Knowledge First: Approaches in Epistemology and Mind [First edition.] 9780191025617, 0191025615, 9780191785023, 0191785024, 9780198716310, 0198716311

Table of contents :
Cover
Knowledge First: Approaches in Epistemology and Mind
Copyright
Contents
List of Contributors
1: Knowledge First: An Introduction
1. Knowledge-First: A Background
2. Knowledge-First: A Motivation
3. Knowledge-First: A Research Program
4. Knowledge-First: The Volume
5. Overview of Chapters
5.1 Foundational issues
5.2 Applications and new directions
References
Part I: Foundational Issues
2: How and Why Knowledge is First
1. Introduction
2. Internalism
3. Externalism
3.1 Reasons-first and the reconciliatory view of perception
3.2 The second reconciliatory view 3.3 A third reconciliatory view?4. Evidence and Justification
References
3: Against Knowledge-First Epistemology
1. Introduction
1.1 Terminology and notation
2. Characterizing Knowledge-First Epistemology
2.2 Reductive and non-reductive knowledge-first epistemologies
2.3 Characterizations of knowledge-first epistemology in conclusion
3. Critique of Reductive Knowledge-First Epistemology
4. Non-reductive Knowledge Norms of Assertion and Action
4.1 Knowledge norms of assertion and their competitors
4.2 Challenges to the knowledge norm 4.3 Non-reductive knowledge-first epistemology in conclusion5. Towards a 'Nothing's First' Alternative: Equilibristic Epistemology
5.1 Equilibristic epistemology
Conclusion
References
4: Mindreading Knowledge
1. Introduction
2. De We Acquire Knowledge First?
2.1 Williamson on the priority of knowledge
2.2 Nagel on the priority of knowledge
2.3 Evidence of infant mindreading
2.4 Animal mindreading
3. Knowledge as a Precondition of Belief
4. Concluding Remarks
References
5: The Cost of Treating Knowledge as a Mental State
1. Introduction
2. The Objection from Internalism 3. Natural Kind Externalism4. Social Externalism
5. Demonstrative Externalism
6. Disjunctivism
7. Counting the Cost
References
6: On Putting Knowledge 'First'
1. Introduction
2. A Quick History
3. A Taxonomy of Priority
3.1 Metaphysical priority
3.2 Representational priority
4. Connections Between Families?
5. The Way Forward
References
7: No Need for Excuses: Against Knowledge-First Epistemology and the Knowledge Norm of Assertion
1. Introduction
2. Knowledge-First Epistemology
3. ... and its Limits
4. The Knowledge Norm of Assertion
5. Excuses, Excuses 6. No Excuses Necessary7. Conclusion: Assertion without Knowledge
References
Part II: Applications and New Directions
8: Acting on Knowledge
1. A nalogies Between Knowledge and Action
2. The Analogy with Intention
3. Knowing is to Believing as Acting is to Intending
4. Directions of Explanation
5. Extending the Analogy
6. Conclusion
References
9: Perception First?
1. Introduction
2. Preliminaries
3. Causal Analyses of Object Perception
4. Counterexamples are Not Enough
5. A Positive Account of Perception?
References

Citation preview

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First Approaches in Epistemology and Mind

edited by

J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis

1

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

3 Great Clarendon Street, Oxford, ox2 6dp, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © the several contributors 2017 The moral rights of the authors have been asserted First Edition published in 2017 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2017944029 ISBN 978–0–19–871631–0 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Contents List of Contributors 1. Knowledge First: An Introduction J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis

vii 1

Part I.  Foundational Issues 2. How and Why Knowledge is First Clayton Littlejohn

19

3. Against Knowledge-First Epistemology Mikkel Gerken

46

4. Mindreading Knowledge Aidan McGlynn

72

5. The Cost of Treating Knowledge as a Mental State Martin Smith

95

6. On Putting Knowledge ‘First’ Jonathan Jenkins Ichikawa and C. S. I. Jenkins 7. No Need for Excuses: Against Knowledge-First Epistemology and the Knowledge Norm of Assertion Joshua Schechter

113

132

Part II.  Applications and New Directions 8. Acting on Knowledge Timothy Williamson

163

9. Perception First? Heather Logue

182

10. Epistemic Supervenience, Anti-individualism, and Knowledge-First Epistemology200 Jesper Kallestrup and Duncan Pritchard 11. Knowledge-First Virtue Epistemology Christoph Kelp

223

12. In Support of the Knowledge-First Conception of the Normativity of Justification246 Anne Meylan

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

vi  contents 13. Sustaining Rules: A Model and Application John Turri 14. ‘More Likely Than Not’: Knowledge First and the Role of Bare Statistical Evidence in Courts of Law Michael Blome-Tillmann

259

278

Index293

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

List of Contributors Michael Blome-Tillmann (McGill University and University of Cambridge) J. Adam Carter (University of Glasgow) Mikkel Gerken (University of Southern Denmark) Emma C. Gordon (University of Edinburgh) Jonathan Jenkins Ichikawa (University of British Columbia) Benjamin W. Jarvis  C. S. I. Jenkins (University of British Columbia) Jesper Kallestrup (University of Edinburgh) Christoph Kelp (University of Glasgow) Clayton Littlejohn (King’s College London) Heather Logue (Leeds University) Aidan McGlynn (University of Edinburgh) Anne Meylan (University of Basel) Duncan Pritchard (University of California, Irvine and University of Edinburgh) Joshua Schechter (Brown University) Martin Smith (University of Edinburgh) John Turri (University of Waterloo) Timothy Williamson (Oxford University)

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

1 Knowledge First An Introduction J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis

1.  Knowledge-First: A Background ‘Knowledge-First’ constitutes what is widely regarded as one of the most significant innovations in contemporary epistemology in the past twenty-five years. Knowledgefirst epistemology is (in short) the idea that knowledge per se is an epistemic kind with theoretical importance that is not derivative from its relationship to other epistemic kinds, such as rationality. Knowledge-first epistemology is rightly associated with Timothy Williamson (2000) in light of his influential book, Knowledge and Its Limits (KAIL). In KAIL, Williamson suggests that meeting the conditions for knowing is not constitutively explained by meeting the conditions for anything else, e.g., justified true belief.1 Accordingly, knowledge is conceptually and metaphysically prior to other cognitive and epistemic kinds. In this way, the concept know is a theoretical primitive. The status of know as a theoretical primitive makes it particularly suitable for use in making substantive constitutive and causal explanations of a number of other phenomena, including the nature of belief, the nature of evidence, and the success of intentional actions.2 As just indicated, Williamson takes the view in KAIL that knowledge—considered as a kind or type—has no constituents. (This should not be confused with the view that instances of knowledge aren’t at bottom physically constituted—Williamson is, in fact, a physicalist.)3 This negative idea seems to be that there are no further kinds that 1   That meeting the conditions for knowing is constitutively explained by meeting the conditions for justified true belief (or: justified, true belief plus some further “x”) has been, especially since the latter half of the twentieth century, the driving assumption behind the epistemological project termed “the analysis of knowledge”. For a recent overview of the analysis of knowledge, as a theoretical project within mainstream epistemology, see Ichikawa and Steup (2014). Cf., Shope (1983) and Jenkins and Ichikawa (2016, this volume). 2   Additionally, the status of know as a theoretical primitive makes it particularly suitable as a normative constraint or rule that governs certain actions (including speech acts such as assertion) and mental states, such as belief. For a recent overview of knowledge norms, see Benton (2015). 3   See, for example, Williamson (2000, §2.2).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

2  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis constitute knowledge when collectively instanced; there is no correct theory that identifies the kind “knowledge” with some mix of distinct epistemic and cognitive kinds meeting specifiable conditions. Nevertheless, in KAIL, Williamson also offers a positive ­characterization of knowledge as the most general factive mental state.4 This further characterization of knowledge is interesting on at least two counts: first—and perhaps more controversially—because it implies that there are factive mental states—remembering that, seeing that, etc.; second, because it suggests that knowledge is, in some sense, the central or most general factive mental state since other factive mental states are more specific ways of knowing.5 The suggestion that there are, among mental states, some that are factive is provocative. Although Putnam-Burge semantic externalism entails that a person’s state of mind— in particular, what these states or mind are directed towards or about—depends on what kind of environment the person has interacted with,6 one might—even after accepting this species of externalism—still resist the idea that a person’s state of mind depends on what the facts—potentially outside of her—are. Moreover, while one might accept that whatever factive mental states there are derive from non-factive ones, this is clearly not Williamson’s idea. Williamson takes factive mental states to be at least on a par with non-factive ones.7 Moreover, with respect to the (allegedly) central mental factive state—knowing—and the central cognitive non-factive correlate—believing— Williamson is clear that the former is no less explanatory than the latter. Even if it is possible to understand knowing as a kind of “apt” believing,8 it is also possible to understand believing as a kind of “botched” knowing.9 If we are to understand what it is to believe at least partly through what it is to know, then, given the prominence of belief in contemporary theories of mind, knowledge appears to be central, not only to epistemology, but also to the philosophy of mind.

2.  Knowledge-First: A Motivation A central project within epistemology is to understand the proper assessment of belief. A central project within the philosophy of mind is to understand what a belief is. A not wholly implausible idea is that these central projects are, in fact, related. To understand better what a belief is, one needs to think about what happens when belief   Ibid., 34–7; 39–40.   For an earlier presentation of this idea, see Williamson (1995). The natural expression of a factive (stative) mental state in natural language is a factive mental state operator (FMSO); Williamson’s position is that knowledge is the most general FMSO. 6   Putnam’s (1975) classic externalist argument insists that mental content is individuated by features of one’s physical environment, whereas Burge’s (1986) argument adverts to features of one’s socio-linguistic environment. For an overview, see Lau and Deutsch (2014, §2) Cf., Carter et al. (2014) for an overview of varieties of externalism in the philosophy of mind and epistemology, and how they interface with one another. 7   See, for example, Williamson (2000, §1.5). 8   For a notable recent defence of this proposal, see Sosa (2015, Ch. 1). 9   Williamson (2000, 47). 4 5

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  3 goes right, and to understand better what happens when belief goes right, one needs to think about what beliefs are. If this idea seems odd, consider it in another context. To understand better what a president of the United States is, it helps to understand better what it looks like when the person occupying that office acts in accordance with the responsibilities and duties of the office. In principle, a person in office might try to ­violate the duties in any number of ways—he might refuse to give the State of the Union address, for example. However, for the purposes of gaining a fundamental understanding, it helps to consider first not the abundance of ways a president might shirk responsibilities and duties, but rather the way a competent president might act. Indeed, it may help to focus first on a case where the surrounding government is also relatively competent. One can subsequently understand all varieties of presidential dysfunction by way of contrast with the case of a harmonious United States federal government. For some time, the dominant approach to the theory of belief has been functionalism (at least broadly construed)—so that, to a first approximation, beliefs are what they do, i.e. believing any particular proposition is largely a matter of occupying a certain role.10 Arguably, beliefs play a number of roles—assertions express them, actions are based on  them, topical understanding consists in them, and so forth. Consequently, the approach of understanding belief by understanding its proper assessment might begin by considering what it is for belief to go right in each of these roles. In turn, we might gain some further insight about epistemic standards by considering what a person might want from her beliefs in each of these roles. A natural suggestion—prompted, perhaps, by the fact that the most prominent tool in our conceptual repertoire for marking the quality of belief is the concept know—is that going right for belief is a matter of knowing. Williamson defends individual theses about the explanatory primacy of knowledge in KAIL—e.g., that it is the standard for proper assertion,11 that it is central to the explanation of action12—and others have defended further theses—e.g. that it is required for topical understanding.13 Arguably, a unifying feature of these individual theses is that the phenomena at issue are closely associated with belief—so that belief might even plausibly be at least partly constituted by its role in each case. A knowledge-first addition to this last plausible idea is that the role that beliefs play generally is parasitic on the role that they play when things go right so that the belief qualifies as knowledge: there are a surfeit of ways for a belief to fail in assertion, in action, in understanding etc., but we understand how beliefs can fail in these ways by considering what happens when they don’t fail—because the subject doesn’t merely believe, but rather knows.

10   For some seminal defences of this view, see Armstrong (1968), Fodor (1968), and Lewis (1972). For a critical overview, see Schwitzgebel (2006, §1.4). 11   Other defenders of this view include DeRose (2002), Hawthorne (2004), and Stanley (2005). 12  See Fantl and McGrath (2002), Hawthorne (2004), Hawthorne and Stanley (2008), and Stanley (2005). 13   See, for example, Sliwa (2015) and Kelp (Forthcoming).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

4  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis Of course, there are other theoretical alternatives for a belief ’s going right—perhaps most obviously, the belief ’s being rational or justified and the belief ’s being true.14 It is well beyond the scope of this introduction to assess in depth whether these alternatives might fare better. However, it may be worth briefly noting some possible deficiencies: a rational or justified belief can fail to be true and a true belief can fail to be rational or justified, but both rationality and truth have claims as minimum standards for a belief ’s having gone right. An irrational belief—even if true—is subject to revision should the subject ever become a better thinker or exert more effort to deliberating. If discovered by others as such, it may undermine one’s credibility in making assertions, and it is hard to see how it could play a central role in one’s understanding of some subject matter (e.g., economics). On the other hand, a false belief—even if rational—may not be the best one to transfer to another person through testimony. Moreover, any false belief may—through various inferences—ultimately lead to a further false instrumental belief—about how to satisfy some particular preferences—even when reasoning is good; and, false instrumental beliefs generally put one no closer to satisfying preferences when acted upon. Knowing has the virtue of entailing both that the belief is true and that the belief is rational, so it avoids all of these problems. In addition, it appears to have the very nice feature of being a widely available commodity—unlike, for instance, an alternative epistemic kind that requires Cartesian infallibility. There may be the additional difficulty in understanding why knowing—rather than mere justified true belief—should matter,15 but there’s at least the theoretical possibility that for a belief to go right, a certain harmony must obtain between mind and world—a certain harmony that does not obtain in “double-luck” cases of mere justified true belief. Thus—while the matter is far from settled—it may be a greater threat to the knowledge-first program that a belief might go right in a multiplicity of nonknowledge ways. More precisely, one might imagine that different activities—e.g., assertion, action, the pursuit of understanding, etc.—make different demands on beliefs with the result that the “optimal” way of going about believing is different for each and not necessarily oriented towards knowing. On the other hand, unless there’s alignment—or at least an available optimal compromise—between this variety of demands on belief, it’s not clear that it should really be belief (rather than some other cognitive output) playing this total variety of roles since it would appear that belief can only be manufactured and managed in one particular way. In fact, it may be that a greater threat still to the knowledge-first program need not find fault with the idea that knowledge per se is the uniquely relevant way that beliefs generally go right, but rather takes issue with the idea that (on-off) belief itself is especially central in the theory of mind. If, for instance, credences—or graded belief— are more important to understanding assertion, action, understanding, theory 14   This idea has been given both constitutive and teleological glosses, in the former case under the description of normativism about belief (e.g., Shah 2003; Shah and Velleman 2005), and in the latter case under the description of epistemic-value truth monism (e.g., Pritchard 2014). 15   For a recent discussion of this point, see Carter, Jarvis, and Rubin (2013). Cf., Pritchard and Turri (2014).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  5 revision, etc., then belief could even turn out to be epiphenomenal.16 Given that it is not altogether clear whether evaluations of credence as knowledge make any sense, knowledge might turn out to be epiphenomenal as well. One way to understand the motivation for some work within the knowledge-first program, then, is to see it as attempting to undercut this kind of threat. Williamson’s views in KAIL on knowledge and evidence may be a case in point vis-à-vis credence and evidential probability. However, we might pick a different example entirely to illustrate the same general point: Williamson’s arguments against the “luminosity” of any mental conditions17— and, thus, in favour of fallible cognitive access to even “internal” sensations or feelings, e.g., of coldness or of pain—may serve to help undercut the possibility that experience might supplant belief as the most important information-carrying unit in psychology. Again, the underlying idea here: arguably, the knowledge-first program puts belief (rather than alternatives) at least very close to the center of cognition as well since, arguably, only beliefs are candidates for knowledge.

3.  Knowledge-First: A Research Program Timothy Williamson is the founder of the knowledge-first movement and has been its principal flagbearer. However, we think that it is useful to appreciate that the knowledgefirst research program is itself distinct and largely independent from the collection of knowledge-first theses Williamson has defended. Because the knowledge-first approach to epistemology and mind is a research program, it is fairly resistant to refutation even if any of these particular theses turns out to be false. By way of example, consider a version of the view that Williamson defends in KAIL about assertion. According to this view, knowledge sets the quality standard for assertion, assertions that fall short of this standard (because the assertor doesn’t know) are ipso facto normatively defective (qua assertion) while assertions that meet the standard are not normatively defective in the same way, and this quality norm rather than any other norms, e.g. Gricean maxims18 of “quantity,” “relation,” or “manner,” provides the real essence of an assertion, i.e. something is an assertion solely because this quality norm (rather than some weaker or stronger one) applies to it.19 One might object to this view on the ground that it fails to account for the fact that assertion is multifarious.20 In certain contexts, assertion might be a tool for transferring beliefs— or perhaps more clearly, even knowledge21—from one person to another; in such cases, the quality standard of knowledge might well be appropriate and applicable. In other   Cf., Carter, Jarvis, and Rubin (Forthcoming).   Williamson (2000, Ch. 4). For a recent defence of Williamson’s anti-luminosity argument against various objections, see Srinivasan (2015). 18   E.g., Grice (1991). 19  See, however, Benton (2014) for further discussion and qualification on this point. Cf., Lackey (2007; 2008) for notable criticisms to both the necessity and sufficiency legs of the knowledge norm of assertion. 20   For a more revisionary line on this score, see Cappelen (2011). 21   This is, for example, the view advanced in Stalnaker’s (1978) knowledge-transfer model of assertion. 16 17

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

6  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis cases, however, assertion might merely be tool for trying out ideas (that one may not even believe), tentatively arguing for a hypothetical position, or even indirectly informing a person about one’s beliefs by directly stating (something like) their contents. In these kinds of cases, other quality norms are applicable. Suppose, for the sake of argument, this particular objection succeeds; the targeted knowledge-first thesis is mistaken precisely because assertion is multifarious. Even still, it would not be clear how much the success of this objection would damage the knowledge-first program. To at least some extent, the objection succeeds because (1) the kind assertion is less natural than one might have initially thought, and relatedly (2) assertion is less straightforwardly and closely related to belief than one might have thought. In particular, we are imagining that assertion turns out chimerical rather than, for instance, always and everywhere an outward reflection of belief that is subject to similar standards as belief is.22 If the kind assertion is less natural than one might have thought, it becomes considerably less important to have an elegant theory of it— let alone a knowledge-based one. Indeed, it might be impressive enough to have a knowledge-based theory of some variety of assertion. Moreover, if assertion in general is not especially closely tied to belief, then, of course, belief—and, consequently, knowledge—might be marginally less important from a theoretical perspective than it might have been otherwise. But, the same condition also considerably lowers the expectation we should have about whether there should be a fully general knowledgebased theory of assertion. All in all, the hypothetical failure of this one particular knowledge-first thesis (about assertion)—at least in this particular way—seems to bear very little on the overall prospects of the knowledge-first program. The more important general point is that it is always possible in principle to refute a knowledge-first thesis by showing that belief itself is not the cognitive vehicle that is most or even especially central to the phenomenon at issue in all possible cases. If this is true in every instance, then both belief and knowledge turn out not to be very interesting. But, as long as it is only true in a fairly restricted range of instances, the knowledge-first program is not especially in peril. It would be in peril if, even within the range of cases where belief is important, knowledge is also not important. But, this is not easy to show either. There are usually a number of different ways that knowledge could be of central importance for any phenomenon where belief plays an important role. It is certainly possible that the knowledge-first program will turn out to fail, but it is not a simple matter of refuting a certain number of the knowledge-first theses that Timothy Williamson has defended. Finally, it is worth emphasizing that, in our opinion, the knowledge-first program does not hinge on accepting every element of Williamson’s metaphysical ­picture of (the state or concept of) knowledge. For instance, Williamson seems to place great emphasis on the fact that knowledge is not “factorizable.” On at least 22   For instance, this might be the case if we had compelling reason to reject the “belief-assertion parallel” according to which, as Williamson puts it, “believing p stands to asserting p as the inner stands to the outer” (255–6). See also Benton (2014, §3a).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  7 some understanding of “factorizable,” knowledge would appear to be factorizable if the following biconditional were true: A subject, S, knows

if and only if: (1) (2) (3) (4)

S believes

; S’s believing

is a cognitive success in the sense that

is true; S believes

as the result of exercising epistemic competences; and (2) is attributable (3).

This “robust competence” theory appears to be a version of the JTB+ theories that Williamson generally dismisses.23 One might argue that success of the knowledge-first program hinges on rejecting this theory as a theory of knowledge, but rejecting this theory as a theory of knowledge does not obviously require rejecting the biconditional itself. Arguably, it only requires rejecting the knowing as the analysandum of the theory. One might accept the biconditional as part of a theory of epistemic competences or even as part of a theory of belief.24 Indeed, as long as the order of explanation or analysis does not flow exclusively from left to right, knowledge may be first in the (ordinary) sense that nothing else comes before it.25 Perhaps, in the context where knowing is not the analysandum of the biconditional, it is misleading to say that knowledge is factorizable; nevertheless, it is arguably strictly true that, so long as the biconditional is true, knowing is a conjunction of four factors (whether or not knowing has any explanatory or analytical priority). So, it’s not clear that knowledge being first—or the relative priority of knowledge—especially depends on Williamson being right about factorization. A similar point could be made about Williamson’s ontological categorization of knowing as part of the mental domain.26 Categorizing knowing in this way might well boost knowledge’s credentials as a state worth understanding for the purposes of understanding the mind. However, if, for example, the appropriate way to understand belief is as an attempt at knowing (where mere belief is simply “botched” knowing), then knowledge looks to play an important role within the philosophy of mind whether or not it is a mental kind per se. Again, the general point is that one has to be careful not to assume that any particular part of Williamson’s metaphysics of (the state or concept of) knowledge is essential to the knowledge-first program. Williamson’s complete picture of knowledge—including not only the metaphysics, but the variety of individual explanatory theses involving knowledge that he defends— is interesting in its own right. Our point in this section is not to dismiss either its truth or appeal. Rather, we wish to point out that, when it comes to the knowledge-first program, Williamson’s particular picture is important as a paradigm, not as the ­substance. In fact, his picture seems to constitute a particularly strong version of 23   Proponents of this kind of view, also referred to as robust virtue epistemology, include most notably John Greco (e.g., 2010; 2012) and Ernest Sosa (2009; 2015). 24   Cf. Carter, Jarvis, and Rubin 2013.    25  Cf. Ichikawa and Jarvis (2013, 138). 26   Williamson 2000 (Ch. 1).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

8  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis a  knowledge-first approach—which is particularly useful for appreciating what a knowledge-first approach might be, but may or may not be the most plausible version. After all, it is possible to think that knowledge is at the center of a wider or narrower range of phenomena, and—as some of our contributors will emphasize—there are a variety of ways in which knowledge exhibits priority (e.g. explanatory, conceptual, ontogenetic, epistemic, metaphysical, etc.) over other phenomena. We think, then, that it is actually a disservice to Williamson’s remarkable contribution to associate the knowledgefirst program too closely with the particular way that he has carried it forward.

4.  Knowledge-First: The Volume Given that there are a wide range of ways in which the knowledge-first program (or something approximating it) might succeed at least in part, there are a wide range of ways to engage with the program as well. A few of our contributors have elected to say something highly general about the program, but many have chosen to focus on specific knowledge-first theses or other aspects of knowledge that may be turn out to be  relevant to evaluating the program at a later stage. In two cases—including the ­contribution from Williamson himself—authors have considered turning the methodology Williamson has applied to the study of knowledge to other phenomena (action and perception). We think this, too, may provide some insight about the knowledge-first program. However, given the diversity of topics covered within the volume, we also hope that the contributions not only help advance the understanding of the knowledge-first program, but also are of interest to philosophers and other researchers curious about other phenomena—including assertion, action, normativity, or mentality—independent of any particular interest in knowledge itself. The articles in this volume fall well short—even collectively—of systematically addressing the virtues and vices of the knowledge-first program. However, we believe this result is appropriate. The purpose of the volume is to provide a relatively openended forum for creative and original scholarship that might ultimately bear on the knowledge-first program, rather than to provide a complete summary of knowledgefirst debates up until the present. Moreover, given that we believe the knowledge-first program is not only ambitious, but also flexible and resilient, it comes as no surprise to us that the volume falls well short—in our minds at least—of vindicating either the program’s adherents or opponents. Our intent is only to make some progress in shedding some further light on the subject matter. We think (and hope) that the ­volume succeeds in that modest aim.

5.  Overview of Chapters In what follows are brief overviews of each specific chapter in the volume, thirteen in total. Though the contributions to the volume differ substantially with regard to which

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  9 aspects of the knowledge-first program receive special focus, we think nonetheless that the contributions fall generally within two broad categories: (i) foundational issues with knowledge-first philosophy (including defenses and critiques of core elements of the project), and (ii) applications and new directions. The first half of the book contains papers which cluster around the former theme, and the second half, the latter.

5.1  Foundational issues We begin with Clayton Littlejohn’s staunch defence of Knowledge-First epistemology, in ‘How and Why Knowledge is First’. Littlejohn’s dialectical aim has, as its focus, a sustained defense of the claim that one cannot have a reason in one’s possession unless it is something that one knows. This view is claimed to have advantages over a different way of thinking about epistemic status. On the ‘reasons-first’ approach to epistemic status, reasons and the possession of them are prior to epistemic status. In reversing this picture, Littlejohn reveals an important sense in which knowledge comes first— viz., in that we first come to have reasons in our possession by coming to know that certain things are true; there is nothing prior to knowing that puts these reasons in our possession. In the course of advancing this picture, Littlejohn furthermore offers a sophisticated defence of Williamson’s knowledge-evidence equivalence, (E=K). In ‘Against Knowledge-First Epistemology’, Mikkel Gerken attacks, on several fronts, what is often cited as a theoretical advantage to regarding knowledge as a theoretical primitive—namely, that knowledge can be used to reductively analyse other epistemic phenomena. As Gerken sees it, proponents of such an approach commit a similar mistake to the one that they charge their opponents with—viz., the mistake of seeking to reductively analyse basic epistemic phenomena in terms of other allegedly more basic or fundamental phenomena. After leveling this charge against reductionist brands of knowledge-first epistemology, Gerken then challenges—taking the knowledge norm of assertion as his critical focus27—non-reductionist brands of knowledge-first epistemology. Gerken concludes by articulating an alternative to knowledge-first methodology that he calls “equilibristic epistemology”. According to equilibristic epistemology there isn’t a single epistemic phenomenon or concept that is “first”. Rather, there are a number of basic epistemic phenomena that are not reductively analysable although they may be co-elucidated in a non-reductive manner. In ‘Mindreading Knowledge’, Aidan McGlynn, like Gerken, is critical of a core element of the knowledge-first program—in McGlynn’s case, the thesis that knowledge is a mental state in its own right.28 McGlynn challenges this thesis by way of ­calling into doubt a prominent empirically oriented strategy for defending it, one 27   For the classic defense of this position, according to which assertion is normatively constrained by the rule that one should assert only what one knows, see Williamson (1996). Other prominent defenders of the view include DeRose (2002), Hawthorne (2004), and Stanley (2005). 28   For detailed criticisms by McGlynn of other aspects of the knowledge-first program, see McGlynn (2014).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

10  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis which has been advanced in recent work by Jennifer Nagel (2013). Nagel draws on work in developmental and comparative psychology with the aim of establishing that the concept of knowledge is acquired before the concept of belief, and she regards this conceptual priority claim as evidence for the metaphysical thesis that knowledge is a mental state. McGlynn argues that, on closer inspection, the results established in the developmental and comparative psychology literature fall well short of conclusively supporting Nagel’s conceptual priority claim, much less the stronger metaphysical claim that knowledge is a mental state in its own right. Martin Smith is also critical of the core knowledge-first thesis that knowledge is a mental state. In his contribution ‘The Cost of Treating Knowledge as a Mental State’, Smith insists that embracing the mental state thesis carries certain costs that have not been widely appreciated. Of particular interest for Smith are costs associated with departing (as Knowledge-First proponents do) with internalism about the mental, where internalism is the thesis that one’s mental states are determined by one’s internal physical state. Internalism about the mental has been widely rejected amongst contemporary philosophers of mind.29 However, Smith argues, although philosophers of mind have converged on the falsity of internalism, the Knowledge-First proponent’s claim that knowledge is a mental state effectively takes us much further from internalism than anything philosophers of mind have converged upon. In ‘On Putting Knowledge “First”’, Jonathan Jenkins Ichikawa and C. S. I. Jenkins claim that various views which travel under the banner of “knowledge first” epistemology betray subtle differences in just how it is that they respectively regard knowledge as  “first”. These differences, they argue, are problematic, in part because it is not straightforward to draw connections between certain of these views, which are under closer inspection more independent than they are often assumed to be. Ichikawa and Jenkins’s aim is, in the main, to tease apart various “knowledge first” claims, and explore what connections they do or do not have with one another, in the service of a clearer understanding of just what the knowledge first theses are and how these theses might be evaluated. Rounding out the chapters focusing on foundational issues is Joshua Schechter’s ‘No Need for Excuses: Against Knowledge-First Epistemology and the Knowledge Norm of Assertion’. In this chapter, Schechter offers a two-tiered critique of the knowledgefirst program. Firstly, he surveys some of the “big-picture” objections to knowledge-first epistemology, and argues that they are not conclusive. Secondly, Schechter shifts his critical attention to a specific thesis endorsed by many knowledge-first epistemologists— the knowledge norm of assertion, and in particular, to the objection that it is intuitively appropriate for someone who has a strongly justified belief that p, but who doesn’t know that p, to assert that p.30 Schechter argues that a standard reply to this objection   For an overview of externalist approaches to mental content, see Lau and Deutsch (2014).   The blamelessness reply to this line of objection is also defended by, along with Williamson (2000, 256–7), DeRose (2002) and Hawthorne and Stanley (2008). 29 30

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  11 on behalf of knowledge-first proponents—viz., that such assertions are improper but that the subject has an excuse and is therefore not blameworthy for making the assertion—is ultimately unworkable.

5.2  Applications and new directions In KAIL’s opening paragraph, Williamson takes, as a starting point for discussion, an analogy between knowledge and action. Knowledge and action are the central relations between mind and world. In action, world is adapted to mind. In knowledge, mind is adapted to world. When world is maladapted to mind, there is a residue of desire. When mind is maladapted to world, there is a residue of belief. Desire aspires to action; belief aspires to knowledge. The point of desire is action; the point of belief is knowledge  (2000, 1).

The principal aim of Williamson’s contribution to this volume, ‘Acting on Knowledge’, is to develop and refine this analogy between knowledge and action in KAIL, the general schema of which is: knowledge is to belief as action is to intention. As Williamson himself articulates his project: The analogy reverses direction of fit (mind to world, world to mind). The knowledge/belief side of the analogy corresponds to the inputs to practical reasoning, the action/intention side to its output. Insofar as desire is an input to practical reasoning, it belongs to the former side (the desire-as-belief thesis is considered sympathetically). When all goes globally well with practical reasoning, one acts on what one knows. Beliefs play the same local role as knowledge, and intentions the same local role as action, in practical reasoning. This is the appropriate setting in which to understand knowledge norms for belief and practical reasoning. Marginalizing knowledge in epistemology is as perverse as marginalizing action in the philosophy of action.

An advantage Williamson claims of developing this analogy rigorously is that opponents of knowledge-first epistemology are challenged to produce an equally systematic and plausible account of the relation between the cognitive and the practical. Heather Logue, like Williamson, investigates an analogy—in her case, an analogy between knowledge and perception. In short, Logue asks: if knowledge is unanalysable, might also perception be? After all, the history of attempts to analyse the perceptual relation have been subject to counterexamples in such a way as to broadly mirror the track record of the post-Gettier literature.31 To the extent that the failure of the postGettier project motivates a knowledge-first approach, it is natural to wonder whether an analogous sort of failure to analyse (in a fashion that avoids counterexamples) the perceptual relation motivates a perception-first approach. Logue however argues that even if the perceptual relation turns out to be unanalysable, this does not necessarily

31   For an overview, see Shope (1983). The locus classicus of the Gettier problem is Gettier (1963). For an overview of various attempts to characterize the perceptual relation, see Crane and French (2016).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

12  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis mean that we should embrace a perception-first approach. Though, as she suggests, there might nonetheless be an alternative motivation for a perception-first approach. Duncan Pritchard and Jesper Kallestrup, in ‘Epistemic Supervenience, AntiIndividualism and Knowledge-First Epistemology’, investigate connections between knowledge-first epistemology and a metaepistemological thesis they’ve defended elsewhere (and in opposition to robust forms of virtue epistemology)32 under the description of epistemic anti-individualism.33 Epistemic anti-individualism is a denial of the epistemic individualist’s claim that warrant—i.e., what converts true belief into knowledge—supervenes on internal physical properties of individuals, perhaps in conjunction with local environmental properties. Kallestrup and Pritchard have two central aims. First, they argue that “epistemic twin earth” thought experiments which reveal robust virtue epistemology (RVE) to be problematically committed to epistemic individualism also show that evidentialist mentalism34 is likewise committed to individualism. Second, they argue that, even though a knowledge-first approach in epistemology is in principle (unlike RVE and evidentialist mentalism) consistent with epistemic anti-individualism, this approach fails to offer a plausible account of epistemic supervenience. This point is, they suggest, a reason to pursue epistemic anti-individualism outside the knowledge-first framework. In ‘Knowledge-First Virtue Epistemology’, Christoph Kelp connects what have thus far been independent knowledge-related research programs: virtue epistemology, which offers an account of knowledge in terms of agents’ intellectual virtues or abilities, and knowledge-first epistemology. Kelp’s primary aim in the chapter is to develop a knowledge-first virtue epistemological accounts of knowledge and justified belief 35 and to show that these accounts compare favourably with their traditionalist cousins. Anne Meylan, in her chapter ‘In Support of the Knowledge-First Conception of the Normativity of Justification’ explores knowledge-first epistemology in connection with the New Evil Demon Problem (NEDP).36 In particular, Meylan argues that the knowledge-first solution to the NEDP fits well with a particular conception of the normativity of justification. Williamson (forthcoming) defends this conception, according to which a justified belief is one that satisfies some sort of “ought” or “should”, a view which runs contrary to the more established view of justified belief, as neither obligatory nor forbidden.37 The position she advances is that the knowledge-first conception of the normativity of justification, which is the one on which the knowledge-first solution to the NEDP relies, is superior to the more traditional view.

32   Prominent defenders of robust virtue epistemology include Greco (2010; 2012), Sosa (2009; 2010; 2015), and Zagzebski (1996). 33   See, for example, Kallestrup and Pritchard (2012; 2013; 2014). 34   The most classic defense of this view is developed in Conee and Feldman (2004). 35   For a complementary discussion of a knowledge-first approach to justified belief, see Kelp (2016). 36   For notable presentations of this problem, see Lehrer and Cohen (1983) and Cohen (1984). For a recent critical overview of the problem, and a survey of responses, see Littlejohn (2009). 37   This is a view that has been defended in various places by, among others, William Alston, e.g., (1989).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  13 In ‘Sustaining Rules: A Model and Application’, John Turri places knowledge-first, in a hitherto unexplored way, as a normative standard for belief. One familiar way to think of knowledge as a normative standard for belief is as a rule that governs the ­propriety of believing, for example: one must: believe that p only if one knows that p.38 Turri’s project explores a comparatively broader sense in which knowledge might be the normative standard for belief: by normatively sustaining cognition, and thereby, inquiry. This result is part of a wider project: to offer conditions under which any rule sustains a practice (cognitive or otherwise). On Turri’s proposal, a rule normatively sustains a practice when the value achieved by following the rule explains why agents continue following that rule—in a way that sustains the pattern of activity. Finally, in ‘“More Likely Than Not” Knowledge First and the Role of Bare Statistical Evidence in Courts of Law’ Michael Blome-Tillmann argues that embracing a knowledge-first approach can help to resolve important epistemological problems in legal philosophy.39 Blome-Tillmann takes as a starting point a puzzle arising from the evidential standard Preponderance of the Evidence and its application in civil procedure. The evidential standard captured by Preponderance of the Evidence is usually glossed as “greater than .5 given the admissible evidence”. But this characterization generates puzzles,40 where our intuitions about whether a defendant should be found liable diverge in case pairs where the evidential probability captured this way is the same. Blome-Tillmann argues that the tension generated by such puzzles can be resolved fairly straightforwardly within a knowledge-first framework.

References Alston, William P. 1989. Epistemic Justification: Essays in the Theory of Knowledge. Ithaca, NY: Cornell University Press. Armstrong, David M. 1968. A Materialist Theory of the Mind. London: Routledge. Benton, Matthew A. 2014. Gricean Quality. Noûs 50(4): 659–878. doi:10.1111/nous.12065. Benton, Matthew A. 2015. Knowledge Norms. Internet Encyclopedia of Philosophy. Retrieved from http://www.iep.utm.edu/kn-norms/#SH1d. Blome-Tillmann, Michael. 2015. Sensitivity, Causality, and Statistical Evidence in Courts of Law. Thought: A Journal of Philosophy 4(2): 102–12. Burge, Tyler. 1986. Individualism and Psychology. Philosophical Review 95(1): 3–45. Cappelen, Herman. 2011. Against Assertion. In Jessica Brown and Herman Cappelen, eds., Assertion: New Philosophical Essays. Oxford: Oxford University Press, 21–48. Carter, J. Adam, Benjamin Jarvis, and Katherine Rubin. 2013. Knowledge: Value on the Cheap. Australasian Journal of Philosophy 91(2): 249–63.

38   As Williamson puts it, “Knowledge sets the standard of appropriateness for belief ”. For discussion on this point, see Williamson (2000, 47) and more recently McHugh (2011) and Littlejohn (2013). Cf., Benton (2015, §3). 39   See also Blome-Tillmann (2015). 40   For example, the “Gatecrasher Puzzle”, as noted in L. J. Cohen (1977).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

14  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis Carter, J. Adam, Benjamin Jarvis, and Katherine Rubin. 2016. Belief without Credence. Synthese, 193(8): 2323–2351. doi:10.1007/s11229-015-0846-6. Carter, J. Adam, Jesper Kallestrup, S. Orestis Palermos, and Duncan Pritchard. 2014. Varieties of Externalism. Philosophical Issues 24(1): 63–109. Cohen, L. Jonathan. 1977. The Probable and the Provable. Oxford: Clarendon Press. Cohen, Stewart. 1984. Justification and Truth. Philosophical Studies 46(3): 279–95. Conee, Earl, and Richard Feldman. 2004. Evidentialism: Essays in Epistemology: Essays in Epistemology. Oxford: Oxford University Press. Crane, Tim, and Craig French. 2016. The Problem of Perception. Stanford Encyclopedia of Philosophy. Retrieved from http://plato.stanford.edu/archives/spr2016/entries/perceptionproblem/. DeRose, Keith. 2002. Assertion, Knowledge, and Context. Philosophical Review 111(2): 167–203. Fantl, Jeremy, and Matthew McGrath. 2002. Evidence, Pragmatics, and Justification. The Philosophical Review 111(1): 67–94. doi:10.2307/3182570. Fodor, Jerry A. 1968. Psychological Explanation: An Introduction to The Philosophy of Psychology. New York: Random House. Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23(6): 121–23. Greco, John. 2010. Achieving Knowledge: A Virtue-Theoretic Account of Epistemic Normativity. Cambridge: Cambridge University Press. Greco, John. 2012. A (Different) Virtue Epistemology. Philosophy and Phenomenological Research 85: 1–26. Grice, H. Paul. 1991. Studies in the Way of Words. Cambridge, MA: Harvard University Press. Hawthorne, John. 2004. Knowledge and Lotteries. Oxford: Oxford University Press. Hawthorne, John, and J. Stanley. 2008. Knowledge and Action. Journal of Philosophy 105(10): 571–90. Ichikawa, Jonathan Jenkins, and Benjamin Jarvis. 2013. The Rules of Thought. Oxford: Oxford University Press. Ichikawa, Jonathan Jenkins, and Matthias Steup. 2014. The Analysis of Knowledge. Stanford Encyclopedia of Philosophy. Retrieved from http://plato.stanford.edu/archives/spr2014/entries/ knowledge-analysis/. Jenkins, C. S. Ichikawa, and Jonathan Jenkins Ichikawa. 2018. On Putting Knowledge First. In J. Adam Carter, Emma C. Gordon, and Benjamin Jarvis, eds., Knowledge First: Approaches in Epistemology and Mind. Oxford: Oxford University Press, 113–31. Kallestrup, Jesper, and Duncan Pritchard. 2012. Robust Virtue Epistemology and Epistemic Anti-Individualism. Pacific Philosophical Quarterly 93 (1): 84–103. Kallestrup, Jesper, and Duncan Pritchard. 2013. Robust Virtue Epistemology and Epistemic Dependence. In Tim Henning and David P Schweikard, eds., Knowledge, Virtue, and Action: Essays on Putting Epistemic Virtues to Work. London: Routledge. Kallestrup, Jesper, and Duncan Pritchard. 2014. Virtue Epistemology and Epistemic Twin Earth. European Journal of Philosophy 22(3): 335–57. Kelp, Christoph. 2016. Justified Belief: Knowledge First-Style. Philosophy and Phenomenological Research 93(1): 79–100.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge First: An Introduction  15 Kelp, Christoph. Forthcoming. Towards a Knowledge-Based Account of Understanding. In Stephen Grimm, Christoph Baumberger, and Sabine Ammon, eds., Explaining Understanding: New Perspectives from Epistemology and the Philosophy of Science. London: Routledge. Lackey, Jennifer. 2007. Norms of Assertion. Noûs 41(4): 594–626. Lackey, Jennifer. 2008. Learning from Words: Testimony as a Source of Knowledge. Oxford: Oxford University Press. Lau, Joe, and Max Deutsch. 2014. Externalism About Mental Content. Stanford Encyclopedia of Philosophy. Retrieved from http://plato.stanford.edu/archives/sum2014/entries/contentexternalism/. Lehrer, Keith, and Stewart Cohen. 1983. Justification, Truth, and Coherence. Synthese 55(2): 191–207. Lewis, David. 1972. Psychophysical and Theoretical Identifications. Australasian Journal of Philosophy 50(3): 249–58. Littlejohn, Clayton. 2009. The New Evil Demon Problem. Retrieved from http://philpapers. org/rec/LITTNE. Littlejohn, Clayton. 2013. XV—the Russellian Retreat. Proceedings of the Aristotelian Society 113(3 pt 3): 293–320. doi:10.1111/j.1467–9264.2013.00356.x. McGlynn, Aidan. 2014. Knowledge First? London: Palgrave Macmillan. McHugh, Conor. 2011. What Do We Aim At When We Believe? Dialectica 65(3): 369–92. doi:10.1111/j.1746–8361.2011.01270.x. Nagel, Jennifer. 2013. Knowledge as a Mental State. Oxford Studies in Epistemology 4: 275–310. Pritchard, Duncan. 2014. Truth as the Fundamental Epistemic Good. In Jonathan Matheson and Rico Vitz, eds., The Ethics of Belief. Oxford: Oxford University Press, 112–29. Pritchard, Duncan, and John Turri. 2014. The Value of Knowledge. Stanford Encyclopedia of Philosophy. Retrieved from https://plato.stanford.edu/entries/knowledge-value/. Putnam, Hilary. 1975. The Meaning of “Meaning”. Minnesota Studies in the Philosophy of Science 7: 131–93. Schwitzgebel, Eric. 2006. Belief. Stanford Encyclopedia of Philosophy. Retrieved from http:// stanford.library.usyd.edu.au/entries/belief/. Shah, Nishi. 2003. How Truth Governs Belief. Philosophical Review 112(4): 447–82. Shah, Nishi, and J. David Velleman. 2005. Doxastic Deliberation. Philosophical Review 114(4): 497–534. Shope, Robert K. 1983. An Analysis of Knowing: A Decade of Research. Princeton: Princeton University Press. Sliwa, Paulina. 2015. IV—Understanding and Knowing. In Proceedings of the Aristotelian Society 115(1, pt. 1): 57–74. Wiley Online Library. http://onlinelibrary.wiley.com/wol1/ doi/10.1111/j.1467-9264.2015.00384.x/full. Sosa, Ernest. 2009. A Virtue Epistemology: Apt Belief and Reflective Knowledge (Vol. 1). Oxford: Oxford University Press. Sosa, Ernest. 2010. Knowing Full Well. Princeton: Princeton University Press. Sosa, Ernest. 2015. Judgment and Agency. Oxford: Oxford University Press. Srinivasan, Amia. 2015. Are We Luminous? Philosophy and Phenomenological Research 90(2): 294–319.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

16  J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis Stalnaker, Robert C. 1978. Assertion. In Peter Cole, ed., Syntax and Semantics: Pragmatics, volume 9. Cambridge, MA: Academic Press, 315–32. Stanley, Jason. 2005. Knowledge and Practical Interests. Oxford: Oxford University Press. Williamson, Timothy. Forthcoming. Justifications, Excuses, and Skeptical Scenarios. In Julien Dutant and Fabian Dorsch, eds., The New Evil Demon. Oxford: Oxford University Press. Williamson, Timothy. 1995. Is Knowing a State of Mind? Mind 104(415): 533–65. Williamson, Timothy. 1996. Knowing and Asserting. Philosophical Review 105(4): 489–523. Williamson, Timothy. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Zagzebski, Linda Trinkaus. 1996. Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical Foundations of Knowledge. Cambridge: Cambridge University Press.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Pa rt I

Foundational Issues

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

2 How and Why Knowledge is First Clayton Littlejohn

1. Introduction It is an extremely common mistake.1 People think that our beliefs cannot attain any sort of positive epistemic status unless they are based on good reasons.2 According to the reasons-first approach to epistemic status, reasons and the possession of them are prior to epistemic status.3 In point of fact, the opposite is true. When you know that something is true, it is true that you have reasons in your possession, but you possess these reasons as a result of coming to know. There is nothing prior to knowing that puts these reasons in your possession. Because of this, the proponents of reasons-first epistemology are mistaken in thinking that the possession of reasons is a necessary precondition for the attainment of status. Thus, there is an important sense in which knowledge comes first. It comes first in the sense that, at the beginning, we come to have reasons in our possession by coming to know that certain things are true. 1   I have benefited from conversations with Maria Alvarez, Robert Audi, Bill Brewer, Wesley Buckwalter, John Callanan, Adam Carter, Elijah Chudnoff, Chris Cowie, Jonathan Dancy, Christina Dietz, Trent Dougherty, Julien Dutant, Mikkel Gerken, Sandy Goldberg, Ali Hasan, Scott Hagaman, John Hawthorne, Jennifer Hornsby, Robert Howell, Nick Hughes, John Hyman, Maria Lasonen-Aarnio, Dustin Locke, Heather Logue, Errol Lord, Jack Lyons, Brent Madison, Aidan McGlynn, Rachel McKinnon, Jon Matheson, Ram Neta, Duncan Pritchard, Susanna Siegel, Chris Tucker, John Turri, and Tim Williamson. I also thank four anonymous referees for their helpful feedback on an earlier draft. 2  An anonymous referee reminded me that this is often assumed in stating Agrippa’s Trilemma. Among the defenders of the view that justification requires the support of reasons/evidence are Comesana (2010), Conee and Feldman (2004), and McDowell (1998). While this view is commonly held, it is not commonly argued for. I have never seen an argument for it. Some authors (e.g., Gibbons (2010)) define ‘reason’ as something that occupies a certain theoretical role (i.e., that of making things reasonable), but this approach trivializes the claim and conflicts with an approach that thinks that we have some independent grip on the notion of a reason. As I think the claim is non-trivial and think we have an independent grip on the notion, I will not follow his lead. 3   These reasons are usually understood as evidence. For arguments that a subject’s reasons for believing things consists of a subject’s (apparent) evidence, see Adler (2002) and Shah (2006). A subject’s reason for ϕ-ing is a motivating reason, a special kind of explanatory reason (i.e., a reason why a subject ϕ’d). An important constraint on a theory of motivating reasons is that they can be good reasons to ϕ. See Dancy (2000). For a general discussion of the relationship between reasons of different kinds and the ontology of reasons, see Alvarez (2010) and Littlejohn (forthcoming).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

20  Clayton Littlejohn Knowledge is first because it is distinctive. There is nothing else that could put us in a position to believe, do, or feel things for reasons. According to the knowledge-first view defended here, you cannot have a reason in your possession unless it is something you know. Reasons are facts and the possession of them requires knowledge. In defending this view, I shall defend a crucial part of the view that your evidence is all and only what you know (E=K). Much of the literature on E=K has been concerned with questions about whether evidence has to consist of truths, whether it can be inferential, and whether Gettier cases cause trouble for E=K.4 Defenders of E=K have done a nice job fending off challenges on these fronts, but some issues haven’t been explored in sufficient depth.5 It seems that knowledge requires justified or appropriate belief.6 If it does, E=K implies that your evidence includes p only if you appropriately or justifiably believe p. If knowing p does require justifiably or appropriately believing p, E=K tells us two surprising things about the possession of evidence. The first is that you cannot possess something as evidence unless you believe it.7 This rules out the possibility that your non-inferential beliefs are justified by virtue of  being supported by evidence.8 Some alternative story has to be told about how   McGlynn (2014) provides a detailed survey and critical discussion of this literature.   The challenges have focused on things like whether evidence can be inferential, whether it must be propositional or factive, whether evidence has to be believed, and whether the externalist implications of E=K are acceptable. For a discussion of whether evidence can be inferential, see Bird (2004) and Neta (2008). For discussion of whether evidence has to be propositional, see Dougherty (2011). For discussion of whether propositional evidence is factive and requires belief, see Littlejohn (2012, forthcoming) and Williamson (2009). For discussion of the externalist implications, see Pritchard (2012). 6   I will treat these notions interchangeably. Maria Alvarez has convinced me (whether she intended to or not) that this holds only for mature humans. There are some non-human animals where it seems that they know things but cannot be held accountable in the way that they would have to be to make sense of the idea that their actions, emotions, or beliefs are justified. On my view, then, the relation between knowledge and justification is akin to the way that utilitarians conceive of the relation between being right and being optimific. A heroic dog might perform optimific actions. It is an open question whether they have acted rightly. They might know that it is time for dinner. It is an open question whether this belief is justified. If this is a good idea, it is probably hers. If it is not, it is probably mine. 7   This clashes with many of the standard claims about propositional justification (e.g., that it does not require belief but does require reasons or evidence). Obviously, I will have to defend it further on. If you wanted to offer a knowledge-first approach to propositional justification, you could do it in terms of what you’re in a position to know, but then you’d have to give up some of the standard views about the relation between propositional justification and basing. 8   For further defense of the idea that there’s nothing that is your reason for believing p if your belief is non-inferential, see Littlejohn (forthcoming) and McGinn (2012). One argument for the view that there is nothing that is your reason for believing p when that belief is non-inferential uses McDowell’s (1978) characterization of a subject’s reasons and points out that on this gloss only that which is known independently from believing p could be your reason for believing p. Readers who want to insist that cases of non-inferential belief are cases in which there’s something that is your reason for believing what you do are free to introduce their own conception of a subject’s reasons, but the standard glosses on it do not allow for non-inferential beliefs to be based on reasons. A simple point that is often neglected is that Unger’s (1975) argument that all propositionally specified reasons have to be known to be a subject’s reasons holds true for all the relevant verbs, even those used to talk about beliefs that are not formed through inference. For further arguments for the knowledge requirement, see Hornsby (2007, 2008) and Hyman (1999). 4 5

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  21 these beliefs attain positive epistemic standing. The second is that possession has to be understood in normative terms. This rules out the possibility that the possession of evidence could play a grounding role, grounding epistemic statuses like justification, rationality, or knowledge.9 The possession of evidence cannot ground such statuses if  the attainment of such statuses is required for a piece of evidence to be in your possession. Proponents of the traditional reasons-first approach are likely to challenge both the idea that possession itself should be understood in normative terms and the idea that possession requires belief. If they want to challenge these consequences of E=K, they will need to offer their own answers to these two questions: The Constitution Question:  What do your reasons for believing what you do ­consist of? The Possession Question:  What does it take to have these reasons? We will see that the answers available to the proponents of the reasons-first view are not very good. The proponents of the reasons-first approach cannot say that our evidence consists of the facts we know, so they turn to perception or perceptual experience to give them an account of constitution or possession. On one version of the view, evidence consists of experiences, and having evidence is simply a matter of  having these experiences. On another more promising version of the view, ­evidence consists of facts made manifest in successful perceptual engagement with our ­surroundings. We will see that neither approach works. Once we see that, we will see that knowledge is distinctive and thus, must come first in the sense that I have claimed. Here is my plan of attack. In Section 2, I argue against internalist versions of reasonsfirst epistemology. Any view that offers acceptable answers to the constitution and possession questions has to explain how it is possible to believe, feel, and do things for reasons that are the facts that we have in mind when we believe, feel, and do things for reasons. Internalist views deny that this is possible, so they cannot explain what needs to be explained. In Section 3, I look at externalist versions of reasons-first epistemology according to which perception provides us with reasons, and I argue that perception is not the kind of thing that could provide you with reasons because perceptual contact doesn’t put you in a position to be guided by reasons that consist of facts. In Section 4, I conclude with an argument for a knowledge-first account of justification. 9   Conee and Feldman (2004) are the leading proponents of the view that evidence does play this grounding role. Littlejohn (2012, forthcoming) and Sylvan and Sosa (forthcoming) criticize their ­evidentialist view on the grounds that the possession of evidence has to be understood in normative terms, although they disagree about what is involved in possession. Beddor (forthcoming) criticizes their grounding claim on the grounds that they use normative notions to identify the theoretical role that evidence occupies.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

22  Clayton Littlejohn

2. Internalism This section looks at some views that combine the reasons-first approach to epistemic status with this supervenience thesis: Internalism:  Necessarily, if two subjects are in the same non-factive mental states, these subjects have the very same evidence.10 According to the internalists, it is not possible for you to have evidence that your systematically deceived counterparts don’t have. If you and your counterparts’ both ϕ and your mental states stand in the same causal relations, you’ll ϕ for the very same reasons. According to some internalists, this is part of the intuitive motivation for the view.11 Many internalists think that it is intuitively obvious that (a) any reasons that your deceived counterpart has, you’ll have and (b) any reasons that you have will be reasons that your deceived counterpart has. If there cannot be any difference in justificatory status without a difference in possessed reasons or evidence, the (purported) fact that these internal duplicates share the same reasons is supposed to explain why these subjects’ beliefs are equally justified. To flesh out the details of this internalist view, internalists have to say something about the constitution and possession of reasons. Internalism doesn’t answer these questions directly. Rather, it only tells us that facts about such reasons (including facts about what’s in your box of evidence) supervene upon your non-factive mental states. This doesn’t tell us whether a subject’s evidence consists of mental states or events (statism), the contents of those states or events (propositionalism), or facts (factualism), but we’ll see that once we see what reasons are and what’s sufficient for their possession that the internalists are mistaken.12 Many internalists now identify a subject’s evidence or a subject’s reasons for her beliefs with states of mind. On this approach, having a reason or a piece of evidence is simply a 10   This is mentalism about evidence, but not (necessarily) about justification. Mentalists often say that propositional justification supervenes upon a subject’s evidence, but we’ll see below that there are some reasons to think that this is not a helpful way to think about things. See Littlejohn (2015) for an attack on the idea that justification could be understood in terms of strong evidential support. 11   See Cohen (1984), Madison (2014), and Silins (2005) for a defense. 12   For defenses of the statist view, see Brueckner (2009), Cohen (1984), Conee and Feldman (2008), Dougherty and Rysiew (2013), Gibbons (2010), Haack (2009), Lyons (2009), McCain (2014), Mitova (forthcoming), and Turri (2009). For defenses of propositionalism, see Dougherty (2011), Fantl and McGrath (2009), Miller (2008), and Schroeder (2008, 2011). For defenses of factualism, see Alvarez (2010), Hornsby (2007, 2008), Hyman (1999), Littlejohn (2012), Unger (1975), and Williamson (2000). Most participants to this debate aren’t terribly concerned about the relationship between propositions and facts and are happy to say that they think of true propositions as facts. Thus, the crucial difference between the propositionalist and factualist view is on whether p’s being a reason depends, in part, upon whether p is true. Because of this difference, I take the propositionalist approach to be unacceptable. Here’s a quick argument. If your reason for ϕ-ing was that p, you ϕ’d because p. (If it is not true that you ϕ’d because p, the claim that your reason for ϕ-ing was that p would be mistaken.) The sentences ‘You fell because the floor was slippery’, ‘The chocolate melted because the radiator was on’, and ‘You went to the store because you were out of gin’ are true only if the floor was slippery, the radiator was on, and you were out of gin. I won’t discuss the propositionalist view further here as I have yet to see any response to this argument.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  23 matter of having an attitude or an experience. A proponent of reasons-first epistemology needs an account of possession that doesn’t require a belief that attains positive standing to possess a reason and this approach would seem to meet this constraint. While Davidson (2001: 141) defended the view that only beliefs could be reasons for belief, most statist internalists defend the view that experiences constitute our reasons for belief. Consider, for example, Dougherty and Rysiew’s experience-first view.13 As they see it, experience is “first . . . in the order of immediacy: in short, experience is where we begin” (2013: 17). Elaborating on this theme, they say: This is where experience is first: in the quest for true belief, justification, knowledge, understanding, and wisdom we have no other starting point than experience. Our experiences . . . are our basic evidence, in the light of which all else that is evident is made evident. Experiences play well the roles that characterize evidence  (2013: 17).

And later: Any chain of cited evidence must end with the way the world appears to us to be. So on this view, experience is first in that it inhabits the ground floor of the intellectual evidence  (2013: 18).

Is this right? Does our ultimate evidence consist of experiences? Davidson thought not. It is instructive to consider his argument. Davidson (2001: 141) argued that the only thing that could be a reason for a belief would be another belief, but on the grounds that only that which stands in a logical relation to a belief could serve as a reason for it. When it comes to belief, nothing could be a reason to believe p unless it stood in some sort of logical relation to p. Relatedly, nothing could function in reasoning as my reason to believe p unless it was something I took to stand in a logical relation to p. If I don’t take there to be some logical connection between p and q, q cannot be what convinced me that p by settling the question whether p. As such, it is hard to see how it could be my reason. A standard statist response is to say that experiences can stand in logical relations because of their contents. This is a muddle. It is a muddle because this assumes that if experiences were like beliefs in having content, they could be like beliefs in being reasons. Alas—beliefs aren’t reasons. As such, arguing that experiences are like beliefs is not a particularly good strategy for showing that experiences are reasons. It is a good strategy for showing that they cannot be reasons. Neither beliefs nor experiences stand in logical relations. Experiences are events. They are datable, coarse-grained particulars. Datable occurrences and coarse-grained particulars don’t themselves stand in logical relations or constrain rational degrees of belief. They don’t entail anything. They aren’t entailed by anything. (Events lack truth-values.) If experiences had contents (more on this further on), their contents might stand in logical relations, but as experiences 13   Experience-first can be understood as a version of the reasons-first view. That’s how I’ll understand it here. The experience-first view can be understood as a view on which a subject’s evidence consists of experiences, but it can be understood to include views like McDowell’s on which reasons are facts that are made manifest in experience.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

24  Clayton Littlejohn aren’t their contents, it doesn’t follow from the fact that contents stand in logical relations that experiences do. Your reasons for feeling, doing, or believing things are things that can figure in reasoning. What figures in reasoning as things that you reason from (i.e., reasons) are things that are propositional (i.e., facts or propositions), not things that have contents that are propositional (i.e., states of mind). Properly understood, Davidson’s logical relations argument rules out both his view that identifies epistemic reasons with beliefs and any version of the experience-first view that identifies epistemic reasons with experiences. What Davidson’s argument shows is that reasons have to be propositions or facts, the things that we have in mind when we believe, feel, or do things for reasons, not things in the head when we believe, feel, or do things for reasons. Our reasons for believing, feeling, and doing things could not be things in the head because (a) our reasons for believing, feeling, and doing things are things that we take to merit or make appropriate these responses, and (b) we all know full well that such things are the things we have in mind, not the states of mind themselves.14 Consider the case of emotion. There is a reason why Agnes is upset and then there is Agnes’ reason for being upset with the neighbors upstairs. When we use the possessive construction to talk about Agnes’ reasons (i.e., motivating reasons), we’re interested in reasons that help us see what it was from Agnes’ point of view that was so upsetting. This has to be something that helps us to understand why, from her perspective, it makes sense to be upset or why it is appropriate to be upset with the neighbors upstairs. If you saw things as she did, it would be something like the fact that they make so much noise and sign for your packages without ever bringing them down. For this reason, a motivating reason has to be something of which Agnes is cognizant, and so there are epistemic constraints on the proper ascription of motivating reasons.15 To be Agnes’ reason for being upset with the neighbors upstairs, something has to be such that Agnes is both cognizant of and averse to. It has to be something that Agnes wants not to be.16 If we follow the statists’ lead in identifying Agnes’ reasons for believing things with states of mind, we’d have to identify Agnes’ reasons for feeling things with those same states of mind. It is possible to believe something for the very same reason that you feel something. If Agnes heads upstairs to confront her neighbors, her reason for believing that she should go confront them could be her reason for heading upstairs and her reason for being upset with the neighbors upstairs. There is no good statist candidate to play the role of Agnes’ reason for being upset with her neighbors, so 14   Following Dancy (2000) we should say that a subject’s reasons are the kinds of things that can be good reasons since it is possible to ϕ for a reason that’s a good reason to ϕ. Since the good reasons there are to ϕ are facts, not states of mind or their false contents, Dancy’s constraint supports the view that reasons are facts. 15   This does not hold for all explanatory reasons, mind you. If you are deceived by a demon, that is one reason why you would think that you had hands. It is not your reason for thinking you have hands! 16   See Gordon (1987) for discussion.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  25 there is no good statist candidate to play the role of Agnes’ reason for believing that she should go complain. Remember that Agnes’ reason for being upset is something that she desires not to be the case. Maybe she believes that the neighbors upstairs signed for one of her packages (a case of wine) and are conspiring to keep it. She wants her wine and so wants them not to keep it. She doesn’t want not to believe that the neighbors are holding onto her wine. Her anger has to do with their actions, not her attitudes. The reason that this is trouble for the statist is that the statist has to identify all of a subject’s reasons with that subject’s attitudes and that’s just a deeply implausible account of what Agnes’ reason is for being upset with her neighbors. Once we see that, it is easy to see that statism is a deeply implausible account of what Agnes’ reason is for believing that she should go complain. It is also a deeply implausible account of what her reason is for heading upstairs to demand the return of her package. Agnes’ reasons are facts about the situation. They are the facts that she has in mind when she believes things, feels things, or does things for reasons. Agnes is not different from us. Her reasons are the kinds of things that our reasons are. The trouble with internalism as defined here is that it conflicts with the idea that Agnes’ reason for being upset could have been something that the neighbors have done or are doing. Notice that, if Agnes’s reason for being upset with the neighbors did not merit this kind of affective response, her emotional response would not be fitting. If, for example, the package she saw being carried upstairs belonged to them, she might be angry with the neighbors, but her being angry with them would not be fitting. A fitting emotional response requires that a subject’s reason for having that emotional response is the fact that merits that response. Any view that implies, as internalism and statism do, that a subject’s reason for her emotional responses could not be the facts that merit such a response imply that fitting emotional responses are impossible. Let us add this to the long list of reasons for rejecting this approach to reasons.17 I’ve argued that it would be a mistake to follow the statists and the internalists in characterizing a subject’s reasons as consisting of states of mind or experiences. A subject’s reasons for believing, feeling, and doing things will be facts about the situation that she has in mind and the proponents of reasons-first or experience-first epistemology have to explain how this is possible.

3. Externalism Any view that gives us acceptable answers to the constitution and possession questions has to accommodate the idea that it is possible to believe, feel, and do things for reasons that consist of the facts you have in mind when you believe, feel, and do things for reasons. Since internalist views deny that your reason for ϕ-ing could be such 17   In Littlejohn (2012), I offer further arguments against statist and internalist approaches to reasons of all sorts and try to address all the main arguments for these approaches in the literature.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

26  Clayton Littlejohn things, these views are unacceptable. We cannot think of your reasons for ϕ-ing as consisting of states of mind or the contents of such states and we cannot think of your reasons for ϕ-ing as consisting of facts about your own mind. Thus, while states of mind and mental events have some sort of rational role to play, they play a role in providing reasons, a role they play without being reasons. My proposal is that only knowledge provides reasons in the sense that matters here. If you ϕ, it is only when you know p that your reason for ϕ-ing could be that p. (If you did not know p, you could still ϕ but could not ϕ for the reason that p until you came to know it.) If this is right, knowledge plays a distinctive role in our story about how we’re guided by reasons. If my opponents are right, something else can play this role. I begin by highlighting three constraints that something has to meet if it is going to figure in an account of possession. The first constraint has to do with ability. Consider Nozick’s (1981) experience machine. Agnes undergoes a series of experiences that dispose her to form false beliefs about her surroundings. It seems to her that she and everyone she cares about are flourishing. In the standard telling, her beliefs are all mistaken. This is not essential to the story. Agnes can be cut off from reality even if some of her beliefs happen to be true.18 Let us suppose that it seems to her that her brother has just crossed the stage at graduation and a smile stretches across Agnes’ face because she believes he just graduated.19 What the lab technicians 18  An anonymous referee reminded me that there will be messy details having to do with the contents of her thoughts. We can control for some of this if we focus on the beliefs she forms soon after the switch. 19   In the bad case, she can ϕ because she believes p without ϕ-ing for the reason that p. In the bad case, the fact that her brother just received his diploma is not her reason for smiling or cheering. In the bad case, there’s nothing that could be Agnes’ reason for being so happy but there are reasons why she’s happy. In the good case, her reason for being happy might well be that her brother just received his diploma. If so, that’s a reason why she is smiling and a reason why she’s happy. In this case it is also true that she is smiling and she is happy because she believes her brother just graduated. There is no competition in describing the good case between the fact about the situation and the fact about her mental life. For further discussion of these kinds of points, see Alvarez (2010) and Littlejohn (2012). Some critics think that it is a weakness of this view that we can’t say what it is in the bad case that was Agnes’ reason for being happy. It is not a weakness; it is a strength. Some critics try to identify that reason but then say obviously false things about what it is (e.g., a falsehood or a state of mind). Other critics criticize my proposal but (wisely) don’t try their hand at specifying what this reason is. An anonymous referee raised the following worry:

Given that she does not know that her brother has graduated, Agnes either lacks a normative reason to be happy or lacks a normative reason to be happy of which she is entitled to avail herself, but does she not have a motivating reason to be happy—a reason for which she is happy—constituted by the proposition that her brother has graduated? There are two things to say in response to this. First, the lack of knowledge has nothing to do with the presence or absence of normative reasons. You can have reasons to ϕ that consist of facts unknown to you in the sense that such reasons apply to you and help determine what you should do (e.g., a reason to revise your beliefs because unbeknownst to you your evidence doesn’t support them or a reason to rethink your ranking of CVs because unbeknownst to you your assessment of applicant quality is biased, etc.). Second, there is nothing that is the subject’s reason for being happy. There are reasons why she is happy, but explanatory reasons need not be a subject’s reasons. For a defense of that point about the relationship between a subject’s reasons (i.e., motivating reasons) and explanatory reasons, see Alvarez (2010). Critics of factualism have the unfortunate habit of ignoring important pieces of linguistic evidence at just this point, which is that attributions of propositionally specified reasons entail correlative knowledge attributions. See, for example, Comesana and McGrath (2014) and Fantl (2015). Everyone seems to think that knowledge

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  27 do not realize is that precisely as Agnes undergoes this experience her brother crosses the stage and accepts his diploma. While she believes correctly that her brother is graduating and is happy because she believes this, her reason for being happy is not that her brother is graduating. She cannot be rationally guided by such a fact not when she’s cut off from reality. If p is your reason for ϕ-ing, you have to have the ability to be guided by the fact that p, and that requires a non-accidental connection to it, one that’s missing from the Gettiered version of Nozick’s experience machine. This is the ability problem, which is the problem of stating a suitable account of what would eliminate the accidental ­connection between belief and fact, so that if you ϕ in the belief that p, your reason for ϕ-ing could be that p.20 The second constraint has to do with rational authority. Other things equal, acquiring evidence is supposed to improve your epistemic position. When you have it, you’re supposed to have the authority to treat the relevant reason as a reason in your deliberation. If your true belief is not rationally held, it would not be proper to include p in your reasoning. Thus, whatever condition or conditions we take to be sufficient for possession, we have to think of these condition or conditions as sufficient for underwriting an undefeated entitlement or warrant to treat p as a reason. The third constraint has to do with the generality of reasons and the role of conceptual capacities in coming to possess them. Davidson’s argument tells us that reasons stand in logical relations—relations in which only facts or propositions can stand. The facts that we’re primarily interested in are those that make perceptually based predicative judgments true. Such judgments are true only if some particular, a, belongs to some category (i.e., the Fs) by being an F. To grasp such reasons, we have to grasp the fact that some particular case belongs to a range of possible cases: A concept, as I will speak, is always of (being) such-and-such. As such it has a certain sort of generality . . . It may be, necessarily, of just one thing. There may even be one thing it is necessarily of. Still, there is a sort of generality it has. Suppose a concept were of being Frege. To fit that concept one could stop nowhere short of being him. But suppose (if you can) Frege had taken to wearing a beret, or had devoted his life to sailing. He would have been different than he was. Still, he would have fit that concept . . . There being at least one such range of cases is one thing one might mean by generality. It is what I will mean here by the generality of the conceptual. The key feature of the conceptual, on its present understanding, is that for anything conceptual there is a specific form of generality intrinsic to it. There is then a range attributions are factive, but this leaves no room for a non-factive account of attributions of motivating reasons. They also tend to ignore the obvious entailment from the attribution of motivating reasons to propositionally specified explanatory reasons (i.e., the entailment that holds between ‘S’s reason for ϕ-ing was that p’ to ‘A reason S ϕ’d was that p’. The latter is obviously factive. For important discussions of the linguistic evidence that supports the factivity of attributions of motivating reasons, see Unger (1975) and Gordon (1987). 20   While Hyman (1999), Littlejohn (2012), McDowell (1998), Mantel (2013), and Unger (1975) have a non-accidental connection requirement in their accounts of possession, not everyone agrees that this is a genuine desiderata. See Hofmann (2014) for discussion.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

28  Clayton Littlejohn which is the range of cases, or circumstances, which would be ones of something instancing that generality  (Travis 2013: 125).

