two weeks. But of course, this is just an error possibility that DeRose cannot eliminate. In this way, then, DeRose’s self-attribution of knowledge would carry a false implicature. And it is just this false implicature that, according to Pragmatic Compatibilists, explains why DeRose’s selfattribution of knowledge would seem wrong. On this view, Non-shifty Invariantism is perfectly compatible with KNA: DeRose knows in both CASE A and CASE B and is, therefore, in a good enough epistemic position to assert that the bank will be open. Sceptical Invariantism (Kyriacou 2019; Schaffer 2004; Unger 1975) traditionally has a somewhat harder time accommodating the shiftiness intuition without giving up on KNA. According to this view, speakers ascribing knowledge speak non-literally, i.e. the speaker meaning does not coincide with the meaning of the sentence uttered. What varies with context, according to these philosophers, is not what the speaker literally says, i.e. the truth-conditional content of the sentence used, but what she means by her use of that sentence. According to Sceptical Invariantism, by uttering ‘x knows that p,’ what the speaker says is ‘x can rule out every conceptually possible alternative to p,’ while what she means is

260  Mona Simion ‘x can rule out every relevant alternative to p.’ As such, truth conditions come apart from assertability conditions; KNA is thus not easily vindicated by this view. Recently, though, Christos Kyriacou (2019) argues for a Sceptical Pragmatic Compatibilism (henceforth SPC): on this deflationary view, although strictly speaking we do not know most of the things we take ourselves to know, KNA concerns a weaker, everyday notion of knowledge – henceforth weak-knowledge that is useful to have, and that we employ in our everyday affairs. As such, on this view, it is this weaker notion that is at stake in KNA, and it is the very same notion that is responsible, just like in the case of Modest Pragmatic Compatibilism, for the implicatures allegedly to blame for generating the shiftiness intuition. A lot of ink has been spilled on whether the Pragmatic Compatibilist account of the bank cases is ultimately satisfactory for accounting of the shiftiness of proper knowledge attributions.6 I will not address this issue here. In contrast, what I will focus on is whether Pragmatic Compatibilism can explain the shiftiness intuition in its full generality. After all, we have seen that this intuition is not restricted to knowledge attributions, but rather it also arises in cases of assertions of propositions that do not feature epistemic concepts. By way of illustration, recall once more the variation of CASE B in which were DeRose to assert simply that the bank is open tomorrow, his assertion would be improper. Note that Pragmatic Compatibilists will have a hard time extending their account of the original CASE B to this variation. The thought that the knowledge attribution, in conjunction with contextually salient error possibilities, generates an implicature to the effect that DeRose can eliminate the salient error possibilities has little traction here. After all, there is no knowledge attribution being made. To the best of my knowledge, Hazlett (2009: 31) is the only champion of Pragmatic Compatibilism that has addressed this issue in print, on behalf of the Moderate variety of the view, so in what follows I will first examine his solution in detail. Further on, though, I will also look at one more possible response on behalf of MPC, as well as a response on behalf of SPC – the sceptical version of the view. Here is Hazlett: say that, in the relevant variation of CASE B, DeRose were to assert that the bank is open tomorrow. Now, by the Gricean maxim of Quality (alternatively, if knowledge is the norm of assertion), this generates the implicature that DeRose knows that the bank is open tomorrow. According to Hazlett, this, in turn, will imply that the speaker can eliminate all the contextually salient error possibilities. For instance, in CASE B, were DeRose to assert that the bank will be open tomorrow, his assertion would generate the implicature that he knows this, which, in turn, would generate the implicature that he can eliminate the contextually salient error possibility that the bank changed its hours in the last two weeks. Since DeRose can do no such thing, the assertion does

Assertion Compatibilism  261 generate a false implicature, i.e. in virtue of generating an implicature that itself has a false implicature. 3.1 Shiftiness without Error Possibilities One immediate worry with this line is that it is easy to come up with cases that are just like the variation of CASE B we have been considering in which there are no contextually salient error possibilities being tabled. The ASPIRIN cases we have been looking at are precisely like that. In these cases, in contrast to the Bank Cases, no error possibilities are raised; still, the Shifty Intuition survives. Perhaps the Pragmatic Compatibilist could respond that the set of contextually salient error possibilities need not be made explicit. It will do if there is an implicit set of such possibilities. For instance, in ASPIRIN2, the Pragmatic Compatibilist could perhaps claim that the possibility that you misremember having bought aspirin becomes implicitly salient. The trouble with “going implicit” is that the solution threatens to be too easy. For this move to enjoy any plausibility whatsoever, we will need an independently plausible account of how the set of implicitly salient error possibilities is generated. Given that what we are dealing with is a context-sensitive phenomenon, it is quite plausible that it is facts about the participants to the conversation that generate the set, such as their interests, what they are attending to, etc. It is hard to deny that these facts vary from one speaker to the next, which begs the question as to just how a single set is settled on. One obvious proposal is that it is the lowest common denominator, as it were, i.e. the intersection of the set of error possibilities that each speaker would generate individually. However, this proposal is tricky because it is just not clear why all the error possibilities the Pragmatic Compatibilist needs will always be generated. For instance, why should we think that, in ASPIRIN2, the possibility of misremembering should always be generated as implicitly salient at this context? After all, since your sister has no idea when and if you bought aspirin – for all she knows, it may well be that you did so that very morning –, considerations pertaining to the reliability of your memory might not be particularly salient for her. Note that if it is possible that the possibility of misremembering is not part of the common denominator set, it looks as though the Pragmatic Compatibilist account of the data is in trouble. Another obvious proposal is that it is the highest common denominator, as it were, i.e. the union of the sets of error possibilities that each speaker would generate individually. This proposal is also tricky because it is not clear why there couldn’t be a participant to a low-stakes conversation whose interests, etc. unduly raise the bar for ‘knowledge.’ But given all this, the Pragmatic Compatibilist who wants to solve the problem posed by cases like ASPIRIN2 shoulders a substantive

262  Mona Simion explanatory burden. Unless this burden has been discharged, we have every reason to remain sceptical about the Pragmatic Compatibilist line here. 3.2 Shiftiness without Stakes Cases like ASPIRIN2 suggest that contextually salient error possibilities are not necessary for triggering the unassertability intuition. Even if it turns out that this problem can be circumvented, perhaps along the lines just outlined, the Pragmatic Compatibilist faces a further problem: they don’t seem to be sufficient for doing the job either. That is, it looks as though, at least in some cases, tabling error possibilities will not do the trick, at least not in the absence of accompanying high stakes (see also Brown 2005 for an acknowledgement of this). If that is the case, however, it starts looking as though proper assertability does not have all that much to do with salient error possibilities to begin with. Rather, it seems more and more plausible that the only thing that matters are variations in stakes. Thus consider the following variation of ASPIRIN1: ASPIRIN1*. You remember having bought aspirin last month. You are heading together with your sister towards your place for dinner, and she lets you know she has a minor headache. You flat out assert: ‘Don’t worry, I have aspirin at home, I remember having bought some.’ Your sister asks: “Maybe you’re mistaken; you do sometimes misremember things,” she remarks. “I have aspirin at home,” I reply. “All right,” she says. I take the intuition here to be that you and your sister are right to put the matter to rest. If that is true, however, tabled error possibilities by themselves – that is, in absence of raised stakes – won’t seem to do the work Pragmatic Compatibilists takes them to do. This, of course, should hardly come as a surprise: just because sceptical worries are ‘in the air’ while I write an epistemology paper, for instance, it hardly follows I do not know and cannot assert that there is a computer in front of me. More is needed to trigger unassertability. 3.3 Implicature-Generating Implicatures and Sceptical Alternatives We still have not reached the end of the obstacles the Pragmatic Compatibilist will have to overcome. To see this, note that she is wheeling in some pretty heavy theoretical machinery, i.e. the idea of an implicature-generating implicature: for her account to work, the implicature that the speaker knows what she says (generated by KNA) in itself

Assertion Compatibilism  263 must be able to generate the further implicature that she can eliminate all contextually salient error possibilities. Again, there is an explanatory burden to be discharged. To begin with, one might wonder whether there is such a thing as implicatures generating implicatures in the first place. Moreover, even if it turns out that the phenomenon exists, it would be good to know whether what we are dealing with here is a fully general phenomenon, or whether it is restricted to particular types of implicatures. And then there is the issue of just how implicatures generate implicatures. And finally, the account had better work, at least for the type of implicature that the Pragmatic Compatibilist wants to invoke. In this way, the burden on the Pragmatic Compatibilist shoulders gets weightier and weightier. Here is a more direct problem for the Pragmatic Compatibilist. It is widely agreed that implicatures must be calculable. And, of course, whether a given implicature is calculable will depend on the cognitive capacities of the agents who are facing the calculation task. But given that this is so, it is surely possible for there to be communities of speakers that are only able to generate simple implicatures, and not implicatures of implicatures. It is also surely possible for some such communities (i) to operate the concept of knowledge and have a word expressing it in their language and (ii) to have variations in stakes. If so, we can generate analogues of DeRose’s bank cases for members of these communities. Note that the Pragmatic Compatibilist account will work just fine for the analogues of DeRose’s original bank cases. After all, all the Pragmatic Compatibilist needs here is a simple implicature. However, the envisaged account for the generalized cases like ASPIRIN2 will just not work. After all, according to the Pragmatic Compatibilist line, what’s going on here is an implicature of an implicature. Since members of the community cannot calculate implicatures of implicatures and since implicatures must be calculable, it follows that there is no such thing as an implicature of an implicature for this community. By the same token, the Pragmatic Compatibilist account is bound to fail here. One could think that Sceptical Pragmatic Compatibilism will do better here, since they can have their cake and eat it too: it is open to these philosophers to argue that weak-knowledge, but not knowledge proper, is pragmatically encroached upon (i.e. a practically shifty relevant alternatives view of weak-knowledge, but not of knowledge, is correct). In turn, this view could be plausibly motivated by the thought that weak-knowledge is an entity that we only employ for practical purposes anyway. The problem with a view like this, however, is that it’s not clear that it leaves room for any philosophically substantive – rather than merely terminological – debate between Sceptical Pragmatic Compatibilism and shifty incompatibilist views of knowledge, such as Pragmatic Encroachment. After all, it looks as though what Pragmatic Encroachment calls ‘knowledge,’ the champion of sceptical

264  Mona Simion pragmatic compatibilism calls ‘weak-knowledge’ (in that both of these phenomena are pragmatically encroached). Also, the ‘real knowledge’ by SPC’s lights looks suspiciously similar to what everybody else in the literature calls ‘epistemic certainty.’ If all this is so, the debate between Compatibilism and Incompatibilism becomes merely terminological – which, in turn, I take it, will not be a satisfactory result for either side. 3.4 Shiftiness without Error Possibilities Last, I would want to look at an alternative possible reply on behalf of MPC that does not rely on the problematic notion of implicaturegenerating-implicature,7 and which, one might think, is closer to the spirit of MPC to begin with. What seems key for people like Brown (2005) and Rysiew (2001) is that in the relevant cases (e.g. DeRose’s Case B) what’s practically important is that the subject be in a very good/strong epistemic position vis-à-vis the proposition in question. Due to considerations of relevance (it’s alleged), a positive claim to knowledge in Case B would then ‘pragmatically impart’ that S’s epistemic position is good enough relative to those heightened standards, and so that S can rule out any salient alternatives. But this doesn’t clearly mean that some such alternatives must be mentioned in advance for the former proposition, which is plausibly false, to be communicated. Why would it be improper for DeRose to assert, in B, simply that the bank is open tomorrow? Because he’d be representing himself (as per the maxim of Quality) as having ‘adequate evidence,’ where such adequacy is (as per the maxim of Relevance) is context-, and so stakes-, sensitive. Given this, in asserting that p, DeRose would be pragmatically communicating that his epistemic position vis-àvis p is much stronger than it plausibly is. The worry for this reply is that the notion of ‘contextually sufficient’ adequate evidence is itself typed: it can be sufficient, e.g. prudentially – i.e. for practical purposes; morally – i.e. for moral purposes; or epistemically – i.e. for epistemic purposes. What the champion of MPC needs, and is missing, is an argument that it is the former and not the latter sense of enough that is at stake here – on pain of either collapsing into pragmatic encroachment (if the former obtains), or begging the question against it (if no argument is given). Furthermore, as I am about to argue, once we recognize and have an argument in favour of prudential normativity being at work in these cases, we no longer need to rely on the pragmatics of language to do the work in explaining the shiftiness intuition.

Assertion Compatibilism  265

4  Functionalist Compatibilism In previous work (e.g. Simion 2018, 2019, 2020), I have defended a Compatibilist position that I dub ‘Functionalist Compatibilism.’ According to this account, once we distinguish between assertion’s epistemic function on one hand, and its prudential function on the other hand, the incompatibility worry sourced in the Shiftiness Intuition dissipates: KNA and NSI (in both of its incarnations) are perfectly compatible. 4.1 The Functionalist Framework Here is the view, in a nutshell: The main epistemic function of assertion is to generate testimonial knowledge in hearers. In the vast majority of the cases, assertion will generate testimonial knowledge that p if and only if the speaker knows that p.8 Thus, in virtue of the function of assertion of generating knowledge in hearers, one is in a good enough epistemic position to make an epistemically proper assertion that p if and only if one knows that p. In turn, whether one knows (or ‘knows’) that p is insensitive to practical matters. Crucially, on this account, what KNA claims is mere epistemic propriety. The way to ascertain whether the requirements at work in one case or another are genuinely epistemic requirements is by looking at the function that is plausibly being served. In traits, artefacts and actions alike, only functions of a type T generate norms of type T; there is such a thing as a biologically properly functioning heart – associated with its biologic function of pumping blood; a practically properly functioning can opener – borne out by its practical function of opening cans; and an aesthetically proper way to engage in dancing waltz – aimed at serving the aesthetic function of exemplifying beauty. We should expect, then, that if there is such a thing as epistemic propriety of assertion, it will somehow be associated with the epistemic function it serves. Now, what epistemic goods is assertion meant to deliver? The plausible answer is that, characteristically, assertions will aim at generating testimonial knowledge in the audience.9 Due to our physical and cognitive limitations, a lot of the knowledge we have is testimonial; thus, assertion is one of our main epistemic vehicles. Functional types in general come with norms of proper functioning. More specifically, these norms specify conditions that will ensure that the relevant function is served reliably enough. Note that assertion will, with very few exceptions, generate knowledge in the hearer if and only if the speaker also knows; it should be pretty clear then that being knowledgeable is the most reliable way for assertion to fulfil its epistemic

266  Mona Simion function. By the same token, there is reason to think that assertion is governed by a knowledge norm. Now, importantly, notice that a given trait/artefact/act can have several functions simultaneously, even several functions of different types; take, for instance, the functions served by food for humans. One important such function will surely be a biological one, a nourishment function. Plausibly enough, though, on top of this, food also serves an aesthetic function, that of generating pleasant gustatory experiences. Now, normally, the aesthetic function complements the nourishing function. It serves survival by increasing the probability of us ingesting nourishing substances. This need not be the case, however; there can be situations where the two functions come in conflict, at which point the more stringent requirement will take precedence. In other words, when there is a conflict between the normative requirements associated with two functions, one requirement may override the other and dictate what’s the all-things-considered good to observe. By way of illustration, think of a case where I am on a desert island and all I can eat in order to stay alive are my boots; surely, against my aesthetic well-being, that is exactly what I should do, all-things-considered. To repeat, the thought here is that we get overriding from conflicts between the normative requirements associated with functions. We can now take this idea and put it to use for our epistemological purposes. To see how, note first that it is highly plausible that the epistemic function is merely one of the many functions served by assertion. One other very important function of assertion, as with action in general, will be a prudential one, serving our practical ends. Again, just like in the case of food, the epistemic function will, in most cases, complement this prudential function. Generating testimonial knowledge in one’s hearer with regard to an imminent threat, or about the whereabouts of resources are paradigm cases. However, again, this need not be the case. The two functions can also come into conflict. For instance, even if one knows that one’s boss is bald, it may not be polite, prudent or relevant to point this out to him ((Brown 2010: 550): here, the prudential function comes in conflict with the epistemic one. What’s more, it’s also plausible that the prudential function overrides the epistemic. The result is that although it is epistemically permissible to assert that one’s boss is bald to his face, it is not all things considered permissible for one to do so. One question that might arise at this stage is:10 if the epistemic function of assertion – that of generating testimonial knowledge – is its main function – which, plausibly, it is – why is it that norms generated by prudential normativity – so by a presumably secondary, prudential function of assertion – override the epistemic ones? Shouldn’t norms generated by its main function be stronger? Two things about this. First, the short answer is ‘no.’ In general, norms generated by the main function of a trait or artefact, etc. need

Assertion Compatibilism  267 not override norms generated by secondary functions. Indeed, prudential and moral norms will plausibly have a high disposition to dictate all-things-considered-permissibility in the normative landscape, whether they are generated by main or secondary functions. Take, for instance, your washing machine: analytically, I take it, its main function is to wash laundry. If, however, laundry washing is dangerous to your life for some reason, the all-things-considered normative picture will be such that it is impermissible for your washing machine to wash laundry. Second, compatibly, notice that a washing machine that fails at washing laundry is, importantly, a bad token of its type, in a way in which a washing machine that fails to save our life is not. That is because, even though main functions need not generate all-things-considered prescriptive demands – i.e. concerning what ought-to-be-done – they do generate all-things-considered evaluative demands – i.e. concerning ought-to-bes. Ought-to-bes correspond to attributive goodness (a la Geach 1956): a washing machine that fails to wash but saves your life thereby is bad attributively, i.e. as a token of its type: it fails to be how it ought-to-be, although it does what it ought to do. 4.2 Modest and Sceptical Functionalist Compatibilism The phenomenon of conflict and overriding is precisely what explains, in my view, the unassertability intuition in high stakes scenarios, no matter what flavour of Non-Shifty Invariantism one prefers – i.e. be it modest or sceptical. Since I myself have a mild preference for the former, I will start with the modest variety thereof. In a nutshell, the account predicts that, just like in the case of the washing machine, in the case of assertion, when the epistemic norm is overridden by practical considerations as in the Bank Cases, what one needs to do is not assert, even if one knows. Compatibly, though, if one were to (all-things-considered impermissibly) assert, one’s assertion would be attributively good, i.e. a good token of its type (see also Kelp (2016) and Simion, Kelp and Ghijsen (2016)). On a modest invariantist variety of my Functionalism, then, in ASPIRIN2, strictly epistemically speaking, if you have memorial knowledge that you have aspirin at home, you are permitted to assert. You are in a position to make an assertion that can fulfil its epistemic function of generating testimonial knowledge in your hearer. However, prudential constraints concerning your nephew’s health come into conflict with these epistemic requirements. They override the epistemic constraint and drive the degree of warrant needed for all-things-considered permissible assertion up to a point the speaker’s epistemic support does not reach. As a result, it will be all-things-considered impermissible for the speaker to assert in such a case, which explains why her assertion would seem improper to us.