To grasp such things requires the use of conceptual capacities, which are capacities to grasp that a particular belongs to a range of cases. If something doesn’t involve the  actualization of any such capacities, then it doesn’t relate us to the general, and if something doesn’t involve the requisite sort of generality, then it cannot be the sort of thing that could be true or false. Since reasons are truths, truths involve this sort of generality, and getting a grip of such generalities requires the use of c­ onceptual capacities, i.e., the account of possession has to implicate these conceptual capacities. Proponents of the knowledge-first approach can tick all the boxes. If knowledge is the ability to be guided by reasons that consist of facts, the ability problem doesn’t arise for E=K.21 Even if we reject this identification, knowledge involves a non-accidental connection to truth. The ability problem arises most sharply in those cases where the connection between belief and truth is accidental. As for the authority problem, it seems that this problem doesn’t arise for E=K because anything known is justifiably believed. To justifiably believe something, you must have sufficient ­epistemic authority to include p in some of your reasoning or deliberation. Finally, it is clear that propositional knowledge involves the exercise of the relevant conceptual capacities. If knowledge is unique in meeting these three constraints, it is distinctive in just the way I claim. Its being distinctive in meeting these three constraints means that there’s nothing apart from knowledge that involves the exercise of the conceptual capacities that must be activated for you to be rationally guided by the fact that a is F where you have both the ability and the authority to treat the fact that a is F as a reason for ϕ-ing. Not everyone thinks that knowledge is unique in this way, for it is now widely believed that perception can meet these constraints, too. If perception can tick all three boxes, it could provide reasons that could be the rational basis for your non-inferential perceptual beliefs. It could do the work I’ve claimed can only be done by knowledge. For it to do this work, however, it would have to tick all the boxes. In what follows, I argue that it cannot.

3.1  Reasons-first and the reconciliatory view of perception In Sections 3.1–3.3, I argue against a conjunction of these claims: Perceptual Sufficiency:  Perceptual relations alone between you and your ­surroundings can put you in the position to believe things for reasons that consist of facts. 21   For defense of this approach, see Hyman (1999). An anonymous referee asked whether this conflicts with the view, defended by Williamson (2000), that knowledge is a state of mind. I think this is an interesting issue and I have to confess that I don’t have any settled view on these matters. It doesn’t conflict with the idea that knowledge is mental, but it does seem to conflict with the idea that knowledge is a state. (It also seems to conflict with the idea that knowledge is an achievement, a claim that I’m quite happy to reject.) Hacker (2013) argues that knowledge is not a state even though it is signified by a stative verb, and I find his arguments to be rather persuasive.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  29 Perceptual Dependence:  Perceptual knowledge is possible only when the ­subject’s perceptual beliefs are held for reasons where these reasons are independently possessed because the subject bears the right perceptual relations to her surroundings. My argument starts from the assumption that perception is fundamentally a relation between a perceiver and things in her surroundings that perception brings into view. 22 If this relationalist picture of perception is correct, we should reject the conjunction of Perceptual Sufficiency and Perceptual Dependence. As the case for the relationalist view has been made elsewhere, I’ll assume that it is correct and also assume that the proponents of reasons-first epistemology will take this view on board. An important question is whether we can reconcile this approach to perception with the claim that perceptual experience has a representational content by virtue of involving the subject’s conceptual capacities. According to reconciliatory views, experience can be both relational and representational: Perception makes knowledge about things available by placing them in view for us. But it is precisely by virtue of having content as they do that perceptual experiences puts us in such relations to things  (McDowell 2013: 144).

To start my case against the use of reconciliatory views by proponents of the reasonsfirst approach, consider the first reconciliatory view: The First Reconciliatory View:  When we have perceptual knowledge that a is F, it is the result of seeing that a is F. By seeing that a is F, we’ll either have the fact that a is F or the fact that we see that a is F as part of our evidence.23 Seeing that a is F is understood as standing in the right visual relation to things in your surroundings and the fact that a is F is understood as the object of visual awareness. Having such facts as reasons requires an appropriate exercise of conceptual capacities (e.g., seeing that a is F requires exercising the conceptual capacities involved in characterizing something as an F), but seeing that a is F is nevertheless a relational affair.24 22   If this claim is correct, perception of things in our surroundings is explanatorily prior to the other features our experiences might have (e.g., their phenomenal character or their representational content). I cannot hope to give a defense of the idea that perception is fundamentally a relation between the perceiver and her surroundings. For defenses of the idea, see Brewer (2011), Logue (this volume), Martin (2009), and Travis (2011, 2013). Ginsborg (2011), Logue (2014), McDowell (2009), and Schellenberg (2011, 2014) all defend reconciliatory views that are supposed to reconcile representational and relational views of perception. 23  According to McDowell (2006), a subject’s reason for believing p is never p, so he thought that a subject’s reason for believing p would be that she saw that p. Pritchard (2012) defends a similar view. Hopp (2012) and Schroeder (2011, forthcoming) hold that p can be a subject’s reason for believing p. I think McDowell is right that your reason for believing p could never be p, but nothing turns on this issue here. 24   McDowell once defended this sort of view, but he has since modified his position. (See his (2009)). A referee asked what sorts of views should count as relational. Views of the kind defended by Burge would not count as relational because the experience we have when we see is not a relation to things in our surroundings, but a representational state that is about things in our surroundings. Someone might think, as Burge does, that the experience you have when you see something in your surroundings is not a perception, but the

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

30  Clayton Littlejohn This view would appear to be the view that McDowell (1998) defends, which he recently described as follows: Suppose I have a bird in plain view, and that puts me in a position to know non-inferentially that it is a cardinal. It is not that I infer that what I see is a cardinal from the way it looks . . . I can immediately recognize cardinals if the viewing conditions are good enough . . . Since my experience puts me in a position to know non-inferentially that what I see is a cardinal, its content would have to include a proposition in which the concept of a cardinal figures: perhaps one expressible . . . by saying “That’s a cardinal”  (2009: 259).

Initially, it will help to focus on this view because it is a version of the reasons-first view that has these two features: Visualism:  Facts about possessed reasons for visual beliefs supervene upon facts about a subject’s visual contact with her surroundings.25 Content Constraint:  If S knows visually that a is F, S’s visual experience has the representational content that a is F. Visualism is a crucial assumption in McDowell’s epistemological argument for his disjunctivist conception of experience according to which your experience can be a matter of some fact, such as the fact that a is F, being made manifest. Such a fact is made visually manifest iff the subject sees that a is F. This fact being made visually manifest is, in turn, supposed to put you in a position to be rationally guided by it (i.e., to be in a position to believe things for the reason that a is F or for the reason that you see that a is F).26 If Visualism is false, it is possible for two subjects to stand in the very same perceptual relations to their environments and yet possess different reasons for their perceptual beliefs. The difference in reasons would not be traced to a difference in what they saw or what their experiences were like, so it would be beyond their ken. The subject that had reasons inaccessible to the second subject would thus enjoy an epistemic benefit because of something beyond her ken. For McDowell, this is not allowed. Since nothing can confer any sort of epistemic benefit upon you unless it is within your ken, there must be some difference in the subjects’ experiences that corresponds to the difference in the reasons they possess. This is just what Visualism says. proponents of the view I have in mind would reject this. The experience you have when you perceive something is the perception, but you could have a very similar experience if you did not perceive things but were merely hallucinating. I would expect the proponents of the first reconciliatory view to be disjunctivists. For this and more of Burge’s thoughts on disjunctivism, see his (2005). For Travis’ thoughts on those thoughts, see Travis (2013: 259–313). 25   The kind of visual contact will include simple seeing, in Dretske’s sense. It is an extensional affair (i.e., seeing the F will just be seeing the G if the F is the G). One of the interesting questions I’ll explore further on is whether we should think of visual recognition as supervening upon such visual relations. For an argument that there must be more to visual recognition than standing in such relations, see Dietz, Knowledge and Recognition (Manuscript) where she argues that recognition just is knowledge. 26   See McDowell (1998).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  31 The Content Constraint captures the idea that there’s a match between the content of perceptual beliefs that constitute knowledge and the content of the perceptual experiences that provide the reasons that put us in a position to know the things we know in forming such beliefs. McDowell acknowledges that the Content Constraint could be true only if the conceptual capacities exercised in believing that a is F are active in the experience you have when you see that a is F. As the passage above makes clear, the content of an experience, which is due to the exercise of conceptual capacities, helps determine what’s presented to us in experience. He would reject the idea that these capacities operate on that which is made present in experience anyway independently from the exercise of these conceptual capacities. Mere sensibility, McDowell would say, doesn’t bring things into view, but only does so in concert with the operation of the understanding.27 One way to put pressure on the first reconciliatory view is to think about environmental luck cases. In these cases, a subject lucks into a true belief by correctly predicating a property to some object she sees, even though she easily could have mistakenly predicated this property to a ringer and thereby ended up with a false belief. In a suitably good case, it seems that we can know by looking that something is a barn, say, which gives us (1): 1. In the good case, it is possible to have perceptual knowledge expressed by, ‘This structure is a barn’. 2. The subject has this perceptual knowledge only because the experience has the same content as this belief and because this experience justifies beliefs with this content [(1), Visualism and the Content Constraint]. 3. In a correlative bad environmental luck case, a subject could stand in the same visual relations to her environment that our first subject stands in with respect to hers, but the belief she expresses by saying ‘This structure is a barn’ would not constitute knowledge. 4. This subject would have the very same reasons for her perceptual beliefs and an experience that would justify the belief expressed by ‘This structure is a barn’, and would believe for the very same reason as the subject in the good case [(3) and Visualism]. 5. The subject in the bad case would thus know that the structure is a barn [(4) and the assumption that if you believe something for the reason that p, you can ϕ for the reason that p if you ϕ in the belief that p].

27   For McDowell, the understanding is operative when our conceptual capacities are operative. In his view, conscious experience is the joint product of sensibility and the operation of understanding, so things are only brought into view for us by having been brought under concepts. (We can work with a very liberal conception of concepts here so that placing something into a category is sufficient for bringing something under a concept.)

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

32  Clayton Littlejohn This is an inconsistent set of claims, so something has to give. Someone could block the reasoning that generates the inconsistency at the outset by denying (1).28 If we do this and leave the background restrictions in place, we’ll end up having to say that perceptual knowledge is limited to knowledge of how things seem to us. Any perceptible feature of a thing can appear to be different than it is. In light of this, it seems we should be able to generate fake barn style cases for a wide range of properties (e.g., a person’s identity, a person’s mood, the colors of surfaces, the kind of thing a thing is, etc.). If the possibility of such cases is sufficient to establish that the scope of non-inferential knowledge is so limited that it could not include properties like being a  barn, it would be difficult to defend the view that our non-inferential knowledge includes knowledge of some other properties of external things (e.g., their colors or shapes). We should be able to tell by looking what color a thing is if we’re in ideal conditions, so (1) should stay. It looks like (5) follows from (4). Suppose you believed that you were out of gin for the reason that one of your guests just poured the remaining gin into the shaker. If this is your reason for believing this, this could be your reason for being disappointed and your reason for reaching for the vodka. It could not be your reason for feeling this or doing that unless you knew that one of your guests just poured the remaining gin into the shaker. You can only believe something for the reason that p if you’re able to believe, do, and feel further things for that reason. Thus, if responding in these further ways requires knowledge of p, your belief in p would have to constitute knowledge. McDowell suggests, in keeping with his view, that a subject in an environmental luck case neither sees that the structure is a barn nor knows that it is a barn. Since he wants to use the visual relations the subject bears to her environment to determine what she’s in a position to know, it looks like he’ll have to abandon Visualism, the Content Constraint, or (3). Unfortunately, it seems that (3) is rather plausible, particularly if you think, as he does, that his view vindicates the intuitions that relationalists appeal to in motivating their view, such as the intuition that, ‘perception places our surroundings in view’ (2008: 14). The presence or absence of unseen fakes should have no bearing whatever upon whether vision places your surroundings in view for you.29 Some writers have tried to turn this objection on its head, arguing that the kind of perceptual relations we stand in when we’re in the environmental luck cases are actually good enough for something epistemically good. Turri (forthcoming) suggests that the relations provide knowledge. Hughes (2014) thinks that it puts us in a position 28  An anonymous referee notes (rightly) that we can block this reasoning by denying that sameness of  reasons implies sameness of knowledge. Indeed, it does. Elsewhere (2012: 133–44), I’ve addressed McDowell’s (1998) arguments for the claim that sameness of reasons implies sameness of knowledge. My aim here is slightly different, which is to argue that experience cannot shoulder the epistemic burden that certain epistemologists require of it. Contrary to what McDowell (1998), Pritchard (2012), and Turri (forthcoming) suggest, when we perceive things and experience places things in view, it does not thereby provide reasons that guarantee that we’ll be in a position to know. 29  An anonymous referee observed that this was not as obvious to her/him as it seems to be to me. See further on for the reasons that support this claim.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  33 to believe things for the reasons that are the facts we see, even if we don’t know them to obtain: In Fake Barns, the fact that there is a barn is (along with various facts about Henry) one of the causes of Henry’s belief that there is a barn. It figures in a causal explanation of why Henry believes what he does  (Hughes 2014: 459).30

If I understand the thought here, the idea is that the target fact (i.e., that the building is a barn) explains the belief that there’s a barn on the hillside in just the same way it would if there weren’t any fakes present. We’d agree that a subject’s reason for doing things could be that the structure is a barn if the subject is in the good case, so we have to admit that it is the subject’s reason in the bad. The guiding idea seems to be that perceptual contact with the subject of predication and veridical experience is sufficient for putting the subject in a position to do things for the reason that’s the target fact because the same explanatory relations hold between that fact and your attitudes in the environmental luck case as hold in the correlative good case in which the fakes are removed. I think this is a mistake. While the causal relations between the barn and its properties might be the same in the good case and bad, it doesn’t follow that the facts about the barn play the same explanatory role in the two cases. Some cases should make this clear: Lucky Penny:  Jill has a lucky penny. She hasn’t seen other pennies before. Her brother stole her lucky penny and took it with him to school. He dropped it. Someone picked it up but later dropped it. It worked its way across the city. A week later Jill was on a school trip when she looked down and saw a penny that happened to be her lucky penny. ‘It is my lucky penny!’ she said.31 White Diamond: Agnes had a bucket of fake diamonds marked ‘diamonds’ that she left in her flat to attract the attention of any jewel thieves that might break in. Each stone in the bucket looked like a real diamond. What she did not realize is that one of the hundreds of stones in the bucket was indeed a real diamond. That stone happened to be sitting on top. A thief knocked the bucket over clumsily, saw the stones spilled across the floor, saw that the bucket was labeled diamonds, and grabbed the first stone that she could, believing it to be a diamond. She left the others because she thought that she heard someone coming. She happened to grab the only diamond from the bucket and believes that she has a diamond. 30   I used to think he was right, but I now think that he is not for reasons discussed in Littlejohn (2014). Like Carter (2013) and Jarvis (2013), I’m skeptical of the idea that the environmental luck cases are cases in which the correctness of your predication is attributable to the exercise of your abilities. 31  An anonymous referee noted that it might be irrational for the subject to judge that this is her lucky penny and that this accounts for why she could not know. Perhaps, but part of the issue is whether she can see that it is her lucky penny. I think she cannot. If propositional seeing is a purely visual matter (something that this same anonymous referee is rightly skeptical of), it is hard to see how the rationality of belief could bear on what is seen.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

34  Clayton Littlejohn These are similar to environmental luck cases insofar they involve veridical experience and simple seeing with nothing that interferes ‘in between’ the subject and the fact. These cases strike me as being clear non-knowledge cases, and it is also clear, I think, that the facts they believe to hold cannot rationally guide them or be their reasons for believing, feeling, or doing things. If the combination of simple seeing and veridical experience bringing about a true belief were sufficient for the target facts to be the causal explanation of the belief, we’d have to say that the fact that Agnes took a diamond explains why she believes she has a diamond and the fact that the penny Jill sees is her lucky penny explains why she believes what she does. I don’t think their beliefs are explained by these facts, for while they see the relevant objects and have veridical experiences, the features of the things that they see don’t enable them to recognize, say, that the rock is a diamond, or that the penny is the lucky penny. That’s because recognizing the penny as the lucky penny or the diamond as a diamond is possible only if these things have a distinctive look that would trigger a disposition to classify these things in the relevant ways. Since they don’t have this look, the correctness of the classification cannot be attributable to the subject’s sensitivity to the features of the objects that would show that the relevant facts obtained (i.e., that the stone is a diamond and that the penny is the particular penny). Once we see that, we see that the subject’s beliefs aren’t explained by the target facts, in which case the case against (3) fails.

3.2  The second reconciliatory view It might seem that the problems that have to do with environmental luck call only for a minor modification of the first reconciliatory view. Could we not just add some further condition or conditions (e.g., safety?) to rule out environmental luck? The Second Reconciliatory View:  When we have perceptual knowledge that a is F, it is the result of seeing that a is F. By seeing that a is F, we’ll either have the fact that a is F or the fact that we see that a is F as part of our evidence. Seeing that a is F is understood as standing in the right visual relation to things in your surroundings when further conditions are met (i.e., conditions that don’t supervene upon the visual relations between the perceiver and her environment). Having such facts as reasons requires an appropriate exercise of conceptual capacities (e.g., seeing that a is F requires exercising the conceptual capacities involved in characterizing something as an F), but seeing that a is F is nevertheless a relational affair.32 I don’t think minor modifications will work. For a starter, it is hard to see that there’s anything left of the idea that a fact is an object of visual awareness once it is conceded that visual relations to a subject’s surroundings alone don’t determine whether the subject sees that a is F. If seeing that a is F is a visual relation, whether you see that a is F   This seems to be the sort of view that Pritchard (2012) defends.

32

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  35 has to supervene upon purely visual relations between a perceiver and her environment, because everything supervenes upon itself.33 If seeing that a is F depends upon further modal conditions (e.g., what would be brought into view in nearby cases), conditions that don’t supervene upon visual relations between a perceiver and her environment, it is just not clear what the reconciliatory view’s positive proposal is. It is clear what the proposal cannot be. It is clear that the proposal cannot be one according to which a perceiver stands in the relation of visual awareness to something that’s her reason and thereby is able to believe things for reasons that she’s visually aware of. Holding relations of visual awareness fixed, we can modify the further conditions and thereby determine whether the subject sees that a is F and whether the subject can believe things for either the reason that a is F, or the reason that she sees that a is F. To put to rest the idea that it is visual awareness that puts us in a position to believe things for reasons, consider an argument distilled from Travis’ work: The Generality Argument 1. Everything we perceive is particular. 2. The facts that we know perceptually aren’t particular, but general, as they are facts that have to do with the categories that the particulars we see belong to. 3. Thus, the objects of perceptual awareness are not the objects of perceptual knowledge.

What we see when we see some particular in our surroundings is not that the particular fits into a class of similar things: But don’t we see that the sun has risen? And don’t we thus also see that this is true? That the sun has risen is no object which sends out rays that reach my eyes, no visible thing as the sun itself is  (Frege, quoted in Travis 2013: 123).

What we see in our surroundings and what we know about these things belong on different sides of Frege’s line. On the left are the particulars that instance generalities and on the right, we find the conceptual, the representational, and the general (Travis 2013: 126). The stuff on the right represents the stuff on the left as belonging to categories or kinds. Once we appreciate this point, we cannot say that perception takes in some fact or something that involves the generality that facts have, and thereby places a reason before us that we might use to come to know how things are. All that perception can do is place particulars in view. At this stage, people often say that they appreciate the force of this point but see no need to abandon the reconciliatory view. Even if the fundamental relation is between a perceiver and the particulars she sees, mightn’t she also have an experience while she sees the things she does where this experience has a 33   McGinn (1999) seems to have thought that seeing that a is F is a matter of being visually aware of a fact, but I find French (2012) and Ranalli’s (2014) arguments to the contrary to be persuasive. For further arguments that we don’t see facts, see Moltmann (2013).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

36  Clayton Littlejohn representational content that enables her to believe things for reasons connected to this content? Provided that we don’t claim to perceive the content, is not such a view perfectly compatible with Travis’ Fregean insight? Ultimately, I think not, but the argument of this chapter does not require any sort of  demonstration of the incompatibility of relationalist with representationalist approaches to perceptual experience. My aim is simply to show that there’s no good reason to think that perception’s power to provide knowledge doesn’t derive from its power to provide reasons.

3.3  A third reconciliatory view? McDowell has backed off the idea that when we see that a is F, this is a matter of being made visually aware of the fact that a is F. Facts, he now concedes, aren’t things that we see or perceive.34 He was moved to abandon this view by Travis’ observation that everything we perceive in our surroundings is particular. Nevertheless, he wants to retain the idea that our conceptual capacities are active in experience. Whereas the relationalist wants to say that conceptual capacities play no role in bringing the particulars we see into view, and only are active when it comes to judging that something brought into view is an instance of some kind, the reconciliatory view says that such conceptual capacities must be active for perception to bring into view what it brings into view. The actualization of such capacities is essential to  experience and doesn’t merely operate on the particulars that are present in experience anyway. Let us consider a third reconciliatory view, which seems to avoid the difficulties mentioned previously by jettisoning the idea that the perceptual relation has facts as one of its relata: The Third Reconciliatory View:  What we see is not ever a fact or something that’s true. Nevertheless, conceptual capacities are active in experience and the things we see are presented as being instances of kinds or as having properties. The conceptual capacities do not operate on things that are present in experience anyway; rather, they are active in bringing particulars into experience, and so, into view. It is because these things are brought into view, in part, because of the operation of these conceptual capacities, our predicative judgments are based on reasons.35 Against this proposal, I shall argue that it is a mistake to think of the representational character of experience as essential to perceptual knowledge. Representation is essential to the account of perceptual knowledge that says that such knowledge arises because our perceptual judgments are based on reasons, but this shows that it is not essential to our understanding of perceptual knowledge that it is constituted by beliefs based on reasons.   See McDowell (2009: 258).   

34

  This seems to be closer to Ginsborg (2011) and McDowell (2009).

35

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  37 The Problem of Transduction 1. If we are not visually aware of facts, either experience has no propositional content, or there is a transductive process that takes non-propositional input (e.g., an object seen) and yields a propositional content for visual experience.36 2. If the former, the propositional content of experience is not epistemically essential in the sense that it is required for knowledge because we have perceptual knowledge. 3. If the latter, the propositional content of experience is not epistemically essential because there is a possible creature with our perceptual knowledge that acquires this knowledge without this intervening process yielding a propositional content of experience. Instead, there is a transductive process that takes non-propositional input and yields content for visual belief. 4. If we are not visually aware of facts, the propositional content of experience is not epistemically essential. 5. We are not visually aware of facts. 6. The propositional content of experience is not epistemically essential.

The guiding thought is that if we think of perception as fundamentally being a relation between the perceiver and the particulars she sees, her experience will only have representational content the accuracy of which turns on how things are with this perceived particular if there is some process that yields a content by taking in the particulars we see and generating some sort of content from that. At the ground level, what is taken in would have to be something we see if, indeed, the perception grounds the representational character of the experience.37 Once we see this, it is hard to see what reason there could be for thinking that this transductive process that takes a non-propositional input and yields something that is the kind of content that can bear logical relations to belief could not simply yield a belief ’s content as opposed to an experience’s content. If the aim is to acquire knowledge, I see no advantage for wiring things up one way rather than another. 36  An anonymous referee asks why there would have to be a transductive process? Why, she/he asks, could we not say that having a visual experience is simply entailed by seeing an object? I think that having an experience is entailed by seeing an object, but the tricky issue is whether having an experience with representational content could simply be so entailed. I would think that we would need a transductive process to understand how perceptual knowledge is possible, because without it, it would be impossible to understand how an experience with the content that it had could be properly grounded in the way that the perceived objects are. Without this grounding, the representation could not reliably provide accurate representations true of the things we see. (Remember that all sides have agreed that the things we are related to in perception is not itself a representation, but a thing that can be represented more or less accurately.) 37  An anonymous referee suggested that even if relational perception is more fundamental than representational experience, it could be that nothing further has to happen for perception to give rise to representational content. I do not think it could be that easy. The perceptual relations that we bear to the things we see could not select some unique way of representing them. If they did, representationalists would have to deny the possibility of two perceivers perceiving the same thing where only one perceived it accurately. They would also have to say that any two perceivers that perceived the same things would have experiences that visually represented the objects as having the very same properties. I do not think that many representationalists would want to say that the possibility of my dog and me seeing a squirrel turns on our having pairs of experiences that attribute all and only the same properties to the squirrel.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

38  Clayton Littlejohn On the pure relationalist view that I would favor, what is present in experience is always particular, and so a’s belonging to the range of cases that makes for an F’s being present cannot be present to the perceiver through her conscious experience of a. What is present is something of a’s that merits the classification of a as being an F.38 The perceiver’s knowledge that a is F depends upon whether she has the ability to classify things as F when aware of the features of a that merit this classification. It does not require some representation of a as being F that’s prior to belief or judgment. By contrast, the proponents of the third reconciliatory view insist that the understanding and the conceptual capacities do not operate upon some particulars that sensibility anyway brings into view without the operation of these conceptual capacities. They think these particulars are only brought into view by an exercise of these capacities in concert with the operation of sensibility. Since they think that the experience and its representational character is grounded in how things are with the particulars brought into view, they must think that the understanding and our conceptual capacities operate on something particular in generating an experience with that content and an experience the accuracy of which turns on how things are with the particulars we ultimately see. What is most unclear is what advantage there is in saying that the conceptual capacities operate on particulars in generating the experience and its representational character, as opposed to saying that conceptual capacities operate on the particulars present in experience through the operation of sensibility alone. So far, the problem is merely a challenge. The challenge can be met if someone can produce an argument that shows that there must be some sort of suitable representational content in place that would then play a role in explaining why the subject believes what she does. It is hard to imagine how this challenge might be met, but let us push this worry a bit further. One way to get representation into the picture, if you are skeptical of the idea that experience is itself representational, is to introduce recognitional abilities into the picture as something that operates upstream from the experience on the things that experience puts us in touch with anyway, quite independently from the exercise of any conceptual capacities.39 On this model, a subject sees an object, a, and the perceptual experience the subject has of a just is the subject seeing a. There are recognitional capacities upstream from belief that account for the subject’s disposition to judge that a is an F. The alternative model that the proponents of reasons-first defend is that the subject sees an object, a, and has a perceptual experience as of a’s being F. The experience has a representational character determined, in part, by a and, in part, by the subject’s discriminatory and recognitional capacities. These capacities account for the subject’s disposition to judge that a is an F. 38   The range of possible particulars that could play this role is extensive. This could be a’s parts, a’s condition, or tropes. 39   For a discussion of the role that recognitional abilities play in the acquisition of perceptual knowledge, see Millar (2011). One of the aims of this chapter is to show that his approach to perceptual knowledge gives us the resources to overcome some of the problems that arise for epistemological disjunctivists like McDowell (1998) and Pritchard (2012).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  39 The main differences between these accounts are these. The first view doesn’t see experience as having a representational character shaped by the exercise of conceptual or recognitional capacities, but the second one does, and sees the operation of these capacities as playing an epistemic role by fixing the content of that experience. I think that this second view is the sort of thing that Ginsborg has in mind when she offers this example: [I]magine two people who have different discriminative capacities with respect to the experience of music. One can discriminate among chords of different qualities (for example, major, minor, and diminished triads), the other among the timbres of different but related instruments (cornet, trumpet, saxophone). It is natural to think that each of them, listening to the same major chord played by a saxophone trio, will hear it differently. Because the first is, as we might put it, sensitive to the harmonic qualities of what she is hearing but not to the timbre, her experience will register its character of being a major triad. The experience of the second, conversely, will register the characteristic saxophone sound of the chord, but not its harmonic quality  (2011: 151).

If her proposal is correct, a certain feature of the sound is present in the experience because of the operation of a discriminatory capacity (which may or may not be a recognitional capacity), but the third reconciliatory view says that what is present in the experience will be the particular sounds the subjects hear. In contrast, any difference between the two subjects’ experiences, the relational view says, will be upstream from experience and will not have any bearing on the representational content of the experience. The proponents of this third reconciliatory view face a trilemma. What is the basis of the relevant predication when the subject makes a perceptual judgment? If the subject judges, say, that some sphere is red, some chord is a minor chord, some wine is acidic, etc., should we say that the basis of the predication is (a) something particular that is present in experience, (b) something representational, or (c) something present in experience that’s representational?40 It looks like we can strike (c) from the list. It was agreed upon by all sides that the way for defenders of the reconciliatory view to reconcile the representational character of experience with relationalism is to acknowledge that what we see is not general. What (c) says is that the basis for the predicative judgment is both seen and representational. This is a mythical beast, and one that would have to straddle Frege’s line. What about (a)? The trouble with this is that if we say that the basis for the predication you make when you judge that, say, this sphere is red, this chord is a minor chord, or this wine is acidic, we are saying that the basis for the predication is something particular, something that cannot stand in the kinds of logical relations that reasons do. If the basis is the particular, there could not be something that is your reason for predicating

40  An anonymous referee suggests that these options need not exclude each other, and I think that that is right. The argument is supposed to show that these options will not do the work alone or in combination.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

40  Clayton Littlejohn this property of this particular. This is fatal to the representationalist view, but it is a view that I have a soft spot for as I prefer the first view to the second. What about (b)? If we opt for (b), can’t we save the reasons-first view by putting representation back into the picture? Putting representation back into the picture comes at a cost. You cannot have predication done for a reason unless there is something general that’s a suitable representation that guides the predication, but if the basis for the predication is something suitably general and representational, notice that this implies that the basis for the predication is not present in perceptual consciousness. Imagine staring at a red sphere in your hand. You have a choice. You can think of what is present to consciousness as the basis for judging that this is red, in which case, the predication of redness is not something done for a reason. You can think of the predication as done for a reason, but the reason for which you would predicate redness of the ball is not something visible and it is not part of something visible. Of the two approaches, the first strikes me as being more plausible. At some point between you and the ball, representation comes into the picture, but the part of the representation that has to do with predication is not itself ever made visible. Thus, it is hard to see how something representational is your basis (even in part) for predicating anything of anything. If the third reconciliatory view cannot deliver the goods for the proponents of reasons-first epistemology, none can. There cannot be a fourth view that does what the previous views cannot, for any modification to the views considered thus far will either take the objects of visual awareness to be facts or particulars, and we’ve seen that none of these views can help the proponents of the reasons-first view make sense of the idea that our perceptual beliefs are based on reasons. There is nothing provided by perception alone that could be your reason for believing things, doing things, or feeling things, so knowledge is epistemically distinctive. Of course, perception helps us acquire knowledge, but it doesn’t do so by providing us reasons that could be in our possession prior to or independent from the possession of perceptual knowledge.

4.  Evidence and Justification I’ve focused thus far on the relationship between knowledge and the possession of evidence or reasons, but I haven’t said much about the relationship between knowledge and the attainment of an epistemic status like justification. I conclude with a brief discussion of the relationship between reasons and resultant epistemic status. On the traditional view of things, justified beliefs are beliefs based on good reasons. This view requires that there’s a way of having or possessing reasons that doesn’t involve the target belief, i.e., the belief whose justificatory standing is supposed to be explained in terms of the rational support of the relevant reasons. So much for that view! In the

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  41 non-inferential case, your belief that p cannot be justified by being based on some reasons that you would possess independently from having formed the belief that p. In the non-inferential case, there is nothing that could be your reason for believing p.41 Let us consider a different approach to the relationship between reasons and epistemic status. Instead of seeing status as turning on what reasons support a belief, we can see status as turning on whether beliefs bring reasons to us that could then support further things (e.g., doxastic, affective, and practical responses). Consider a simple closure principle for justification: J-Closure:  If you justifiably believe X and know that Y is a logical consequence of X, you can justifiably infer Y if you come to believe Y by means of competent deduction. A natural explanation of J-Closure is one that appeals to this connection between the justificatory status of a belief and reasons: J-Reasons:  If you justifiably believe X and can justifiably infer X’s known consequences, you’d be able to ϕ for the reason that X.42 To get a feel for J-Reasons, suppose you justifiably believe that Agnes was the one who finished off the gin. This belief might be inferential or non-inferential. If it is inferential, it might be based on deductive reasoning or some sort of non-deductive reasoning. All we know about this belief is that it is justifiably held. It could, presumably, be justifiably held, even if you have lost the original grounds for forming it. Suppose you deduce that Agnes once drank gin. What J-Reasons says, in effect, is that if this is something that you can justifiably believe, in part, because you justifiably believe that Agnes was the one who finished off the gin and because you can justifiably believe all manner of known consequences of the first belief, you’re able to believe, feel, or do something for the reason that Agnes was the one who finished off the gin. If J-Reasons is part of the best explanation of J-Closure, perhaps we should accept it. The idea would be that competent deduction transmits a status (i.e., justification) because beliefs that attain that status enable you to believe further things for reasons. 41   In saying that there is nothing that is your reason for believing, say, Agnes is on the sofa if this belief is non-inferential, I am not saying that there is no reason why you believe this. There is. You see her and you recognize her. Such things can explain why you believe and explain how you know without being your reasons for believing anything at all. You needn’t have any attitude whatever to such things for them to explain why you believe or how you know, but they could not be your reason for anything if you did not have any attitude toward them. An anonymous referee has urged me to recognize the difference between saying (i) the belief ’s status as knowledge is not a matter of the belief ’s being formed in view of some reason and (ii) the belief that is knowledge being based on a reason that consists of the fact known (i.e., that p) or that she sees that p. As my main concern is with (i), I’m happy to acknowledge this distinction and note that once we see that (i) is true, we can see that a standard story about the acquisition of knowledge is one we should stop telling. Having said that, I have some residual worries about how (ii) is developed. I don’t think that your belief that p is true can be sustained by the fact that you see that p, not if talk of ‘seeing that p’ is really just a way of talking about the subject’s knowledge. (We would not want to say that the fact that sustains my belief that it is raining outside is the fact that I know that it is.) 42   What matters is whether you know X’s consequences. Thanks to an anonymous referee for raising this.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

42  Clayton Littlejohn If this idea is correct, it has a rather surprising consequence. If this idea is correct, the distinction between justification and knowledge breaks down. J-Reasons implies this surprising thesis: J-K:  If you justifiably believe X, you know X. This is because of this link between ability and knowledge: A-K:  You cannot ϕ for the reason that X unless you know X. Once we see that you cannot ϕ for the reason that X unless you know X, we can see that you cannot justifiably believe what you don’t know.43 Justified beliefs are beliefs that provide you with reasons that can be your reasons for believing things, feeling things, or doing things. If you believe, feel, or do things for reasons, these reasons must be known. Thus, the original belief must itself constitute knowledge. Why is this? Why must beliefs constitute knowledge to be justifiably held? The obvious answer is that the justificatory status of a belief depends upon whether it can provide rational support by providing us with reasons. It seems that a plausible candidate for the fundamental norm of belief is one that tells us that beliefs are supposed to provide us with reasons: RN:  You shouldn’t believe p unless your belief that p is true can provide you with reasons that can be your reasons for ϕ-ing.44 It turns out that only knowledge can play this role in providing reasons. Since only knowledgeable beliefs do that, we now know why belief is governed by this norm: KN:  You shouldn’t believe p unless you know p. Since these norms determine what’s justified, right, permitted, appropriate, etc., the best you can hope for if you believe what you don’t know is an excuse. We now have the beginning of an argument for the claim that knowledge is the fundamental norm of belief, and one that explains why we should think of the aim of belief as knowledge, not merely truth.45 Once we see that knowledge, not perception, is what provides us with reasons, we can see that possessing evidence requires a commitment to truth that’s not present in perceptual contact with our surroundings. In turn, we have to think of the justificatory status of beliefs as turning on what support these beliefs can provide, not exclusively on what supports these beliefs. 43   This view has been defended by McDowell (1998), Steglich-Petersen (2013), Sutton (2007), Unger (1975), and Williamson (forthcoming) although the arguments that these authors offer differ from the one offered here. I believe that this is the first attempt to derive a standard of justification like KN from claims about the connection between knowledge and ability. 44   This norm differs from the kind of norm that McKinnon (2013) defends for assertion for her norm focuses on the reasons that support belief in a proposition whereas this norm focuses on whether a belief can provide reasons that support other things. As I think that some things are known without being based on reasons, I would not want to introduce a norm that requires of each justified belief that it is supported by reasons. 45   For further defenses of the view that knowledge is the fundamental epistemic norm, see Adler (2002), Littlejohn (2013), Sutton (2007), and Williamson (2000).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  43

References Adler, J. 2002. Belief ’s Own Ethics. Cambridge, MA: MIT University Press. Alvarez, M. 2010. Kinds of Reasons. Oxford: Oxford University Press. Beddor, B. Forthcoming. Evidentialism, Circularity, and Grounding. Philosophical Studies. Bird, A. 2004. Is Evidence Non-Inferential? Philosophical Quarterly 54: 252–65. Brewer, B. 2011. Perception and its Objects. Oxford: Oxford University Press. Brueckner, A. 2009. E=K and Perceptual Knowledge. In P. Greenough and D. Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 5–12. Burge, T. 2005. Disjunctivism and Perceptual Psychology. Philosophical Topics 33: 1–78. Carter, A. 2013. A Problem for Pritchard’s Anti-Luck Virtue Epistemology. Erkenntnis 78: 253–75. Cohen, S. 1984. Justification and Truth. Philosophical Studies 46: 279–95. Comesana, J. 2010. Evidentialist Reliabilism. Noûs 44: 571–600. Comesana, J. and M. McGrath. 2014. Having False Reasons. In C. Littlejohn and J. Turri, ed., Epistemic Norms. Oxford: Oxford University Press, 59–81. Conee, E. and R. Feldman. 2004. Evidentialism. Oxford: Oxford University Press. Conee, E. and R. Feldman. 2008. Evidence. In Q. Smith, ed., Epistemology: New Essays. Oxford: Oxford University Press, 83–104. Dancy, J. 2000. Practical Reality. Oxford: Oxford University Press. Davidson, D. 2001. Subjective, Intersubjective, Objective. Oxford: Oxford University Press. Dietz, C. Manuscript. Knowledge and Recognition. Dougherty, T. 2011. In Defense of Propositionalism about Evidence. In T. Dougherty, eds., Evidentialism and Its Discontents. Oxford: Oxford University Press, 226–35. Dougherty, T. and P. Rysiew. 2013. Experience First. In J. Turri, M. Steup, and E. Sosa, eds., Contemporary Debates in Epistemology, 2nd edition. Oxford: Wiley-Blackwell, 17–21. Fantl, J. 2015. What is it to be happy that p? Ergo 2(12): 267–97. Fantl, J. and M. McGrath. 2009. Knowledge in an Uncertain World. Oxford: Oxford University Press. French, C. 2012. Does Propositional Seeing Entail Propositional Knowledge? Theoria 78: 115–27. Gibbons, J. 2010. Things That Make Things Reasonable. Philosophy and Phenomenological Research 81: 335–61. Ginsborg, H. 2011. Perception, Generality, and Reasons. In A. Reisner and A. Steglich-Petersen, eds., Reasons for Belief. Cambridge University Press, 131–57. Gordon, R. 1987. The Structure of Emotion. Cambridge: Cambridge University Press. Haack, S. 2009. Evidence and Inquiry, Expanded Edition. Amherst, NY: Prometheus Books. Hacker, P. 2013. The Intellectual Powers. Oxford: Wiley-Blackwell. Hofmann, F. 2014. Gettier for Justification. Episteme 11: 305–18. Hopp, W. 2012. The (Many) Foundations of Knowledge. In D. Zahavi, ed., The Oxford Handbook of Contemporary Phenomenology. Oxford: Oxford University Press, 327–48. Hornsby, J. 2007. Knowledge in Action. In A. Leist, ed., Action in Context. Berlin: De Gruyter, 285–302. Hornsby, J. 2008. A Disjunctive Conception of Acting for Reasons. In A. Haddock and F. Macpherson, eds., Disjunctivism: Perception, Action, Knowledge. Oxford: Oxford University Press, 244–61. Hughes, N. 2014. Is Knowledge the Ability to ϕ for the Reason that P? Episteme 11: 457–62. Hyman, J. 1999. How Knowledge Works. Philosophical Quarterly 49: 433–51.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

44  Clayton Littlejohn Jarvis, B. 2013. Knowledge, Cognitive Achievement, and Environmental Luck. Pacific Philosophical Quarterly 94: 529–51. Littlejohn, C. 2012. Justification and the Truth-Connection. Cambridge: Cambridge University Press. Littlejohn, C. 2013. The Russellian Retreat. Proceedings of the Aristotelian Society 113: 293–320. Littlejohn, C. 2014. Fake Barns and False Dilemmas. Episteme 11: 369–89. Littlejohn, C. 2015. Stop Making Sense? On a Puzzle About Rationality. Philosophy and Phenomenological Research. DOI: 10.1111/phpr.12271. Littlejohn, C. Forthcoming. Reasons and Theoretical Rationality. In D. Star, ed., The Oxford Handbook of Reasons and Normativity. Oxford: Oxford University Press. Logue, H. 2014. Experiential Content and Naive Realism: A Reconciliation. In B. Brogaard, ed., Does Perception Have Content? Oxford: Oxford University Press, 220–41. Logue, H. 2018. Perception First. In J. A. Carter, E. Gordon, and B. Jarvis, eds., Knowledge-First Epistemology. Oxford: Oxford University Press, 182–99. Lyons, J. 2009. Perception and Basic Beliefs. Oxford: Oxford University Press. McCain, K. 2014. Evidentialism and Epistemic Justification. Abingdon: Routledge. McDowell, J. 1978. Are Moral Requirements Hypothetical Imperatives? Proceedings of the Aristotelian Society (Supplementary Volume) 52: 13–29. McDowell, J. 1998. Meaning, Knowledge, and Reality. Cambridge, MA: Harvard University Press. McDowell, J. 2006. Reply to Dancy. In C. Macdonald and G. Macdonald, eds., McDowell and his Critics. Oxford: Blackwell, 134–42. McDowell, J. 2009. Having the World in View: Essays on Kant, Hegel, and Sellars. Cambridge, MA: Harvard University Press. McDowell, J. 2013. Perceptual Experience: Both Relational and Contentful. European Journal of Philosophy 21: 144–57. McGinn, C. 1999. The Concept of Knowledge. In C. McGinn, ed., Knowledge and Reality: Selected Essays. Oxford: Oxford University Press, 7–36. McGinn, M. 2012. Non-Inferential Knowledge. Proceedings of the Aristotelian Society 112: 1–28. McGlynn, A. 2014. Knowledge First? London: Palgrave Macmillan. McKinnon, R. 2013. The Supportive Reasons Norm of Assertion. American Philosophical Quarterly 50: 121–35. Madison, B. 2014. Epistemological Disjunctivism and the New Evil Demon. Acta Analytica 29: 61–70. Mantel, S. 2013. Acting for Reasons, Apt Action, and Knowledge. Synthese 190: 3865–88. Martin, M. 2009. The Reality of Appearances. In A. Byrne and H. Logue (eds.), Disjunctivism: Contemporary Readings. Cambridge, MA: MIT University Press, 91–116. Millar, A. 2011. How Visual Perception Yields Reasons for Belief. Philosophical Issues 21: 332–51. Miller, C. 2008. Motivation in Agents. Noûs 42: 222–66. Mitova, V. Forthcoming. Truthy Psychologism about Evidence. Philosophical Studies. Moltmann, F. 2013. Abstract Objects and the Semantics of Natural Language. Oxford: Oxford University Press. Neta, R. 2008. What Evidence Do You Have? British Journal for the Philosophy of Science 59: 89–119. Nozick, R. 1981. Philosophical Explanations. Cambridge, MA: Harvard University Press.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

how and why knowledge is first  45 Pritchard, D. 2012. Epistemological Disjunctivism. Oxford: Oxford University Press. Ranalli, C. 2014. Luck, Propositional Perception, and the Entailment Thesis. Synthese 191: 1223–47. Schellenberg, S. 2011. Perceptual Content Defended. Noûs 45: 714–50. Schellenberg, S. 2014. The Relational and Representational Character of Perceptual Experience. In B. Brogaard, ed., Does Perception Have Content? Oxford: Oxford University Press, 199–219. Schroeder, M. 2008. Having Reasons. Philosophical Studies 139: 57–71. Schroeder, M. 2011. What Does it Take to ‘Have’ a Reason? In A. Reisner and A. Steglich-Petersen, eds., Reasons for Belief. Cambridge: Cambridge University Press, 201–22. Schroeder, M. Forthcoming. Knowledge is Belief for Sufficient (Objective and Subjective) Reason. In J. Hawthorne and T. Gendler, eds., Oxford Studies in Epistemology 5. Oxford: Oxford University Press. Shah, N. 2006. A New Argument for Evidentialism. Philosophical Quarterly 56: 481–98. Silins, N. 2005. Deception and Evidence. Philosophical Perspectives 19: 375–404. Steglich-Petersen, A. 2013. Truth as the Aim of Justification. In T. Chan, ed., The Aim of Belief. Oxford: Oxford University Press, 204–26. Sutton, J. 2007. Without Justification. Cambridge, MA: MIT University Press. Sylvan, K. and E. Sosa. Forthcoming. The Place of Reasons in Epistemology. In D. Star, eds., The Oxford Handbook of Reasons and Normativity. Oxford: Oxford University Press. Travis, C. 2011. Objectivity and the Parochial. Oxford: Oxford University Press. Travis, C. 2013. Perception: Essays after Frege. Oxford: Oxford University Press. Turri, J. 2009. The Ontology of Epistemic Reasons. Noûs 43: 490–512. Turri, J. Forthcoming. Knowledge Judgments in ‘Gettier’ Cases. In J. Sytsma and W. Buckwalter, eds., A Companion to Experimental Philosophy. Oxford: Wiley-Blackwell. Unger, P. 1975. Ignorance. Oxford: Oxford University Press. Williamson, T. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, T. 2009. Replies to Critics. In P. Greenough and D. Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 279–384. Williamson, T. Forthcoming. Justifications, Excuses, and Sceptical Scenarios. In F. Dorsch and J. Dutant, ed., The New Evil Demon. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

3 Against Knowledge-First Epistemology Mikkel Gerken

1. Introduction This chapter begins with some preliminary discussion of the nature of knowledge-first epistemology. On this basis I argue that reductive knowledge-first epistemologists commit a mistake similar to the one with which they charge their opponents. This is the mistake of seeking to reductively analyze basic epistemic phenomena in terms of other allegedly more fundamental phenomena. I then turn to non-reductive knowledge-first epistemology and argue that it fails in its application to epistemic norms of action and assertion. Knowledge norms of assertion and action have been thought to exemplify the virtues of the knowledge-first approach (Williamson 2000, Hawthorne and Stanley 2008). So, if this instance of knowledge-first epistemology is problematic, it raises general doubts about the program. To conclude on a positive note, I return to the methodological level and outline the contours of an alternative ‘equilibristic methodology.’ It has it that a number of core epistemic and doxastic phenomena are so interrelated that they cannot be reductively analyzable although they may be co-analyzed non-reductively. Thus, I suggest that there isn’t a single epistemic phenomenon, word or concept that is “first”. Rather, several core epistemic phenomena are (roughly) equally explanatorily basic. This approach preserves some grains of truth in knowledge-first epistemology: The idea that knowledge is not reductively analyzable in terms of more basic epistemic phenomena as well as the idea that knowledge is explanatorily basic. However, it denies that knowledge is most explanatorily basic. Since no single epistemic phenomenon is first, knowledge is not first.

1.1  Terminology and notation I use the term ‘warrant’ as the genus for non-factive epistemic rationality. I use ‘entitlement’ for an externalist species of warrant and ‘justification’ for internalist species of it.1 1   The terminology derives from Burge (2003). See Graham (2012), Gerken (2013a,  2013b, and ­forthcoming a).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  47 This terminology clashes with other uses. Often ‘justification’ is used in a sense similar to my use of ‘warrant.’ Moreover, Williamson and others use ‘warrant’ as ‘a term of art, for that evidential property (if any) which plays the role of property C in the correct simple account of assertion’ (Williamson 2000: 243). The property C, in turn, marks the epistemic position required by assertion (Williamson 2000: 241ff). Such terminological clashes are irksome, but almost unavoidable. To minimize terminological confusion, I sometimes take the liberty to use my preferred terminology in paraphrasing and discussing views articulated with other terminologies. I will mainly discuss knowledge itself. But where it is necessary to discuss words or concepts, I will mention with single quotes and underlining, respectively. For example, the word ‘knowledge’ typically expresses the concept knowledge. On occasion, I use the phrase ‘the phenomenon of knowledge’ to refer to knowledge itself in a somewhat neutral manner. Thus, the phrase is not intended to suggest the appearance of knowledge. Rather, I use the phrase to avoid begging questions as the nature of knowledge. I will use phrases such as ‘epistemic phenomena’ in a similar manner.

2.  Characterizing Knowledge-First Epistemology I begin by considering some characterizations of the knowledge-first program. It will transpire that it is not trivial to characterize a knowledge-first program that is worthy of distinction.

2.1  Inadequate characterizations of knowledge-first epistemology A good number of the broad theses and methodological stances that knowledge-first epistemologists hold dear fail to constitute a distinctive knowledge-first epistemology. The reason for this is that almost everyone holds these broad theses and stances dear. Indeed, ardent foes of knowledge-first epistemology, such as myself, accept many of them. This point is not trivial because knowledge-first epistemology is sometimes motivated by presupposing that opposition to it is in tension with the broad theses and methodological stances in question. But knowledge-first epistemologists and their opponents should agree that such a motivation is misguided. Opponents should reject it because it is a straw man argument. Proponents should reject it because it is a straw man argument. Moreover, proponents should reject it because the substantive core of the knowledge-first agenda may drown in overly programmatic appeal to theses and stances that almost everyone accepts. I do not include all of the theses and methodological stances figuring in the literature. Likewise, I avoid exegesis. Rather, I discuss a few representative ideas. Knowledge-first epistemology is not merely subscription to the thesis that knowledge cannot be given a reductive analysis. If it were so, a philosopher who harbors general skepticism about reductive analyses would ipso facto be a knowledge-first epistemologist. Likewise, knowledge-first epistemology must be more than the claim that knowledge is a primitive epistemic concept or phenomenon. For someone who

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

48  Mikkel Gerken regards a wide range of epistemic concepts or phenomena as primitive may not be a knowledge-first epistemologist. Indeed, this is the equilibristic view that I develop in Section 5. Furthermore, knowledge-first epistemology is not merely the contention that knowledge is central to epistemology. If this were the case, most epistemologists (before and after Williamson’s inauguration of knowledge-first epistemology) should be categorized as knowledge-first epistemologists. Finally, knowledge-first epistemology is not merely the idea that the word ‘knowledge’ is central to our folk epistemological talk or that the concept knowledge is central to our epistemological thought. These ideas are also accepted by most epistemologists—including opponents of knowledge-first epistemology (Gerken 2015a, 2017, forthcoming b). The theses mentioned are examples of ideas so weak and so widespread that they do not—alone or in conjunction—constitute any knowledge-first epistemology worthy of distinction.

2.2  Reductive and non-reductive knowledge-first epistemologies A substantive characterization of knowledge-first epistemology may begin with Williamson’s idea that the program ‘reverses the direction of explanation predominant in the history of epistemology’ (Williamson  2000: my italics). I italicize ‘reverses’ because it indicates a commitment to the idea that knowledge can be taken as basic and used to analyze other epistemic phenomena. We may distinguish between reductive and non-reductive knowledge-first ­epistemologies. The former is the view that knowledge can be used in reductive analyses or explanations of other epistemological phenomena.2 Such analyses often take, at least nominally, the form of a X = K thesis, where ‘X’ denotes an epistemic or doxastic ­phenomenon. But reductive knowledge-first analyses may also take the form of a biconditional. Such versions only purport to establish co-extensionality albeit sometimes a necessary co-extensionality. In either case, reductive accounts are the clearest examples of a distinctive knowledge-first epistemology. In contrast, it is not easy to characterize a non-reductive knowledge-first epistemology. It is not altogether clear whether Williamson is best characterized as a reductive knowledge-first theorist. At times, he appears to have reductive ambitions—for example, he defends the E = K thesis in terms of necessary co-extensionality of knowledge and evidence (Williamson 2000: 186). But often he is quite cautious.3 For example, he suggests an analysis of belief in terms of knowledge but makes the following qualification: ‘Although a full-blown conceptual analysis of believes in terms of knows is too much to expect, we can still postulate a looser connection along those 2   Maybe we need to require that knowledge can be used as the analysans in reductive analysis of a substantive range of epistemological or doxastic phenomena. Thanks to the editors. 3   Note also that Williamson holds that even if the concepts evidence and knowledge ‘ . . . are a priori, it does not follow that one is prior to the other’ (Williamson 2000: 186). The notion of priority is not explicated but a charitable reading of this remark attributes cautiousness about reductionism to Williamson. But such explicit cautioning has at times disappeared in subsequent debates.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  49 lines’ (Williamson 2000: 47; original italics which Williamson uses to mention the concepts. See also Williamson 2011: 211). Such explicit cautionings in Williamson’s inauguration of knowledge-first epistemology have sometimes gone amiss in subsequent knowledge-first epistemologizing. For example, in a later introduction, Williamson sketches an account of belief in terms of knowledge without any non-reductionist qualification or cautioning (Williamson 2013a: 4). Part of the elusiveness derives from the fact that Williamson is not always explicit about whether his aim is to give working explanations or theorizing explanations (Williamson 2009a in response to Cassam 2009). However, other knowledge-first theorists clearly have reductionist ambitions. Sutton explicitly pursues a reductionist knowledge-first epistemology sloganized as ‘knowledge first and last’ and Williamson has expressed sympathy for it (Sutton 2007).4 Despite Williamson’s role as originating knowledge-first epistemology, my aims are systematical rather than exegetical. My main point here is that arguing against reductionist knowledge-first epistemology is not to argue against a straw man. One notorious knowledge-first view that I will set aside is the thesis that justification or justified belief is knowledge (J = K) or, alternatively, that it requires knowledge. Each of these views entails that justification (warrant in my terminology) is factive (Sutton 2007, Littlejohn 2012). Williamson also entertains the idea that ‘ . . . a belief is fully justified if and only if it constitutes knowledge’ (Williamson 2013a: 5. See also 2011: 214). Though my aim is to defuse some ambitious versions of knowledge-first epistemology, I will not address views according to which justification (warrant) is factive. One reason for this is that such a view introduces special dialectical problems. A standard argumentative strategy is to argue that the target view leads to problematic consequences. But I have a hard time identifying epistemological consequences of the view that justification (warrant) is factive that I find more problematic than the view itself. Indeed, I am inclined to regard an argument that knowledge-first epistemology entails factivity of justification (warrant) as close to a reductio. But such an argument against knowledge-first epistemology is unlikely to be dialectically effective. Consequently, I will set aside the view that full justification (warrant) is knowledge in order to focus on aspects of knowledge-first epistemology where I hope that my engagement may be more productive. As mentioned, it is more difficult to characterize a non-reductive knowledge-first epistemology than a reductive program. However, one interesting approach consists in articulating substantive or even constitutive epistemic norms in terms of knowledge,   Indeed, Williamson scholars might take notice of Williamson’s blurb about Sutton’s book (Sutton 2007):

4

‘(Sutton) argues with vigor and originality for a knowledge-centered theory of justification, testimony, and inference that flies in the face of received opinion but grows ever more plausible the more one thinks about it.’ Despite the endorsement of a reductionist knowledge-first epistemology, I am hesitant to cite such a blurb as a ‘smoking gun.’ The discursive context differs from the scholarly contexts in which we normally articulate our commitments.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

50  Mikkel Gerken without purporting to provide a reductive analysis. To claim that knowledge is a constitutive necessary or epistemically sufficient condition on, for example, assertion is not to provide a reductive analysis of assertion in terms of knowledge. Rather, it is to claim that the postulated connection to knowledge partly characterizes assertion and helps distinguish it from other speech acts (Williamson 2000, Turri 2010, 2011). The claim that knowledge is the constitutive epistemic norm of assertion marks an interesting and distinctive non-reductive knowledge-first thesis (McGlynn  2014). Moreover, this approach seems to align with at least one of Williamson’s ways to promote the knowledge-first program: ‘I have shown elsewhere how a knowledge-first methodology casts light on such matters as the nature of indiscrimination and the norm of assertion’ (Williamson 2013a: 6–7. Here ‘elsewhere’ refers to Williamson 1996, 2000, 2007). As it will transpire, I think that it is an overstatement that knowledge-first epistemology has been shown to provide such insights. Yet it is important to acknowledge the methodological point that a broad knowledge-first program may pursue such insights without commitment to a reductionist methodology.

2.3  Characterizations of knowledge-first epistemology in conclusion The preceding discussion suggests that it is not trivial to characterize a knowledge-first program in epistemology. Among the characterizations that succeed in delineating a distinctive program in epistemology, we may distinguish between reductionist and non-reductionist versions. This calls for a piecemeal approach. Consequently, I will criticize reductionist and non-reductionist versions of knowledge-first epistemology in turn.

3.  Critique of Reductive Knowledge-First Epistemology As mentioned, knowledge-first epistemologists are fond of setting forth X = K theses where ‘K’ denotes knowledge and ‘X’ denotes some other epistemic or doxastic phenomenon. Notable examples involve E = K for evidence (Williamson 2000) and J = K for justification (Sutton 2007. See also Williamson 2013a). Sometimes, the equations are shorthand for biconditionals. For example, Sliwa argued for the thesis that ‘An agent understands that p if and only if she knows that p’ (Sliwa 2015: 58. See also Kelp 2014, Khalifa 2011, 2013). Sometimes an X = K thesis may simply label or ­sloganize a more modest analyses.5 To get clearer on this issue, let us have a closer look at the E = K thesis. In the case of E = K, Williamson is clear that the equation is at the level of the extension and that this does not entail any equation at the level of the concepts: ‘E = K equates the extensions of the concepts knowledge and evidence in any possible ­situation . . . By itself, E = K does not equate the concepts themselves; nor is it to be   See, e.g., Williamson on belief (Williamson 2000).

5

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  51 read as an analysis of either the concept evidence or the concept knowledge . . . ’ (Williamson 2000: 186). Since the claim is not one of conceptual identity, there is no simple argument, by transitivity of identity, to a commitment to the view that the concepts are identical. But if several further X = K theses are introduced, the result would be a long equation at the level of extensions—e.g., K = E = J = U . . . = X. Such a long equation would be at risk of reducing too many too distinct phenomena to knowledge. This should give rise to caution about adding yet a X = K (or X iff K) thesis to the existing equations (or biconditionals). It should be noted, however, that the X = K theses may or may not be intended as identity statements. Moreover, knowledge does not always stand alone on the righthand side of the respective X = K equations. But sometimes the definiens consists in little but knowledge. For example, Sutton articulates the J = K thesis as follows: ‘a subject’s belief that p is justified if and only if he knows that p’ (Sutton 2007: 8. See also Sliwa’s  2015 articulation of a knowledge-understanding biconditional). However, knowledge may also be the central aspect of the definiens of the various definienda. If so, various X = K statements may be best understood as shorthand slogans for more elaborate analyses at the level of the extension. Despite these qualifications, reductively analyzing doxastic or epistemic phenomena in terms of knowledge runs the risk of glossing over epistemologically important categories.6 Although all the phenomena in question may well be constitutively related to knowledge, the relationship is not plausibly that of identity. Of course, this worry is best substantiated by detailed piecemeal criticism of specific reductionist knowledgefirst analyses.7 But since I will direct my more specific critiques at non-reductionist knowledge-first claims, I will restrict myself to the point above and some broad methodological criticisms of the reductionist knowledge-first program. An overarching methodological worry with reductionist knowledge-first epistemology is that it is little but a traditional reductionist theory of knowledge in reverse. But a key problem with traditional reductionist theory of knowledge is its reductionism. As Williamson and others point out, the fact that a non-circular reductionist theory of knowledge has not been found may owe to the fact that there is no such analysis to be found: ‘The increasingly gerrymandered definitions were obvious signs of a degenerating research program. Most of them, if correct, seemed to make knowledge too grue-like to be worth analyzing. But in any case, they succumbed one after another to counterexamples’ (Williamson 2013a: 2). Although I have my doubts, I will here not address the historical question as to whether this is an entirely accurate characterization of post-Gettier theory of 6   For a reductive knowledge-first epistemologist who seeks to reduce other epistemic and doxastic concepts to the concept knowledge, the analog worry will be that this approach glosses over too many conceptual differences. For a reductive knowledge-first epistemologist who seeks to reduce other epistemic and doxastic words to the word ‘knowledge’, the analog worry will be that this approach glosses over too many linguistic differences. 7   See, for example, the extensive criticism of E = K (Schiffer 2009, McGlynn 2014).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

52  Mikkel Gerken knowledge. My point here is that critics of knowledge-first epistemology may agree with such a ‘negative abduction’. But they may do so on the grounds that there are few, if any, non-circular reductive analyses to be found in general. So, such negative abduction arguments do not compromise non-reductive non-knowledge-first ­epistemologies. Furthermore, if what drives the negative case for knowledge-first epistemology is the fact there are few, if any, non-circular reductive analyses to be found in general, it is rather dubious to replace traditional reductionism with knowledgefirst reductionism. It is one thing to cast traditional epistemology as a degenerating research program that consists of reductive analyses in which knowledge is the analysandum. It is another to do so and simultaneously pursue similar reductive analyses of other epistemic phenomena in which knowledge is the analysans. Even if knowledge is in some sense more fundamental than other epistemic and doxastic phenomena, why think that it may serve as analysans in a reductive analysis of them? For example, it is often argued that the concept knowledge is ontogenetically prior to the concept belief (Nagel 2013 gives an empirical case for this assumption, and Aidan McGlynn’s chapter (this volume) criticizes it). But even if knowledge is ontogenetically prior, there is a considerable step from there to the assumption that it is ‘first’ in an epistemological theorizing. Indeed, this is a giant step for those among us who reject that epistemology is primarily about conceptual analysis (Williamson 2007). I return to this point in Section 5. In a nutshell: Why isn’t reductionist knowledge-first epistemology prone to the very same sort of methodological problems that knowledge-first advocates claim beset prior reductionist epistemology? As far as I am aware, knowledge-first epistemologists have done little to answer to this question. This may simply indicate that there are few, if any, reductive knowledge-first epistemologists—I am genuinely unsure about this. But, as noted, if knowledge-first epistemology is not reductionist, it is not clear whether there is a distinctive knowledge-first program. Moreover, a good range of ongoing knowledge-first epistemology appears to be a reductionist epistemology in which knowledge plays the role of analysans rather than that of analysandum. Unfortunately, knowledge-first proponents are not always as clear about this as they could be. It would be helpful if knowledge-first proponents would be more explicit as to whether various X = K theses are set forth as reductive analysis, as a mere slogan or as something in between. In the cases in which a X = K thesis is set forth as a reductive analysis, the methodological queries above should be addressed head on. If not, the nature of the analysis or account that the X = K thesis sloganizes should be specified. The present broad methodological criticism is not meant to be the final word on reductionist knowledge-first epistemology. But perhaps it will encourage some reflection. More constructively, it motivates a non-reductionist alternative—equilibristic epistemology—that I set forth in Section 5. Meanwhile, I will turn to non-reductive knowledge-first epistemology.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  53

4.  Non-reductive Knowledge Norms of Assertion and Action Is there a distinctive knowledge-first epistemology that avoids the problems of reductionism? As mentioned, candidates for such a program may be those that involve constitutive knowledge norms. The view that knowledge is the constitutive epistemic norm of assertion in combination with the idea that knowledge is unanalysable, distinguishes a brand of knowledge-first epistemology. Thus, the knowledge account of epistemic norms marks one principled way of carrying out a non-reductive, but nevertheless distinctive, knowledge-first agenda (cf. McGlynn 2014: 17–18). Furthermore, the debates over epistemic norms have ramifications for other areas of epistemology. For example, knowledge norms have been invoked in arguments for pragmatic encroachment theories of knowledge.8 Likewise, DeRose argues that the ‘knowledge account of assertion demands a contextualist account of knowledge and is simply incredible without it’ (DeRose 2002: 182). However, Williamson (2005) and Turri (2010) uphold knowledge norms but reject pragmatic encroachment and contextualism. Given their significance, the epistemic norms of assertion and action are good candidates for an assessment of an important strand of non-reductive but distinctive knowledge-first epistemology. Such an assessment is a comparative exercise (Williamson 2013a, Benton  2014). Here I will focus on the epistemic norms of assertion. I have argued against knowledge norms of elsewhere (Gerken 2011, 2015a, 2017). So, here I will extend some of these arguments to bear on the knowledge norm of assertion. On this basis, I will argue that these problems indicate more general challenges for this brand of non-reductive knowledge-first epistemology.

4.1  Knowledge norms of assertion and their competitors Williamson articulates his knowledge norm as a necessary condition: One must: assert p only if one knows that p (Williamson 2000: 241). However, DeRose accepts a biconditional version and Hawthorne expresses sympathy for it (DeRose 2002: 217 and Hawthorne 2004: 23 fn. 58). Moreover, knowledge-first epistemologists tend to accept a biconditional knowledge norm of action (see Hawthorne and Stanley’s 2008 Reason-Knowledge principle. See also Fantl and McGrath 2002,  2009,  2012). Of course, proponents of a biconditional knowledge norm only uphold the epistemic sufficiency of knowledge for assertion. Every proponent of a knowledge norm of assertion, accepts that assertion is subject to other extra-epistemic norms such as norms of relevance (Grice 1989). Hence, there is no reduction of assertion to knowledge. Rather, knowledge is argued to be the single constitutive epistemic norm of assertion that distinguishes it from other speech acts (Williamson 2000, Ch. 11). 8   See, e.g., Fantl and McGrath (2009,  2012), Hawthorne (2004), Stanley (2005), and Hawthorne and Stanley (2008). See Brown (2008, 2010), Gerken (2011, 2014, 2015a, 2017) for elaboration and criticism.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

54  Mikkel Gerken Given the constitutivity claim, it is natural—although not mandatory—to accept the epistemic sufficiency of knowledge for assertion. Moreover, a biconditional knowledge norm is more apt to serve as cornerstone of a knowledge-first program than only a necessary condition. In any case, it will be worth considering a more ambitious biconditional knowledge norm of assertion which I label ‘KNAS’ for short (from Brown 2010 with minor reformulations): KNAS S is in a good enough epistemic position to assert that p iff S knows that p. Linguistic data cited in favor of KNAS includes patters of reasonable complaints about assertions without knowledge, Moore paradoxes, lottery paradoxes, and more (Benton 2014). Thus, the main dynamic of the contemporary debate has consisted in how opponents of knowledge norms responding to such phenomena (see McGlynn 2014 and McKinnon 2015 for surveys concluding that knowledge norms are not well supported). Here I seek to change this dynamic a bit by focusing on challenges to KNAS. Elsewhere, I have been in the defensive mode (Gerken 2015a). But, as Williamson rightly emphasizes, the knowledge norm of assertion must be evaluated in comparison with its alternatives: ‘In the long run, knowledge-first epistemology too should be judged by its fruitfulness as a research program, compared to its competitors’ (Williamson 2013a: 6–7). A fruitful comparison requires critical discussion of both phenomena that supports and compromises knowledge norm. Here I focus on the latter. Moreover, it will be important to present not only methodological but also substantive alternatives to knowledge-first epistemology. So, I will briefly present my favored ‘Warrant-Assertive Speech Act’ norm of assertion and note some of my reasons for preferring it to the knowledge norm.9 WASA In a conversational context, CC, in which S’s assertion that p conveys that p, S meets the epistemic conditions on appropriate assertion that p (if and) only if S’s assertion is appropriately based on a degree of warrant for believing that p that is adequate relative to CC.10 Although WASA is more complicated that KNAS, the idea is simple enough. The conversational context that S is in determines the threshold of warrant that S needs for asserting that p. The conversational context is constituted by a wide range of factors which include but are not restricted to S’s rational beliefs or presuppositions about: (i) alternative assertions (including qualified assertions). (ii) the availability of evidence for the asserted content (or what is conveyed by it). 9  I do not want to repeat myself. So, the presentation will be condensed and rely on other work (Gerken 2011, 2012a, 2012b, 2014, 2015a, 2015c, 2017). 10   Compared to (Gerken 2012a) there is a slight change in formulation of WASA, and DJA below. It pertains to parenthetical right-to-left direction of the biconditional. But the issue can be set aside here.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  55 (iii) the urgency of conveying the asserted content (or what is conveyed by it). (iv) the relevant stakes. (v) social roles and conventions.11 For example, if S rationally believes the stakes are high, further evidence is easily ­available and it is non-urgent to assert right away, then the degree of warrant required for assertion can exceed the degree of warrant that is required for knowledge (Brown 2008, Gerken 2012a, 2015c). But in a conversational context in which the stakes are so low that it would be more costly to gather further evidence than to assert something false, the degree of warrant required for assertion may fall below the degree of warrant that is required for knowledge.12 WASA involves contextual variation in the degree of warrant that is required for assertion. In this regard it differs from binary non-factive epistemic norms (Douven 2006, Lackey 2007, Kvanvig 2009, Madison 2010). But WASA moreover permits for a variation in the required kind of warrant (Gerken 2012a, 2015c, 2017). In some contexts, an externalist kind of warrant—entitlement—will do. Assume, for example, that Sally comes to believe that Jane is at the conference on the basis of catching a glimpse of her crossing the lobby in the distance. Even if Sally forgets the source of her belief, we may assume that it is entitled (Burge 1996: 38). Many externalists— myself included—may even regard Sally as knowing that Jane is at the conference. If someone casually asks Sally whether Jane has arrived, and the interaction must be brief, she may meet the epistemic conditions on asserting that Jane is at the conference in virtue of being entitled. Of course, in many contexts, Sally should qualify this assertion or weaken its content unless she improves her epistemic position. All I  claim is that in some contexts an adequate degree of entitlement may meet the warrant-demand. However, in many, perhaps most, conversational contexts the asserter is reasonably expected to be able to articulate some reasons in favor of the content of her assertion.13 I call such conversational contexts ‘discursive contexts.’ In discursive contexts, the speaker must possess an internalist kind of warrant, ‘discursive justification’ that   The list is not exhaustive. But since this is not the place for theory-building, I’ll work with what I’ve got.   Because there are a number of context-determiners, the relationship between stakes and the epistemic position required for assertion (and action) is not a simple linear one. For example, the conversational context may also be configured such that high stakes lower the warrant-demand on assertion. Among the candidate cases is one in which the relevant costs of improving one’s epistemic position are higher than the relevant costs of asserting something false (see Gerken 2015b, 2017). 13  I take the notion of reasonability here to be set by the standard of a reasonable interlocutor (cf. Lawlor 2012, Ch. 5). There are similar norms of assertion around. McKinnon sets forth a ‘supporting reasons’ norm according to which the speaker should always be in a position to provide contextually adequate reasons (McKinnon 2015). I regard this requirement as too strong. One may meet the epistemic norm of assertion by possessing mere entitlement that one cannot articulate as reasons. Grice’s formulation is ‘Do not say that for which you lack adequate evidence’ (Grice 1989). I take this to be too evidentialist. These are in-house disputes insofar as Grice’s and McKinnon’s norms involve gradable, non-factive epistemic properties (pace Benton who argues that Grice’s norm—despite its official formulation—is a knowledge norm (Benton 2016)). 11

12

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

56  Mikkel Gerken requires that the agent can articulate her epistemic grounds as reasons (Gerken 2012a, 2015c; See also Leite  2005). Thus, I articulate a species of WASA, the Discursive Justification-Assertion account, DJA, as follows: DJA In the discursive conversational context, DCC, in which S’s assertion that p conveys that p, S meets the epistemic conditions on appropriate assertion that p (if and) only if S’s assertion is appropriately based on a degree of discursive justification for believing that p that is adequate relative to DCC. I find that the epistemic norm WASA and its internalist species DJA provide a richer account of a wide range of linguistic phenomena than norms that appeal to knowledge and nothing else. This is, in part, because of the pluralism about warrant. On this note, let me move from presentation to argumentation.

4.2  Challenges to the knowledge norm The knowledge norm is in dispute in part because there are prima facie counterexamples to both directions of the biconditional, KNAS. I will discuss some of such candidate counterexamples (some new, some old). This is not the place to defend any one of them in detail. But I hope to say enough to consider their methodological ramifications for a non-reductive knowledge-first epistemology. Specifically, I will argue that the apparent simplicity of the knowledge-first approach to epistemic norms is deceptive. The knowledge-first approach requires an increasingly complex apparatus of auxiliary assumptions and maneuvers that render it no more tidy than the tradition it seeks to replace. The case above in which Sally forms the belief that Jane is at the conference but forgets its source (a glimpse) provides a methodologically interesting example. A core assumption in externalist epistemology is that Sally may know that Jane is at the conference although she has no access to and, hence, cannot articulate any reasons for asserting so. Of course, this assumption is compatible with the idea that knowledge that p is a necessary condition on epistemically appropriate assertion that p. But, in certain discursive conversational contexts, Sally may not be in an epistemic position to assert that Jane is at the conference. Insofar as Sally’s lack of access to reasons is an ­epistemic deficit, the case may form a novel type of counterexample to the sufficiency claim of KNAS (see Gerken 2012a, 2013b, 2015c for an argument that the ability to access and articulate reasons is epistemically significant). I will address the prospects for knowledge-first responses to the challenge from discursive conversational contexts below. But for the sake of comparison, note that  this issue is straightforwardly addressed by the distinction between warrant generally (including entitlement) and discursive justification and the corresponding distinction between WASA and DJA. Furthermore, different species of knowledge may be characterized by reference to the entitlement-justification distinction and

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  57 I think that they should be.14 In contrast, someone who thinks that knowledge must be explanatorily first will have a hard time distinguishing between species of knowledge by reference to difference in kinds of warrant. So, a knowledge-first proponent must either reject that there are different species of knowledge or find another way of individuating between them.15 Likewise, an account of the case at hand will be not be straightforward if all that we can appeal to in assessing assertion epistemically is knowledge simpliciter. The present approach may also account for much of the linguistic data that knowledge-first proponents have themselves appealed to. There is an important methodological point to be drawn from this. The knowledge norm is often motivated by the fact that it can account for the prominence of ‘knowledge’ in everyday epistemic assessment of assertion (Williamson 2000, Hawthorne and Stanley 2008). It is natural to say ‘you shouldn’t have asserted that p since you didn’t know that p.’ Elsewhere, I have addressed such linguistic data in the context of the epistemic norms of action (Gerken 2011, 2015a, 2017). Transposed to the case of assertion, I’ve argued for the following idea: Normal Coincidence* In normal cases of epistemic assessment, the degree of warrant necessary for S’s  knowing that p is frequently necessary and very frequently sufficient for the ­epistemic permissibility of S’s asserting that p. Given this principle, it is reasonable to suppose that ‘knows’ serves us well as ‘communicative heuristic‘ to covey complex epistemic information in a simple and reasonably accurate manner. In consequence, KNAS may be seen as a crude but useful folk epistemological principle. It tends to be reasonably accurate in normal cases of epistemic assessment. But sometimes it is inaccurate and this is what the counterexamples to it indicate (Gerken 2015a, Sect. 5C. I elaborate in Gerken 2017 Chapters 6 and 7). The methodologically important point is that both directions of KNAS may well be overgeneralizations from paradigm cases. In many such cases, someone who knows that p also meets the epistemic conditions on assertion that p. In many other cases, someone who does not know that p does not meet the epistemic conditions on assertion that p. Despite the importance of paradigm cases, the fact that there are more complex cases in which knowledge and the epistemic conditions come apart, puts pressure on knowledge-first methodology. Specifically, the notion we should start and end with paradigm cases because we have a better grasp of them seems problematic if there are significant exceptions from such cases. And, as opponents of knowledge norms have pointed out, there is no shortage of exceptions. 14   Sosa 2007 and Graham 2012 provide similar distinctions between kinds of knowledge. I elaborate my own epistemic pluralism in Gerken 2013a, 2013b and forthcoming a. 15   Note that individuating kinds of knowledge by reference to its source—e.g., knowledge by perception, testimony, memory and so forth—will not do if there are internalist and externalist species of each kind (Gerken 2013b).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

58  Mikkel Gerken For instance, I may not be in a position to assert that Ortcutt is a spy, even though I know that this is the case if my evidence is not strong enough to bring Ortcutt to court, and asserting so will give Ortcutt a chance to escape (for elaboration, see the case SPY in Gerken 2014: 732–3). Such exceptions to the sufficiency direction of KNAS may be most forceful in first-person past tense knowledge ascriptions. Consider, for example, Brown’s case, AFFAIR, in which S knows that S*’s wife is cheating but withholds asserting so until he has more conclusive evidence (for elaboration, see Brown 2008). Here saying ‘I knew but I was not in a position to assert so until I had stronger evidence’ is both natural and appropriate. As mentioned, the case of externalist (entitlementbased) knowledge in discursive contexts may be another problematic case. It may be a case in which Sally knows that p although she is not in an epistemic position to assert that p outright due to her lack of reasons to back up her assertion. It is important to acknowledge that there is an assortment of moves available to knowledge-first theorists. It may be argued on internalist grounds that the speaker does not know. Another popular response is to embrace pragmatic encroachment and argue that the speaker no longer knows given that the stakes have risen. It may also be argued that the speaker in fact meets the epistemic norm although she violates some other non-epistemic norm. Furthermore, it can be argued that the speaker only knows that it is highly probable that, for example, the wife is cheating. Likewise, it may be argued that the cases rest on confusing having a reason with having a sufficient reason. Moreover, it may be argued that the relevant speech act is not an assertion and, therefore, not subject to the knowledge norm but to another more demanding epistemic norm.16 The space for such maneuvering is very wide in philosophy. Indeed, this is one ­reason why principles that are central to a philosophical program are rarely refuted by way of counterexample alone. However, each of these responses is associated with considerable costs. Consider, the claim that the additional requirements on assertion have nothing to do with its epistemic norm but with an extra-epistemic norm. This response leaves a good deal to explain. Why can the subject meet those additional requirements by improving her epistemic position? Why is it apt to criticize someone who asserts on the basis of knowledge in epistemic terms? In the relevant cases, it can be reasonable to complain: ‘you shouldn’t have asserted that since you didn’t have any reasons/strong enough evidence for believing it.’ It seems ad hoc to regard reasonable complaints in terms of knowledge as strong evidence for KNAS and then dismiss similar complaints that seem to compromise it. 16   For ideas that are broadly congenial to the listed responses, see Hawthorne and Stanley 2008 for the pragmatic encroachment response, Williamson 2000 for the non-epistemic norm response (thanks also to a referee), Ichikawa 2012 for the sufficient reason response, Hawthorne and Stanley 2008 for the knowledge of probability response, and Turri  2010,  2011 for the different speech act response according to which the conversational context to bear on the type of speech act rather than, as contextualists, would have it, the content. I take the central problem with this view to be one that Turri attributes to a referee: ‘the scale of speech acts might not be as fine-grained as the scale of epistemic standards’ (Turri 2010, fn. 24). I address some of these objections in (Gerken 2017).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  59 Of course, I do not want to simply dismiss any possible responses—or some combination thereof—prior to a proper development of them. However, the onus is on knowledge-first theorists to develop these responses at a level of detail that admits of criticism.17 It is unsatisfactory to just point to a response strategy and assume that it can handle all troublesome cases. Moreover, a combination of responses may well be required to address the variety of counterexamples indicated above. If so, the sufficiency of knowledge for assertion requires a complex of interacting auxiliary assumptions. We find a similar dynamic in the candidate counterexamples to the necessity of knowledge for assertion. Such cases include cases of extremely well-warranted belief (perhaps accompanied by low stakes, lack of further available evidence, and no opportunity for alternative assertions) and Gettier-style cases (Gettier 1963; Gerken 2011). Again, there is a wide range of moves available to the knowledge-first epistemologists. For example, they may invoke an excuse maneuver according to which the asserter is not meeting the epistemic norm of assertion but rather excused from violating it (Williamson 2000, forthcoming; DeRose 2002; Littlejohn 2012). Or they may argue that the speaker does have knowledge but only of the proposition that the content of the assertion is probable (Hawthorne and Stanley 2008). Or they may reject that the speech acts in question are assertions (Turri 2010, 2011). Again, the range of such maneuvers is very wide. I have argued against some of these moves (Gerken 2011a, 2012a, 2014, 2015a, 2017). Since space does not permit for detailed argumentation here, I will make a couple of general methodological points. The first point is that the responses to the cases of candidate counterexamples to the knowledge norm of assertion are controversial and may seem ad hoc. Consequently, the dynamic begins to look somewhat familiar in some respects. The knowledge norms may be defended against counterexamples by revising assumptions about knowledge (or ‘knowledge’) or by invoking auxiliary assumptions. The latter may concern excuses, knowledge of probability, the speech act of assertion and so forth. However, the auxiliary assumptions that knowledge-first epistemologists have set forth have been argued to be problematic (McGlynn  2014). For example, I have argued that the appeal to excuses is either a deus ex machina or subject to substantive problems or collapsing into a warrant account (Gerken 2011). Of course, knowledge-first proponents may respond by specifying the operative notion of an excuse accordingly. To briefly consider a more specific instance of this dynamic, consider my objection that the excuse maneuver is at the risk of collapsing into a warrant account. If one is excused from asserting that p without knowing that p in virtue of being very well warranted in believing that p, then the speaker’s degree of warrant and the fact that the belief is false appears to do the explanatory job (Gerken 2011). One response might consist in qualifying the account of excuses such that one cannot be excused from 17   It is only fair to acknowledge that those who uphold a combination between knowledge norms and pragmatic encroachment have done a reasonable job on this account. I provide some criticism in Gerken 2011, 2015a, and, especially, in 2017.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

60  Mikkel Gerken asserting falsely in virtue of being very well warranted. I regard this further qualification as implausible and in need of independent motivation. Indeed, it strikes me as coming fairly close to gerrymandering the notion of excuse to salvage the knowledge norm. Of course, knowledge-first proponents see things otherwise. But the methodological point is that the alleged advantage of knowledge-first epistemology—namely, that it is simpler than the alternative—appears to have all but evaporated. The initially straightforward knowledge norms that accounted well for paradigm cases appear to require an increasingly complex apparatus of, for example, excuses. So, although the knowledge norms themselves may be preserved in their original form, a comprehensive account of all the cases requires very considerable auxiliary assumptions. The dialectic is subtle because the response rarely consists in fiddling with the knowledge norm itself. So, the contrast is that between a simple knowledge norm which can be preserved by adopting a complex of auxiliary assumptions and a more complex epistemic norm, such as WASA, that can deal with a wider set of cases— including the counterexamples to KNAS—without auxiliaries.18 Something similar may be said of the responses that consist in revising basic assumptions about knowledge itself. For example, some (but not all)19 pragmatic encroachers will respond to Brown’s AFFAIR case by claiming that since the stakes are high, the speaker no longer knows. Similarly, some (but not all) contextualists will claim that the utterance of ‘S knows that p’ is no longer true. Of course, there is great debate about whether this response marks a refreshing insight or severe problem. I take it to be a problem (Gerken 2015b; Anderson 2015). So, does Williamson, who develops a response according to which higher-order knowledge is required as the stakes go up (Williamson 2005, but see Gao, forthcoming for criticism). But here I will make a different point. As I have indicated, the myopic focus on stakes as determining the epistemic position one must be in for one to be in a position to assert that p or act on (the belief that) p is unfortunate. A wide range of factors are determinants of the relevant epistemic position. This is important in its own right. (For one thing, it speaks against the idea that raising stakes invariably raise the ­epistemic requirements on action or assertion—see Gerken 2017). But it yields a ­particular challenge to proponents of knowledge norms. Assume, for the sake of argument, that the epistemic position required for assertion/action may be altered by variation in things such as availability of evidence, urgency, social roles, alternative assertions/courses of action, etc. If the assumption that urgency, for example, may determine whether someone knows is too much of a bullet to bite even for pragmatic encroachers, then a different response is required to cases that are driven by variations in urgency rather than in stakes. While I have not encountered a response to this objection (first noted in Gerken 2011, Sect. 6 and 7 and elaborated in 2017), a

18   Thanks to Hawthorne, Williamson, Yli-Vakkuri, and a referee for helping me better see this aspect of the dialectic clearer. 19   Note that the stakes pertain to the cuckold husband, not the speaker.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  61 response may, of course, be developed. However, the risk of gerrymandering in the face of this objection should be obvious. A word of qualification: I do not mean to deny that progress may come from developing responses to objections to knowledge norms.20 My main point is that the responses that knowledge-first proponents invoke to address troublesome cases are by no means free of the risk of gerrymandering and ad hoc–ness (see also McGlynn 2014, Ch. 8). At present, the debates over the objections and responses to knowledge norm bear resemblance to clashes of paradigms in sciences. Whereas knowledge-first epistemologists regard the auxiliary assumptions as natural aspects of their account, opponents regard them as forlorn attempts to salvage the knowledge norms from obvious counterexamples. However, many of the challenges to the knowledge norms are so novel that responses to the challenges are still under development. Indeed, some challenges remain unaddressed. So, we do not yet have a history of increasingly complex auxiliary assumptions. Perhaps we will have to witness fifty years of fiddling with, for example, the excuse maneuver before the picture is clear enough for an assessment.

4.3  Non-reductive knowledge-first epistemology in conclusion The preceding reflections suggest a more prescriptive methodological point: We should not only work with schematic knowledge ascriptions in the abstract or with paradigm cases of knowledge and assertion. Rather, we should consider a wide variety of cases that exemplify various configurations of parameters of the conversational context. Doing so may reveal that some motivations of knowledge norms are overly programmatic overgeneralizations. For example, considering a wider set of cases reveals some of the problems with the step from the reasonable assumption that ‘knows’ is prominent in assessment of assertion to the conclusion that knowledge is the norm of assertion (Gerken 2015a, 2017). I favor WASA and its species, DJA, to knowledge norms in part because I take them to provide more accurate explanatory accounts of the sort of phenomena that knowledgefirst epistemologists have invoked. Reflection on the counterexamples suggest that our use of ‘knows’ in criticizing assertion works a communicative heuristic—i.e., as cognitively cheap shorthand uses that work well for us because they normally align with the contextually variable warrant-demand on assertion.21 Thus, the prominent roles of ‘knowledge’ in our folk epistemological talk and thought are perfectly compatible with alternatives to knowledge norms. (Or so I have argued in the case of action, cf.  Gerken  2011,  2015a). As suggested above, these arguments transpose mutatis mutandis to the case of assertion. 20   This cuts both ways, however. Gettier did teach us something important about knowledge, justification and their relationship. Moreover, the post-Gettier tradition has not been altogether futile. Although it has admittedly produced some epistemological garbage, it has also produced important insights. Epistemic externalism is among them. 21   For the same reason, varieties WASA and DJA are apt to contribute to a strict purist invariantist account of cases that are supposed to motive pragmatic encroachment about knowledge or ‘knowledge.’ Doing so is a key ambition of Gerken 2017.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

62  Mikkel Gerken Let me conclude with a basic but important point. Opposition to the knowledgefirst agenda or specific alternative principles, such as WASA, does not yield any commitment to any ‘experience-first’ epistemology (Dougherty and Rysiew  2013a,  b). Such an epistemology is, as Williamson argues, hard to distinguish from a traditional ‘appearance-first’ epistemology, which he rightly criticizes (Williamson  2013b). However, the staging of this debate is unfortunate. Williamson comes close to suggesting that knowledge-first epistemology aligns with externalist epistemology in general. Dougherty and Rysiew respond by adopting a radically internalist ‘experience-first’ alternative. What I find unfortunate is that this juxtaposition suggests (although it does not entail) a false dilemma between two extremes: knowledge-first and experience-first. This alleged choice is a dilemma because both options are unattractive. It is a false dilemma because there are viable alternatives. One such alternative is a pluralistic view that includes both epistemic externalist and epistemically internalist species of warrant (Gerken 2013b, forthcoming a). Likewise, a broader methodological alternative is to abandon the idea that any single epistemic phenomenon is ‘first.’ I will not address the internalism/externalism dispute here (but see Gerken 2013b, forthcoming a). But it is worth reemphasizing that WASA allows that one meets the epistemic requirements on assertion by externalist warrant (entitlement) alone. Such warrant need not be based on appearances, experiences, reasons, or any such. So, one may reject knowledge-first epistemology and accept the most prominent brands of externalism (see, e.g., Goldman  1979, Burge  2003, Sosa  2007, and Graham  2012). Although this is a basic point, it is apt to illustrate why the above-mentioned dilemma is a false dilemma. My broader aim has been to question whether specific non-reductive knowledgefirst epistemological theses are plausible and robust enough to constitute a knowledgefirst program. But neither specific theses nor a knowledge-first agenda may be refuted by way of counterexamples. However, I have argued that such counterexamples generate an objection-response dynamic that involves a complex apparatus of auxiliary assumptions that may become increasingly ad hoc or gerrymandered. So, although the challenges to non-reductive knowledge-first epistemology in the guise of knowledge norms of assertion and action are non-conclusive, I hope that they will encourage some reflection among knowledge-first epistemologists. Along the way, I have indicated serious competitors to the substantive non-reductive knowledge-first theses. In the next section, I extend this point to the methodological level and argue that there is a less radical alternative to knowledge-first epistemology.

5.  Towards a ‘Nothing’s First’ Alternative: Equilibristic Epistemology Let me briskly sketch what strikes me as a more promising methodological way forward. I will not be able to develop this methodological alternative in satisfying detail (although I say more in Gerken 2017, Ch. 3). But perhaps I can convey two broad points.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  63 First, I hope that the sketch will speak to various suggestions that an alternative to a knowledge-first epistemology involves a return to a ‘degenerating research program’ of providing ‘sterile and inward-looking’ reductive analyses of knowledge that are ‘small-minded and old-fashioned’ (Williamson 2013a: 2).22 Second, by sketching a methodological alternative, I can indicate what I take to be a grain of truth in non-reductive knowledge-first epistemology. For the alternative is compatible with the assumption that reflective judgments about knowledge may help us understand other epistemic phenomena and, crucially, vice versa. Indeed, my positive message is that a good deal can be learned by isolating grains of truth to be found in non-reductive knowledge-first epistemology.

5.1  Equilibristic epistemology I label my alternative methodology ‘equilibristic epistemology’ because it pursues a ‘wide reflective equilibrium’ between folk epistemological judgments and epistemological principles, as well as interrelations between different epistemic phenomena.23 One of its central claims is that, among the most basic epistemic and doxastic phenomena, no one phenomenon is sufficiently explanatorily primary to serve as a primary explainer of the others. Thus, I do not advocate a warrant-first, rationality-first, or belief-first epistemology. Nothing is first. Another aspect of equilibristic epistemology is that the tacit principles of our folk epistemology may diverge from the most plausible epistemological principles. Thus, critical reflection on intuitive judgments about cases may inform epistemological theorizing and vice versa. Elsewhere, I outline principles of this latter aspect of the equilibristic epistemology (Gerken 2017, Chapter  3). Here I focus on its former aspect—the one that pertains to explanatory primacy and lack thereof. It can be articulated in the form of a broad principle: Principle of Non-Primacy The most fundamental epistemic and doxastic phenomena, concepts and words are non-reducible and no one of these phenomena, concepts or words can serve as the primary explainer of the others. The Principle of Non-Primacy is a negative one. But a more positive corollary of it is that fundamental phenomena (concepts, words) must be co-investigated. By considering fundamental epistemic and doxastic phenomena in relation to each other, important constitutive connections may be discovered. Some aspects of this viewpoint are inspired by Strawson’s non-reductive methodology (Strawson 1992). However, I part 22   The quoted phrases are from (Williamson 2013a). But it is unclear whether they can be attributed to Williamson who takes them to exemplify “analytic philosophy’s reputation” or attributes them to “analytic philosophers” (Williamson 2013a: 2). 23   For recent discussions of reflective equilibrium, see (Kelly and McGrath  2010, Walden  2013, and Cath 2016).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

64  Mikkel Gerken ways with Strawson at important crossroads.24 The broad idea that is congenial to Strawson is, roughly put, that we should abandon reductive analysis in favor of nonreductive elucidation which may be circular (Strawson 1992, Ch. 2). In epistemology, this means that we should abandon attempts to reductively analyse knowledge in terms of belief, rationality and so forth. But it also means that we should not simply reverse this strategy and seek to reductively analyse epistemic rationality, evidence or belief in terms of knowledge. Rather, we should consider epistemic and doxastic phenomena such as knowledge, belief, normal circumstances, warrant, epistemic luck, evidence, and so forth in relation to each other. Identifying constitutive relations between such phenomena does not require that any of these phenomena may be non-circularly reduced to other allegedly primary phenomena. Thus, counterproductive constraints of reductionist epistemology may be shed. The equilibristic methodology departs radically from Strawson’s insofar as it is not focused on conceptual analysis or on linguistic analysis of our ordinary epistemic vocabulary. Strawson’s approach inherits elements of the ‘linguistic turn’ that I want to distance myself from. The primary objects of study are not primarily linguistic or ­conceptual but the epistemic subject matter. So, with regard to the explanandum, the equilibristic epistemology aligns with Williamson’s methodology, rather than with Strawson’s (Williamson 2007). Epistemological questions should often be taken at face value. The primary object of study for the theory of knowledge is knowledge. It is not the word ‘knowledge’ or the concept knowledge. Oddly enough, however, proponents of knowledge norms of action and assertion sometimes rely heavily on ordinary language phenomena such as the claim that it is natural to use ‘knows’ in epistemic assessment (Williamson 2000; Hawthorne and Stanley 2008; see Gerken 2015a, 2017 for criticism). Likewise, the step from the purported ontogenetic primacy of the concept knowledge to the idea that knowledge is first in epistemology would be less implausible if epistemology was primarily concerned with conceptual analysis.25 My approach differs from the Strawsonian methodology by upholding a firmer distinction between the context of discovery and the context of justification in ­epistemological theorizing.26 Due to Strawson’s focus on conceptual analysis he emphasizes that “. . . the acquisition of the theoretical concepts of the special disciplines presupposes and rests upon the possession of the pre-theoretical concepts of ordinary life” (Strawson 1992: 21). Strawson’s claim is a claim about the genesis of theoretical concepts. As such, it has some plausibility.27 However, once we shift the focus from concepts to their referents, 24   Strawson’s book is from 1992 but represents lectures from 1968–1987. The relevant discussion is ‘­virtually unchanged’ from 1968 (Strawson 1992: vii). 25   Even the step from conceptual ontogenesis to conceptual composition is very controversial. 26   For Reichenbach’s famous distinction and recent work on it, see (Reichenbach 1938 and Schickore and Steinle 2006). 27   Given some qualifications that I will not go into here.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  65 it ceases to be plausible that the ontogenetic earliness of concepts corresponds to the explanatory primacy at the level of extension. Thus, I am skeptical about the idea that the development of our concepts must correspond to a hierarchy of ontological or explanatory basicness at the level of the extension. We acquire the concept water before we acquire the concepts hydrogen and oxygen. But at the level of the extension, water is not more basic than hydrogen and oxygen. On the contrary, if there is an asymmetric basicness relation, it is the opposite of the order of conceptual acquisition. This is so even if we could not, as a practical matter, come to acquire the concepts hydrogen and oxygen without first possessing the concept water. Similarly, even if we must acquire the concept knowledge before, for example, the concept belief, this would not provide evidence that knowledge itself is more explanatorily basic than belief itself.28 In epistemology, we should also be open to the idea that some of the most explanatorily basic epistemic phenomena may only be identified at later stages of inquiry. For example, the phenomena that I denote by the label ‘normal epistemic circumstances’ may be explanatorily basic (Gerken  2013a). But in the genesis of explanation, it is invoked after knowledge and rationality are invoked. Perhaps it is only psychologically possible to identify it via reflection on judgments about knowledge and epistemic rationality. Likewise, we should even be open to the idea that properties which may only be recognized through overt theorizing may turn out to be explanatorily basic. The modal property of safety may be a candidate.29 Even if we could only recognize and conceptualize this modal property via judgments about knowledge, we may, once it is recognized, work with it in relation to other phenomena such as the metaphysics of modality. In the best case, we may, thereby, improve our understanding of safety. Perhaps, we may even be able to put an improved understanding of safety to use in refining our previous understanding of  knowledge. Whether or not safety is a good candidate of a basic explanatory ­phenomenon is a disputed matter. But this dispute is hardly methodologically misguided in virtue of the fact that safety was not properly recognized until 1970 (Dretske 1970, 1971). As mentioned, the assumption that the word ‘knowledge’ (or the concept knowledge) is ontogenetically prior to, for example, the word ‘belief ’ (or the concept belief) is itself controversial (for this controversy, see Nagel 2013 vs. Aidan McGlynn’s chapter in this volume). But even if true, the assumption should not automatically be taken as a r­ eason to reflect an explanatory primacy relation in epistemology. If knowledge is ontogenetically ‘first’ in our conceptual development, this may be a reason to think that it is first 28   This line of reasoning extends from concepts to linguistic terms. So, likewise, I am skeptical about the idea that all of the most explanatorily basic epistemic phenomena correspond to vernacular terms (see Gerken 2015b). 29   I mention safety as a candidate without any commitment. This is due to the fact that a number of knowledge-first epistemologists are theorizing about safety in a manner that appears to be compatible with the equilibristic modus operandi that I am sketching (Williamson 2000).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

66  Mikkel Gerken in the context of discovery. But even if knowledge is first in the context of discovery, this does not indicate that it is explanatorily first in the context of justification (pun intended). The equilibristic epistemology does not entail or even suggest that there are no asymmetric explanatory basicness relations to be found. Cassam (2009) characterizes explanatory basicness as follows: ‘A concept C is more explanatorily basic than another concept D if and only if C can be explained without using D but D cannot be explained without using C’. As mentioned, the aim of equilibristic epistemology is not primarily conceptual analysis, but substantive explanation. However, Cassam’s idea of explanatory basicness may be transposed from the conceptual to the substantive realm. Indeed, doing so provides some specification of a central claim of equilibristic epistemology that is among the most basic epistemic and doxastic phenomena, no one is so significantly more explanatorily basic than the others that may be said to be a primary explainer. This claim is compatible with Cassam’s idea that there are asymmetric explanatory relations between epistemic concepts, words, and phenomena. But acknowledging this much does not entail that any one phenomenon is ‘first’ in the sense of being significantly more explanatorily basic than other basic phenomena. This assumption also tells against a quantitative account of explanatory primacy based on the claim that knowledge figures prominently in the explanans of more epistemological explanations than any other epistemic or doxastic phenomenon.30 Such an approach faces multiple challenges. First and foremost, the quantitative approach would at best characterize a rather thin knowledge-foremost program rather than a principled knowledge-first program. This point is a variation of the initial point that to say that knowledge is ubiquitous, central, and prominent in our folk epistemology does not distinguish an epistemological program. Second, it is a controversial claim that knowledge figures prominently in the explanans in more successful explanations than any other notion. As I have argued in the case of epistemic norms, the knowledge-first approaches may lack in explanatory power what they have in simplicity. Finally, there appear to be areas of epistemology in which a fruitful investigation may be conducted without centrally invoking knowledge. For example, it can be both feasible and fruitful to consider varieties and properties of reliability in their own right. Likewise, a principled argument for explanatory primacy does not follow from the fact that the knowledge-first approach has inspired fruitful work in new directions.31 I believe that it has. But the assumption that starting with judgments about knowledge and considering their consequences is fruitful would at best show that this is a fruitful modus operandi for one part of the epistemological project. Unless it is argued that this is the most basic part, there is no sense in which knowledge is argued to be explanatorily primary. But much equally foundational work in epistemology   This idea was non-committally suggested to me by John Hawthorne.   This idea was non-committally suggested to me by Lizzie Fricker.

30 31

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  67 barely mentions knowledge.32 Moreover, it seems to me that much of the interesting work on, for example, safety or luminosity carried out by self-declared knowledgefirst theorists could, in principle, have been carried out without commitment to knowledge-first epistemology. Finally, to note that Williamson is an extraordinarily ingenious and resourceful epistemologist who benefits from starting with knowledge is not to provide a principled characterization of a knowledge-first ­program.33 In general, then, thin methodological claims to the effect that we should often begin with judgments about knowledge seem too thin to constitute a distinctive knowledgefirst epistemology. However, Williamson and the best of his followers should be credited for arguing that knowledge is as explanatorily basic as any epistemological phenomenon.34 Given this assumption, reflection on knowledge and judgments about cases of knowledge may be used constructively in theoretical elucidations of other epistemic phenomena such as belief, epistemic luck, etc. This, I believe, is a grain of truth in knowledge-first epistemology that may be well worth preserving. However, the equilibristic epistemology that I have sketched is able to preserve this grain of truth. To sum up, the positive case for knowledge-first epistemology hinges to some extent on a misguided assumption that rough and ready ordinary language phrases that serve our everyday communication about matters epistemic may inform epistemology in a fairly straightforward manner. But while our folk epistemological practices should inform epistemology, they should not inform it in any straightforward manner. As epistemologists we should critically assess our folk epistemology (Gerken 2017). The negative case for knowledge-first epistemology, in turn, has tended to target radical internalist alternatives or to presuppose that the opposition is committed to pursuing reductionist accounts of knowledge. I have tried to convey that there are better alternatives available and sketched some of them. I hope to someday say more about equilibristic methodology for philosophy more generally.35 But here the broad label has mainly served as a broad contrast in my assessment of the knowledge-first agenda.

Conclusion I have argued that both reductive and non-reductive knowledge-first approaches are problematic. However, reflection on the alternative that I—in lieu of a better label— call ‘equilibristic epistemology’ suggests that a less radical methodology is available to epistemologists.   I’m afraid that I have already used my quota of self-citations.   I mention this because several self-identifying knowledge-first epistemologists have suggested—only half in jest—that knowledge-first epistemology is the conjunction of Williamsonian views in epistemology. 34   Of course, this not quite what they argue. But just as Columbus was mistaken about the continent that he had encountered, his encounter was nevertheless an important achievement. 35   I say a good deal more in (Gerken 2017). 32 33

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

68  Mikkel Gerken Although my general assessment of knowledge-first epistemology is negative, it is important to acknowledge that the movement has offered an interesting perspective shift. Even if knowledge (or ‘knowledge’ or knowledge) is in no distinctive way ‘first’, such a perspective shift may provide an illuminating perspective on core epistemological phenomena.36 For example, knowledge norms of assertion may reflect important aspects of our folk epistemology. However, our folk epistemology is important to understand, even though it diverges systematically from the most plausible epistemological principles.37 Moreover, the knowledge-first agenda has been an antidote to the at times myopic focus on providing a reductive analysis of knowledge. However, recognizing these contributions is a far cry from providing substantive grounds for adopting a substantive knowledge-first program. The fact that other epistemic phenomena can be illuminated by reflection on knowledge is perfectly compatible with the fact that knowledge may be illuminated by reflection on other epistemic phenomena. Opposition to knowledge-first epistemology is committed to uphold this much—but it is not committed to a reductive analysis of knowledge. So, if the negative motivation for knowledge-first epistemology hinges on such a  charge, it is rather feeble. There are non-reductionist alternatives—including ­epistemically externalist ones—and the equilibristic epistemology sketched here is among them. Nevertheless, there are important grains of truths in the contemporary knowledgefirst theory. My concern is that an overly radicalized and overly programmatic knowledge-first agenda distracts from those truths. For example, it may be a grain of truth in knowledge-first epistemology that knowledge is basic and unanalysable. The ironic mistake of the most radical strands of the knowledge-first program is that of reversing, rather than abandoning, the reductionist ambitions of the tradition it criticizes. This reaction strikes me as an overreaction. The alternative I have suggested consists in taking knowledge, belief, rationality, normal circumstances, etc., to be explanatorily basic, non-reducible phenomena. According to an equilibristic methodology, we should explore constitutive relations between these epistemic and doxastic phenomena. In this manner, knowledge may illuminate other epistemological and doxastic phenomena—and vice versa.38

36   Compare with Thomson’s famous perspective shift from the rights of the fetus to the rights of a pregnant woman in the debates about the ethics of abortion (Thomson 1971). Whether one ultimately agrees with Thomson’s argument or not, the shift in perspective marks an important intellectual contribution. There are many disanalogies between these cases, of course, and I am not suggesting that they are on a par in all respects. 37   This is one of the key theses of (Gerken 2017) in which I provide a number of examples of divergences between folk epistemology and epistemology proper. 38   For written comments, I am grateful to the trio of editors, an anonymous OUP referee who provided a helpful knowledge-first perspective, Emil Møller, and Bob Beddor. The paper was presented at the University of Geneva, Yonsei University, and Oxford University (all in the first half of 2014) and I benefited from the discussions on these occasions (and yet more discussions after another talk in Oxford’s Philosophical Society in 2015). Finally, I am grateful to Timothy Williamson for clarifying some exegetical issues and for several discussions over the years. The chapter is dedicated to Teo.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  69

References Anderson, Charity. 2015. Knowledge and Reasons for Action. Episteme 12(3): 343–53. Benton, Matthew. 2014. Knowledge Norms. Internet Encyclopedia of Philosophy. Retrieved from http://www.iep.utm.edu/kn-norms/. Benton, Matthew. 2016. Gricean Quality. Noûs 50: 167-89. Brown, Jessica. 2008. Subject-Sensitive Invariantism and the Knowledge Norm for Practical Reasoning. Noûs 42(2): 167–89. Brown, Jessica. 2010. Knowledge and Assertion. Philosophy and Phenomenological Research 81(3): 549–66. Burge, Tyler. 1996. Our entitlement to self-knowledge. Proceedings of the Aristotelian Society 96: 91–116. Burge, Tyler. 2003. Perceptual Entitlement. Philosophy and Phenomenological Research 67: 503–48. Cassam, Quassim. 2009. Can the concept of knowledge be analyzed? In Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 12–30. Cath, Yuri. 2016. Reflective Equilibrium. In H. Cappelen, T. Gendler, and J. Hawthorne, eds., Oxford Handbook of Philosophical Methodology. Oxford: Oxford University Press, 213–30. DeRose, Keith. 2002. Assertion, knowledge, and context. Philosophical Review 111(2): 167–203. Dougherty, Trent and Patrick Rysiew. 2013a. What Is Knowledge-first Epistemology? In M. Steup and J. Turri, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Blackwell, 10–16. Dougherty, Trent and Patrick Rysiew. 2013b. Experience First. In M. Steup and J. Turri, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Blackwell, 17–22. Douven, Igor. 2006. Assertion, Knowledge and Rational Credibility. Philosophical Review 115(4): 449–85. Dretske, F. 1971. Conclusive Reasons. Australian Journal of Philosophy 49: 1–22. Dretske, Fred. 1970. Epistemic Operators. Journal of Philosophy 67: 1007–23. Fantl, Jeremy and Matthew McGrath. 2002. Evidence, pragmatics, and justification. Philosophical Review, 111: 67–94. Fantl, Jeremy and Matthew McGrath. 2009. Knowledge in an Uncertain World. Oxford: Oxford University Press. Fantl, Jeremy and Matthew McGrath. 2012. Arguing for shifty epistemology. In Jessica Brown and Mikkel Gerken, eds., Knowledge Ascriptions. Oxford: Oxford University Press, 55–74. Gao, Jie. forthcoming. Against the Iterated Knowledge Account of High-stakes Cases. Episteme. Gerken, Mikkel. 2011. Warrant and Action. Synthese 178(3): 529–47. Gerken, Mikkel. 2012a. Discursive Justification and Skepticism. Synthese 189(2): 373–94. Gerken, Mikkel. 2012b. On the Cognitive Bases of Knowledge Ascriptions. In Jessica Brown and Mikkel Gerken, eds., Knowledge Ascriptions, Oxford: Oxford University Press, 140–70. Gerken, Mikkel. 2013a. Epistemic Reasoning and the Mental. London: Palgrave Macmillan. Gerken, Mikkel. 2013b. Internalism and Externalism in the Epistemology of Testimony. Philosophy and Phenomenological Research 87(3): 532–57. Gerken, Mikkel. 2014. Same, same but different: The epistemic norms of assertion, action and practical reasoning. Philosophical Studies 168(3): 745–67. Gerken, Mikkel. 2015a. The Roles of Knowledge Ascriptions in Epistemic Assessment. European Journal of Philosophy 23(1): 141–61. Gerken, Mikkel. 2015b. How to do things with knowledge ascriptions. Philosophy and Phenomenological Research 90(1): 223–34.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

70  Mikkel Gerken Gerken, Mikkel. 2015c. The Epistemic Norms of Intra-Scientific Testimony. Philosophy of the Social Sciences 45(6): 568–95. Gerken, Mikkel. 2017. On Folk Epistemology. How we Think and Talk about Knowledge. Oxford: Oxford University Press. Gerken, Mikkel. Forthcoming a. Entitlement—its scope and limits. In P. Graham and N. J. L. L. Pedersen, eds., Epistemic Entitlement. Oxford: Oxford University Press. Gerken, Mikkel. Forthcoming b. The New Evil Demon and the Devil in the Details. In V. Mitova, ed., The Factive Turn in Epistemology. Cambridge: Cambridge University Press. Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23(6): 121–3. Goldman, Alvin. 1979. What is Justifed Belief? In G. Pappas, ed., Justification and Knowledge. Dordrecht: D. Reidel, 1–23. Graham, Peter. 2012. Epistemic Entitlement. Noûs 46(3): 449–82. Grice, Paul, H. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press. Hawthorne, John. 2004. Knowledge and Lotteries. Oxford: Oxford University Press. Hawthorne, John and Jason Stanley. 2008. Knowledge and Action. Journal of Philosophy 105(10): 571–90. Ichikawa, Jonathan. 2012. Knowledge Norms and Acting Well. Thought 1(1): 49–55. Lackey, Jennifer. 2007. Norms of assertion. Noûs 41(4): 594–626. Lawlor, Krista. 2012. Assurance—An Austinian View of Knowledge and Knowledge Claims. Oxford: Oxford University Press. Leite, Adam. 2005. A localist solution to the regress of epistemic justification. Australasian Journal of Philosophy 83(3): 395–421. Littlejohn, Clayton. 2012. Justification and the Truth-Connection. Cambridge: Cambridge University Press. Kelly, Thomas and Sarah McGrath. 2010. Is reflective equilibrium enough? Philosophical Perspectives 24(1): 325–59. Kelp, Christoph. 2014. Knowledge, Understanding, and Virtue. In Abrol Fairweather, ed., Virtue Scientia: Virtue Epistemology and Philosophy of Science. Dordrecht: Springer, 347–60. Khalifa, Kareem. 2011. Understanding, Knowledge, and Scientific Antirealism. Grazer Philosophische Studien 83(1): 93–112. Khalifa, Kareem. 2013. Understanding, grasping, and luck. Episteme 10(1): 1–17. Kvanvig, Jonathan L. 2009. Assertion, knowledge, and lotteries. In Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 140–60. McGlynn, Aidan. 2014. Knowledge First? London: Palgrave Macmillan. McGlynn, Aidan. 2017. Mindreading Knowledge. In J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis, eds., Knowledge-First Approaches in Epistemology and Mind. Oxford: Oxford University Press, 72–94. McKinnon, R. 2015. The Norms of Assertion: Truth, Lies, and Warrant. London: Palgrave Macmillan. Madison, B. J. C. 2010. Is Justification Knowledge? Journal of Philosophical Research 35: 173–91. Nagel, Jennifer. 2013. Knowledge as a Mental State. Oxford Studies in Epistemology 4: 275–310. Reichenbach, Hans. 1938. Experience and Prediction. Chicago, IL: University of Chicago Press. Schickore, Jutta and Friedrich Steinle, eds. 2006. Revisiting Discovery and Justification: Historical and Philosophical Perspectives on the Context Distinction. Dordrecht: Springer.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

against knowledge-first epistemology  71 Schiffer, Stephen. 2009. Evidence = Knowledge: Williamson’s Solution to Skepticism? In Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 183–202. Sliwa, Paulina. 2015. Understanding and Knowing. The Proceedings of the Aristotelian Society 115(1, part 1): 57–74. Sosa, Ernest. 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, Volume 1. Oxford: Clarendon Press. Stanley, J. 2005. Knowledge and Practical Interests. Oxford: Oxford University Press. Strawson, P. 1992. Analysis and Metaphysics: An Introduction to Philosophy. Oxford: Oxford University Press. Sutton, J. 2007. Without Justification. Cambridge, MA: MIT Press. Thomson, Judith Jarvis. 1971. A defense of abortion. Philosophy and Public Affairs 1(1): 47–66. Turri, John. 2010. Epistemic invariantism and speech act contextualism. Philosophical Review 119(1): 77–95. Turri, John. 2011. The Express Knowledge Account of Assertion. Australasian Journal of Philosophy 89(1): 37–45. Walden, Kenneth. 2013. In defense of reflective equilibrium. Philosophical Studies 166(2): 243–56. Williamson, Timothy. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, Timothy. 2005. Contextualism, subject-sensitive invariantism and knowledge of knowledge. Philosophical Quarterly 55(219): 213–35. Williamson, Timothy. 2007. On Being Justified in One’s Head. In M. Timmons, J. Greco, and A. Mele, eds., Rationality and the Good. Oxford: Oxford University Press, 106–22. Williamson, Timothy. 2007. The Philosophy of Philosophy, Oxford: Blackwell. Williamson, Timothy. 2009a. Reply to Quassim Cassam. In Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 285–92. Williamson, Timothy. 2011. Knowledge First Epistemology. In Sven Bernecker and Duncan Pritchard, eds., Routledge Companion to Epistemology. London: Routledge, 208–18. Williamson, Timothy. 2013a. Knowledge First. In M. Steup and J. Turri, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Blackwell, 1–9. Williamson, Timothy. 2013b. Knowledge Still First. In M. Steup and J. Turri, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Blackwell, 22–4. Williamson, Timothy. Forthcoming. Justifications, Excuses, and Sceptical Scenarios. In Julien Dutant and Fabian Dorsch, eds., The New Evil Demon. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

4 Mindreading Knowledge Aidan McGlynn

1. Introduction Is knowing a mental state in its own right, as believing is, or is it at best a mental state in an attenuated sense due to being a species of belief?1 Timothy Williamson defends the former account of the nature of knowing in Knowledge and its Limits (2000). There he argues that knowing and other factive propositional attitudes are sufficiently like ­non-factive attitudes, such as believing and desiring, that there should be an initial presumption in favour of taking the factive attitudes to be mental states in their own right, too (2000: 22). Moreover, he contends that the strongest objections to knowing counting as a mental state in its own right—that we do not seem to enjoy the same kind of privileged access to our own states of knowing as we do to our uncontroversially mental states, and that knowing does not really play a role in the explanation of action—fail to dislodge that presumption. The case against knowing being a mental state in a full-blooded sense, Williamson argues, is no better than that against broad attitudes—attitudes that have their contents fixed in part by factors that do not supervene on the internal state of the subject—and he offers reasons to think that both cases fail.2 1   This chapter not only extends, but supersedes the brief and problematic discussion of this topic in Knowledge First? (McGlynn 2014: 183–6). A version was presented at a workshop on knowledge first philosophy at the University of Edinburgh in November 2014, and I am grateful to all of the participants to that event for feedback and discussion. Richard Price encouraged me to set up a ‘session’ on academia.edu, which led to very useful feedback on a complete draft of the paper in February 2015; thanks here is owed to Derek Anderson, Axel Barceló, Kati Farkas, Tim Kenyon, Kristina Musholt, and especially Jennifer Nagel and Richard Price. Further thanks are owed to Giada Fratantonio, Mikkel Gerken, Guy Longworth, Bryan Pickel, Lukas Schwengerer, Till Vierkant, the editors of this volume, and an anonymous referee for comments and discussion, which led to a host of improvements. I finished the initial draft of the chapter shortly before I became aware of David Rose’s then-forthcoming paper ‘Belief is Prior to Knowledge’ (2015), which makes a number of the same points in response to Nagel. However, the differences of emphasis and detail seem great enough to justify another discussion of the issues; where there is overlap, I have credited Rose’s paper in the references, and I am grateful to him for a helpful email exchange. This chapter was written before the recent discussion about replication in social psychology began in earnest, and so I do not engage those issues even where one might think they bear on my topic. 2   For some discussion of the relationship between the thesis that knowledge is a mental state in its own right and other aspects of Knowledge-First philosophy (such as the claims that belief, action, and assertion are governed by knowledge norms) see McGlynn 2014: 19, and Williamson’s contribution to this volume.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  73 The opposed view that knowledge is not a mental state in its own right has not often been explicitly articulated (although see Williamson 2000: 55–6 for discussion of some examples).3 However, it is arguably presupposed by the tripartite account of knowledge sunk by the counterexamples in Gettier 1963, and by many of the subsequent attempts to revise or replace the tripartite account. If such accounts are taken as addressing issues concerning the nature of knowledge, and not merely the concept of knowledge, then they are naturally read as suggesting that to know is to have a belief that meets further mental and non-mental conditions. Recent defenders of this kind of account of knowing have tended to concede the force of Williamson’s defence of his position from the two objections mentioned above. However, they have argued that we have not been given sufficiently strong motivations to take the disquieting step from externalism about the contents of our mental states, of the sort familiar from Putnam and Burge (e.g., Putnam 1975, and Burge 1979), to an externalism about the mental thoroughgoing enough to admit factive attitudes such as knowing (see Leite 2005, Fricker 2009, and McGlynn 2014: Ch. 8).4 In a fascinating recent paper, Jennifer Nagel (2013) has proposed that examining the nature and development of our mindreading capacities enables us to get some empirical purchase on the debate between these two very different accounts of the nature of knowing. Our mindreading—or ‘theory of mind’—capacities are those that allow us to attribute mental states to minded creatures (or to things that appear suitably minded), on the basis of their verbal and non-verbal behaviour together with other relevant cues and background information. The idea now is that we can associate each of the rival views of the metaphysical nature of knowledge with a corresponding account of how we go about attributing knowledge to each other (and to ourselves). The natural account for one who holds that knowing is a mental state in its own right is that tasks involving the recognition and attribution of knowledge are simply tasks for one’s mindreading system, just as tasks involving the attribution of belief are. In ­contrast, Nagel suggests that the rival view, on which knowing is a hybrid of mental components and non-mental components, should be associated with the view that the attribution of knowledge involves a corresponding blend of mindreading (to figure out what the target believes and whether she meets any further mental conditions on knowing) together with whatever capacities are involved in determining the truth of the proposition in question (2013: 284). Nagel then argues that the empirical data supports the former account. She holds that the available evidence supports the claim that an ability to attribute knowledge is not dependent on an ability to attribute belief. In fact, she suggests that the dependency 3   There are, of course, intermediate views here; one might hold that knowledge is unanalysable but not a mental state, for example, see Fricker (2009), McGlynn (2014: 168), and Smith, Ch. 5 this volume). Thanks to Kati Farkas and the editors for stressing this point. 4  I am very sympathetic to Martin Smith’s contribution to this volume, in which he argues that Williamson’s version of externalism about the mind is more radical and problematic than familiar and widely-accepted versions of content externalism.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

74  Aidan M c Glynn runs in the other direction, and that this is evidence that the concept of knowledge is prior to that of belief. Moreover, she argues on this basis that the thesis that knowing is a mental state in its own right is ‘one of the central principles of our intuitive mindreading system’ (2014: 303) and to the extent that one adopts a ‘non-skeptical’ view of  our mindreading system, one will see its explanatory and predictive success as ­confirming that thesis. There are a number of points at which one might challenge this line of reasoning. One might wonder whether in general one can draw metaphysical conclusions about the nature of knowing from such claims about the concept that we, or our mindreading systems, exercise (e.g. Gerken, Ch. 3 this volume, and Ichikawa and Jenkins, Ch. 6 this volume), or one might have more specific worries about Nagel’s claim that a ‘non-skeptical’ view of our mindreading capacities requires that one accept the truth of the potentially philosophically loaded principles on which it operates (McGlynn  2014: 186). One might also wonder whether the claim that the concept of knowledge is prior to that of belief should really be understood as entailing that the concept of knowledge is acquired first (and perhaps, for some creatures, last).5 However, I will not push these kinds of worries here, not because I think that they are unworthy of being explored, but to keep the discussion somewhat focused and manageable. Instead, I will concentrate my attention entirely on the initial step of Nagel’s argument, namely, her attempt to show that there is empirical evidence that speaks firmly in favour of the claim that the concept of knowledge is acquired prior to the concept of belief (see also Rose 2015). It is possible to discern several distinct lines of argument in Nagel’s paper (though she tends to let them run into each other). First, she offers evidence from developmental psychology in favour of the thesis that children acquire the concept of knowledge before they acquire the concept of belief.6 She claims that ‘knows’ and its cognates appear earlier in a child’s vocabulary and are used much more frequently than ‘belief ’ and ‘think’, and the likes, and that a generalization of the point holds cross-linguistically. Moreover, she argues that there is more direct evidence that shows that children can successfully attribute knowledge and ignorance to others before they can successfully attribute beliefs. Second, Nagel reinforces this first line of argument with an appeal to comparative psychology, arguing that chimpanzees, for example, have at least a partial grasp of the concept of knowledge, but lack the concept of belief. Finally, Nagel contends that a prior grip on the distinction between knowledge and ignorance is needed even for simple tasks involving the attribution of belief. It is worth noting that the third 5   Thanks to Derek Anderson and Richard Price for raising these issues. See Butterfill (2013) and Roessler (2013) for further ways in which one might criticize Nagel’s argument. 6   It is now standard to draw a distinction between concept possession and concept mastery. For example, Burge’s (1979) ‘arthritis man’, who believes he has arthritis in his thigh, has not fully mastered the concept of arthritis, but he may still count as possessing it (as evidenced by his willingness to defer to the right members of his linguistic community when his mistake is revealed). I take the issues discussed in the present chapter to concern possession rather than mastery, and this should guide the interpretation of the discussion when it is put in terms that may be ambiguous between these (for example, ‘understanding’). Thanks to Mikkel Gerken and the editors for prompting this clarification.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  75 consideration supports a stronger priority thesis than the first two, since it suggests that an ability to attribute belief not only develops later than an ability to attribute knowledge, but in fact presupposes an ability to attribute knowledge. This chapter divides into two main sections, with the bulk of the discussion falling in the first. Section 2 examines whether the empirical evidence really does favour Nagel’s conclusion that there are creatures—young children and chimpanzees—that exercise the concept of knowledge in mindreading, but not the concept of belief. I consider both Nagel’s argument and an earlier suggestive remark due to Williamson, and I’ll argue that no clear-cut conclusion to this effect is warranted, particularly in light of our lack of clear criteria for when we should say that a creature possesses a given concept, and in light of recent evidence that even very young infants can attribute false beliefs. It will be helpful to say a little about the import of this last point now. Stemming from Wimmer and Perner’s work in the early eighties (1983), the standard picture was that children pass tasks involving attributing a false belief to an individual around the age of four, with subsequent studies suggesting that this could be lowered towards the age of three by making the tasks less demanding on children’s cognitive resources.7 So, when Williamson’s book was published in 2000, it was widely held that it was sometime in the fourth year of life that children first manifest a clear of grasp of representational states, and that they should not be credited with the concept of belief until then. However, standard false belief tasks are verbal, in the sense that questions are asked to elicit a communicative response from the subject (though in some cases, this involves only pointing). For example, two children are shown some chocolate being hidden at a particular location. One of the children is asked to leave the room, and the chocolate is hidden in a different location. The child that witnessed this move is then asked ‘If we asked [name of child who left] where the chocolate is, what would she say?’, or similar. The child has to either name or point to a location; they ‘pass’ the false belief task if they answer that the other child will take the chocolate to be in the original hiding place, not its actual hiding place. Since 2005 it has begun to be more widely accepted that children are able to successfully attribute false belief, and so manifest a grasp of the representational nature of belief, much earlier. Onishi and Baillargeon’s (2005) pioneering study suggested that children can do this as young as 15 months, and subsequent studies have corroborated their results (e.g. Scott and Baillargeon 2009). Indeed, more recent work has pushed the age at which children can pass false belief tasks significantly lower—perhaps as low as 7 months (Kovács, Téglás, and Endress 2010). The tasks in these experiments are non-verbal, instead measuring looking-times, relying on the well-established premise that infants look longer when their expectations are violated. Based on this, it seems that very young infants expect an agent to look for an object where it was when they last saw it, and are surprised when they look in a different location, even though the infant saw the object being moved to that location. The theoretical gloss that has been   See Goldman (2006: 73–4) for discussion and further references.

7

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

76  Aidan M c Glynn put on this is that the infant has attributed a false belief about the object’s location to the agent, and expects the agent to act on that belief, rather than on the real location of the object. Now, this shift in when children are believed to be able to attribute false beliefs looks like it should have a huge bearing on arguments that a grasp of belief develops later than an understanding of knowledge, and so we’d expect to feel the effects of that shift between Williamson’s discussion and Nagel’s. But in fact, these issues do not really have any impact on Nagel’s discussion at all; she briefly mentions Onishi and Baillargeon’s paper in a footnote, but she does not even acknowledge the relevance it might have for her argument.8 One aim in Section 2 will be to try to bring out why this evidence of early ability to attribute belief is problematic for Nagel (and to do so without assuming any particular controversial interpretation of these results). Finally, Section 3 briefly questions Nagel’s further argument for her stronger priority claim—that an ability to attribute knowledge is a precondition of an ability to attribute belief.

2.  De We Acquire Knowledge First? 2.1  Williamson on the priority of knowledge My target in this section is the claim that evidence from developmental work on ­children and from comparative work on chimpanzees shows that there are creatures that mindread knowledge but are unable to mindread belief, and so suggests that the concept of knowledge is acquired first. As Nagel notes (2014: 292 fn. 21), Williamson has already made a proto-version of this point (2000: 33 fn. 7): ‘A further ground for suspicion of analyses of the concept knows in terms of the concept believes is that they seem to imply that the latter concept is acquired before the former. Data on child development suggest, if anything, the reverse order (see Perner 1993: 145–203 for discussion of relevant work). Crudely: children understand ignorance before they understand error. Naturally, the data can be interpreted in various ways, and their bearing on the order of analysis depends on subtle issues in the theory of concepts.’

We should note that Williamson’s intentions are much less ambitious than Nagel’s. It is clear from this passage that Williamson is only hoping to cast (further) doubt on the claim that the concept of knowledge can be analysed in terms of the concept of belief whereas, as we have seen, Nagel hopes to establish that knowledge is both conceptually and metaphysically prior with appeal to these kinds of considerations. Williamson does not spell out how the data discussed in Perner 1993 might support the claim that the concept of knowledge is acquired before the concept of belief. 8   In Nagel’s brief summary of her argument in her more recent book (2014: 105), she adds: ‘To complicate matters further, recent evidence suggests that young children and even infants do have some sort of partial and implicit capacity to recognize false beliefs, although the scope and significance of this capacity are still unclear.’ However, she does not indicate how she might avoid or overcome this complication.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  77 Moreover, on the face of it, that data supports no such thing, even if we bracket the subsequent developments that will occupy us below. It is true that Perner argues that children exercise the concept of knowledge before they pass standard verbal false belief tasks, but he also argues that their understanding of knowledge is deficient until they come to acquire a representational theory of mind (of precisely the sort manifested by passing the false belief tasks). Specifically, Perner takes the empirical evidence to suggest that children initially take whether a subject gets things right, or even just whether they are successful in their actions, as a criterion for whether they know. It is only with the development of a representational theory of mind around the age of four that they grasp that knowing requires ‘a properly caused mental representation of the known fact’ (1993: 145, italics in original): The heavy reliance on success as a best indicator for “knowing” gives way to understanding that access to sufficient and reliable information is a necessary prerequisite for knowing. (1993: 145) Acquisition of the concept of knowledge proceeds in two stages. Children start by using knowledge as a theoretical construct within a theory of behavior that explains successes and failures. Later, as children acquire the concept of representation at about 4 years, they reconceptualize what they know about knowledge within a representational theory of mind. (1993: 169, italics in original)

It is left a little unclear whether Perner takes there to be a single concept of knowledge that is retained throughout this development. He is somewhat prone to talking of the ­‘children’s concept of knowledge’ (1993: 145) and ‘the adult concept of knowledge’ (1993: 146), as if these might be distinct. However, the second of the passages quoted above ­suggests that this is not how Perner really thinks of the ‘reconceptualization’ that happens when a child acquires the concept of representation; rather, the child gains a new understanding of knowledge by bringing it in under a representational theory of mind. Perner explicitly relates this account of the development of a child’s concept of knowledge to his preferred account of the development of their concept of belief, s­ uggesting that the data on when children first reliably pass standard false belief tasks ‘accords well with the acquisition of a representational theory of knowledge discussed in chapter 6’ (1993: 179). Despite this, there are some indications that Perner holds that children do not really manifest a grip of the concept of belief until they pass standard false belief tasks; at the very least, we may observe that Perner explicitly holds that children have a concept of knowledge prior to gaining a representational theory of mind, while appearing somewhat reluctant to endorse the analogous claim regarding belief. However, it is not clear what might justify such a difference between his stance on the concept of belief and his stance on the concept of knowledge. One reason to be suspicious of such a difference is that children pass other tasks that apparently involve the attribution of belief at an earlier age (Nichols and Stich 2003: 74). Perner acknowledges this point, but he argues that the tasks in question can be passed if the child employs a simple behavioral rule; no reasoning about mental states—no genuine mindreading—is required (1993:

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

78  Aidan M c Glynn 176–7). For example, children aged 3 years and 2 months can reliably pass the so-called ‘discrepant belief task’ (Wellman and Bartsch 1988); the child knows that there are bananas in both the cupboard and the refrigerator, and knows that the protagonist wants a banana and believes that there are only bananas in the cupboard, and is asked ‘Where will [name of protagonist] look for bananas?’. According to Perner, a child may correctly respond that the protagonist will only look in the cupboard, not because she is acting on a false belief, but ‘due to a simple rule that “thinking” predisposes people to act in relation to the thought-about object’ (1993: 177).9 I do not find this particular rule-based explanation all that clear or compelling. More importantly, there is a methodological worry here. It is now a widely recognized point in comparative psychology that it is very difficult to empirically distinguish the thesis that an animal’s success at a task is due to mindreading from the thesis that it is due to following a behavioural rule which the animal has learned, which directly links the circumstances in question to the resulting behaviour (e.g., Povinelli and Vonk 2004). Perner (2010) has recently made a parallel point concerning human children (though he does observe that there is an obvious difference, namely that we are fairly sure that adult humans engage in mindreading (2010: 250)). In response, some prominent defenders of the claim that nonhuman animals and very young children really do engage in mindreading have offered theoretical arguments rather than empirical ones, namely that the recipe for generating empirically adequate rules is parasitic on results generated by researchers working on the supposition that animals and young children really do engage in mindreading, and that the resulting theory is less parsimonious (e.g., Carruthers  2011: chapter  8, Carruthers  2013, Fletcher and Carruthers  2013, Nagel  2014: 297-8).10 For example, concerning the evidence that nonhuman primates have some grasp of knowledge, Nagel writes: It is possible to maintain a strict behavior rules’ interpretation of the recent experimental results, for example, by positing that what the subordinate animal really knows is not that the dominant knows where the food is, but that any dominant will go for food that has recently been in its line of sight (Povinelli and Vonk, 2003). However, this strategy may put us on a slippery slope: if we are allowed to formulate behavior rules ad hoc, then a behavior rules explanation could be produced for any pattern of apparent mental state ascription, even one involving advanced social interactions amongst adult humans (Fletcher and Carruthers, 2013). The fact that a certain interpretative strategy is in principle available—as behaviorism continues to be available, in principle, to explain even human social interaction—does not mean that it is advisable, or that it is the strategy that would yield the best explanation of our data. Considerations of parsimony and fruitfulness also matter. 9   Perner offers the following example: ‘if I state, “I am thinking of eating Chinese tonight,” then one can reasonably predict that I will be heading to the restaurant that serves Chinese food rather than the one that specializes in Italian Cuisine’. I do not find either the rule or the example particularly clear, particularly since the utterance in Perner’s example just looks like a clear expression of an intention to eat Chinese food on the night in question. 10  Povinelli and Vonk claim that the methodological advantages instead lie with their rule-based approach (2004: 9): compare Ruffman (2014: 275–7).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  79 Whatever one makes of this dispute, it is clear that deflationary, rule-based explanations of successes on tasks thought to involve mindreading are readily available. Given this, it is very likely that we will be able to find one that can be used to explain away young children’s success on the discrepant belief task, and any other tasks that suggest that they have a partial or mistaken grasp of belief prior to passing standard false belief tasks. But clearly that is a double-edged sword, since we will very likely be equally well placed to explain away successes on tasks that suggest that children have a partial or mistaken grasp of knowledge prior to acquiring a representational theory of mind. Appeals to rule-based explanations do not seem to discriminate between knowledge and belief in the right kind of way to avoid this parity.11 So, on the face of it, there is little in Perner’s 1993 discussion that supports the conclusion that children acquire the concept of knowledge before they acquire the concept of belief. Rather, Perner’s discussion suggests that gaining an adult’s understanding of either concept is at least a two-stage process, with the second stage involving the acquisition of a representational theory of mind sometime in a child’s fourth year. It might be noted in reply that throughout his discussion, Perner relies on an account of the concept of knowledge that would be unacceptable to Williamson, and I have been relying somewhat on that aspect of Perner’s discussion in my response. Following Dretske  1981, Perner characterizes knowledge as a mental representation of a fact that is created by access to reliable information about that fact (1993: 146), and this is exactly the kind of analysis of the concept of knowledge that Williamson rejects. But that observation only serves to reinforce my point that it is very unclear how Perner’s discussion offers Williamson what he wants. Any argument for the priority of knowledge based on Perner’s discussion will have to involve cherry-picking conclusions and offering alternative explanations of the significance of many of the results that Perner reports, since Perner himself takes these results to suggest that children progress ­step-by-step towards something like the Dretskean conception of knowledge. Perhaps there is a principled way to pick out data from Perner’s discussion that supports only those conclusion that Williamson would endorse, but Williamson certainly does not show us how that might be done.

2.2  Nagel on the priority of knowledge As I stressed above, Nagel’s aims are significantly more ambitious than Williamson’s, and she draws on a broader range of the available developmental and comparative ­evidence. However, I worry that Nagel fails to heed the note of caution sounded in the final sentence of the passage from Williamson quoted above; she does not sufficiently 11   An anonymous referee has pointed out that Perner (1989) contains a number of other suggestions about how to resist the conclusion that passing the discrepant belief task requires a grasp of belief. I lack the space to discuss these earlier proposals, but see Wellman (1992: 82–8) for relevant critical discussion.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

80  Aidan M c Glynn consider alternative interpretations of the data, nor does she spell out in any detail the connections between the data she cites and issues about concept acquisition and possession. As noted in the introduction, Nagel argues for the claim that the concept of knowledge is prior to that of belief by citing three sources of evidence. First, she cites Shatz, Wellman, and Silber 1983 and Bartsch and Wellman 1995 as showing that children use ‘know’ earlier and more frequently than they use ‘believe’ or ‘think’ (2013: 292–3). Nagel presents this as evidence ‘relevant to the question of whether the concept of knowledge is prior to the concept of belief ’ (2014: 292), though its significance is left rather unclear in her discussion. The suppressed premise is presumably something to the effect that the order of concept acquisition is reflected in the order in which the corresponding verbs are learned. Stated this plainly, the claim perhaps looks a bit too crude to be plausible, and though Nagel is not explicit on this point, she almost certainly has in mind a more nuanced version, such as that articulated by Bartsch and Wellman (1995: 17): Even though language development does not map onto conceptual development in any strict sense, an analysis of discourse can nonetheless provide an important window onto conception. We intend to show that systematic examination of children’s language provides considerable insight into children’s understanding of beliefs and desires. At least it does so if appropriate data are available and careful methods of analysis are employed.

Let us take Nagel’s appeals to this data about children’s discourse about the mind in this spirit. The claim about the higher frequency of uses of ‘know’ is difficult to dispute (see Bartsch and Wellman 1995: 29), but it is also very difficult to see its relevance to questions about conceptual priority, and Nagel does not speak to this issue at all. The relevance of the claim that ‘know’ makes its way into children’s vocabularies earlier is much more apparent, assuming Bartsch and Wellman’s methodology is sound, but it is less clear that this claim is actually supported by the work cited by Nagel. Shatz, Wellman, and Silber’s discussion is mostly focused on a single child, Abe, and while he used ‘know’ earlier than ‘think’, the authors in fact take his first genuine references to mental states to involve ‘think’ rather than ‘know’ (1983: 306 and 315: compare Rose 2015: 388–9 on this point).12 Later in the same paper, the authors report a second study involving thirty children, and they note that the children in this study preferred either ‘know’ or ‘forget’ (1983: 315). However, the numbers involved are still very small indeed, since the majority (all but six) of the children failed to produce an utterance that the authors counted as involving genuine reference to a mental state during the study at all. With such a small group, it is hard to see how we could have any confidence that we’re dealing with a remotely representative sample of English-speaking children of that age, and so this does not seem like good evidence of the priority of the concept 12   Of course, ‘think’ is hugely ambiguous, and we cannot conclude from what we’re told that Abe’s first references to mental states were references to beliefs.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  81 of knowledge. As far as I have been able to determine, Bartsch and Wellman make no claims and offer no evidence that speaks directly to the matter of whether ‘know’ or  ‘think’ comes first.13 So I am not convinced that the evidence from discourse ­analysis cited by Nagel supports her claim that children use ‘know’ first as strongly as she suggests.14 Despite this reservation, let us suppose that Nagel’s claim is correct. Even given the methodology laid out by Bartsch and Wellman, it is not clear how much weight this point should be allowed to carry, given the more recent evidence from nonverbal s­ tudies (discussed in more detail later) that suggests that infants can, in fact, manifest a grasp of the concept of belief very early on. Such studies might be taken to show that focusing on the use of mental state terms offers an unduly conservative picture of when children acquire mental state concepts like belief. At the very least,

13   In Nagel’s paper she unfortunately didn’t offer any page references to Bartsch and Wellman’s book, and so it was left unclear where to look for the relevant discussion. I’m grateful to Nagel for following up on this issue via email, though I remain committed to the claim in the text. Table 3.1 (Bartsch and Wellman 1995: 41) gives examples of utterances involving ‘know’ and ‘think’, but we cannot assume that these are the earliest examples that the authors found, and so we cannot conclude anything about which word was used earlier from this table. Indeed, there is some evidence that the authors did not use the earliest example of a use of ‘think’ that was available to them. The earliest example of a use of ‘think’ offered by Table 3.1 is from the child Abe at 3 years and 3 months, and Shatz, Wellman, and Silber report that Abe first used ‘think’ to refer to a mental state when he was 2 years and 8 months old (1983: 306). Nagel also drew my attention to Bartsch and Wellman’s results on when children first refer to sources of knowledge (Table 3.9, 1995: 62) and when they first employ ‘think’ in contrastive constructions (Table 3.3, 1995: 50), observing the former typically happens before the latter. Since Bartsch and Wellman do not take reference to sources of knowledge to be necessary for genuine reference to knowledge, nor use in a contrastive to be necessary for genuine reference to belief, it is hard to see how this is evidence that children typically refer to knowledge first (see Rose 2015: 387–91 for further discussion). Rose reaches the same conclusion as I do: ‘Bartsch and Wellman (1995) . . . do not present a timetable regarding the emergence of “think” and “know” in genuine mental state ascription’ (2015: 389). 14   Nagel also cites Tardif and Wellman (2000) in support of the claim that children learning languages other than English (specifically Mandarin and Cantonese) also learn a word for the concept of knowledge first. However, the sample sizes of these studies are also incredibly small; there were eighteen children studied in total, with ten in the study on Mandarin and eight in the study on Cantonese, and not all of these children made a relevant utterance in the time-frame of the studies (compare Rose 2015: 390). So again, any evidence supplied here for a general claim about the priority of the concept of knowledge looks like it must be very weak indeed. Moreover, Tardif and Wellman are unable to rule out one of the most obvious alternative explanations of why Chinese-speaking children might use words for ‘think’ and ‘believe’ later than their English-speaking counterparts (despite in general using verbs earlier), namely that such words figure less frequently in adult speech in Chinese, and in particular in adult speech directly at children. They found that Mandarin-speaking caregivers do use words for ‘think’ and ‘believe’ notably less frequently to children, and that Mandarin-speaking children tend towards ‘relatively late development and minimal reference to thinking’ (2000: 31), and they note that they haven’t determined whether Cantonese-speaking caregivers refer to thinking and believing similarly infrequently (2000: 37). Tardif and Wellman do caution against adopting the general view that the order in which children learn mental state verbs depends on ‘adults’ conversational topics’ (2000: 37), but they do take seriously the more specific hypothesis that Chinese-speaking children refer to thinking and believing later than their English-speaking counterparts because such references are so rare in their caregivers’ input. Perhaps for this reason, Tardif and Wellman themselves hedge somewhat on the issue of whether their studies suggest that children acquire a term for knowledge before a term for thinking or believing (2000: 35). Thanks to a referee and to Jennifer Nagel for encouraging me to say more about this issue, and see Rose 2015: 395 for further discussion.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

82  Aidan M c Glynn it seems like this kind of evidence concerning linguistic development will be highly defeasible, and depending on how the issues discussed below work out, may have already been defeated. Nagel also offers more direct evidence of her conceptual priority claim, based on  children’s success and failure on certain mindreading tasks. For example, she cites Hogrefe, Wimmer, and Perner (1986) in support of the claims that recognizing that someone knows cannot invariably implicate the ability to recognize what they believe and that children grasp the concept of knowledge before they grasp the ­concept of belief. In Hogrefe, Wimmer, and Perner’s first study, pairs of children are shown that a familiar container contains what one would expect it to contain; they are show a dominoes box contains dominoes, or a box for paperclips contains paperclips, or a Smarties tube contains Smarties chocolates, and so on. One child is then asked to leave, and the contents of the container are replaced with something unexpected. The child who witnessed the switch is then asked the following two questions in random order: (1) ‘Does [name of first child] know what is in the container or does he not know?’, and (2) ‘If we ask [name of first child] what is in the box, what will he say?’15 Nagel writes (2013: 295–6): Among three-year-olds, 39 percent answered question (1) correctly, and only 6 percent answered question (2) correctly; four-year olds improved to 81 percent and 44 percent, and five-year olds were right 88 percent and 76 percent of the time on the two questions. If we generally made judgments about the presence or absence of knowledge by attributing belief and then evaluating the truth or falsity of this belief, we would not expect to see such a lag between the capacity to recognize the absence of knowledge and the capacity to attribute a false belief.