268  Mona Simion Similarly, DeRose does know that the bank will be open on Saturday in Bank Case B if he knows in Bank Case A. It is, though, prudentially and all-things-considered inappropriate for him to say so. What about a sceptical variety of Functionalist Compatibilism? How would such a view go? I will start, once again, with my preferred variety of this view: on this account, standards for knowledge are very high – although not non-achievable. We can know from super-safe processes like deduction or perception of nearby large dry goods, but that’s about it: knowledge from flimsy processes, such as long-term memory or testimony, is not easy to get. Notice that the view is not as far fetched as sceptical views are usually taken to be – in that it is much more in line with folk intuitions about knowledge: often ordinary folk are reticent to ascribe knowledge based on mere memory or testimony (see, e.g., Lackey 2011; Simion 2016 for discussion). To see it for testimony, think of a modified version of Lackey’s (2008) case of the ‘Chicago Visitor’: upon arriving at the Chicago station, Morris gets off the train and asks a passer-by for directions to the Sears Tower. The passer-by, knowledgeable about the geography of Chicago, tells Morris that the Sears Tower is to the left. Lackey’s reported intuition (and mine) is that Morris thereby comes to know that the Sears Tower is to the left. The sceptical invariantist – and many undergraduate students, for that matter – will deny this claim. One way to see the sceptical point here is to imagine that, while on his way to the Sears Tower, a tourist mistakenly takes Morris to be a local and asks him which way to go to the Sears Tower, Many have the intuition that Morris should not outright assert ‘To the left!,’ but rather hedge along the lines of ‘I’m not from around here myself, but a local told me it’s to the left.’ This intuition is water to the sceptical invariantist’s mill. On this view, De Rose does not know that the bank will be open on Saturday in either of the Bank Cases; after all, he’s basing his corresponding assertion on a non-knowledge conferring memory from two weeks ago. Epistemically, then, he should not assert. That being said, practical considerations override the epistemic norm in both cases: in both cases, he is under practical pressure from his wife’s question and the proximity of the bank to decide one way or another – to stop at the bank or not. In the first case, it’s prudentially best if he does not stop – which explains the prudential permissibility of the corresponding assertion. In the second case, it is prudentially best if he stops – which is why it is prudentially impermissible to assert outright that the bank will be open tomorrow. The last option that needs to be discussed is how Functionalist Compatibilism will deal with a strong version of sceptical invariantism, according to which we pretty much have no knowledge at all. Importantly for our purposes here, note that the super-sceptical variety of this view can also afford the above explanation of the intuitive shiftiness in the

Assertion Compatibilism  269 bank cases, in terms of the prudential normativity at work in the contextualist cases. The extra problem, though, with this type of view, is that having a knowledge norm of assertion does not make sense to begin with on super-scepticism, since there is no such thing as knowledge to be had and to base one’s assertions on. My Functionalism cannot solve this problem; neither is it, though, a problem that it is called to solve. The framework put forth in this chapter merely concerns itself with rendering KNA and non-shifty invariantism compatible in the presence of the shiftiness intuition. What this latter problem makes clear, however, is that super-sceptical varieties of invariantism are not compatible with KNA on different grounds, independent of the shiftiness intuition (see also Hawthorne 2004). Solving this problem falls beyond the scope and the ambition of this chapter.

5 Conclusion This chapter has defended Assertion Compatibilism: a biconditional norm of assertion need not imply a view about knowledge (or ‘knowledge’) that takes it to be sensitive to practical matters. I have argued here that there are problems with the extant compatibilist views on the market. Pragmatic Compatibilism – be it of the modest or sceptical variety –, attempting to explain the Shiftiness Intuition in Gricean terms, was shown to face the generality objection: that is, it fails to adequately account for variations in assertability in cases in which no epistemic concepts are deployed and no error possibilities are on the table. I have argued that my favourite compatibilist view, Functionalist Compatibilism, compares favourably with the competition in that it succeeds in its compatibilist ambition. This view makes use of normative overriding for explaining the intuitive variability of proper assertion with stakes. What varies with practical matters is all-things-considered propriety: epistemic propriety, together with the epistemic standard at stake, are invariant.

Notes 1 “Anyone who has even passing knowledge of analytic epistemology in the last fifty or so years knows that moderate invariantism is the orthodoxy. It is the view to beat” (Fantl and McGrath 2009: 37). 2 Champions of KNA include DeRose (e.g. 2002), Hawthorne (e.g. 2004), Kelp (e.g. 2016), Simion (e.g. 2016, 2019, 2020), Turri (2011), Unger (1975) and Williamson (1996, 2000). 3 To my knowledge, the term was coined in Fantl and McGrath (2012). 4 Austin (1979: 180), for instance, observes that, while in normal contexts the fact that your hat is in the hall seems to be good enough reason for me to say that you are in, when a lot hinges on it, I would be quite reticent to do the same. 5 The locus classicus for the defence of the necessity claim involved in KNA is Williamson (2000). For support for the sufficiency claim, see Simion (2016).

270 Mona Simion 6 See, e.g., (DeRose 2009). 7 I would like to thank very much for this suggestion to an anonymous referee who reviewed my manuscript ‘Shifty Speech and Independent Thought’ for Oxford University Press. 8 See Lackey (2008) for exceptions. 9 For arguments to this effect, see, e.g., Simion (2018, 2020), Goldberg (2015), Kelp (2016), and Reynolds (2002). See Graham (2010) for a defence of a true belief generation function of assertion. 10 Thanks a lot to James Beebe for pressing me on this.

References Austin, J. (1979). Wittgenstein’s Criterial Semantics. Oxford: Oxford University Press. Brown, J. (2005). Williamson on Luminosity and Contextualism. The Philosophical Quarterly, 55(219): 319–327. Brown, J. (2010), Knowledge and Assertion. Philosophy and Phenomenological Research, 81: 549–566. DeRose, K. (1992). Contextualism and Knowledge Attributions. Philosophy and Phenomenological Research, 52(4), 913–929. DeRose, K. (2002). Assertion, Knowledge, and Context. Philosophical Review, 111(2): 167–203. DeRose, K. (2009). The Case for Contextualism: Knowledge, Skepticism, and Context. Vol. 1, Oxford: Clarendon Press. Fantl, J. and McGrath, M. (2009). Knowledge in an Uncertain World. Oxford University Press. Fantl, J. and McGrath, M. (2012). Arguing for Shifty Epistemology. In J. Brown and M. Gerken (eds.), Knowledge Ascriptions. Oxford: Oxford University Press, 55–74. Geach, P. (1956). Good and Evil. Analysis 17 (2):33 - 42. Goldberg, Sanford C. (2015). Assertion: On the Philosophical Significance of Assertoric Speech. Oxford: Oxford University Press. Graham P.J. (2010). Testimonial Entitlement and the Function of Comprehension. In Duncan Pritchard, Alan Millar and Adrian Haddock (eds.), Social Epistemology. Oxford: Oxford University Press, 148–174. Hawthorne, J. (2004). Knowledge and Lotteries. Oxford: Oxford University Press. Hazlett, A. (2009). Knowledge and Conversation. Philosophy and Phenomenological Research, 78(3): 591–620. Kelp, C. (2016). Assertion: A Function First Account. Nous, Online First. Kyriacou, C. (2019). Assertion and Practical Reasoning, Fallibilism and Pragmatic Skepticism. Acta Analytica, Online First. Lackey, J. (2008). Learning from Words. Oxford: Oxford University Press. Lackey, J. (2011). Assertion and Isolated Second-Hand Knowledge. In J. Brown and H. Cappelen (eds.), Assertion. Oxford: Oxford University Press: 251–275. Reynolds, S. L. (2002). Testimony, Knowledge, and Epistemic Goals. Philosophical Studies, 110 (2): 139 –161. Rysiew, P. (2001). The Context-Sensitivity of Knowledge Attributions. Noûs, 35(4): 477–514.

Assertion Compatibilism  271 Schaffer, J. (2004): Skepticism, Contextualism, and Discrimination, Philosophy and Phenomenological Research, 69: 138–155. Simion, M. (2016). Assertion: Knowledge Is Enough. Synthese, 193(10): 3041–3056. Simion, M. (2018). Assertion: The Context Shiftiness Dilemma. Mind & Language, 34: 503–517.Online. Simion, M. (2019). Saying and Believing: The Norm Commonality Assumption. Philosophical Studies, 176(8): 1951–1966. Online. Simion, M. (2020). Shifty Speech and Independent Thought. Manuscript. Simion, M., Kelp, C. and Ghijsen, H. (2016). Norms of Belief. Philosophical Issues, A Supplement to Nous: Epistemology and Mind Issue (Jack Lyons and Chris Kelp, eds.), 26(1): 375–392. Turri, J. (2011). The Express Knowledge Account of Assertion. Australasian Journal of Philosophy, 89 (1): 37–45. Unger, P. (1975). Ignorance: A Case for Scepticism. Oxford: Clarendon. Williamson, T. (1996). Knowing and Asserting. Philosophical Review, 105(4): 489–523. Williamson, T. (2000). Knowledge and Its limits. Oxford: Oxford University Press.

14 Knowledge and Loose Talk Alexander Dinges

1 Introduction Skeptical invariantists maintain that the expression “knows” invariably expresses an epistemically extremely demanding relation. This leads to an immediate challenge. The knowledge relation will hardly if ever be satisfied. Consequently, we can rarely if ever apply “knows” truly. This seems troublesome because we make knowledge ascriptions all the time and because, intuitively, this is perfectly all right.1 A prominent response strategy on behalf of skeptical invariantism appeals to loose talk (see, e.g., Conee, 2005: 52–53; Davis, 2007; Fantl and McGrath, 2009: 194; BonJour, 2010: 73; Dinges, 2016; Kyriacou, 2017, 2020). Consider the following sentences: 1 Hannah arrived at 3pm. 2 Hannah is 6 foot 2. 3 The table is flat. Utterances of (1) and (2) can be proper even if Hannah didn’t arrive at precisely 3pm and isn’t exactly 6 foot 2, and utterances of (3) can be proper even if the table isn’t perfectly flat. Skeptical invariantists can argue that knowledge ascriptions work just like that. We accept knowledge ascriptions when we don’t strictly know things for just the same reasons for which we accept, e.g., ascriptions of flatness to objects that are less than perfectly flat. We speak loosely in each case. This paper argues that such appeals to loose talk on behalf of skeptical invariantism fail. It also presents a closely related, more promising response to the challenge above. Skeptical invariantism comes out as an account of knowledge ascriptions that combines the virtues of invariantism and contextualism. It posits a stable, context-independent knowledge relation that can unify the theoretical roles that knowledge plays in philosophy, psychology, legal theory, ethics, etc. At the same time, it accommodates the flexible, relatively weak demands on knowledge ascriptions in ordinary discourse. It achieves all that in a semantically and

Knowledge and Loose Talk  273 pragmatically precise, yet parsimonious framework that employs only familiar mechanisms that we have to posit for independent purposes anyway. For starters, I introduce some formalism and terminology (Section 2). Then I look at some striking parallels between loose talk and knowledge ascriptions. They motivate the appeal to loose talk on behalf of skeptical invariantism, and they indicate some desiderata for any plausible account of knowledge ascriptions (Section 3). I then turn to familiar theories of loose talk and their application to the case of “knows.” Numerous accounts of loose talk have been suggested in the literature. On many of these accounts, loose talk is a semantic phenomenon, where the semantic value of, e.g., “3pm” varies with the context of utterance, denoting smaller or larger ranges of points in time (e.g., Krifka, 2007; DeRose, 2012: 714–717; Solt, 2014). If any of these accounts is correct, the appeal to loose talk on behalf of skeptical invariantism is self-refuting. After all, skeptical invariantists are invariantists and hence they cannot countenance semantic variability in “knows.”2 Meanwhile, I show that familiar pragmatic accounts of loose talk won’t help the skeptical invariantist either, because they make false predictions when applied to “knows.” The upshot is that if any extant account of loose talk is on the right track, loose talk is of no avail to skeptical invariantists (Section 4). Hoek (2018, 2019) specifically explains loose talk as a type of what he calls conversational exculpature. Following the results from the previous section, knowledge ascriptions don’t instantiate this type of conversational exculpature, but they may still instantiate the general phenomenon. Indeed, I argue that such an approach looks promising if we adopt independently motivated assumptions about the dynamics of pragmatic presuppositions familiar from Blome-Tillmann’s (2014) works on epistemic contextualism (Section 5).

2 A Little Bit of Formalism In presenting the subsequent material, it will be useful to work in a regimented system, and I will assume a simple intensional semantics for this purpose. Thus, I will assume that sentences express propositions represented as sets of possible worlds, which result from composing semantic values of atomic expressions. Consider, for instance, the following semantic clauses for “knows” derived from familiar attempts to analyze knowledge. knows = {〈S, p, w〉  | At w, i S has a true belief with the content p, and ii This belief is justified/safe/reliably formed}.

274  Alexander Dinges In the assumed semantic framework, these semantic clauses determine the propositions expressed by a sentence of the form “S knows that p” as follows: S knows that p = { w |  〈 S , that p , w〉 ∈ knows

}

In the given system, we can clarify skeptical invariantism. This view is compatible in spirit with any of the semantic clauses just proposed for “knows.” We just have to interpret the respective second condition as being invariably maximally demanding on the respective dimension. In what follows, I will understand skeptical invariantism in this way, i.e., as the view that the semantic value of “knows” invariably entails the maximum of some pertinent epistemic quantity, like justification, safety, reliability, etc. It will be useful in what follows to have a term that, by definition, expresses the demanding relation that “knows” expresses according to skeptical invariantism, and I will use “knowsS” for this purpose. Here are some possible semantic clauses for “knowsS.” knowsS = {〈S, p, w 〉  | At w, i S has a true belief with the content p, and ii This belief is perfectly justified/safe across every possible world/100% reliably formed}. It will also be useful to have a term that stipulatively denotes the kind of relation that normally holds between a subject and a proposition when a knowledge ascription seems appropriate. This relation is much less demanding than knowledgeS , and it corresponds to the semantic clauses outlined for “knows” before. Let’s use “knowsO” for this purpose. knowsO = {〈S,  p, w 〉  | At w, i S has a true belief with the content p, and ii This belief is justified/safe/reliably formed}. Properly speaking, there isn’t just one such relation, because the precise requirements for proper knowledge attribution vary from context to context. But for present purposes, we can simplify matters a bit by ignoring this variation. Later on, I will consider how the contextsensitivity in the interpretation of “knows” can be accounted for on my proposal. To fix some further terminology, I will somewhat tendentiously refer to the proposition expressed by a knowledge ascription when “knows” is read as “knowsS” as the strict reading of that knowledge ascription, and I will refer to the proposition expressed by a knowledge ascription when

Knowledge and Loose Talk  275 “knows” is read as “knowsO” as the loose reading of that knowledge ascription. Similar distinctions can be drawn in obvious ways for sentences (1) to (3) above. We can take, e.g., “3pm” to denote a precise point in time to arrive at the strict reading of (1), while we get the loose reading when we take “3pm” to denote, say, a range of points in time surrounding this precise point in time.

3 Parallels This section presents a number of striking parallels between, on the one hand, cases of loose talk as in (1) to (3) and knowledge ascriptions on the other. Semantic contradictions. Even in contexts where the proper use of sentences like (1) to (3) would require only the truth of their loose reading, it is normally infelicitous to conjoin them with a sentence that contradicts only the strict reading. Thus, sentences of the following form sound odd (e.g., Lasersohn, 1999: 535; Lauer, 2012: 390; Solt, 2014: 521; Carter, 2019: 8; Hoek, 2019: 171; Moss, 2019: 263). 4 # Hannah arrived at 3pm, but she didn’t arrive before 3:02. Exactly parallel phenomena are familiar from the debate on so-called concessive knowledge attributions. Here too it seems that even in contexts where the proper use of a knowledge ascription would merely require the truth of its loose reading, it is infelicitous to directly contradict the strict reading. 5 # Hannah knows the bank will be open tomorrow, but she cannot rule out that it burns down over night. Embeddings. The loose reading of (1) to (3) typically embeds under various types of logical operations (Davis, 2007: 408–409; Carter, 2019: 5–6; Hoek, 2019: 172). 6 a b c

Hannah isn’t 6 foot 2. If Hannah is 6 foot 2, then she needs the larger suit. Is Hannah 6 foot 2?

For instance, the more or less trivial fact that Hannah isn’t exactly 6 foot 2 normally doesn’t suffice for the proper use of (6.a). Intuitively, this requires that Hannah isn’t even around 6 foot 2. The same goes for knowledge ascriptions, where the loose reading also embeds (Kindermann, 2019).

276

Alexander Dinges

7 a b c

Hannah doesn’t know the bank will be open. If Hanna knows the bank will be open, we can go home now. Does Hannah know the bank will open?

For instance, a proper use of (7.a) seems to require that Hannah fails to knowO that the bank will be open. The mere failure to knowS doesn’t suffice. Pedantry. While challenging only the strict reading of sentences like (1) to (3) normally sounds pedantic, it still has the effect that the proper use of these sentences subsequently requires the truth of the strict reading (e.g., Lewis, 1979: 351–354; Grice, 1989: 45; Klecha, 2018: 92–93). 8

A: Hannah will arrive at 3pm. B: In fact, she will arrive at 2:58pm. # A: True. Still, she will arrive at 3pm.

In the case of knowledge ascriptions, so-called salient alternative effects bring out parallel phenomena. 9

A: Hannah knows the bank will be open. B: But she cannot rule out that a meteor destroys it overnight. # A: True. Still, she knows the bank will open.