However, Nagel only reports the results of the first of seven experiments, and Hogrefe, Wimmer, and Perner themselves note a potential worry one might have with that first study, namely that the ‘Ignorance question’ and the ‘Belief question’ are not very closely matched. Here is the precise formulation of the questions again (1986: 569): Ignorance Question:  ‘Does [name of other child] know what is really in that box?’ Belief Question:  ‘If we ask [name of other child], what will he [she] say is in the box?’ The Ignorance question is a simple yes/no question, while the Belief question requires the child to produce the name of the contents, and one might worry that this makes the knowledge/ignorance version of the task easier than the false belief task for reasons that do not bear on which concepts the child possesses. In particular, we might suggest that correctly answering the Belief question, unlike the Knowledge question, requires a  child to inhibit an answer which instead gives the content of her own belief on the matter (e.g. Leslie 1994). The child believes (say) that there are paperclips in the Smarties container, and the suggestion is that, when asked what the other child will say 15   This is the so-called ‘unexpected content’ version of the false belief task. The version sketched in the introduction is the ‘unexpected location’ version.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  83 is in the container, the first child has to suppress an answer that expresses the content of their own belief; since the Knowledge question is a yes/no question, the content of the child’s belief about what is in the container manifestly is not at all suitable to be an answer to the question, and so there is no need for the child to inhibit this answer. More fully, the idea is that children’s answers initially have a strong ‘reality bias’, and it is only when their capacities such as those for management of their cognitive processes—their ‘executive function’ capacities—are sufficiently developed that this bias is overcome.16 Hogrefe, Wimmer, and Perner offer several variants of their first experiment that are designed to make the Belief question and the Ignorance question more alike, but they never manage to address the worry just mooted.17 Nagel is aware of such worries with the conclusion drawn by Hogrefe, Wimmer, and Perner, and in a footnote (2013: 296 fn. 24) she addresses them by appealing to two studies reported in Wellman and Liu 2004. Though Nagel does not explicitly discuss the objection raised in the previous paragraph, Wellman and Liu (2004: 537) do offer a serious concern with it, namely that there are other tasks involving attributing desires and beliefs that are different to one’s own that children pass much earlier than they pass standard false beliefs tasks (including the so-called discrepant belief task, and a similar task involving attributing discrepant desires), and these other tasks also require children to inhibit their own answers. In light of this, let us consider a different reply to Nagel’s appeal to Hogrefe, Wimmer, and Perner’s work in support of the conclusion that children acquire the concept of knowledge before the concept of belief. This worry is that drawing this conclusion from their work requires that one treats the standard false belief task as something like an ‘acid test’ for possessing the concept of belief. As I noted in my discussion of Perner, it is not clear why we should accept this, given that there are several other tasks involving the attribution of belief that children can pass before the standard false belief task: for example, the discrepant belief task discussed earlier, as well as tasks 16   See Apperly (2011: 22–6) for useful discussion of executive function and its relevance to performance on standard false-belief tasks. In light of the evidence already mentioned that children pass nonverbal false belief tasks much earlier than the standard verbal versions, some contemporary defenders of a mindreading module have modified Leslie’s explanation to apply more specifically to tasks involving language (e.g., Carruthers 2011: 250–1, 2013: 153–4). I will discuss Carruthers’s position in a little more detail below. One of the editors suggested that this proposal might concede too much to Nagel, since it might suggest that possessing the concept of belief requires more sophistication than possessing that of knowledge. However, the idea is that if we can adequately explain children’s difficulties on certain false belief tasks by appealing to the lack of development in their capacities for things like executive function, that undercuts what motivation there is for the view that they lack the concept of belief. 17   I lack the space to argue for this claim at length, but here is a brief version. Hogrefe, Wimmer, and Perner’s experiment 3 changes the Belief question to ‘Does he think that there are matches in the box or does he think there is chocolate in this box?’ (1986: 572), which requires the child to inhibit answering with the disjunct they know to be correct. Their experiment 5 keeps the same Belief question as experiment 3, but features a new Ignorance question which asks which of two characters knows the location of a familiar objection (1986: 575). However, answering a forced-choice question in which both answers are potential locations of the chocolate requires inhibiting an answer based on one’s own knowledge of the current location of the chocolate; answering a forced-choice question in which both answers are potential knowers does not.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

84  Aidan M c Glynn involving deception.18 In fact, Wellman and Liu (2004) argue that children typically pass the discrepant belief task before they are able to pass ignorance/knowledge tasks, which perhaps hints at a more complicated developmental story than the one that Nagel wants to press on us.19

2.3  Evidence of infant mindreading More importantly, we need to assess the import of recent evidence suggesting that ­children grasp false belief much earlier than their performance on standard (verbal) false belief tasks indicates. As noted above, Onishi and Baillargeon 2005 presented ­evidence from looking-time studies suggesting that infants grasp false belief as young as 15-months-old. Their conclusion rests on the well-established premise that infants look longer when their expectations are violated. What Onishi and Baillargeon found was that 15-month-old infants look longer when an agent, who was present when a desired object was put in one location but absent when it was relocated, looks for the  object in its actual location rather than its original location. This finding has been replicated several times (e.g. Scott and Baillargeon 2009), sometimes with even younger children; for example, Surian, Caldi and Sperber (2007) present a similar ­finding for 13-month-olds and, even more strikingly, Kovács, Téglás, and Endress (2010) claim to find sensitivity to false belief in 7-month-olds.20 Onishi and Baillargeon were inclined to take their results to support the conclusion that ‘15-month-old infants expect an actor to search for a toy where she believes, rightly or wrongly, that it is hidden’ (2005: 257), and this interpretation has recently been defended at length by Carruthers (2011: chapter 8 and 2013) and others. Of course, one who takes this line incurs an obligation to explain why children systematically fail ­verbal false belief tasks for at least another 21 months, but Carruthers has argued that a modified version of Leslie’s explanation discussed is available. Again, the basic idea is that the child’s answers to questions about what another believes initially show a strong reality bias, and what changes is the child’s capacity to inhibit these reality-biased answers. He writes (2013: 142): ‘ . . . the domain-specific mechanism implicated in mindreading competence is functional early in infancy. (I shall tentatively suggest: by the middle of the first year of life.) This provides 18   Chandler, Fritz, and Hala (1989) show that children as young as two and a half are able to spontaneously employ deceptive strategies when playing a competitive hide-and-seek game, and they take this to be evidence that such children understand the possibility of false belief. 19   For related points and further discussion of Wellman and Liu (2004), see Rose 2015: 392–5. Unlike Rose, I do not find much support for the priority of the concept of belief over that of knowledge here (much as I would like to). Both Butterfill (2013: 314) and Roessler (2013: 325–7) also emphasize that the developmental picture looks more complicated when one looks at just how little about knowledge children seem to grasp around the age they first pass standard false belief tasks. 20   There may also be some support from studies that exploit the disposition of toddlers to help, rather than relying on looking times (e.g., Buttelmann, Carpenter, and Tomasello  2009), though I’m grateful to  Jennifer Nagel for pointing out that Perner has raised worries about such evidence from the helper ­paradigm (2014: 297).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  85 infants with the concepts and core knowledge necessary to represent the mental states of other agents, of all basic types (including beliefs that are false and appearances that are misleading). While the operations of this system probably become more streamlined and efficient with age, its representational capacities do not alter in any fundamental way. Rather, what matures over time are the quality and extent of the interactions between this system and executive, attentional, and planning mechanisms.’

What Carruthers does is give this familiar kind of explanation a language-specific twist. Our default is to attribute our own beliefs to others when it is possible to do so smoothly and consistently with assumptions about shared background knowledge (2011: 237–8), and Carruthers thinks that ‘[t]here is surely every reason to expect that these reality-based defaults would be present in the language-use of young children’, while there is no reason to expect such a default to be present which respect to other forms of action (2011: 251).21 This kind of mentalistic explanation of infants’ performance on non-verbal false-belief tasks is not very conducive to Nagel’s argument. A flat-footed point is that as studies lower and lower the age at which infants are taken to pass false-belief tasks, there is less  evidence of any kind of developmental lag between the concept of knowledge and that of belief, of the sort that drives Nagel’s argument. However, it is not clear how much weight we can place on this point, since some proponents of mentalist explanations of infants’ successes do still posit a (brief) developmental lag (e.g. Scott and Baillargeon 2009), and since the evidence that infants can pass the relevant tests in their first year of life seems thinner than the evidence that they can do so by around 15-months.22 A better point is that the kind of interpretation of the evidence offered by Carruthers explains any such lag in terms of the on-going development of ‘the quality and extent of the interactions between [one’s mindreading] system and executive, attentional, and planning mechanisms’; it thereby undercuts the motivation for positing a gap in when children first acquire different mental state concepts. On the other hand, there are much more minimalist interpretations of what is going on in these kinds of studies. We have already mentioned one of these, according to which infants have learned a number of rules directly linking certain circumstances to behaviour of a particular sort, and other minimalist alternatives have been proposed (e.g. Heyes  2014a, discussed in Scott and Baillargeon  2014 and Heyes  2014b, and Ruffman 2014, discussed in Perner 2014 and Scott 2014).23 These also aim to explain 21   This is not the entirety of Carruthers’s account of the gap between nonverbal and verbal success on false-belief tasks (see 2011: 251–2), but it suffices for our present purposes. A number of readers have ­suggested to me that Carruthers’s claim that there is no reality-based default for non-verbal action seems unmotivated. I agree that this aspect of his account stands in need of further support, though since I am not defending his account, I will not attempt to supply that here. 22   I am indebted to an anonymous referee here, who drew my attention to relevant passages from Scott and Baillargeon and pointed out that some of my original claims were rather incautious. Thanks also to Jennifer Nagel for relevant discussion. 23  Though Nagel does not discuss these issues in her paper, in correspondence she has expressed ­sympathy for Ruffman’s minimalist approach.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

86  Aidan M c Glynn children’s success on mindreading tasks in ‘low-level’ rather than mentalistic terms, but without a commitment to this involving children learning rules which link certain kinds of circumstances with certain kinds of behaviour. For example, according to Ruffman, children represent what an individual has perceived and what she has not (where perception is understood ‘non-mentalistically’ (2014: 270)). Moreover, the familiarization stages of some of the experiments that are claimed to provide evidence of early mindreading will lead a child to form expectations about when that individual will act in certain ways, based on what that individual has perceived and what she has not. Since these expectations are violated in the false belief conditions, the child looks for longer.24 Some interesting middle-ground between robustly mentalistic and austerely minimalist interpretations has recently been explored by Butterfill and Apperly (2013). They propose that infants (and nonhuman primates) have a minimal theory of mind; they do track what others see, know, and believe, but they do not do this by representing those states. They offer a non-mental analogy to explain how one might track something without representing it (2013: 607): What could someone represent that would enable her to track, at least within limits, the toxicity of potential food items? Here the most straightforward answer (she could represent their toxicity) is clearly not the only one. After all, someone might track toxicity by representing odours or by representing visual features associated with putrefaction, say. Suppose Sinéad has no conception of toxins but represents the odours of food items and treats those with foul odours as dangerous to eat, so she would not normally offer them to friends of family nor conceal them from competitors. This brings nutritional and competitive benefits obtaining which depends on facts about toxicity. If Sinéad tends to behave in this way because of these benefits, representing odours enables her to track, in a limited but useful range of cases, toxicity. Our question, roughly put, is whether belief has something like an odour.

Butterfill and Apperly offer a detailed construction of the kinds of states that a child would need to represent in order to show the capacities for tracking belief and knowledge suggested by their success on tasks studied in the empirical literature on mindreading. We are asked to consider a subject called Lucky who is only able to represent ‘nonintentional behaviour’, and to ask what more she would need in order to track the desires, beliefs, and knowledge of others (and herself). The construction proceeds by introducing four principles. The first is ‘that bodily movements form units which are directed at goals’, where a goal, in the relevant sense is ‘simply an outcome to which an action is directed’ (2013: 614). In order to state the other principles, we need the concept of an agent’s field at a given time. This is a set of objects that are proximate to an agent, reasonably well lit, more or 24   This is obviously just a rough sketch of the kind of explanation Ruffman offers of some of the results (for example, Scott and Baillargeon 2009); the details need not detain us here. It is worth stressing, though, that Ruffman’s account of other related results is quite different; it is in part this which leads to Scott’s (2014) charge that his explanations are merely post-hoc.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  87 less in front of that agent’s eyes, and not hidden by any opaque barriers ‘unless the object was recently in motion and not behind a barrier’ (2013: 614). So, the objects in an agent’s field are ‘approximately those that the agent can perceive’ (plus, it seems, those which the agent has recently seen move or be moved into an occluded location). An agent is encountering an object if it is in her field. The second principle is then that ‘one cannot goal-directedly act on an object unless one has encountered it’ (2013: 615). As Butterfill and Apperly stress, they are not invested in the truth of this implausible principle. Rather their claim is that an agent who grasped it could track what another sees and perhaps knows, and form some limited expectations about their behaviour on that basis, without representing those states. The third principle again requires a new concept, that of registration, which Butterfill and Apperly implicitly define in terms of encounters and successful action. One ­registers an object at a location if one most recently encountered it at that location. A registration is correct if that object is indeed at that location, and correct registration is necessary for successful action involving that object as a goal (2013: 617). Armed with this principle, Lucky can predict that the relevant actions of an individual who has not correctly registered an object will not be successful, and can infer that an individual who has successfully acted has correctly registered the objection in question. Butterfill and Apperly argue that these three principles are already enough to explain many of the apparent mindreading capacities of nonhuman animals (2013: 615, 618), including those that Nagel appeals to in support of the claim that chimpanzees reason about states of knowledge (Nagel 2013: 297). The fourth principle is the really crucial one for our purposes, since it is the principle that Butterfill and Apperly appeal to in order to explain the success of very young infants on nonverbal false beliefs tasks (such as that described in Onishi and Baillargeon 2005). Here is how they state the principle (2013: 619): ‘when an agent ­performs a goal-directed action with a goal that specifies a particular object, the agent will act as if the object were in the location she registers it in’. Let Lucky be an infant in the false belief condition in Onishi and Baillargeon’s study. Here’s how we can describe things, given this fourth principle. Lucky sees an individual register an object. Then, in the absence of that individual, the object’s location is moved, so that this individual’s registration is no longer correct. Still, Lucky will expect that such an individual will look for that object where they register it, and that expectation is violated in the false belief condition. Some commentary will be useful here. First, though I will not discuss this here, Butterfill and Apperly are clear that these four principles may require supplementation or modification (2013: 620–1). Second, as already stressed, Butterfill and Apperly are not arguing for the truth of their four principles; the principles are clearly, at best, overgeneralizations. Third, they are also not arguing for the truth of the explanations they offer in terms of those principles (2013: 619). Rather, they are offering one possible account of what other than those states themselves a subject could represent in order to track desires, beliefs, and knowledge. Fourth, Butterfill and Apperly have much

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

88  Aidan M c Glynn to say concerning how an account like the one they offer might be empirically ­distinguished from both a richly mentalistic picture and a rule-based minimalist approach (2013: 621–8), but that important issue is not one we need to discuss here. Fifth, the idea is not that children shed their minimal theory of mind as they begin to  genuinely represent mental states. Butterfill and Apperly take there to be some ­evidence suggesting that adult humans may continue to rely on minimal mindreading abilities, and so the overall picture will have to be much more complex (see e.g. Apperly 2011: Chapter 5, and Butterfill and Apperly 2013: 608–10). Sixth, it is important to note that these principles give an account of how a creature could track both knowledge and belief without representing those states themselves; there is no suggestion that the account might apply to belief, but not to knowledge. In this section I have sketched several relatively well-developed accounts of the kind  of results offered by Onishi and Baillargeon, and by others subsequently. My aim has not been to defend any particular account, nor has it been to argue for the superiority (on empirical or theoretical grounds) of the hypothesis that children genuinely mindread over more minimalist accounts of their successes.25 The aim has rather to be to convey something of a sense of the space of options, with an eye towards bringing out just how odd and unmotivated an account that suited Nagel’s purposes would have to be; it would require a relatively full-blooded mentalistic interpretation of children’s early success on tasks involving the attribution of knowledge or ignorance, but a relatively minimalist explanation of their success on nonverbal false belief tasks.26 I do not want to claim that such an account would inevitably be either internally inconsistent or empirically inadequate. But I do strongly suspect that such an account would be rather ad hoc, and so very difficult to motivate. Moreover, Nagel does not indicate how one might go about developing such an account in the same level of detail as the accounts I have considered here. So, the suggestion that the best account of the mindreading abilities displayed by children is one on which they grasp knowledge but not belief looks rather suspect to me. It is worth stressing again that this way of bringing out why the evidence of early understanding of belief is problematic for Nagel does not require us to take that evidence as supporting the richly mentalist conclusion favoured by Onishi and Baillargeon themselves (as well as by Carruthers and others). Rather, the worry is that even if alternative interpretations of the data prove superior, they are likely to furnish more deflationary 25   Nor do I wish to suggest that these are the only possible or actual accounts worth taking seriously; for example, a very interesting alternative not discussed here can be found in the recent collaborative work between Perner and Johannes Roessler (e.g., Perner and Roessler  2010 and Roessler and Perner  2013). Roessler appeals to their account of early mindreading in reply to Nagel in his 2013. 26   Contrary to a suggestion from an anonymous referee, I do not think that it is enough for Nagel’s purposes if there proves to be some gap between children’s ability to attribute knowledge and their ability to attribute belief, of the sort posited in Rose and Baillargeon (2009). The crucial questions concern what explains the gap, and whether an explanation in terms of a lag in the possession of the relevant concepts is plausible. The mere fact of a gap, should it prove to be a fact, doesn’t by itself demonstrate that the explanation Nagel needs for her argument, in terms of a conceptual priority of knowledge over belief, is the most plausible explanation—or even a plausible explanation—of that gap.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  89 explanations of children’s (and chimpanzees’) success on tasks involving knowledge and ignorance too. Here I am in agreement with Butterfill (2013: 315): If we take 2- and 3-year-old children’s abilities to discriminate knowledge and ignorance as ­evidence that they are deploying a concept of knowledge, then it will be hard to justify denying that early sensitivity to belief does not involve deploying a concept of belief. If, on the other horn of the dilemma, we insist that these discriminatory abilities are not sufficient for concept possession, then we can no longer infer from 2- and 3-year-old children’s failure on standard false belief tasks that children deploy a concept of knowledge before they deploy a concept of belief.27

2.4  Animal mindreading I have mentioned Nagel’s appeal to mindreading in chimpanzees in passing throughout this section, and I want to close by briefly considering this issue a little more explicitly. Now, we seem to have broadly the same options here as we do when interpreting the apparent mindreading abilities of young children; we might offer a richly mentalistic interpretation (e.g. Carruthers 2011: 254–9, Fletcher and Carruthers 2013); or we might attribute a minimal theory of mind to primates and other nonhuman animals which explains how they track certain mental states by representing something else (Butterfill and Apperly 2013); or we might offer a more thoroughly minimalist explanation of their successes on the relevant tasks, perhaps in terms of behavioural rules (e.g. Povinelli and Vonk 2004). What makes animal mindreading interestingly different, from the point of view of our current concerns, is that the available evidence all seems to suggest that even animals as intelligent and social as chimpanzees cannot pass suitably modified false belief tasks (e.g. Kaminski, Call, and Tomasello 2008). It is not obvious what we can conclude from this, though. Nagel wants us to conclude that chimpanzees grasp knowledge but not belief, but as Apperly has argued, any understanding they have of the former is extremely partial and rudimentary—perhaps even more so than that of young children (2011: 53)—and we simply do not have a clear enough set of criteria for concept-possession and understanding to draw the conclusion that this suffices to be credited with the concept of knowledge (2011: 52–3, 109).28 27   See Roessler (2013: 325–7) for related discussion. I agree that it is good to note the proponents of the knowledge norm of the assertion here. However, the relevant sections do not merely negative criticism. Rather, they consist of comparative arguments–specifically, I argue that my own norm is superior all things considered. Consequently, be give the reader a more accurate impression of the debate if the opposing “warrant-assertive speech act norm” was also noted. It was first developed in Gerken (2012) and elaborated over a series of papers culminating in Gerken (2017). 28   Nagel discusses Apperly’s argument on this point, but somehow reaches the conclusion that he is in agreement with her. She writes (2013: 299 fn25): ‘[Apperly] grants the general priority of knowledge ­recognition over belief recognition. His overall assessment of the comparative literature is that “there is good evidence that some non-human species ‘understand something about’ seeing or knowing, but do not understand these mental states completely, and have no understanding of other mental states, such as believing” (2011: 109)’. This is rather misleading, however. The phrase Apperly quotes here, ‘understand something about’, echoes his own earlier use of that phrase in his book, and there he suggests that the kind of conclusion Nagel wants to draw ‘initially appears a reasonable gloss of current findings’, but he quite clearly goes on to argue that this initial appearance is misleading (2011: 52). The passage that Nagel quotes

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

90  Aidan M c Glynn Likewise, it seems much too quick to conclude that chimpanzees lack the concept of belief just on the grounds that they fail false belief tasks. That is again to assume that such tasks furnish some kind of ‘acid test’ for concept possession, and we have good reason to be sceptical that they can do so.29 My discussion of the comparative literature has admittedly been even less conclusive than my discussion of the developmental literature that Nagel appeals to. The main difference between the two, I have conceded, is that there is considerably less scope for making the argument that even animals like chimpanzees really do possess the concept of belief. As Apperly says (2011: 53), a ‘never say never’ policy is probably wise here, and I have already cautioned against concluding too quickly that such animals lack the concept of belief. But as things stand, there seems to be no evidence speaking positively in favour of the claim that they do grasp belief. However, I have contended that it is still far from clear that the available evidence favours the claim that chimpanzees possess the concept of knowledge, but not that of belief: firstly, because more deflationary explanations of the knowledge- and ignorance-attributing abilities shown by chimpanzees remain in play; and secondly, because we do not have a solid enough grip on what kind of evidence might justify us in granting that a creature has a particular mental concept to allow us to draw the conclusion Nagel wants with any confidence.

3.  Knowledge as a Precondition of Belief Nagel’s other argument for the priority of the concept of knowledge starts from a p ­ roblem raised by Apperly (2011: 118-9) concerning how mindreading is so much as possible. Apperly’s problem concerns the tractability of mindreading; given the sheer variety of alternative possibilities, how is it that we (and children of the right age) are able to ­recognize the correct response to the false belief test and other tasks involving the ­attribution of belief?30 Nagel’s response to this problem is also a development of Apperly’s: is of a piece with this, but she takes half a sentence from it out of context. Here it is in context: ‘Moves towards assessing children’s mindreading with a more diverse range of tasks are surely a good thing, but one thing this will not do is settle questions about when children really have a concept such as “belief ”. The same problem arises in the comparative literature, where there is good evidence that some non-human species “understand something about” seeing or knowing, but do not understand these mental states completely, and have no understanding of other mental states, such as believing. This evidence may very well be an accurate reflection of the abilities and limits of these species. The problem is that this undermines the utility of discussing mindreading concepts, as understood by the normative account of mindreading, because it is just not clear what it means to have half a concept of seeing or knowing, or half a “theory of mind” that includes some such mental states and not others.’ (Apperly 2011: 109). Apperly is no ally of Nagel here; rather, he is applying pressure to the entire framework within which she draws her conclusion.   Rose (2015: 396) makes a related point.   Nagel writes that the difficulty is in understanding how adult-level mindreading is ‘computationally possible, given the open-ended character of the information that might be task-relevant’ (2014: 299). I am now very doubtful that my earlier response to Nagel’s second argument (McGlynn 2014: 185) really engages the problem, so understood; that is one important respect in which I take the present chapter to supersede that discussion. Thanks to Tim Kenyon and the editors for pointing out some unclarities in this section in an earlier draft of this chapter. 29 30

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  91  . . . according to Apperly, we ordinarily construct simple and stereotyped models of situations involving other agents, drawing on our background knowledge of interactions like the ones we are observing [ . . . ] The default state of knowledge plays an important role in this process of situation modeling. What makes it reasonable to grade “Sally thinks the object is in the round box” as the right answer to the problem (or to attribute this mental state to Sally in a real-world version of the problem) is that this agent would know that the object is in the round box if things had gone for her as they normally do in a person’s momentary absence. If recently having seen something somewhere were not generally a reliable source of knowledge of its location, we would have no reason to attribute that particular belief to Sally. So a prior sense of how knowledge ordinarily arises helps to make belief attribution a tractable problem.  (2013: 299–300, footnote omitted)

As noted above, Nagel is here arguing for a stronger priority claim; the concept of knowledge is not only ‘first’ in the sense of being acquired earlier, but an ability to ­exercise it is a precondition of being able to successfully attribute belief. However, Nagel makes some remarkably strong claims in this passage, and as far as I can see, she does not offer anything in support of them. In particular, it is not clear why we should accept that ‘If recently having seen something somewhere were not generally a reliable source of knowledge of its location, we would have no reason to attribute that particular belief to Sally’. Why is it not enough to appreciate that a person will normally believe that an (inanimate) object is where they last saw it being placed, in the absence of reasons to think it has been moved? Why does one also need to grasp something about how ‘knowledge ordinary arises’? Nagel does not explain; it is just asserted.

4.  Concluding Remarks In this chapter, I have critically discussed Jennifer Nagel’s empirical case for the thesis that the concept of knowledge is prior to that of belief, and so that knowledge is a mental state in its own right. As I noted in the introduction, there are various points at which her argument might be questioned, but this chapter has pursued the strategy of arguing in detail that the empirical literature does not support the claim that the concept of knowledge is prior to that of belief, either in the sense that it is acquired first (and perhaps also last, in the case of some nonhuman animals), or in the stronger sense that possessing the former is a kind of precondition of possessing the latter. I have not argued for any particular alternative account of the relationship between the concept of knowledge and the concept of belief, but I have instead suggested that the currently available evidence is compatible with a range of different accounts that do not support Nagel’s priority claim. Relatedly, I have also tried to cast doubt on the idea that any of the mindreading tasks Nagel discusses offer anything like an ‘acid test’ for possession of a particular concept; with Apperly, I am inclined to think that there are not going to be sharp criteria here, and I take Nagel’s argument to be in trouble to the extent that it presumes otherwise.31 31   It is for the same reason that I’m sceptical of Rose’s (2015) claim that there’s empirical evidence for the priority of belief over knowledge.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

92  Aidan M c Glynn I certainly do not wish to rule out the possibility that a strong case for a knowledge first picture of the mind can be made on empirical grounds, and Nagel’s paper is highly suggestive in this respect. But where Nagel already finds evidence for the priority of knowledge, I find only a morass of fascinating but inconclusive studies which are open to various theoretical interpretations, and which have unclear bearing on issues of conceptual priority or the metaphysics of knowing. Given this, I think we should be sceptical that a powerful empirical case has been made for the claim that knowledge is a mental state in its own right, and against the claim that it is a species of belief.

References Apperly, Ian. 2011. Mindreaders: The Cognitive Basis of “Theory of Mind”. New York: Psychology Press. Bartsch, Karen and Henry Wellman. 1995. Children Talk About the Mind. New York: Oxford University Press. Burge, Tyler. 1979. Individualism and the Mental. Midwest Studies in Philosophy 4: 73–121. Buttelmann, David, Malinda Carpenter, and Michael Tomasello. 2009. Eighteen-Month-Old Infants Show False Belief Understanding in an Active Helping Paradigm. Cognition 112: 337–42. Butterfill, Stephen. 2013. What Does Knowledge Explain? A Commentary on Jennifer Nagel, “Knowledge as a Mental State.” In Tamar Szabó Gendler and John Hawthorne, eds. Oxford Studies in Epistemology, Volume 4. Oxford: Oxford University Press, 309–20. Butterfill, Stephen and Ian Apperly. 2013. How to Construct a Minimal Theory of Mind. Mind and Language 28: 606–37. Carruthers, Peter. 2011. The Opacity of Mind: An Integrative Theory of Self-Knowledge. Oxford: Oxford University Press. Carruthers, Peter. 2013. Mindreading in Infancy. Mind and Language 28: 141–72. Chandler, Michael, Anna Fritz, and Suzanne Hala. 1989. Small-Scale Deceit: Deception as a Marker of Two-, Three-, and Four-Year-Olds’ Early Theories of Mind. Child Development 60: 1263–77. Dretske, Fred. 1981. Knowledge and the Flow of Information. Oxford: Wiley-Blackwell. Fletcher, Logan and Peter Carruthers. 2013. Behavior-Reading Versus Mentalizing in Animals. In Janet Metcalfe and Herbert Terrace, eds., Agency and Joint Attention. Oxford: Oxford University Press, 82–99. Fricker, Elizabeth. 2009. Is Knowing a State of Mind? The Case Against. In Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 31–59. Gerken, Mikkel. 2012. Discursive Justification and Skepticism. Synthese 189(2): 373–94. Gerken, Mikkel. 2017. On Folk Epistemology. How we think and talk about knowledge. Oxford: Oxford University Press. Gerken, Mikkel. 2018. Against Knowledge-First Epistemology. In J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis, eds., Knowledge First: Approaches in Epistemology and Mind. Oxford: Oxford University Press, 46–71. Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23: 121–3. Goldman, Alvin. 2006. Simulating Minds: The Philosophy, Psychology, and Neuroscience of Mindreading. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

mindreading knowledge  93 Heyes, Cecilia. 2014a. False Belief in Infancy: a Fresh Look. Developmental Science 17: 647–59. Heyes, Cecilia. 2014b. Rich Interpretations of Infant Behavior are Popular, but are They Valid? A Reply to Scott and Baillargeon. Developmental Science 17: 665–6. Hogrefe, G.-Juergen, Heinz Wimmer, and Josef Perner. 1986. Ignorance versus False Belief: A Developmental Lag in Attribution of Epistemic States. Child Development 57: 567–82. Ichikawa, Jonathan Jenkins, and C. S. I. Jenkins. 2018. On Putting Knowledge ‘First’. In J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis, eds., Knowledge First: Approaches in Epistemology and Mind. Oxford: Oxford University Press, 113–31. Kaminski, Juliane, Josep Call, and Michael Tomasello. 2008. Chimpanzees Know What Others Know, but Not What They Believe. Cognition 109: 224–34. Kovács, Ágnes Melinda, Ernő Téglás, and Angsar Denis Endress. 2010. The Social Sense: Susceptibility to Others’ Beliefs in Human Infants and Adults. Science 24: 1830–4. Leite, Adam. 2005. On Williamson’s Arguments that Knowledge is a Mental State. Ratio 18: 165–75. Leslie, Alan. 1994. ToMM, ToBy, and Agency: Core Architecture and Domain Specificity in Cognition and Culture. In Lawrence Hirschfield and Susan Gelman, eds., Mapping the Mind: Domain Specificity in Cognition and Culture. Cambridge: Cambridge University Press, 119–48. McGlynn, Aidan. 2014. Knowledge First? Basingstoke: Palgrave Macmillan. Nagel, Jennifer. 2013. Knowledge as a Mental State. In Tamar Gendler and John Hawthorne, eds., Oxford Studies in Epistemology, Volume 4. Oxford: Oxford University Press, 273–308. Nagel, Jennifer. 2014. Knowledge: A Very Short Introduction. Oxford: Oxford University Press. Nichols, Shaun and Stephen Stich. 2003. Mindreading: An Integrated Account of Pretence, SelfAwareness, and Understanding Other Minds. Oxford: Oxford University Press. Onishi, Kristine H. and Renée Baillargeon. 2005. Do 15-month-Old Infants Understand False Beliefs? Science 308: 255–8. Perner, Josef. 1989. Is “Thinking” Belief? Reply to Wellman and Bartsch. Cognition 33: 315–19. Perner, Josef. 1993. Understanding the Representational Mind. Cambridge, MA: MIT Press. Perner, Josef. 2010. Who Took the Cog Out of Cognitive Science? Mentalism in an Era of ­Anti-Cognitivism.’ In Peter Frensch and Ralf Schwarzer, eds., Cognition and Neuropsychology: International Perspectives on Psychological Science (Volume 1). Hove and New York: Psychology Press, 241–61. Perner, Josef. 2014. Commentary on Ted Ruffman’s “Belief or Not Belief: . . . ” . Developmental Review 34: 294–5. Perner, Josef and Johannes Roessler. 2010. Teleology and Causal Understanding in Children’s Theory of Mind. In Jesús Aguilar and Andrei Buckareff, eds., Causing Human Actions: New Perspectives on the Causal Theory of Action. Cambridge, MA: MIT Press, 199–228. Povinelli, Daniel and Jennifer Vonk. 2003. Chimpanzee Minds: Suspiciously Human? Trends in Cognitive Science 7: 157-60. Povinelli, Daniel and Jennifer Vonk. 2004. We Don’t Need a Microscope to Explore the Chimpanzee’s Mind. Mind and Language 19: 1–28. Putnam, Hilary. 1975. The Meaning of “Meaning”. Minnesota Studies in the Philosophy of Science 7: 131–93. Roessler, Johannes. 2013. Knowledge, Causal Explanation, and Teleology. In Tamar Gendler and John Hawthorne, eds., Oxford Studies in Epistemology, Volume 4. Oxford: Oxford University Press, 321–2.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

94  Aidan M c Glynn Roessler, Johannes and Josef Perner. 2013. Teleology: Belief As Perspective. In Simon Baron-Cohen, Helen Tager-Flusberg, and Michael Lombardo, eds., Understanding Other Minds: Perspectives from Developmental Social Neuroscience. Oxford: Oxford University Press, 35–50. Rose, David. 2015. Belief is Prior to Knowledge. Episteme 12: 385–99. Ruffman, Ted. 2014. To Belief or Not to Belief: Children’s Theory of Mind. Developmental Review 34: 265–93. Scott, Rose M. 2014. Post Hoc Versus Predictive Accounts of Children’s Theory of Mind: A Reply to Ruffman. Developmental Review 34: 300–4. Scott, Rose M. and Renée Baillargeon. 2009. Which Penguin is This? Attributing False Beliefs About Object Identity at 18 Months. Child Development 80: 1172–96. Scott, Rose M. and Renée Baillargeon. 2014. How Fresh a Look? A Reply to Heyes. Developmental Science 17: 660–4. Shatz, Marilyn, Henry Wellman, and Sharon Silber. 1983. The Acquisition of Mental Verbs: a Systematic Investigation of the First Reference to Mental State. Cognition 14: 301–21. Smith, Martin. 2018. The Cost of Treating Knowledge as a Mental State. In J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis, eds., Knowledge First: Approaches in Epistemology and Mind. Oxford: Oxford University Press, 95–112. Surian, Luca, Stefania Caldi, and Dan Sperber. 2007. Attribution of Beliefs by 13-Month-Old Infants. Psychological Science 18: 580–6. Tardif, Twila and Henry Wellman. 2000. Acquisition of Mental State Language in Mandarin- and Cantonese-Speaking Children. Developmental Psychology 36: 25–43. Wellman, Henry. 1992. The Child’s Theory of Mind. Cambridge, MA: MIT Press. Wellman, Henry and Karen Bartsch. 1988. Young Children’s Reasoning About Beliefs. Cognition 30: 239–77. Wellman, Henry and David Liu. 2004. Scaling of Theory-of-Mind Tasks. Child Development 75: 523–41. Williamson, Timothy. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, Timothy. 2018. Acting on Knowledge. In J. Adam Carter, Emma C. Gordon, and Benjamin W. Jarvis, eds., Knowledge First: Approaches in Epistemology and Mind. Oxford: Oxford University Press, 161–81. Wimmer, Heinz and Josef Perner. 1983. Beliefs About Beliefs: Representation and Constraining Function of Wrong Beliefs in Young Children’s Understanding of Deception. Cognition 13: 103–28.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

5 The Cost of Treating Knowledge as a Mental State Martin Smith

1. Introduction My concern in this chapter is the claim that knowledge is a mental state—a claim that Williamson places front and centre in Knowledge and Its Limits (Williamson 2000). While I am not by any means convinced that the claim is false, I do think it carries certain costs that have not been widely appreciated. One source of resistance to this claim derives from internalism about the mental—the view, roughly speaking, that one’s mental states are determined by one’s internal physical state. In order to know that something is the case, it is not, in general, enough for one’s internal physical state to be a certain way—the wider world must also be a certain way. If we accept that knowledge is a mental state, we must give up internalism. One might think that this is no cost, since much recent work in the philosophy of mind has, in any case, converged on the view that internalism is false. This thought, though, is too quick. As I will argue, the claim that knowledge is a mental state would take us to a view much further from internalism than anything philosophers of mind have converged upon.1

2.  The Objection from Internalism The claim that knowledge is a mental state is a centrepiece of the knowledge first movement in epistemology—the first claim that one encounters in Williamson’s Knowledge and Its Limits (2000). While there might be various watered-down ways to interpret this claim, Williamson is clear that it should be taken in a literal, unflinching way. In his view, knowledge is as much a mental state as belief, desire, fear, hope, intention, etc. For me to know something—that it is raining outside, that per capita 1   This chapter was presented at the University of Edinburgh in November 2014 and greatly benefited from  the discussion on this occasion. Particular thanks go to Zoe Drayson, Jesper Kallestrup, Clayton Littlejohn, Heather Logue, and Aidan McGlynn. I am also grateful to Adam Carter, Emma Gordon, Ben Jarvis, and Keith Wilson for very helpful comments on earlier drafts of this chapter.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

96  Martin Smith greenhouse gas emissions are higher in Qatar than in any other nation, that the velocipede was invented in 1817—is just for my mind to be in a particular state. An objection immediately suggests itself: knowledge, unlike belief, desire, fear, hope, intention, etc., is factive—if I know that per capita greenhouse gas emissions are higher in Qatar than in any other nation, it has to be the case that per capita greenhouse gas emissions are higher in Qatar than in any other nation. But that is not a fact about my state of mind—it involves the wider world. One way to make this kind of objection more precise is by appealing to internalism about the mental. According to internalists, one’s mental states are determined by the internal physical state of one’s body and brain—any two individuals with the same internal physical state will have the same mental states. I might currently know that per capita greenhouse gas emissions are higher in Qatar than in any other nation, but I could be in the same internal physical state even if that were false, in which case I wouldn’t know it. Whether I know something is not, in general, determined by my internal physical state. If my mental states are determined by my internal physical state, as internalists would have it, then knowledge is not a mental state. We might call this the objection from internalism.2 One might think that this objection can be swiftly dismissed—after all, internalism is a view that has largely fallen from favour in the philosophy of mind, succumbing to a number of well-known arguments and thought experiments. In ‘Is knowing a state of mind?’ (1995) and in chapter 2 of Knowledge and Its Limits (2000), Williamson argues that, once we have given up the restrictive internalist picture of the mind, and embraced externalism, no meaningful impediment remains to treating knowledge as a mental state. As he puts it ‘An externalist conception frees one to affirm that knowledge is a mental state’ (Williamson 1995, 533). In this chapter I argue that treating knowledge as a mental state would, in fact, have us recoil far further from internalism than any of the well-known arguments and thought experiments take us. Much of this chapter consists of a survey of these arguments and thought experiments, along with a slightly new way of mapping out their purported externalist consequences. The survey will not be especially detailed or deep—just detailed enough to extract the consequences that concern me. I conclude by returning to the view that knowledge is a mental state, and showing that it takes us, in effect, into uncharted externalist territory.3 2   It is sometimes suggested that internalism should be reformulated in such a way as to include dualist theories on which one’s mental states are constitutively independent of the external world, but not determined by one’s internal physical state. I will not attempt such a reformulation here—but it is worth noting that a dualist of this kind would, I imagine, have just as much reason to resist classifying knowledge as a mental state as one who signs up to the materialist internalism under consideration here. 3  In ‘Is knowing a state of mind? The case against’ (2009), Elizabeth Fricker also takes issue with Williamson’s use of standard externalist arguments to soften one’s resistance to the idea that knowledge is a mental state, arguing that the acceptance of the former and rejection of the latter makes for a perfectly stable, defensible overall view. I agree with this, but aim, in a way, to go further. I will argue that there are in fact very strong reasons for otherwise committed externalists to stop short of granting that knowledge is a mental state.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   97 In the end, nothing that I say here will constitute a decisive refutation of the view that knowledge is a mental state. Whether we do decide to treat it as one will ultimately depend on how we weigh the costs and benefits of doing so—and I will not attempt to come to any overall judgment about this. What I do hope to show here is that the costs of treating knowledge as a mental state have, in at least one respect, been seriously underestimated. Before proceeding, it may be worth remarking briefly on what it means to deny that knowledge is a mental state. At a minimum, to deny this is to accept that a person’s mental states do not, on their own, settle what that person knows—that two individuals with the same mental states might have different knowledge, owing to differences in the worlds they inhabit. It is sometimes suggested that the alternative to treating knowledge as a mental state is to treat it instead as a metaphysical hybrid or composite of mental and non-mental factors (Williamson 2000, 5, Section 1.3, Nagel 2013, 275, Section 2). This, though, seems a very tendentious way of putting things—almost as though anyone who denies that knowledge is a mental state is committed to finding some analysis or factorization of knowledge into mental and non-mental components. But the denial that knowledge is a mental state can be combined perfectly well with doubts about whether it is analysable and, indeed, with general doubts about the project of philosophical analysis (Fricker 2009, Section 3, McGlynn 2014, 168). Many of Williamson’s arguments in the opening chapters of Knowledge and Its Limits are directed, first and foremost, against the possibility of analysing or factorizing knowledge. In spite of the way that Williamson sometimes presents things, these arguments cannot simply be co-opted to support the view that knowledge is a mental state.

3.  Natural Kind Externalism According to the causal theory of reference, the referents of certain proper names and common nouns are determined by the particulars and natural kinds that are causally related to their usage, and not by any ideas or descriptions that users associate with them. The causal theory attracted a number of supporters in the 1970s and 1980s due largely to the pioneering work of Kripke (1972) and Putnam (1975), and a corresponding view about the contents of beliefs followed in its wake—in order for one to hold beliefs about certain particulars or natural kinds, it is necessary that one have an appropriate causal relation to them. In Putnam’s classic ‘Twin Earth’ thought experiment, we are introduced to a remote planet, exactly like the Earth, but completely devoid of water—that is, H2O. Instead of water, a different chemical compound—XYZ—flows in the rivers, fills the lakes and oceans, falls as rain, and quenches the thirst of the Twin Earthers. XYZ has all of the same macroscopic properties as water—it is clear, odourless, boils and freezes at 100 and 0o C at sea level, etc. Furthermore, the Twin Earthers, who speak like us, even call it ‘water’ (and ‘eau’ and ‘mizu’, etc.). Finally, we might imagine that it is prior to 1750 and, without access to microscopes and the like, no one on Earth or Twin Earth would be

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

98  Martin Smith capable of distinguishing these two substances. According to Putnam, while those on Earth and those on Twin Earth might associate exactly the same ideas and descriptions with the term ‘water’, when the former use the term it refers to H2O, and when the latter use the term, it refers to XYZ. The lesson here is not just a semantic one, however—there are also implications for the contents of beliefs and other mental states (McGinn 1977, Burge 1979, n. 2, Putnam 1982, Ch. 1). When a person on Earth sincerely utters the sentence ‘there is water in my flask’ he expresses a belief that is true iff there is H2O in his flask. When his duplicate on Twin Earth utters the same words, he doesn’t express a belief about H2O—neither he, nor anyone on his planet, has ever come into contact with this substance. Rather, his belief is true iff there is XYZ in his flask. These two individuals may be in exactly the same internal physical state (at least if we ignore the fact that the human body contains water). Nevertheless, they hold beliefs with different contents—the former holds beliefs about H2O and no beliefs about XYZ, while the latter holds beliefs about XYZ and no beliefs about H2O. If we accept this verdict then we are led to a view we might call natural kind externalism—the content of certain beliefs involving natural kind concepts depends on what natural kinds are present in the believer’s environment. Similar thought experiments could doubtless be used to motivate similar conclusions about other kinds of mental states—desires, fears, hopes, intentions, etc.—but I will focus here on beliefs. There are, of course, a number of objections to natural kind externalism and to the standard verdicts about Twin Earth thought experiments—but I won’t explore them. While I am inclined to think that this view is broadly correct, I have no stake in defending it here—or in defending any of the various kinds of externalism I will ­canvass. My aim, rather, is to show that, even if all of these kinds of externalism be simultaneously accepted, the claim that knowledge is a mental state still represents a considerable further step in an externalist direction. If we accept natural kind externalism, then we will have to abandon the letter of internalism—one’s mental states will not be determined exclusively by one’s internal physical state. But whether natural kind externalism represents a major departure from internalism very much depends on one’s motivation for being an internalist in the first place. One might be moved to accept internalism as a result of philosophical theories that one holds—internalism may be entailed, for instance, by a number of once-fashionable, reductionist views in the philosophy of mind, such as the identity theory and classical functionalism (see, for instance, Smart 1959, Armstrong 1968). If this was one’s reason for accepting internalism, then perhaps it would have to be accepted to the letter—one could tolerate no divergence from it. But there are far more pedestrian thoughts that seem to point us in an internalist direction. Here is one: it shouldn’t be possible to influence or interfere with a person’s state of mind without causally interacting in some way with that person’s sense organs or body. If this was the reason one was inclined towards internalism, then one might be quite open to qualifying or compromising the view in various ways.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   99 Suppose a person on Earth was transported, without his knowledge, to Twin Earth. Natural kind externalism does not commit us to the view that this person’s beliefs would immediately change. On the contrary, those who have considered such a ­scenario tend to agree that the beliefs the exile expresses using the term ‘water’ would continue to be about H2O—at least for a while (see, for instance, Burge 1988, Boghossian 1989). If he arrives with his flask and says to a thirsty Twin Earther ‘there is water in my flask’, he would be expressing a true belief about H2O whereas, if the Twin Earther were to say ‘there is water in your flask’, he would be expressing a false belief about XYZ. The two would, at this point, be speaking slightly different languages, although neither would be in a position to notice the difference. However, after a substantial period of time has elapsed and the exile has had a good deal of causal ­interaction with XYZ—drinking it, bathing in it, etc.,—while his interactions with H2O recede into the distant past, it seems that he will have become, in all relevant respects, like a Twin Earther himself. Natural kind externalism predicts that, at some point, the contents of his beliefs will shift.4 This could happen even if the individual remains forever oblivious to his teleportation to Twin Earth. Let us introduce the term ‘switching’ to refer to a change in a person’s mental states that is not mediated by any change in the person’s internal physical state. Typically, when a person’s beliefs, desires, fears, hopes, intentions, etc., change, this will be the result of some new experience or of the acquisition of new information or of some conscious reasoning or unconscious processing, or of remembering something or forgetting something, etc. All of these processes involve very definite changes in a person’s sensory systems and brain. Switching is something altogether different—a change in a person’s mental states that does not involve any of these processes and does not, in effect, involve the person’s body at all. If a person’s mental states change while his internal physical state remains completely static then this will, of course, count as a case of switching—but, so too, will cases in which the changes in a person’s internal state are incidental or irrelevant to a given mental state change. Switching strikes us as a peculiar idea—it is strange, perhaps even disconcerting, to think that this is something that could actually happen to us. In the case just described, however, the mental states of the exile do switch—he loses his beliefs about H2O and acquires beliefs about XYZ. While he may undergo all kinds of changes in his internal physical state after his teleportation, the mental state shift is not due to any of these— they could have unfolded in exactly the same way if he had remained on Earth. Natural kind externalism does, then, allow for the possibility of switching—but the only way this can come about is via the substitution of one environment for another that contains a superficially indistinguishable natural kind, in such a way that an unwitting 4   Natural kind externalists may disagree about just how soon this shift can occur. Even if it takes some time for the exile’s ‘water’ beliefs to latch on to XYZ, one might hold that they disconnect from H2O— becoming indeterminate in content—shortly after arrival. Whatever view one adopts, the change will not be immediate—requiring, at the very least, that the exile’s ‘water’ beliefs come partly under the causal control of XYZ.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

100  Martin Smith believer remains oblivious to the change. As we have seen, such switching is slow—it takes a substantial period of time, after the substitution, for the switching to take effect. Furthermore, such switching is proximal—it involves changes to the natural kinds with which the believer is causally interacting during the period of the switch. The final thing to observe about this kind of switching is that it seems exceedingly contrived and farfetched (Warfield 1992). Given the way the world actually is, the conditions required for such switching are rarely, if ever, realized. At worst, then, natural kind externalism, as described here, lands us with the possibility of slow, proximal, rare switching. For an internalist motivated by an instinctive aversion to the idea of switching, the move to natural kind externalism need not seem like a major concession.

4.  Social Externalism The Twin Earth thought experiment purports to show that the contents of one’s natural kind beliefs can depend upon the nature of the world outside of one’s body. A few years after Putnam first described Twin Earth, another famous thought experiment, devised by Tyler Burge, purported to extend this conclusion to virtually all beliefs (Burge 1979, 1986). Burge has us imagine an individual who is unaware that arthritis is a condition of the joints and complains to her doctor of arthritis in her thigh. When she says ‘I have arthritis in my thigh’ she expresses a (necessarily) false belief. Now consider another individual, with exactly the same history and internal states, who happens to be part of a linguistic community in which the term ‘arthritis’ is used to refer to a broader class of rheumatoid ailments which happens to include the very complaint she has in her thigh. According to Burge, when this second individual says to her doctor ‘I have arthritis in my thigh’ she is not expressing a false belief about arthritis, a condition exclusively of the joints—neither she nor anyone in her community has beliefs about arthritis. Rather, she is expressing a true belief about the broader category that her linguistic community recognizes. If we accept this verdict, we are led to a view that we might call social externalism—the content of certain beliefs can depend upon facts about how terms are used in the believer’s linguistic community. Unlike the Twin Earth thought experiment, this thought experiment seems to provide a general recipe by which the content of almost any belief could be shown in principle to depend on facts about the wider world. There is nothing unusual about the term ‘arthritis’—if the meaning of this term, on the lips of one individual, can be shown to depend on how it is used in her linguistic community, then, presumably, the same could be shown about a great many of the terms that we use. When it comes to the meanings of our words, we are not the final arbiters—we are disposed to defer to the common usage of the linguistic community we are a part of. Burge himself takes his externalist conclusion to have a very broad scope and to extend to any belief involving theoretical, observational, natural, or artefactual kind concepts (Burge 1979, Section IIb).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   101 Social externalism may allow for a kind of mental state switching as defined in the last section—one could simply imagine a person travelling between the linguistic communities envisaged in Burge’s thought experiment. Once again, though, those who have considered such scenarios agree that this person would not immediately lose her beliefs about arthritis upon arriving in the new community. It is only after she has become a fully fledged member of the new community and has clocked up a good deal of causal and linguistic interaction with its members that the contents of these beliefs might eventually shift. And this shift could, of course, happen without it being due to any change in the individual’s internal state and without the individual ever realizing that the two communities use the word ‘arthritis’ in different ways. Once again, the only sort of switching that is in prospect here is switching that is slow, requiring a substantial period of time to elapse, and proximal, requiring changes to the linguistic community with which the believer interacts during this period. Interestingly, though, switching of this kind would not seem to require farfetched or contrived scenarios to bring it about. On the contrary, there need be nothing out of the ordinary about an individual travelling between subtly different linguistic communities all the while remaining oblivious to the differences between them. Indeed, as Ludlow (1995) has argued, this is something that may commonly happen—as when people move between Britain and the United States without realizing that words such as ‘chicory’, ‘jelly’, ‘pants’, etc., mean different things in British and American English. Consider an American who has just arrived in Britain and sincerely says ‘Chicory is rich in vitamin A’, thereby expressing a belief about cichorium endivia. Social externalism may suggest that, once she has lived in Britain for enough time, and become a part of the new linguistic community, this belief may be lost. Her disposition to assert ‘Chicory is rich in vitamin A’ could still be present, but these words will now express a new belief about cichorium intybus. Ludlow suggests that this kind of switching may even occur when people move across social groups that have their own internal conventions for the use of certain terms (Ludlow 1995, 46–49). Philosophers, for instance, use terms like ‘realist’ and ‘pragmatist’ in a specialized way that differs from general public usage. Perhaps someone who started socializing more and more with philosophers, while remaining ­unaware of this difference, would eventually come to express different beliefs with these words. There is, of course, room for reasonable disagreement amongst social externalists as to whether switching can happen quite so easily as this. There may, for instance, be versions of social externalism that place more emphasis on the original linguistic community in which a person acquired the use of a term and which, as a result, make the contents of that person’s beliefs far more resilient to these sorts of changes. However we fill in the details, though, social externalism will make switching into a more realistic prospect than natural kind externalism and thus, along one dimension at least, it represents a further step away from internalism.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

102  Martin Smith

5.  Demonstrative Externalism Natural kind externalism, as noted in Section 3, is closely associated with a view about how certain proper names and common nouns refer. A related view about demonstrative expressions—expressions such as ‘this’ and ‘that’ used to pick out objects being directly perceived—is connected with externalism of a third type. According to Kaplan (1979, 1989), Evans (1982, Ch. 6), and others, when one uses a demonstrative expression, one’s assertion does not merely have descriptive content—rather, its content may literally involve the perceived object being demonstrated. Suppose I see an apple on the table and, pointing towards it, utter the words ‘That apple is overripe’. On the present view, if it had been a different apple sitting on the table, then the content of my assertion would have involved a different object and, thus, would have been a different content, even if the new apple looked exactly the same to me and I was in exactly the same internal state. This may be a view about language but, once again, a view about the contents of beliefs is waiting in the wings (Evans, 1982, Ch. 6). As a general methodological rule, we should attribute to people beliefs with the contents of their sincere assertions and hold back from attributing to people beliefs with contents that they would not, or could not, sincerely assert. This forces us to conclude that, in the two scenarios just described, I hold different beliefs, despite being in the same internal state. In the first scenario, I  hold a belief about a particular apple. In the second scenario, I hold a belief about a second, different apple and no belief at all about the first apple, to which I am in no position to refer with a demonstrative. We might term this demonstrative ­externalism—the view that the contents of certain beliefs involving demonstrative concepts can depend upon the observed objects being demonstrated. My internal state could, of course, be the same even if there were no apple at all—if I was hallucinating, say. In this case, according to one version of demonstrative externalism, the utterance ‘That apple is overripe’ would have no content and would fail to express any belief. Although I’d be in the same internal state as someone who holds a genuine belief, I would be suffering from an illusion of belief (Evans 1982, Section 2.2, appendix to Ch. 6; McDowell 1986). Like social externalism, demonstrative externalism may allow for a new kind of mental state switching. For the demonstrative externalist, switching could, in principle, be accomplished by surreptitiously substituting objects within a person’s immediate vicinity. Suppose I am examining an apple on the table before me and am willing to assert things like ‘That apple is overripe’, ‘That apple is a Golden Delicious’, etc. Suppose that, while I am momentarily distracted, someone replaces the apple with another of the exact same size, shape, and colour. Even if I do not notice the subterfuge, according to the demonstrative externalist, my beliefs will change—while I remain willing to assert the very same sentences, they will now express beliefs with new contents. This change will not of course be due to any corresponding change in my internal physical state.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   103 One thing that is remarkable about this kind of switching is that it would appear to take place very fast. Demonstrative externalism seems to predict that the content of a demonstrative belief should change the very moment a perceptual, demonstrative connection to a new, substituted object is established. While this kind of switching may not require any farfetched mechanisms per se, it does require that circumstances conspire against a perceiver in a very particular way. For the switching to occur, it must be that a perceived, demonstrated object is substituted for another, indistinguishable object without this impinging in any way on the internal state of the perceiver. While such switching may occur in practice, we would expect it to be a rare event. Finally, the switching envisaged here would still appear to be proximal—after all, the substitution must occur within the range of the believer’s perceptual faculties, in order for the objects to be candidates for demonstration. One clarification is important however—in describing this sort of switching as proximal, I don’t mean to suggest that the substitution must necessarily take place close to the believer’s body. Suppose I am shown an apple on the other side of the world over Skype and utter those same words—‘That apple is overripe’. One might take the view that this apple is literally being perceived and that it enters into the content of my utterance and my belief. If so, then some appropriate sleight of hand on the other side of the world could elicit a fast mental state switch. This switching would still, however, be categorized as proximal—it is triggered by an event in a part of the world with which, however physically far away it might be, I am in intimate causal contact. The boundaries of a believer’s environment, in the relevant sense, aren’t drawn at some fixed distance from his body. The relevant notion of an environment is a causal one—it includes those parts of the world with which one is causally interacting in a relatively direct way. I have no interest in trying to specify precisely what a person’s environment includes—the notion is not meant to be a precise one. It is enough that there be clear cases of events that occur within a person’s environment and clear cases of events that occur outside of it, even if there are a great many borderline cases. It is interesting to note that, according to Evans, an important condition for an object to feature in the content of a demonstrative utterance is that one be capable of locating that object within egocentric space—the space of possibilities of bodily action (Evans 1982, Sections 6.2–6.4). This condition is not fulfilled in cases of the type just described, leading Evans to speculate that these kinds of cases should, perhaps, be assimilated to cases of ‘deferred ostention’ in which ‘this’ and ‘that’ arguably are used as shorthand for a description (Evans 1982, Section 6.2). If this is right then, while I’m Skyping, the apple on the other side of the world is not a part of the content of any of my beliefs. Rather, the belief I express will have a descriptive content and substituting apples will not affect a switch in my mental states. It may, indeed, be a general consequence of Evans’s view that switching of the present kind can only be brought about by events that take place close to a believer’s body—but I won’t investigate this further here. According to demonstrative externalism, as I’ve defined it, certain beliefs expressed using ‘this’ and ‘that’ depend for their content on one’s relations to the immediate

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

104  Martin Smith environment. In addition, demonstrative externalists often endorse a closely related view about certain beliefs expressed using indexical expressions such as ‘I’, ‘he’, ‘she’, ‘here’, etc. Evans (1982, Ch. 6, Appendix 6) describes a case in which a person and his duplicate on Twin Earth both think about their current location as being ‘here’. If both were to sincerely assert ‘It is sunny here’ then, on Evans’s view, they would express beliefs with different contents—contents that involve different locations—even though these locations appear exactly the same and the two individuals are in the same internal state. Evans also describes a kind of switching case in which a person confined to a hospital bed keeps muttering to himself ‘It was so hot here yesterday’. While the person takes himself to be lying still in the dark, his bed is actually in constant motion on silent, well-oiled wheels. When he repeatedly utters this sentence he thinks he is reiterating the same belief over and over—but, according to Evans, he is actually expressing a new belief each time, as his location will have changed (Evans 1982, Ch. 6, Appendix 6). This sort of switching can also, perhaps, be categorized as fast, proximal, and rare. The switching is taken to occur from moment to moment, in response to changes in the person’s immediate surroundings and, while this kind of set-up may be practically possible, it would also be exceedingly unusual.

6. Disjunctivism Demonstrative externalism predicts that there can be a mental difference between a person who is veridically perceiving the world and one who is hallucinating or taken in by an illusion, even though the two may be in exactly the same internal state. This claim is characteristic of another influential kind of mental state externalism—namely, disjunctivism about perception (see, for instance, Hinton 1967, Snowdon 1980, McDowell 1982). According to disjunctivists, seeing that something is the case, feeling (in the sense of touch) that something is the case, hearing (directly, and not in the sense of hearing testimony) that something is the case, etc., are all mental states in their own right—different to the mental states of merely seeming to see or hear or feel, etc., that something is the case, as when one suffers a hallucination or is fooled by an illusion. Further, disjunctivists typically insist that veridical perceptual states have no mental element in common with hallucinations and illusions—no ‘highest common factor’, as it is sometimes put. The category of ‘perceptual experience’, which is supposed to subsume all three types of state, is a heterogeneous or disjunctive one. Relatedly, disjunctivists often hold that, when one sees or feels or hears, etc., that something is the case, one’s mind reaches right out into the world and literally involves external objects and facts—and that this is a fundamental, defining feature of these mental states that is not shared by hallucinations and illusions. When it comes to the externalist character of disjunctivism, however, many of these further commitments are not needed. For externalism to hold, it is enough that there be some mental difference between veridical perception, hallucination and illusion, each of which can be

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   105 associated with the same internal state—it doesn’t matter exactly how the difference is explained or how profound it is taken to be. It is possible, in principle, for a veridical perception to become an indistinguishable hallucination or illusion, or vice versa, without there being any accompanying change in the perceiver’s internal state. In such a case, disjunctivism predicts that the perceiver will undergo mental state switching. Suppose I wander into a room and see that there is a blue cube sitting atop a white podium. Suppose that, in an instant, the cube is vaporized and a perfect hologram of the cube projected in its place. Suppose the change does not impact on my sense organs or body in any way. According to the disjunctivist, this event will trigger a mental state switch—I will have gone from seeing that there is a blue cube on a white podium to merely seeming to see that there is a blue cube on a white podium. The switching would appear to be fast—indeed, it would appear to be instantaneous. The moment the cube is destroyed, it is no longer the case that there is a blue cube on the podium, and I can no longer be said to see that there is a blue cube on the podium. If this kind of example is characteristic of these switching cases, then they will, clearly, be very rare. In fact, I think it is characteristic—for a veridical perception to seamlessly change into a hallucination or illusion, or vice versa, is an extraordinary occurrence. Finally, this switching must be proximal, for much the same reason as the switching permitted by demonstrative externalism must be proximal—it requires changes in the objects and states of affairs that one is perceiving. Natural kind, social and demonstrative externalists all emphasize ways in which the content of a given type of mental state may depend upon external factors. Disjunctivists, however, find a role for external factors in determining the type of mental state that bears a given content—the type of attitude one has towards that content. For this reason, it is sometimes described as an attitude externalism (Williamson 2000, Ch. 2). The view that knowledge is a mental state should also, of course, be characterized as a kind of attitude externalism in this sense. In fact, disjunctivism can provide a certain model for how to think about this view. The disjunctivist draws a mental contrast between successful states on the one hand, and failed or botched states on the other—even though the two kinds of state may be indistinguishable for the one who instantiates them. The successful states include seeing, hearing, feeling, etc., that something is the case, and the botched states include merely seeming to see or hear or feel, etc., that something is the case. One who treats knowledge as a mental state also draws a mental contrast between a successful state and a botched state that may be indistinguishable from the inside. On this view, the successful state is knowing that something is the case, and the botched state is merely believing that something is the case. On both views, the difference between the ­successful states and the botched states does not lie with non-mental factors—but with mental factors that one is not in a position to detect. For the disjunctivist, when one sees, or hears, or feels, etc., that something is the case, one’s mind reaches right out into the world and involves external objects and facts.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

106  Martin Smith If knowledge is a mental state then, presumably, it is also a state in which one’s mind reaches right out into the world (indeed much further out into the world—though more on this in Section 7). One who accepts that knowledge is a mental state might follow the disjunctivist example even further and insist that knowledge and mere belief have no mental element in common. Williamson, though, stops short of this, remaining open to the standard idea that knowledge involves belief and that belief is a mental state (Williamson 2000, Section 1.5). For Williamson, belief can serve, in one sense, as the ‘highest common factor’ of knowledge and mere belief (though he denies of course, that knowledge can be ‘factorized’ into belief and further components). If we do accept the existence of factive mental states such as seeing, hearing, and feeling that something is the case, it might seem like a small step to admit one more factive mental state—namely, knowing that something is the case. For Williamson, once we embrace factive mental states, knowledge naturally takes its place as the most general such state—the factive mental state that one is in when one is in any factive mental state (Williamson 2000, Section 1.4). As I explore in Section 7, focusing on mental state switching offers a rather different perspective on this.

7.  Counting the Cost I have examined a number of externalist views through the lens of mental state switching. Each of these theories allows for different kinds of switching possibilities, which I have mapped out relative to three dimensions—a speed dimension, a range dimension, and a frequency dimension. Table 5.1 presents the results thus far: Table 5.1  Switching possibilities Natural Kind Externalism Social Externalism Demonstrative Externalism Disjunctivism

Speed

Range

Frequency

Slow Slow Fast Fast

Proximal Proximal Proximal Proximal

Rare Common Rare Rare

How, then, should we fill in the row for the view that knowledge is a mental state? If the arguments of this final section are on the right track, then this view allows for a kind of switching that is unlike anything countenanced so far. I currently believe, and know, that my local fishmonger has a blue shopfront. Suppose that, while I am on a trip overseas, the shopfront is restyled and painted red. As soon as I return home and set eyes upon it, I will lose my belief that my local fishmonger has a blue shopfront. This, of course, will be a perfectly standard case of mental state change, in which my beliefs are revised in light of new information. But if knowledge is a mental state, then this is not the first time that the new paint job will have had an effect on my state of mind. Rather, this will have happened days or weeks

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   107 before, while my body was thousands of miles away and in no causal contact with the fishmonger—for this is the point at which I will have ceased to know that my local fishmonger has a blue shopfront. And this change, whenever it happens, won’t be due to any change in my internal state, which will continue exactly as it would have if the shopfront had not been touched. The switching that occurs here is fast—the moment the red paint is applied, and it ceases to be true that my local fishmonger has a blue shopfront, I can no longer know it. Furthermore, switching of this kind would presumably be common—the events described are in no way unusual. Already we find ourselves confronting something new—the possibility of fast, common switching. But what is perhaps most striking about the switching described here is that it is not proximal. When the switching takes place, it is triggered by events completely outside of my environment at that time— events in a part of the world that is both physically remote from my body and in no causal contact with it. Consider, in addition, the following example discussed by Jennifer Nagel (2013, 289): If knowledge is a mental state, then on the evening of 14 April 1865, a kind of mass mental state switching would have swept the globe. Horrified theatregoers who witnessed Lincoln’s assassination might have ceased to believe that Lincoln is president of the United States—but countless millions more will have ceased to know it, including people on the other side of the world to Washington.5 If knowledge is a mental state, then the range within which switching may be triggered will be effectively limitless. Consider a person stranded in a disabled space probe drifting in deep space billions of miles from the surface of the Earth and with no means of communication. If knowledge is a mental state, then this person can still have influence on the minds of friends, relatives, and acquaintances on Earth, and they can still have influence on his. If he shaves his head, for instance, then people on Earth may go from knowing that he has a full head of hair to merely believing it. And this change in their state of mind will presumably be simultaneous with the head shaving, even though it may take light several hours to travel the distance between Earth and the probe. The mental state switch will take place long before the information that it has taken place could, in principle, reach the Earth. It may even appear as though treating knowledge as a mental state leads to the possibility of instantaneous action at a distance, and thus serves to settle what would usually be regarded as a very controversial empirical issue. But this will only be so if we think of a person’s mental state switch as an event spatially located within that person’s body or brain—and this is presumably an unwelcome remnant of internalist thinking that should be given up. If we treat knowledge as a mental state, then a person’s mental state switch may be better thought of as an event that can take place at any distance 5   Since Lincoln did not die until the morning of the 15 April, one might argue that it was only at this point that he strictly ceased to be the president of the United States. I put this potential complication aside for present purposes. There are, in any case, countless further historical examples that we could consider instead of this one.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

108  Martin Smith from the person’s body and brain, or an event that can be spread out over a vast distance, or even an event that has no determinate spatial location whatsoever. At the very least then we can fill in the final row of the table (Table 5.2) as follows: Table 5.2  Switching possibilities Natural Kind Externalism Social Externalism Demonstrative Externalism Disjunctivism Knowledge is a mental state

Speed

Range

Frequency

Slow Slow Fast Fast Fast

Proximal Proximal Proximal Proximal Distal

Rare Common Rare Rare Common

Uniquely amongst the kinds of externalism I have considered, the view that knowledge is a mental state allows for fast, distal, common mental state switching. But this is not yet the end of the story. Social externalism, I claimed, would make switching a common occurrence. While I stand by this claim, it is important to keep it in perspective. The kind of switching that social externalism permits may be common relative to the kinds of switching permitted by natural kind and demonstrative externalism, but it is hardly something that we would expect to undergo on a daily basis. Indeed, depending on the precise version of social externalism being entertained, switching is something that may happen on only a handful of occasions during a lifetime. But how often does knowledge turn into ignorance? As I sit writing, it is easy enough to recall examples of this from my own recent past. This morning I believed, and knew, that my bin was on the kerb, having put it there myself. My belief persisted until, some hours later, I saw it by the gate. My knowledge, however, will have been lost some time prior to this when my neighbour, seeing that the rubbish had been collected, wheeled it in. This loss of knowledge, whenever it happened exactly, won’t have been accompanied by any change in my internal state. A week ago I believed, and knew, that my one-year-old daughter had four teeth. That belief persisted until yesterday, when I noticed that a fifth tooth had come through. But my knowledge will have been lost some time in the intervening week, without any corresponding change in my internal state. If knowledge is a mental state, then these will both have been bona fide mental state switches. Furthermore, mass switching events, like the one triggered by Lincoln’s assassination, would themselves be relatively common occurrences if knowledge were a ­mental state—an unexpected side-effect of a globalized culture obsessed with celebrity. The table above suggests that treating knowledge as a mental state would lead to common switching—but this, in truth, is to play down the consequences of the view. To do justice to it, what is needed is a further point on the frequency spectrum—rampant switching—where none of the other kinds of externalism take us.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   109 Table 5.2 may also mislead in another respect. To suggest that treating knowledge as a mental state allows for fast switching is, once again, to underestimate the consequences of the view. This claim might seem surprising—after all, I have used the term ‘fast’ to describe even cases in which switching is instantaneous, such as the switching permitted by disjunctivism. Presumably, then, ‘fast’ must cover the very extreme end of the speed spectrum—how could switching be any speedier than instantaneous? The idea that knowledge requires a kind of insulation or protection against the possibility of error is one that is widespread in contemporary epistemology. Suppose one believes P via method M at time t. Say that one’s belief is safe iff it could not easily have been the case that one falsely believed P via M at t. This is sometimes spelled out in terms of possible worlds: say that one’s belief is safe iff there is no close possible world in which one believes P via M at t and P is false. A number of philosophers, Williamson included, have been attracted to the idea that knowledge requires safety—in order for one to know P at time t, one must safely believe P at time t, in something like the sense just defined (Williamson 1992, 2000, Ch. 5; Sosa 1999a, 1999b, Pritchard 2005, Ch. 6; Smith, 2016, Ch. 6). With this in mind, return to the case of Lincoln’s assassination. Booth, allegedly, decided to wait until a particular line in the play had been spoken before opening fire. Arguably, he could have easily decided to shoot Lincoln a few minutes earlier than he did or, put differently, there are close possible worlds in which he does shoot Lincoln a few minutes earlier than at the actual world. If this is right then, on the evening of 14 April 1865, the belief that Lincoln is president of the US would have ceased to be safe at least several minutes before it ceased to be true. Several minutes before the assassination, one’s belief that Lincoln is president, while still true at the actual world, would be false at some close possible worlds at which it is held via its actual method. If knowledge requires safety, and is a mental state, then this is the point at which the mental states of people the world over would have switched—prior to the event that would appear responsible for the switch. What we seem to confront here is a case of backwards switching, where an event triggers a mental state switch in the past. One might, of course, resist this description of the case. Perhaps what these reflections about safety show is that the assassination itself is not what truly triggers the mental state switch—rather other events are responsible, such as Booth approaching Ford’s theatre, his crouching outside the presidential box, pistol in hand, etc. And these events, we might insist, really do take place before the mental state switch does. The trouble with this re-description, though, is that if Booth had not gone on to assassinate Lincoln, it is not clear that there would have been any mental state switching at all, even if these other events had still taken place. We might consider another case in which there are no obvious precursors to an event that renders a known proposition false. Suppose I decide, on a complete whim, to shave my head, purely to deprive my friends, relatives, and acquaintances of the knowledge that I have a full head of hair. If knowledge is a mental state, then my actions

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

110  Martin Smith will bring about a small spate of mental state switches. But when do these switches take place? Suppose the impulse to shave my head suddenly strikes me at 10 a.m., and I act on it immediately. This impulse could, however, have easily struck me at, say, 9:50 a.m. or 9:40 a.m., instead. As such, there are close possible worlds at which I shave my head at these earlier times. The switch must, then, have already occurred by 9:40 a.m.—before my action is performed, and before I even have any inkling of performing it. Once again, one could insist that there is some causal precursor of my action already simmering away in my brain by 9:40 a.m., and that this is what is truly responsible for the mental state switching. But, even if such a precursor really could be identified, it is not something that I have any awareness of or any control over—and yet it seems that I am the one responsible for switching the mental states of my friends and relatives. It seems that the switch is a direct result of my action and my decision, and not of some subconscious activity in my brain before the decision was even made. Although the switch might occur at 9:40 a.m., it occurs because of what I do at 10:00 a.m. One might attempt to avoid these results by joining Neta and Rohrbaugh (2004), Comesaña (2005), Baumann (2014), and some others in denying that knowledge truly requires safety. In order for the present point to stand, though, it is enough that a lack of safety preclude knowledge in the cases just described, whether or not it does so in all possible cases that we might imagine. Indeed, at a bare minimum, all that is really needed to construct backwards switching cases of the present sort is that a lack of safety sometimes precludes knowledge of a knowable proposition that can change its truth value over time. Whether someone can be described as knowing something, at a given time, can depend upon what happens at later times. This, in and of itself, is nothing remarkable— whether a decision can be described as life-changing, whether a coincidence can be described as fortunate, whether a wound can be described as mortal, whether a belief can be described as true, at a particular time, can all depend upon what happens at later times. The observation only becomes remarkable if we insist that knowledge is a mental state. For then, one’s state of mind at a given time can depend upon what happens at later times—as though one’s mind reaches out not only into the external world, but out into the future as well.6 Even after a person has died, it may be possible, for a short time at least, to exert an influence, by my actions, on what that person counted as knowing shortly before his death. Again, this need not, in and of itself, be particularly Earthshattering—but what of the thought that I can still exert an influence, by my actions, on this person’s mental states? 6   If, as most agree, it is possible to know contingent truths about the future, then treating knowledge as a mental state will lead immediately to the result that the nature of person’s mental states, at a given time, can depend on what happens at later times. I currently believe that the sun will come up tomorrow. Whether I currently know this depends, among other things, on what happens tomorrow. If knowledge is a mental state, then what mental states I am currently in depends on what happens tomorrow. To obtain this result, we need not engage in any substantial theorizing about knowledge. Some substantial theorizing is required though to generate cases of backwards switching, such as those described in the body text.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

the cost of treating knowledge as a mental state   111 Treating knowledge as a mental state may not lead to the possibility of instantaneous action at a distance, but it is in genuine danger of leading to a possibility that might be considered even more contentious—namely, backwards causation. Even if we deny that a mental state switch has any definite spatial location, it seems much more difficult to deny that it has a definite temporal location—that it takes place at a specifiable time. In this case, the preceding examples will genuinely involve effects that temporally ­precede their causes. If we accept that knowledge is a mental state, we will have to abandon internalism about the mental and embrace externalism instead. But the binary internalist/­ externalist dichotomy is a very skewed way of dividing up the underlying logical space. There is only one way to be an internalist, but one can be an externalist in many ways. In truth, there is a broad spectrum of possible views, with internalism occupying one extreme end and externalism per se covering everything else. Treating knowledge as a mental state doesn’t merely nudge us from the internalist end of the spectrum, as many of the best-known externalist arguments and thought experiments arguably do—it takes us close to, if not all the way to, the opposite extreme. As I have noted, none of this is to say that knowledge categorically must not be treated as a mental state. The picture of the mind that this view foists upon us is one that we could, no doubt, inure ourselves to. But it is not one that should be accepted lightly.