B’s utterance arguably challenges only the strict reading of the initial knowledge ascription. Correspondingly, it sounds somewhat pedantic, while still raising the demands for subsequent uses of knowledge sentences.3 Relevant differences. Depending on the context, the loose reading of sentences like (1) to (3) can be more or less demanding, allowing for more or less slack. It’s a widely shared assumption that the amount of slack allowed in any given context is determined at least in part by what differences count as relevant (Lasersohn, 1999: 526; Hoek, 2018: 173; Klecha, 2018: 97; Carter, 2019: 19). Take sentence (1). In a context where Hannah is dropping by for a coffee, differences of a couple of minutes don’t matter and this sentence will be used with a corresponding amount of slack. The admissible range of points in time surrounding 3pm is relatively wide. Meanwhile, in a context where we are planning our next diamond heist, much smaller differences may matter, and this reduces this range. This phenomenon seems reflected in familiar stakes effects on knowledge attributions. Consider the bank cases (DeRose, 1992). Hannah wants to decide whether she should wait in line at the bank to deposit her paycheck or return on the next day, a Saturday, when the lines are shorter. She has been to the bank two weeks before on a Saturday, and it was open then. Based on this, she says, “I know the bank will be open tomorrow.” Intuitively, the requirements for the proper use of this

Knowledge and Loose Talk  277 sentence shift with what is at stake. If it is very important to deposit the paycheck before Monday, we apply stricter standards. For instance, we require safety across a wider range of possible worlds.4 This phenomenon is easily re-described to show that relevant differences affect the admissible amount of slack in the case of knowledge ascriptions. In the bank cases, Hannah wants to decide whether she should wait in line or come back the other day. If the stakes are low, it doesn’t matter whether, e.g., her belief is safe across every possible world or just the nearby ones, because she will head home either way. Meanwhile, if the stakes are high, this difference becomes relevant, because Hannah may decide to wait in line already when her belief fails to be safe across relatively distant possible worlds. The requirements for proper knowledge attributions vary accordingly. In this way, relevant differences affect the admissible amount of slack. Slack-regulators. The target expressions in sentences like (1) to (3) all come with an associated set of so-called “slack regulators” (Lasersohn, 1999: 525; Klecha, 2018: 95; Carter, 2019: 4–5; Hoek, 2019: 177–178; Moss, 2019: 262). These expressions reduce the admissible amount of slack, so that the requirements for proper use come closer to the respective strict reading. Consider the following pairs: 10 a b

Hannah arrived at 3pm. / Hannah is 6 foot 2. / The table is flat. Hannah arrived at exactly 3pm. / Hannah is exactly 6 foot 2. / The table is perfectly flat.

In the b-sentences, “exactly” and “perfectly” feature as a slack regulator that raise the standards for proper usage. Slack regulators show an interesting kind of unidirectionality (Carter, 2019: 5–6). 11 a b

Hannah is 6 foot 2, but she isn’t exactly 6 foot 2. ? Hannah isn’t exactly 6 foot 2, but she is 6 foot 2.

The first sentence in this pair sounds perfectly fine, while the reversed, second sentence sounds comparatively odd. Here the parallels to knowledge ascriptions are least clear. In particular, familiar slack regulators sound off when applied to “knows.” 12 # Hannah perfectly/exactly knows that the bank will be open. Still, there are candidates for slack regulators. Moss (2017: 113), for instance, suggests “for sure” as a potential slack regulator (see also Ludlow, 2005: 19–20; Kyriacou, 2019b: 16 for other candidates). 13 a b

Hannah knows the bank will be open. Hannah knows for sure that the bank will be open.

278  Alexander Dinges Adding “for sure” arguably raises the standard, such that the second sentence comes out as intuitively more demanding than the first. The indicated asymmetry may be felt too, though the intuitions are shaky (one reviewer denies them entirely). 14 a b

Hannah knows that the bank will be open, but she doesn’t know for sure that it will be open. ? Hannah doesn’t know for sure that the bank will be open, but she knows that it will be open.

In summary, there are striking parallels between knowledge ascriptions and cases of loose talk with respect to a whole range of data points that have been crucial to each debate. This lends strong prima facie credibility to the idea that knowledge ascriptions are cases of loose talk.

4 Pragmatic Accounts of Loose Talk Despite these parallels, appeals to loose talk are of no avail to skeptical invariantism. As indicated, many authors propose semantic treatments of loose talk, which are, by definition, incompatible with skeptical invariantism. Familiar pragmatic accounts are problematic too, because they clash with the data outlined when applied to knowledge ascriptions, as I subsequently show. 4.1 Lasersohn’s Pragmatic Halos Lasersohn (1999) presents the first substantial pragmatic account of loose talk. On his view, sentences (1) to (3) express their respective strict reading, which is normally obviously false or at least obviously unjustified. We normally don’t possess the kinds of measurement instruments that would be required to ascertain, e.g., that Hannah arrived at exactly 3pm. To make this compatible with the observation that we can properly use these sentences nonetheless, Lasersohn proposes to modify the norms of assertion. A commonly accepted norm of assertion ties the appropriateness of literal assertions to the truth of the proposition expressed. A literal assertion of S in context C is proper only if the proposition expressed by S in C is true. On this norm, utterances of sentences (1) to (3) come out as improper most of the time, assuming they are literal and assuming Lasersohn’s semantic commitments. He thus proposes the following replacement norm. A literal assertion of S in context C is proper only if one proposition in the pragmatic halo of S in C is true.

Knowledge and Loose Talk  279 This norm predicts that utterances of sentences (1) to (3) are often proper, despite the fact that they express obviously false or unjustified propositions. To see this, an account of the pragmatic halo of a sentence in context is required, and Lasersohn provides such an account along the following lines.5 The starting idea is to associate each atomic expression in a given sentence with a pragmatic halo. The pragmatic halo of a given atomic expression comprises a set of objects of the same logical type as the original denotation, such that the difference between these objects and the original denotation is “pragmatically ignorable” (526). Take “3pm” in sentence (1). Its halo comprises points in time surrounding the precise point in time 3pm, such that each of these points in time differs from 3pm in pragmatically ignorable ways. What counts as pragmatically ignorable depends on context. If, for instance, we only care about Hannah’s rough time of arrival, the halo of “3pm” might comprise all points in time in the interval of 3pm ± 5 minutes. Where HC gives you the halo of an expression in such a context C, we can write: HC (3pm) = {t | t ∈ [2 : 55pm …3 : 05pm]} The halo of entire sentences is compositionally derived from the halos of the atomic expressions, by deriving one proposition for each combination of elements in these halos. HC (Hannah arrived at 3pm) = {{w |  〈 x, y, w 〉 ∈ R} | x ∈ HC (Hannah); y ∈ HC (3pm);R ∈ HC (arrived at)} In the case at hand, this gives us propositions to the effect that Hannah arrived at t, for each point in time t in the interval [2:55pm … 3:05pm]. We can now see that utterances of, e.g., sentence (1) can be proper on the suggested norm of assertion. Given this norm, they are proper only if one of the propositions in the halo is true, i.e. only if Hannah arrived within the interval of 3pm ± 5 minutes. This condition is easily satisfied in everyday contexts. Promisingly, ordinary knowledge ascriptions come out as proper too, even assuming skeptical invariantism. For concreteness, consider a skeptical invariantist who adopts the following, reliabilist clause for “knows.” knows = {〈S, p, w 〉  | At w, i S has a true belief with the content p, and ii This belief is 100% reliably formed}.

280  Alexander Dinges Depending on the context, the difference between 100% reliability and, say, 80% reliability may not matter. If we face low stakes, for instance, one’s decisions may be the same either way. We thus obtain a halo for “knows” along the following lines, where NC is the lowest percentage that differs irrelevantly from 100% in C. HC(knows) = {{〈S, p, w 〉  | At w, i S has a true belief with the content p, and ii This belief is N% reliably formed }|N ≥ NC}. In our low stakes bank case, for instance, NC may be 80, such that the halo of “knows” comprises one relation for any N in the interval [80 … 100]. Given the indicated norm of assertion, a knowledge ascription is thus proper when the subject satisfies one of these relations, i.e., when she has a pertinent true belief that has been formed on a basis that is at least 80% reliable. This condition is satisfiable, and hence ordinary knowledge ascriptions come out proper. This promising outlook notwithstanding, skeptical invariantists end up with false predictions about embedded knowledge ascriptions. Consider “S doesn’t know that p.” According to Lasersohn, the halo of this sentence is derived by composing the elements of the halos of its atomic expressions. As Lasersohn (1999: 548) himself points out, the halo of every atomic expression always comprises its literal denotation. This is because the “difference” between the literal denotation and itself is always pragmatically ignorable. For instance, the halo of “knows” always comprises the literal denotation. The halo of “S doesn’t know that p” therefore comprises the proposition it literally expresses, which results from composing the literal denotations of its atomic expressions. On standard assumptions about negation, this is the following proposition:

{w |  〈 S

,  that p , w 〉 ∉   knows

}

Assuming the above skeptical invariantist semantics, this proposition is obviously true when it comes to most everyday propositions, for we hardly if ever use 100% reliable belief formation methods. This means that there is an obviously true proposition in the halo of most knowledge denials. Consequently, knowledge denials should be obviously proper most of the time, which is not what we want. When people have a decent amount of evidence, it is normally improper to deny them knowledge, as indicated above.6 4.2 Klecha’s Optimality Account Klecha (2018) proposes a different pragmatic account of loose talk, which also starts from the assumption that sentences (1) to (3) express

Knowledge and Loose Talk  281 their obviously false or unjustified strict reading. Unlike Lasersohn, he retains the familiar norm of assertion, which ties the appropriateness of literal assertions to the truth of the proposition expressed. But he argues that we use sentences like (1) to (3) non-literally in ordinary contexts. The non-literal message they convey corresponds to their loose reading, which is often true. According to Klecha, respective utterances are proper for this reason (see also Davis, 2007, who appeals to conversational implicature). Klecha proposes an optimality theoretic account to spell out how people derive the non-literal loose reading (see also Krifka, 2002). According to optimality theory, hearers generally use the following strategy to derive what speakers want to convey with their utterances. They ask, which belief state do I have to attribute to the speaker such that the sentence she uttered comes out as an optimal choice? Once they have figured this out, they take the content of the respective belief state as the message conveyed. Klecha offers an account of optimality that supposedly allows one to derive loose readings in this way. Here is a simplified statement of Klecha’s proposal. The central notion in Klecha’s account is that of the coarsening of a proposition. It relies on a certain way to partition the set of possible worlds with which I will begin. In any given context, people have things they want to achieve, they have so-called “domain goals” (Roberts, 2012). Depending on what the world is like, one or another means will be conducive to these goals, but not every aspect of the world matters. Suppose Hannah wants to drop by for a coffee, and I want to make preparations. In this context, it doesn’t matter to me whether Hannah arrives at 3pm or 3:01pm, maybe because I will set up the table some minutes before 3pm anyway. We can therefore use the domain goals to induce a partition on the set of possible worlds, where each partition cell contains a set of worlds such that whichever of these worlds we occupy, the recommended means to achieve our domain goals are the same. Let’s call the partition so-induced in a given context C, πC . We can now define the coarsening, KC , of a proposition p in a context C. Intuitively, coarsening makes propositions less precise, so that they speak only to the domain goals. Formally, coarsening gives you the union of the partition cells in πC that overlap with p (Figure 14.1). KC (p) = ∪ {p′ ∈ πC |  p′ ∩ p ≠ ∅} To illustrate, the coarsening of the strict reading of (1) no longer tells us that we find ourselves in a world where Hannah arrived at exactly 3pm. Instead, it tells us that we find ourselves in world from a partition cell containing worlds where this is so. Thus, we could still be in a world where Hannah arrived at 3:01pm as long as such a world is clustered together with a world where she arrived at 3:00pm.

282  Alexander Dinges → →

Figure 14.1 Schematic depiction of the coarsening of two different propositions. Lines demarcate partion-cells induced by πC . The grey area on the left covers the worlds in which the proposition is true. The result of coarsening is shown in grey on the right.

With this idea of coarsening at hand, Klecha suggests that the proposition conveyed by a sentence S in context C is determined as follows (simplifying a bit; see Klecha, 2018: 108). The subsequent principle explicates the optimality theoretic idea that the conveyed message corresponds to the content of the belief state that make the chosen sentence optimal. The proposition conveyed by S in C = ∪{KC (B) | B ∈ BC  and S ∈ opt(SC | B)}. BC stands for the set of belief states the speaker is in, for all the audience knows (see Klecha, 2018: 107). A belief state is represented as the set of possible worlds that are consistent with it. SC denotes the set of “utterance alternatives,” i.e., roughly, the set of sentences that the speaker can relevantly utter in the context at hand, including the actually uttered sentence, S (see Klecha, 2018: 108). The set opt (SC | B) is the set of optimal sentences among SC on the assumption that the speaker is in belief state B. The proposition conveyed thus corresponds to the union of the coarsenings of the possible belief states on behalf of the speaker relative to which the actually uttered sentence is optimal compared to its alternatives. What is missing is a definition of the set opt (SC | B), the set of optimal sentences relative to a belief state. Klecha offers three constraints that together determine this set. One constraint is Faithfulness, where sentences rank highly roughly when the speaker believes the coarsening of the proposition expressed (see Klecha, 2018: 109). Another constraint is Manner, where sentences rank highly when they are perspicuous in a relevant sense, e.g., when they are brief and use “round” expressions such as “3pm” instead of “3:01pm” (see Klecha, 2018: 98–99, 109– 110; see also Krifka, 2002). For our purposes, we can focus exclusively on the third constraint called Informativity, which ranks sentences as follows. Informativity(S) ≥ Informativity(S)′ iff KC ( S

) ⊆ KC ( S′ )

Knowledge and Loose Talk  283 A sentence is more informative than another, and thus pro tanto more optimal, when the coarsening of the proposition expressed by the former sentence is stronger than the coarsening of the proposition expressed by the latter. This principle gets Klecha in trouble when it comes to knowledge denials. To see this, consider a context like the bank cases where we only care about whether Hannah knowsO that the bank will be open, because this will settle our decision whether to wait in line. In such a context, πC partitions the set of possible worlds into worlds where Hannah at least knowsO that the bank will be open and worlds where she doesn’t at least knowO that.

πC = {w | At w, Hannah at least knowsO that the  bank will be open},{w | At w, Hannah doesn’ t at least knowO that the bank will be open}

   

Now look at the result of coarsening the proposition expressed by “Hannah doesn’t know that the bank will be open.” On a skeptical invariantist semantics, coarsening this proposition leads to a complete triviality, the entire set of possible worlds. This is because lacking knowledgeS is compatible with both lacking and having knowledgeO. The intersection with every partition cell is non-empty (Figure 14.2). The knowledge denial thus comes out as maximally uninformative. Its coarsening is entailed by the coarsening of every sentence, and hence no sentence is less informative. Such utterly uninformative sentences shouldn’t be optimal relative to any belief state we are normally prepared to ascribe.7 This means that, when we derive the proposition conveyed by the principle above, i.e., by unioning the belief states relative to which the uttered sentence is optimal, we end up with the empty set. Klecha’s theory predicts that knowledge denials typically convey the equivalent of a contradiction, which should make them trivially improper.8 This is evidently no improvement on Lasersohn’s theory, which made knowledge denials trivially proper.

→

Figure 14.2 S chematic depiction of the coarsening of a knowledge denial in a bank case context. Lines demarcate partion-cells in πC . The grey area on the left covers the worlds in which the knowledge denial is true on a skeptical invariant semantics. The result of coarsening is shown in grey on the right.

284  Alexander Dinges 4.3 Hoek’s Conversational Exculpatures Hoek (2018, 2019) provides another pragmatic account of loose talk. Like Klecha, he denies the literality of typical utterances of (1) to (3) to square their propriety with standard norms of assertion. Instead of appealing to optimality theory, though, he derives the conveyed message through a mechanism that he dubs conversational exculpature. Conversational exculpature is a type of pragmatic repair strategy that hearers apply under certain conditions. Hearers apply this repair strategy when the proposition literally expressed by an utterance fails to be wholly relevant to the question under discussion (henceforth, QUD). The notion of being wholly relevant is defined against the background of the following, familiar understanding of a question. A question Q is a partition of the space W of possible worlds. (2019: 173) Here, each partition cell stands for one complete answer to the question Q. Take the question how many cows are there on this ranch. This question induces one partition cell for each answer of the form there are exactly n cows on this ranch, containing just those worlds where this answer is true. Against this background, we can define what it means to be wholly relevant. A proposition p ⊆ W is wholly relevant to a question Q if and only if no partition cell of Q contains worlds where p is false and also worlds where p is true. (2019: 173) To illustrate, the responses there are exactly three cows on this ranch and there are between three and five cows on this ranch are both wholly relevant to the question above. Meanwhile, the response there are exactly three cows and a donkey on this ranch is not because it is true in only some of the worlds in the three-cows partition cell, namely, those where there is also a donkey (notice that coarsening in Klecha’s sense leads to a proposition that is wholly relevant to the question/partition πC). Here is how the so-triggered repair strategy works. The basic idea is to replace the proposition expressed with another proposition, which is then taken as being conveyed. Which proposition replaces the proposition expressed? In making utterances, speakers often speak as if certain assumptions were true, while in fact, they are knowingly false or uncertain (Hoek, 2018: 160–161). These pragmatic presuppositions, together with the QUD, determine the proposition conveyed by sentence S in context C, as follows.