References Armstrong, D. 1968. A Materialist Theory of Mind. London: Routledge. Baumann, P. 2014. No luck with knowledge? On a dogma of epistemology. Philosophy and Phenomenological Research 89(3): 523–51. Boghossian, P. 1989. Content and Self-knowledge. Philosophical Topics 17: 5–26. Burge, T. 1979. Individualism and the mental. Midwest Studies in Philosophy 4: 73–121. Burge, T. 1986. Individualism and psychology. Philosophical Review 95: 3–45. Burge, T. 1988. Individualism and self-knowledge. Journal of Philosophy 85: 649–63. Comesaña, J. 2005. Unsafe knowledge. Synthese 146(3): 395–404. Evans, G. 1982. The Varieties of Reference. Oxford: Clarendon Press. Fricker, E. 2009. Is knowing a state of mind? The case against. In P. Greenough and D. Pritchard, eds., Williamson On Knowledge. Oxford: Oxford University Press, 31–59. Hinton, J. 1967. Visual experiences. Mind 76: 217–27. Kaplan, D. 1979. ‘Dthat’. In P. French, T. Uehling, and H. Wettstein, eds., Contemporary Perspectives in the Philosophy of Language. Minneapolis: University of Minnesota Press. Kaplan, D. 1989. Demonstratives’, in J. Almog, H. Wettstein, and J. Perry, eds., Themes From Kaplan. New York: Oxford University Press, 481–564. Kripke, S. 1972. Naming and Necessity. Oxford: Blackwell. Ludlow, P. 1995. Externalism, self-knowledge and the prevalence of slow-switching. Analysis 55: 45–9. McDowell, J. 1982. Criteria, defeasibility and knowledge. Proceedings of the British Academy 68: 455–79.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

112  Martin Smith McDowell, J. 1986. Singular thoughts and the extent of inner space. In J. McDowell and P. Pettit, eds., Subject, Thought and Context. Oxford: Clarendon Press, 137–68. McGinn, C. 1977. Charity, interpretation and belief. Journal of Philosophy 74: 521–34. McGlynn, A. 2014. Knowledge First? Basingstoke: Palgrave Macmillan. Nagel, J. 2013. Knowledge as a mental state. In T. Gendler and J. Hawthorne, eds., Oxford Studies in Epistemology, Volume 4. Oxford: Oxford University Press, 273–308. Neta, R. and G. Rohrbaugh. 2004. Luminosity and the safety of knowledge. Pacific Philosophical Quarterly 85(4): 396–406. Pritchard, D. 2005. Epistemic Luck. Oxford: Clarendon Press. Putnam, H. 1975. The meaning of “meaning”. Minnesota Studies in the Philosophy of Science 7: 131–93. Putnam, H. 1982. Reason, Truth and History. Cambridge: Cambridge University Press. Smart, J. 1959. Sensations and brain processes. Philosophical Review 68: 141–56. Smith, M. 2016. Between Probability and Certainty: What Justifies Belief. Oxford: Oxford University Press. Snowdon, P. 1980. Perception, vision and causation. Proceedings of the Aristotelian Society 81: 175–92. Sosa, E. 1999a. How to defeat opposition to Moore’ Philosophical Perspectives 13: 137–49. Sosa, E. 1999b. How must knowledge be modally related to what is known? Philosophical Topics 26: 373–84. Warfield, T. 1992. Privileged self knowledge and externalism are compatible. Analysis 52: 232–7. Williamson, T. 1992. Inexact knowledge. Mind 101: 217–42. Williamson, T. 1995. Is knowing a state of mind? Mind 104: 533–65. Williamson, T. 2000. Knowledge and Its Limits. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

6 On Putting Knowledge ‘First’ Jonathan Jenkins Ichikawa and C. S. I. Jenkins

1. Introduction There is a New Idea in epistemology. It goes by the name of ‘knowledge first,’ and it is particularly associated with Timothy Williamson’s book Knowledge and Its Limits (Williamson 2000). In slogan form, to put knowledge first is to treat knowledge as basic or fundamental, and to explain other states—belief, justification, maybe even content itself—in terms of knowledge, instead of vice versa. The idea has proven ­enormously interesting, and equally controversial. But foundational questions about its actual ­content remain relatively unexplored. We think that a wide variety of views travel under the banner of ‘knowledge first’ (and that the slogan doesn’t help much with  differentiating them). Furthermore, it is far from straightforward to draw ­connections between certain of these views; they are more independent than they are often assumed to be.1 Our project here is exploratory and clarificatory. We mean to tease apart various ‘knowledge first’ claims, and explore what connections they do or do not have with one  another. Section 3 provides a taxonomy, and Section 4 explores connections. The result, we hope, will be a clearer understanding of just what the Knowledge-First theses are. Section 5 concludes with some brief suggestions as to how we think the ­various theses might be evaluated. Section 2 begins with a brief summary of the historical context in which the knowledge-first programme arose. It provides insights into what exactly the programme is by disentangling elements of the tradition to which it is a reaction.

2.  A Quick History To begin, we want to offer a brief recap of a certain now-(in)famous portion of the ­history of epistemology. We imagine that the majority of this summary will be familiar to many of our readers; however, as our subsequent discussion will pick up various   The authors contributed to this chapter equally; they are listed alphabetically.

1

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

114  Jonathan Jenkins Ichikawa and C. S. I. Jenkins strands and themes from this section, it is important to describe the relevant background explicitly before proceeding. Our story is certainly an oversimplification in some ways, but it should be sufficient for our purposes in this chapter. During much of the twentieth century, it was widely held that analytic philosophy was in the business of finding analyses. We shall address the question of just what counts as an ‘analysis’ in a moment. For now, it is enough to note that the project of finding a correct analysis was often pursued by way of presenting putative sets of ­necessary and sufficient conditions for something. Analyses so expressed could then be criticized by identifying putative counterexamples, whether actual or merely possible. The canonical history of epistemology has it that, prior to 1963, it was widely assumed that the correct analysis of knowledge was the view now known as the ­‘standard’ analysis, or the ‘JTB’ analysis (for ‘Justified True Belief ’). The standard ­analysis of knowledge says that for any subject S and any proposition p, the following are individually necessary and jointly sufficient conditions for S knows that p: 1. p 2. S believes that p 3. S is justified in believing that p. Precursors of this view are sometimes attributed to Plato, who says in the Meno that knowledge is distinguished from mere true belief in being ‘tied down’, so that it cannot easily escape or be lost as mere true belief can. Views more closely resembling the above formulation of the JTB analysis are attributed by Edmund Gettier to Roderick Chisholm and A. J. Ayer.2 Gettier (1963) published a short paper in which he described some possible ­situations that he claimed were counterexamples to the sufficiency direction of the standard analysis of knowledge: that is, cases where conditions 1–3 are all met, but where S nevertheless does not know that p. The cases Gettier described were situations in which S held a justified true belief which was inferred from a justified false belief. These cases were widely accepted as cases of JTB without knowledge, and thus as ­providing a decisive refutation of the JTB analysis. In response to Gettier’s counterexamples, which involve S deriving a justified true belief from a justified false one, new analyses of knowledge were developed, specifically designed to block these counterexamples. These often took the form of ‘JTB+X’ analyses, the idea being that each of 1–3 in the standard analysis is a genuinely necessary condition on knowledge, but 1–3 are not jointly sufficient for knowledge (as ­demonstrated by the counterexamples), so a fourth condition needs to be added. Of particular note here is the ‘No False Lemmas’ (hereafter ‘NFL’) analysis (see e.g. Clark 1963): 2   Neither of those authors phrased their conditions for knowledge exactly this way. The now-usual f­ ormulation, in terms of justification, is so in large part because of Gettier (1963). As Dustin Locke reminded us, Kaplan (1985) argues that the canonical history is inaccurate for this reason.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  115 1. p 2. S believes that p 3. S is justified in believing that p 4. S’s belief that p is not inferred from any falsehood In response to the NFL analysis, further problem cases were developed, in which S’s  justified true belief is not inferred from a false belief but arrived at in some other way. For example, Carl Ginet’s now-famous fake barn case3 was offered as a  case of  ­justified true belief without knowledge, but one in which there is no false belief from which S’s belief in the target proposition is derived. In response, epistemologists attempted to articulate different ‘fourth conditions’, or to complicate their ­analyses further.4 New (putative) counterexamples continued to appear in response to the various proposed analyses of knowledge (see Shope 1983 for a detailed survey). Eventually, however, the whole project of proposing new analyses and then presenting new counterexamples to them was called into question. Linda Zagzebski (1994) influentially argued that counterexamples broadly similar to those of Gettier could be produced in response to any putative analysis of knowledge meeting certain conditions, and provided a recipe for the construction of such counterexamples. The conditions were that the proposal be of the form JTB+X, with X logically independent of truth. The recipe for producing a counterexample to such an analysis involves two steps: first, think of a case in which S’s belief in p is justified, and condition X is met, but p is false, then alter the case to one in which p is true but only by luck.5 Edward Craig put further pressure on the project of analysing knowledge, citing as evidence the failure (as he saw it) of the post-Gettier project (Craig 1999, 1). Craig argues that what he calls the intuitive extension of the concept of knowledge (i.e. the collection of cases to which the concept intuitively applies) is in serious tension or perhaps outright conflict with the intuitive intension of the concept (i.e. our intuitive ideas about what the criteria for knowing are). If this is so, he says, then the project of analysis is hopeless and should be replaced with a kind of ‘conceptual genealogy’, attempting to understand knowledge by first asking ‘what the concept of knowledge does for us, what its role in our life might be’, and then figuring out ‘what a concept ­having that role would be like’ (Craig 1999, p. 2).

3   The case was made famous by Goldman (1976). Goldman later attributed it to Ginet. See, e.g., Goldman (2009), 29. 4   In addition to these challenges to the sufficiency of JTB and the various JTB+Xs, some challenges were made to the necessity of the individual J, T and B conditions. In particular, some challenged the justification condition, and proposed new tripartite analyses with the justification condition replaced with something else. One well-known suggestion made by Alvin Goldman (see, e.g., Goldman 1967) replaced the justification condition with the condition that S’s belief that p was causally connected [in the right sort of way] to the fact that p; Goldman (1976) invoked a reliability requirement. Either of these can be interpreted either as a development of the justification condition or a replacement of it—see Kornblith (2008). 5   Zagzebski (1996) attempted a virtue-theoretic account that avoids the challenge.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

116  Jonathan Jenkins Ichikawa and C. S. I. Jenkins Developing the trend begun by Zagzebski and Craig, Williamson (2000) also expresses scepticism about the project of analysing knowledge into necessary and ­sufficient conditions. But Williamson finds a different moral in the failure (as he sees it) of the analytic project. Williamson diagnoses the problem as an error in order of explanation: where the traditional analytic project attempts to analyse knowledge into  component features (truth, belief, justification, etc.), Williamson argued that ­epistemologists should instead take knowledge as primitive, or basic, and use it to understand other things. This is the headline news of Williamson’s ‘knowledge first’ approach to epistemology. It is our purpose in this chapter to get a better sense about what being ‘first’ might amount to in such a context, and hence pin down a few possible things one might mean by calling oneself a ‘knowledge-first’ epistemologist. We shall investigate some of the things that Williamson himself might mean by ‘knowledge-first’, but our focus is not on exegesis; our aim is to get clear about the options. In order to do this, we now return to the deferred task of discussing exactly what an ‘analysis’ of knowledge, in the traditional sense, was supposed to be. One thing to note is that ‘analysis’ is not the only word used to describe the traditional project involving the pursuit of necessary and sufficient conditions. Sometimes the traditional proposals are referred to as ‘analyses’, but at other times they may be called ‘accounts’ or ‘theories’ of knowledge, with this sort of terminological difference typically going unremarked upon.6 When the project is explicitly described as ‘analysis’, it is sometimes said to be an analysis of knowledge, and other times an analysis of the concept of knowledge; again, this difference is not often remarked upon.7 When it was a concept being analysed, there would also often be talk of analysing the meaning of the word ‘know’, perhaps because it was assumed that the word meaning is identical with (or good evidence ­concerning) the concept.8 Gettier himself wrote of ‘definitions’ of knowledge. With the benefit of hindsight (and of, e.g., Kornblith 2007) we can see that it is highly problematic to assume that investigating one’s concept of knowledge is the same thing as investigating knowledge; in the same way, one can raise serious concerns about how much word meanings have to do with either knowledge or the concept of knowledge. Given these complexities, in interpreting this stage in the history of epistemology we should allow for the possibility that there was not really a single traditional project. It might be more accurate to say that multiple projects were being pursued within that tradition, but these were not on the whole clearly distinguished from one another. Be that as it may, whatever was going on, the search for necessary and sufficient conditions for the truth of S knows that p was part of it. But not just any old set of necessary and sufficient conditions would do. It was obviously not permissible to offer {S knows   See, e.g., p. 33 (‘analysis’) vs. p. 19 (‘account’) of Williamson (2000).   See, e.g., p. 5 vs. p. 8 of Williamson (2000). 8   See, e.g., Craig (1999, 1): ‘The standard approach to questions about the concept of knowledge has for some time consisted in attempts to analyse the everyday meaning of the word “know” and its cognates. Such attempts have usually taken the form of a search for necessary and sufficient conditions . . . ’ . 6 7

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  117 that p} as the relevant set of conditions; such ‘circular’ conditions were considered unacceptable. An obvious way to be circular in the relevant sense is to include knowledge explicitly in the proffered conditions, but accounts would also be subject to criticism if the conditions were couched in terms considered to be ‘less basic’ than knowledge itself. For example, analysing knowledge as ‘non-Gettiered justified true belief ’ is a ­non-starter, on the plausible assumption that we understand being ‘Gettiered’ only by reference to knowledge. What might basicness amount to here? One possibility is metaphysical priority, of a kind currently associated with notions like grounding, dependence, and the in virtue of relation. Another possibility is the kind of psychological priority we might associate with the idea that we apply the concept knows9 to a situation only via an application of  the concept believes. Yet another possibility is conceptual priority of the sort ­associated with conceptual composition (the idea of one concept being composed of others). There is also a kind of temporal priority to be considered; perhaps the order in which we learn things is relevant to what counts as ‘more basic’ than what. There may be interactions and interrelationships between these various kinds of basicness; for example, if we learn what belief is before we learn what knowledge is, perhaps that suggests that we apply the concept knows via applications of the concept believes. Be that as it may, the different kinds of basicness are different and we should not start out by either conflating them or assuming any deep connections between them (e.g., that belief will turn out to be ‘more basic’ than knowledge in either all of these senses or none of them). Knowledge, the knowledge first epistemologist will say, is not subject to traditional analysis in more basic terms; it is unsuitable for decomposition into more basic components; it should be treated as the explanans rather than the explanandum, and so on. It is not what the traditional project took it to be. But since there seem to be multiple different ‘basicness’ claims tangled together in the traditional project(s), we might correspondingly expect to see, not a single clearly identifiable knowledge first thesis, but a tangle of different claims. In his (2000), Williamson is clearly concerned at some points with the state of knowledge—is it a mental state?—at some points with the word ‘knows’—is it a factive mental state operator?—and at some points with the concept knows—is it  a  hybrid of a mental and a non-mental concept? Which of them is he saying is ‘first’, and in what sense? It is clear that Williamson at least intends to make some priority claims with respect to concepts. He often says that the orthodoxy he is challenging is the view that belief is ‘conceptually prior’ to knowledge (see e.g. pp. 2–3, 5, 8–10) and his discussion of what he sees as the ‘degenerating research programme’ of traditional analysis makes very clear that he thinks that programme is concerned with the analysis of concepts. 9   We’ll use small caps to indicate reference to concepts and ‘single quotes’ to indicate reference to words. We use words by typing them.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

118  Jonathan Jenkins Ichikawa and C. S. I. Jenkins Williamson writes that ‘one would not expect the concept knows to have a non-trivial analysis in somehow more basic terms’ (p. 31). But Williamson also makes claims about the word ‘knows’ that concern unanalysability into more basic things (or decomposability into parts). For example, in calling ‘knows’ a factive mental state operator, he means (among other things) to say that it is ‘an unanalysable expression’ in the sense that it is ‘not synonymous with any complex expression whose meaning is composed of the meanings of its parts’ (p. 34). Then again, he also makes some priority claims about knowledge itself (not the ­concept or the word), such as: ‘[k]nowing does not factorize’ (p. 33). He states that he aims to find an approach to epistemology on which knowledge is ‘central, and not ­subordinate to belief ’ (p. 5). And the thesis of his first chapter is explicitly that ‘[k]­nowing is a state of mind’ (p. 21). Other of Williamson’s relevant-looking claims are less easy to categorize as falling neatly into one of the mentioned camps, as (for example) when he says that one can ‘characterize’ knowledge ‘without reference’ to belief and ‘characterize’ belief ‘by reference’ to knowledge (p. 5). There is of course no contradiction in holding various ‘knowledge-first’ theses; it may be that Williamson shifts between the various claims mentioned here for the straightforward reason that he is interested in defending all of them. Nevertheless, it is philosophically worthwhile to articulate and consider the views independently. Our aim in this chapter is to do some work to untangle the various different notions of basicness or priority that could be in play, both in the traditional project(s) and, correspondingly, in the knowledge-first movement. All forms of knowledge-first epistemology are still new and controversial, and our project is not to defend or reject it in any of its versions, although we will indicate and motivate some of our sympathies along the way. We think that convincing defences or rejections of knowledge-first theses will only be ­possible in the light of clarification as to what is at stake in these debates.

3.  A Taxonomy of Priority As we see it, there are two families of notions of priority, which correspond to two families of versions of the traditional project, and hence to two kinds of ‘knowledgefirst’ views. We call them metaphysical and representational families of priority. We begin with the metaphysical families. As alluded to earlier, one sense in which one might think that, for example, belief is more basic than knowledge, is that knowledge facts obtain partly in virtue of belief facts, rather than vice versa. Several metaphysicians have argued in recent decades that there is a hierarchical structure to reality;10 such views are able to provide a sense of priority that can give content to the questions surrounding knowledge first epistemology. For example, according to one   Prominent examples include Fine (2012), Koslicki (2012), and Sider (2011).

10

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  119 metaphysical interpretation of the traditional project, epistemologists were attempting to account for the nature of knowledge by identifying metaphysically prior states (belief, justification, etc.) which ground states of knowledge. By contrast, representational families of priority locate the priority in question not in the metaphysics of knowledge itself, but in features of the way we think or talk about knowledge. For example, questions about whether one concept has another as a constituent are questions about priority in a representational sense. We are not sure how best to make sense of talk of whether ‘belief is conceptually prior to knowledge,’ which seems to straddle the families. Such language appears to use, rather than mention, the terms ‘belief ’ and ‘knowledge’, which is suggestive of the metaphysical family. But if ‘conceptual priority’ has to do with concepts, the use of this phrase suggests representational interpretations. We regard such statements as calling out for further interpretation.11 The metaphysical and representational families of priority claims are very different from one another, in at least two important kinds of ways. First, they have different subject matters: metaphysical priority claims consider how knowledge itself is related to states like belief, etc.; representational priority claims relate ways of thinking or talking about knowledge to ways of thinking or talking about other states. Second, methodologically speaking, they are answerable to significantly different kinds of considerations. The truth or falsity of metaphysical priority claims is a matter of how the world is, rather than how we represent it. By contrast, representational priority claims are tied up deeply with human cognition; they are answerable to psychology in a way that metaphysical priority claims are not.12, 13 Sections 3–4 consider what relations there might be between the two families. First, however, it is helpful to make further intra-family distinctions, to help clarify the kinds of views that are on offer. We begin with the metaphysical family.

11   Here is one interpretation of the language that might make good sense of some uses: it is a metaphysical claim, as the use of ‘belief ’ and ‘knowledge’ suggests, but it maintains that the metaphysical priority claim itself is a conceptual truth. The notion of conceptual truth need not be understood as a truth about concepts; it may rather be a matter of, for example, what Paul Boghossian calls epistemic analyticity: a sentence’s being such that ‘grasp of its meaning can suffice for justified belief in the truth of the proposition it expresses’ (Boghossian  2003). Whatever the merits of this notion, however, it does not seem a good ­candidate for making sense of Williamson’s discussion: Williamson holds that there are no conceptual truths in the sense of Boghossian’s epistemic analyticities (see Williamson 2006). Something more like Jenkins’s notion of conceptual truth, i.e., truth knowable through conceptual examination (see Jenkins 2008), may point to a more viable interpretation of Williamson, as it is unaffected by the arguments of Williamson (2006). 12   The particular cases of priority claims about knowledge are somewhat complicated by the fact that knowledge itself is tied up with psychology; one of the metaphysical views in the neighbourhood, for example, is that knowledge (itself) is a mental state. So, certain ways one might precisify the distinction in  play are unavailable—we cannot say, for instance, that the difference lies in whether any priority is mind-dependent. 13   On the view that concepts are abstract Fregean senses belonging to a ‘third realm’, concepts are ­perhaps not so directly answerable to psychology as they are on the view that concepts are mental representations. However, conceptual priority so understood would still be much more closely bound up with human cognition than metaphysical priority of a non-representational kind. Thanks to a referee.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

120  Jonathan Jenkins Ichikawa and C. S. I. Jenkins

3.1  Metaphysical priority One way to hold a metaphysical version of the Knowledge-First thesis is to consider knowledge to be a fundamental part of reality. Ted Sider (see e.g., Sider 2011) is a clear exemplar of a metaphysician discussing this sense of fundamentality (though not with respect to knowledge). According to Sider, there is, as an objective matter of fact, a structure to reality: some properties14 are more fundamental than others, and they are the ones that carve nature at its joints. The property of being an ­electron, for example, is purportedly more joint-carving than the property of being a blender. Here is an example of one knowledge-first thesis: knowledge is a joint-carving property. This kind of claim can come in a stronger or a weaker form. The strong form, ­suggested by a flat-footed reading of the claim that knowledge is first, would have it that knowledge is absolutely fundamental; it is among the most basic features of reality. This claim is very strong, and we see little to commend it. It entails, for example, that there can be no metaphysical reduction of the mental in terms of physical particles.15 While this might be true, it seems odd to think of it as a commitment of knowledge-first epistemology. The more moderate version of the metaphysical priority claim would treat knowledge as relatively fundamental; it would say that, although some properties—those at play in physics, perhaps—are more fundamental than knowledge, knowledge is more fundamental than any other epistemic properties (such as belief and justification). Here, then, are two metaphysical priority claims one might be interested in considering: M-1.  Knowledge is an absolutely fundamental feature of reality. M-2.  Knowledge is a relatively fundamental feature of reality. On some ways of understanding what metaphysical priority amounts to (including Sider’s), it is closely related to naturalness in approximately David Lewis’s (1983, 1984) sense. Thus, some ways of spelling out (M-1) and (M-2) would have them assert respectively that knowledge is a perfectly natural or a relatively natural property. A quite different notion of metaphysical priority, however, can be understood in terms of parthood: on this conception, parts are metaphysically prior to wholes. One reading of the traditional project is that knowledge has belief, etc., as constituents; in the same way that legs are parts of tables, belief and justification were thought to be parts of knowledge. So we can articulate a version of the knowledge first thesis that denies that knowledge has such parts: 14   Sider also discusses the fundamentality of things belonging to other ontological categories, but we shall restrict our attention to properties here. 15   This claim depends on the relatively uncontroversial assumption that knowledge is a state that is at least partly mental, but does not require commitment to the more controversial assumption that knowledge is a wholly mental state.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  121 M-3.  Knowledge does not have belief, etc., as proper parts. We pause to make two quick observations here. First, there is also a stronger knowledge-first thesis available in terms of parthood—namely, that states like belief in fact have knowledge as a proper part. This, however, does not seem plausible, since belief does not entail knowledge, but the existence of wholes does entail that of parts. Thus, we focus attention here only on (M-3), which is in effect more a denial of a ‘knowledge last’ claim than anything stronger. Second, parthood-priority and fundamentality-priority might be connected, but it  is not obvious that they are. Since fundamentality theorists have tended to treat small things (like physical particles) as fundamental, the two notions of priority may coincide, but we see no reason one must say this. (Northeast England is part of England, but it is not obvious that it comes closer to marking a joint in nature. For an extreme way of denying that parthood entails fundamentality, consider the priority monism of Schaffer 2010.) This question bears some relation to the question of whether knowledge is a mental state. M-4.  Knowledge is a mental state. At least on its face, (M-4) is not any kind of priority claim, but there may be some ­connections with priority. Williamson argues for (M-4), contrasting his view with one according to which knowledge is a hybrid state, consisting in a combination of mental and non-mental conditions. On the assumption that if knowledge is  not a mental state then it is such a hybrid state, denying (M-4) would require ­denying (M-3). Another strand in the debate that appears to be related to metaphysical priority is the idea that knowledge is causally efficacious, and, in particular, that it figures into the best causal explanations of significant amounts of behaviour. M-5.  Knowledge plays important causal roles. As before, although this is not itself a priority claim, comparison with the natural ­contrast case indicates why it would be associated with one. In many cases in which one both knows and acts, there is a dispute about whether it is one’s knowledge that explains the action, or one’s mere belief.16 Insofar as there is a sense in which relative fundamentality is associated with figuring as explanans rather than explanandum,17 the claim that knowledge does the work is tantamount to a kind of priority claim of knowledge over belief (at least in these instances).

  See e.g., Williamson (2000, 62–3), Nagel (2013, 282–3).   See e.g., Jenkins (2013), especially Section 3 (which surveys some of the connections between explanation and fundamentality in the extant literature on fundamentality). 16 17

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

122  Jonathan Jenkins Ichikawa and C. S. I. Jenkins

3.2  Representational priority We turn now to distinguishing a few different senses of representational priority. To say that there is a representational priority—for instance, of belief over knowledge—is not to make a claim about the world itself. Rather, it is to make a claim about the way we think about or talk about the world. For example, according to one interpretation of ‘conceptual analysis’, the aim of the post-Gettier literature was literally to analyse the concept knows—i.e., to identify its component parts. Just as to analyse a chemical compound is to discover the more fundamental elements of which it is made, so might one treat conceptual analysis as the attempt to discover, of a given concept, the more fundamental concepts of which it is comprised. One may read Gettier, then, as having refuted the view that the concept knows is identical with the compound concept justified true belief. One kind of statement of a ‘knowledge first’ thesis in this neighbourhood would deny that knows has conceptual constituents of the traditionally envisaged kind. For instance: R-1.  The concept knows does not have believes as a part. A stronger claim might also be made: instead of merely rejecting the traditional analysis project, one might reverse it, suggesting that a concept like believes is in fact a  complex concept that includes knows as a constituent. Unlike its metaphysical ­analogue, this does not strike us as clearly implausible—it would imply the interesting thesis that thinking about belief requires thinking about knowledge; it would not imply the obviously false thesis that belief entails knowledge. R-2.  believes has knows as a part. Another kind of stronger conceptual claim is this, more general, one: R-3.  knows does not have any other concept as a part. As with our list of metaphysical theses, these are meant to be representative, not exhaustive. (For example, one can also generate statements of representational knowledge first claims involving concepts like justification instead of belief, in the obvious ways.) Jennifer Nagel articulates one of the questions at issue in the knowledge first debate as (R-1), describing evidence she takes to bear on ‘whether the concept of knowledge is in some sense prior to the concept of belief, or whether it is composed from that concept and further conditions.’18 It is clear that (R-1) is among the claims Williamson is interested in defending.19 We are unaware of any serious defence of (R-2), although one could interpret certain remarks Williamson and other knowledge-first theorists have made along these lines.20   See Nagel (2013, 292).    19  See e.g., Williamson (2000, 33).   See e.g., Williamson (2000, 3): ‘Perhaps the inference from knowledge to belief derives from a conceptualization of belief in terms of knowledge . . . ’ , 18 20

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  123 Note that one could hold (R-3) on extremely general grounds. Fodor (1998), for example, argues that lexical concepts (approximately, concepts corresponding to single words in natural languages) quite generally fail to decompose. All lexical concepts, Fodor argues, are atomic. Since knows is a lexical concept, it has no conceptual ­constituents, so (R-3) is true. (Though note that Fodor’s view applies equally against (R-2).) Fodor’s famous claims here are highly relevant for the truth of any kind of ­representational priority claim in this neighbourhood, yet the knowledge first literature has not much engaged Fodor.21 It seems to us that either this has been a mistake, or representational priority claims like (R-1), (R-2), and (R-3) are not ultimately where the action is in the knowledge first debate.22 A different kind of representational priority claim concerns whether knowledge attributions involve other kinds of epistemic judgments, or vice versa. Jennifer Nagel articulates what a traditional (i.e., knowledge last) view of this kind would look like: [I]n attributing knowledge of a proposition to some target agent, we attribute the mental state of belief, while simultaneously judging the agent’s situation to satisfy various non-mental conditions.  (Nagel 2013, 282)

Nagel rejects this traditional view,23 writing that ‘in attributing [knowledge] to others we would not ordinarily have to pass through intermediate stages of attributing the weaker mental state of belief . . . ’ (p. 284). Here, then, is another pair of knowledge first theses: R-4.  Knowledge ascriptions need not involve tacit belief ascriptions. R-5.  Belief ascriptions are parasitic on our capacity for knowledge ascriptions. In support of both claims, Nagel cites evidence from developmental psychology and non-human primate studies that suggests that belief attributions are in various ways more cognitively difficult than knowledge attributions. These claims fit somewhat naturally with this one, which Nagel also defends: R-6.  Children learn the concept knows before the concept believes. There are many other representational knowledge-first theses that we could mention, but these should suffice for now. Some intra-family connections seem plausible here; we can see a common-sense empirical case for expecting (R-6) to go along with (R-4) and (R-5)—and it looks like (R-2) at least will require (R-6).   Williamson (2000, 31, n. 4) briefly mentions Fodor as relevant.   Quassim Cassam (2009, 19–20) complains that Williamson attacks a straw man in defending (R-1): ‘We have so far found no clear evidence either that analytic epistemologists who rely on conceptual analysis to establish necessary and sufficient conditions for propositional knowledge actually think that concepts like knows and has a justified true belief are identical or that they are committed to thinking this. Why, then, does Williamson make so much of the issue of concept identity . . . ?’ 23   One can find something like a statement of this traditional view in this passage from Jay Rosenberg (2002, 1–2): ‘We correctly judge that S knows that p whenever, from our de facto epistemic perspective, we judge S able adequately to justify his belief that p.’ 21 22

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

124  Jonathan Jenkins Ichikawa and C. S. I. Jenkins But what of inter-family connections? Are there logical or evidential connections to be drawn between metaphysical and representational priority views? We turn now to this question.

4.  Connections Between Families? Knowledge-first theses tend to be held or denied together. As we have been emphasizing, for instance, Williamson is clearly interested in defending views from both families in Knowledge and Its Limits. But how natural a unity do they form? Is there any reason one should assume that metaphysical knowledge first theses should stand or fall together with representational ones? As we will explain further on, we’re not sure. We are open to the idea that there might be reasons to accept views from both families or neither; later in this section we will point to some possible ways we can see arguments to this effect being attempted. We are rather confident, however, that there is no obvious connection between metaphysical and representational knowledge first theses. And this does seem to put us at odds with at least some epistemologists who assume such connections in their engagements with knowledge first epistemology. Consider, for example, the fuller context of a passage from Jennifer Nagel quoted in Section 2.2: Knowledge, in [Williamson’s] view, is a stronger mental state than belief, but it is nevertheless a purely mental state in its own right, so in attributing it to others we would not ordinarily have to pass through intermediate stages of attributing the weaker mental state of belief and also recognizing the believed proposition as true.  (Nagel 2013, 184, our emphasis)

Nagel’s use of ‘so’ links a statement of our (M-4) to one of our (R-4). She does not remark on the inference, apparently considering it obvious. But we see no obvious ­justification for it. As we suggested above, we are reasonably comfortable with the move from (M-4)—knowledge is a mental state—to (M-3)—knowledge does not have belief as a component. But we don’t see any reason to assume that the latter metaphysical claim should justify any claim in particular about how knowledge attributions made by humans work. We see no contradiction in the idea that knowledge does not have belief as a component, but that nevertheless, we only detect its presence by means of a step involving recognition of belief. After all, we only detect the presence of the measles virus by means of a step involving symptoms or antibodies, but that doesn’t mean those symptoms or antibodies are components of the virus. Maybe there is something more specifically untenable about the combination of (M-4) and the denial of (R-4), but Nagel doesn’t tell us what it is. Elizabeth Fricker also assumes a connection between metaphysical and conceptual knowledge first theses. Fricker writes: Williamson maintains that ‘knows’ has no analysis ‘of the standard kind’—this being one that factors knowing into a conjunction of mental and non-mental components, notably the mental

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  125 state of (rational) belief plus truth and some other factors. Call this thesis NASK. If NASK were false, ‘know’ having an a priori necessary and sufficient condition in terms of belief plus some other (non-factive) mental and non-mental components, this would establish the falsity of KMS [‘knowledge is a mental state’]: knowing would be revealed a priori to be a conjunctive ‘metaphysically hybrid’ state.  (Fricker 2009, 33)

And she adds in a footnote: Here I follow Williamson in ruling out the possibility of an error theory—that our concept ‘knows’ could be complex, while it in fact denotes a simple state. It is doubtful whether this is even coherent, and it can surely be discounted.

Fricker here seems to be arguing that denying (R-1) requires denying (M-3); she calls a view according to which knows has believes as a component, but where knowledge does not have belief as a part but merely entails it, an ‘error theory’ of dubious coherence. But why? It is not true in general that features of concepts must be shared by the properties to which they refer. (The concept pink shirt has no colour.) Why, then, is Fricker so convinced that knows could not be a complex concept while denoting a simple state? A complex concept like carrie’s favourite property can denote a simple property, after all: Carrie’s favourite property might be identity, or ­electronhood (or any property that the reader thinks is simple). Conversely, simple concepts can apparently denote complex referents: a concept like . . . is water can denote the complex property of being H2O. If there is some particular reason why knows could not work like that, we are owed an argument. We do think it is possible to identify one particular kind of combination of claims, in the vicinity of those Fricker discusses, that one cannot coherently make. However, this only works if one takes concept-identity talk to be literally a claim of numerical identity. Here is a toy model illustrating the kind of thing we mean. Suppose for the sake of argument that there are no Gettier cases—that there is no possible case of justified true belief without knowledge. Suppose also that the concept knows is in fact identical to the complex concept justified true belief. One cannot coherently say: 1.  The concept knows is identical to the complex concept justified true belief. 2.  Knowledge is not identical to justified true belief. The problem with that particular combination is that, on the assumption that the concept identity in claim 1 is in fact true, the thought expressed by claim 2 is the same thought as: 3.  Knowledge is not identical to knowledge. And 3 is, of course, false. But the fact that one cannot coherently accept 1 and 2 is an almost trivial, and quite specific, consequence about what a concept identity claim would commit one to. Significantly, this does not earn us a right to say anything more general about whether complex concepts can denote simple states. The only way we can make a move like the

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

126  Jonathan Jenkins Ichikawa and C. S. I. Jenkins one from 2 to 3 is to assume, not just that knows is a complex concept, but that it is ­literally identical with some particular complex of other concepts. Another statement of a possible connection between the conceptual and the ­metaphysical readings of knowledge first comes from Johannes Roessler: Mentalism is a claim about the state of knowing, not about the concept of knowledge. You might wonder, therefore, whether as a metaphysical thesis, it may not simply be neutral on ­conceptual issues. This would be a mistake, though. Mentalism arguably does commit one to the rejection of the traditional priority view. Crudely put, the idea is this. Suppose believes is analytically prior to knows. This would imply that there is a successful reductive analysis of knows in terms of the more basic concepts believes, is true, and so on. These more basic concepts would, in turn, enable us to identify the circumstances whose obtaining would be sufficient for the state of knowing. So knowledge would turn out to be a ‘composite’ state, involving among other things, the mental element of believing that p and the (typically) non-mental element of p being true. If this were so, believing that p would be a kind of mental state that’s necessary for knowing that p, but there would be no kind of mental state that is necessary and sufficient for knowing that p. So knowing would not be a mental state.  (Roessler 2013, 322–23)

It seems that Roessler is here proposing that (M-4), via something like (M-3), entails something like (R-1). It seems to us, however, that the reasoning offered is suspect. At one point this inferential move is made: a. There is a successful reductive analysis of knows in terms of more basic concepts, which enable us to identify the circumstances whose obtaining would be sufficient for the state of knowing. b. So, knowledge is a ‘composite’ state, involving among other things, the element of believing that p and the element of p being true. But merely identifying sufficient circumstances for a state to obtain via a conceptual analysis does not, we contend, by itself tell us much about what the state in question is like. In particular, it does not by itself tell us whether the state is composite (let alone of what it is composed). To demonstrate this, note that a conceptual analysis of carrie’s favourite property might enable us to identify sufficient conditions for being Carrie’s favourite property: it is sufficient that something be (1) a property, and (2) favoured by Carrie above all other properties. But that property itself may be as simple and non-composite as the reader cares to imagine. We do not agree with Nagel, Fricker, and Roessler that one can straightforwardly infer metaphysical priority claims from representational ones, or vice versa. We do not rule out the idea that there may be connections, but as far as we can tell, no interesting connections have so far been established. This represents a limited point of agreement with a line famously advanced by Hilary Kornblith (2007); we close this section with a brief consideration of this connection. Kornblith’s discussion of knowledge is not tied directly to knowledge first epistemology, but he is one of the authors who has written most explicitly about the relationship between metaphysical and representational projects in epistemology. He

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  127 holds the rather strong view that epistemologists should not take any interest in the concept knows, but should just investigate the state of knowledge. The proper concern of epistemology, Kornblith writes, is knowledge itself. And the concept knows is not even interesting as a guide to the state, according to Kornblith. Among the reasons he offers for this line is that the concept knows refers to a natural kind, and natural kinds, he thinks, are just not the sorts of things about whose natures we can learn anything substantive through conceptual investigation. (He thinks that empirical enquiry is the appropriate methodology.) For those not already convinced that knowledge is a natural kind and that concepts provide us with no way to learn anything substantive about natural kinds, he adds: It really doesn’t matter, for present purposes, whether knowledge and other targets of philosophical analysis are natural kinds . . . . [E]ven if the topics of philosophical interest typically correspond to . . . socially constructed kinds, it remains true that the concepts of the folk, and the concepts of philosophers as well, need not accurately characterize these socially constructed categories . . . . And once we recognize that our concepts, whether the concepts at issue are those of the folk or of theoreticians, may fail to characterize the categories they are concepts of, the philosophical interest of our concepts thereby wanes.  (2007, 36–7)

In this passage, Kornblith argues that because concepts can mislead us as to the true nature of what they represent, we cannot even use the concept knows as a guide to the state of knowledge. The conclusion is that epistemologists should simply forget about studying the concept knows altogether: [O]ur concepts are not plausibly viewed as the target of philosophical understanding . . . it is the extra-mental phenomena themselves which are the real targets of philosophical analysis: knowledge, justification, the good, the right, and so on, not anyone’s concepts of these things. (2000, 46).

Kornblith is in agreement with us, then, that there is no straightforward route from representational claims to metaphysical ones. However, we see two central points where Kornblith’s approach diverges from ours. First, we do not assume that it is metaphysical, rather than representational, questions that are central to epistemology. There is a clear tradition of epistemological investigation of the nature of the concept knows, the semantics of ‘knows’, and the psychology of knowledge ascriptions. We are pluralists about epistemology; we see no obvious reason not to think these are of significant epistemological interest in their own right, alongside questions about knowledge itself. Second, where we said only that we see no obvious connection between representational and metaphysical claims, Kornblith takes himself to argue that there couldn’t be any. We do not accept Kornblith’s argument. As one of us has pointed out elsewhere,24 Kornblith’s argument from the possibility of mischaracterization is not compelling, because the mere possibility of error is not in   See Jenkins (2014).

24

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

128  Jonathan Jenkins Ichikawa and C. S. I. Jenkins general a good reason for scepticism. That is to say, the mere possibility that knows may  fail to reveal the true nature of knowledge does not show that it doesn’t do so.  Perhaps some version of Kornblith’s conclusion can be derived with the aid of ­additional premises to the effect that (1) knowledge is a natural kind and (2) nothing substantive can be learned about natural kinds by examining concepts. But those are strong and controversial premises. None of this shows that Kornblith’s position is wrong; perhaps the concept knows tells us nothing of substance about knowledge. But if so, some argument for that claim is needed.

5.  The Way Forward We conclude with some brief suggestions about how one might investigate the various knowledge first claims we have been articulating. What considerations might one look to in favour of or against each view? We begin with (M-2), which is the thesis that knowledge is relatively metaphysically fundamental, and with a kind of consideration similar in spirit to certain arguments made by Williamson (2000, 62–3). Suppose appeals to knowledge appear in a lot of the best explanations we can give of human action, and that we cannot explain the same actions comparably well by mentioning only belief (or true belief, or justified true belief, etc.). If knowledge has some such important role to play in our theorizing about observed phenomena, this may provide some support for (M-2), at least in the presence of further assumptions about the connections between fundamentality and theoretical roles. If, for example, we agree with Lewis (1984) that the perfectly natural (or fundamental) properties are the ones that appear in the basic axioms of a minimally adequate theory of the world, we might also take it more generally that figuring in somewhat simple yet rather informative theories evinces relative naturalness or fundamentality. The same sort of claims could be supported if instead of (or as well as) appearing in good explanations of action, knowledge turned out to play multiple important roles in our other theorizing. Still within the realm of human action, we might think here of  the knowledge-action norms defended by (Hawthorne and Stanley  2008, Fantl and McGrath 2009). More broadly, we might think of the idea that knowledge is a ­constitutive norm of assertion (Williamson 2000, 238–69), the Williamsonian proposal that the best way to understand evidence is simply to identify it with knowledge (Williamson 2000, 184–208), and the idea (hinted at by Williamson and developed in Bird (2007) and Ichikawa (2014)) that, in a total reversal of the aims of the traditional project, justification might be analysed in terms of knowledge.25 Turning to the other core metaphysical claims, as we mentioned in Section 2.1, we do not imagine there is much support for (M-1), the view that knowledge is absolutely   Ichikawa (2017) defends (M-2) along these kinds of lines.

25

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  129 metaphysically fundamental. But we, of course, invite metaphysicians of knowledge to prove us wrong about the plausibility and interest of this strong thesis. (M-3), the claim that knowledge does not have belief, etc., as proper parts strikes us as more likely to be defended by actual philosophers than (M-1). However, any serious defence of (M-3) has its work cut out; it will need to include at minimum a robust metaphysics of knowledge (are we talking about a property or relation? a kind of state? or what?) and then adduce some appropriate and compelling notion of parthood for things of that kind (do properties literally have parts?), before finding specific reason to discount the obtaining of a parthood relation between knowledge and (say) belief. Another possible strategy for those interested in defending (M-2) or (M-3), or both, would be to argue for a corresponding claim about concepts or other representations (to which we will turn in a moment) and then further argue that we should expect our concepts and the world to line up in the relevant respects (contra e.g., Kornblith 2007; see Section 3 in this chapter). We do not at the moment have any suggestions about how to argue for the necessary linkages, however. On the representational side, if one buys Fodor’s view about the atomic nature of lexical concepts, one gets (R-1)–(R-3), which are all about the rejection of conceptual parthood relations, for free. Independently of Fodorian premises, however, considerations in favour of specific theses from this group might be based upon conceptual analysis if one believes in such a thing: for example, if one thinks it is possible to tell by conceptual analysis, or through the use of intuition or what have you, that the concept knows does not have concept believes as a part (R-1), one has support of a kind for (R-1) whether or not one is a Fodorian about concepts in general. Yet another route to these theses about conceptual parthood might proceed via some of the other representational knowledge first thesis about (say) the order of ­conceptual acquisition: for example, if we established (R-6), that children learn the concept knows before the concept believes, we might argue that this at least suggests (and maybe even entails) that (R-1) is true, i.e. that the concept knows does not have concept believes as a part. Similarly, (R-4), the claim that belief ascriptions are parasitic on our capacity for knowledge ascriptions, might be thought of as at least suggestive of the truth of (R-1). As for such psychological questions as whether children learn the concept knows before the concept believes, or vice versa, it seems to us pretty clear that the role of armchair epistemological theorising is extremely limited. These are claims that are properly in the domain of empirical psychology; armchair philosophical reflection should be informed by, but cannot replace, such enquiry.26

26   Thanks to Dustin Locke, John Gregg, and Lisa Miracchi for detailed comments on a draft of this c­ hapter. An ancestral of this paper was presented at the ‘Gettier Problem at 50’ conference in Edinburgh in 2013; thanks to Jennifer Nagel, Timothy Williamson, and all the other participants there for helpful feedback. Thanks also to audiences at University of Waterloo and University of California, Davis, where a version of this paper was also presented, and to three anonymous commentators on the penultimate draft.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

130  Jonathan Jenkins Ichikawa and C. S. I. Jenkins

References Bird, Alexander. 2007. Justified Judging. Philosophy and Phenomenological Research 74(1): 81–110. Boghossian, Paul. 2003. Epistemic Analyticity: A Defense. In Hans-Johann Glock, Kathrin Glür, and Geert Keil, eds., Grazer Philosophische Studien: Fifty Years of Quine’s Two Dogmas. Leiden: Brill, 15–35. Cassam, Quassim. 2009. Can the Concept of Knowledge be Analysed? In Patrick Greenough, Duncan Pritchard, and Timothy Williamson, eds., Williamson on Knowledge. Oxford: Oxford University Press, 12–30. Clark, Michael. 1963. Knowledge and Grounds: A Comment on Mr. Gettier’s Paper. Analysis 24(2): 46–8. Craig, E. 1999. Knowledge and the State of Nature: An Essay in Conceptual Synthesis. Oxford: Oxford University Press. Fantl, Jeremy and Matthew McGrath. 2009. Knowledge in an Uncertain World. Oxford: Oxford University Press. Fine, Kit. 2012. Guide to Ground. In Fabrice Correia and Benjamin Schnieder, eds., Metaphysical Grounding: Understanding the Structure of Reality. Cambridge: Cambridge University Press, 37–80. Fodor, Jerry A. 1998. Concepts: Where Cognitive Science Went Wrong. Oxford: Oxford University Press. Fricker, Elizabeth. 2009. Is Knowing a State of Mind? The Case Against. In Duncan Pritchard and Patrick Greenough, eds., Williamson on Knowledge. Oxford: Oxford University Press, 31–60. Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23(6):121–3. Goldman, Alvin I. 1967. A Causal Theory of Knowing. Journal of Philosophy 64(12): 357–72. Goldman, Alvin I. 1976. Discrimination and Perceptual Knowledge. Journal of Philosophy 73(November): 771–91. Goldman, Alvin I. 2009. Williamson on Knowledge and Evidence. In Patrick Greenough, Duncan Pritchard, and Timothy Williamson, eds., Williamson on Knowledge. Oxford: Oxford University Press, 73–91. Hawthorne, John and Jason Stanley. 2008. Knowledge and Action. Journal of Philosophy 105(10): 571–90. Ichikawa, Jonathan J. 2014. Justification is Potential Knowledge. Canadian Journal of Philosophy 44(2): 184–206. Ichikawa, Jonathan J. 2017. Contextualising Knowledge: Epistemology and Semantics. Oxford: Oxford University Press. Jenkins, C. S. I. 2008. Grounding Concepts: An Empirical Basis for Arithmetical Knowledge. Oxford: Oxford University Press. Jenkins, C. S. I. 2013. Explanation and Fundamentality. In Miguel Hoeltje, Benjamin Schnieder, and Alexander Steinberg, eds., Varieties of Dependence (Basic Philosophical Concepts Series). Munich: Philosophia Verlag, 211–42. Jenkins, C. S. I. 2014. ‘Intuition’, Intuition, Concepts and the A Priori. In Anthony Booth and Darrell Rowbottom, eds., Intuitions. Oxford: Oxford University Press, 91–115. Kaplan, Mark. 1985. It’s Not What You Know that Counts. Journal of Philosophy 82(7): 350–63.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

on putting knowledge ‘first’  131 Kornblith, Hilary. 2007. Naturalism and Intuitions. Grazer Philosophische Studien 74(1): 27–49. Kornblith, Hilary. 2008. Knowledge Needs No Justification. In Quentin Smith, ed., Epistemology: New Essays. Oxford: Oxford University Press, 5–23. Koslicki, K. 2012. Varieties of Ontological Dependence. In Fabrice Correia and Benjamin Schnieder, eds., Metaphysical Grounding: Understanding the Structure of Reality. Cambridge: Cambridge University Press, 186–213. Lewis, David K. 1983. New Work for a Theory of Universals. Australasian Journal of Philosophy 61: 343–77. Lewis, David K. 1984. Putnam’s Paradox. Australasian Journal of Philosophy 62: 221–36. Nagel, Jennifer. 2013. Knowledge as a Mental State. Oxford Studies in Epistemology 4: 275–310. Roessler, Johannes. 2013. Knowledge, Causal Explanation, and Teleology. Oxford Studies in Epistemology 4: 321–32. Rosenberg, Jay. 2002. Thinking about Knowing. Oxford: Oxford University Press. Schaffer, Jonathan. 2010. Monism: The Priority of the Whole. Philosophical Review 119(1): 31–76. Shope, Robert K. 1983. The Analysis of Knowing: A Decade of Research. Princeton, NJ: Princeton University Press. Sider, Theodore. 2011. Writing the Book of the World. Oxford: Oxford University Press. Williamson, Timothy. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, Timothy. 2006. Conceptual Truth. Aristotelian Society Supplementary Volume 80: 1–14. Zagzebski, Linda. 1994. The Inescapability of Gettier Problems. Philosophical Quarterly 44(174): 65–73. Zagzebski, Linda. 1996. Virtues of the Mind. Cambridge: Cambridge University Press.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

7 No Need for Excuses Against Knowledge-First Epistemology and the Knowledge Norm of Assertion Joshua Schechter

1. Introduction Since the publication of Timothy Williamson’s Knowledge and its Limits, knowledge-first epistemology has become increasingly influential within epistemology.1 This chapter discusses the viability of the knowledge-first program. It has two main parts. In the first part, I briefly present knowledge-first epistemology (Section 2) as well as ­several big-picture reasons for concern about this program (Section 3). While these considerations are pressing, I concede, however, that they are not conclusive. To determine the  viability of knowledge-first epistemology will require philosophers to carefully evaluate the individual theses endorsed by knowledge-first epistemologists, as well as to compare it with alternative packages of views. In the second part of the ­chapter, I  ­contribute to this evaluation by considering a specific thesis endorsed by many ­knowledge-first epistemologists—the knowledge norm of assertion. According to this norm, roughly speaking, one should assert that p only if one knows that p. In Section 4, I present and motivate this thesis. I then turn to a familiar concern with the norm: In many cases, it is intuitively appropriate for someone who has a strongly justified belief that p, but who doesn’t know that p, to assert that p. Proponents of the knowledge norm of assertion typically explain away our judgments about such cases by arguing that the relevant assertion is improper but that the subject has an excuse and is therefore not blameworthy for making the assertion (Section 5). In Section 6, I argue that that this response will not work. In many of the problem cases, it is not merely that the subject’s assertion is blameless. Rather, the subject positively ought to make the assertion. Appealing to an excuse cannot be used to adequately explain this fact. (Nor can we 1  Proponents of knowledge-first epistemology include Williamson (2000), Hawthorne and Stanley (2008), and Littlejohn (2013). See McGlynn (2014) for a comprehensive critical discussion. My discussion of knowledge-first epistemology is heavily indebted to Williamson (2000).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  133 explain this fact by appealing to some other, quite different, consideration.) Finally, Section 7 concludes the chapter by briefly considering whether we should replace the knowledge norm of assertion with an alternative norm. I argue that the most plausible view is that there is no norm specifically tied to assertion.

2.  Knowledge-First Epistemology Knowledge-first epistemology was originally developed as a reaction to the post-Gettier attempts to provide a conceptual analysis of the concept of knowledge. Gettier (1963) famously argued against analysing knowledge as justified true belief. In doing so, he provided cases in which a subject had a justified true belief that did not count as knowledge.2 Gettier’s paper led to a long series of papers proposing new analyses of knowledge—the no false lemmas view, the no undefeated defeater view, casual theories, sensitivity theories, and many others.3 Each of these proposed analyses were subject to further counterexamples. As time went on, proposed analyses grew increasingly baroque and ad hoc. As Williamson (2013, 2) writes, the program of providing an analysis of knowledge displayed the “signs of a degenerating research program”. The repeated failure of the post-Gettier literature to provide a successful analysis of the concept of knowledge suggested that we should perhaps take the concept of knowledge to be unanalysable. Indeed, since the force of the counterexamples did not turn on whether the proposed accounts were intended to provide conceptual analyses of the concept of knowledge or metaphysical reductions of the state of knowing,4 there was also some impetus in favor of taking the state of knowing to be irreducible (or, at least, to not be reducible along the lines of anything like the “traditional” justified true belief account). Once it is claimed that knowledge is unanalysable and irreducible (in anything like the traditional way), we have a new primitive to deploy. We can try to put this primitive to work in providing conceptual analyses and metaphysical reductions. For instance, instead of analysing the concept of knowledge in terms of the concept of belief and the concept of justification, one might try to analyse the concept of belief and the concept of justification in terms of the concept of knowledge. Alternatively, or in addition, one might try to reduce the cognitive state of believing and the epistemic status of being justified to the state of knowing. One might try to go further and relate knowledge to other epistemic notions, such as inquiry and evidence. One might go 2   The dialectic in Gettier’s paper is more complicated than the typical presentation suggests. Gettier presents an abstract argument to the conclusion that there can be a justified true belief that does not count as knowledge. (One can competently deduce a justified true belief from a justified false belief. Such a belief will not count as knowledge.) The role of Gettier’s original cases was to illustrate this general argument. The subsequent literature, however, became explicitly case-based. 3   See Clark (1963), Lehrer and Paxson (1969), Goldman (1967), and Nozick (1981), respectively. See Shope (1983) for an exhaustive presentation of the literature up to the book’s publication. 4   My use of the term “state” is not meant to commit me to what Williamson (2000) means when he calls knowledge a mental state. Every claim I make about the state of knowing could be rephrased as a claim about the knowledge relation.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

134  Joshua Schechter still further and relate knowledge to non-epistemic notions, such as deliberation, assertion, and action. Knowledge-first epistemology is the result of going down this intellectual path. Proponents of knowledge-first epistemology endorse the claim that knowledge is unanalysable and irreducible (in anything like the traditional way).5 But they go much further than this. They suggest that epistemology (and perhaps the philosophy of mind) should be reoriented with knowledge at its base. Much of our mental lives—our cognitive states and their epistemic statuses—and our behavior—actions, assertions, and their assessment—should be understood in terms of knowledge. How exactly this should go varies from theorist to theorist. But what doesn’t vary among knowledgefirst epistemologists is the endorsement of the claim that knowledge is of fundamental importance to the understanding of our mental lives—it is an “unexplained explainer”.6 Knowledge-first epistemology, then, shouldn’t be understood as a single thesis or package of theses. Rather it is what we might call a “program”. To be a knowledge-first epistemologist requires endorsing the claims that the concept of knowledge is unanalysable and that the state of knowing is irreducible (in anything like the traditional way). But it also requires taking a certain stance—treating knowledge as being of first importance, and trying to put the concept of knowledge and the state of knowing to use in doing significant philosophical work.7 There are many specific theses that are endorsed by individual knowledge-first ­epistemologists. They concern the concept of knowledge and its relation to other ­concepts; the metaphysics of the state of knowing and its relation to other cognitive states and epistemic statuses; normative connections between knowledge, assertion, and action; the purpose of cognition and communication; and much else besides. Being a knowledge-first epistemologist does not require that one endorse any specific collection of these theses. But knowledge-first epistemologists will typically endorse many such theses. In particular, knowledge-first epistemologists will typically endorse many of the following theses (or variants thereof):8 Concepts: • The concept of belief (and the concepts of other cognitive states) can be analysed in terms of the concept of knowledge. For instance, perhaps believing that p can be analysed as treating p as if one knows it. 5   It is compatible with knowledge-first epistemology to endorse positive claims about the nature of knowledge. For instance, Williamson (2000) endorses a “safety from error” requirement: If one knows that p, one could not have easily been wrong in a similar case. This is not meant as a conceptual analysis or a reduction of knowledge, since according to Williamson, “similar case” should itself ultimately be understood in terms of knowledge. 6   Williamson (2000, 10). 7   In principle, these two components of the knowledge-first program are separable. One could claim that the concept of knowledge is unanalysable and the state of knowing is irreducible without also claiming that the concept of knowledge and the state of knowing can be used to do significant philosophical work, and vice versa. 8   Many of these theses can be found—explicitly or implicitly—in Williamson (2000). See McGlynn (2014) for additional citations.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  135 • The concept of justification (and the concepts of other epistemic statuses) can be analysed in terms of the concept of knowledge. For instance, perhaps having a justified belief that p can be analysed as either knowing that p or having a belief that p that would have been knowledge had the world cooperated. Psychology: • Normal human subjects acquire the concept of knowledge before they acquire the concept of belief or the concept of justification. Metaphysics: • The knowledge relation is a metaphysically privileged relation. In the terminology of Lewis (1983), it is “perfectly natural”. Alternatively, the knowledge relation is more natural than any other cognitive relation (such as the belief relation) or epistemic property (such as being justified). • Belief and other cognitive states can be characterized in terms of the state of knowing. • Justification and other epistemic statuses can be characterized in terms of the state of knowing. • The characterization of belief in terms of knowledge is part of the essence of belief. It provides the real definition of belief. (And similarly for other cognitive states.) • The characterization of justification in terms of knowledge is part of the essence of justification. It provides the real definition of justification. (And similarly for other epistemic statuses.) • Seeing that p, remembering that p, and understanding that p each entail knowing that p. More generally, knowing is the most general propositional attitude that is factive and that constitutes a mental state. Behavior: • In predications, explanations, and rationalizations of a subject’s behavior, one should appeal to the subject’s knowledge and not merely to the subject’s beliefs. Connections within Epistemology: • A subject’s total evidence is all and only the propositions that the subject knows. • Knowledge terminates inquiry. Knowing that p (or knowing that it is not the case that p) closes the question of whether p. Normative Connections: • Knowledge norm of assertion: One ought to assert that p only if one knows that p. • Knowledge norm of reasoning: One ought to use the proposition that p as a premise in practical or theoretical reasoning only if one knows that p. • Knowledge norm of belief: One ought to believe that p only if one knows that p.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

136  Joshua Schechter • Knowledge norm of action: One ought to act as if p only if one knows that p. • The knowledge norm of assertion uniquely characterizes assertion from among the speech acts. (And similarly for the other normative connections.) • The real definition of assertion is that it is the speech act that obeys the knowledge norm. (And similarly for the other normative connections.) Teleology: • The central purpose of cognition is to generate knowledge.9 • The central purpose of communication is to transmit and pool knowledge.10 • Belief aims at knowledge. Meaning and Reference: • What one knows (or is in a position to know) in part determines the reference of one’s linguistic and mental representations. Philosophical Methodology: • Epistemology should primarily focus on issues concerning knowledge rather than justified belief or rational credence. • The philosophy of mind should primarily focus on issues concerning knowledge rather than belief. What motivates these theses (or variants thereof)? Many of these theses can be ­motivated individually. For instance, since it will be the focus of the second half of this chapter, consider the knowledge norm of assertion. According to this thesis, one ought to assert that p only if one knows that p. This thesis has been argued for in the following ways.11

(i) When someone makes an assertion, it can be questioned by asking, “How do you know that?” An assertion can be rejected by saying, “You do not know that.” Requests for answers to yes/no questions can be appropriately responded to with “I do not know.” And so forth. The best explanation of this body of data is provided by the knowledge norm. (ii) There is something inappropriate about asserting a Moorean paradoxical sentence of the form “p but I don’t know that p.” The best explanation of this is provided by the knowledge norm. (iii) If one has merely statistical grounds for believing that someone who purchased a ticket in a lottery is highly unlikely to win, it is inappropriate to tell them, “You’re not going to win”. The best explanation of this is provided by the   There are other important purposes of cognition—e.g., planning and deliberation.   There are other important purposes of communication—e.g., coordination with others. 11   See Williamson (2000), which develops considerations originally put forward in Unger (1975) and Slote (1979). 9

10

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  137 k­ nowledge norm combined with the claim that subjects cannot know lottery propositions on merely statistical grounds.12 Beyond the individual motivations for these theses, there is a more general ­motivation—the theses “hang together” nicely. Taken as a whole, they provide an attractively tidy theoretical account of mental life, with knowledge at its center. A central purpose of cognition is the generation of knowledge. When all goes well, our beliefs count as knowledge. In our theoretical and practical reasoning, we reason from premises that we know. In communication, we exchange knowledge with others. We act on what we know. And so on. That is what happens when things are going (very) well. Of course, things don’t always go so well. Sometimes we have “mere beliefs”—beliefs that do not count as knowledge. Sometimes our mere beliefs are justified, sometimes not. We can explain what mere beliefs are and what justification is in terms of knowledge. We can explain what it is for a course of reasoning or an assertion to be improper in terms of knowledge. And so on. The simplicity and elegance of this picture provides us with reason to endorse it.