Knowledge and Loose Talk  285 The proposition conveyed by S in C = the unique proposition p, if there is any, such that i p is wholly relevant to the QUD, and ii p is equivalent to the proposition expressed conditional on the pragmatic presuppositions (Hoek, 2019: 174).9 Consider a case where, according to Hoek, this mechanism is at work independently of loose talk. Suppose the QUD is what type of hat does Hannah wear, which induces one partition cell for each type of hat there is. You respond, “She wears the same type of hat Sherlock Holmes wears.” This response isn’t wholly relevant. Take, e.g., the partition cell induced by the response that Hannah wears a straw-hat. The given response is true in only some of the worlds in this partition cell, namely, those where both Hannah and Sherlock Holmes wear a strawhat. This triggers conversational exculpature, and the proposition conveyed is determined as follows. Naturally enough, the speaker speaks as if the Sherlock Holmes stories are true by “referring” to Sherlock Holmes. Given this presupposition, there is a unique proposition that is (i) wholly relevant to the QUD and (ii) conditionally equivalent to the proposition expressed. This is the proposition that Hannah wears a deerstalker. This proposition is evidently wholly relevant. It is conditionally equivalent to proposition expressed for the following reason. If Hannah wears the same hat as Sherlock Holmes does while the Holmes stories are true, then she wears a deerstalker. After all, Holmes wears a deerstalker. Conversely, if Hannah wears a deerstalker while the Holmes stories are true, then she wears the same hat as Sherlock Holmes does. Other wholly relevant propositions—such as the proposition that Hannah wears a straw-hat—aren’t conditionally equivalent to the proposition expressed. The proposition that Hannah wears a deerstalker is therefore conveyed (which is desirable, because if this proposition is conveyed, the speaker is not committed to the existence of Sherlock Holmes). Let’s turn to loose talk. According to Hoek, loose talk is a special case of conversational exculpature, where the contextual presupposition involved takes the specific shape of a scale presupposition. Scale presuppositions are presuppositions induced when speakers select an expression from a conventionally restricted scale of alternative expressions. Their content is, roughly, that the true description of the object in question involves one of these expressions. For instance, speakers of English conventionally use expressions from the following scale to describe personal heights, at least in contexts where precision is not required. {… “5 foot 11,” “6 feet,” “6 foot 1,” “6 foot 2,” …}

286  Alexander Dinges By uttering, “Hannah is 6 foot 2,” you select an expression from this scale, thereby indicating that you are following the indicated convention. By following this convention, you speak as if Hannah’s height was located on the indicated scale. Hence, you (falsely) presuppose that Hannah is an exact number of inches tall. This is a scale presupposition. Let’s see how this scale presupposition figures in a specific application of conversational exculpature to loose talk. Suppose we want to know Hannah’s height to the nearest inch, a question that induces partition cells containing worlds where Hannah is n ± ½ inches tall. You respond, “Hannah is 6 foot 2.” Given Hoek’s semantic commitments, this response isn’t wholly relevant because it is true in only some of the worlds in the 6 foot 2 ± ½-inches-cell. This triggers conversational exculpature. By the reasoning above, your use of “6 foot 2” induces the scale presupposition that Hannah is an exact number of inches tall. Given that, there is a unique proposition that is (i) wholly relevant to the QUD and (ii) conditionally equivalent to the proposition expressed. This is the proposition that Hannah is 6 foot 2 ± ½ inches. Evidently, this proposition is wholly relevant. It is conditionally equivalent to the proposition expressed for the following reason. If Hannah is 6 foot 2, then Hannah is 6 foot 2 ± ½ inches independently of any additional assumptions. Meanwhile, if Hannah is 6 foot 2 ± ½ inches and we assume, via the scale presupposition, that Hannah is an exact number of inches tall, then she must be 6 foot 2. The proposition that Hannah is 6 foot 2 ± ½ inches is therefore conveyed, which is as desired. This account doesn’t have specific problems with embedded knowledge ascriptions (see below). Unfortunately, though, it doesn’t even make the right predictions when it comes to unembedded ones. On the suggested account, loose talk arises against the background of a scale presupposition, which, in turn, depends on a scale of expressions, which encodes conventional ways to describe, e.g., the height of a person. To apply Hoek’s account to knowledge ascriptions, we need such a scale for “knows,” containing expressions conventionally used to describe each other’s epistemic states. No scale suits our purposes, however. Consider the range of expressions that we conventionally use to describe each other’s epistemic states. Sentences like “I know/heard/read/ saw/recall that p” are all used regularly in this context. Therefore, we should end up with a scale like this. {“know,” “hear,” “read,” “see,” “recall,” …} But this scale doesn’t give us what we want. The proposition conveyed with a knowledge ascription should be some proposition to the effect that S at least knowsO that p. This would deliver the correct predictions about when knowledge ascriptions are appropriate. So, if Hoek’s account of loose talk is correct, and knowledge ascriptions are a case

Knowledge and Loose Talk  287 of loose talk, this proposition should be conditionally equivalent to the proposition expressed, i.e., the proposition that S knowsS that p. But that’s not the case. Even given the assumption that we are located on the indicated scale, the proposition that S at least knowsO that p doesn’t entail that S knowsS that p. For instance, S may knowO that p because S has read that p. In that case, S is in one of the states on the scale and still lacks knowledgeS. One could instead propose the following scale, which is familiar from discussions of so-called Horn-scales (e.g., Levinson, 2000: 87). {“believe,” “know”} But again, the proposition that S at least knowsO that p doesn’t entail that S knowsS that p, even when it is stipulated that we are located on this scale. Indeed, the proposition that S at least knowsO that p entails the scale presupposition because it entails that S believes that p. As such, the scale presupposition cannot generate any novel entailments. One might suggest the following scale instead:10 {“merely believe,” “know”} To assess this scale, we first have to get clear on what “merely believing” is. One idea would be that merely believing p is believing p without any evidence. This would lead to correct predictions, e.g., in a bank case context, where we only care about whether Hannah at least knowsO that the bank will be open (henceforth B) and where the QUD thus entails the bi-partite partition familiar from the discussion of Klecha’s account. According to skeptical invariantism, the proposition expressed by “Hannah knows B” is the proposition that Hannah knowsS B, which isn’t wholly relevant. This triggers conversational exculpature. On the assumption that Hannah is located on the above scale interpreted as described, the proposition that Hannah knowsS B is uniquely conditionally equivalent to the wholly relevant proposition that she at least knowsO B. For if Hannah at least knowsO B, while being located on the above scale, then the only option is that she knowsS B. Meanwhile, if Hannah knowsS B then she at least knowsO B independently of any additional assumptions. Unfortunately, it just seems false that, whenever we make appropriate knowledge ascriptions in ordinary contexts, we speak as if we are located on the indicated scale, i.e., as if we either knowS something or lack evidence entirely. Suppose I say, “Hannah doesn’t know who perpetrated the crime. She found John’s weapon, but someone might have placed it there. Meanwhile, Sarah knows who did it, because she talked to the neighbor who saw everything.” I am acknowledging a whole range of possible epistemic positions here, and if I invoke any scale at all, it is the rich scale initially provided, which doesn’t yield the right interpretation.

288  Alexander Dinges “Merely believing” something can be interpreted in other ways. Two natural options would be as believing without knowingS and as believing without knowingO. On either interpretation, however, we face by now familiar problems. On the interpretation as believing without knowingS , we cannot establish the conditional equivalence between knowingS and at least knowingO. This is because our scale presupposition now says that we either believe p without knowingS p or knowS p. Given that knowledgeS entails belief, this just means that we believe p. Even granting this assumption, knowingO p doesn’t entail knowingS p. On the interpretation as believing without knowingO, we can derive the required equivalence (readers can verify this for themselves). But once more, it seems implausible that we typically presuppose that our epistemic position is either weaker than knowledgeO or as strong as knowledgeS. In the crime example above, the speaker acknowledges many epistemic states (testimonial and perceptual belief, for instance) that often amount to knowledgeO while being insufficient for knowledgeS.11

5 A Novel Approach The appeal to loose talk on behalf of skeptical invariantism fails if we assume that current theories of loose talk are on the right track. This puts strong pressure on proponents of skeptical invariantism who defend their view by appeal to loose talk. In what follows, I explore a closely related defense of skeptical invariantism, which appeals to the more general phenomenon of conversational exculpature instead. This approach is more promising. To make this clear, I begin with a brief presentation of epistemic contextualism as proposed by Lewis (1996) and spelled out by Blome-Tillmann (2014). This will serve as a helpful background later on. 5.1 Epistemic Contextualism Lewis (1996) presents a familiar contextualist semantics for “knows” along the following lines: knows C = {< S, p, w >| At w, S’ s evidence eliminates all ¬p worlds, except for those that are properly ignored in C} On this view, “knows” expresses differently demanding epistemic relations depending on which possible worlds are “properly ignored” in the context hand. Blome-Tillmann (2014) elaborates on this proposal, offering specific rules for when a possible world counts as properly ignored. He largely accepts most of Lewis’ rules, specifically, the Rules of Actuality, Resemblance, Belief, Reliability, Method and Conservatism

Knowledge and Loose Talk  289 (31). Crucially, though, he replaces Lewis’ Rule of Attention by the Rule of Presupposition (20; I rephrased this rule for presentational purposes). Rule of Presupposition A possible world w is properly ignored in C only if the pragmatic presuppositions in C entail that w doesn’t obtain. For instance, we must presuppose that the bank hasn’t changed its hours in order to properly ignore worlds where it has. Following Stalnaker (1978), Blome-Tillmann (2014: 30) defines the relevant notion of pragmatic presupposition as follows. x pragmatically presupposes p in C ↔ x is disposed to behave, in her use of language, as if she believed p to be common ground in C. Importantly, on this conception of pragmatic presupposition, one can pragmatically presuppose p even if one doesn’t believe p. One must only behave as if one took p to be common ground. This aligns with Hoek’s characterization of pragmatic presupposition also in terms of “speaking as if.” For instance, we pragmatically presuppose that the bank hasn’t changed its hours once we speak as if this was common ground, independently of whether this is actually so. To accommodate data such as those outlined in Section 3, BlomeTillmann argues that pragmatic presuppositions vary, e.g., with salient alternatives and stakes (36–37). In the low stakes bank cases, for instance, we pragmatically presuppose that the bank hasn’t changed its hours, behaving as if this was common ground. When someone makes this possibility salient, by mentioning it, we stop presupposing this. Similarly, high stakes force us to take this possibility seriously, which again leads us to abandon the respective presupposition. This variation in pragmatic presuppositions leads to variations in which worlds our evidence needs to eliminate before we can properly ascribe “knowledge,” via the Rule of Presupposition. And this explains, e.g., why knowledge ascriptions are sensitive to stakes and salient error-possibilities. 5.2 Knowledge Ascriptions and Conversational Exculpature Now consider a skeptical invariantist who adopts the following, maximally demanding, invariant variation on Lewis’ semantics for “knows.” knowsS = knows = {< S, p, w >| At w, S’s evidence  eliminates all ¬p worlds} Such a skeptical invariantist faces the familiar challenge to explain why we properly ascribe knowledge all the time, given that our evidence hardly ever eliminates every ¬p world. Based on Blome-Tillmann’s

290  Alexander Dinges assumptions about pragmatic presuppositions, conversational exculpature yields a promising response. Let’s begin with a stipulatively defined notion of properly treating a possible world as ruled out. Properly treating a possible world as ruled out is just like properly ignoring it, except that the following rule replaces the Rule of Presupposition. Rule of Presupposition* A world w is properly treated as ruled out in C only if the pragmatic presuppositions in C entail that the target subject’s evidence eliminates w. Properly treating a world as ruled out, unlike properly ignoring it, depends on presuppositions about our evidence, particularly, presuppositions about the evidence available to the target subject, i.e., the subject to whom knowledge is ascribed. Properly treating a world was ruled out requires the presupposition that this subject can rule w out. I submit that we properly treat many possible worlds as ruled out in ordinary contexts. More specifically, I submit that the worlds we properly treat as ruled out are more or less the same worlds as the worlds that, according to Blome-Tillmann, we properly ignore (see below for some important divergences). For instance, in the low stakes bank case, we properly ignore worlds where the bank has changed its hours, where a meteor destroys it overnight, where an evil demon deceives us, etc. This is because we presuppose that the bank hasn’t changed its hours, etc. Blome-Tillmann at least finds this assumption “intuitive” (21), and I agree. It is likewise intuitive, I think, that we properly treat the indicated worlds as ruled out. For instance, we plausibly presuppose that Hannah can rule out that the bank has changed its hours, that she can rule out that a meteor destroys it overnight, that she can rule out that an evil demon deceives us, etc. We plausibly treat all of these worlds as if it was settled that they don’t obtain. Importantly, in the case of both properly ignoring and properly treating as ruled out, the respective presuppositions need not align with our beliefs or our knowledge. We may believe or know neither that the bank hasn’t changed its hours nor that Hannah can rule this out. We make the relevant presuppositions nevertheless by speaking as if they were true. The situation here is analogous to the cases of conversational exculpature discussed above, which also featured knowingly false presuppositions such as scale presuppositions or presuppositions about Sherlock Holmes. With this in mind, let’s turn to how conversational exculpature applies to knowledge ascriptions. First, consider the QUD and the pragmatic presuppositions in, e.g., the bank cases. As for the QUD, we are interested

Knowledge and Loose Talk  291 in Hannah’s evidential position, for this bears on our decision to wait in line at the bank or to come back the other day. However, not every aspect of her evidential position matters. Consider, for instance, possible worlds that we properly treat as ruled out. By the Rule of Presupposition*, we presuppose that Hannah’s evidence eliminates those worlds and so we presumably don’t care about them. A natural QUD would be, “Which possible worlds that aren’t properly treated as ruled out does Hannah’s evidence eliminate?” As for the pragmatic presuppositions, consider once more the possible worlds we properly treat as ruled out. By the Rule of Presupposition*, we pragmatically presuppose that Hannah’s evidence eliminates them, i.e., we pragmatically presuppose that Hannah’s evidence eliminates all possible worlds that are properly treated as ruled out. With these assumptions about the QUD and the pragmatic presuppositions in place, conversational exculpature makes promising predictions about knowledge ascriptions. Consider an utterance of “Hannah knows that the bank will be open (B)” in the bank cases. According to the above version of skeptical invariantism, this utterance expresses the proposition that Hannah’s evidence eliminates all ¬B-worlds. This proposition isn’t wholly relevant to the QUD, which triggers conversational exculpature. Conversational exculpature then predicts that the knowledge ascription conveys that Hannah’s evidence eliminates all ¬B-worlds that aren’t properly treated as ruled out. Here is why. The latter proposition is evidently wholly relevant to the QUD. It is unique in also being conditionally equivalent to the proposition expressed, i.e., the proposition that Hannah’s evidence eliminates all ¬B-worlds. If Hannah’s evidence eliminates all ¬B-worlds, then her evidence eliminates all ¬B-worlds that aren’t properly treated as ruled out, independently of any additional assumptions. Meanwhile, if her evidence eliminates all ¬B-worlds that aren’t properly treated as ruled out, then her evidence eliminates all ¬B-worlds, given the pragmatic presupposition that her evidence eliminates all worlds that are properly treated as ruled out. Knowledge ascriptions thus convey that the subject’s evidence eliminates all ¬p-worlds that aren’t properly treated as ruled out. More precisely, they convey that the subject knowsO that p, where the exact requirements for knowledgeO shift with the context as follows. knowsO C = {< S, p, w >| At w, S’ s evidence eliminates all ¬p-worlds, except for those that are properly treated as ruled out in C} Given that we properly treat as ruled out more or less the same worlds as the worlds that we properly ignore, as argued above, it follows that ordinary knowledge ascriptions are often proper, even assuming skeptical invariantism. For ruling out the remaining worlds is typically feasible (that’s at least what contextualists like Lewis and Blome-Tillmann assume).

292  Alexander Dinges I can also make good on the promissory note above that I will be able to explain why the propriety conditions for ordinary knowledge ascriptions vary with the context. On my view, this is due to variations in which possible worlds are properly treated as ruled out, where these variations in turn are due to variations in our pragmatic presuppositions. Such variations shift the proposition conveyed with knowledge ascriptions and thereby the conditions under which they are appropriate. My account thus mimics the precise and detailed predictions of BlomeTillmann’s contextualism, while sticking with an invariantist semantics. Unlike loose use accounts, I can also explain the data from Section 3. Let’s begin with knowledge denials, which spelled doom for Lasersohn’s and Klecha’s theory of loose talk when applied to “knows.” Conversational exculpature easily explains why knowledge denials are neither trivially proper nor trivially improper. Take an utterance of “Hannah doesn’t know B” in the bank cases. According to skeptical invariantism, this utterance expresses the proposition that Hannah’s evidence doesn’t eliminate all ¬B-worlds (i.e., that Hannah doesn’t knowS B). Once more, this proposition isn’t wholly relevant to the QUD, which triggers conversational exculpature. Conversational exculpature then predicts that the knowledge denial conveys that Hannah’s evidence doesn’t eliminate all ¬B-worlds that aren’t properly treated as ruled out (i.e., that Hannah doesn’t knowO B). This proposition is unique in being wholly relevant and conditionally equivalent to the proposition expressed. It is evidently wholly relevant given the QUD. It is conditionally equivalent to the proposition expressed—that Hannah’s evidence doesn’t eliminate all ¬B-worlds—for the following reason. If Hannah’s evidence doesn’t eliminate all ¬B-worlds that aren’t properly treated as ruled out, then her evidence doesn’t eliminate all ¬B-worlds, independently of any additional assumptions. Meanwhile, if her evidence doesn’t eliminate all ¬B-worlds, then her evidence doesn’t eliminate all ¬B-worlds that aren’t properly treated as ruled out, given the presupposition that her evidence eliminates all worlds that are properly treated as ruled out. When it comes to salient alternative and stakes effects, I can once more mimic Blome-Tillmann’s contextualist account. Following BlomeTillmann, salient error-possibilities and high stakes shift our presuppositions and thereby what we properly treat as ruled out. This affects the proposition conveyed by knowledge ascriptions via conversational exculpature and thereby the conditions under which these utterances are appropriate. For instance, in the low stakes bank case, we properly treat worlds as ruled out where the bank has changed its hours, by presupposing that Hannah’s evidence eliminates them. Meanwhile, in the high stakes bank case, we give up this presupposition. Consider, finally, the assumed slack regulator “for sure.” Knowledge ascriptions are no longer cases of loose talk, and correspondingly, we can no longer apply Hoek’s suggested account of slack regulators,

Knowledge and Loose Talk  293 which is tied to loose talk and doesn’t carry over to conversational exculpature in general. He suggests that familiar slack regulators shift the governing scale to a more fine-grained scale (2019: 177–178). Since I have abandoned scales, this can’t be what’s going on. As indicated at the outset, though, the parallel between knowledge ascriptions and loose talk is feeblest when it comes to slack regulators, and familiar slack regulators don’t apply to “knows.” So this outcome seems just what we want. At the same time, we only have to modify Hoek’s account slightly to offer a plausible account of “for sure.” According to Hoek, slack regulators express some kind of interpretation instruction, namely, an instruction to use a finer-grained scale than one would have used otherwise. Our target expression “for sure” doesn’t express just that instruction but something similar, namely, an instruction to abandon at least some of the pragmatic presuppositions that one would otherwise have used to interpret the target utterance. Consider an utterance of “Hannah knows for sure that the bank will be open” as made in the low stakes bank case. One would have presupposed, e.g., that Hannah’s evidence eliminates worlds where the bank changes its hours, but “for sure” instructs one to abandon such presuppositions. Once these presuppositions have been abandoned, conversational exculpature yields a more demanding interpretation, namely, an interpretation on which Hannah’s evidence eliminates, e.g., the possibility of changed opening hours. 5.3 Comparison Given the close parallel to Blome-Tillmann’s contextualism, let me briefly explore whether the suggested view has any distinctive advantages. Here are two initial candidates. First, Blome-Tillmann arguably has problems with certain kinds of third-personal knowledge ascriptions. Emily says, “Joe is directing plays again,” thereby pragmatically presupposing that Joe directed plays before (henceforth P). She subsequently says, “Nathan knows P.” This knowledge ascription threatens to be trivially true on Blome-Tillmann’s contextualism. It is true on this account iff Nathan’s evidence eliminates all ¬P-worlds, expect for those that are properly ignored. But why couldn’t Emily properly ignore just every ¬P-world? The Rule of Presupposition, for instance, tells us that we can properly ignore a ¬P-world only if our pragmatic presuppositions entail that it doesn’t obtain. Since Emily pragmatically presupposes P, this condition is satisfied for every ¬P-world. Maybe Lewis’ other rules help to explain why Emily cannot properly ignore some ¬P-worlds, but Ichikawa (2015) suggests that we can construct the case in such a way that they don’t.12 This problem doesn’t arise on my exculpature account. On this account, the knowledge ascription conveys that Nathan’s evidence eliminates every ¬P-world,

294  Alexander Dinges except for those that are properly treated as ruled out. The Rule of Presupposition* tells us that we can properly treat a ¬P-world as ruled out only if we presuppose that, in this case, Nathan’s evidence eliminates it. But we don’t presuppose that in the case at hand. We presuppose that ¬P-worlds don’t obtain. It doesn’t follow that we presuppose that Nathan’s evidence eliminates them. As indicated, such presuppositions often go together. In third-person cases, however, they tend to come apart, in a way that favors the exculpature account. Second, Blome-Tillmann himself grants that his version of contextualism posits an “entirely novel and previously unheard of type of context-sensitivity” (124) by tying semantic context-sensitivity to pragmatic presuppositions. While he plausibly suggests that this need not be a problem, it still seems that, other things being equal, it would be nice to avoid such commitments. Skeptical invariantists can avoid such commitments if they see knowledge ascriptions as just another instantiation of conversational exculpature. Relatedly, and even if Blome-Tillmann’s form of context-sensitivity was precedented, Grice’s “Modified Occam’s Razor” favors skeptical invariantism as outlined. Senses are not to be multiplied beyond necessity. In at least one sense of “sense,” BlomeTillmann posits many senses where skeptical invariantists posit just one (see Hazlett, 2007 for a pertinent construal of the Razor).