3.  . . . and its Limits As attractive as this picture is, there are reasons for concern. This section presents three big-picture worries for knowledge-first epistemology. While they are not conclusive, they do provide reasons to be uneasy about the program. The first worry stems from the post-Gettier literature. Proponents of knowledgefirst epistemology argue that the post-Gettier literature shows that the concept of knowledge is unanalysable and that the state of knowing is irreducible. I am sympathetic to this view. But they also argue for the central importance of knowledge in our  mental lives. My view is that the post-Gettier literature illustrates that this is a mistake.13 Consider some Gettier-style case in which a subject has a justified true belief that is  not knowledge. For instance, here is a version of an example originally due to Russell (1912): Suppose that Jones wonders what time it is and briefly glances at his trusty wall clock. Suppose that the clock reads “2:05” and so Jones comes to believe that it is 2:05 pm. Suppose that it really is 2:05 pm. However, suppose that, unbeknownst to Jones, the clock stopped exactly 12 hours ago. In this case, Jones has a true belief that it is 2:05 pm. This belief is justified, since it is based on good evidence that the time is 2:05 pm. But, intuitively, Jones’s belief does not count as knowledge. Thinking about this case, it seems clear that Jones is in a very good epistemic ­position. His belief about the time is true. His belief fits the evidence that is available 12   There have been several responses to these motivations in the literature. See, for example, Hill and Schechter (2007), Lackey (2007), Kvanvig (2009), and Whiting (2013). 13   Kaplan (1985) and Kvanvig (2003) also argue that a moral of the post-Gettier literature is that knowledge is not an important epistemic status.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

138  Joshua Schechter to him. His belief was formed on the basis of the evidence. If he acts on his belief about the time—rushing off because he has a 2:15 pm appointment, say—he will be acting in a way that is appropriate given the facts. He will be acting in a way that makes sense given his evidence. If he is asked what time it is and he asserts that it is 2:05 pm, he’ll be asserting a truth. He’ll be asserting something that is reasonable for him to assert given his evidence. And so on. In every sense that seems important, Jones is doing quite well. To be sure, Jones lacks knowledge. He doesn’t know what time it is. Put baldly like that, it can sound like Jones is not in a great epistemic position, after all. But this strikes me as an illusion due to how we talk—when we say things like “S does not know what time it is”, we are typically trying to communicate that S is unsure, getting it wrong, or otherwise in some highly problematic state with regard to his view of the time.14 Focusing on the fact that Jones’s belief is both true and justified, however, it does not seem like Jones is missing anything important. What this suggests is that knowledge is actually a rather marginal epistemic status. It is important to have true beliefs so that we correctly represent the world. It is important to respect the evidence and, more generally, to have justified beliefs (and rational ­credences). But it is difficult for me to see that, in addition, it is important to have knowledge, too. There are features of Jones’s belief that might seem problematic. It is lucky that Jones’s belief turned out to be true. Jones’s belief about the time is bound up with his false belief that the clock is in good working order. Jones’s belief was formed via an unreliable process—glancing at a stopped clock.15 Jones’s belief is not safe in the sense that Jones could easily have come to have a false belief about the time. And so on. Nevertheless, these features of Jones’s belief do not make it any more problematic than some genuine cases of knowledge. There are cases of knowledge that have similar features. For instance, someone can win a History of the Dodo volume in a raffle and by reading it come to believe that the first recorded mention of dodo birds was in 1598. This belief counts as knowledge despite the element of luck involved. Someone can miscount the number of chairs in a room as eleven (rather than twelve), and then infer that the number of chairs in the room is less than thirty. The inferred belief counts as knowledge despite the fact that it is bound up with a false belief about the exact number of chairs present. There are differences between the kind of luck involved in Jones’s belief and the kind of luck involved in winning a raffle—it was lucky that Jones’s belief was true, whereas it was lucky that the raffle entrant came to have a belief about dodos at all.16 There are also differences between the way in which a false belief is bound up 14   See Weiner (2009, 166) for a related view. Here, Weiner argues that the concept of knowledge is a “Swiss Army” concept in the sense that knowledge is a concept “that is not important in itself, but that provides an economical way of summing up several other concepts that are important in themselves”. 15   Jones’s belief was also formed via a reliable process—glancing at a clock that one has no reason to think is not working. 16   Pritchard (2005) distinguishes between five ways in which a belief may be epistemically lucky and argues that three of them are compatible with knowledge and two are not.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  139 with Jones’s belief about the time and the way in which a false belief is bound up with the chair counter’s belief about there being fewer than thirty chairs. But it is difficult to see how anything of serious epistemic significance could depend on these subtle differences. The point is not specific to Russell’s clock case. The same holds true for other Gettierstyle cases. In all of these cases, when we focus on the details of the case, the fact that the subject doesn’t count as knowing is of little intuitive significance. We are used to talking in terms of knowledge and treating knowledge as the most significant epistemic status. But this may be more an accident of human language and psychology than a guide to the epistemological joints of nature. The second worry about knowledge-first epistemology can be raised by asking the following questions: What could the state of knowing be such that it does everything that proponents of knowledge-first epistemology suggest it does? How could there be a single kind of state that plays all (or many) of the roles listed above? Proponents of knowledge-first epistemology argue that the state of knowing is irreducible, and so cannot answer these questions by providing a reductive account of knowing. But that does not mean that the questions are illegitimate. Proponents of knowledge-first epistemology make the striking claim that there is a single kind of state that plays very many different roles in our mental life. In supporting this claim, they owe us some understanding of how this could be so. What is needed, then, is not a reductive account, but some kind of picture that explains what knowledge is such that it can be the unexplained explainer at the heart of our mental lives. There are pictures that could explain the centrality of knowledge to our mental lives. If one of these pictures were attractive, that would buttress the case of knowledge-first epistemologists. But if there were no attractive picture, that would seem to cast doubt on the program of knowledge-first epistemology. The trouble, then, is this: every ­picture that I can envision that would explain the centrality of knowledge is grossly implausible. Consider a broadly Cartesian picture of knowledge, according to which knowledge entails infallibility, indefeasibility, and rational certainty.17 Suppose, too, that we could always tell when we had such knowledge. If it were possible for us to have a wide range of such Cartesian knowledge, one could make sense of the centrality of knowledge in our mental lives.18 It would be extremely important to generate such knowledge, ­communicate such knowledge, act on such knowledge, and so forth. Of course, this picture of knowledge is highly implausible.19 If knowledge were required to conform to these constraints, we would have precious little, if any, k­ nowledge. Perhaps some of our beliefs in simple logical, mathematical, and conceptual truths are infallible, indefeasible, and rationally certain. I suspect not, given that there can be   Descartes himself did not impose such strong requirements on everyday knowledge.   See BonJour (2011). 19   Williamson (2000) agrees. His anti-luminosity argument is targeted in part against such a Cartesian view. 17 18

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

140  Joshua Schechter rational disagreements even over basic logical principles. But even if we could have some knowledge that satisfied the Cartesian constraints, none of our everyday ­knowledge about ourselves or the world would count. Alternatively, consider a picture of knowledge according to which knowledge entails being in some intimate metaphysical relation with the relevant fact. Such a view could be developed by analogy to a disjunctivist view of perception according to which perception provides subjects with a distinctive kind of contact with objects in the world, at least in the “good case”.20 On the analogous picture of knowledge, in the “good case” of believing—that is, knowing—subjects have a distinctive kind of metaphysical contact with the fact that p. Such a picture of knowledge might be able to explain the central importance of knowledge to our mental life, at least if we could explain what “contact” comes to and why having contact of this sort is itself so important. (We would also need to explain why having a true belief or a justified true belief does not entail having contact with the relevant fact.) Yet, this picture is implausible, too. Consider some of the many different ways in which one can gain knowledge. For instance, one can gain knowledge by testimony. If knowledge entails a distinctive kind of contact with the facts, the picture presumably would be that when a hearer gains knowledge by the testimony of someone who knows, the speaker’s contact with the relevant fact is somehow transmitted to the hearer along with the belief in the relevant proposition. This seems incredible. Similarly, one can gain knowledge by inductive reasoning. Again, it is difficult to believe that an inductive inference can yield some kind of intimate metaphysical relation to the truth of the conclusion, even given an intimate metaphysical relation to the truth of the premises.21 How could it be that being in contact (whatever that is) with assorted facts about the nearby environment can put us in contact with facts about distant locations? How could it be that being in contact with assorted facts about the present and recent past can put us in contact with facts about the distant future?22 Given that these pictures of knowledge are implausible, it is difficult to see how else to explain the multifarious roles that knowledge plays in our mental lives according to knowledge-first epistemology. This is not to say that knowledge-first epistemologists do not have arguments for their claims—they have provided a battery of arguments in support of their individual theses, and there is a nice theoretical unity to the ­combination of their views. But in the absence of an attractive picture that illuminates 20   See Hinton (1967), Snowdon (1980), McDowell (1982), and Martin (2002). See Pritchard (2012) and the papers in Byrne and Logue (2008) for recent discussions of disjunctivism. 21   The difficulty here is perhaps most acute for an inductive inference to a conclusion about a particular case that does not go via some inductive generalization—for example, an inductive inference to the conclusion that the next marble will be red that does not go via the intermediate conclusion that all marbles are red. 22   There are metaphysical relations that are transmitted via testimony—for instance, standing in a causal relation with the fact that makes the belief true. The point is that there is no good candidate for such a relation that (a) can be transmitted via testimony and inductive inference, (b) is present in all cases of knowledge, and (c) explains the centrality and importance of knowledge. Thanks to Louis deRossett for pressing me on this issue.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  141 how it could be that there is a cognitive state that can simultaneously play all the roles that knowledge is supposed to play, there is reason for concern. The third worry about knowledge-first epistemology stems from the diversity of our cognitive lives. There are many different kinds of cognitive states beyond outright belief.23 For instance, there are conditional beliefs. There are degrees of confidence. There are assumptions, presuppositions, hypotheses, speculations, hunches, guesses, expectations, convictions, and so on. If part of knowledge-first epistemology is to ­characterize all of these cognitive states in terms of knowledge, that strikes me as a tall order. Similarly, there are positive epistemic statuses beyond justification and ­knowledge. For instance, there is wisdom. There are various kinds of understanding.24 There are specific intellectual virtues, such as open-mindedness and creativity. If part of knowledge-first epistemology is to characterize all of these epistemic statuses in terms of knowledge, that also strikes me as a tall order. In a similar vein, in providing an account of our cognitive lives, one should not focus entirely on beliefs and other cognitive states. There are also transitions between cognitive states. These transitions are governed by rules of inference (and belief-forming methods, more generally). There are important epistemological questions about such transitions and rules. For instance, one can look at a course of reasoning that yields an unjustified belief and ask where exactly the mistake was made. Was one of the subject’s initial beliefs unjustified? Or did the lack of justification stem from an epistemic mistake in one of the subsequent transitions in thought? It is difficult to see how these questions can be understood as ultimately concerning knowledge. A knowledge-first epistemologist could, I suppose, claim that a transition between beliefs is faulty if it is not the application of a rule of inference that preserves k­ nowledge when the world is cooperating. (This is analogous to the suggestion that a belief is ­justified if it is knowledge or would have been knowledge had the world cooperated.) But this proposal faces two difficulties. First, we need an account of what it is for the world to cooperate. Second, and more importantly, there is a kind of Euthyphro ­problem here. It is plausible that a rule of inference preserves knowledge in part because it doesn’t involve any mistake. It is less plausible that a rule of inference doesn’t involve a mistake because it preserves knowledge. This suggests that there is a more fundamental epistemic status than knowledge—namely, whether a rule of inference involves an epistemic mistake. The diversity of our cognitive lives—involving many different kinds of cognitive states, many different epistemic statuses, and including transitions and rules of ­inference—suggests that the attractively tidy account provided by knowledge-first epistemologists may be a bit too tidy an account. 23   I am here distinguishing cognitive states from conative states such as desiring, wishing, planning, intending, and the like. 24   See Grimm (2006), Khalifa (2011), Sliwa (2014), and Kelp (2015) for proposals on how to reduce (propositional) understanding to knowledge.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

142  Joshua Schechter I should concede that these worries are not conclusive. (Indeed, I suspect that knowledge-first epistemologists won’t be strongly moved by them.) There may be responses to each of them. And, even if there aren’t, other considerations in favor of knowledge-first epistemology may outweigh the worries. All I would like to claim here is that these considerations provide reasons for concern about the program of ­knowledge-first epistemology. To fully determine the viability of the knowledge-first approach will require ­philosophers to carefully evaluate the specific theses endorsed by knowledge-first epistemologists and to determine their costs and benefits. It will also require philosophers to compare knowledge-first epistemology with alternative packages of views. In this, I agree with Williamson (2013, 6–7), who writes, “In the long run, knowledge-first epistemology . . . should be judged by its fruitfulness as a research program, compared to its competitors.” Indeed, I expect that any verdict on knowledge-first epistemology will only emerge over the next few decades. The remainder of this chapter contributes to the evaluation of knowledge-first epistemology by considering a thesis endorsed by many of its proponents (as well as by other philosophers)—the knowledge norm of assertion. According to this thesis, roughly speaking, one should assert that p only if one knows that p. In what follows, I will be focusing on a relatively fine-grained issue. There is a popular line of objection to the norm; in many cases, it seems appropriate for someone who has a strongly justified belief that p, but who does not know that p, to assert that p. Proponents of the ­knowledge norm of assertion typically explain away our judgments about such cases by arguing that the relevant assertion is improper but that the subject is not blameworthy for making the assertion because the subject possesses an excuse. In what ­follows, I argue that this response to the objection does not work. My discussion does not, of course, conclusively show that knowledge-first epistemology faces serious ­trouble. Indeed, it does not even conclusively show that the knowledge norm of ­assertion is unacceptable. But it puts pressure on the knowledge norm of assertion, and derivatively on the ­program of knowledge-first epistemology.

4.  The Knowledge Norm of Assertion Roughly speaking, the knowledge norm of assertion is as follows: One ought not to assert that p unless one knows that p. In other words, there is something improper about asserting what one does not know to be true.25 As stated above, there are several different motivations that have been put forward in support of this norm—for instance, it is appropriate to respond to an assertion by 25   Versions of the knowledge norm of assertion have been defended by Unger (1975), Slote (1979), Williamson (2000), DeRose (2002), Reynolds (2002), Adler (2002), Hawthorne (2004), Stanley (2005), Engel (2008), Schaffer (2008), Turri (2010), Benton (2011), and Blaauw (2012). See McGlynn (2014) for detailed discussion.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  143 asking the speaker “how do you know?”, it is inappropriate to assert claims of the form “p and I don’t know that p”, and it is inappropriate to assert lottery propositions believed on merely statistical grounds. To this we could add another, bigger picture, line of thought. A central purpose of communication is to transmit and pool knowledge. It is, therefore, important that there be a speech act tied to the transmission of knowledge. This speech act is assertion. This explains why an assertion of something that one does not know counts as improper.26 The statement of the knowledge norm of assertion above is somewhat rough. There are many choice points in developing a specific thesis. Here are a few of them. First, what is the particular normative status that appears in the knowledge norm? Must subjects make assertions only if they have the relevant knowledge? Or, instead, ought subjects to make assertions only if they have the relevant knowledge? Do they have reason not to make an assertion if they lack the relevant knowledge? Are they criticizable for asserting when they lack knowledge? Do they lack the authority to make an assertion when they lack knowledge? Or is some other normative status at issue?27 In what follows, to use neutral terminology, I’ll ask about whether an assertion is “proper” or “improper”. Second, is knowledge meant to be necessary, or both necessary and sufficient for the propriety of an assertion? For simplicity, I’ll assume that the knowledge norm states a necessary condition.28 So far as I can tell, nothing in what follows will hang on this. Third, does the norm directly target the propriety of the assertion or the propriety of the assertor? That is, does the norm state that there is something wrong with an assertion that is not based on knowledge or that there is something wrong with the person making the assertion? The prevailing view seems to be that the norm directly targets the propriety of the assertion, rather than the assertor. (In the literature, the norm is often stated as a claim about assertors, but this seems mainly to be for presentational convenience.) I find this somewhat puzzling—the intuitions that are used to motivate the norm seem to involve judgments about the relevant assertor and not merely the assertion. But so far as I can tell, nothing in what follows will hang on this, either. Fourth, what kind of normativity is at issue in the knowledge norm? Is it epistemic, moral, pragmatic, conventional (perhaps like the rules of etiquette or of spelling), a distinctively representational or communicative kind of normativity, or “plain vanilla” normativity (if there is such a thing)? One might think that the relevant kind of ­normativity must be epistemic since knowledge is, after all, an epistemic notion. But this is a mistake—one can have an epistemic requirement built into a non-epistemic norm. (Compare: an epistemic requirement can be built into a legal norm or into the rules of a game.) A better thought to have here is that the knowledge norm of assertion   A version of this motivation is put forward in Williamson (2000, 267).   Williamson (2000) moves between “must”, “criticizable”, and “authority”. It is not apparent to me how these statuses are meant to be connected. 28  See Brown (2010) and Lackey (2011) for objections to the sufficiency of knowledge for proper assertion. 26 27

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

144  Joshua Schechter flows from the function or purpose of the practice of assertion. The kind of normativity at issue in the knowledge norm depends on what this function or purpose is. For instance, on one view, the central function of the practice of assertion is to enable the transmission and pooling of knowledge. Since this is an epistemic function, the relevant kind of normativity is epistemic. This is perhaps the most plausible view for a knowledge-first epistemologist to take. However, one could instead motivate a knowledge norm starting with the claim that the central function of assertion is to enable agents to coordinate with one another. This would support a moral or pragmatic kind of normativity. And other options are live possibilities, too. Fifth, does the norm govern assertions by ordinary subjects or does it, first and foremost, govern assertions by idealized subjects? Here is an analogy that may help to make this contrast more salient. Probabilists typically argue that probabilistic coherence is a norm of rational credence. But this norm is usually understood as a constraint that, in the first instance, governs ideal reasoners. How this norm ends up bearing on ordinary reasoners turns out to be a delicate matter. What should we say about the knowledge norm of assertion? The prevailing view here seems to be that the norm applies to all assertions, whether by ordinary or idealized subjects. Sixth, is the knowledge norm of assertion a fundamental norm? Or is it grounded in  some more basic norm or norms (together with some descriptive facts)? For instance, is the knowledge norm of assertion somehow grounded in the knowledge norm of action? Seventh, and finally, what exactly is the connection between the norm and assertion? Does the knowledge norm of assertion uniquely characterize the speech act of assertion, out of all of the speech acts? Is it an analytic truth about assertion? Is it an essential feature of assertion? Does it provide the real definition of assertion? Williamson (2000) suggests that the knowledge norm is constitutive of assertion. On this view, the knowledge norm of assertion is analogous to a rule of a game. One counts as playing a game just in case (and by virtue of the fact that) one is subject to the rules of the game. Analogously, one counts as making an assertion just in case (and by virtue of the fact that) one is subject to the knowledge norm.29 Different answers to these questions generate different candidate knowledge norms of assertion. So far as I can tell, the differences between these candidate norms will largely be irrelevant to the discussion that follows. (There are exceptions. It will be important that the knowledge norm applies to ordinary subjects and not merely to idealized subjects. The distinction between the norm targeting the assertion or the assertor will also turn out to be relevant.) What should we think about the knowledge norm of assertion, however it is developed? There is an obvious objection to the norm—one that has loomed large in the

29   See Maitra (2011) and Johnson (2017) for objections to the analogy between the knowledge norm of assertion and the rules of a game.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  145 l­ iterature: in many cases, an assertion that p can be proper if the assertor has a strongly justified belief that p, even if the assertor does not know that p.30 Consider a Gettier-style case, such as the case of Jones and the stopped clock. Suppose that Jones glances at the clock and is then immediately asked the time. Suppose that Jones answers that it is 2:05 pm. Intuitively, Jones’s assertion is proper despite the fact that he does not know that it is 2:05 pm. (Moreover, Jones’s assertion seems just as proper as it would have been had he genuinely known that it is 2:05 pm.) And similarly for other Gettier-style cases. Indeed, there are many cases in which an assertion that p seems proper if the ­assertor has a strongly justified belief that p, even if the belief is false. For instance, suppose that Smith has strong evidence that the fastest way to get to Thayer Street from a certain location is to make a right on Benefit Street and then a left on Waterman Street. (Suppose that Smith lives in the area and knows the streets well.) When asked by someone “What is the fastest way for me to get to Thayer from here right now?”, it is intuitively proper for Smith to respond by asserting “The fastest way for you to get to Thayer is to make a right on Benefit and then a left on Waterman.” This is so even if, unbeknownst to Smith, Benefit Street is currently closed to traffic due to an unexpected problem with a water main. The fact that Smith’s belief is false does not intuitively make his assertion improper.31 One can push this general line of objection still further. In many cases, an assertion that p can be proper if the assertor believes that p, even if the belief is unjustified.32 For  example, suppose that Robinson strongly believes that there is life on Europa on the basis of wishful thinking. (In case it matters, we can suppose that Robinson does not believe that his belief that there is life on Europa is at all defective.) Suppose that Robinson is asked “Is there life somewhere other than the Earth?” Suppose that Robinson responds “Yes, there is life on Europa.” I maintain that Robinson’s assertion is proper. Since Robinson believes that there is life on Europa, it is not improper for Robinson to say so. Indeed, if Robinson has no reason to avoid answering the question or to try to mislead his interlocutor, Robinson positively ought to assert that there is life on Europa. Of course, there is something wrong with Robinson—he has an unjustified belief. But the fact that his belief is unjustified does not entail that his assertion is improper. That is a different matter. One might think that there is a straightforward way to see that my claim about the propriety of Robinson’s assertion is false. In particular, one might draw an analogy between assertion and belief. According to this line of thought, assertion is the outer

30   Versions of this objection appear in Douven (2006), Hill and Schechter (2007), Lackey (2007), Brown (2008), Koethe (2009), Kvanvig (2009), Gerken (2011), and McKinnon (2013). See McGlynn (2014) for discussion. 31   For a similar case, see McKinnon (2013, 122). 32   This claim is not widely endorsed in the literature.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

146  Joshua Schechter acceptance of a claim and believing is the inner acceptance of a claim.33 Now consider a case of inference from unjustified beliefs. For instance, suppose that a thinker has unjustified beliefs in p and in if p then q. Suppose that the thinker comes to believe that q by competently performing a Modus Ponens inference. The thinker’s belief that q will (typically) be unjustified. By analogy, one might think, if a subject’s belief that p is unjustified and hence epistemically improper, and the subject asserts that p, based on this belief, the assertion will also be improper. Just as the impropriety of a belief can be inherited from the impropriety of the beliefs that it is inferred from, so, too, can the impropriety of an assertion be inherited from the impropriety of the belief that it is based on. Or so goes the line of thought. This line of thought, however, is mistaken. Belief is not analogous to assertion. Believing is a state and assertion is an action. The right analogy is not between asserting that p and believing that p. It is between asserting that p and coming to believe that p.34 And, once we see that, the case of a competent inference from unjustified premises actually tells in the other direction. If a thinker comes to believe that q by competently performing a Modus Ponens inference from unjustified premises, the output belief will be unjustified. But the forming of the belief that q—the Modus Ponens inference itself—is not improper. The inference itself does not inherit any impropriety from the impropriety of its inputs. Given that the thinker strongly believes that p, and that if p, then q, and given that the question of whether q is salient, the thinker positively ought to draw the inference and form the belief that q. True, the thinker is in an epistemically problematic state. But the mistake was not in the inference to q. The mistake was in believing that p, and that if p, then q, to begin with. The analogous is true for Robinson. Robinson’s assertion that there is life on Europa is not improper. Robinson’s belief that there is life on Europa is unjustified. But the act of asserting that there is life on Europa does not inherit any impropriety from the belief that it is based on. Given that Robinson strongly believes that there is life on Europa, and given that Robinson has no reason to avoid answering the question or to try to be misleading, Robinson positively ought to make the assertion. The cases of Jones, Smith, and Robinson tell against the knowledge norm of assertion. If an assertion that p can be proper when the subject does not know that p but has a justified true belief that p, a justified false belief that p, or even an unjustified belief that p, then the knowledge norm of assertion is false. Indeed, analogous problems face other proposed knowledge norms, such as knowledge norms for reasoning, belief, and action.35 So there is a serious worry here for proponents of knowledge norms. 33   See Adler (2002). Also see Williamson (2000, 255–6), who draws the analogy between assertion and occurrent belief. 34   Is there something tied to assertion to which believing is analogous? Perhaps we could say that in making an assertion, an assertor undertakes a kind of commitment. Then we could say that believing is analogous not to assertion, but to assertoric commitment. I am not sure this is correct. But even if it is, my topic concerns the norms governing assertion, not the norms governing assertoric commitment. 35   There is no analogue of the case of Robinson that can be used against the knowledge norm of belief, but there are analogues of the cases of Jones and Smith.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  147 Proponents of the knowledge norm of assertion (and other knowledge norms) are, of course, aware of this issue. There is a standard response. In the next section, I present this response and examine some arguments for the claim that it cannot be made to work.

5.  Excuses, Excuses The intuition that an assertion can be appropriate in many cases where the assertor does not have knowledge—but instead has a justified true belief (e.g., the case of Jones) or a justified false belief (e.g., the case of Smith)—seems to be shared widely. In response to such problem cases, the standard response is to explain away our i­ntuitions. According to this response, in a problem case, the relevant assertion is not, in fact, proper. Instead, our intuition that the assertion is appropriate tracks a different normative feature. Although the assertion violated the norm governing assertion, the a­ ssertor is blameless for this violation. This is because the assertor has an excuse for the violation. Our intuition about the appropriateness of the assertion tracks blamelessness, not propriety. This explains away our intuition about the case.36 It is worth saying just a bit about excuses here.37 Very generally, there are normative constraints on action. When the norms are complied with, the action is proper.38 When the norms are violated, the action is improper. However, it can happen that the relevant norms are not complied with, but the agent is not blameworthy for the violation. This is what happens when the agent has a sufficient excuse. For example, suppose that I promised to pick up my friend from the airport today. It is a norm on behavior that one must do what one has promised to do. If my car unexpectedly (and through no fault of my own) breaks down while I am driving to the airport and I am unable to pick up my friend, I have violated the norm. But I am blameless in violating the norm—through no fault of my own, it was impossible for me to comply with it. That my car broke down excuses my violation.39 36   See Williamson (2000, 256–7), DeRose (2002), and Hawthorne and Stanley (2008). (DeRose uses a somewhat different terminology—instead of talking about excuses, he talks about “secondary propriety”. I think this is merely a terminological difference.) See Sutton (2007, 80), for an analogous response to challenges to the claim that knowledge closes inquiry. See Littlejohn (forthcoming) and Williamson (forthcoming) for detailed presentations of the excuse response as targeted against the Cohen and Lehrer (1983) and Cohen (1984) New Evil Demon problem. 37   See Austin (1956). 38   More precisely, when the moral norms are complied with, the action is morally proper. When the prudential norms are complied with, the action is prudentially proper. And so forth. 39   There are two other categories of blameless norm violation that should be distinguished from having an excuse. First, the violation of a norm can be justified if the agent had sufficiently good reason for violating it. For instance, I might have strong moral grounds not to pick up my friend from the airport, such as if there was an emergency that required me to take someone else to the hospital. (This notion of justification should not be identified with the notion(s) of justification used in epistemology.) Second, as Strawson (1962) argues, a subject can be exempt from the requirement to comply with a norm if the subject lacks the relevant rational capacities required to be held responsible for complying with the norm. For instance, young children are exempt from certain norms on behavior. See Littlejohn (forthcoming) for a discussion of the justification/excuse/exemption distinction in epistemology.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

148  Joshua Schechter What could be the excuse for violating the knowledge norm of assertion in the case of Jones or Smith? There are several options. One could point to any (or several) of the following facts as excusing the norm violation: The assertor reasonably believes the asserted proposition. The assertor reasonably believes that she knows the asserted proposition. In asserting the proposition, the assertor was trying to comply with the norm that governs assertion. The assertor is generally disposed to comply with the norm governing assertion, and the assertion was an exercise of the underlying categorical basis of that disposition. It is through no fault of the assertor that the assertion violates the norm. The assertor reasonably believes that it is highly likely that she is complying with the norm. And so forth. There are several concerns with this general line of response that appear in the ­literature. For instance, Gerken (2011) argues against versions of the excuse response where the assertor has an excuse because she reasonably believes that she knows the proposition asserted. Gerken points out that it requires significant cognitive sophistication to believe that one knows something. Presumably, someone in a Gettier-style case or a case of a justified false belief could lack the sophistication required to have such a belief. Nevertheless, intuitively, if such a person asserted the proposition she justifiably believed, the assertion would be appropriate. There are two main ways an advocate of the excuse response could try to avoid this concern. First, one could identify the excuse not with the fact that the subject reasonably believes that she knows the relevant proposition, but with something less cognitively sophisticated. For instance, one could say all that is required is that the subject’s evidence be sufficiently strong to make a belief that she knows the proposition reasonable, even if the subject is incapable of forming that belief. Similarly, one could identify the excuse with the fact that the subject reasonably believes the proposition, or with the fact that her assertion is the exercise of the categorical basis for her disposition to assert only what she knows. Second, one could say that a subject who lacks the cognitive sophistication to believe that she knows the relevant proposition (and is not culpable for this lack) is exempt from the requirement that she have such a belief. Instead, for such a cognitively limited subject, something less is required.40 Either of these approaches strikes me as providing a plausible response to Gerken’s challenge.41 A different concern appearing in the literature is that subjects in Gettier-style cases (and cases of justified false belief) intuitively do not owe an apology to their addressees, and so there could not be any violation of a norm.42 This concern can be answered by pointing out that not every norm violation requires an apology. In particular, if a norm violation is excusable, an apology is typically unnecessary.43

40   Compare: young children have fewer epistemic obligations to check up on their reasoning than do competent adults. 41   But see Gerken (2011) for responses to these suggestions. 42   See Brown (2008) and Kvanvig (2011) for versions of this concern. 43   See Littlejohn (2012) and McGlynn (2014).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  149 A closely related concern is that subjects in Gettier-style cases and cases of justified false belief do not need any kind of excuse for their assertions to be appropriate.44 I agree with this claim, but I do not see that our intuitions about the cases are clear enough that we can simply read this fact off of our intuitions. The claim is one that must be earned by careful argumentation. So, I do not think that this concern is ­dialectically compelling, either. There are additional worries that one might have about the excuse response.45 Here are two more. The first concerns the target of the norm. The prevailing view in the ­literature seems to be that the knowledge norm of assertion concerns a normative property of the assertion, not of the assertor. But the claim that the assertor is blameless for making the assertion is a claim about the assertor, not the assertion. If our ­intuitions about assertions in Gettier-style cases (and cases of justified false belief) concern the normative status of the assertor rather than the assertion, then these intuitions cannot be used to motivate the prevailing view. If instead they concern the normative status of the assertion rather than the assertor, then the excuse response cannot be used to explain them away. This suggests that there is a problem with combining the excuse response and the prevailing view. I find this worry somewhat compelling. However, it is not dialectically very strong. Proponents of the excuse response can simply respond that our intuitive judgments are not fine-grained enough to clearly distinguish between the normative status of the assertion and the normative status of the assertor. A second additional worry concerns the assertion of a proposition that the assertor unjustifiably believes, such as in the case of Robinson’s assertion that there is life on Europa. Suppose that I am right and, when we carefully consider the case, Robinson’s assertion is intuitively appropriate. An advocate of the excuse response will have to find a way to argue that Robinson has a sufficient excuse. The problem is that many of the potential excuses listed earlier are not ones that apply in the case of Robinson. For instance, Robinson does not reasonably believe the proposition. Robinson does not reasonably believe that he knows the proposition. And so forth. A potential response here, I suppose, is to claim that all that Robinson needs to have in order to count as blameless is a belief in the proposition asserted, or perhaps a belief that he knows the proposition asserted. The problem with this response is that it makes excusably violating the knowledge norm of assertion something that is remarkably easy to do—one need only believe that one is complying with it. That is not typically how excuses work. Merely believing that one is complying with a norm is not sufficient to count as blamelessly violating it. That said, my guess is that proponents of the knowledge norm of assertion will simply reject my intuition about Robinson and say that his norm violation is not blameless after all. This strikes me as a bullet to bite. But perhaps it is not a very big one.

  See Douven (2006) and Lackey (2007) for versions of this concern.   I am not aware that the following concerns appear in the literature.

44

45

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

150  Joshua Schechter These concerns about the excuse response, then, are not dialectically very powerful. In the next section, I present a more powerful objection. The objection is a little bit involved—it has a few moving parts—but it is, I think, rather compelling.

6.  No Excuses Necessary Consider the case of an assertion in a Gettier-style case (e.g., the case of Jones) or in the case of a justified false belief (e.g., the case of Smith). For concreteness, I will focus on the case of Jones, but nothing will hang on this. According to the excuse response, Jones’s assertion is improper but Jones is blameless for asserting it, since Jones has an excuse. The trouble is that it is not merely the case that Jones’s assertion is blameless. Rather, Jones positively ought to make the assertion.46 When asked, he should say that it is 2:05 pm. If Jones does not say that it is 2:05 pm (and has no reason not to cooperate with his interlocutor), then he is criticizable for not asserting what he believes about the time.47 That Jones has an excuse for making the assertion, and is therefore blameless for so doing, does not explain this fact.48 A proponent of the excuse response might reply by saying that there are two distinct considerations involved. Jones has an excuse. That explains why he is blameless in making his assertion. In addition, there is a second consideration that explains why Jones positively ought to make the assertion. For instance, perhaps there is a norm of politeness that requires Jones to respond to his interlocutor (if he has no reason not to cooperate). There are difficulties facing this specific proposal.49 But there are also more general problems with the suggestion that there are two separate considerations in play. First, whenever possible, it is better to avoid multiplying considerations, especially when the phenomena being explained are, on the face of it, closely related. Second, if there is a 46   This already casts some doubt on the excuse response. Typically, when one positively ought to do something, no excuse is necessary in order to count as blameless in doing it, at least when there is a single kind of normativity in play. I do not want to rest much weight on this point, however, since it might be claimed that there are two kinds of normativity in play—for instance, perhaps the kind of blamelessness at issue is epistemic and the kind of ought at issue is not. 47   At the very least, asserting that it is 2:05 pm is a good thing for Jones to do. That is all that is really needed for the argument in this section. 48   See Schechter (2013) for a version of this point in a different context. 49   Here are two difficulties. First, a norm of politeness may be able to explain why Jones ought to respond in some way to his interlocutor—after all, it would be rude for him to remain silent. But such a norm cannot explain why Jones ought to respond by asserting what he believes the time to be. There are many other things that Jones could instead assert that would be equally polite—for instance, “I’m very sorry, I cannot help”. Second, a norm of politeness cannot handle all of the relevant cases. Suppose that instead of being asked the time, Jones notices that someone nearby is wondering what time it is—frantically looking around for a clock, stopping people to ask if they have a watch, and so forth. Even if Jones is not asked what the time is, so long as Jones has no reason to avoid engaging with this person, there is still some sense in which Jones positively ought to say that it is 2:05 pm. At the very least, that would be a good thing for him to do. A norm of politeness cannot explain this fact. This suggests that it would be better to appeal to a norm of helpfulness instead of a norm of politeness. In a way, I think this is right, and I elaborate further later on.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  151 separate consideration that can explain why Jones positively ought to make the a­ ssertion, there is the danger that this consideration could be also used to explain the intuitive data alleged to support the knowledge norm. So, there is the danger that there would be no work left for the knowledge norm to do. More importantly, recall where the knowledge norm of assertion is supposed to come from. On the most plausible versions of knowledge-first epistemology, the practice of assertion has a function or purpose. The knowledge norm of assertion flows from this function or purpose. Assuming that this is right, the reason that Jones positively ought to assert that it is 2:05 pm would also seem to flow from the function or purpose of assertion. For instance, suppose that the central function of assertion is to enable the transmission and pooling of knowledge. Assuming this view, it is plausible to think that Jones ought to respond to his interlocutor because (very roughly) doing so is the kind of thing that tends to contribute to the transmission and pooling of knowledge. Alternatively, suppose that the central function of assertion is to help agents to coordinate more effectively. Assuming this view, it is plausible to think that Jones ought to respond to his interlocutor because (very roughly) doing so is the kind of thing that tends to contribute to effective coordination. And so forth. What this suggests is that there are not two separate considerations in play. The reason why Jones is blameless in making the assertion and the reason why Jones positively ought to make the assertion are the very same reason or, at least, are closely connected reasons. The natural view for proponents of the excuse response to take is to agree that the fact that Jones has an excuse does not explain why Jones ought to make the assertion he does. It is not the existential claim that Jones has an excuse that explains why Jones ought to make the assertion. Rather, it is the fact that constitutes his excuse that explains why he positively ought to make the assertion. On this view, one and the same fact is significant in two different ways—it functions both as an explanation of blamelessness and as an explanation of an ought. So far, the proponent of the excuse response has been able to respond to the objection. But things are about to get a bit stickier. Proponents of the excuse response should not claim that the kind of excuse at issue in the problem cases for the knowledge norm of assertion is a special kind of excuse that is specific to the problem cases, or even to assertion more generally. Any account of the kind of excuse at issue should show how such excuses fit in with a general picture of norms and norm compliance. Otherwise, the resulting view will be unwieldy and ad hoc.50 How can the excuses that feature in the problem cases be tied to a more general account of excuses for norm violations? The most sophisticated proposal of which I am aware is due to Williamson.51 On Williamson’s account, when there is a norm 50   See Hill and Schechter (2007) for an articulation of the concern that the excuse response commits Williamson to an overly complex view. 51   See Williamson (forthcoming). Also see Littlejohn (forthcoming) for a related view.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

152  Joshua Schechter governing a practice, there is typically also a secondary norm to be the sort of person who complies with the original or primary norm.52 Williamson suggests that this ­secondary norm further generates a tertiary norm to do in any situation what someone who complies with the secondary norm (that is, what someone who is disposed to comply with the primary norm) would do. For instance, there is a norm to keep one’s promises. This generates the secondary norm to be the sort of person who keeps one’s promises. This further generates the tertiary norm to do in any situation what someone who is disposed to keep their promises would do. Williamson further claims that while some of the normative force of the original norm “rubs off ” on the secondary and tertiary norms, the derivative norms typically do not “inherit the full normative status” of the original norm.53 For instance, in breaking a promise, one is breaking one’s word, not living up to a commitment one has undertaken, etc. In not being the sort of person who keeps one’s promises—so long as one has not actually broken a promise—one has done no such thing. Finally, Williamson claims it is often a good excuse for violating a primary norm that one is complying with the tertiary norm it generates. (He also claims that if one complies with both the secondary and tertiary norms, this typically strengthens the excuse.) In particular, if an agent breaks a promise in a certain situation, the fact that the agent is doing what someone who keeps their promises would do in that situation will often excuse the norm violation. There is much to say in favor of Williamson’s account. It is elegant. It fits with familiar examples of norm-governed behavior. (Williamson focuses on legal norms and the norms governing promising. One might also add the cases of spelling, grammar, ­etiquette, and games.) However, in the current context, there is a problem. Namely, Williamson’s account does not sufficiently explain why it is that, in the problem cases for the knowledge norm of assertion, the assertor ought to make the assertion. Applying Williamson’s account, the knowledge norm for assertion generates a secondary norm: one should be the sort of person who asserts only what one knows.54 It further generates a tertiary norm: in any given situation, one should assert what the sort of person who asserts only what they know would assert. Let us grant that Jones is the sort of person who asserts only what he knows. Let us also grant that in asserting that it is 2:05 pm, Jones is asserting what someone who is the sort of person who asserts only what they know would assert. So, Jones is complying with the secondary and ­tertiary norms. This could explain, perhaps, why Jones is blameless in making the assertion. But it does not sufficiently explain why Jones positively ought to make the assertion and why Jones would be criticizable if he failed to make the assertion. 52   “Complies with the norm” should be understood as a generic—to be the sort of person who complies with a norm does not require that one actually complies with the norm in any given situation. 53   Williamson (forthcoming). 54   See Lasonen Aarnio (2010), which makes use of a secondary norm on belief in order to try to explain away our intuitions that knowledge can be defeated.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  153 One might think that there is a simple reason why Jones is criticizable if he does not make the assertion. In not making the assertion, he would be violating the tertiary norm. This is not quite right—the primary norm does not mandate any assertions at all, and so neither does the tertiary norm. But this problem is easily fixed. Presumably, there is a general norm of helpfulness that entails that one should respond to reasonable questions about the time. (Perhaps such a norm is tied to Grice’s maxim of quantity.)55 In combination with the knowledge norm of assertion, this explains why one should tell someone the time when asked, if one knows it. In combination with the tertiary assertion norm, the line of thought goes, this mandates that one should tell someone what one believes the time to be if doing so is what someone who is helpful and is the kind of person who asserts only what they know would assert. So, the tertiary norm (in combination with a general norm of helpfulness) does mandate that Jones make the assertion, after all. What, then, is the problem? The problem is that the kind of mandate that tertiary norms provide is too weak to account for the force of Jones’s requirement to assert what he takes the time to be. As Williamson admits, secondary and tertiary norms typically do not “inherit the full normative status” of the primary norm. Not doing what someone who is the sort of person who keeps their promises would do is very different from breaking a promise. One ought to keep one’s promises. One also ought to do what the sort of person who keeps their promises would do. But the former ought is stronger and more “fullblooded” than the latter ought. Breaking a promise is a much more serious violation than not doing what the sort of person who keeps their promises would do. In the Jones case, however, not asserting that it is 2:05 pm seems every bit as bad— and bad in the very same way—as it would be if Jones knew that it was 2:05 pm. One way to see this is to think about Jones and twin Jones. Jones and twin Jones are in exactly the same situation—they have both just glanced at a clock that reads 2:05 and they have both been asked for the time—except for one difference. Jones glanced at a clock that stopped twelve hours ago and so does not know the time. Twin Jones glanced at a working clock and does know the time. I maintain that Jones and Twin Jones face equally strong pressure to assert that it is 2:05 pm. The oughts that apply to them are exactly on a par. A different way to see the point is to consider a diachronic case where the subject moves from knowledge to a merely justified belief and back again.56 It is a bit easier to make use of the case of Smith rather than the case of Jones, so let us turn to that case. Suppose that on Monday, Smith counts as knowing that the fastest way to get to Thayer Street from a certain location is to make a right on Benefit Street and then a left on Waterman Street. Suppose that Smith is asked by a passerby “What is the fastest way for   See Grice (1975).   See White (2014) for diachronic cases used against Williamson’s externalist response to the New Evil Demon argument. 55 56

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

154  Joshua Schechter me to get to Thayer from here right now?” Smith ought to respond by saying that the fastest way to get to Thayer Street is to make a right on Benefit Street and then a left on Waterman Street. Suppose that on Tuesday, unbeknownst to Smith, Benefit Street is closed due to a water main break. Suppose that Smith is asked by another passerby “What is the fastest way for me to get to Thayer from here right now?” Smith again ought to respond by saying that the fastest way to get to Thayer Street is to make a right on Benefit Street and then a left on Waterman Street. Suppose that on Wednesday, ­everything is back to normal. Suppose that Smith is asked by a yet another passerby “What is the fastest way for me to get to Thayer from here right now?” Yet again, Smith ought to respond by saying that the fastest way to get to Thayer Street is to make a right on Benefit Street and then a left on Waterman Street. Indeed, the strength of the oughts on all three days seems just the same. Not giving the passerby directions would be just as bad on any of the days, and in just the same way. Williamson’s account cannot accommodate this point.57 The objection to the excuse response thus involves the interaction of three ­constraints on an account of the kind of excuse at issue in the problem cases for the knowledge norm of assertion. First, it is a constraint that the account not merely explain why the assertions are blameless, but also explain why the assertors ought to assert what they do. Second, it is a constraint that the account explain the full normative strength of this ought—and in particular, why the ought is as strong in a problem case as it is in a corresponding knowledge case. Third, it is a constraint that the kind of excuse at issue not be some special kind of excuse particular to assertion. Instead, the account must fit with a general account of norm-governed behavior. The trouble is that it is difficult to see how these constraints can be satisfied simultaneously. I have illustrated this by focusing on Williamson’s account. But it is difficult to see that an alternative account can do better. It is worth noting that the dialectical situation would seem to be analogous for other proposed knowledge norms—including the knowledge norms of reasoning, belief, and action. It can be appropriate for a subject to use the proposition that p as a premise in reasoning, to believe that p, or to act as if p, despite not knowing that p. In response to problem cases for one of these norms, it might be suggested that in such cases, the ­subject has violated the norm, but has done so blamelessly, since the subject has an appropriate excuse. But, again, this response underdescribes the situation. In many problem cases, the subject positively ought to reason with the proposition, believe the proposition, or act as if the proposition is true. And it seems difficult to provide an 57   Williamson might claim that Smith does not have knowledge of the fastest route on Monday or Wednesday, since Smith’s belief is not safe. We can modify the case to avoid this response. For instance, we can change the case so that Smith is asked for directions on days that are months apart. We can also specify that the water main break is not caused by a random fluke (and thus avoid any resemblance to a lottery case). Presumably, there is some variant of the case on which Smith loses and gains knowledge. We can substitute this variant for the case presented in the text.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  155 account of the kind of excuse at issue in these problem cases that also explains where this positive ought comes from, accommodates the full normative strength of the ought, and does so in a non-ad hoc way.58

7.  Conclusion: Assertion without Knowledge Let’s take stock. In the first part of this chapter, I briefly presented the program of ­knowledge-first epistemology. I then put forward three big-picture worries for this program: (i) the post-Gettier literature suggests that knowledge is a marginal status; (ii) for knowledge-first epistemology to be tenable, we need an explanation of what knowledge could be such that it is central to our mental lives; and (iii) the diversity of our mental lives—involving many different kinds of cognitive states and epistemic ­statuses and involving transitions and rules in addition to cognitive states—casts doubt on the centrality of knowledge. I also conceded that, while these objections are pressing, they are not conclusive. Determining the viability of knowledge-first epistemology requires carefully evaluating the individual theses endorsed by proponents. In the second part of the chapter, I turned to a specific thesis that is widely endorsed by knowledge-first epistemologists—the knowledge norm of assertion. I presented a common objection to this thesis: in many cases, it is intuitively appropriate for someone who has a strongly justified belief that p, but who does not know that p, to assert that p. I then presented the standard response to this objection. In such a case, the assertion is improper, but the assertor is blameless because the assertor possesses a sufficiently good excuse. I argued that extant objections to this response are not dialectically very powerful. Instead, I presented a new (and somewhat complicated) objection. In its essentials, the objection is that it is difficult to provide a non-ad hoc account of the assertor’s excuse that explains why the assertor ought to make the assertion, where this ought is just as strong as it would be in a case of knowledge. This, I conclude, provides us with reason to reject the knowledge norm of assertion. (Analogous objections provide us with reason to reject the knowledge norms of reasoning, belief, and action.) Derivatively, this provides us with some reason to reject knowledge-first epistemology. Supposing we reject the knowledge norm of assertion—how, then, should we replace it? The discussion earlier provides some support for moving to a different norm. If we give up on the excuse response, the cases of Jones, Smith, and Robinson suggest that knowledge, truth, and justification are not necessary conditions on proper assertion. That leaves belief, or perhaps belief that one knows.59 Perhaps, then, we should accept one of these two norms. 58   If anything, the problem for these other norms is worse than for the knowledge norm of assertion. The suggestion that there are two separate considerations in play—one that explains blamelessness and one that explains the positive ought—is even less plausible for the cases of reasoning, belief, and action. 59   Bach and Harnish (1979) argue that assertions express beliefs. Bach (2008) further argues that this generates a belief norm on assertion. See Williamson (2000) for discussion of the belief that you know norm.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

156  Joshua Schechter I suspect not. There are assertions that are intuitively appropriate in which the a­ ssertor does not believe the proposition asserted. Suppose, for example, that one is tutoring a student in physics and knows that if one puts in all of the required hedges, the student will get lost in the weeds.60 Similarly, suppose that some situation is urgent, and the most straightforward thing one could do to provide the required aid is to say something one knows to be false.61 In each of these cases, it strikes me as appropriate to assert claims that one knows to be false.62 Proponents of various norms—the belief norm, the justified belief norm, the belief that one knows norm, the knowledge norm—will try to handle these cases by saying that the assertions violate the norm of assertion, but that the educational needs of the student or the urgency of the situation makes the violation blameless. This maneuver may work here. It is not subject to the same objections that face the excuse response discussed earlier. But it would certainly be more satisfying if we did not have to categorize these assertions as norm violations to begin with. It would be theoretically preferable if we could adopt an account on which these assertions are normatively on a par with other intuitively appropriate assertions. The natural account to reach for here is a broadly Gricean account. In particular, one might think that in the case of the physics student or the case of the urgent situation, what makes the assertion appropriate is that it is the most helpful thing to say in context—the assertor takes the proposition to be more relevant to the listener’s informational needs than any alternative proposition that could be asserted.63 On this approach, there are no norms specifically tied to assertion.64 Instead, there are general pragmatic principles governing communication, such as a principle that tells us to do what is communicatively most helpful in context. This approach requires that we find some other way, not involving norms, to distinguish assertion from the other speech acts.65 It also requires that we respond to the positive arguments that have been ­presented for the proposed norms of assertion. That will take a lot of work. But it strikes me as the most promising approach to take.66

  For a related case, see McKinnon (2013, 124).   For a related case, see Williamson (2000, 256). 62   For additional cases, see the examples of “selfless assertions” in Lackey (2007). 63   This is related to Grice’s maxims of relevance and quantity. See Weiner (2005), Hill and Schechter (2007), and Lackey (2007) for applications of Gricean principles in responding to arguments in favor of the knowledge norm of assertion. 64   See Sosa (2009) and Johnson (2017) for arguments that assertion does not have a constitutive norm. 65   See MacFarlane (2011) for a survey of approaches. 66   Thanks to Stewart Cohen and Timothy Williamson for discussion of these issues. Thanks to David Black, David Christensen, Louis deRossett, Mikkel Gerken, Jeremy Goodman, Christopher Hill, Casey Johnson, and David Plunkett for helpful comments on this chapter. Thanks to the students in David Plunkett’s undergraduate seminar at Dartmouth for useful questions and remarks. Thanks to the three editors of this volume for numerous suggestions that led to improvements. This chapter was written while I was a visiting fellow at the University of Connecticut Humanities Institute. I am grateful to the University of Connecticut for its hospitality. 60 61

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  157

References Adler, Jonathan. 2002. Belief ’s Own Ethics. Cambridge, MA: MIT Press. Austin, John L. 1956. A Plea for Excuses. Proceedings of the Aristotelian Society 57: 1–30. Bach, Kent. 2008. Applying Pragmatics to Epistemology. Philosophical Issues 18: 68–88. Bach, Kent and Robert Harnish. 1979. Linguistic Communication and Speech Acts. Cambridge, MA: MIT Press. Benton, Matthew. 2011. Two More for the Knowledge Account of Assertion. Analysis 71: 684–87. Blaauw, Martijn. 2012. Reinforcing the Knowledge Account of Assertion. Analysis 72: 105–8. BonJour, Laurence. 2011. The Myth of Knowledge. Philosophical Perspectives 24: 57–83. Brown, Jessica. 2008. The Knowledge Norm for Assertion. Philosophical Issues 18: 89–103. Brown, Jessica. 2010. Knowledge and Assertion. Philosophy and Phenomenological Research 81: 549–66. Byrne, Alex and Heather Logue (eds.). 2008. Disjunctivism: Contemporary Readings. Cambridge, MA: MIT Press. Clark, Michael. 1963. Knowledge and Grounds: A Comment on Mr. Gettier’s Paper. Analysis 24: 46–8. Cohen, Stewart. 1984. Justification and Truth. Philosophical Studies 46: 279–95. Cohen, Stewart and Keith Lehrer. 1983. Justification, Truth, and Coherence. Synthese 55: 191–207. DeRose, Keith. 2002. Assertion, Knowledge, and Context. Philosophical Review 111: 167–203. Douven, Igor. 2006. Assertion, Knowledge, and Rationality. Philosophical Review 115: 449–85. Engel, Pascal. 2008. In What Sense is Knowledge the Norm of Assertion? Grazer Philosophische Studien 77: 45–59. Gerken, Mikkel. 2011. Warrant and Action. Synthese 178: 529–47. Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23: 121–3. Goldman, Alvin. 1967. A Causal Theory of Knowing. Journal of Philosophy 64: 357–72. Grice, H. Paul. 1975. Logic and Conversation. In Peter Cole and Jerry Morgan, eds., Syntax and Semantics, Vol. 3. New York: Academic Press, 41–58. Grimm, Stephen. 2006. Is Understanding a Species of Knowledge? British Journal for the Philosophy of Science 57: 515–35. Hawthorne, John. 2004. Knowledge and Lotteries. Oxford: Oxford University Press. Hawthorne, John and Jason Stanley. 2008. Knowledge and Action. Journal of Philosophy 105: 571–90. Hill, Christopher and Joshua Schechter. 2007. Hawthorne’s Lottery Puzzle and the Nature of Belief. Philosophical Issues 17: 102–22. Hinton, John M. 1967. Visual Experiences. Mind 76: 217–27. Johnson, Casey Rebecca. 2017. What Norm of Assertion? Acta Analytica. DOI: 10.1007/ s12136-017-0326-3 Kaplan, Mark. 1985. It’s Not What You Know that Counts. Journal of Philosophy 82: 350–63. Kelp, Christoph. 2015. Understanding Phenomena. Synthese 192: 3799–816. Khalifa, Kareem. 2011. Understanding, Knowledge, and Scientific Antirealism. Grazer Philosophische Studien 83: 93–112. Koethe, John. 2009. Knowledge and the Norms of Assertion. Australasian Journal of Philosophy 87: 625–38.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

158  Joshua Schechter Kvanvig, Jonathan. 2003. The Value of Knowledge and the Pursuit of Understanding. Cambridge: Cambridge University Press. Kvanvig, Jonathan. 2009. Assertion, Knowledge, and Lotteries, in Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 140–60. Kvanvig, Jonathan. 2011. Norms of Assertion. In Jessica Brown and Herman Cappelen, eds., Assertion: New Philosophical Essays. Oxford: Oxford University Press, 233–50. Lackey, Jennifer. 2007. Norms of Assertion. Noûs 41: 594–626. Lackey, Jennifer. 2011. Assertion and Isolated Second-Hand Knowledge. In Jessica Brown and Herman Cappelen, eds., Assertion: New Philosophical Essays. Oxford: Oxford University Press, 251–75. Lasonen Aarnio, Maria. 2010. Unreasonable Knowledge. Philosophical Perspectives 24: 1–21. Lehrer, Keith and Thomas Paxson. 1969. Knowledge: Undefeated Justified True Belief. Journal of Philosophy 66: 225–37. Lewis, David. 1983. New Work for a Theory of Universals. Australasian Journal of Philosophy 61: 343–77. Littlejohn, Clayton. 2012. Justification and the Truth Connection. Cambridge: Cambridge University Press. Littlejohn, Clayton. 2013. The Russellian Retreat. Proceedings of the Aristotelian Society 113: 293–320. Littlejohn, Clayton. Forthcoming. A Plea for Epistemic Excuses. In Fabian Dorsch and Julien Dutant, eds., The New Evil Demon Problem. Oxford: Oxford University Press. McDowell, John. 1982. Criteria, Defeasibility and Knowledge. Proceedings of the British Academy 68: 455–79. MacFarlane, John. 2011. What is Assertion? In Jessica Brown and Herman Cappelen, eds., Assertion: New Philosophical Essays. Oxford: Oxford University Press, 79–96. McGlynn, Aidan. 2014. Knowledge First? New York: Palgrave Macmillan. McKinnon, Rachel. 2013. The Supportive Reasons Norm of Assertion. American Philosophical Quarterly 50: 121–35. Maitra, Ishani. 2011. Assertion, Norms, and Games. In Jessica Brown and Herman Cappelen, eds., Assertion: New Philosophical Essays. Oxford: Oxford University Press, 277–96. Martin, Michael. 2002. The Transparency of Experience. Mind and Language 17: 376–425. Nozick, Robert. 1981. Philosophical Explanations. Cambridge, MA: Harvard University Press. Pritchard, Duncan. 2005. Epistemic Luck. Oxford: Oxford University Press. Pritchard, Duncan. 2012. Epistemological Disjunctivism. Oxford: Oxford University Press. Reynolds, Steven. 2002. Testimony, Knowledge, and Epistemic Goals. Philosophical Studies 110: 139–61. Russell, Bertrand. 1912. The Problems of Philosophy. London: Williams and Norgate. Schaffer, Jonathan. 2008. Knowledge in the Image of Assertion. Philosophical Issues 18: 1–19. Schechter, Joshua. 2013. Rational Self-Doubt and the Failure of Closure. Philosophical Studies 163: 429–52. Shope, Robert. 1983. The Analysis of Knowing: A Decade of Research. Princeton: Princeton University Press. Sliwa, Paulina. 2014. Understanding and Knowing. Proceedings of the Aristotelian Society 115: 57–74.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

no need for excuses  159 Slote, Michael. 1979. Assertion and Belief. In Jonathan Dancy, ed., Papers on Language and Logic. Keele: Keele University Press, 177–90. Snowdon, Paul. 1980. Perception, Vision and Causation. Proceedings of the Aristotelian Society 81: 175–92. Sosa, David. 2009. Dubious Assertions. Philosophical Studies 146: 269–72. Stanley, Jason. 2005. Knowledge and Practical Interests. Oxford: Oxford University Press. Strawson, Peter. 1962. Freedom and Resentment. Proceedings of the British Academy, 48: 1–25. Sutton, Jonathan. 2007. Without Justification. Cambridge, MA: MIT Press. Turri, John. 2010. Epistemic Invariantism and Speech Act Contextualism. Philosophical Review 119: 77–95. Unger, Peter. 1975. Ignorance: A Case for Skepticism. Oxford: Clarendon Press. Weiner, Matthew. 2005. Must We Know What We Say? Philosophical Review 114: 227–51. Weiner, Matthew. 2009. Practical Reasoning and the Concept of Knowledge. In Adrian Haddock, Alan Millar, and Duncan Pritchard, eds., Epistemic Value. Oxford: Oxford University Press, 163–82. White, Roger. 2014. What is My Evidence that Here is a Hand? In Dylan Dodd and Elia Zardini, eds., Scepticism and Perceptual Justification. Oxford: Oxford University Press, 298–322. Whiting, Daniel. 2013. Stick to the Facts: On the Norms of Assertion. Erkenntnis 78: 847–67. Williamson, Timothy. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, Timothy. 2013. Knowledge First. In Matthias Steup, John Turri, and Ernest Sosa, eds., Contemporary Debates in Epistemology, Second Edition. Malden, MA: Wiley-Blackwell, 1–9. Williamson, Timothy. Forthcoming. Justifications, Excuses, and Sceptical Scenarios. In Fabian Dorsch and Julien Dutant, eds., The New Evil Demon Problem. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Pa rt I I

Applications and New Directions

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

8 Acting on Knowledge Timothy Williamson

1. Analogies Between Knowledge and Action Knowledge and its Limits starts its exposition of the knowledge-first approach to epistemology with a structural analogy between knowledge and action as the two key relations between mind and world (Williamson 2000, 1, 6–8). This chapter aims to reconsider the relation between knowledge and action, and refine the analogy.1,2 Consider the opposition between knowledge-first and belief-first epistemology— the approach still taken for granted without argument by many contemporary epistemo­logists. Methodologically, knowledge-first epistemology starts with the distinction between knowledge and ignorance, belief-first epistemology with the distinction between belief and unbelief—typically with a further distinction of belief into justified and unjustified, and perhaps another into degrees of belief. On a popular view in the philosophy of mind, the core of all intentional mental life is ‘belief-desire ­psychology’. Belief and desire are paired. Typically, they are understood as propositional attitudes with opposite directions of fit (Humberstone  1992). With belief, we are ­supposed to fit the mind to the world, with desire to fit the world to the mind. In Anscombe’s famous comparison, the shopper tries to fit his purchases to his shopping list, while the detective shadowing him tries to fit her list to his purchases (1957: 56). Belief and desire are supposed to be mirror images, in a mirror that reverses direction of fit. Thus, it is natural to ask: what stands to desire as knowledge stands to belief? When all goes well in fitting the mind to the world, there is knowledge. Therefore, what

1   Earlier versions of this material were presented at the Universities of Geneva, Oxford, Stanford, Sussex, and Xiamen, the Collège de France, a conference on the Factive Turn at the University of Vienna, a Summer School at the University of Cologne, and as the 2015 Gustav Bergmann Lecture at the University of Iowa. I thank the audiences there and two anonymous referees for helpful comments and questions. 2   For exposition, development, and defence of my knowledge-first approach in general see Williamson (2007: 247–77, 2009, 2011, 2013a,b). For an account of action in a similar spirit to the one here, see Levy (2013). See also Hyman (2015) for an account of action coordinated with a broadly knowledge-first approach to epistemology. Of course, some analogies between knowledge and action can be developed even outside the knowledge-first framework; Sosa (2010) is a recent example.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

164  Timothy Williamson stands to desire as knowledge stands to belief should presumably be what there is when all goes well in fitting the world to the mind. For example, contrast believing that the window is open with desiring that the window be open. In a simple case of knowing that the window is open, one believes that the window is open because the window is open. In a simple case with the reverse direction of fit, the window is open because one desires that the window be open. One intentionally brought it about that the window is open, for instance, by opening it or asking someone else to do so. Intentionally bringing something about is what there is when all goes well in fitting the world to the mind. In that sense, action is what stands to desire as knowledge stands to belief. Here and henceforth, ‘action’ is to be read as ‘intentional action’; ­eating is acting but digesting is not. Knowledge and its Limits developed that structural analogy between knowledge and action, while reversing the usual explanatory order by treating knowledge as more fundamental than belief and action as more fundamental than desire. Call ‘cognitive’ those aspects of intelligent life which concern fitting mind to world, and ‘practical’ those which concern fitting world to mind. Thus, it is proposed, knowledge stands to the cognitive as action stands to the practical. If so, to marginalize knowledge in one’s account of things cognitive, as many epistemologists have done, is as perverse as it would be to marginalize action in one’s account of things practical. The analogy between knowledge and action favours a knowledge-centred approach to epistemology. This chapter substantiates the analogy between knowledge and action. However, its way of doing so is not quite the same as in Knowledge and its Limits. For, if action is the image of knowledge in the mirror reversing direction of fit, it does not follow that desire is the image of belief. Desire entered the original analogy because it is paired and contrasted with belief in the standard internalist picture of belief-desire psychology as  the core of intelligent life. But that picture itself goes naturally with belief-first ­epistemology. If we start with knowledge and action instead, we may see belief as corresponding more naturally to something other than desire. For desire is not connected more closely to action than belief is. Indeed, even the once-standard view emphasizes that belief and desire are needed equally, and more or less symmetrically, as inputs to practical reasoning if it is to issue in action. On that view, to get chocolate, an intense desire for it is impotent without a belief about how to get it, just as that belief is impotent without the desire.3 Beliefs and desires are together meant to supply the premises of practical reasoning—its inputs. If we seek something closer to action, as the analogy requires, the obvious place to look is at the conclusion of the practical reasoning, i.e., its output, not at the premises. But the conclusion of a piece of practical reasoning should not be just another desire; it should be an intention to do or be something. Indeed, Anscombe describes the shopping list written by the shopper as an expression of ­intention, not merely of desire. In forming the intention to φ, one puts φing on one’s   The belief might be a partial degree of belief.

3

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  165 to-do list. This suggests that we can improve the analogy by substituting intention for desire: knowledge is to belief as action is to intention.4 This chapter substantiates that more refined analogy. Where does desire fit into the refined analogy? What is the mirror image of desire as knowledge is the mirror image of action and belief of intention? Suppose that desire is on the same side of the analogy as action. Then, since desire is further than intention from action, the mirror image of desire should be on the same side of the analogy as knowledge, but further than belief from it. That might make it some sort of appearance, suspicion, or level of credence short of outright belief. But then the mirror image of desire is further than desire itself from action, since it is further than outright belief from action, and outright belief is at least as far as desire from action. That is an awkward result, for belief and desire combine as inputs to practical reasoning. A very different proposal is instead to put desire on the same side of the analogy as knowledge, and assimilate desire to belief. On such a desire-as-belief view, very roughly, to desire something is to believe that it is good. To desire that P is to believe that it is good that P. Here ‘it is good that P’ is to be read as neither entailing nor presupposing that P; one might say ‘it would be good if P’ or ‘it would be good for it to be the case that P’. The relevant sense of ‘good’ is generic.5 In particular, it is by no means exclusively moral; a chocolate cake can look good.6 On the desire-as-belief view, it is presumably irrational to desire something that is unlikely on one’s evidence to be good: a reasonable result. As for direction of fit, the idea is, roughly, that the desire that P requires fitting world to mind with respect to the proposition that P, but fitting mind to world with respect to the proposition that it is good that P (Humberstone 1992: 61). The desire-as-belief view helps to differentiate desire from intention. Buridan’s ass does not, and should not, believe that the bale of hay on the right is better than the bale on the left, or that the bale on the left is better than the bale on the right. He can still avoid starvation by intending to eat the bale on the right, and not intending to eat the bale on the left (first), or by intending the reverse. He has no need to desire the bale on the right more than the bale on the left, or vice versa. In a loose sense, we can apply the 4   See Holton (2008; 2014) for an extended case for the analogy between outright belief and outright intention, both of which are understood as stable all-or-nothing states. 5   Humberstone (1987) presents a strong case for the desire-as-belief view, based on considerations of expressibility. That Humberstone uses ‘desirable’ where I use ‘good’ does not reflect any disagreement between us. For a critique of the desire-as-belief view see Lewis (1988; 1996). For two responses to Lewis see Byrne and Hájek (1997) and Bradley and List (2009). However, unlike Humberstone, Lewis and those responding to him focus on degrees of desire as degrees of belief, whereas the present concern is with outright desire as outright belief. 6   Of course, one can always refuse to believe that things are as they appear. The cake may look delicious even after one has tried it and knows it to be disgusting (compare the Müller-Lyer illusion). One may still feel a strong urge to eat the cake, which one controls, because one does not believe that it is good to eat it. On the desire-as-belief view, one does not desire to eat it; one refuses the temptation to form the desire/ belief. That verdict is supported by the absence of any such desire from one’s practical reasoning, which is quite different from giving some weight to it in the reasoning, but letting it be outweighed by other desires. Talk of the intensity of desires may conflate several independent dimensions: how good one is tempted to believe something is, how strong the temptation is, and how far one gives way to it.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

166  Timothy Williamson term ‘practical reasoning’ to the not-irrational process by which Buridan’s ass, if sensible, gets from its initial beliefs and desires to an intention to eat the bale on the right, or an intention to eat the bale on the left, despite the quasi-random nature of the process. Obviously, the desire-as-belief thesis raises much larger issues about the nature of the good than can be broached here. It will not be assumed in what follows, but it ­provides one attractive way of incorporating desire into the account below. In any case, we can continue to treat desires as inputs to practical reasoning, and intentions as outputs. Having selected intention as the most promising analogue to belief, we can now develop the analogy between knowledge and action in more detail, although still somewhat schematically.

2.  The Analogy with Intention We can tabulate the proposed analogy: knowledge belief truth falsity fitting mind to world input to practical reasoning

action intention success failure fitting world to mind output from practical reasoning

Of course, each column must be understood with respect to a fixed content. On the left-hand side, when the knowledge condition is that one knows that P, the belief condition is that one believes that P, the truth condition is that P, and the falsity condition is that not P. Similarly, on the right-hand side, when the action condition is that one intentionally φs, the intention condition is that one intends to φ, the success condition is that one φs, and the failure condition is that one does not φ. If one so preferred, one could use ‘realization’ or ‘fulfilment’ in place of ‘success’ and ‘non-realization’ or ‘non-fulfilment’ in place of ‘failure’. Whatever words one uses as the correlates on the action side of ‘truth’ and ‘falsity’, care is needed. They must be read with respect to the specific value of ‘φ’ at issue. When I shoot at a target and miss, my intention to shoot succeeds, but my intention to hit the target fails. Similar care is needed with ‘intentionally’. When I throw a dart at a ­dartboard and hit, I intentionally hit some point on the board but do not intentionally hit that particular point. An apparent asymmetry between the two columns is that the contents on the knowledge side were just treated as propositional while those on the action side were not. More precisely, the schematic letter ‘P’ used in explaining the left column stands

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  167 for a declarative sentence (such as ‘The door is open’), while the schematic letter ‘φ’ used in explaining the right column stands for a verb phrase (such as ‘open the door’). However, this asymmetry is largely an artefact of presentation. The verb phrase still needs a subject, for instance, ‘one’ in the schemata above. Even in ‘One intends to φ’, the unpronounced implicit subject of ‘to φ’ is mandatorily coreferential with ‘one’. When the subject of the embedded verb phrase differs from the subject of ‘intend’, it must be made explicit.7 For instance, you may intend a room to be used by guests (more awkwardly: you intend that the room be used by guests). A controlling mother may intend her son to become a doctor (more awkwardly: she intends that he become a doctor). Similarly, if I name the shopper who is leaving a trail of sugar ‘Perry’, I may intend Perry to be warned that he is making a mess, without realizing that I am Perry (at least, in those very words). Even when the speaker makes the co-reference explicit to exclude such confusions of identity, that still makes the intender sound self-estranged. ‘Joe intends himself to have a drink’ suggests that Joe may have suffered brain damage, and be capable of getting himself a drink only by some compensatory strategy that involves directing himself as he might direct someone else. But that is exactly what one would expect. For if the speaker envisaged a normal case of Joe intending to have a drink, going to the trouble of adding the redundant word ‘himself ’ would be pointless and prolix, thereby violating Grice’s conversational maxim of manner (Grice 1989: 27). Thus, hearers are entitled to assume that the speaker added ­‘himself ’ to signal something abnormal in how the intender represents himself in the intention; the natural hypothesis is that he represents himself in a way ­similar to the way he might represent someone else in a typical use of ‘Joe intends him to have a drink’, that is, in a third-personal way. The difference between ‘Joe intends to have a drink’ and ‘Joe intends himself to have a drink’ is a non-semantic conversational effect, predictable on quite general grounds; it provides no evidence against the assumption that the objects of intentions are as propositional as the objects of belief.8 Given what has been said, the success condition for an intention to bring it about that P is that one brings it about that P, not merely that somehow or other P. If you intend to open the door, but the wind blows it open before you can get there, your intention failed, because you did not do what you intended to do. By contrast, if you merely intend the door to be open, that is a different and perhaps less ambitious intention, whose success condition is just that the door is open. It is crucial to respect such ­distinctions in what follows. 7   ‘Subject’ here means the grammatical subject of the verb, rather than the agent of the action described; the subject in ‘I intend to be cheered’ is ‘I’, not the cheerers. 8  To some ears, ‘Joe intends himself to have a drink’ sounds less strange when the time of the intended drinking is comparatively distant. Presumably, the reason is that one’s present relation to one’s actions in the distant future tends to be less distinctively first-personal in nature than is one’s present relation to one’s actions in the immediate future (though not wholly third-personal either).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

168  Timothy Williamson Like the contents of beliefs, the contents of intentions need not be about the agent. Nevertheless, in practice, it may be more common for intentions than beliefs to be about the agent, and for those that are not to be mediated by those that are. The value of intending the book to be on the table normally depends on also intending to put it on the table, or to ask someone else to do so, or the like. By contrast, the value of believing that the book is on the table does not normally depend on also believing that one can see that it is on the table, or that it appears to one that it is on the table, or the like. The range of what we reasonably believe is much wider than the range of what we reasonably intend. We can see further than we can reach. These ‘statistical’ differences are significant, both robust and important, but they do not undermine the structural similarities below. Although the grammatical asymmetry between ‘believe that P’ and ‘intend to φ’ can be misleading, as just noted, I will continue to use both forms, since they are the most natural ones. They concentrate attention on typical cases. If anything, they make my task of bringing out the analogies harder rather than easier. Moreover, the ‘intend to φ’ construction is in principle less limited than it seems, since ‘intend X to φ’ and ‘intend that P’ are at least roughly equivalent to the admittedly rather artificial forms ‘intend to be such that X φs’ and ‘intend to be such that P’ respectively, where ‘Y is such that P’ is read as trivially equivalent to ‘P’, not as implying some dependence on Y.9 For instance, you intend to be such that the room is used by guests, and the mother intends to be such that her son becomes a doctor. When all goes well, you are intentionally such that the room is used by guests, and the mother is intentionally such that her son becomes a doctor. The ‘intentionally’ may require some dependence on you and the mother respectively, even though the ‘such that’ does not. The range of the variable ‘φ’ is not limited to paradigmatic actions. One may intend to stay in bed and not to get up, or to be spontaneous, or whatever. For typical action verbs ‘φ’ there is a temptation to treat ‘φ’ as equivalent to ‘intentionally φ’, because φ-ing is usually intentional. For instance, most talking is intentional talking. However, the temptation should be resisted, for several reasons. First, it risks making the meanings of ordinary verbs unnecessarily paradoxical, since the content of the intention (talking) would already involve that very intention. Second, it gives the wrong results for agents who classify what they are doing differently from the speaker. Someone may yell without intentionally yelling because their intention is merely to speak. Third, the tendency to treat ‘φ’ as implying ‘intentionally φ’ may be explained as a Gricean conversational implicature, since once it is commonly assumed that normal φ-ing is intentional φ-ing, hearers will expect speakers to warn them of abnormalities, and so take the absence of warning to indicate normality. In any case, it aids clarity to consider verbs where both 9   The equivalence is like that between the sentences λx(A)t and A in a formal language, where in λx(A)t the predicate λx(A) is applied to the singular term in subject position t, and the variable x does not occur free in A. Note also that one must take care when generalizing the equivalences: the definite description in ‘intend the F to φ’ may easily be read as taking wider scope than ‘intend’, and so must then be read accordingly in ‘intend to be such that the F φs’.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  169 intentional and unintentional cases are within the bounds of normality. One can drop a hat intentionally or unintentionally. In such examples, the contribution of ‘intentionally’ stands out. The difference in grammatical role between the adverb in ‘intentionally φs’ and the verb in ‘knows that P’ also obscures the underlying symmetry. For instance, intentionally shouting is an action while knowing that one shouted is a state. Normally, one knows that one shouted after one shouted, as long as one still believes that one shouted, but one intentionally shouts only while shouting, not as soon as one intends to shout. This looks like a significant metaphysical difference. But it results from our comparing the main verbs of the respective sentences: the main verb of ‘I intentionally shouted’ is ‘shouted’ whereas the main verb of ‘I knew that I shouted’ is ‘knew’, not ‘shouted’. One could restore the grammatical and metaphysical symmetry by comparing ‘I was intentionally such that I shouted’ with ‘I was knowingly such that I shouted’, and more generally ‘S is intentionally such that P’ with ‘S is knowingly such that P’, sacrificing ­linguistic naturalness for the sake of fine-grained accuracy. For present purposes, the more natural constructions are good enough.