6 Conclusion Skeptical invariantists must explain how ordinary knowledge ascriptions can be appropriate. Appeals to loose talk are of no avail if we assume that extant accounts of loose talk are on the right track. Skeptical invariantists can still appeal to the general phenomenon of conversational exculpature to explain proper knowledge ascription once they take on board familiar assumptions about the dynamics of pragmatic presuppositions. The resulting view combines the virtues of invariantism and contextualism. It posits a unique knowledge relation, and it offers an account of the context-sensitivity of ordinary knowledge ascriptions, which is on a par, in terms of detail and predictive power, with the most elaborate versions of contextualism. This is achieved while relying solely on the independently motivated pragmatic machinery of conversational exculpature. Overall, skeptical invariantism comes out as a highly attractive position, not the last resort that it is often taken to be.

Acknowledgements I am grateful to Wayne Davis, Katharina Felka, Daniel Hoek, Miguel Hoeltje, Dirk Kindermann, Christos Kyriacou, Alex Steinberg, Emanuel

Knowledge and Loose Talk  295 Viebahn, Julia Zakkou and an anonymous reviewer for very helpful comments on earlier versions of this chapter.

Notes 1 Davis (2007) and Kyriacou (2017, 2018), for instance, defend moderate skeptical invariantism, where knowledge is not so rare after all. It is still rarer than ordinary knowledge ascriptions, so the worry above still arises and my discussion of this worry remains relevant. 2 On Carter’s (2019) dynamic semantic account of loose talk, expressions like “3pm” have a fixed static semantic value and a variable dynamic semantic value. I assume that skeptical invariantists want to hold both these semantic values fixed so that Carter’s account is unavailable to them. See also Dinges (MS) for independent concerns with Carter’s account. 3 For pertinent results from experimental philosophy, see, e.g., Gerken et al. (2020) and the references therein. 4 The existence of these effects is somewhat controversial, but see, e.g., Dinges and Zakkou (2020) and the references therein for supportive findings. 5 As I understand Lasersohn, loosely used sentences do not assert (or convey) anything other than what they express, and thus they are literal. They correspondingly aim to add obviously false or unjustified propositions to the common ground. This is permissible on the given norm, and we tolerate such propositions in the common ground, as long as their falsity is harmless, by way of a truth in the pragmatic halo. For a competing interpretation on which Lasersohn proposes a kind of non-literality, see Carter (2019: 7). My worries apply regardless. 6 Similar problems arise when we apply Lasersohn’s account to embeddings of (1) to (3); see, e.g., Carter (2019: 7) and Hoek (2019: 172). The relevance theoretic account of loose talk in Sperber and Wilson (1985) faces similar concerns because, on this account, loose talk can only weaken the proposition expressed. See Hoek (2018: 159). 7 The inference here is a bit compressed and the details depend on exactly which constraints on optimality we adopt and how they are ranked. I hope the inference is intuitive enough as it stands. See Dinges, MS for details. 8 Klecha (2018: 108–109) points out a rescue strategy for when the conveyed proposition is the empty set. But this rescue strategy is designed to deal with the specific phenomenon of unidirectionality, and it is of no avail in the present context. See Dinges, MS for details. 9 To be precise, Hoek (2018: 160) appeals to contextual presuppositions (following Simons, 2002) rather than pragmatic presuppositions. I use the latter notion because it is more familiar and seems to do the job as well. See below for more on pragmatic presuppositions. 10 Thanks to Wayne Davis. 11 Another potential scale would be {“know,” “know for sure,” “know with absolute certainty”}. Thanks to Christos Kyriacou. But the corresponding scale presupposition entails that S knowsS that p because, according to skeptical invariantism, “know” already has this demanding interpretation. It seems implausible that speakers typically presuppose that all available epistemic states are at least that strong; see the crime case above. 12 See Blome-Tillmann (2015) and Ichikawa (2017: 23n25) for further discussion.

296  Alexander Dinges

Bibliography Blome-Tillmann, M. (2014), Knowledge and Presuppositions (Oxford: Oxford University Press). ——— (2015), “Ignorance, presuppositions, and the simple view”, Mind, 124/496: 1221–1230. BonJour, L. (2010), “The myth of knowledge”, Philosophical Perspectives, 24/1: 57–83. Carter, S. (2019), “The dynamics of loose talk”, Noûs, Early View: 1–28. Conee, E. (2005), “Contextualism contested”, in M. Steup and E. Sosa (eds.), Contemporary Debates in Epistemology (Malden, MA: Blackwell Publishing), 47–56. Davis, W. A. (2007), “Knowledge claims and context. Loose use”, Philosophical Studies, 132/3: 395–438. DeRose, K. (1992), “Contextualism and knowledge attributions”, Philosophy and Phenomenological Research, 52/4: 913–929. ——— (2012), “Replies to Nagel, Ludlow, and Fantl and McGrath”, Philosophy and Phenomenological Research, 84/3: 703–721. Dinges, A. (MS), “Generalizing loose talk”. ——— (2016), “Skeptical pragmatic invariantism: Good, but not good enough”, Synthese, 193/8: 2577–2593. Dinges, A. and J. Zakkou (2020), “Much at stake in knowledge”, Mind & Language: 1– 21. https://doi.org/10.1111/mila.12300. Fantl, J. and M. McGrath (2009), Knowledge in an Uncertain World (Oxford: Oxford University Press). Gerken, M., J. Alexander, C. Gonnerman et al. (2020), “Salient alternatives in perspective”, Australasian Journal of Philosophy, 98 (4): 792–810. Grice, H. P. (1989), Studies in the Way of Words (Cambridge, MA: Harvard University Press). Hazlett, A. (2007), “Grice’s razor”, Metaphilosophy, 38/5: 669–690. Hoek, D. (2018), “Conversational exculpature”, The Philosophical Review, 127/2: 151–196. ——— (2019), “Loose talk, scale presuppositions and QUD”, Proceedings of the 22nd Amsterdam Colloquium, 171–180. Ichikawa, J. J. (2015), “Ignorance and presuppositions”, Mind, 124/496: 1207–1219. ——— (2017), Contextualising Knowledge: Epistemology and Semantics (Oxford: Oxford University Press). Kindermann, D. (2019), “Knowledge embedded”, Synthese, Online First 1–21. Klecha, P. (2018), “On unidirectionality in precisification”, Linguistics and Philosophy, 41/1: 87–124. Krifka, M. (2002), “Be brief and vague! and how bidirectional optimality theory allows for verbosity and precision”, in D. Restle and D. Zaefferer (eds.), Sounds and Systems. Studies in Structure and Change. A Festschrift for Theo Vennemann (Berlin: De Gruyter Mouton), 429–448. ——— (2007), “Approximate interpretations of number words. A case for strategic communication”, in G. Bouma, I. Krämer and J. Zwarts (eds.), Cognitive Foundations of Interpretation (Amsterdam: Koninklijke Nederlandse Akademie van Wetenschapen), 111–126.

Knowledge and Loose Talk  297 Kyriacou, C. (2017), “Bifurcated sceptical invariantism”, Journal of Philosophical Research, 42: 27–44. ——— (2018), “From moral fixed points to epistemic fixed points”, in C. Kyriacou and R. McKenna (eds.), Metaepistemology. Realism and Anti-Realism (Cham: Springer International Publishing), 71–95. ——— (2020). ‘Assertion and practical reasoning, fallibilism and pragmatic skepticism’. Acta Analytica 35:543–561 ——— (2019b), “Semantic awareness for skeptical pragmatic invariantism”, Episteme, First View: 1–19. Lasersohn, P. (1999), “Pragmatic halos”, Language, 75/3: 522–551. Lauer, S. (2012), “On the pragmatics of pragmatic slack”, Proceedings of Sinn und Bedeutung 16: 389–402. Levinson, S. C. (2000), Presumptive Meanings: The Theory of Generalized Conversational Implicature (Cambridge, MA: MIT Press). Lewis, D. (1979), “Scorekeeping in a language game”, Journal of Philosophical Logic, 8/1: 339–359. ——— (1996), “Elusive knowledge”, Australasian Journal of Philosophy, 74/4: 549–567. Ludlow, P. (2005), “Contextualism and the new linguistic turn in epistemology”, in G. Preyer and G. Peter (eds.), Contextualism in Philosophy. Knowledge, Meaning, and Truth (Oxford: Oxford University Press), 11–50. Moss, S. (2017), Probabilistic Knowledge (Oxford: Oxford University Press). ——— (2019), “Full belief and loose speech”, Philosophical Review, 128/3: 255–291. Roberts, C. (2012), “Information structure in discourse. Towards an integrated formal theory of pragmatics”, Semantics and Pragmatics, 5/6: 1–69. Simons, M. (2002), “Presupposition and relevance”, in Z. G. Szabó (ed.), Semantics vs. Pragmatics (Oxford: Clarendon Press), 329–355. Solt, S. (2014), “An alternative theory of imprecision”, Proceedings of SALT 24: 514–533. Sperber, D. and D. Wilson (1985), “Loose talk”, Proceedings of the Aristotelian Society, 86: 153–171. Stalnaker, R. C. (1978), “Assertion”, in P. Cole (ed.), Pragmatics (New York: Academic Press), 315–332.

15 Knowledge Claims and the Context of Assessment Wayne A. Davis

As defined in this volume, skeptical invariantism is the view that the meaning of ‘know’ is invariant, and so strong that it is very difficult or even impossible to know anything. To say that its meaning is contextually invariant is to say that a sentence of the form ‘S knows p’ expresses the same proposition and has the same truth conditions in any context of use or assessment. A well-known difficulty for any form of invariantism is that whether people make or accept a knowledge claim varies from context to context without any evident change in the subject’s belief, its truth, or its justification. This has led to (indexical) contextualism, on which an instance of ‘S knows p’ expresses a different relation between S and the proposition that p, and thus has different truth conditions, in different contexts of use. I will examine another leading alternative to skeptical invariantism: John MacFarlane’s (2005a, 2005b, 2014) assessment relativism, which consists of a relativist semantics for ‘S knows p’ and a set of norms. On MacFarlane’s semantics, there are contexts of assessments relative to which it is easy for instances of ‘S knows p’ to be true, and others for which it is difficult or impossible. I will argue that the assessment-relative semantics has little explanatory value, is subject to some of the objections to contextualism, and does not provide monadic truth conditions even when conjoined with an assessment-relative semantics for truth. Intensions varying with parameters that have no privileged value cannot represent propositions or the monadic truth predicate. I will argue briefly that MacFarlane’s norms are untenable given his semantics. What is relative to contexts of assessment, I suggest, are conditions of appropriateness. Like many other terms with strict and invariant meanings, loose uses of ‘know’ are appropriate in some contexts but not others.1

1 Relativist Semantics Like contextualism, MacFarlane’s relativism takes instances of ‘S knows p’ to have truth values that are relative in some way to contexts and vary with their epistemic standards. The epistemic standards S must meet are those of the assessor’s context, however, rather than the speaker’s

Knowledge Claims  299 context. MacFarlane’s relativism differs from contextualism in holding that an instance of ‘S knows p’ expresses the same proposition, and the same relation between S and the proposition that p, relative to every context of use. It thus differs markedly from ‘It is raining here,’ which describes different places when uttered in different contexts. Kaplan (1977: 494) distinguished the context of use from a circumstance of evaluation. A sentence like ‘I am a Berliner’ may be false as used by John F. Kennedy in a speech in 1961, but true as used there when evaluated in a hypothetical circumstance in which Kennedy resides in Berlin. Kaplan took circumstances (of evaluation) to be worldtime pairs, but allowed that additional parameters might be needed to evaluate some expressions. MacFarlane believes an epistemic standards parameter is needed to evaluate knowledge claims. MacFarlane (2014: 60) characterizes a context of assessment as “a possible situation in which a use of a sentence might be assessed.” MacFarlane’s rules assign truth values relative to circumstances whose worlds and times are those of the context of use and whose epistemic standards are those of the context of assessment. The principal question I wish to investigate is whether the truth value of ‘S knows p’ is relative to an epistemic standard. We can concentrate on this issue by restricting our attention to contexts of use and assessment characterized by the present time and the actual world. Circumstances will be differentiated only by their epistemic standards. We will assume with MacFarlane that (an instance of) ‘S knows p’ is neither indexical (beyond tense) nor ambiguous, and so expresses the same proposition K in all contexts of use or assessment. This will require assuming that ‘S’ and ‘p’ are themselves neither indexical nor ambiguous. The truth value of ‘S knows p’ will be the truth value of K. We will further assume with the relativist that for each proposition K, there is a unique function i(K) from world-time-standard triples to truth values. 2 On MacFarlane’s view, i(K) assigns T to ‹w, t, s› iff ‘p’ is true and S’s epistemic position meets standard s in w at t. S’s epistemic position is characterized mainly by the evidence S has for ‘p.’ But we also need to assume that S’s epistemic position with respect to ‘p’ meets s only if S believes p, and is not in a Gettier situation. Focusing on the actual world and the present time, and letting s(C) be the epistemic standard in context of assessment C, we can formulate a simple rule for truth relative to C. 1 Relative-Truth Rule: ‘S knows p’/K is true relative to C iff ‘p’ is true and S’s epistemic position with respect to ‘p’ meets s(C). Two additional constraints are necessary to define a relativist semantics.3 2 Relativist Constraints: a Positive: i(K) varies with s(C). b Negative: No s(C) is such that K is true iff K is true relative to s(C).

300  Wayne A. Davis The positive constraint means that instances of ‘S knows p’ are true relative to some assessment contexts and false relative to others because the epistemic standards in those contexts differ. The negative constraint means that no epistemic standard is privileged. No standard is distinguished from others the way the actual world is distinguished in modal logic. It is the combination of these independent constraints, I shall argue, that makes relativism untenable.

2 Relativist vs. Contextualist Semantics One objection to contextualism is its implication that a speaker B denying ‘S knows p’ is not expressing disagreement with a speaker A who affirms ‘S knows p’ in a context with different standards. For on contextualism, the sentence expresses different propositions in the different contexts. So B is denying something other than what A affirmed. It seems clear, however, that A and B are expressing disagreement.4 Relativism claims to provide the correct result because it takes ‘S knows p’ to express the same proposition in different contexts. Even relativists have worried about how well their theory accounts for disagreement, though, given that relativism also allows that both speakers’ claims may be true relative to their different contexts and that neither speaker has made a mistake.5 More on this in Section 3. Another objection to contextualism is that if it were true, ‘knows’ should block disquotational truth conditions the ways ‘here’ and other indexicals do.6 Contrast the following sentences: 3 a

‘It is raining here’ is true as used in any context iff it is raining here.

b

‘Strasbourg is in France’ is true as used in any context iff Strasbourg is in France.

Sentence (3)(a) is unambiguously false because it may be raining here in Washington even though ‘It is raining here’ is false as used in the Atacama desert. The sentence expresses different propositions when used in the Atacama then it does here on the right side of (3)(a). In contrast, (3)(b) seems unambiguously true (given our restriction to actual present contexts) as long as the sentence referred to on the left side has the same meaning that it has on the right. If we believe that Strasbourg is in France, and thus affirm the right side of (3)(b), then we cannot consistently deny that ‘Strasbourg is in France’ is true in some other context the way we can deny that ‘It is raining here’ is true in another context when we believe that it is raining here. Similar biconditionals with ‘know’ seem like (3)(b) rather than (3)(a). 4 ‘Sue knows Strasbourg is in France’ is true as used in any context iff Sue knows Strasbourg is in France.

Knowledge Claims  301 If we believe that Sue knows Strasbourg is in France, and thus affirm the right hand side of (4) to be true, it seems inconsistent for us to deny that ‘Sue knows Strasbourg is in France’ is true in any (actual, present) context. We cannot do it the way we can consistently deny that ‘It is raining here’ is true in another context when we believe that it is raining here. Relativism seems to avoid this problem because it takes ‘S knows p’ to express the same proposition in any context. Assessment relativism has a parallel problem, though. It allows that the proposition expressed by ‘Sue knows Strasbourg is in France’ may be true relative to assessment contexts with low epistemic standards while false relative to contexts with high epistemic standards. Let CL and CH be such contexts. Assuming that a biconditional is false relative to any context of assessment relative to which its components have different truth values, relativism rules that (5) below is false relative to CL .7 For relative to CL , the right side of (5) is true. Yet the left side is false relative to CL; for relative to any context of assessment, the knowledge claim is false relative to a context of assessment CH with a sufficiently high epistemic standard. 5 ‘Sue knows that Strasbourg is in France’ is true as assessed in any context iff Sue knows that Strasbourg is in France. If ‘Sue knows that Strasbourg is in France’ does express the same proposition in any context, then (5) should seem as undeniable as (3)(b) and (4). (4) and (5) thus provide strong evidence for invariantism. We will return to this in Section 4. The principal strength of contextualism is its ability to account for a number of standard test cases. Each consists of two contexts. The epistemic standard is low in one, high in the other; other things are equal. Knowledge is affirmed in the low-standard context, denied in the high-standard context. Both claims are natural and appropriate. 2.1 The Parking Case CL . Dick and Jane are discussing how often they forget where they parked their cars, and have to hunt for them. Jane asks, “Does Joe know his car is in the driveway?” Dick answers “Yes. He just told me that.” CH. In nearly identical circumstances, Jack and Jill are discussing car thefts, noting that cars have been stolen from even the safest neighborhoods. Jill asks, “Does Joe know his car is in the driveway?” Jack answers “No. He remembers parking it there, but hasn’t checked on it recently.” MacFarlane claims that relativism explains such variability as well as contextualism.

302  Wayne A. Davis Why is it that I’ll happily assert ‘Joe knows that his car is parked in his driveway’ when standards are low, and ‘Joe doesn’t know that his car is parked in his driveway’ when standards are high? The relativist semantics affords a simple explanation: the former sentence is true as used and assessed in a context where standards are low, and the latter is true as used and assessed in a context where the standards are high. (MacFarlane 2005a: 218–219)8 I believe this explanation is mistaken even if we grant that ‘Joe knows his car is in his driveway’ is true relative to s(CL), while its negation is true relative to s(CH). For on relativism, the sentence expresses the same proposition K D in both contexts. So any difference in the truth value of ‘Joe knows his car is in the driveway’ between CL and CH is not due to a difference in what is expressed. And while it is true that K D is assessed in a context where standards are low, namely CL , it is just as true that K D is assessed in other contexts where standards are high, including CH. If ‘knows’ has a relativist semantics, we should recognize in both CL and CH that K D is true relative to s(CL) and false relative to s(CH). How can relativism get from that knowledge to the prediction that we properly believe and assert K D when we happen to be in context CL rather than CH?9

3 In Situ Detachment MacFarlane’s explanation of the parking case seems to rely on the following reasoning, with ‘D’ abbreviating ‘His car is in the driveway’: 6 a b c d e

‘Joe knows D’ is true relative to CL . ∴ When assessed in CL , ‘John knows D’ is true. ‘Joe knows D’ is assessed by A in CL . ∴ ‘Joe knows D’ is true. ∴ A may assert ‘Joe knows D.’

Whether (e) follows from (d) can be questioned,10 but I am concerned with the reasoning from (a) and (c) to (d). Conclusion (d) may appear to follow, but only if we equivocate on (b). On one interpretation, (b) is a mere restatement of (a)—another way of saying that ‘Joe knows D’ is true relative to CL . On that interpretation, (b) obviously follows from (a). But if (b) is equivalent to (a), then (d) does not follow from (b) and (c). One way to see this is to note that on relativism, the premises of (6) are compatible with those of (7), even though their conclusions are contradictory: 7 a b c d

‘Joe knows D’ is not true relative to CH. ∴ When assessed in CH , ‘John knows D’ is not true. ‘Joe knows D’ is assessed by B in CH. ∴ ‘Joe knows D’ is not true.

Knowledge Claims  303 Line (6)(b) has another interpretation, however, on which it entails If ‘John knows D’ is assessed in CL , it is true. On this interpretation, (6)(d) clearly follows from (6)(b) and (c). But on this conditional interpretation, (6)(b) does not follow from (a). For (6)(a) is compatible with (7)(a).11 ‘Relative to CL’ cannot be detached from ‘true,’ even by someone who happens to be in CL . If K D is true (false) relative to C did mean or entail if K D is assessed in C, then it is true (false), as the quoted passage from MacFarlane may suggest,12 then relativism’s assumption that ‘Joe knows D’ expresses the same proposition in all contexts would lead to the result that the proposition is both true and not true given the relativist assumption that some people are in low-standards contexts and others in highstandards contexts. This contradiction cannot be avoided by asserting that ‘true’ is relative too. For we are deriving the contradiction in our context of assessment.13 An analogy might make it more obvious that truth relative to C does not entail truth if or when assessed in C. Suppose we use (8) to stipulate a meaning for ‘trall,’ which has no meaning in English. 8 a b

S is trall relative to group R iff S is taller than most members of R. ‘S is trall’ is true relative to context C iff S is taller than most people in C.

From these stipulations, A can never infer that Obama is simply trall, only that he is trall relative to a group. This is so even if A is in a group of violinists and knows that Obama is taller than most members of that group. The fact that A may also be in a group of basketball players, relative to which Obama is not trall, makes the point especially clear. Suppose we further stipulate that standards are high in R iff most members of R are over 6ʹ tall, and that standards are low in R iff most members of R are under 5ʹ tall. That would not enable us to assert that ‘Obama is trall’ is true when standards are low, only that the sentence is true relative to a group with low standards. The same sentence might be false relative to another group with high standards. Contextualists avoid contradiction by holding that ‘Joe knows D’ expresses different propositions in different contexts. Relativists can avoid contradiction by saying nothing about monadic truth beyond (2)(b). A stronger alternative is to say that knowledge claims “have truth values only relative to contexts” (MacFarlane 2007a: 23).14 Only implies that while the dyadic predicate ‘x is true relative to C’ applies to knowledge claims, the monadic ‘x is true’ does not.15 This alternative would be true, for example, if ‘Joe knows D’ were like ‘Obama is trall.’ Neither (b) nor (d) would be sanctioned in (6). But then any support for (6)(e) is gone. It is hard to see how this radical form of relativism could account for the intuitions cited by contextualists, which differ from one context to another.

304  Wayne A. Davis For the relative truth values do not differ (or appear to differ) from one context to another. This radical relativism would also entail that second-order knowledge claims are true relative to no contexts unless (1) is rejected. For (1) entails that ‘S knows that S knows p’ is true relative to C only if ‘S knows p’ is true (monadically). This problem can be avoided by replacing ‘p’ is true in (1) with ‘p’ is true relative to C. But then nothing remains of the facticity of knowledge. The inference from ‘p’ is true relative to C to ‘p’ is true or p is invalid.16 MacFarlane’s official view, however, is that relativism can provide an account of monadic truth for knowledge claims. Given (1) and (2), the question is “How?”

4 Monadic Truth and Semantics It might be suggested that the relativist can endorse Kaplan’s rule relating monadic truth to truth in a context: 9 An occurrence of a sentence, or the assertion made by uttering it, is true just in case the proposition expressed or asserted is true at the circumstance of the context.17 This rule is possible for Kaplan only because for every occurrence of a sentence, there is a unique context of use whose world is the actual world. On relativism, however, the relevant contexts are assessment contexts, of which there are many in the actual world. The occurrence may be true relative to some, false relative to others. Since the context (of assessment) would have a false uniqueness presupposition, the Kaplanian definition of monadic truth is unavailable.18 Montminy (2009: Section 8) has suggested using a modal analogy. A standard principle of modal logic equates monadic truth with truth in the actual world.19 10 P is true iff P is true in the actual world. This definition of monadic truth is possible because the actual world α is the unique world w in which whatever is true is true in w, and vice versa. But according to relativism, there are indefinitely many contexts in the actual world in which a proposition is assessed. ‘S knows p’ is true relative to some actual contexts of assessment, and false relative to others. ‘True’ cannot be equated with ‘true relative to the actual context of assessment,’ because that has a false uniqueness presupposition. And given the negative constraint on a relativist semantics, there is no standard s such that ‘S knows p’ is true iff it is true relative to ‹α, s›.

Knowledge Claims  305 MacFarlane (2011a: 442; 2014: 37–38, 93) takes the monadic truth predicate of natural languages to be governed by the “Equivalence Schemas” in (11). 11 a b

‘p’ is true iff p. The proposition that p is true iff p.

Thus ‘Joe knows his car is in the driveway’ and the proposition it expresses are true iff Joe knows his car is in the driveway. These equivalences hold provided the instances of ‘p’ have the same interpretation on both sides of ‘iff.’ ‘That is hot’ is true iff that is hot will be false if the first occurrence of ‘that’ is referring to a lion and the second is referring to a car (recall (3) above), or if ‘hot’ means “very warm” on the left and “stolen” on the right; but then the sentence is not an instance of the Equivalence Schema. 20 Because it is also necessary that ‘p’ be non-paradoxical, the schemas cannot be considered full definitions of truth. Moreover, two sentences or propositions may be equivalent even though the fact represented by one in no way explains or determines the fact represented by the other, particularly if the equivalences are material. Consider, for example, ‘2 is the square root of 4’ is true iff 2 is the cube root of 8, or ‘London is in England’ is true ≡ Berlin is in Germany. The Equivalence schemas are important because they hold in virtue of what it is for a non-paradoxical sentence or proposition to be true (i.e., to satisfy the monadic truth predicate of natural languages). For a sentence or proposition to be true is for what it says to be the way things are—for the sentence or proposition to “correspond to reality” in a non-tendentious sense. ‘2 is the square root of four’ is true because the sentence says that 2 is the square root of four and the number is the square root of four. The sentence is not true in virtue of the fact that 2 is the cube root of eight. Because the equivalence schemas hold in virtue of the meaning of ‘true,’ it is not only false but incoherent to say something like Joe knows D, but the proposition that Joe knows D is not true (cf. MacFarlane 2014: 37). The same account can be given of what it is for a proposition to be true in (relative to) a possible world, circumstance, context, or story w. For ‘Joe knows D’ to be true in w is for Joe to know D in w—for things in w to be the way the sentence says they are. Hence ‘p’ is true in w iff p in w is another Equivalence Schema. Its grounding principle is incompatible with relativism. Relativism holds that a proposition can be true relative to one circumstance but not another when their standards differ even if their worlds are the same. A proposition can clearly correspond to the way things are in one circumstance but not another when their worlds are different. Then the way things are in those circumstances is the way things are in different possible worlds. But if two circumstances have the

306  Wayne A. Davis same world, then the way things are in those circumstances is the same. Any differences in their epistemic standards will have no bearing on the truth of a proposition unless it says something about those standards. Even on relativism, the proposition that Joe knows D says nothing about the epistemic standards of any circumstance of evaluation. So whether it corresponds to the way things are in a circumstance does not depend on its epistemic standards. MacFarlane (2014: 129) claims that “even if we specify a state of the world, there is no saying whether the proposition is true until we specify the relevant [standard].” But a possible world is simply a complete specification of the way things are or could be. In any possible world, Joe either knows D or he does not. So things either are the way the proposition that Joe knows D says they are or they are not. The world completely determines whether or not the proposition is true. That is why (4) is true (for actual contexts). The contextualist differs from MacFarlane in maintaining that ‘Joe knows D’ says different things in different contexts. Whether the proposition it expresses in a given context is true is completely determined by the way things are in that context. So contextualism does not face the same problem. The semantics of a nonindexical predicate is standardly represented in a formal semantics by an intension, a function assigning an extension to every circumstance of evaluation. ‘True’ applies to propositions (and sentences expressing them), so its intension is a function from circumstances to sets of propositions (or sentences expressing them). Specifically, the intension of ‘true’ assigns to circumstance E the set of propositions true at E. Any adequate semantics will assign the proposition that P is true to the extension of ‘true’ at E iff it assigns P to the extension of ‘true’ at E, at least when P is non-paradoxical. A standard rule for ‘iff’ will ensure that the propositions expressed by instances of the Equivalence Schemas are in the extension of ‘true’ at every E, yielding (12): 12 ‘p’ is true iff p is true at every E. If what is true at E is determined by the world of E, then the truth of instances of (11)(a) will follow from (12) via (10): if a proposition is true at every circumstance, it is true at the circumstance with the actual world, and thus is true. MacFarlane believes that the relativist can adopt this same approach. [I]t is easy to give a semantics for monadic “true” and “false” that works in an analytic relativist framework and ratifies the disquotational inferences. 21 Roughly stated: “true” expresses a property, truth, whose extension at a circumstance of evaluation is the set of propositions that are true-at that circumstance of evaluation. (MacFarlane 2011a: 442)

Knowledge Claims  307 Given this semantics for “true,” every instance of the Equivalence Schema will be true at every circumstance of evaluation…. (MacFarlane 2014: 93)22 If the intension of a knowledge claim varies with a parameter other than worlds, then the intension MacFarlane’s semantics assigns to ‘true’ must also vary with a parameter other than worlds. If the truth predicate is genuinely monadic, however, then it cannot be represented by an intension varying with a parameter other than worlds. 23 Let us say that a function from circumstances to extensions is open if it varies with a parameter that has no privileged value. 24 Otherwise the intension is closed. If worlds are the only parameters, then intensions are automatically closed. If there are other parameters, an intension can still be closed as long as it does not vary with the non-world parameters. To see why assigning an open intension does not suffice to represent the semantics of a monadic predicate, recall (8). Suppose we assign ‘trall’ the intension θ whose value at any circumstance ‹w, R› is the set G iff G is the set of all objects taller than most members of R in w. This may suffice for us to understand (13): 13 Obama is trall relative to U.S. presidents. Given θ, we can plausibly infer that (13) is true iff Obama is taller than most U.S. presidents. It seems reasonable to assume a semantics for phrases of the form ‘Adj relative to R’ that delivers (13) when θ is assigned to ‘trall.’ But θ does not enable us to understand (14): 14 Obama is trall. We have no idea what (14) would convey, what set it says Obama belongs to. For (14) does not in any way indicate a reference group to which Obama is being compared. The intension θ stipulated for ‘trall’ does not single out any one reference group. We who are assessing (14) belong to countless reference groups, as does Obama, but nothing in θ makes any of them relevant. The best we could say about (14) is that it is incomplete: it does not express a proposition or complete thought. The verb phrase ‘is trall’ is itself incomplete. To form a monadic predicate, ‘is trall’ must be combined with a prepositional phrase specifying a reference group, like ‘relative to U.S. presidents.’ A monadic predicate must be represented by a closed intension. If a semantics could account for the Equivalence Schemas the way MacFarlane says, then (15) would be ratifiable by assigning the proposition expressed by ‘Obama is trall’ to the extension of ‘true’ at just those circumstances ‹w, R› at which ‘Obama is trall’ is true, making (15) true at every circumstance ‹w, R›.

308

Wayne A. Davis

15 ‘Obama is trall’ is true iff Obama is trall. But (15) has no intelligible content. If a semantics assigns θ to ‘trall,’ ‘Obama is trall’ does not express a proposition. The semantics therefore cannot assign the proposition expressed by ‘Obama is trall’ to the extension of ‘true’ at any E, and does not generate the absurd result that (15) is true at every E. A semantics for ‘true’ will be as inadequate as that for ‘trall’ if its meaning is represented by a function from circumstances to extensions that is open because it varies with a standards parameter. But the intension MacFarlane proposes for truth will have to be open if the intensions for knowledge claims are open. On MacFarlane’s relativist semantics for ‘knows,’ for example, the following may both be true: 16 a b

K D is true relative to ‹α, s(CL)›. K D is not true relative to ‹α, s(CH)›.

If the extension of ‘true’ at a circumstance is the set of propositions true relative to that circumstance, then the following must also both be true: 17 a b

‘K D is true’ is true relative to ‹α, s(CL)› ‘K D is true’ is not true relative to ‹α, s(CH)›.

This means that the intension for ‘true’ on a relativist semantics must be open. As we saw with ‘trall,’ an open intension cannot represent the semantics of a monadic predicate. If the intension of ‘true’ were open, ‘K D is true’ would be like ‘Obama is trall,’ and would not express a proposition. For the very same reasons, an open intension cannot adequately represent the semantics of an instance of ‘S knows p’ if it expresses a proposition. An open intension might tell us when the proposition is true relative to a standard, but not when it is true. An intension varying with non-world parameters may represent a propositional function, predicate concept, or incomplete proposition. But only a closed intension can represent a proposition, something that can be believed or disbelieved and true or false. The intension relativists propose for Joe knows his car is in the driveway could only represent it as having the meaning of the open sentence Joe knows his car is in the driveway relative to s, or the indexical sentence-type Joe knows relative to this standard, not something that is truth-evaluable and capable of being believed or disbelieved.25 An open intension would similarly fail to represent the content of ‘S knows p’ as an object of agreement or disagreement. If ‘S knows p’ has an assessment-sensitive semantics, then two people can no more disagree about whether Joe knows his car is in the driveway than they can about whether Obama is trall.