3.  Knowing is to Believing as Acting is to Intending We can now develop the structural analogy in more detail. The most obvious difference between knowledge and belief is that knowledge entails truth while belief does not; it is compatible with falsity. If you know that P, it follows that P, but sometimes even though not P you still believe that P.10 The most obvious difference between action and intention is that action entails ­success, while intention does not; it is compatible with failure. If you intentionally φ, it follows that you φ, but sometimes even though you never φ, you still intend to φ. Since knowledge entails truth and belief does not, a fortiori belief does not entail knowledge; otherwise, it would have to entail truth, too. But it is very plausible that the converse does hold: knowledge entails belief. When you know that P, you believe that P. Given a case where you know that P, typically in some possible variant case you believe that P, but things went wrong earlier in the process and not P; since your belief is just as firm in the former case, you believe that P there, too.11 10   People are occasionally misled into denying that knowledge entails truth. Telling a story dramatically, one may say ‘He knew that the shop was only a mile away, so he decided to walk’, then later reveal that he was mistaken: the shop was twenty miles away. In such cases, one uses language projectively, to show how things looked from the protagonist’s point of view, not how they really were. One trusts hearers to understand, at least by the end of the story (unfortunately, if they are literal-minded, they may not). One might just as easily have said, The shop was only a mile away […] it turned out to be twenty miles away’, speaking projectively. One would not thereby have been using the phrase ‘a mile’ to mean twenty miles. Similarly, when one said ‘He knew that the shop was only a mile away […] it turned out to be twenty miles away’. one did not thereby use ‘know’ non-factively. Any words whatever can be used projectively, so projective uses show nothing special about the meaning of any word in particular. 11   One might get the contrary impression because it is sometimes misleading to attribute belief but not misleading to attribute knowledge. If you know that I know that Mary is in China, it would be misleading

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

170  Timothy Williamson Since action entails success and intention does not, a fortiori intention does not entail action, otherwise it would have to entail success too. But it is very plausible that the converse does hold: action entails intention. When you intentionally φ, you intend to φ. Given a case where you intentionally φ, typically, in some possible variant case you intend to φ, but things go wrong later in the process and you fail to φ; since your intention is just as firm in the former case, you intend to φ there, too. Knowledge entails true belief. If one knows that P, P, and one believes that P. But not  the converse: not even true belief entails knowledge. Sometimes, even though P and one believes that P, one does not know that P. If Mary is in town and I believe that she is in town simply through misidentifying someone else as her, I do not know that she is in town.12 Action entails successful intention. If one intentionally φs, one φs and intends to φ. But not the converse: not even successful intention entails action. Sometimes, even though one φs and intends to φ, one does not intentionally φ. A would-be assassin may accidentally run over and kill the man he intends to kill, without intentionally killing him. Since belief and truth are necessary but not even jointly sufficient for knowledge, many epistemologists have found it natural to ask: what must be added to belief and truth to make knowledge? Schematically, they seek a solution in ‘X’ of the equation: knowledge = belief + truth + X In other words, a necessary and sufficient condition for S to know that P is to have this conjunctive form: (K1) S believes that P; (K2) P; (K3) S satisfies condition X with respect to ‘P’. Such a solution might be regarded as an analysis, or even definition, of knowledge, stating not just necessary and sufficient conditions for knowing, but what knowledge is in some more fine-grained sense. A crucial constraint on a solution is that X must not be specified in terms of ‘know’, on pain of circularity. For example, one for you to say ‘TW believes that Mary is in China’, because doing so might cause your hearers falsely to believe that you do not know that I know that Mary is in China. For they might reason thus: if you did know that I knew that Mary was in China, you would have made the more informative and useful statement ‘TW knows that Mary is in China’ instead; therefore, since you did not make the latter statement, you did not know that I knew that Mary was in China. Thus, on the assumption that knowledge entails belief, we can still explain why saying ‘TW believes that Mary is in China’ is true but conversationally misleading when saying ‘TW knows that Mary is in China’ is true and not conversationally misleading (Grice 1989). Such cases pose no threat to the proposed entailment from knowledge to belief. For more discussion of the relation between knowledge and belief see Williamson (2000: 41–8). One might even suspect that believing is too weak a condition to be appropriately paired with knowledge (for arguments that belief is weak see Hawthorne, Rothschild, and Spectre 2016). Perhaps knowing that P requires being sure that P (as proposed by Ayer 1956: 34), in an appropriately undemanding non-Cartesian sense of ‘sure’, not just believing that P (thanks to Jeremy Goodman for discussion on this issue). One might understand both being sure and intending as states of having made up one’s mind: to be sure that P is to have made up one’s mind that P; to intend to φ is to have made up one’s mind to φ. The present chapter could easily be reworked along such lines. For the sake of familiarity, I stick with ‘belief ’ here. 12   We sometimes sloppily count all the students who gave the correct answer in a quiz as ‘knowing the answer’, even if they were just guessing. That does not show that ‘knowledge’ ever means true belief. We sloppily count them as knowing even if they do not believe their answers to be correct.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  171 would get an almost trivially true equation by substituting ‘S knows that P’ for ‘X’, but it would not constitute a definition or analysis of the kind sought. Since intention and success are necessary, but not even jointly sufficient, for action, many philosophers of action have found it natural to ask: what must be added to intention and success to make action? Schematically, they seek a solution in ‘Y’ of the equation: action = intention + success + Y In other words, a necessary and sufficient condition for S to intentionally φ is to have this conjunctive form: (A1) S intends to φ; (A2) S φs; (A3) S satisfies condition Y with respect to ‘φ’. Such a solution might be regarded as an analysis, or even definition, of action, stating not just necessary and sufficient conditions for action, but what action is in some more fine-grained sense. A crucial constraint on a solution is that Y must not be specified in terms of ‘action’, on pain of circularity. For example, one would get an almost trivially true equation by substituting ‘S intentionally φs’ for ‘Y’, but it would not constitute a definition or analysis of the kind sought. The attempt to solve the equation for knowledge generated an extensive, but still unsuccessful, research programme. Many solutions, of increasing complexity, were proposed, but unless circular, always eventually succumbed to counterexamples, of increasing complexity, to their necessity or sufficiency. The mind-to-world direction of fit for belief suggests an instructive case in point: causal theories of knowledge. At its simplest, such a theory identifies X with the condition that S believes that P because (causally) P. Such theories succumbed to the problem of deviant causal chains: the causation from truth to belief may be of the wrong kind for knowledge. For example, a brain disease causes someone to misinterpret his doctor as telling him that he has that disease, which causes him to believe that he has it. The three conditions (K1)–(K3) hold, but he does not know that he has the disease. Attempts are still made to solve the  problems for attempted analyses of knowledge by adding epicycles, but none has worked.13 The attempt to solve the equation for action generated an extensive but still unsuccessful research programme. Many solutions were proposed, of increasing complexity, but unless circular always eventually succumbed to counterexamples, of increasing complexity, to their necessity or sufficiency. The world-to-mind direction of fit for intention suggests an instructive case in point: causal theories of action. At its simplest, such a theory identifies Y with the condition that S φs because (causally) S intends to φ. Such theories succumbed to the problem of deviant causal chains: the causation from intention to success may be of the wrong kind for action. For example, my intention to drop the Ming vase causes my hands to sweat and become slippery, which in turn 13   Goldman (1967) is the classic presentation of a causal theory of knowledge, already elaborated in response to various difficulties. He later abandoned it (1976), though for different reasons from those ­discussed here. Shope (1983) details many failed attempts to analyse knowledge.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

172  Timothy Williamson causes me to drop the vase. The three conditions (A1)–(A3) hold on the proposed reading, but I do not intentionally drop the vase. Attempts are still made to solve the problems for attempted analyses of action by adding epicycles, but none has worked.14

4.  Directions of Explanation Why expect the equations for knowledge and action in Section 3 to have solutions? One might think: since belief and truth are parts of knowledge, and intention and success are parts of action, it is just a matter of identifying the respective remaining parts. But such thinking relies on a naïve application of the part–whole distinction. Here is an analogy. Being coloured is necessary for being red. One might, therefore, wonder what other necessary conditions for being red one must add to being coloured to make a condition sufficient as well as necessary for being red. Such a conjunction, presented in terms of those individually necessary and jointly sufficient conditions, might be regarded as a definition of redness, an analysis of being red into its component parts. This would correspond to the attempt to solve this equation in the variable ‘Z’: red = coloured + Z But what could Z be? One gets an almost trivially true equation by substituting ‘red’ itself for ‘Z’, or ‘red if coloured’, but that would violate the usual non-circularity ­constraint. Even if being red can be reduced to a property specified in physical terms, that would not fit the displayed pattern, since it would lack ‘coloured’ as a component. The moral is: do not assume that a necessary condition for something is a component of a non-circular definition or analysis of that thing. Thus, although knowledge entails belief and truth, and action entails intention and success, that is not good evidence that those equations for knowledge and action have solutions. The equations have a more specific attraction for internalists about the mind, who hold that purely mental phenomena are constitutively (not causally) independent of the world external to the subject. To them, the equations look like promising first steps towards separating out knowledge and action into their postulated purely mental and purely non-mental components, with belief and intention on the purely mental side and truth and success on the purely non-mental side, at least to a first approximation. Of course, even internalists will admit that it is not as simple as that. For one thing, we have beliefs about our mental states: in particular, we have beliefs about our own beliefs and about our own intentions. Thus, truth does not always fall on the nonmental side. We also have intentions about our own mental states: for instance, to think about a mathematical problem. Thus, success does not always fall on the non-mental 14   The essays on action in Davidson (2001) provide a good entry-point for this debate. See White (1968) for the state of philosophy of action simultaneous with early reactions to Gettier. Of course, as with accounts of knowledge in terms of belief, accounts of action in terms of intention do not all take the form of the displayed equations; see Sosa (2015). Obviously, space and time do not permit arguments here and now against all the attempts that have been made.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  173 side either. For another thing, belief and intention are not purely mental states by ­internalist standards, since their contents typically depend constitutively on the e­ xternal environment (Putnam  1973, Burge  1979). The internalist aspiration to separate the purely mental from the purely non-mental remains at a largely programmatic stage. Indeed, the failures of all the many attempts to analyse knowledge and action in such terms provide powerful inductive evidence against those reductive ways of thinking. The project of analysis is, in Imre Lakatos’s phrase, a degenerating research programme. Nor should we assume that the constitutive dependence of belief and intention on the external is limited to their contents. It extends to the attitudes to those contents, too. To be intentions, rather than mere fantasies, mental states must be somehow poised to produce action. And to be beliefs, rather than mere fantasies of another flavour, other mental states must be somehow poised to lead through practical reasoning to intentions. Of course, the contents of some beliefs never, in fact, occur as premises in practical reasoning, and the contents of some intentions never, in fact, occur as a conclusion in  practical reasoning (intentions may form through some non-rational process). Nevertheless, belief is a sort of state disposed to make the premises of practical reasoning, and intention is a sort of state disposed to make the conclusion of practical reasoning, even if those dispositions are not always manifested.15 No belief ’s content immunizes it from relevance to action, for example in deciding whether to take a bet, how to answer in a quiz, or what to say in conversation. If you have no disposition whatever to rely on it as a premise in practical reasoning, it is no belief. This is not to deny that a brain in a vat could have beliefs and intentions, linked by practical reasoning, for the brain’s intentions could still be disposed to produce action, even if its unfortunate circumstances prevent the disposition from manifesting itself appropriately.16 But without even such unmanifested dispositions, we have, at best, a system of states misinterpretable as beliefs and intentions. It is obviously wrong-headed to try to understand intention without reference to action. Given the connection between belief and intention, it is also wrong-headed to try to understand belief without at least indirect reference to action. Furthermore, if knowledge is to belief as action is to intention, it is wrong-headed to try to understand belief without reference to knowledge, or to try to understand intention without at least indirect reference to knowledge. A natural starting-point is the case of acting on what one knows. Practical reasoning is the primary nexus between knowledge and action; it puts the two sides of the structural analogy together. In basic cases of acting on what one knows, causation 15   The disposition must be stronger than the overridden disposition in fn. 6 to assume that the chocolate cake is delicious. 16   When an unkind external environment makes reference fail, something locally like practical reasoning may occur even though no content is provided for what plays the local role of knowledge or action, so there is no content to be believed or intended, so no belief or intention. The local role determines at most the attitude of belief or intention to the content, if any. Nor is it obvious that what makes reasoning practical is local.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

174  Timothy Williamson flows from the external environment through perception to the brain and back through intervention into the external environment. Of course, not all knowledge is just perception of the external environment, not all action is just intervention into that environment, and beyond the most basic cases, complex feedback loops involve cycles of both perception and action. Not only does knowledge generate action, but also action generates knowledge; for example, through actions of looking, listening, touching, and investigating more generally. It is not mysterious in principle why the ability to acquire knowledge and act on it should give a creature an advantage in dealing with its environment (including other members of the same species). Presumably, the gross evolutionary advantage of having a mind is that it enables one to interact with a complex, changing environment in ways more complex and flexible than can mindless things like plants. Acting on knowledge is the central case of such interaction. When all goes well in one’s whole cognitive-practical system (not just one’s brain), one acts on what one knows. But things often go badly. One believes that P when not P, or (perhaps as a consequence) intends to φ, but never φs. In practical reasoning, the false belief plays the same local role as knowledge (it makes a premise), and the failed intention plays the same local role as action (it makes the conclusion), but they do not play the same global roles. Such malfunctions have to be understood in relation to what happens when the whole system is functioning well. Why do mere true belief (short of knowledge) and mere successful intention (short of action) not constitute functioning well? In such cases, one may still get what one wants or needs. But for a system to bring about what is immediately needed or wanted is not generally enough for it to be functioning well. For example, suppose that it is the function of the heart to pump the blood. Imagine that, as a result of an injury, someone’s heart in some bizarre way causes their blood to circulate by means other than pumping. The heart brings about what is immediately needed and wanted, but it is not functioning properly, since it is not pumping the blood. Similar problems arise even if the function of the heart is characterized more abstractly, for instance, as being to ­regulate the flow of blood. In some still more bizarre way, the heart may cause something else to intervene and regulate the flow of blood. Again, the heart brings about what is immediately needed and wanted, but it is not functioning properly, since it is not regulating the flow of blood. Thus, on quite general grounds, one would expect cases where the cognitive system malfunctions; there is no knowledge, but there is still true belief. On the same general grounds, one would expect cases where the practical system malfunctions; there is no action, but there is still successful intention. When the cognitive-practical system functions well, one acts on what one knows. Other cases resemble that nexus locally, but not globally. On the basis of such local resemblances, the nexus still counts as practical reasoning, the premises still count as believed, even if not known, and the conclusion still counts as intended, even if not acted on. Belief is what tends to play the local role of knowledge with respect to action, whether or not it plays the global role. Similarly, intention is what tends to play the local role of action with respect to knowledge, whether or not it plays the global role.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  175 These remarks are too schematic to constitute analyses of belief and intention in terms of knowledge and action. Nevertheless, they involve a direction of explanation from more global distinctions between good and bad functioning, to more local distinctions neutral between good and bad functioning. Cognitive malfunctioning is understood as a deviation from good cognitive functioning, and practical malfunctioning as a deviation from good practical functioning. A natural worry about that order of explanation is that it seems to turn its back on an explanatory strategy that has proved immensely successful in the natural sciences and elsewhere: understanding a system by analysing it into its local components and studying their local interactions. To understand how the heart pumps blood, we must employ that strategy. Similarly, to understand how agents act on what they know, will not we have to analyse the cognitive-practical system into local components such as belief and intention, and study their local interactions with each other and with ­relevant bits of the environment? Do not cognitive science and cognitive psychology flourish by already following such a decompositional strategy? Of course, we have already seen that categories such as belief and intention are themselves functional, in ways unsuited to the full decompositional strategy. Nevertheless, they might be regarded as somehow prefiguring the sought-for analyses in more detail than do knowledge and action. What is not in question is that, in any particular case, knowledge and action, or belief and intention, must be implemented by specific mechanisms susceptible to decompositional analysis. Corresponding explanatory theories of those mechanisms can, in principle, be given. But, in general, a given functional kind can be implemented by many different mechanisms. For example, pumps can be classified by how they move the fluid: there are direct lift, displacement, and gravity pumps. Similarly, a theory of a mechanism that implements one sort of knowledge will not apply to knowledge of other sorts. It may apply to visual but not to auditory knowledge, or to human but not to extra-terrestrial knowledge. A trade-off is likely between a more detailed theory at a lower level of generality and a less detailed theory at a higher level of generality. The choice depends on the theorist’s aims. Like most philosophers, epistemologists and philosophers of action typically ­operate at a very high level of generality. They seek and propose principles about, or  analyses of, all knowledge or belief, all action or intention—actual or possible. Although, for obvious reasons, their examples tend to be of human knowledge or belief, human action or intention, they rarely write any such restriction into their ­theories. This aspiration to a high level of generality helps explain their attitudes to cognitive psychology. Even if it were discovered that all human knowledge had a given property π, most epistemologists would not take that to justify including π in their theory of knowledge if there could be creatures whose knowledge lacked π. Indeed, theories in epistemology and the philosophy of action are typically supposed to apply to infinite minds as well as finite ones. Analyses are supposed to provide necessary and sufficient conditions.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

176  Timothy Williamson Thus, epistemologists and philosophers of action typically lack the implementationoriented aims and methods that would justify them in applying the decompositional strategy. Belief-centred epistemology and intention-centred philosophy of action should gain no prestige by association with the achievements of that strategy in the natural and social sciences; the level of generality is too different. In both philosophy and the natural and social sciences, rigorous methods can often be applied at a high level of generality when the decompositional strategy is inappropriate. If you want to find out, in general, how many objects of a given shape and size (say, balls) can be packed into a container of a given shape and size (say, a cubical box), do not ask what the objects are made of; use geometry, not chemistry. In knowledge-centred epistemology, logic and mathematics are applicable at equally high levels of rigour and generality: for instance, epistemic logic and Bayesian probability theory. The decompositional strategy poses no threat to epistemology centred on knowledge rather than belief, or to philosophy of action centred on action rather than intention.17 Another feature of epistemology and the philosophy of action also distances them from descriptive, decompositional inquiry: both are typically pursued as normative or prescriptive inquiries, by internalists as much as by externalists. They discuss reasons for belief and reasons for action, apply standards of rationality to both, and assess ­practical reasoning as valid or invalid. The distinction between justified and unjustified belief is central to much internalist epistemology. Violations of the norms are ­recognized as such, in order not to be counted as counterexamples to principles about the normative categories. The normative aspirations of epistemology and the philosophy of action are sometimes thought to pose a different threat to the externalist forms of those disciplines, on the grounds that agents can only be governed by norms accessible to them. That is, agents are supposed to be always in principle in a position to determine whether they satisfy a given governing norm. It is then assumed that internalist norms of rationality or justification are accessible to agents in that sense, while externalist norms are not: in sceptical scenarios, we are in no position to recognize our own failures to know or to act. Although we can be judged with respect to norms that obviously violate the accessibility constraint, such as norms of keeping our promises and of obeying the law, internalist norms of rationality are supposed to be somehow special in a sense that excludes most of the familiar norms. But it is arguable that no non-trivial norms satisfy 17   This chapter discusses general states and actions themselves, rather than our words or concepts for them. But the internalist order of explanation is no better supported at the level of linguistic understanding or conceptual grasp. Native speakers of a natural language typically understand its common words not by dictionary definitions in simpler terms but by paradigm examples of its applications (in any case, they could not grasp all concepts by analysis into simpler concepts, on pain of an infinite regress). Even ‘know’ and ‘act’ can be learnt through paradigm examples. There is no evidence that paradigms of believing and of intending are in any sense prior to paradigms of knowing and of acting. If anything, one might expect the opposite, since the external conditions on knowing and acting make their absence more open to thirdparty observation. For a discussion of recent psychological evidence that the ability to ascribe knowledge is basic to the human capacity for mindreading see Nagel (2013).

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  177 the internalist accessibility condition.18 If so, our only non-trivial norms are externalist. When we recognize that the internalist constraint cannot be met, our reaction should not be to seek norms that come as close as possible to meeting the internalist constraint, however many epicycles that quest forces us to add. Rather, we should try to understand the mistakes about normativity that made such a constraint seem compulsory (rather than merely attractive) in the first place. Having done that, we should be open to simpler, more natural, and differently motivated norms, even if they violate the internalist constraint quite blatantly.19 The picture of mind outlined earlier predicts knowledge norms of a functional sort for both action and belief. When practical reasoning functions well, one acts on what one knows: one uses the premise that P only if one knows that P. Thus, if practical ­reasoning uses the premise that P when the agent does not know that P, the reasoning has functioned badly. That is a natural reading of the knowledge norm for action: act only on what you know (see Hawthorne and Stanley  2008 for more discussion). Moreover, since any belief plays the local role of knowledge in practical reasoning, as a premise, any belief short of knowledge can make practical reasoning malfunction. That is a natural reading of the knowledge norm of belief: believe only what you know. When all goes well cognitively, one believes that P only if one knows that P. When one merely believes that P without knowing that P, something has gone wrong cognitively. Similarly, when all goes well practically, one intends to φ only if (sooner or later) one intentionally φs. When one merely intends to φ, but never intentionally φs, something has gone wrong practically.

5.  Extending the Analogy Much exploration is needed to identify appropriate analogues on one side for phenomena on the other. Here are three examples. First, trying, intermediate between intending and acting. Trying to φ entails more than intending to φ, but less than intentionally φ-ing. Presumably, intentionally φ-ing entails trying to φ, which in turn entails intending to φ, but both converses fail. Since trying to φ does not entail φ-ing (some attempts fail), a fortiori it does not entail intentionally φ-ing. It is less trivial that intending to φ does not entail trying to φ. Attempts vary greatly in their seriousness, from the well-prepared climber struck by lightning just before reaching the summit to the ignorant tourist who gives up as soon as the mountain comes into view. Although one might try to classify even the mere intention to φ as a minimal attempt to φ, it is more natural to classify someone who died of a

18   See the anti-luminosity argument in Williamson (2000). For an explanation of how it avoids the objection in Berker (2008), see Srinivasan (2015a). Srinivasan (2015b) explores the normative implications of the arguments. 19   For a fuller discussion of some of the confusions underlying internalist thinking about norms of belief see Williamson Forthcoming.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

178  Timothy Williamson heart attack immediately on forming the intention to become a lion-tamer as not having had the chance to try. Trying resembles acting not just in the local context of practical reasoning (intention) but further forward downstream (towards its destination), although the resemblance need not go all the way. What is the analogue of trying with the opposite direction of fit? Suppose that the analogy works for a propositional attitude we may call blanking. Thus, blanking is intermediate between believing and knowing: knowledge blank belief

action attempt intention

More specifically, knowing that P entails blanking that P, which in turn entails believing that P, but both converses fail. Indeed, blanking that P does not even entail that P. Blanking resembles knowing not just in the local context of practical reasoning (belief) but further back upstream (towards its source), although the resemblance need not go all the way. We can expect a wide range of cases, just as for trying. But capricious, utterly irrational belief should not be enough for blanking. The agent’s cognitive faculties should not be malfunctioning too badly too close to their use in practical reasoning. Perhaps agents with good general cognitive dispositions in bad sceptical scenarios still blank that P when they believe that P. We might loosely call blanking ‘believing reasonably’. Second, an example from the other side of the analogy: (sensory) perceiving. We may assume that perceiving that P entails that P, indeed, that it entails knowing that P (Williamson 2000: 33-41). In some ways, perceiving is an even closer analogue than knowing to acting. Just as acting typically involves direct output from the agent to the external environment, so perceiving typically involves direct input from the external environment to the agent. Just as acting involves a causal flow from the agent, which constrains what can be intentionally done, so perceiving involves a causal flow to the agent, which constrains what can be perceived. For instance, if it is necessary that P, only in very limited cases can one intentionally be such that P or sensorily perceive that P; there is no similar limitation on knowing that P. But to replace knowing by ­perceiving as the analogue of acting would also incur significant theoretical costs, for the contents of perception play no distinctive role in practical reasoning; any content of knowledge will do. The underlying difference is that the category of practical reasoning is not perfectly symmetrical between the two sides of the analogy, since it is constrained primarily in terms of its output. In principle, we could have an analogous category constrained primarily in terms of its input, to the contents of perception; but such a category looks much less interesting, since it leaves the point of the reasoning unspecified. It is more fruitful to keep knowing as the analogue of acting, while

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  179 analogizing the special role of sensory perception in knowledge to the special role of something like bodily action in the more general category of action.20 Third, reasons for action. Given limitations of space, suffice it here to say that the evidence, though easy to misconstrue, strongly indicates that the reasons for which one acts are truths one knows. Anything else, at best, appears to be a reason for which one is acting. There are often said to be two sorts of reasons—objective reasons and subjective reasons—but that is like saying that there are two sorts of gold, objective gold and subjective gold, or real gold and apparent gold. Appearing to be something is not a way of being that thing.21 One methodological advantage of maintaining this close overall analogy between epistemology and the philosophy of action is that it enhances our ability to test ­theories on each side, because their analogues on the other side may be easier to evaluate. For instance, sceptical arguments tempt us to deny that we ever really know anything, but are we tempted to deny that we ever really act?

6. Conclusion ‘Knowledge first’ is not just a slogan in epistemology. It adverts to a systematic, unified view of the mind in the world, on which the central feature of intelligent life is acting on what one knows. In that sense, the view gives both acting and knowing explanatory priority over believing and intending. But if we ask what makes such action intelligent, the answer lies in its connection with knowledge. That makes sense of the slogan ‘Knowledge first’ not only in epistemology, but also in the philosophy of mind and action. On the view sketched in this chapter, what is fundamental to mind is not a bunch of monadic ‘qualitative’ properties making up an inner world, but rather a network of relations between the agent and the environment: relations such as seeing, referring, loving, hating, and all sorts of ways of acting intentionally on things. It is an illusion that the way to pure mind is by abstracting from such relations, as wrong-headed as the idea that the way to pure soccer is by abstracting from the other team. To eliminate 20   The asymmetry noted in the text may also help to explain why closure principles for action and i­ntention sound so much worse than the corresponding closure principles for knowledge and belief. For instance, suppose that it is clear to all concerned that ‘She avenged an insult’ entails ‘She was insulted’. Then, at least usually, when ‘She knows/believes that she avenged an insult’ is true, so is ‘She knows/believes that she was insulted’. But it is much less clear that, even usually, when ‘She will intentionally avenge an insult’ is true, so is ‘She will intentionally have been insulted’, or when ‘She intends to avenge an insult’ is true, so is ‘She intends to have been insulted’. In this respect too, ‘intend’ may behave more like ‘perceive’: it is not clear that, even usually, when ‘I could see that she had avenged an insult’ is true, so is ‘I could see that she had been insulted’ (my visual perception that she had avenged an insult might depend on background ­non-visual knowledge that she had been insulted). 21   For relevant considerations see Hyman (1999,  2010, and  2015), Alvarez (2010), and Hawthorne and Magidor Forthcoming. Philosophers often seem to misunderstand the bearing of ‘reasons’ through a literal-minded treatment of projective talk of the sort mentioned in n. 10, often converging with internalist preconceptions.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

180  Timothy Williamson the other terms of the relations is to eliminate what matters about mind. The point of mind is to involve the world. Far from obscuring the nature of mind by contaminating it with impurities from the external environment, knowledge and action are the fullest expressions of mind. Epistemological responses to knowledge-first epistemology often fail to engage with this bigger picture. From the standpoint of an abductive comparison of theories, they are just not offering a clear alternative view of the mind in the world to rationalize their dissent from the epistemological details. At that level, they are not even competitors. Epistemology is always guided by background assumptions about the nature of mind. When those assumptions are controversial, or ought to be, we need to make them ­explicit, to reflect on them critically. It would be foolish to evaluate Descartes’ epistemology without considering its relation to his philosophy of mind and body—yet it is still common to rely on more or less Cartesian ways of thinking in epistemology without reflecting on their relation to the philosophy of mind. In effect, this chapter has developed more specific challenges to putative alternatives to the knowledge-first approach. How do they theorize the relation between the two directions of fit? Do they accept knowledge as the epistemological analogue of action? If not, what—if anything—do they put in its place? Do they even accept the centrality of action to the philosophy of action?

References Alvarez, Maria. 2010. Kinds of Reasons. Oxford: Oxford University Press. Anscombe, Elizabeth. 1957. Intention. Oxford: Blackwell. Ayer, Alfred. 1956. The Problem of Knowledge. London: Macmillan. Berker, Selim. 2008. Luminosity regained. Philosophers’ Imprint 8: 1–22. Bradley, Richard, and Christian List. 2009. Desire-as-belief revisited. Analysis 69: 31–7. Burge, Tyler. 1979. Individualism and the mental. Midwest Studies in Philosophy 4: 73–122. Byrne, Alex, and Alan Hájek. 1997. David Hume, David Lewis, and decision theory. Mind 106: 411–28. Davidson, Donald. 2001. Essays on Actions and Events, Second Edition. Oxford: Clarendon Press. Goldman, Alvin. 1967. A causal theory of knowing. Journal of Philosophy 64: 357–72. Goldman, Alvin. 1976. Discrimination and perceptual knowledge. Journal of Philosophy 73: 771–91. Grice, Paul. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press. Hawthorne, John, and Ofra Magidor. Forthcoming. Reflections on reasons. In Daniel Star , ed., The Oxford Handbook of Reasons and Normativity. Oxford: Oxford University Press. Hawthorne, John, Daniel Rothschild, and Levi Spectre. 2016. Belief is weak. Philosophical Studies 173: 1393–404. Hawthorne, John and Jason Stanley. 2008. Knowledge and action. Journal of Philosophy 105: 571–90. Holton, Richard. 2008. Partial belief, partial intention. Mind, 117: 27–58. Holton, Richard. 2014. Intention as a model for belief. In Manuel Vargas and Gideon Yaffe, eds., Rational and Social Agency: Essays on the Philosophy of Michael Bratman. Oxford: Oxford University Press, 12–37.

OUP CORRECTED PROOF – FINAL, 10/06/2017, SPi

Acting on Knowledge  181 Humberstone, Lloyd. 1987. Wanting as believing. Canadian Journal of Philosophy 17: 49–62. Humberstone, Lloyd. 1992. Direction of fit. Mind 101: 59–83. Hyman, John. 1999. How knowledge works. Philosophical Quarterly 49: 433–51. Hyman, John. 2010. The road to Larissa. Ratio 23: 393–414. Hyman, John. 2015. Action, Knowledge, and Will. Oxford: Oxford University Press. Levy, Yair. 2013. Intentional action first. Australasian Journal of Philosophy 91: 705–18. Lewis, David. 1988. Desire as belief. Mind 97: 323–32. Lewis, David. 1996. Desire as belief II. Mind 105: 303–13. Nagel, Jennifer. 2013. Knowledge as a mental state. Oxford Studies in Epistemology 4: 275–310. Putnam, Hilary. 1973. Meaning and reference. Journal of Philosophy 70: 699–711. Shope, Robert. 1983. The Analysis of Knowing: A Decade of Research. Princeton: Princeton University Press. Sosa, Ernest. 2010. Value matters in epistemology. Journal of Philosophy 107: 167–90. Sosa, Ernest. 2015. Judgment and Agency. Oxford: Oxford University Press. Srinivasan, Amia. 2015a. Are we luminous? Philosophy and Phenomenological Research 90: 294–319. Srinivasan, Amia. 2015b. Normativity without Cartesian privilege. Philosophical Issues 25: 273–99. White, Alan, ed. 1968. The Philosophy of Action. Oxford: Oxford University Press. Williamson, Timothy. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, Timothy. 2007. The Philosophy of Philosophy. Oxford: Wiley-Blackwell. Williamson, Timothy. 2009. Replies to critics. In Patrick Greenough and Duncan Pritchard, eds., Williamson on Knowledge. Oxford: Oxford University Press, 279–384. Williamson, Timothy. 2011. Knowledge first epistemology. In Sven Bernecker and Duncan Pritchard, eds., Routledge Companion to Epistemology. London: Routledge, 208–18. Williamson, Timothy. 2013a. Knowledge first. In Mathias Steup, John Turri, and Ernest Sosa, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Wiley-Blackwell, 17–21. Williamson, Timothy. 2013b. ‘Knowledge still first’, in Mathias Steup, John Turri, and Ernest Sosa, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Wiley-Blackwell, 22–4. Williamson, Timothy. Forthcoming. Justifications, excuses, and sceptical scenarios. In Julien Dutant and Fabian Dorsch, eds., The New Evil Demon. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

9 Perception First? Heather Logue

1. Introduction As anyone who has taken a typical introductory epistemology course will be painfully aware, attempts to give necessary and sufficient conditions for knowing that p have been beset by countless counterexamples1. Gettier (1963) is widely credited with bringing down the analysis of knowledge as justified true belief (although see Russell 1948, 170–1). Although Smith’s belief that either Jones owns a Ford or Brown is in Barcelona is true and justified (since he saw Jones driving around in a Ford, etc.), it is not knowledge—the Ford that Jones was tooling around in was one he was renting while his own Ford was in the shop with repairs, and, as a lucky coincidence, Brown just happened to be in Barcelona (Gettier 1963, 122–3). Attempts to supplement the justified true belief analysis (for example, with a “no defeaters” condition (e.g., Levy 1977), or a “no false lemmas” condition (e.g., Clark 1963)), and attempts to give an alternative analysis (for example, in terms of reliably formed belief (e.g., Goldman 1967) or sensitive belief (e.g., Nozick 1981, Ch. 3)), haven’t fared any better (see Zagzebski 1994).2 In Knowledge and its Limits, Timothy Williamson suggests that we might take this record of failure to suggest that knowledge doesn’t admit of analysis after all (Williamson 2000, Ch. 1). In particular, while we might reasonably hope to identify several necessary conditions for knowing that p, we should abandon the hope that we  will end up with a list of them that is jointly sufficient for knowing that p.3 1  Thanks to audiences from the Centre for Metaphysics and Mind at the University of Leeds, the  Knowledge First workshop and the Epistemology Research Group at the University of Edinburgh, the University of Southampton, the University of Hertfordshire, and the Mind and Reason group at the University of York, as well as to the editors of this volume, for helpful questions and comments on previous versions of this chapter. 2   I am going to assume, just for the sake of illustration, that all attempts at analysing knowledge fail. I certainly do not regard the brief remarks in this section as a decisive argument for this assumption. 3   As Williamson notes, the fact that we can identify a necessary condition for something does not mean that there have to be any further conditions that we can add to it so as to yield jointly sufficient conditions for it. For example, a necessary condition for being red is being coloured, but there is not any further condition we can add to being coloured that would give us jointly sufficient conditions for being red—short of the condition of being red itself (Williamson 2000, 32).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  183 Williamson’s diagnosis is that we have been going about things the wrong way around—rather than trying to explain knowledge in terms of other epistemic states (e.g., belief, justification), we should instead take knowledge as basic and explain other epistemic states in terms of it. In other words, knowledge should be first in the order of explanation of epistemic states. For example, one might give an account of believing that p in terms of knowing that p—e.g., “[t]o believe p is to treat p as if one knew p” (Williamson 2000, 46). Notably, attempts to analyse the perceptual relation have also been widely regarded as unsuccessful. Williamson suggests that this is part of a broader pattern: “On quite general grounds, one would not expect the concept knows to have a non-trivial analysis in somehow more basic terms. Not all such concepts have such analyses, on pain of infinite regress; the history of analytic philosophy suggests that most of philosophical interest do not. ‘Bachelor’ is a peculiarity, not a prototype.” (2000, 31). It is plausible that we are especially motivated to give philosophical analyses of the notions that play ­central roles in our lives (e.g., knowledge, perception, causation, justice), but they are central precisely because they are the “building blocks” in terms of which other related notions are built. So, the very feature that motivates us to analyse these notions (their centrality) is rooted in that which makes such analyses impossible. In light of this apparent pattern, this chapter considers whether a “perception first” approach is in order. Section 2 distinguishes two ways of fleshing out the knowledge first approach, and clarifies the topic of this chapter (viz., the perceptual relation). Section 3 surveys a representative series of attempts to analyse the perceptual relation, along with corresponding counterexamples. Section 4 argues that even if the perceptual relation is unanalysable, this does not necessarily mean that a perception first approach is in order; nevertheless, there is an alternative motivation for such an approach, which I will sketch briefly. Section 5 concludes with a suggestion concerning a modest positive account of the perceptual relation.

2. Preliminaries There are at least two different ways of fleshing out the knowledge first approach (cf. Jenkins Ichikawa and Ichikawa Jenkins, this volume). One is in terms of conceptual analysis: The concept of knowledge cannot be broken down into other concepts that are necessary and sufficient for its application; rather, the concept of knowledge is a constituent in the analyses of other mental concepts. The other way of fleshing out the “knowledge-first” approach is in terms of metaphysical grounding claims: It is not the case that one knows that p in virtue of being in certain other mental states; rather, one is in certain other mental states in virtue of knowing that p. These are distinct ways of fleshing out the approach because concepts are mental representations, and (at least prima facie) it is an open question whether the structure of mental representations mirrors the facts about metaphysical priority. Since I am primarily interested in the metaphysics of perception—in virtue of what does one count as perceiving

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

184  Heather Logue something?—the relevant comparison here is the metaphysical interpretation of the knowledge first approach. I take it that a metaphysical “X-first” approach only concerns the relation between X and other personal-level mental phenomena. For example, a metaphysical knowledge first approach claims only that knowledge is first in the order of personal-level mental explanation (assuming along with Williamson that knowledge is a mental state). It surely isn’t meant to entail that there are no further subpersonal information processing states or facts about neurons in virtue of which one is in a state of knowledge. So, too, with the perception first approach—the claim is only that related personal-level mental states (e.g., experience, phenomenal character, perceptual justification) are metaphysically grounded in the obtaining of the perceptual relation. This is, of course, compatible with the undeniable claim that a subject perceives in virtue of being in certain subpersonal information-processing and neural states. Before we consider whether a metaphysical perception first approach can be ­motivated by a dismal track record of analyses of perception, we need to clarify the explanandum. In particular, the focus here is the perceptual relation that obtains between a subject and entities in her environment—e.g., objects, property instances, volumes, events, and so on. (I focus mainly on object perception for simplicity’s sake.) We must distinguish perceiving entities in one’s environment from perceiving that p (i.e., that such-and-such is the case). According to Williamson, perceiving that p is a way of knowing that p—it is knowledge that p acquired via perception of entities in one’s environment (2000, Section 1.4). Even if one wants to deny that perceiving that p is a way of knowing that p (Pritchard 2012, Part One, Section 5; McDowell 2009, 158), we should still distinguish between “epistemic” and “non-epistemic” perceiving (Dretske 1969). By contrast, the sort of perception that will be our focus in this chapter is a relation to objects, property instances, and the like (e.g., a banana and its yellowness), rather than propositions (e.g., that the banana is yellow). This sort of perception might afford knowledge that something is the case (perceiving the banana and its yellowness might give one knowledge that it is yellow); but it doesn’t entail that the perceiver has such knowledge, or even that she is in a position to have such knowledge (as in the case of many non-human animals and “pre-conceptual” human children). We must also distinguish the obtaining of the perceptual relation from perceptual experience. As I have just said, perception is a relation between a subject and entities in her environment. By contrast, a perceptual experience is a mental state that a subject could be in even if she doesn’t bear the perceptual relation to anything in her environment. Veridical and illusory perceptual experiences involve the obtaining of the perceptual relation. In both cases, the subject perceives things in her environment; although in the case of illusion, the object perceived may well appear to be a way that it is not. But total hallucinations—the sort of experience that would be had by a “brain in a vat”—by definition don’t involve perceiving things in one’s environment.4 4   Some use the label ‘perceptual experience’ to refer to only those experiences that involve the obtaining of the perceptual relation (see, e.g., Schellenberg 2011). This is merely a terminological difference; I am

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  185 Moreover, perceptual experience is the bearer of a distinctive sort of phenomenal character, and is that which generates and justifies beliefs about one’s environment and can play a role in explaining why a subject (attempts to) act in certain ways. For ­example, there is something it is like to have an experience as of a yellow, crescentshaped thing (be it veridical, illusory, or hallucinatory), which is different from what it is like to have an experience of a red, round thing, or from what it is like to be tickled.5 And other things being equal, an experience of a yellow, crescent-shaped thing (again, be it veridical, illusory, or hallucinatory) will generate the belief that there is a yellow, ­crescent-shaped thing before one, and will provide at least some justification for this belief. Finally, alongside certain background beliefs (e.g., that bananas are yellow, crescent-shaped things) and desires (e.g., the desire to eat a banana), such an experience may be part of an explanation of why one tries to grasp something in one’s environment. (Of course, if the experience is a non-veridical illusion or hallucination, the grasping is likely to be unsuccessful in some respect or other—one may well end up grasping something that isn’t a banana, or come up empty-handed.) This is not to say that the perceptual relation and perceptual experience aren’t intimately related. As I suggest in Section 2, the having of a perceptual experience is a plausible necessary condition on the obtaining of the perceptual relation. But the point is simply that, although the perceptual relation and perceptual experience are intimately related, there is a distinction between them.

3.  Causal Analyses of Object Perception Now that we have clarified the explanandum, let us now turn to attempts to analyse it. For simplicity’s sake, I am just going to focus on object perception—things would go down somewhat differently in the case of perception of property instances, volumes, events, and so on, but I think we would end up in pretty much the same place. Remember that since we are focusing on a metaphysical perception-first approach, we should read the ‘iff ’ claims as in virtue of (i.e., metaphysical grounding) claims: one perceives an object o in virtue of the fact that the conditions on the right-hand side of the biconditional obtain.6 Also, note that the analyses I am proposing aren’t as rigorously worked out as they could be. This is unnecessary for my purposes, since I plan to argue that the mere fact that attempts at analysis have a dismal track record is insufficient to motivate a metaphysical perception first approach. Nevertheless, I need to illustrate the back-and-forth between analyses and counterexamples in order to get this motivation on the table. using ‘perceptual’ more broadly to reflect the fact that some experiences that don’t involve the obtaining of the perceptual relation (at least seem to) have significant phenomenological and epistemological commonalities with those that do. 5   Although note that some deny that there is anything it is like to hallucinate (Fish  2009, Ch. 4; Logue 2012a). 6   For in-depth discussion of the notion of metaphysical grounding, see Correia and Schnieder (2012).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

186  Heather Logue A related point worth flagging is that I am not claiming to have established that there is no counterexample-free analysis of object perception. I do not take the arguments to amount to a decisive refutation of any possible analysis of object perception. If there were a viable analysis, of course, that would take the wind out of the sails of m ­ otivating a metaphysical perception-first approach on the basis of there being no such analysis. In this and Section 3, I simply argue for the following conditional claim: even if there is no viable analysis of object perception, it still wouldn’t follow that we ought to adopt a metaphysical perception first approach. With the preliminaries out of the way, let us commence the illustrative quest for an analysis. As with the case of knowledge, it is natural to begin by looking for necessary conditions, and building from there in hopes of getting a set of individually necessary and jointly sufficient conditions. Just as belief, truth, and justification are plausible ­necessary conditions on knowledge, it is plausible that there is a causal condition on object perception. In particular, we might reasonably suggest that S perceives o only if o causes an experience in S (Grice 1961).7 Just for the sake of illustration, let us pause to remind ourselves why this necessary condition isn’t sufficient for object perception on its own. Proposal 1:  S perceives o iff o causes an experience in S. Here is a flatfooted counterexample to this analysis: my mother caused the experience my cat is having of me right now by causing my existence, but my cat is not perceiving my mother. Here is another somewhat flatfooted counterexample: my brain is causing the experience I am having right now, but I am not perceiving my brain. The general problem that these silly counterexamples illustrate, of course, is that one does not perceive all the objects that are among the causes of one’s experience (Price 1932, 69). We need a condition that singles out the object perceived among the many things that play a role in causing an experience, both distal (my mother) and more proximal (my brain). We can rule out proximal causes easily enough with an “externality” condition— requiring the object to be before the relevant sense organ, rather than “behind” it, so to speak (Lewis 1980, 239). For simplicity’s sake, I’ll leave this condition out in what follows. But how might we rule out the more distal causes? It is tempting to appeal to the fact that it is I who perceptually appear to my cat to have certain properties, whereas my mother does not perceptually appear to be any way at all to my cat.8 Proposal 2:  S perceives o iff o causes an experience in S, and o perceptually appears to be some way to S.9 7   The analyses I will consider are analyses of conscious perception. If you want an analysis of unconscious perception, simply replace ‘experience’ with ‘unconscious perceptual state’. Thanks to Elselijn Kingma for suggesting the need for this clarification. 8   Another possibility is to appeal to the idea that I am in some sense a more direct cause of my cat’s experience than my mother is. However, specifying what exactly is meant by directness here, and how direct of a cause an object would have to be in order to be perceived, would be far from straightforward. 9   To my knowledge, no one has endorsed this analysis. This is presumably because it is circular—both metaphysically construed, as I am about to explain, and also when understood as a claim about concepts (as could be shown by an argument analogous to the one I am about to give in the main text).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  187 This proposal avoids distal cause counterexamples: as I just noted, my mother does not perceptually appear to be any way to my cat. However, the proposed analysis will not do as an account of that in virtue of which one counts as perceiving an object. Recall that we are after a metaphysical grounding explanation of object perception—what are the facts in virtue of which one perceives an object? And an object perceptually appears some way to a subject in virtue of the subject perceiving that object, rather than the other way around. For example, we may ask: what makes it the case that, say, this banana appears to me to be yellow, as opposed to some other object? Presumably, the fact that I perceive this banana, and not (e.g.) the bananas across the street in the café. In short, since we are after a metaphysical explanation of object perception, we need conditions on the right-hand side of the biconditional that are not themselves explained in terms of object perception. Here is a proposal that satisfies this desideratum: Proposal 3:  S perceives o iff o causes an experience in S, and it perceptually appears to S that there is something that has certain properties, and o has at least some of those properties.10 On the face of it, general perceptual appearances (to the effect that something is F . . . etc.) don’t require metaphysical explanation in terms of perception of ­particular objects. Now, we can’t say that o must have all of the properties that specify how something perceptually appears to S. Otherwise, the analysis cannot capture the fact that subjects of non-veridical illusions perceive objects in their environment (e.g., a situation in which one perceives a green banana that looks yellow). So, the most we can say is that the object must have some of the properties. But this limitation opens the door to counterexamples. First, we might want to say that it is possible to see an object without g­etting any of its properties right (a maximally non-veridical ­illusion)—if so, then the condition proposed is not necessary. Second, distal cause counterexamples to sufficiency come back onto the scene—unless there is a nonarbitrary specification of how many of the properties the object must have, my cat could end up counting as perceiving my mother when looking at me (if my mother and I are suitably similar). Moreover, this kind of analysis is notoriously susceptible to so-called “deviant” causal chain counterexamples to sufficiency—cases in which the experience is caused by an object that meets the further conditions, but in a way which is intuitively incompatible with perceiving. Consider a case in which S is given a drug that generates ­hallucinations as of a red apple when the light hitting her retina reaches a certain intensity—which just happens to be reflected off a red apple (cf. Lewis 1980, 242). Intuitively, S doesn’t perceive the particular red apple that reflected the light that hit her retina; after all, she would have had an experience of a red apple no matter which particular object reflected light hitting her retina.

  Analyses of this sort are discussed and rejected in Tye (1982, 314–15).

10

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

188  Heather Logue The counterfactual diagnosis of why S does not count as perceiving the red apple suggests a counterfactual patch for the analysis of object perception. Proposal 4:  S perceives o iff o causes an experience in S, and the experience counterfactually depends on o (cf. Goldman 1977; Lewis 1980).11 If the object were sufficiently different in respect of at least one of its perceptible properties, then (holding the perceptual conditions fixed) the experience would have been different—in the sense that different properties would appear to be instantiated.12 This proposal accommodates the possibility of maximally non-veridical illusion: one’s experience could counterfactually depend on an object even if it gets all of the thing’s properties wrong. (The perceptual conditions would, no doubt, have to be extremely unfavourable in order to generate such an illusion—but even if they are held fixed, a sufficiently drastic change in the object’s perceptible properties would yield a change in how it perceptually appears to be.) This proposal also avoids the kind of deviant causal chain counterexample just described: the experience of a red apple doesn’t counterfactually depend on the red apple before S (as she would have had the same experience if the apple had been yellow). Unfortunately, the proposal falls prey to another kind of deviant causal chain counterexample—namely, ones in which counterfactual dependence is secured in a deviant way. Suppose that the information you receive via your eyes fails to generate an experience for some reason (perhaps some crucial pathway has been severed), but a benevolent puppetmaster saves the day by stimulating your brain to give you experiences as of what is actually before you. The object o causes an experience in the puppetmaster, which (given her benevolent plans) causes her to cause an experience in you. This experience counterfactually depends on o: if o had been different in certain respects, the benevolent puppetmaster would have ensured that your experience was appropriately different.13 But intuitively, you don’t see o. So, it seems that the counterfactual dependence condition is not enough to rule out deviant causal chain counterexamples. One might think that what has gone wrong in the counterexamples I have outlined is that the causal chains described do not support knowledge or successful action. For example, my cat does not perceive my mother because her experience does not put her in a position to (say) climb on my mother’s lap. Or the subject who takes the hallucinogenic drug does not perceive the apple because her experience 11   Note that Lewis is only after a sufficient condition for intransitive seeing—see p. 241. (For analyses in a similar spirit, see Jackson 1977, 171; Tye 1982, 322.) 12   Notice that this way of spelling out the new condition uses the notion of a perceptible property. If the facts in virtue of which a property is perceptible are of the same sort as those in virtue of which an object is perceived, then this way of fleshing out the analysis may be problematic for the same reason that Proposal 2 was problematic. Nevertheless, I will set this worry aside. 13  One might suggest that we do not have genuine counterfactual dependence in this case, as the dependence of the experience on o is subject to the whims of the puppetmaster. However, I take it that we have genuine counterfactual dependence as long as the worlds in which the puppetmaster does not help you out are sufficiently distant from the world under consideration, and we can always suppose that the puppetmaster is really, really nice.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  189 does not put her in a position to know that there is an apple before her. So perhaps we  could avoid the counterexamples by adding a condition about supporting ­knowledge or action. Proposal 5a:  S perceives o iff o causes an experience in S, and the causal chain leading up to the experience puts (a suitably cognitively equipped) subject in a position to know things about it.14 Proposal 5b:  S perceives o iff o causes an experience in S, and the causal chain leading up to the experience puts (a suitably physically equipped) subject in a ­position to successfully act on it (cf. Coates 2007, 134).15 However, neither of these analyses avoids all of the counterexamples that have been raised. In particular, a subject in a “benevolent puppetmaster” case arguably has ­knowledge about her environment and can successfully act on things in it. For the experiences in such a case seem to play the same epistemic and behavioural role of ­testimony. In general, one can know things about one’s environment, and be able to act on things in it, by being told what is in it (e.g., “there is a yellow, crescent-shaped thing directly to your left”). It is natural to regard the hallucinations induced by the puppetmaster as a sort of non-verbal testimony, and hence as endowed with the same ­epistemic and behavioural powers as ordinary verbal testimony. Of course, generating experiences in a subject is unlike verbal testimony in various respects. For example, it does not seem right to say that the puppetmaster is asserting that your environment is a certain way; and if the puppetmaster ceases to be benevolent and starts generating misleading experiences, it does not seem right to say that the puppetmaster is lying to you.16 This can be granted; even so, these differences are not sufficient to show that the puppetmaster-generated experiences do not play the same epistemic and behavioural role as garden-variety verbal testimony. The point is that information transmission between subjects—a category which includes, but is not limited to, verbal assertion— can be a source of knowledge, and support successful action. So, given that the subject in a benevolent puppetmaster case does not perceive objects in her environment, the conditions proposed are not sufficient for object perception after all. One might object that the particular kind of information transmission involved in the benevolent puppetmaster case cannot afford knowledge in the same way that 14   This proposal naturally emerges from a metaphysical knowledge first approach that takes Williamson’s claim that “knowledge is . . . the starting point, the unexplained explainer” (2000, 10) as far as it can reasonably go. That is, knowledge is not just prior to other epistemic states (e.g., belief, justification), but to mental states in general (including perception). On this view, there are no further personal-level mental facts in virtue of which one knows that p that aren’t other knowledge facts; given that one’s evidence is just one’s knowledge (Williamson 2000, Ch. 9), what justifies perceptual belief must itself be knowledge (rather than a non-epistemic state like perception or experience). I do not have the space to argue against this view here, but (as will briefly emerge in the final section) I think it takes the metaphysical knowledge first approach too far. (For related discussion, see Logue 2011, 275.) 15   These are, of course, just rough sketches of what analyses in these terms would look like. 16   Thanks to one of the editors for raising this point.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

190  Heather Logue t­ estimony does. For, in the case of verbal testimony, the reader or hearer can decide to reject it if, say, she deems it untrustworthy. But one might think that we cannot reject our experiences on these grounds; we just find ourselves with them, and we do not normally have independent information about their sources that we could use to assess them for trustworthiness. The idea is that getting knowledge via the transmission of information requires at least the possibility of evaluating the source of that information in order to determine whether one should accept the message (so to speak).17 So, one might conclude that the experiences generated by the benevolent puppetmaster do not afford knowledge (even if they do support successful action). It is true that we do not typically have experience-independent information about the sources of our experiences (aside from cases in which, e.g., the subject knows that she has just taken hallucinogenic drugs). But we can (and sometimes do) rely on the contents of our experiences in order to assess their reliability. For example, if I had a crystal-clear experience as of the banana on my desk sprouting legs and doing a little tap dance, arguably, I ought to conclude that my experience is not to be trusted (on grounds of its incoherence with my beliefs about what kinds of things bananas can get up to). Similarly, if the puppetmaster ceased her policy of benevolence, and started throwing tap-dancing bananas into your experience every now and again, that should lead you to doubt the veridicality of your experiences—regardless of whether you knew about the puppetmaster arrangement. (If you don’t know about the arrangement, you wouldn’t evaluate the source of the information qua puppetmaster; but you should still conclude that the source of your experiences, whatever it is, is not generating experiences that you can take at face value.) In short, it seems that a subject in a benevolent puppetmaster case is in a position to evaluate the source of her experiences (under some description or other), and so there is no obstacle to knowledge here after all. Alternatively, one might defend the analyses under discussion by embracing the conclusion that the subject in a benevolent puppetmaster case perceives objects in her environment. If we flesh out the case in the right way (such that it is maximally ­plausible that the subject has the relevant epistemic and behavioural capacities over an extended period of time), why not say that she perceives things in her environment? Sure, the causal pathway to her experiences is non-standard. But we should allow that subjects can perceive things by non-standard means—how is the situation of the ­subject in a puppetmaster case relevantly different from, say, someone who sees by way of a prosthetic eye?18 It is hard to say what exactly the relevant difference is, but there does seem to be one.19 It has something to do with directness: the connection between light being   Thanks to one of the editors for this suggestion.   Thanks to Paul Coates for this suggestion. 19   I do not think it has anything to do with the possibility that the puppetmaster could decide to lie on a whim—for we could just build it into the description of the case that she is honest to a fault, and that does not make me any more inclined to say that the subject perceives things in her environment. (The subject is getting knowledge about her environment all right, just not via perception.) 17 18

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  191 reflected off of an object, the prosthetic eye, and the subject’s experience seems “tight enough” to count as perceptual, whereas the connection between light being reflected off of an object, the puppetmaster’s eye, her experience, and her inducing a corresponding experience in the subject does not. We might try to refine this idea in order to yield another yet necessary condition for our analysis, but (as indicated in n. 8 in this chapter) I would not hold my breath. In any case, embracing the claim that the subject in a (suitably elaborated) puppetmaster case genuinely perceives things in her environment strikes me as too big a bullet to bite. The relevant epistemic and behavioural capacities are just “detachable” enough from perception to cast doubt on analyses of the latter in terms of the former. Let us suppose (for the sake of argument) that this whirlwind tour through various attempts at analysing object perception indicates that there is no such analysis to be had. Does this mean that we should embrace a metaphysical perception first approach?

4.  Counterexamples are Not Enough This section argues that a failure of an analysis of X (e.g., knowledge, object perception) is insufficient to motivate a metaphysical “X-first” approach. I identify two sources of analysis failure, and argue that only one of these sources motivates a metaphysical X-first approach.20 Moreover, I suggest that while analyses of knowledge might fail for a reason that would motivate a knowledge-first approach, analyses of object perception don’t fail for an analogous reason. Assuming that metaphysical analyses of a phenomenon (knowledge, perception) are doomed to fail, how exactly is this fact related to the claim that the phenomenon is first in the order of metaphysical explanation? If an “X-first” approach to the phenomenon at issue is correct, then metaphysical analyses of the phenomenon in terms of other personal-level mental states are doomed to fail.21 For to say that a phenomenon is first in the order of metaphysical explanation is just to say that there are no further personal-level mental states in virtue of which that phenomenon obtains. But this conditional is not what is at issue here. What we want to know is whether the converse is true: if metaphysical analyses of a given phenomenon in terms of other personal-level mental states are doomed to fail, then should we fly the metaphysical X-first flag? Not necessarily. For the failure of analysis might be attributable to another source. For example, metaphysical analyses of a phenomenon could be doomed to fail because the analysandum and the analysans are metaphysically “on a par”—perhaps neither obtains in virtue of the other. The fact that X is first in the order of metaphysical ­explanation is sufficient but not necessary for the failure of a metaphysical analysis of   Many thanks to Daniel Elstein and Paolo Santorio for suggesting this and convincing me of it.   Note that this is compatible with there being a true claim of the form “S Xs . . . iff . . .” . All that a metaphysical X-first approach rules out is that a certain kind of metaphysical elaboration of such a claim is true—that S Xs in virtue of the facts mentioned on the right-hand side of the biconditional. 20 21

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

192  Heather Logue X; so, we cannot infer from the mere failure of metaphysical analyses of X that a ­metaphysical X-first approach is in order. We need to investigate the source of the ­failure to see whether it supports such an approach. In the case of knowledge, Gettier convinced (most of) us that knowledge is not merely the sum of the mental element of justified belief, and the non-mental element of truth.22 And the dreary history of post-Gettier “JTB+” epistemology suggests that we cannot sidestep all counterexamples by adding anything to either the mental element (e.g., a “no false lemmas” requirement (e.g., Clark 1963)), or to the non-mental element (e.g., a “no defeaters” requirement (e.g., Levy 1977)).23 Indeed, it does not seem possible to identify a mental element that is sufficient for knowledge in combination with truth (and perhaps other non-mental factors).24 Let us call a mental element that is sufficient for X in combination with non-mental factors a maximal mental element of X.25 One reason an analysis of knowledge might fail is that it assumes that knowledge is “built up” out of a maximal mental element and a purely non-mental remainder (a presupposition we will question in due course). If we cannot identify a maximal mental element of knowledge, an analysis that presupposes that there is such a thing cannot get off the ground. After all, a necessary condition on a successful analysis of X is identifying the “building blocks” that give rise to it.26 And an inability to identify one of the essential building blocks suggests that we are going about things the wrong way around. The fact that we cannot identify a maximal mental element out of which knowledge is built may be taken to indicate that metaphysical explanation goes in the opposite direction—that knowledge is actually more fundamental than any other alleged mental element of it (e.g., belief, justification). Plausibly, the justified-true-belief (JTB) analysis fails for this reason (which miscasts justified belief as the maximal mental element), and one might suggest that this is the root of the failure of JTB+ successor analyses as well.27 22   Although see, e.g., Hetherington (1998) for a JTB+ account, Sartwell (1991) for a defence of the claim that knowledge is simply true belief, and Weinberg, Nichols, and Stich (2001) for a sceptical take on the intuitions about Gettier cases. 23   Of course, not all defeaters are non-mental. But the point here is that some of them are (e.g., easily available evidence that would undermine one’s justification for believing the proposition at issue). 24   One might suggest that this element is simply belief—more on this later. 25   By ‘element’, I mean something like proper part—so if Williamson is correct that knowledge is a mental state, it still is not a maximal mental element of itself in the sense used here. Also, careful readers will notice two important differences between this notion and Williamson’s “virtual C”. Virtual C “. . . is the  strongest narrow [i.e., purely internal] condition which C [a broad condition, such as knowledge] entails” (Williamson 2000, 66–7). First, it is no part of virtual C that it is sufficient for C in conjunction with non-mental factors. Second, the notion of a maximal mental element is neutral about whether mental states must be narrow. 26   Thanks to Helen Yetter-Chappell for suggesting the building blocks metaphor. 27   Cf. Williamson (2000, 49–51). Although note that Williamson’s focus is on attempts to analyse knowledge as a conjunction of mental and non-mental conditions, given the assumption that mental conditions are purely internal (i.e., determined by what is going on inside the subject’s body). Our focus here is broader: attempts to analyse knowledge as a conjunction of mental and non-mental conditions, regardless of whether mental conditions are assumed to be purely internal. One need not be an internalist about the mental in Williamson’s sense to go in for metaphysical analyses of mental phenomena.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  193 However, there is a less radical moral one might draw from the failure of such a­ nalyses. Instead of going all the way to a metaphysical knowledge-first approach, one might instead suggest that we simply abandon the assumption that an analysis of knowledge requires identification of a maximal mental element. It is natural to see reliabilist (e.g., Goldman 1967) and sensitivity-based (e.g., Nozick 1981, Ch. 3) analyses in this light. One might be tempted to think that these are views on which belief is the maximal mental element, to which we need to add reliability or sensitivity in order to get knowledge. But this is not right—the reliability or sensitivity of a belief is not a purely non-mental affair. The facts in virtue of which a belief is the product of a reliable belief-forming process, and in virtue of which it counts as sensitive, include mental facts (e.g., whether one engages in random leaps of faith in the course of reasoning, or irrationally resolves to believe that one was abducted by aliens come what may). Note also that even some JTB+ successor analyses do not assume a maximal mental element of knowledge. For example, an analysis of knowledge as justified undefeated true belief (e.g., Levy  1977) does not seem to be committed to a maximal mental ­element, insofar as the defeasibility condition straddles the divide between the mental and the non-mental. It offers a mental element, but does not go as far as specifying a maximal mental one (i.e., one that is sufficient for knowledge in conjunction with the relevant non-mental factors). The belief that (say) there is a banana on my desk is undefeated (in part) because both (a) I do not believe that someone’s been going around replacing bananas with plastic replicas, and (b) there is not any easily accessible evidence floating about to give me reason to believe that such tomfoolery is afoot. While (a) is a mental matter (a fact about what I believe), (b) is a non-mental matter (a fact about what things beyond my mind would give me reason to believe). The point is that whether a belief is undefeated is not just a matter of psychology, and hence having an undefeated justified belief is not a candidate for being the maximal mental element of knowledge. Let us take stock. There are two morals we could draw from the failure to identify a maximal mental element of knowledge. One is that a metaphysical knowledge-first approach is in order. The other is that an analysis of knowledge simply does not require a maximal mental element. What would recommend the first reaction over the ­second? One possibility (which I do not have the space to defend fully here) is that an analysis of knowledge that does not specify a maximal mental element is, in some sense, cheating. That is, even if we can come up with counterexample-free analyses in terms of reliability, sensitivity, or some sort of defeasibility condition, one might think that the job is not yet done. For one might suggest that what we really wanted from an analysis was a specification of the general facts in virtue of which a knower’s belief counts as sensitive, or as the product of a reliable belief forming process, or as ­undefeated. And one might expect this to involve specification of a maximal mental ­element—in order to specify the subject’s contribution to reliability (etc.). No doubt, not everyone will agree that we should expect this from an analysis of knowledge. But for my purposes, all I want to insist upon is the following conditional: if we cannot

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

194  Heather Logue identify a maximal mental element of knowledge, and analyses that do not identify one are (for some reason or other) inadmissible, then a metaphysical knowledge-first approach is in order. Now let us consider whether we have this sort of support for a metaphysical object perception-first approach. It seems not, as it is easy to identify the maximal mental ­element of perception: namely, perceptual experience. The tricky part entirely concerns the non-mental component: namely, the causal chain leading up to the experience. As is illustrated by the discussion in the previous section, the problem for analysing object perception is entirely a matter of ruling out deviant causal chains in a principled way. And the mere fact that we have a difficult time ruling out deviant causal chains does not seem to support the claim that perception is more fundamental in the order of explanation than experience. The problem is simply a matter of specifying which kinds of causal chains are compatible with perception, and which ones are not.28 In short, the failure of analyses of object perception is not rooted in an inability to identify a ­maximal mental element, and so we do not have a motivation for a metaphysical ­perception-first approach that is analogous to the one just sketched in the case of knowledge. Moreover, the root of the failure (an inability to rule out deviant causal chains leading up to the experience) does not seem to provide any support for a metaphysical perception-first approach.29 Indeed, there is a plausible alternative explanation of the failure of analyses of object perception. The difficulty may just be rooted in the fact that there is vagueness in the non-mental element of object perception.30 It is tempting to conclude that there is ­simply no fact of the matter as to precisely when a causal chain gets too weird or indirect to count as supporting object perception. This vagueness might be reflected in ­differing reactions to putative counterexamples: one person’s intended counterexample might well be regarded by another person as a case of really weird perception. We tolerate some deviation from the “standard” type of causal chain leading up to experience, but perhaps it is a vague matter exactly how much and what sort of deviation is compatible with object perception. Notice that, in spite of this vagueness, perception can still be factored into a maximal mental element (experience) and a non-mental element (the causal chain leading up to experience)—it is just that the vagueness makes it impossible to give a principled characterization of the non-mental element. So, the reason why analyses of object perception have failed to command general 28   Of course, ruling out deviant causal chains is a concern in the case of knowledge, too, and one might appeal to non-mental conditions in attempts to do so (e.g., a requirement that there are no non-mental defeaters). Rather, the point is that this is not the only concern in the case of knowledge. Thanks to Martin Smith for raising this point. 29   Note that Williamson’s argument that seeing is a prime condition (2000, 69–70) does not establish that it cannot be factored into a mental element and a non-mental one. If a condition is prime, then it cannot be factored into internal and external conditions that are jointly sufficient for it; but absent the assumption that mental states are purely internal, it does not follow that it cannot be factored into mental and non-mental conditions that are jointly sufficient for it. (As noted earlier, one need not be an internalist about the mental to go in for the project of metaphysical analysis of mental states.) 30   Thanks to Jennifer Carr for this suggestion.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  195 assent has nothing to do with an inability to identify a maximal mental element; and is more likely to be simply a matter of vagueness inherent in the non-mental element.31, 32 It is natural to jump to the conclusion that failed analyses of knowledge and object perception are symptoms of a failure to recognize that knowledge and perception are mental primitives, in that they cannot be decomposed into a maximal mental element and the non-mental remainder. But one moral of this section is that there are alternative explanations of such failures that we have to rule out before the inference to this conclusion is warranted. The other moral is that the alternative explanation in the case of object perception is more compelling. That being said, there is another way to motivate a metaphysical perception first approach. Recall that, on such an approach, we are to give metaphysical grounding explanations of perceptual experience in terms of perception, rather than the other way around.33 This claim sounds an awful lot like Naïve Realism (also known as Relationalism): the view that veridical perceptual experience fundamentally consists in the subject perceiving things in her environment (see, e.g., Campbell  2002, Martin 2004, Fish 2009, Brewer 2011, Logue 2012b). For example, the veridical experience I am having right now of the banana on my desk fundamentally consists in my  perceiving the banana (and some of its properties, like its yellowness and crescent-shapedness).34 As I understand it, this talk of fundamentality is shorthand for the claim that the obtaining of the perceptual relation is the metaphysical ground of certain features of veridical perceptual experience: namely, its phenomenal character, its epistemic role, 31   One might object that there is a way of making the problem of analysing knowledge exactly analogous to the problem of analysing object perception—namely, by claiming that the maximal mental element is belief, and that the problem of analysis is entirely a matter of ruling out deviant causal chains leading up to it. However, for the reasons given previously, the claim that belief is the maximal mental element of knowl­ edge is not plausible. This is also suggested by the fact that externalists typically accept further mental conditions on knowledge. For example, one’s reliably caused belief wouldn’t count as knowledge if it conflicts with other beliefs one has. (See Goldman (1979, Section III) for an analogous point about reliabilist justification.) Thanks to one of the editors for this suggestion. 32   Another (compatible) explanation might be that while we can (usually) tell whether a causal chain leading up to experience is deviant, we do not have a grip on the facts in virtue of which it counts as deviant (or not)—i.e., we do not have a sense of exactly which standing causal regularities need to be in place for perception to occur. (Thanks to one of the editors for this suggestion.) One way of developing this idea would be to say that we are sort of like chicken sexers when it comes to object perception; we can sort the cases correctly, but we don’t have full access to the basis on which we do so. Another way of developing it would be to say that our failure to grasp the details of the required causal regularities is rooted in the sort of vagueness described in the main text. 33   Recall that this is compatible with there being a true claim of the form “S perceives o iff . . . ” . What a metaphysical perception first approach rules out is that a certain kind of metaphysical elaboration of such a claim is true—that S perceives o in virtue of further personal-level mental facts mentioned on the righthand side of the biconditional. 34  Note that some Naïve Realists explicitly extend this type of account to illusory experiences (Brewer 2008; Fish 2009, Ch. 6). But let us set illusions aside for simplicity’s sake. Also note that, largely as a result of commitments to different broader metaphysical views, Naïve Realists differ about the precise way in which properties figure in perceptual experience (compare, e.g., Brewer 2008; Fish 2009, Ch. 3.1). However, this complication is not relevant for our purposes.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

196  Heather Logue and its role in guiding action. For example, Naïve Realism holds that my experience has the phenomenal character it does in virtue of my perceiving the banana’s yellowness and crescent-shapedness. Moreover, other things being equal, I am inclined to believe that there is a yellow, crescent-shaped thing before me, and (sceptical arguments aside) this belief would be justified to a degree sufficient for knowledge. Naïve Realism maintains that these epistemic facts obtain in virtue of my perceiving the banana and its yellowness and crescent-shapedness. Finally, if I have the background belief that bananas are yellow and crescent-shaped, and the desire to eat a banana, other things being equal I will reach for the yellow crescent-shaped thing. Naïve Realism holds that part of the explanation of this behaviour is that I perceive the yellow, crescent-shaped thing before me. In short, for the Naïve Realist, phenomenal, epistemic, and behavioural features of veridical perceptual experience “bottom out” in the obtaining of the perceptual relation—just as a metaphysical perception first approach would have it.35 By contrast, Naïve Realism’s rivals hold that these features bottom out in something else—e.g., perceptual representational states (as a view known as Intentionalism or Representationalism would have it), or acquaintance with sense-data (as sense-datum theories claim). Of course, the phenomenal, epistemic, and behavioural features of total hallucinations (which by definition don’t involve the obtaining of the perceptual relation) cannot be grounded in an actual obtaining of the perceptual relation. Naïve Realism leads to disjunctivism about perceptual experience, which is roughly the view that the personal-level mental grounds of total hallucination are different from those of veridical experience (see, e.g., Logue 2014, Section 5). This follows from the claim that veridical experience is grounded in the obtaining of the perceptual relation, and the fact that total hallucinations do not involve the obtaining of this relation—it has to be the case that total hallucinations are grounded in something else. I do not have the space here to explore how this issue interacts with the metaphysical perceptual first approach; ­suffice it to say that, depending on what one’s account of total hallucination is, the metaphysical perception-first approach might be limited in various respects.36 In summary, Naïve Realism entails some form of a metaphysical perception-first approach. This means that we can motivate a metaphysical perception first approach by arguing for Naïve Realism. Outlining and defending such an argument is a task 35   Note that a metaphysical perception first approach is compatible with recognizing a causal condition on the obtaining of the perceptual relation. One might suspect otherwise on the grounds that Naïve Realism is plausibly interpreted as holding that the things perceived are constituents of the subject’s experience, and one might think that a constituent of something cannot also be a cause of it. However, just as a battle can both cause and constitute a war, and a person can both cause and constitute a conversation (Child 1992, 309–10), a banana can both cause and constitute my experience. 36   As M.G.F Martin emphasizes (1997, 88–90)—albeit in different terms—Naïve Realism is sufficient but not necessary for a metaphysical perception-first approach (see also Williamson 2000, 48, n. 9). For example, one might go in for a rather limited perception first approach on which the obtaining of the perceptual relation grounds the object-dependent representational content of experience, while in turn the representational content does all the important explanatory work with regard to phenomenal character, post-perceptual epistemic states, and action. See Schellenberg (2010) for a view in this vicinity.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  197 beyond the scope of this chapter. For present purposes, my point is simply that any ­motivation for Naïve Realism is ipso facto a motivation for some version of a metaphysical perception-first approach.