Knowledge Claims  309

5 Temporal Relativity I have argued that the semantics of knowledge claims and the monadic truth predicate cannot be adequately represented by assigning them intensions that vary with a standards parameter. Would parallel reasoning entail that circumstances of evaluation cannot have a time parameter, despite common practice? If so, would that cast doubt on my arguments against relativism? One reason for a time parameter is that terms or predicates typically have different extensions when evaluated at different times in the same world. Because different people are senators at different times, it makes no sense to speak of the extension of ‘senator’ unless a particular time is indicated (or unless time is being ignored for simplicity). There is no time τ such that the extension of ‘senator’ (simpliciter) = the extension of ‘senator’ at ‹α, τ›. The semantics of ‘senator’ must be represented by an open intension. This provides no support, however, for the relativist thesis that the semantics of knowledge sentences and the monadic truth predicate can be represented by intensions with a standards parameter. MacFarlane (2009: Section 2; 2014: 94) offers “temporalism” as a relevant analogy, as does Kölbel (2009: Section 2), and relies on its coherence to rebut objections to relativism. Temporalism resembles indexicalism in holding that the truth values of occurrences of a sentence like ‘The sun is setting’ vary with the time of utterance, while resembling relativism in holding that all occurrences express the same proposition; call it ‘T.’ While I can only sketch the reasons here, I believe the coherence of temporalism is questionable.26 MacFarlane takes an occurrence of ‘The sun is setting’ to be true iff T is true relative to the time of the utterance.27 But if T is true relative to 6:00 pm and false relative to 7:00 pm, it must refer to no particular time, and so can be neither true nor false simpliciter. Then it is hard to see how a sentence expressing it could have a monadic truth value. ‘The sun is setting’ is true but the proposition that the sun is setting is not true seems incoherent, in conflict with the Equivalence Schemas (Section 4). And if I know that T is neither true nor false, how could I either believe or disbelieve it? If we grant that T can be an object of belief, other problems emerge. We noted in Section 2 that relativism claims to be superior to contextualism in its ability to account for genuine disagreement. Temporalism, on the other hand, predicts disagreement when there is none. Suppose I am looking out the window as the sun sets, and tell my wife in a different time zone that the sun is setting. Looking out her window, my wife will believe that the sun is not setting. On temporalism, we disagree because there is a proposition I believe that she disbelieves. But there is no disagreement here at all.28 Similarly, I can properly be described as informing my wife that the sun is setting. But on temporalism, I did not because she knows it is not setting (cf. Baghramian and Carter 2019: 67).

310  Wayne A. Davis Suppose my wife is with our son, who mistakenly believes that the sun is setting. One argument for temporalism is that even though it is incorrect to say that my son and I agree that the sun is setting, there is a clear sense in which he and I believe the same thing (MacFarlane 2014: 50). There is also a clear sense, though, in which we believe different things, which follows from the fact that what I believe is true and what he believes is false. ‘The same thing’ in the first case means “the same type of thing.” In that sense, two people who say “I love you” say the same thing even though they express different beliefs. The problem for temporalism is its implication that my son and I believe one and the same proposition, not just two propositions of the same type. In that case, what we believe could not differ in truth value. The occurrences of ‘The sun is setting’ in (18) have important differences: 18 a b c

The sun is setting. The sun is setting at 6:00 pm, September 1, 2020. The sun is setting at 7:00 pm, September 1, 2020.

(18)(a) is a complete sentence equivalent to ‘The sun is setting now,’ which is true when used at some times and not others.29 The same word sequence is not a complete sentence in (18)(b), and does not have a truth value, although (18)(b) itself does. (18)(a) is a temporal indexical, expressing different propositions in different contexts. In any context of use, (18)(a) is used to say that the sun is setting at a particular time indicated by the context. If I uttered it now, it would be true iff the sun is setting at the present time (9:30 am, March 29, 2021). ‘The sun is setting’ has the same meaning in (18)(b) as it has in (18)(c), but does not express a complete proposition. In (18)(b) and (c), ‘The sun is setting’ expresses part of a proposition that is not itself a proposition. We could describe the non-propositional part as true relative to a time, but it has no monadic truth value and is not an object of belief or disbelief, agreement or disagreement. In these respects, ‘The sun is setting’ in (18)(b) and (c) is like the incomplete sentence ‘The sun is setting at,’ and more like the open sentence ‘The sun is setting at t’ than (18)(a).

6 Assessment, Assertion, and Retraction Rules Perhaps the relativist can explain the variation of knowledge claims via relativist rules for proper assessment and assertion. [K]nowledge claims are always properly assessed in light of the standards in play at the assessor’s current context. (MacFarlane 2005a: 219; see also 2011a: 444; 2014: 116, 188)

Knowledge Claims  311 Assert φ in c only if φ is true as used in c and assessed from c. (MacFarlane 2012: 140; see also 2014: 103–104) I believe MacFarlane is proposing the following: 19 Assessment Rule: When in C, A should assess ‘S knows p’ as true iff ‘S knows p’ is true relative to C. Given (1), this says that when in C, A should assess ‘S knows p’ as true iff ‘p’ is true and S’s epistemic situation with respect to ‘p’ meets s(C). In the parking case, the assessment rule tells us that because Dick is in CL , he should take K D to be true, while Jack should take –K D to be true because he is in CH. Assuming that Dick and Jack assert what they assess to be true, MacFarlane’s assessment rule seems to explain the difference in their knowledge claims.30 By making truth relative to a context the norm for asserting or assessing a knowledge claim in the context, MacFarlane (2005b: 321; 2011a: 444; 2012: 138–141) believes he has made relativism a testable semantic hypothesis. This is questionable. First, MacFarlane’s relativist semantics plays no role in the derivation of predictions. Rule (19) does not depend on the Relativist Constraints, and can be accepted when (2)(a) and (b) are rejected. Hence (19) is compatible with skeptical invariantism. Second, the arguments presented above show that (19) would not be rational given MacFarlane’s relativist semantics. Given the Relativist Constraints, (19) would endorse the fallacy of in situ detachment (Section 3). The open intensions assigned to instances of ‘S knows p’ by a relativist semantics represent propositional functions or predicates rather than propositions. Open intensions do not represent objects of belief or bearers of monadic truth values. On relativism, therefore, instances of ‘S knows p’ do not express contents that can be assessed as true or false, nor anything whose truth or justification we can commit ourselves to or defend. Since the difference between contexts of use and assessment play no role in (19), MacFarlane recognizes that a contextualist could also accept it. What distinguishes relativism from contextualism, he maintains, is a retraction rule, according to which A should be committed to retracting a knowledge claim in a subsequent context of assessment if it is false relative to that context (MacFarlane 2012: 140; 2014: 108–116, 191–192). One of the well-known objections to contextualism is that such a change would not prompt or warrant retraction because ‘S knows p’ would simply assert a different proposition in the new context (Section 2). MacFarlane’s retraction rule, however, is problematic given the Relativist Constraints for the same reasons his assertion and assessment rules are. Even if those problems are waived, there is no evident reason why

312  Wayne A. Davis A should be committed in Cu to retract in Ca an assertion (or assessment) made in Cu when A knows it will remain true in Cu (Marques 2014: 370–371). And if the falsity of something relative to Ca would be a reason to retract it, it should be a reason not to assert it (or assess it to be true) in the first place.31

7 Appropriateness Relativity We have been exploring problems with the relativist thesis that an instance of ‘S knows p’ expresses a proposition that may be true relative to one context of assessment and false relative to another because the epistemic standards differ. Is anything other than truth dependent on the epistemic standard characterizing an assessment context? In Davis (2007, 2015), I explain the variation in knowledge claims in the parking and other test cases in terms of variably strict usage. I believe that ‘S knows p’ has invariant truth conditions that are strict but satisfiable: S’s certainty that p is completely and non-defectively justified. S’s certainty that p is “completely” justified provided it is either self-evident or based on evidence sufficient to establish the truth of ‘p’ in any actual context. I believe that ‘Joe knows his car is in the driveway’ is strictly speaking false and its negation true in both contexts of the Parking Case. Joe has good evidence that his car is in the driveway: he remembers parking it there and thefts in his neighborhood are rare. But his belief is not completely justified: someone might have stolen it after he parked it. I  argue that the speakers in the low-standards contexts of the test cases are speaking loosely. The sentences they use express the same propositions in all contexts, but the speakers imply different propositions in different contexts. The speakers say the same thing in all contexts, but mean different things. What Dick means in CL is that he is close enough to knowing for the purposes of C L . That is true. The purposes of the high-standard context require greater evidence than those of the low, so Jack denies in CH what Dick affirms in CL , thereby implicating that Joe is not close enough to knowing for the purposes of CH. That is also true. Loose use is a common form of implicature, similar to hyperbole but more ubiquitous. Essential to hyperbole is the intention that one’s audience recognize that what is said is strictly speaking false. Knowledge claims in lottery cases are usually hyperbolic, but those in the parking case are not. Loose use may or may not be appropriate. If Jack speaks as loosely in CH as Dick does in CL , Jack’s knowledge claim would be inappropriate. The purpose of the conversation in CH is to rule out the possibility that Joe’s car has been stolen. The evidence Joe has that his car is in the driveway does not make Joe close enough to knowing for that purpose. So it would be inappropriate for Jack to use ‘Joe knows his car’s in the driveway’ loosely in CH. The standards of evidence “in play” determine the appropriate level of strictness, not the truth value, of knowledge claims.

Knowledge Claims  313 Thus Dick’s saying that Joe knows he won’t win the lottery would be appropriate as loose use in CL , but what Dick said would be strictly speaking false. Even when we are in a high-standard context, we should judge that it was appropriate for Dick to say that Joe knows his car’s in the driveway because Joe was close enough to knowing for Dick’s purposes in CL . But in a high-standards context, we should not assert that Dick’s claim was true or assess it to be true. For what he said was not strictly speaking true, and Joe was not close enough to knowing for our purposes in CH for us to say even loosely that Dick’s statement was true. MacFarlane objects to a loose use account. [T]he loose use strategy requires that speakers are normally aware that their knowledge claims are not strictly true. If that were so, however, skepticism would be universally accepted as true but uninteresting. And it isn’t. (MacFarlane 2011c: 541; 2014: 179) But loose use is not hyperbole. It is perfectly normal for people to speak loosely when they do not know or care whether their claims are strictly true. When I say “It is 8:20” because my watch reads “8:20,” for example, I do not believe it is exactly 8:20, yet nor do I believe it is not. I do not know how accurate the watch is. What I believe is that the time is within a minute or so of 8:20. My purposes rarely require greater precision. More importantly, the invariantist should not maintain that all variation in knowledge claims has the same source as that in the parking and other standard test cases. It is thoroughly implausible, in my view, that when people say “I know I have a hand,” they are speaking either loosely or falsely. If they are persuaded to say otherwise in an epistemology class, the transformation is much more profound than what happens to Hannah in DeRose’s (1992: 913) bank case. It is more like what happens in a philosophy of religion class when theists become atheists. Finally, if knowledge claims had a relativist semantics, the most we could say about skepticism is that it is true relative to skeptical standards and false relative to non-skeptical standards. That would be uninteresting. Is my strict invariantism a skeptical invariantism? As ‘skeptical invariantism’ is defined in this volume, that depends on whether the truth conditions I indicated for ‘S knows p’ are so strong that it is “very difficult or impossible to know anything.” I do deny that many of our knowledge claims are strictly speaking true, such as those in the low-standards context of the Parking Case. But I do not believe it is difficult or impossible to know things if knowing requires complete and non-defectively justified certainty. I am certain that I have a hand now, that I either do or do not have a hand, that I am thinking about whether I have a hand, and that countless other things are the case. I believe my certainty in these cases is completely and non-defectively justified. It is quite easy to

314  Wayne A. Davis get such knowledge.32 So while my view may entail that there are many things we do not know that others believe we do know, it does not entail that it is very difficult or impossible to know anything, or that we know very little, without assumptions that I believe are unjustified.

Notes 1 An early version of this paper was presented at the Context, Perspectives and Relative Truth Conference held at the University of Bonn, June 9–12, 2011. I would like to thank Christos Kyriacou, John Greco, members of the Bonn audience, and anonymous reviewers for many helpful comments. 2 See Egan, Hawthorne, & Weatherson (2005: 157ff), MacFarlane (2005a: 223; 2009: 236–237; 2011c: 536, 537; 2012: 135; 2014: 188–189) and Kölbel (2008: 16–17, 24; 2009: §2). 3 Zimmerman (2007: 316), Kölbel (2009: 383–384, 386–387), MacFarlane (2014: 49, 61–62, 64–65, 67, 188–189), and Baghramian & Carter (2019: 3–6, 37). 4 Cf. Hawthorne (2004: 520), Richard (2004: 215ff), Stanley (2005: 115, 144), Williamson (2005: 220–221), McFarlane (2005a: 202–203, 209–210; 2007: 18–21; 2011c: 540; 2014: 131, 181), Davis (2007: 428–430; 2015), Kölbel (2008: 23–24; 2009: 390–392), Cappelen & Hawthorne (2009: 54– 57), Baghramian & Carter (2019: 54), and Marques (2014: 363). Contrast Cohen (1999: 65, 80, 83) and DeRose (1992: 924; 1995: 46ff; 2009: 49–51, 158–159, 177). See Brogaard (2009: 221) for a related problem. 5 See Richard (2004: 230, 239), Stanley (2005: 147), MacFarlane (2007a; 2014: 136), Montminy (2009: 352–353), Cappelen & Hawthorne (2009: 14–16; 126–128), Kölbel (2009: 390–391), Ferrari (2016: 522–523), and Baghramian & Carter (2019: 12–13). 6 See Cappelen & Lepore (2005: 105ff), Egan, Hawthorne, & Weatherson (2005: 139), Cappelen & Hawthorne (2009: 11, 20), and Davis (2013a: §2). 7 Egan, Hawthorne, & Weatherson (2005: 159–160) observe that relativism has such “awkward” consequences. See also MacFarlane (2014: 199). 8 See also Kölbel (2009: 391). Egan, Hawthorne, & Weatherson (2005: 152ff) make similar claims for relativism with respect to epistemic modals and relative terms such as ‘tastes great’ and ‘huge,’ as do Kölbel and MacFarlane. 9 Cf. Evans (1985: 349–350), Zimmerman (2007: 322, 335–336), Marques (2014: 365), and Baghramian & Carter (2019: 35–36). 10 See for example Williamson’s (2000: Ch. 11) knowledge norm of assertion and Grice’s (1989: 27) two submaxims of Quality. 11 John knows D’ is true in C H ’ has the same ambiguity as (6)(b). Cf. Dever (2015: 615–616) on at. 12 See also Montminy (2009: 351–353), and MacFarlane’s (2011a: 443, 446 fn 6) suggestion that a relativist could accept the explanatory priority of monadic truth by taking ‘x is true at a context c’ to mean something like “x would express a true proposition if used at c.” 13 Compare and contrast Richard (2004: 225, 235). 14 See also Brogaard (2009: 216–7, 219; 2002: 11), MacFarlane (2011a: 445; 2011c: 536, 543; 2012: 135; 2014: 67, 189), and Baghramian & Carter (2019: 1, 8). 15 It could also mean that ‘x is true’ is always elliptical for ‘x is true relative to C’ (Baghramian & Carter 2019: 8, 31, 54–55). But I do not call such a predicate “monadic.”

Knowledge Claims 315 16 Cf. Stanley (2005: 145–147). MacFarlane (2014: 194) responds that relativism can account for facticity because ‘p is true relative to C’ follows from ‘S knows p is true relative to C.’ But that is a different inference, from which facticity does not follow. 17 See Kaplan (1989: 522), Richard (2004: 233), MacFarlane (2007b: 246; 2014: 77), Kölbel (2008: 7–8, 18, 24), and Brogaard (2009: 216–7). 18 Kölbel (2008: 20, 24; 2009: 386–387), Brogaard (2009: 219), and MacFarlane (2011c: 538–539, 542; 2014: Section 4.7). 19 See e.g. Kripke (1963: 71), Crossley & Humberstone (1977: 16), Evans (1985: 351), Nelson & Zalta (2012: 154), Davis (2013c: Section XI), and MacFarlane (2014: 49). 20 One way to make the interpretation requirement explicit is to say that ‘p’ is true on i iff pi is true and The proposition that pi is true iff pi . 21 The “disquotational inferences” are those of the form ‘p’ is true, ∴ p licensed by the Equivalence Schema. 22 See also Richard (2004: 231), MacFarlane (2009: 243; 2011b: Section 6.4; 2014: 37–38, 93, 106–107, 197), and Cappelen & Hawthorne (2009: 13–14). 23 In Davis (2013b: Section III) I argued against MacFarlane’s “nonindexical contextualism,” and his related assumption that a three-place relation (e.g., S’s knowing p at t) can be represented by an intension that varies with a fourth parameter other than a world. 24 I assume for simplicity that circumstances have exactly one world parameter. 25 Cf. Zimmerman (2007: 322–323) and Cappelen & Hawthorne (2009: 134–135). 26 Cf. Richard (1981, 2003), Evans (1985: 348–352), Zimmerman (2007: 323, 335, 342), and Baghramian & Carter (2019: 67). For a brief introduction, see Speaks (2019: §2.3.2). For an extended defense of temporalism (ineffective in my view), see Brogaard (2012). 27 See also Kölbel (2008: 2–4, 17–18), MacFarlane (2011c: 538; 2012: 136; 2014: 50) and Brogaard (2009: 216–217). Compare and contrast Kaplan (1977: 502–504) and Cappelen & Lepore (2005: 11, 61, 155). 28 Cf. Dever (2015: 610) and Ferrari (2016: 526). Brogaard (2012: 73–76) considers a similar case, in which, she insists, there is no “interesting disagreement” because the speakers are in different conversational contexts. In my case, they are conversing with each other. 29 As Perry (1979) showed, this is not to say that ‘the sun is setting’ expresses the same thought or belief as ‘The sun is setting at 6:00 pm’ when it is uttered at 6 pm. Contrast Brogaard (2012: 12, 17, 28). 30 The suggested norm for assertion is problematic in denying that it is ever permissible to assert what is false, even if one is fully justified in believing it or has a compelling reason to lie. But these are orthogonal concerns. 31 MacFarlane (2014: Section 12.1) might appear to address this objection. But what he tries to argue is that it can be rational to make an assertion one will be obliged to retract when one comes to occupy a relevantly different context. An implication of Section 3 is that A does not have to change contexts to recognize that a proposition is false relative to another context. 32 I also believe it is easy to know that many things are not known. I know that I do not know I have three hands, for example, and that I do not know I have both two and three hands. Knowing what we don’t know is often very important.

316  Wayne A. Davis

References Baghramian, M. & Carter, A. J. (2019) Relativism. In The Stanford Encyclopedia of Philosophy, (Spring 2021 Edition). E. N. Zalta, ed. URL = Brogaard, B. (2009) Introduction to Relative Truth. Synthese, 166, 215–229. Brogaard, B. (2012) Transient Truths. New York: Oxford University Press. Cappelen, H. & Hawthorne, J. (2009) Relativism and Monadic Truth. Oxford: Oxford University Press. Cappelen, H. & Lepore, E. (2005) Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act Pluralism. Oxford: Blackwell Publishing. Cohen, S. (1999) Contextualism, Skepticism, and the Structure of Reasons. Philosophical Perspectives, 13, 57–89. Crossley J. N. & Humberstone, L. (1977) The Logic of “Actually”. Reports on Mathematical Logic, 8, 11–29. Davis, W. A. (2007) Knowledge Claims and Context: Loose Use. Philosophical Studies, 132, 395–438. Davis, W. A. (2013a) Minimizing Indexicality. Philosophical Studies, 168, 1–20. Davis, W. A. (2013b) On Nonindexical Contextualism. Philosophical Studies, 163, 561–574. Davis, W. A. (2013c) The Semantics of Actuality Terms: Indexical vs. Descriptive Theories. Noûs, 49 (3): 470–503. Davis, W. A. (2015) Knowledge Claims and Context: Belief, Philosophical Studies, 172, 399–432. DeRose, K. (1992) Contextualism and Knowledge Attributions. Philosophy and Phenomenological Research, 52, 913–929. DeRose, K. (2009) The Case for Contextualism: Knowledge, Skepticism, and Context, Vol. 1. Oxford: Oxford University Press. Dever, J. (2015) Eternalism, Temporalism, and Neutralism. Inquiry, 58, 606–616. Egan, A., Hawthorne, J. & Weatherson, B. (2005) Epistemic Modals in Context. In Contextualism in Philosophy: Knowledge, Meaning, and Truth, G. Preyer & G. Peter, eds. Oxford: Oxford University Press, 131–168. Evans, G. (1985) Does Tense Logic Rest on a Mistake? In Collected Papers, A. Phillips, ed. Oxford: Clarendon Press, 343–363. Ferrari, F. (2016) Assessment-Sensitivity. Analysis, 76, 516–526. Grice, H. P. (1989) Studies in the Way of Words. Cambridge: Harvard University Press. Hawthorne, J. (2004) Replies. Philosophical Issues, 14, 510–523. Kaplan, D. (1977) Demonstratives. In Themes from Kaplan, J. Almog, J. Perry, & H. Wettstein, eds. (1989). Oxford: Oxford University Press, 481–563. Kölbel, M. (2008) Introduction: Motivations for Relativism. In Relative Truth, M. García-Carpintero & M. Kölbel, eds. Oxford: Oxford University Press, 1–38. Kölbel, M. (2009) The Evidence for Relativism. Synthese, 166, 375–395. Kripke, S. (1963) Semantical Considerations for Modal Logics. Acta Philosophica Fennica, 16, 83–94. Reprinted in Reference and Modality, ed. L. Linsky, pp. 63–72. Oxford: Oxford University Press, 1971.

Knowledge Claims  317 MacFarlane, J. (2005a) The Assessment Sensitivity of Knowledge Attributions. In Oxford Studies in Epistemology, Vol. 1, T. Gendler & J. Hawthorne, eds. Oxford: Oxford University Press, 197–233. MacFarlane, J. (2005b) Making Sense of Relative Truth. Proceedings of the Aristotelian Society, 105, 321–339. MacFarlane, J. (2007a) Relativism and Disagreement. Philosophical Studies, 132, 17–31. MacFarlane, J. (2007b) Semantic Minimalism and Nonindexical Contextualism. In Context-Sensitivity and Semantic Minimalism, G. Preyer & G. Peter, eds. Oxford: Oxford University Press, 240–250. MacFarlane, J. (2009) Nonindexical Contextualism. Synthese, 166, 231–250. MacFarlane, J. (2011a) Simplicity Made Difficult. Philosophical Studies, 156, 441–448. MacFarlane, J. (2011b) Epistemic Modals Are Assessment-Sensitive. In Epistemic Modality, B. Weatherson & A. Egan, eds. Oxford: Oxford University Press, 144–178. MacFarlane, J. (2011c) Relativism and Knowledge Attributions. In Routledge Companion to Epistemology, S. Bernecker & D. Pritchard, eds. New York: Routledge, 536–544. MacFarlane, J. (2012) Relativism. In Routledge Companion to the Philosophy of Language, G. Russell & D. Graf, eds. New York: Routledge, 132–142. MacFarlane, J. (2014) Assessment Sensitivity: Relative Truth and its Applications. Oxford: Oxford University Press. Marques, T. (2014) Relative Correctness. Philosophical Studies, 167, 361–373. Montminy, M. (2009) Contextualism, Relativism and Ordinary Speakers’ Judgments. Philosophical Studies, 143, 341–356. Nelson, M. & Zalta, E. (2012) A defense of contingent logical truths. Philosophical Studies, 157, 153–162. Perry, J. (1979) The Problem of the Essential Indexical. Noûs, 13, 3–21. Richard, M. (2003) Introduction to Part I. In Time, Tense, and Reference, A. S. Jokic & Q. Smith, eds. Cambridge, MA: MIT Press, 25–45. Richard, M. (2004) Contextualism and Relativism. Philosophical Studies, 119, 215–242. Rysiew, P. (2001) The Context-Sensitivity of Knowledge Attributions. Noûs, 35, 477–514. Speaks, J. (2019) Act Theories and the Attitudes. Synthese, 196, 1453–1473. Stanley, J. (2005) Knowledge and Practical Interests. Oxford: Oxford University Press. Williamson, T. (2000) Knowledge and its Limits. Oxford: Oxford University Press. Williamson, T. (2005) Contextualism, Subject-Sensitive Invariantism, and Knowledge of Knowledge. Philosophical Quarterly, 55, 213–235. Zimmerman, A. Z. (2007) Against Relativism. Philosophical Studies, 133, 313–348.

Contributors

Nevin Climenhaga (Australian Catholic University) Is a Senior Research Fellow in the Dianoia Institute of Philosophy at Australian Catholic University. He received his PhD in 2017 from the Philosophy Department at Notre Dame, with a graduate minor in History and Philosophy of Science. He joined ACU in 2018, after completing a one-year postdoc at Notre Dame as the Managing Editor for Studies in History and Philosophy of Science. He has published articles on probability, confirmation, inference to the best explanation, causal inference, the use of intuitions in philosophy, and the problem of evil. Annalisa Coliva (University of California, Irvine) Is a Professor of Philosophy at University of California, Irvine. Her research focuses on epistemology, philosophy of mind and history of analytic philosophy. She has published widely on G. E. Moore and L. Wittgenstein, on scepticism, relativism and self-knowledge, as well as on perception, concepts and the first person. She is Chair of the Department of Philosophy at UC, Irvine, Co-director of the Minor in Medical Humanities and Deputy director of COGITO Research Centre (Bologna). Wayne A. Davis (Georgetown University) Is a Professor of Philosophy at Georgetown University). His research interests are centered in philosophy of mind, philosophy of language, epistemology, and logic, and are focused mainly on the nature of mental states (particularly belief, desire, and thought) and the concept of meaning. Professor Davis has taught at UCLA (1976), Rice (1977), Washington University (1978), and Georgetown. (1979-Present). He is Editor of Philosophical Studies. Alexander Dinges (Friedrich-Alexander-Universität Erlangen-Nürnberg) Is a fixed-term lecturer at the Friedrich-Alexander-Universität ­Erlangen-Nürnberg. Before that, he was principal investigator in

320 Contributors the DFG  research project The Semantics and Pragmatics of Knowledge Claims based at the Universität Hamburg. He did his PhD at the ­Humboldt-Universität zu Berlin. He works primarily in epistemology and philosophy of language. In epistemology, he works on the semantics, pragmatics and psychology of knowledge ascriptions. He coordinates the research network The Semantics and Metasemantics of Context-­ Sensitive Language. Davide Fassio (Zhejiang University, University of Johannesburg) Is a ZJU 100 Young Professor in the Department of Philosophy at Zhejiang University and a Research Associate at the University of Johannesburg. Before that, he was a postdoctoral research fellow at the University of Geneva and a visiting fellow at the philosophy departments of the University of Southampton and King’s College London. His main interests are in Epistemology, Ethics and Philosophy of Mind. In particular, he’s interested in the nature of belief and its relation with truth (the aim of belief) and the relations between knowledge and practical rationality. Michael Hannon (University of Nottingham) Is an Assistant Professor of Philosophy at the University of Nottingham. He works in epistemology, particularly the social and political aspects of epistemology. He has co-edited two volumes on political epistemology and is currently writing a book for Routledge titled Political Epistemology: An Introduction. In 2019, Oxford University Press published his book, What’s the Point of Knowledge? He is also a Visiting Research Fellow on the “Knowledgeable Democracy” project at VU Amsterdam, as well as a Research Fellow at the Institute of Philosophy in London. In 2018, he received a British Academy Rising Star Award for a project on ‘The Role of Truth in Politics’ which allowed him to establish the Political Epistemology Network. He received his PhD from the University of Cambridge, King’s College. Christos Kyriacou (University of Cyprus) Is a Lecturer at the University of Cyprus and received his PhD in Philosophy from the University of Edinburgh. His main interests lie in epistemology, metaethics and their intersection. He has published articles in Synthese, Erkenntnis, Episteme, Journal of Ethics and Social Philosophy, Philosophical Psychology and elsewhere as well as in volumes. Krista Lawlor (Stanford University) Is a Professor of Philosophy at Stanford University. In philosophy of mind, she works on issues about coreference and confusion. In epistemology, she works on a variety of issues including the nature of assurance, the semantics of knowledge ascription, self-knowledge, memory and inference, and J.L. Austin’s contributions to epistemology. She is the

Contributors  321 author of Assurance: An Austinian View of Knowledge and Knowledge Claims (Oxford UP, 2013) and New Thoughts About Old Things: Cognitive Policies as the Ground of Singular Concepts (Routledge, 2001). Robin McKenna (University of Liverpool) Is a Lecturer in Philosophy at the University of Liverpool. Before coming to Liverpool, he was a postdoc on the ERC Project “The Emergence of Relativism” at the University of Vienna and a postdoc with the Episteme Research Group at the University of Geneva. He completed his PhD at the University of Edinburgh. Most of his work is in epistemology, but he is also interested in philosophy of language, philosophy of science and ethics. Within epistemology, he is increasingly interested in applied epistemology, feminist epistemology and social epistemology more broadly. Duncan Pritchard (University of California, Irvine) Is a Distinguished Professor of Philosophy at the University of Irvine, California. He works mainly in epistemology, and has published widely in this field, including Epistemic Luck (Oxford UP, 2005), The Nature and Value of Knowledge (co-authored, Oxford UP, 2010), Epistemological Disjunctivism (Oxford UP, 2012), Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing (Princeton UP, 2015), and Scepticism: A Very Short Introduction (Oxford UP, 2019). Genia Schönbaumsfeld (University of Southampton) Is a Professor of Philosophy at the University of Southampton who specializes in Wittgenstein, Epistemology, Kierkegaard and the Philosophy of Religion. Before coming to Southampton, Genia studied Philosophy at St. Hilda’s College, Oxford, Trinity College, Cambridge and the University of Vienna. She has been a Visiting Fellow at New College, Oxford and a Visiting Professor at the University of Regensburg. From 2003 to 2006 she held a ‘Hertha Firnberg’ research fellowship at the University of Vienna, awarded by the Austrian Science Fund. She is the author of A Confusion of the Spheres – Kierkegaard and Wittgenstein on Philosophy and Religion (OUP 2007) and of The Illusion of Doubt (2016). Currently, she is working on a book on Wittgenstein’s conception of religion for Cambridge University Press as well as on a new research project aiming to bring Kierkegaard’s philosophy into dialogue with mainstream epistemology. Genia is an Associate Editor of Philosophical Investigations and Editorial Board member of Anthem Studies in Wittgenstein. Mona Simion (University of Glasgow) Is Deputy Director of the COGITO Epistemology Research Centre at the University of Glasgow. She also sits on the Management Committee of the British Society for Theory of Knowledge, the Steering Committee of the Social Epistemology Network, and on the Editorial Board of

322 Contributors the Philosophical Quarterly. Her research is in epistemology (epistemic norms, social epistemology, knowledge first epistemology), philosophy of language (assertion, conceptual engineering, contextualism), ethics (wellbeing, blame, trust, distributive justice, media ethics), and feminist philosophy (epistemic injustice, gender concepts). She is Principal Investigator on a major 5-year ERC-funded research project – KnowledgeLab: Knowledge-First Social Epistemology (2021–2025). She is also Co-­ Investigator on the Leverhulme Trust-funded ‘Virtue Epistemology of Trust’ project (2020–2023) and a Principal Investigator on the ‘Dimensions of Wellbeing’ industry-funded project (2021–2026). Before this she was Principal Investigator on the Mind Association-funded ‘Epistemic Norms and Epistemic Functions’ project (2016–2019). Her PhD (2016) is from Arché, University of St Andrews. Gregory Stoutenburg (York University of Pennsylvania) Is a Visiting Assistant Professor of Philosophy at York College of Pennsylvania. He specializes in epistemology, with substantial overlapping interests in philosophy of language and philosophy of mind. His work mostly concerns the nature of knowledge, how our talk about knowledge works, and what it is to be reasonable in believing something. In his work these issues are connected by the idea that we are implicitly committed to demanding standards for knowledge, ones that imply that our knowledge is very limited. Kevin Wallbridge (University of Southampton) Works primarily on issues in epistemology, the philosophy of language, and the philosophy of mind and has published in Synthese, Australasian Journal of Philosophy, Pacific Philosophical Quarterly, Logos & Episteme, and Philosophia. His research explores how various issues in semantics, linguistics, and cognitive science (e.g. regarding modal semantics, free choice effects, and the compositional structure of belief) bear on theories in epistemology. He is a notable defender of the controversial sensitivity condition in epistemology. He is currently a temporary lecturer at the University of Southampton, and before that he was a postdoctoral research fellow at the University of Cyprus. His PhD is from the University of Edinburgh.

Index

abductive inference 105 abilities 173, 180, 183, 185, 188, 204 abominable conjunctions 115, 230–231 absolute 9, 14–15, 19, 26–27, 41–42, 138, 160–161, 173–174, 220 accuracy 139–140, 148 achievement 109, 144, 186, 281 action 31, 43, 65–66, 71–73, 77, 106, 134–135, 197–199 adjectives 105, 155 aesthetic 265–266 analysis 52–53, 59, 64, 82, 95, 161–162, 165, 231 assertion 15, 27, 64, 89, 103–104, 131, 213, 221, 240, 244–245, 255–260, 264–269, 279–281, 302–304, 310–313 assessment-sensitivity 298, 308–310 attributor 16, 36, 38, 44, 48, 81 bank cases 73–74, 93, 130, 133–134, 159–160, 223–224, 256–264, 267–269, 275–278, 280, 283, 287, 289–293 basing 21–27, 67, 73, 111, 156–158, 162–165, 178, 183, 205, 268–269, 280 Bayes 62 bias 110, 113, 140, 142–143, 178–188, 199–206 capacities 137, 173, 177, 179, 188, 263 causality 148, 161, 165–167, 218 certainties 14, 41, 57–61, 72–73, 110, 157–160, 165–167, 173–174, 195–196, 199–201, 214, 222–227, 235–238, 240–243, 313

closure 22–24, 66–71, 81, 196, 206, 230, 245–247 compatibilism 255–256, 259–265, 267–269 contextualism 16, 34–38, 41–45, 68–75, 83, 175, 213, 220–226, 229, 243, 256–264, 272–273, 288, 291–294, 298–303, 311 credences 117, 182–183 debiasing 186 deduction 22, 66–75, 246, 268 defeasible 157, 172, 228 defeater 133, 183, 187, 228 deflationism 104 denotation 273, 275, 279–280 Descartes 17, 174–176 disjunctivism 26, 81, 227–228, 249 dogmatism 106, 115, 163–165, 179, 186–187 doubt 83, 113–116, 172–179, 181–187, 215–221, 226–227, 235–247 evidence 45, 61–75, 108–111, 129–135, 140, 156–158, 177–183, 216–218, 224–226, 230, 235–236, 244, 264, 287–294, 312 exaggeration 87–88 expertise 203–204 externalism 105, 182–183, 225–226 factivity 26, 104, 227–228 faith 75, 243 fallibilism 20–21, 26–27, 57–69, 72–74, 80–82, 104–106, 138, 157, 172–175, 228 functionalism 38, 40, 47, 60, 216–219, 246, 265–269

324 Index Gettier 61, 72–75, 82 hinges 217–218, 221–231, 235–237, 241–247 hyperbole 87–88, 312–313 ignorance 94, 96, 143, 188 ignoring 81, 86, 96, 116–118, 205, 288–293 illusion 104, 179, 219, 222, 235, 240–242 imagination 20, 108 implicature 88, 91, 96, 105, 259–264, 281, 312 implicit 28, 90–94, 184–185, 199–206, 261 indexicals 35–36, 155, 298–300, 308–310 infallibilism 14–15, 18–19, 21, 57–75, 80–81, 90–97, 103–106, 157, 172–177, 228 informativity 282 instrumental rationality 186–187 intension 273, 306–309, 311 interest-relative 68–75, 130, 138, 155–156, 160, 166 justification 59–60, 73, 76, 80, 90–96, 106, 112, 129–131, 138, 157, 172–173, 182–183, 221, 225, 229–230, 237–243 loose use 74, 86–87, 105, 272–278, 280–281, 284–288, 292–294, 312–313 lotteries 67, 255, 312–313 Meditations 174, 176 modal conditions 112, 130, 300 Mooreanism 75, 112–120, 154, 196, 213, 216–219, 221–229, 236–240, 246, 255 naturalized epistemology 155, 162 nonsense 219–222, 228, 239, 245

norms 65, 104, 255, 258, 260, 266–269, 278–281, 311 objective 34, 39–42, 47, 49, 117–118, 179, 240–242 open-mindedness 107–108, 110 paradox 19–21, 23–25, 48, 231 pragmatics 37, 84, 86–87, 90–96, 103–106, 157, 259–264, 278–280, 284–285, 289–294 probability 41, 57–73, 117–118, 162, 185 prudential 132–133, 264–269 rationality 136, 186, 229–231 relativism 139, 298–306, 308–313 reliabilism 41–44, 59, 91, 106, 139–142, 178, 188, 201, 204, 217, 273–274, 279–280, 288 responsibilism 106, 197–198 safety 112–120 self-defeat 83, 104–105, 187 situationism 195–201 stability 37–38, 141–144, 197 temporalism 309–310 testimony 25, 34, 45, 75, 96, 109, 131, 140, 160–165, 184, 204–205, 265–268, 288 underdetermination 24–25, 114 value 43, 47, 58–61, 66, 69, 72–75, 131, 136–137 vice 106–114, 133, 187, 199–200 virtue 102–103, 106–115, 197–201, 204–205 warrant 112, 183, 246 Wittgenstein 213, 216–229, 235–238, 240–247