5.  A Positive Account of Perception? As readers of Knowledge and its Limits will recall, there are two strands to the ­knowledge-first approach. There is a negative claim, to the effect that knowledge cannot be analysed in terms of other personal-level mental states. On the metaphysical version of the approach, this is because knowledge is first among epistemic states in the order of metaphysical explanation. However, Williamson also provides what he calls a “modest positive account” (Williamson 2000, 33). The (alleged) fact that knowledge cannot be analysed does not mean that we cannot say anything illuminating about it. In particular, Williamson holds that knowledge is the most general factive mental state: if I am in any factive mental state at all (like remembering that p, or regretting that p), I thereby know that p (Williamson 2000, sec. 1.4). Our focus so far has been on the analogue of the negative claim in the perceptual case: that perception does not admit of metaphysical analysis, and that this is because it is first in the order of metaphysical explanation of related mental phenomena (experience, phenomenal character, etc.). I have argued that while perception may well be first in the order of metaphysical explanation (e.g., because Naïve Realism is true), we cannot infer this from the (alleged) fact that perception does not admit of metaphysical analysis—as there is a simpler explanation for the failure of analysis in this case. In any event, one might wonder whether there is a modest positive account of the perceptual relation that is analogous to Williamson’s account of knowledge. A natural suggestion is that perception is the most general form of (something like) awareness of the world in virtue of it impinging on one’s body, which comes in a variety of determinates—e.g., seeing, hearing, touching, smelling and tasting.37 One might dismiss this suggestion as banal. However, while it is uncontroversial that seeing, hearing, and so forth are determinates of perceiving, it is a matter of debate how to characterize perceiving in general. So, for instance, the suggestion just made is not regarded as a truism in the literature on individuating the senses (Matthen 2015, Section 1). But rather than take a stand on this account, I would like to close by sketching an alternative to which I am particularly sympathetic. While knowledge may be first in the order of explanation of other epistemic states (belief, justification), one might think that perception is metaphysically prior to perceptual knowledge. In particular, perceiving objects and property instances is that in virtue of which one is in a position to 37   Another possibility (suggested to me by Martin Smith) is that perceiving an object might be the most general causal “object-entailing” mental state. The kind of state at issue is the analogue of factivity for objects, with a causal condition attached. The idea is that perceiving an object comes in various forms: attending to it, noticing it, and so on. However, I am not optimistic about this proposal, as I find it difficult to extend the list beyond the two forms of perceiving already mentioned.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

198  Heather Logue p­ erceive that p. For example, perceiving this banana and its yellowness is what puts me in a position to perceive that it is yellow. Perceiving that p is a way of knowing that p (or at least, a way of being in a position to know that p), and the claim is that such epistemic states are grounded in perceiving entities in one’s environment.38 Moreover, it would be natural to add that perception is part of what grounds one’s capacity for successful action in one’s environment. I am able to pick up this banana, peel it, and eat it partly in virtue of the fact that I see and feel it. In short, the suggested positive account is that perceiving entities in one’s environment grounds knowledge about it and successful action in it. As before, one might complain that this account, while positive and plausible, is not news. Williamson’s positive account of knowledge is informative—it makes a claim about knowledge that most (if not all) of us did not already believe explicitly. But the claim that perception grounds perceptual knowledge and successful action might seem just plain obvious, and so rather uninteresting as a positive account of perception. I suspect that the proposed positive account of perception might seem obvious to many. But it does not seem to be obvious to everyone. As we saw at the end of Section 3, one might try to give an analysis of perception in terms of knowledge or successful action. On the metaphysical construal of these analyses, one perceives something in virtue of being in a position to know things about it, or in virtue of being able to act successfully on it. The positive account just suggested has things the other way around: perceiving entities in our environment is what grounds our knowledge about it and our successful action in it. Note that if this account is correct, knowledge is not first in the broader scheme of things after all—perception is prior. Although I do not have the space to defend it here, this strikes me as the right result.

References Brewer, Bill. 2008. How to Account for Illusion. In A. Haddock and F. Macpherson, eds., Disjunctivism: Perception, Action, Knowledge. Oxford: Oxford University Press, 169–80. Brewer, Bill. 2011. Perception and Its Objects. Oxford: Oxford University Press. Campbell, John. 2002. Reference and Consciousness. Oxford: Oxford University Press. Child, William. 1992. Vision and Experience: The Causal Theory and the Disjunctive Conception. Philosophical Quarterly 42(168): 297–316. Clark, M. 1963. Knowledge and Grounds: A Comment on Mr. Gettier’s Paper. Analysis 24: 46–8. Coates, Paul. 2007. The Metaphysics of Perception. New York: Routledge. Correia, Fabrice, and Benjamin Schnieder, eds. 2012. Metaphysical Grounding: Understanding the Structure of Reality. Cambridge: Cambridge University Press. Dretske, Fred. 1969. Seeing and Knowing. London: Routledge & Kegan Paul. Fish, William. 2009. Perception, Hallucination, and Illusion. Oxford: Oxford University Press. 38   Of course, this does not mean that perceiving something entails that one is in a position to know anything about it. A creature that has the capacity for perception but lacks the capacity for belief cannot be in a position to know anything, given that belief is necessary for knowledge.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

perception first?  199 Gettier, Edmund. 1963. Is Justified True Belief Knowledge? Analysis 23: 121–3. Goldman, Alvin. 1967. A Causal Theory of Knowing. Journal of Philosophy 64(12): 357–72. Goldman, Alvin. 1977. Perceptual Objects. Synthese 35(3): 257–84. Goldman, Alvin. 1979. What Is Justified Belief? In George Pappas, ed., Justification and Knowledge. Boston: Reidel, 1–25. Grice, H. P. 1961. The Causal Theory of Perception. Proceedings of the Aristotelian Society 35(Suppl): 121–52. Hetherington, Stephen. 1998. Actually Knowing. Philosophical Quarterly 48(193): 453–69. Jackson, Frank. 1977. Perception: A Representative Theory. Cambridge: Cambridge University Press. Jenkins Ichikawa, Jonathan, and C. S. I. Jenkins. 2018. On Putting Knowledge ‘First’. In J. Adam Carter, Emma Gordon, and Benjamin W. Jarvis, eds., Knowledge First: Approaches to Epistemology and Mind. Oxford: Oxford University Press, 113–31. Levy, Steven. 1977. Defeasibility Theories of Knowledge. Canadian Journal of Philosophy 7(1): 115–23. Lewis, David. 1980. Veridical Hallucination and Prosthetic Vision. Australasian Journal of Philosophy 58: 239–49. Logue, Heather. 2011. The Skeptic and the Naive Realist. Philosophical Issues 21: 268–88. Logue, Heather. 2012a. What Should the Naive Realist Say about Total Hallucinations? Philosophical Perspectives 26: 173–99. Logue, Heather. 2012b. Why Naïve Realism? Proceedings of the Aristotelian Society 112: 211–37. Logue, Heather. 2014. Disjunctivism. In Mohan Matthen, ed., Oxford Handbook of the Philosophy of Perception. Oxford: Oxford University Press, 198–216. McDowell, John. 2009. Wittgenstein’s ‘Quietism’. Common Knowledge 15: 365–72. Martin, M. G. F. 1997. The Reality of Appearances. In M. Sainsbury, ed., Thought and Ontology. Milan: FrancoAngeli, 81–106. Martin, M. G. F. 2004. The Limits of Self-Awareness. Philosophical Studies 120: 37–89. Matthen, Mohan. 2015. The Individuation of the Senses. In Mohan Matthen, ed., Oxford Handbook of the Philosophy of Perception. Oxford: Oxford University Press, 567–86. Nozick, Robert. 1981. Philosophical Explanations. Cambridge, MA: Harvard University Press. Price, H. H. 1932. Perception. London: Methuen. Pritchard, Duncan. 2012. Epistemological Disjunctivism. Oxford: Oxford University Press. Russell, B. 1948. Human Knowledge: Its Scope and Limits. London: Allen and Unwin. Sartwell, Crispin. 1991. Knowledge Is Merely True Belief. American Philosophical Quarterly 28(2): 157–65. Schellenberg, Susanna. 2010. The Particularity and Phenomenology of Perceptual Experience. Philosophical Studies 149(1): 19–48. Schellenberg, Susanna. 2011. Perceptual Content Defended. Nous 45(4): 714–50. Tye, Michael. 1982. A Causal Analysis of Seeing. Philosophy and Phenomenological Research 42: 311–25. Weinberg, Jonathan M., Shaun Nichols, and Stephen Stich. 2001. Normativity and Epistemic Intuitions. Philosophical Topics 29: 429–60. Williamson, Timothy. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Zagzebski, Linda. 1994. The Inescapability of Gettier Problems. Philosophical Quarterly 44(174): 65–73.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

10 Epistemic Supervenience, Anti-individualism, and Knowledge-First Epistemology Jesper Kallestrup and Duncan Pritchard

1.  Epistemic Supervenience With very few exceptions—e.g., Keith Lehrer (1997)—epistemologists have embraced the claim that epistemic properties (E) supervene on non-epistemic properties (N), such as natural, descriptive, or physical properties. The relevant type of supervenience is typically that of strong, individual supervenience, stated in terms of metaphysical necessity.1 Consider the following formulation of epistemic supervenience: Epistemic Supervenience (ES) Necessarily, if an individual S has epistemic property E, then S has some non-epistemic property N such that, necessarily, any individual S* with N also has E. Strictly speaking, the supervenience relation is reflexive, transitive, and non-symmetric, yet it is best understood in the epistemic domain as holding asymmetrically—i.e., ­non-epistemic properties do not plausibly supervene on epistemic properties. Also, all ES says is that a pattern of variation holds between epistemic and non-epistemic ­properties. It is a further metaphysical question why these sets of properties co-vary; assuming ES is not to be regarded as an inexplicably brute fact, an explanatory account is owed of why S has E in virtue of having N. Following Terence Horgan (1993), call such an account ‘superdupervenience’. Moreover, assuming the supervenience relation holds asymmetrically, we also need an explanation of why S does not have N in virtue of having E. Putting the two explanatory accounts together will show why N is ontologically prior to E.2 Thus, suppose property realization is what superdupervenience amounts to. In that case, ES has it that whatever particular N realizes E is necessarily sufficient for E, yet might not have 1   For a helpful overview of the different kinds of supervenience, and of strong individual supervenience in particular, see McLaughlin and Bennett (2011). Note that, unless we stipulate otherwise, we will henceforth understand supervenience in terms of strong individual supervenience. 2   For more details, see McLaughlin (1995).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  201 realized E. In short, E is asymmetrically necessitated by its realizer N. Before we consider distinct epistemic properties and the nature of the non-epistemic properties on which they allegedly supervene, one may well ask why ES should be true in the first place. Ernest Sosa (1991) has argued that all epistemic properties supervene on non-evaluative properties, because all evaluative properties supervene on non-evaluative properties, and all epistemic properties are evaluative properties. The conclusion of Sosa’s argument entails ES, but not the other way around, given that some properties, such as moral properties, are evaluative but not epistemic.3 Even more generally, physicalists agree that if their view is actually true, then all (positive, non-indexical) properties of our world (strongly or globally) supervene on physical properties of our world, where a property is physical just in case it occurs in current physical theory, or an improved version thereof. Since all epistemic properties are non-physical on such a theory-based conception, this physicalist thesis entails ES, but not the other way around, given that some non-physical properties—e.g., moral properties—are non-epistemic. Let us now ponder which epistemic properties ES might pertain to. A useful strategy is to begin with the paradigmatic property of having (propositional) knowledge, and  then factor out other epistemic properties that must be instantiated if having ­knowledge is. Certainly, since knowledge entails truth, having knowledge is a property that supervenes at least in part on non-epistemic properties, as long as truth is regarded as non-epistemic. Reflect that at least in the case of knowledge N is going to be a highly extrinsic matter. After all, the contents of S’s knowledge are frequently about features— environmental, historical, etc.,—which are well beyond S. Knowledge also entails (dispositional) belief; or so we shall be assuming throughout.4 But then if both belief and truth are necessary, yet jointly insufficient, conditions on knowledge, we can define warrant as the additional condition, whatever that may be, such that when satisfied converts true belief into knowledge.5 Put differently, from knowledge we can factor out warrant as that which turns true belief into knowledge. We can then ask the question of whether being warranted supervenes on non-­ epistemic properties without having to worry about the factivity of knowledge.6 Yet, warrant is still a very complex epistemic property. For instance, being warranted implies believing p non-defeatedly, where this latter property involves the absence of various types of (rebutting and undercutting) epistemic defeat.7 For each of these, the question can be raised of whether they supervene on non-epistemic properties. However, we can set aside epistemic defeat by focusing on the slightly less complex property of having a justified belief. Justification aims for truth: a belief is likely to be true if believed justifiably, i.e., believing p justifiably raises the objective probability of p.   See Turri (2010).    4  See, for instance, Rose and Schaffer (2013).   Here we follow Plantinga (1993) in his generic specification of warrant. 6   Focusing on warrant has the additional advantage that we avoid vexed questions about the supervenience of belief content, or the physical vehicle that expresses that content, on non-epistemic properties. For more on these supervenience claims, see Kallestrup (2011). ES should preferably be a claim about the supervenience of purely epistemic properties on non-epistemic properties. 7   For a helpful recent survey of different kinds of epistemic defeater, see Sudduth (2015). 3 5

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

202  Jesper Kallestrup and Duncan Pritchard Still, a belief can remain (propositionally) justified even if defeated, indeed even if false.8 The connection between justification and truth is not that of entailment. Having a justified belief in that sense is typically taken to be a property that supervenes on nonepistemic properties. That may have to do with the epistemic basing-relation, which is what converts propositional justification—i.e., having justification to believe—into doxastic justification—i.e., having a justified belief. That is to say, S’s belief is properly based, or well-founded, just in case S believes the target proposition p on the basis of the propositional justification that S has for believing p. Arguably, being properly based supervenes on non-epistemic properties, and so one might think that having a justified belief supervenes in part on those non-epistemic properties on which being properly based supervenes.9 Finally, by considering propositional justification we can ignore questions about both epistemic defeat and epistemic basing. Typically, even the fairly simple epistemic property of having justification to believe is taken to supervene on non-epistemic properties. To say that S has justification to believe p means that there is justification available to S to believe p. For instance, if the justification that S has consists of e­ vidence, then the evidence is accessible to S such that S is in a position to believe p on the basis of that evidence. With these different supervenience claims in mind, let us now turn to consider the nature of the various non-epistemic properties on which our epistemic properties— warrant, doxastic justification, and propositional justification—supervene. Our contention is that these base properties have been assumed by many epistemologists to be what we call individualistic properties, i.e., physical or mental properties that are ­internal to the bodily boundaries of S.10 However, epistemic twin earth scenarios show that none of our epistemic properties supervene on individualistic properties. Drawing on our previous work—see especially Kallestrup and Pritchard (2011; 2012; 2013)—we use Sosa’s robust virtue epistemology (2011; 2013) in Section 2 as a test case to determine whether warrant supervenes on individualistic properties. Our answer is negative. In Section 3 we extend our epistemic twin earth scenario to demonstrate that neither doxastic nor propositional justification supervenes on individualistic properties. For this purpose, we use Earl Conee and Richard Feldman’s (2004) evidentialist and mentalist stripe of epistemic internalism as a case in point. Finally, in Section 4, we argue that while Timothy Williamson’s (2000) knowledge-first approach to epistemology is compatible with epistemic anti-individualism, the approach confronts a problem about providing an explanatory account of ES. The upshot is that a prima facie case is made for embedding epistemic anti-individualism outside of the knowledge-first framework.   As in the case of misleading defeaters.   For a helpful recent overview of work on the epistemic basing relation, see Korcz (2015). 10   We shall not discuss the extended mind/cognition thesis in this chapter, as in Clark and Chalmers (1998). Everything we say applies mutatis mutandis if individualistic properties are understood to include mental properties or cognitive processes that extend beyond the bodily boundaries of the individual in question. For more on how to extend the extended mind/cognition thesis to epistemology, see Carter and Kallestrup (2016). 8

9

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  203

2.  Robust Virtue Epistemology Any virtue-theoretic account of knowledge says that believing truly through epistemic virtue is a necessary condition on knowledge. Robust virtue epistemology (RVE) says that such a virtue-theoretic condition is also sufficient for knowledge. According to Sosa’s (2011; 2013) version of RVE, knowledge has a triple-A structure.11 Knowledge implies true belief, and a belief which is true is accurate. Knowledge also implies a belief which is formed out of epistemic virtues. A belief formed on the basis of reading tea leaves or crystal ball gazing cannot constitute knowledge, even if true. But when formed through reliable cognitive abilities, which is how Sosa prefers to view epistemic virtues, the belief is adroit. Accuracy and adroitness are, however, still insufficient for knowledge, as is illustrated by standard Gettier-cases involving intervening epistemic luck. Consider a familiar case: through exercising S’s reliable visual apparatus S forms the belief that there is a sheep in the field. In actual fact, S sees a rock which looks just like a sheep, but S still believes truly because a sheep is hidden behind the rock.12 So, to rule out the possibility that the truth of a belief be down to such happenstance, Sosa requires aptness for knowledge—i.e., that the belief be accurate because adroitly formed.13 Apt belief, for Sosa, is knowledge.14 Now recall our definition of warrant as that which converts true belief into ­knowledge. On Sosa’s view, the conversion consists exclusively in the aptness of the belief, i.e., in its accuracy being through adroitness. Since no separate (modal or otherwise) condition is needed, his theory of knowledge is a version of RVE. When S believes truly, all that is required for that belief to qualify as knowledge is that S meets a virtuetheoretic condition. And for S to meet that condition is for S’s belief to be true as a result of S exercising a pertinent cognitive ability. As Sosa (1991) accepts ES, an intriguing question is whether warrant, when ­construed in terms of aptness, supervenes on individualistic properties. Sosa (2007, 29; 2009, 135) explicitly conceives of cognitive abilities in terms of cognitive dispositions which have physical bases wholly resident in whoever has those dispositions.15 This 11   For two prominent defences of alternative versions of RVE, see Zagzebski (e.g.,  1996;  1999) and Greco (e.g., 2009). 12   This Gettier-style example was originally due to Chisholm (1977, 105). 13   In more recent work, Sosa (2013; 2015) proposes a triple-S analysis of a complete competence comprising an innermost S-competence, which is the seat (or skill), an inner SS-competence, which is the combination of seat and shape, and a complete SSS-competence, which is the conjunction of seat, shape, and situation. The connection between the triple-A analysis of a performance and the triple-S analysis of a competence is the following: a performance is apt when its success manifests competence, which happens just in case the innermost skill causally produces the success in combination with the appropriate shape and situation. The seat of the competence is determined as the causal basis for a success-response of an object when subjected to a stimulus in certain shape and situation combinations. 14   Or rather, apt belief is what Sosa calls ‘animal’ knowledge. In contrast, reflective knowledge is apt belief that one’s first-order belief is apt. 15   Interestingly, in the specific context of social epistemology, Sosa (2007, 93–8; 2015, 86–90) argues that testimonial knowledge involves complex competences that are socially seated, or seated in a group collectively. But our claim is that all kinds of warrant fail to supervene on individualistic properties of S.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

204  Jesper Kallestrup and Duncan Pritchard suggests that adroitness is an epistemic property that supervenes on individualistic properties. After all, dispositional properties are arguably instantiated, at least as a matter of nomological necessity, whenever their base properties are instantiated. But, by the lights of Sosa’s RVE, adroitness is only one ingredient of warrant. The truth of the belief must also be because of adroitness, where the ‘because of ’ at issue is to be understood in terms of the manifestation of a cognitive disposition. So, the question is whether disposition manifestation, thus understood, is a property that supervenes on individualistic properties. Consider the following thesis: Strong Robust Virtue Epistemology (SRVE) Necessarily, if individual S manifests cognitive disposition CD (to form true beliefs), then S has individualistic properties I, such that, necessarily, any other individual S* who has I also manifests CD. Cases involving intervening epistemic luck provide a reason to reject SRVE. Suppose S and twin-S possess the same cognitive dispositions in virtue of being individualistic duplicates who normally occupy the same knowledge-friendly environment; call it their global environment. Add that they share causal histories, if you like. Indisputably, S comes to know that there is a sheep in the field on the basis of looking at a real sheep in circumstances in which there are no fake sheep. However, twin-S forms a true belief with the same content on the basis of looking at a sheep-shaped rock behind which a real sheep is hidden from view. The intervening epistemic luck arguably prevents ­twin-S’s belief from counting as knowledge. Importantly, Sosa’s RVE can handle such cases without conceding that warrant supervenes on individualistic properties: S’s belief is true through the exercise of S’s visual disposition, but twin-S’s belief is true through environmental coincidence. Since S’s belief is apt and twin-S’s belief is inapt, the former, but not the latter, counts as knowledge.16 Let’s instead consider a Gettier-case involving environmental epistemic luck. Suppose S comes to know there’s a barn on the basis of looking at a real barn in circumstances in which there are no fake barns in the vicinity. In contrast, S’s individualistic duplicate twin-S forms a true belief with the same content on the basis of looking at a real barn in circumstances in which twin-S might easily have formed the same belief on the basis of looking at a fake barn.17 Intuition has it that the environmental epistemic luck prevents twin-S’s belief from qualifying as knowledge, but in this case no resources seem available to Sosa’s RVE to account for an epistemic difference between S and twin-S. Just as in the case of intervening epistemic luck, both S and twin-S possess the same cognitive dispositions in virtue of their individualistic indiscernibility and 16   Here is a residual worry: if Sosa admits that warrant fails to supervene on individualistic properties, the question arises of how knowledge can be viewed as a cognitive achievement for which the individual deserves full credit. Is part of the credit not due to knowledge-conducive features of the proximate environment? For a negative answer, see Jarvis (2013). 17   The ‘barn façade’ case was originally described in Goldman (1976), and credited to Carl Ginet.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  205 occupying global environments that are equally conducive to knowledge. Against this background, there is no reason why S and twin-S’s beliefs should differ in respect of their aptness. In particular, even the truth of twin-S’s belief is down to an exercise of her disposition to form visual beliefs. After all, the local physical conditions that need to obtain for twin-S to manifest that disposition are identical to those that obtain in the case of S, such as a clear view of a real barn in close proximity. As we have argued elsewhere (Kallestrup and Pritchard 2011; 2013), the large number of barn facsimiles in twin-S’s regional environment, beyond her local environment containing the real barn that she is currently facing, presents a nearby possibility of error which renders her barn-belief unsafe (and, for that matter, insensitive)—i.e., her cognitive success (true belief) is such that it could very easily have been formed in the same way and yet have been cognitive failure (false belief).18 Figure 10.1 illustrates our epistemic twin earth scenario: Scenario I

Scenario II

Global: Real barns

Global: Real barns

Regional: Real barns

Regional: Fake barns

Local: Real barn S

Local: Real barn Twin-S

Figure 10.1  Epistemic Twin Earth.

The explanation Sosa offers of why S has knowledge is that S’s cognitive success is because of her cognitive ability. That is, her belief is apt. The challenge, however, is to explain why twin-S lacks knowledge, since her belief seems to be no less apt. The fact that S and twin-S are individualistic duplicates embedded in physically identical global environments means that one cannot possess a cognitive ability the other lacks. And the fact that S and twin-S are currently located in physically identical local environments rules out the possibility that only one of them manifests that ability. So, if S’s belief is apt, then so too must be twin-S’s belief. Indeed, Sosa (2007) himself admits that some easy error-possibilities, such as those presented by barn-façades, are perfectly compatible with the aptness of twin-S’s belief. 18  The distinction between environmental and intervening epistemic luck is drawn, and further e­ laborated on, in Pritchard (2009a; 2009b, Chs. 3–4; 2009c; 2012a; Pritchard, Millar, and Haddock 2010, Chs. 2–4).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

206  Jesper Kallestrup and Duncan Pritchard Consequently, Sosa must concede that twin-S has knowledge.19 This means that Sosa would endorse the following thesis about the manifestation of cognitive dispositions: Weak Robust Virtue Epistemology (WRVE) Necessarily, if individual S manifests cognitive disposition CD (to form true beliefs), then S has individualistic properties I under local conditions L such that, necessarily, any other individual S* who has I under L also manifests CD. The problem is of course that many epistemologists follow Alvin Goldman (1976) in taking environmental epistemic luck to undermine knowledge simpliciter no less than intervening epistemic luck does. If that is right, then not only does warrant fail to supervene on individualistic properties, as illustrated by twin scenarios involving intervening epistemic luck, but warrant fails also to supervene on the conjunction of individualistic and local properties, as illustrated by twin scenarios involving environmental epistemic luck. Instead, warrant supervenes, at least in part, on non-epistemic properties pertaining to (regional) environmental features beyond S’s immediate vicinity (or local environment). Put differently, warrant is not just anti-individualistic; warrant is radically anti-individualistic, in that it is incompatible with both SRVE and WRVE.20

3.  Evidentialist Mentalism Having argued that warrant fails to supervene on individualistic properties, we turn now to the question of whether other epistemic properties supervene on such ­properties. Amongst those epistemologists who endorse ES vis-à-vis doxastic or ­propositional justification, there’s a familiar disagreement over the precise character of  the base properties. Ironically, while these base properties are obviously not ­themselves epistemic, this disagreement often pertains to epistemic features of these properties. Here we have in mind the dispute between epistemic internalists and ­externalists, which primarily concerns doxastic justification, although the former occasionally ­formulate their view in terms of propositional justification.21 Call the 19   Sosa (2007) attempts to explain away the intuition that twin-S lacks knowledge by claiming that, while environmental epistemic luck is compatible with animal knowledge, such luck does undermine reflective knowledge. We argue in Kallestrup and Pritchard (2011) that when reflective knowledge is construed as meta-apt belief, a corresponding argument from epistemic twin scenarios shows that environmental ­epistemic luck equally undermines reflective knowledge. A bolder tack is instead to adopt Turri’s (2011) revisionary theory according to which knowledge simpliciter is incompatible with intervening e­ pistemic luck, yet fully compatible with environmental epistemic luck. 20   At root, the question in play here is whether knowledge demands safety, viz., that when one knows, one’s cognitive success (true belief) could not have very easily been (by the same method of belief-formation) cognitive failure (false belief). For further discussion of safety, see Sainsbury (1997), Sosa (1999), Williamson (2000), and Pritchard (2002; 2005; 2007; 2012a; 2012c; 2015a). For a recent exchange on the specific question of whether knowledge demands safety, see Hetherington (2013) and Pritchard (2013). 21  For example, the evidentialist epistemic internalist position defended by Conee and Feldman (2004)—and which we will be considering in detail further on in the chapter—is expressed in terms of propositional justification.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  207 non-epistemic properties on which being justified supervenes ‘justifiers’. Justifiers are the properties that confer justification on beliefs. Thus, epistemic internalists hold that the property of being doxastically justified supervenes on justifiers that are internal to S’s mind (or  cognitive perspective), while epistemic externalists maintain that this property supervenes in part on justifiers that are external to S’s mind. The question is what this internal/external distinction delineates. So-called accessibilists claim that the internal should be understood in terms of what is reflectively accessible, in the sense of what one can become consciously aware of through reflection. So, on their view, being doxastically justified supervenes on ­justifiers to which S has reflective access. Consider: Accessibilism Necessarily, if individual S’s belief that p is justified, then S has reflective access to justifiers j such that, necessarily, any other individual S* who believes p and has reflective access to j also holds a belief that is justified. To say that S’s belief that p is justified just in case she has reflective access to the properties that justify that belief can be taken in at least two distinct ways. Weakly understood, what is required is conscious awareness through reflection of the presence of those properties (i.e., the justifiers) that justify belief in p. Strongly understood, what is required is also conscious awareness through reflection that those properties justify belief in p.22 Whether accessibilism counts as epistemic individualism depends on whether S has reflective access only to her individualistic properties. For instance, if S is deemed to have reflective access to justificationally relevant environmental properties beyond the skin-and-skull of S, then accessibilism as a stripe of epistemic internalism is compatible with epistemic anti-individualism. Likewise, if by the lights of accessibilism, being internal is cashed out in terms of reflective access, then epistemic externalism is the view that the property of being doxastically justified supervenes on properties to which S need not have reflective access. Thus understood, epistemic externalism is compatible with epistemic individualism if the base properties (i.e., justifiers) on which this epistemic property supervenes are individualistic, yet such that S need not have reflective access to them. We shall not here delve further into the details.23 The point to keep in mind is that given the different ways the internal/external distinction can be drawn, the dispute over epistemic internalism is different from the dispute over epistemic individualism. Let’s instead turn to ponder a different version of epistemic internalism, namely ­evidentialist mentalism.24 Evidentialist mentalists and accessibilists agree that the   For two key defences of accessibilism, see Chisholm (1977) and Bonjour (1985, Ch. 2).   For more details, see Kallestrup (2015). 24   For the core defence of evidentialist mentalism, see Conee and Feldman (2004). For some useful discussions of the debate between accessibilists and mentalists, see Steup (1999), Pryor (2001, Section 3), Bonjour (2002), Pappas (2005), and the papers collected in Dougherty (2011). 22 23

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

208  Jesper Kallestrup and Duncan Pritchard properties on which doxastic justification supervenes are internal to S’s mind (or cognitive perspective). They also agree that justification consists in reasons or evidence, understood broadly as signs or marks of truth. But they disagree about what it means for such base properties to be internal. Evidentialist mentalists hold that only mental properties are internal to S’s mind, understood broadly to include occurrent or dispositional mental states, events, and conditions.25 So, on their view only mental properties can serve as base properties (i.e., justifiers), and hence all reasons or evidence consist in mental states. More precisely, Conee and Feldman’s version of evidentialist mentalism (2004, 83, 204; 2008, 83) says that, as a matter of necessity, S justifiably believes p if and only if S’s evidence (on balance) supports p.26 That is to say, the evidential facts and facts about doxastic justification are necessarily equivalent. Their mentalism then says that necessarily, there can be no evidential difference between any two possible individuals without a mental difference. That is to say, the evidential facts strongly supervene on the (totality of the) mental facts. Putting the two together, Conee and Feldman (2004, 56) accept that the mental fixes the justificatory status of beliefs such that if individuals S and S* have the same mental properties, then they are necessarily justificational duplicates.27 Bearing in mind that occurrent or dispositional mental states, events, and conditions all count as individualistic properties, consider the following formulation of evidentialist mentalism. Evidentialist Mentalism (EM) Necessarily, if individual S’s belief that p is justified to degree d, then S has individualistic properties i such that, necessarily, any other individual S* who believes p and has i also holds a belief which is justified to degree d. Thus formulated, EM is not merely a claim about the supervenience base for doxastic justification. This view also takes the gradability of doxastic justification into account: it says that the degree (or extent) to which S is justified in believing p supervenes on her mental properties. As Conee and Feldman (2001, 234) put it, it is a consequence of the supervenience of justification on the mental that ‘if any two possible individuals are exactly alike mentally, then they are alike justificationally, e.g., the same beliefs are ­justified for them to the same extent.’ Observe that the non-epistemic properties on which doxastic justification supervenes are characterized metaphysically in terms of

25   Here we shall equate an occurrent state with a conscious state, and we shall take a dispositional state to be a non-occurrent state. Hence, dispositional states are not conscious states. 26   Conee and Feldman (2008, 85–86) sketch different kinds of evidence. Their preferred notion is that of justifying evidence: for S to have evidence for p is for S to have a reason to believe p (i.e., justifying evidence is something S could cite as a justifying basis for belief in p. Note also that evidence, on their usage, is always evidence for someone—i.e., it is always something that someone has. 27   Elsewhere Conee and Feldman (2004, 101) maintain that justification supervenes on evidence. Given their mentalist claim that evidence supervenes on the mental, it follows via transitivity of supervenience that justification supervenes on the mental.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  209 the mental nature of certain states, processes, and conditions.28 So, evidentialist mentalists demarcate what is internal to an individual’s mind in terms of such mental factors. So far, we have formulated evidentialist mentalism in terms of doxastic justification, but that raises the question of whether the basing relation can be fleshed out in mentalist terms. Conee and Feldman (1985, 93; 2004, 94) offer an account of that relation in terms of a belief fitting the evidence. The idea is roughly that S’s belief that p is wellfounded (or properly based) if and only if having that belief is justified for S and S has justifying evidence e on the basis of which S believes p, where the latter is understood in terms of S’s belief fitting e. In response, Goldman (2012) and Juan Comesaña (2005) have argued that the prospects for explaining this fitting relation in evidentialist mentalist terms look dim. For instance, if S’s belief that p fits e only if that belief is caused by e, then it is hard to see how epistemic fitting could be a mental relation. Feldman (2005) suggests that, since all evidential facts are mental facts, we can understand the causal relation between e and the belief that p as an instance of mental-to-mental causation, and therefore as a mental fact. But this proposal rests on a mistake: what makes mentalto-mental causation different from mental-to-physical or physical-to-physical causation has to do with the nature of the causal relata. Causation itself is homogeneous. In these latter cases, there is no temptation to conceive of causation as involving mental forces, spirits, or what have you. Nor therefore should there be in the former case. Let’s therefore make the charitable assumption that evidentialist mentalism is only (or primarily) a view about the supervenience of propositional justification on mental properties. Nothing hangs on this since our objection applies equally to this weaker view. Given that the relevant mental properties count as individualistic, consider: Evidentialist Mentalism* (EM*) Necessarily, if individual S has justification to degree d to believe p, then S has individualistic properties i such that, necessarily, any other individual S* who has i also has justification to degree d to believe p. In other words, whatever property turns S’s belief into a belief for which S has justification supervenes on individualist properties of S. In the following we shall mount a dependence thesis to the effect that propositional justification depends negatively on physical features of the environment beyond the individual who has that justification. Our dependence thesis is established by an ­epistemic twin earth scenario, which shows that when we vary such physical features, S and twin-S can differ justificationally, despite being individualistic duplicates.29

28   As Conee (2007, 51) puts it, ‘mentalism is the thesis that for epistemic purposes, the “internal” is the mental.’ See also Conee and Feldman (1985, 55). 29   The following epistemic twin earth scenario and the concomitant epistemic dependence thesis are developed in more detail in Kallestrup (2015).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

210  Jesper Kallestrup and Duncan Pritchard Our twin earth scenario in favor of the negative epistemic dependence thesis vis-àvis propositional justification bears some resemblance to the new evil demon (NED) problem for process reliabilism, originally raised by Keith Lehrer and Stewart Cohen (1983). The NED case aims to show that reflectively inaccessible factors, or factors beyond conscious awareness, are unnecessary for justification. In effect, our twin earth scenario turns the tables on the NED case against the reliabilist. Before we proceed with our argument, it is thus worth revisiting the NED problem. Process reliabilism, remember, is the view that a belief is justified only if the process by which it was produced (and sustained) is reliable. Now imagine S in the actual (or normal) world and individualistic duplicate twin-S in a world where everything seems to be the way things are in the actual world. S and twin-S are experiential and doxastic duplicates, i.e., they share indistinguishable perceptual experiences. The only difference is that twin-S is deceived by a Cartesian demon into falsely believing everything S truly believes about the external world. Hence, only S’s beliefs about the external world are justified if process reliabilism is true. Twin-S’s beliefs are unjustified if being justified requires being produced by a reliable process. After all, the processes by which twin-S’s beliefs were produced led exclusively to false beliefs, and so those processes, unlike the processes by which S arrived at beliefs, are wholly unreliable.30 However, S and twin-S should intuitively enjoy the same justificational statuses (even though only S attains knowledge). NED is thus a challenge to the process reliabilist to accommodate an entrenched intuition that twin-S’s beliefs are justified no less than S’s beliefs. A pressing question to ask is: what underlies the judgment that S and twin-S should hold beliefs with identical justificational statuses when the processes by which they were produced led to markedly different truth-ratios? Cohen (1984, 282) is explicit that his argument hangs on a normative (or deontological) conception of justification, according to which, ‘if S’s belief is appropriate to the available evidence, he is not to be held responsible for circumstances beyond his ken.’ In contrast, what NED purports to show is that the reliabilist’s probabilistic notion of justification is flawed in that reliability is unnecessary for justification. More precisely, there is no conceptual connection between such justification and objective truth-frequency. Rather, Cohen (1984, 284–85) maintains that being justified is simply a matter of being reasonable or rational.31 The only conceptual connection between justification and truth is at the doxastic level: if conditions C justify a belief for individual S, then C entails, not that S’s belief is probably true, but that S believe that certain conditions obtain which make that belief probably true.32   Cohen (1984, 281–323).   Interestingly, pretty much the same conception of justification undergirds the epistemic intuition that is operative in contemporary internalist defenses of the NED problem, such as in Huemer (1999) and Wedgwood (2002). 32   Reliabilists take the connection between justification and truth to be captured by the conceptual claim that if conditions C justify a belief for individual S, then C makes that belief probably true. They typically reject the infallibilist claim that C entails the truth of S’s belief if C justifies that belief for S. 30 31

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  211 Interestingly, Feldman (2003) also explicitly employs NED against process reliabilism. Indeed, Conee and Feldman (2004, 155) take S in the normal world and twin-S in the demon world to be mental twins, and so to be justified (to the same degree) in believing the same propositions. But this conflicts with their acceptance (2004, 252–3) that justification is always an indication of truth. After all, the alleged justification that the radically deceived twin-S has is systematically misleading. Thus, Feldman (2005, 277) understands justification in terms of evidence or good reasons comprising the perceptual experience and doxastic states that S has to go on in forming beliefs, but he is adamant that such reasons are not to be interpreted deontologically in terms of ­fulfilling epistemic duties. Likewise, Conee (1992, 664–5) maintains that justification is truth-oriented in that justification of p for S is evidence for S of the truth of p. Moreover, p is then justified for S when it is evident to S that p is true. In that sense, possession of evidence is an indication of truth. And Conee and Feldman (2004, 252) identify all evidence that S can have for p with indications to S that p is true. Consequently, all justification that consists in evidence for p bears on the truth of p. In fact, Conee and Feldman (2004, 62–3) take their view to be defensible solely on nondeontological grounds.33 Since any view that construes justification (even in part) as a matter of objective probability, as opposed to subjective rationality, will face the problem posed by NED, the foregoing shows that EM* is no better placed than process reliabilism vis-à-vis this problem. However, NED involves a skeptical hypothesis which raises a number of vexed issues to do with the nature of skeptical doubt. Let us instead consider a scenario closer to home.34 Consider the following dry earth twin scenario. First, consider S, who is on earth where water plays the stereotypical role of being the clear, potable liquid which fills the 33   If justification is taken not to require an objective indication of truth, then it is hard to explain how justification can be a guide to knowledge. We defined warrant as that which converts true belief into knowledge. Given that justification is an indispensable component of warrant, justification plays a key role in converting true beliefs into knowledge. Absent cases of improper basing and epistemic defeat, ­justification is exactly what makes the difference between true belief and knowledge. Justification may play additional deontological roles, but if we sever any connection with truth, we cannot take justification as a guide to knowledge. 34   Other philosophers have also argued that NED or similar cases pose a problem for epistemic internalist accounts. Gibbons (1996) used the possibility of unpossessed evidence that one ought to have possessed to argue against accessibilism, by relying on cases where two introspectively identical individuals differ in respect of whether they fulfil their epistemic obligations. Lyons (2013) points out that NED presents a problem for what he calls ‘seemings internalism’, the view that S is prima facie justified in believing that p if S is appeared to as if p. Lyons’s argument hinges on the claim that memory, deduction, etc., are conditionally reliable even in the demon world. Moon (2013) uses NED against forms of epistemic internalism which rely on S and deceived duplicate twin-S being internally identical except that a number of S*’s un-accessed internal states are deleted by a demon and then replaced with phenomenologically indistinguishable occurrent states. Goldberg (2012) defines being doxastically justified as the property that turns true ­un-Gettiered belief into knowledge. On the basis of a NED case, he then concludes that no property that is internal in the internalist sense is the property of being doxastically justified. Our twin scenario is different from all of these. It relies neither on the distinction between categorically and conditionally reliable ­processes, nor on the distinction between accessed and unaccessed mental states, nor on intuitions about knowledge-ascriptions.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

212  Jesper Kallestrup and Duncan Pritchard oceans and rivers, falls from the sky, etc.,—in short, the watery stuff. On the basis of a veridical perceptual experience as of a jug containing water, S forms the true belief that there is water in the jug. S is normally highly reliable when it comes to distinguishing (by sight, taste, etc.) between water and relevantly different liquids. Moreover, the proximate conditions in which S makes her perceptual observation are normal. Against this background, S forms a justified belief in that S bases her belief on the available justification that is provided by her perceptual experience. Now contrast S with her individualistic duplicate twin-S who has recently and unwittingly been whisked off to dry earth, which is just like earth except there is no water. Despite all appearances, on dry earth the lakes, rivers, taps, etc., all run bone dry. The entire community on dry earth mistakenly believes that there is a stuff which they call ‘water’ and which has all the watery properties. Based on a hallucinatory experience as of there being water in the jug, twin-S forms the false belief that there is water in the jug. Apart from the missing water, the perceptual conditions are normal both in this case and in general. Even so, when on dry earth twin-S is obviously not reliable at all in telling when water is present, and so the pressing question is whether twin-S has justification for her belief, let alone can form a justified belief. From a purely deontological point of view, any justificational discrimination against twin-S seems unmotivated. The beliefs twin-S uses ‘water’ to express are as epistemically responsible as the beliefs that S forms about water. From the point of view of subjective rationality, neither is to be blamed for their doxastic behavior. But now switch attention to a non-deontological conception of justification on which holding a justified belief is a matter of believing on a basis that objectively probabilifies that belief. It seems equally obvious that S’s belief is justified in a way (or at least to an extent to which) twin-S’s belief is not. For twin-S to form a belief on the basis of what looks like water is almost bound to lead her astray. The experiential evidence twin-S has available is likely to produce very few, if any, true beliefs about water. Not so in the case of S whose belief-forming method vis-à-vis water is highly reliable. The experiential ­evidence S has to go on is highly conducive to the formation of true beliefs. Given that evidentialist mentalists endorse a probabilistic conception of justification according to which justification consists in evidence which supports p if and only if it raises the (objective) probability of p, they predict that S and twin-S must differ justificationally. But since S and twin-S are stipulated to be individualistic duplicates, our dry earth scenario presents a counterexample to EM*.35 Note that adopting a 35   A friend of evidentialist mentalism may question whether in our dry earth twin scenario S and twin-S are individualistic duplicates for the reason that they fail to share content-bearing mental states due to the physical differences in their environments. Thus, Conee (2007, 57–8, 63) explicitly adopts wide content mental states as part of the supervenience base for propositional justification. In response, one should first note that the content of their beliefs—i.e. that there is water in the jug—is descriptive, which therefore need not be individuated by environmental features. Secondly, many semantic externalists accept that tokens of ‘water’ express the concept of water even in a dry earth scenario. This means that S’s and twin-S’s utterances of sentences containing ‘water’ both express propositions containing that concept, and hence, both are in belief states the content of which includes that concept. For more details see Kallestrup (2011; 2015).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  213 hybrid view on which justification is partly deontological (to do with fulfilling obligations) and partly probabilistic (to do with objective likelihood) provides no answer to this problem. As long as justification requires as a component the claim that a belief be formed on a basis which makes its truth objectively likely, there will be a clear difference in the justificational status between S’s and twin-S’s beliefs. Perhaps on the hybrid view that difference is one of degree of justification, but EM* is formulated in terms of  the degree to which an individual has justification to believe p. Consequently, ­propositional justification fails to supervene on individualistic features. The evidentialist mentalist might object that the justificational difference between S and twin-S is, after all, down to a mental difference, namely S having a veridical experience while twin-S undergoes a hallucinatory experience. In reply, the scenario can be tweaked to involve an illusory experience instead. Suppose twin-S* occupies an ­environment in which water (i.e., H2O) appears to have watery properties that it actually lacks. On the basis of an experience of water having some such property, twin-S* forms a false belief about water. Again, the individualistic duplicates S and twin-S* ­differ ­justificationally. Of course, the evidentialist mentalist might again insist that the justificational difference between S and twin-S* is due to the mental difference between having veridical and illusory experiences. A better response is to query whether these mental differences should count as ­justificationally relevant by the lights of evidentialist mentalism. We have already seen that Conee and Feldman (2004, 155) explicitly take S in the normal world and twin-S in the demon world to be mental duplicates, despite the fact that only the former has a veridical experience. In fact, they seem to assume that the more transparent to awareness a perceptual experience is, the more justified a belief is, if based on that experience.36 Thus, since the differences in veridicality between S’s and twin-S/twin-S*’s experiences are opaque to awareness—i.e., these experiences are phenomenologically indistinguishable—they need not make for a justificational difference. Also, note at this juncture that the arguments which Conee and Feldman (2001; 2004) bring to bear in support of evidentialist mentalism all involve pairs of cases in which two individuals differ justificationally in virtue of mental differences of which they are transparently aware. Here is one of their illustrative examples: Bob and Ray are sitting in an air-conditioned hotel lobby reading yesterday’s newspaper. Each has read that it will be very warm today and, on that basis, each believes that it is very warm today. Then Bob goes outside and feels the heat. They both continue to believe that it is very warm today. But at this point Bob’s belief is better justified.  (Conee and Feldman 2001, 3)

Conee and Feldman take the mental change that Bob undergoes when he feels the heat outside to enhance the justification for his belief—a change that involves distinctive sensational features of which Bob is transparently aware.37   See Conee and Feldman (2008, 100).   For more detailed discussion of responses to epistemic dry earth scenarios on behalf of evidentialist mentalism, see Kallestrup (2015). 36 37

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

214  Jesper Kallestrup and Duncan Pritchard The upshot is that as far as Conee and Feldman are concerned, S and twin-S/twin-S* are mental duplicates, and so given EM* also justificational duplicates. But S and ­twin-S/twin-S* ought to differ justificationally if, as Conee and Feldman accept, justification is to be understood, even in part, as an objectively probabilistic notion.38

4.  Knowledge-First Epistemology According to Williamson’s (2000) knowledge-first approach, no individual S is guaranteed reflective, introspective or otherwise privileged access to her evidence. On his view, S may have some evidence to which she lacks any such access; indeed, she may be entirely mistaken about what her evidence is. Given that S’s evidence determines whether she has justification to believe a proposition p, this approach is at odds with the accessibilist claim that S has privileged access to the factors which determine whether she has justification to believe p. Instead, Williamson claims that only knowledge justifies belief. Since what justifies belief is evidence, it follows that only knowledge constitutes evidence. S’s evidence is thus equated with her knowledge (Ep = Kp). So, instead of using non-epistemic ­concepts such as those of causation and reliability to characterize the concept of ­knowledge, Williamson reverses the traditional order of explanation by explicitly deploying the concept of knowledge to elucidate, perhaps only a posteriori, the ­concepts of justification and evidence: The concept knows is fundamental, the primary implement of epistemological inquiry. (Williamson 2000, 185)

Now, if the total evidence available equals the total knowledge available, then S is never guaranteed to be in a position to know what her evidence is. For if evidence were epistemically transparent in that sense, then Ep and KEp would be equivalent. Given Ep = Kp, that would make Kp and KKp equivalent, but (claims Williamson) Kp fails to entail KKp. What then about epistemic individualism? It should be obvious that Williamson would reject the epistemic individualist view that the properties of being justified and having evidence supervene on individualistic properties. He maintains that: E = K is an externalist theory of evidence, in at least the sense that it implies that one’s evidence does not supervene on one’s internal physical states. But if knowing is a mental state [. . .], then one’s evidence does supervene on one’s mental states. (Williamson 2000, 191) 38   Interestingly, some non-classical versions of epistemic internalism explicitly disavow a commitment to NED. This is true, for example, of epistemological disjunctivism, a proposal rooted in the work of McDowell (e.g., 1995) and developed at length in Pritchard (2008; 2012b; 2015b; cf. Neta and Pritchard 2007). See, in particular, Pritchard (2011), which explicitly contrasts epistemological disjunctivism with evidentialist mentalism. The problems facing views such as evidentalist mentalism when it comes to its adherence to NED thus offer prima facie support for these non-classical proposals, which are, in principle, better placed to accommodate the arguments in support of epistemic anti-individualism. For an important recent discussion of the relationship between NED and epistemic internalism, see Littlejohn (2012).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  215 Notice that Williamson seems to be drawing the internal/external distinction around the bodily boundaries of S: The internal will be identified with the total internal physical state of the agent at the relevant time, the external with the total physical state of the external environment.  (Williamson 2000, 51)39

The reason these epistemic properties fail to supervene on individualistic properties is presumably that they supervene on knowledge, which is factive and, in most cases, pertains to environmental features beyond S. As a claim about individuation, i.e., about what makes something knowledge, knowledge is not a property that is internal to the bodily boundaries of S. That is consistent with knowledge being instantiated by S, and so in that sense being located within the body of S. But knowledge is also a mental state, according to Williamson. Indeed, knowledge is the most general factive mental state— i.e., it is the mental state S is in if she is in any factive mental state at all. It follows that Williamson is a mentalist, or an evidentialist mentalist, in the sense that the properties of being justified and having evidence supervene on the mental states of knowing. Put differently, all justifiers are mental states. What distinguishes Williamson’s view from evidentialist mentalism as defended by Conee and Feldman (e.g., 2008, 87–8) concerns the nature of the mental states that can serve as evidence. While the latter regard beliefs as indirect evidence, only nondoxastic states, such as perceptual experiences, are ultimate evidence. Importantly, as we argued in Section 3, the veridicality of an experience adds no justificatory value. Conee (2007, 62–6) tentatively suggests that knowledge states can play a justifying role. However, due to the factivity of knowledge, it is unclear how, by Conee and Feldman’s lights, such states can constitute evidence as they are opaque to awareness and so not wholly internal to the mind.40 In contrast, since Williamson only allows for states of knowledge to constitute evidence, on his view only factive mental states can serve as evidence. So, Williamson’s mentalism is non-individualistic in that the mental states on which evidence supervenes are all factive, whereas Conee and Feldman’s mentalism is individualistic in that the mental states on which justification (or evidence) supervenes are all non-factive. For Williamson justifiers are thus not individualistic properties in the way they are for Conee and Feldman. He is, therefore, not committed to any version of epistemic individualism; indeed his knowledge-first approach clearly implies anti-individualism with respect to the property of having ­evidence. The problem is rather, as we shall now argue, that this approach fails to offer an explanatory account of ES.

39   What Williamson is saying here is admittedly consistent with an extended conception of the mental/ cognitive, but as mentioned in n. 10, we are setting the extended mind/cognition thesis aside in this chapter. 40   See also Conee and Feldman (2008, 99–101). Wedgewood (2002) defends a version of mentalism according to which justification supervenes on non-factive mental states, as well as explanatory relations between those states.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

216  Jesper Kallestrup and Duncan Pritchard Williamson (1995; 2000, 28, 47–8, 51; 2009) argues that knowledge does not a priori factor into separate conditions—belief, truth, and something else—which are individually necessary and jointly sufficient for knowledge. No such reductive analysis of the state of knowing is forthcoming. That is to say, knowledge is no compound of a mental state and a non-mental condition. Rather, knowledge is a sui generis mental state, or more specifically, the most general factive propositional attitude. To say that knowledge is a factive attitude means that ‘not-p and S knows that p’ is a contradiction. Factive attitudes matter to us. Williamson (2000, 39) takes their distinctive value to consist in having an essence that involves a matching between mind and world. To shift focus away from such factive states is to forego the importance of that matching relation in our epistemic pursuits. Knowing makes demands on the world that believing does not. Notably, since believing is a non-factive attitude, believing involves no matching between mind and world. Here is Williamson: [. . .] to know is not merely to believe while various other conditions are met; it is to be in a new kind of state, whose essence involves the world. What is required is [. . .] rejection of a conjunctive account of knowing.  (Williamson 1995, 539; cf. Williamson 2000, 46)

Williamson (2000, 33ff) says that factive, stative attitudes are expressed in natural language by factive mental state operators (FMSO), which combine syntactically with other expressions in the way verbs do, yet admit of no semantic analysis. More precisely, an FMSO Φ takes as subject a term S for something animate, and as object a sentence prefixed by ‘that’, but Φ is not synonymous with any complex expression the meaning of which is composed of the meanings of its constituent expressions. So, Φ is semantically unanalyzable: its sole semantic role is to denote an attitude. Examples include the verbs ‘know’, ‘see’, and ‘remember’. Against this backdrop, Williamson (2000, 37) claims that if Φ is any FMSO, then ‘S knows that p’ is entailed by ‘S Φs that p’. This proposal is the linguistic analogue of the claim that S is in a knowledge state with content p if S has any factive and stative attitude to p. It follows that ‘believe truly’ is not an FMSO, because S can believe truly without knowing. Indeed, that ‘believe truly’ is not an FMSO is also a corollary of the semantic unanalysability of FMSOs. The mental state that ‘believe truly’ picks out is that of believing. Truth adds nothing mental to the state of believing. Believing truly is both a factive and stative attitude, but for an attitude to entail knowing, the mental state itself must be sufficient for truth. So, given that ‘seeing’ and ‘remembering’ are FMSOs, their referents must be mental states sufficient for truth. While Williamson holds that knowledge is a sui generis mental state, he is not denying that knowledge entails belief. That is to say, belief is a necessary condition on knowledge, but belief is no component of knowledge.41 Instead, believing should be analyzed in terms of knowing: to believe p is to treat p as if p is known—i.e., when p is believed, 41   See Williamson (2000, 41ff). Notice that this stance raises the question of whether a true sentence of the form ‘S knows that p’ ascribes two distinct mental states to S.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  217 p is treated in ways that are similar to the ways in which p is treated when known (e.g., as a premise in practical reasoning). Thus, knowledge sets the gold standard for belief. The more believing approximates to knowing, the more appropriate believing becomes. Belief simply aims at knowledge, and belief that fails to hit the target is a kind of botched knowing.42 So, Williamson makes two related claims. First, the linguistic (or conceptual) claim that ‘know’ is semantically unanalysable in that ‘know’ is no synonym for a complex expression comprising ‘truth’, ‘belief ’, and some other expression. We can take ‘knowledge’ to be shorthand for ‘warranted true belief ’, but, in that case, the meaning of ‘warrant’ cannot be explicated without using ‘knowledge’. Second, the metaphysical claim that the referent of ‘know’ is a sui generis mental state, rather than a motley of mental and non-mental components. That is to say, the mental state of knowing cannot be factorized into belief, truth, and something else which is not knowledge. The pressing question is now: if, as Williamson (2000, 186) claims, all S’s knowledge serves as the foundation for her evidence and justified beliefs, then on which nonepistemic properties do these epistemic properties supervene? Unless knowledge itself supervenes on non-epistemic properties, the latter cannot supervene on knowledge on pain of violating ES, according to which all epistemic properties supervene on nonepistemic properties. After all, knowledge is a canonical epistemic property. But given what Williamson says about the nature of evidence and justification, only knowledge seems to be what constitutes the base on which such epistemic properties supervene. Importantly, one can consistently deny (with Williamson) that the concept of ­knowledge is susceptible to a reductive analysis, while maintaining that what is expressed by that concept supervenes on non-epistemic properties, as witnessed by what Sosa (1991, 153–4) calls pessimism.43 Indeed, one can consistently deny (with Williamson) that knowledge states are metaphysically hybrid, subject to decomposition into purely subjective and objective components, while maintaining that such states supervene on non-epistemic properties (or states).44 So, neither the semantic unanalysability of ‘know’, nor the metaphysical indivisibility of the referent of ‘know’ provide any reason to reject ES. Indeed, to also deny ES is implausible as epistemic properties would be entirely autonomous, free-floating from the non-epistemic domain. For instance, beliefs are ordinarily regarded as justified only if some non-epistemic properties (i.e. justifiers) confer justification on them. Otherwise, justification would ­magically spring into existence and then inscrutably attach itself to beliefs. 42   Williamson offers surprisingly little by way of actual argument for the primacy of knowledge over belief vis-à-vis mental states. True, only knowledge is factive, and so only knowledge is a matching attitude between mind and world, but no substantial account is given of why such a matching relation can obtain only if knowledge is a sui generis, non-decomposable, mental state. 43   In philosophy of mind, for instance, many physicalists accept that mental concepts, and phenomenal concepts in particular, cannot be reductively analysed, yet they take all such concepts to express properties that supervene on non-mental properties. 44   Again, few physicalists regard phenomenal consciousness as a metaphysical hybrid, but that is perfectly consistent with supervening on physical properties.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

218  Jesper Kallestrup and Duncan Pritchard Moreover, Williamson ought to be sympathetic to ES. In Section 1, we offered two arguments in support of ES, the second of which hangs on a physicalist commitment to the supervenience of everything non-physical on the physical. Since all epistemic properties are non-physical by the lights of a theory-based conception of the physical, ES follows. As Williamson (2000, 52) accepts the physicalist thesis that ‘the total internal physical state of the subject and the total physical state of the external environment jointly determine the total state of the world’, ES is a claim he ought to endorse. The problem is that ES conflicts with putting knowledge first in an account of evidence and justification; or so we contend. Reflect finally that, even if Williamson were able to accommodate ES, a second problem would appear. Suppose that evidence and justification supervene on nonepistemic properties, because these epistemic properties supervene on knowledge, which in turn supervenes on non-epistemic properties. Remember that supervenience is transitive.45 As far as we are aware, Williamson has offered no positive account of which non-epistemic properties constitute the base on which knowledge supervenes. And the question is, of course, whether knowledge is still being put first if such an account would provide a supervenience base for evidence and justification in terms of wholly non-epistemic properties. Be that as it may. The non-epistemic properties on which knowledge can be supposed to supervene arguably include experiential properties. For example, Williamson (2000, 197) tentatively suggests that experiences can provide evidence by conferring the status of evidence on propositions of which all evidence consists. Now we need an explanatory account of why our epistemic properties (to do with evidence and justification) supervene on certain experiential or other nonepistemic properties—i.e., we need to know how exactly the epistemic superdupervenes on the non-epistemic. Of course, Williamson may insist that ES is a primitive fact of which no further explanation or elucidation is needed. But to regard ES as an inexplicably brute fact goes well beyond what the knowledge-first approach involves in terms of the concept of knowledge or the state of knowing not being subject to reductive analysis or metaphysical decomposition. Indeed, such an extraordinary claim sits badly with the widely recognized demand for superdupervenience in other branches of philosophy, including physicalist supervenience, which Williamson explicitly accepts. If, in general, the necessary co-variance between physical and non-physical properties is no fundamental fact about our world, then there is no reason to think that the epistemic domain is an exception. Surely, any such co-variance between distinct properties, epistemic or not, holds in virtue of some explanatory metaphysical relation such as grounding, realization, or what have you. We conclude that there is a significant explanatory lacuna inherent within Williamson’s knowledge-first approach to epistemology.46 45   As mentioned earlier, supervenience is also reflexive, and so knowledge trivially supervenes on knowledge. The key point is that ES demands that knowledge also supervenes on non-epistemic properties. 46   For a very helpful recent critical appraisal of knowledge-first epistemology—albeit one which explores very different ground to that set out here—see McGlynn (2014).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  219

5. Conclusion Our reflections on epistemic supervenience have shown that the case for epistemic anti-individualism, which undermines canonical versions of robust virtue epistemology, also extends to a prominent variety of epistemic internalism in the form of evidentialist mentalism. This further strengthens the hand of the epistemic anti-individualist. Moreover, although epistemic anti-individualism is, in principle, compatible with the influential knowledge-first approach advocated by Williamson, we have also seen that reflecting on the nature of epistemic supervenience has exposed a new challenge for this approach. In particular, Williamson has more explanatory work to do when it comes to defending knowledge-first epistemology than hitherto recognized. This last point thus offers prima facie support for an alternative approach to knowledge— favoured by the present authors—which embraces epistemic anti-individualism while eschewing the knowledge-first framework.47, 48

References BonJour, L. 1985. The Structure of Empirical Knowledge, Cambridge, MA: Harvard University Press. BonJour, L. 2002. Internalism and Externalism. In P. Moser, ed., Oxford Handbook of Epistemology. Oxford: Oxford University Press, 234–64. Carter, A. and J. Kallestrup. 2016. Extended Cognition and Propositional Memory. Philosophy and Phenomenological Research 92(3): 691–714. Chisholm, R. 1977. Theory of Knowledge, Second Edition. Englewood Cliffs, NJ: Prentice-Hall. Clark, A. and D. Chalmers. 1998. The Extended Mind. Analysis 58: 7–19. Cohen, S. 1984. Justification and Truth. Philosophical Studies 46: 279–96. Comesaña, J. 2005. We Are (Almost) All Externalists Now. Philosophical Perspectives 19: 59–76. Conee, E. 1992. The Truth Connection. Philosophy and Phenomenological Research 52: 657–69. Conee, E. 2007. Disjunctivism and Anti-Skepticism. Philosophical Issues 17: 16–36. Conee, E. and R. Feldman. 1985. Evidentialism. Philosophical Studies 48: 15–34. Conee, E. and R. Feldman. 2001. Internalism Defended. American Philosophical Quarterly 38: 1–18. Conee, E. and R. Feldman. 2004. Evidentialism. Oxford: Oxford University Press.

47   In particular, we have elsewhere argued for a conception of knowledge which includes both a virtuetheoretic and an anti-luck condition. See, for example, Pritchard, Millar, and Haddock (2010, Ch. 3), Pritchard (2012a), and Kallestrup and Pritchard (2011; 2013). 48   An earlier version of this chapter was presented (by JK) at the ‘Knowledge-First’ workshop held at the Eidyn research centre, University of Edinburgh, in November 2014. We are grateful to the audience on this occasion, including J. Adam Carter, Clayton Littlejohn, Heather Logue, Aidan McGlynn, and Martin Smith. Thanks also to J. Adam Carter, Emma Gordon, and Ben Jarvis for detailed comments on an earlier version of this chapter. This chapter was generously supported by the award of a grant from the John Templeton Foundation, as part of the Philosophy and Theology of Intellectual Humility Project at Saint Louis University.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

220  Jesper Kallestrup and Duncan Pritchard Conee, E. and R. Feldman. 2007. Externally Enhanced Internalism. In S. Goldberg, ed., Internalism and Externalism in Semantics and Epistemology. Oxford: Oxford University Press, 51–67. Conee, E. and R. Feldman. 2008. Evidence. In Q. Smith, ed., Epistemology: New Essays. Oxford: Oxford University Press, 83–104. Dougherty, T., ed. 2011. Evidentialism and its Discontents. Oxford: Oxford University Press. Feldman, R. 2003. Chisholm’s Internalism and Its Consequences. Metaphilosophy 34: 603–20. Feldman, R. 2005. Justification is Internal. In M. Steup and E. Sosa, eds., Contemporary Debates in Epistemology. Oxford: Blackwell, 270–84. Gibbons, J. 1996. Externalism and Knowledge of Content. Philosophical Review 105: 287–310. Goldberg, S. 2012. Epistemic Extendedness, Testimony, and the Epistemology of InstrumentBased Belief. Philosophical Explorations 15: 181–97. Goldman, A. 1976. Discrimination and Perceptual Knowledge. Journal of Philosophy 73: 771–91. Goldman, A. 2012. Reliabilism and Contemporary Epistemology: Essays. Oxford: Oxford University Press. Greco, J. 2009. Achieving Knowledge. Cambridge: Cambridge University Press. Hetherington, S. 2013. There Can be Lucky Knowledge. In M. Steup and J. Turri, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Blackwell, 164–76. Horgan, T. 1993. From Supervenience to Superdupervenience: Meeting the Demands of a Material World. Mind 102: 555–86. Huemer, M. 1999. A Defense of the Given. Philosophical Review 108: 128–30. Jarvis, B. 2013. Knowledge, Cognitive Achievement, and Environmental Luck. Pacific Philosophical Quarterly 94: 529–51. Kallestrup, J. 2011. Semantic Externalism. New York: Routledge. Kallestrup, J. 2015. Epistemic Anti-Individualism. Unpublished manuscript. Kallestrup, J. and D. H. Pritchard. 2011. Virtue Epistemology and Epistemic Twin Earth. European Journal of Philosophy 22(3): 335–57. Kallestrup, J. and D. H. Pritchard. 2012. Robust Virtue Epistemology and Epistemic AntiIndividualism. Pacific Philosophical Quarterly 93: 84–103. Kallestrup, J. and D. H. Pritchard. 2013. Robust Virtue Epistemology and Epistemic Dependence. In T. Henning and D. Schweikard, eds., Knowledge, Virtue and Action: Putting Epistemic Virtues to Work. London: Routledge, Ch. 11. Kallestrup, J. and D. H. Pritchard. 2016. Dispositional Robust Virtue Epistemology versus Anti-Luck Virtue Epistemology. In M. Fernandez Vargas, ed., Performance Epistemology. Oxford: Oxford University Press, 31–50. Korcz, K. A. 2015. The Epistemic Basing Relation. Stanford Encyclopaedia of Philosophy. Retrieved from http://plato.stanford.edu/entries/basing-epistemic/. Lehrer, K. 1997. Self-Trust: A Study of Reason, Knowledge, and Autonomy. Oxford: Oxford University Press. Lehrer, K., and S. Cohen. 1983. Justification, Truth, and Coherence. Synthese 55: 191–207. Littlejohn, C. 2012. Justification and the Truth-Connection. Cambridge: Cambridge University Press Lyons, J. 2013. Should Reliabilists Be Worried About Demon Worlds? Philosophy and Phenomenological Research 86: 1–40.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Anti-Individualism and Knowledge-First Epistemology  221 McDowell, J. 1995. Knowledge and the Internal. Philosophy and Phenomenological Research 55: 877–93. McGlynn, A. 2014. Knowledge First? London: Palgrave Macmillan. McLaughlin. 1995. Varieties of Supervenience. In E. E. Savellos and U. Yalcin, eds., Supervenience: New Essays. Cambridge: Cambridge University Press, 16–59. McLaughlin, A., and K. Bennett. 2011. Supervenience. Stanford Encyclopaedia of Philosophy. Retrieved from http://plato.stanford.edu/entries/supervenience/index.html Moon, A. 2013. Remembering Entails Knowing. Synthese 190: 2717–29. Neta, R., and D. H. Pritchard. 2007. McDowell and the New Evil Genius. Philosophy and Phenomenological Research 74: 381–96. Pappas, G. 2005. Internalist versus Externalist Conceptions of Epistemic Justification. Stanford Encyclopaedia of Philosophy. Retrieved from http://plato.stanford.edu/entries/justep-intext/. Plantinga, A. 1993. Warrant: The Current Debate, Oxford: Oxford University Press. Pritchard, D. H. 2002. Resurrecting the Moorean Response to the Sceptic. International Journal of Philosophical Studies 10: 283–307. Pritchard, D. H. 2005. Epistemic Luck. Oxford: Oxford University Press. Pritchard, D. H. 2007. Anti-Luck Epistemology. Synthese 158: 277–97. Pritchard, D. H. 2008. McDowellian Neo-Mooreanism. In A. Haddock and F. Macpherson, eds., Disjunctivism: Perception, Action, Knowledge. Oxford: Oxford University Press, 283–310. Pritchard, D. H. 2009a. Apt Performance and Epistemic Value. Philosophical Studies 143: 407–16. Pritchard, D. H. 2009b. Knowledge. Basingstoke, UK: Palgrave Macmillan. Pritchard, D. H. 2009c. Knowledge, Understanding and Epistemic Value. In A. O’Hear, ed., Epistemology (Royal Institute of Philosophy Lectures). Cambridge: Cambridge University Press, 19–43. Pritchard, D. H. 2011. Evidentialism, Internalism, Disjunctivism. In T. Dougherty, ed., Evidentialism and its Discontents. Oxford: Oxford University Press, 362–92. Pritchard, D. H. 2012a. Anti-Luck Virtue Epistemology, Journal of Philosophy 109: 247–79. Pritchard, D. H. 2012b. Epistemological Disjunctivism. Oxford: Oxford University Press. Pritchard, D. H. 2012c. In Defence of Modest Anti-Luck Epistemology. In T. Black and K. Becker, eds., The Sensitivity Principle in Epistemology. Cambridge: Cambridge University Press, 173–92. Pritchard, D. H. 2013. There Cannot be Lucky Knowledge. In M. Steup and J. Turri, eds., Contemporary Debates in Epistemology, Second Edition. Oxford: Blackwell, 152–64. Pritchard, D. H. 2015a. Anti-Luck Epistemology and the Gettier Problem. Philosophical Studies 172: 93–111. Pritchard, D. H. 2015b. Epistemic Angst: Radical Scepticism and the Groundlessness of Our Believing. Princeton, NJ: Princeton University Press. Pritchard, D. H., A. Millar, and A. Haddock. 2010. The Nature and Value of Knowledge: Three Investigations. Oxford: Oxford University Press. Pryor, J. 2001. Highlights of Recent Epistemology. British Journal for the Philosophy of Science 52: 95–124. Rose, D. and J. Schaffer. 2013. Knowledge Entails Dispositional Belief. Philosophical Studies 166: 19–50. Sainsbury, M. 1997. Easy Possibilities. Philosophy and Phenomenological Research 57: 907–19.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

222  Jesper Kallestrup and Duncan Pritchard Sosa, E. 1991. Knowledge in Perspective: Selected Essays in Epistemology. Cambridge: Cambridge University Press. Sosa, E. 1999. How to Defeat Opposition to Moore. Philosophical Perspectives 13: 141–53. Sosa, E. 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, Volume 1. Oxford: Clarendon Press. Sosa, E. 2009. Reflective Knowledge: Apt Belief and Reflective Knowledge, Volume 2. Oxford: Clarendon Press. Sosa, E. 2011. Knowing Full Well. Princeton: Princeton University Press. Sosa, E. 2013. Epistemic Agency. Journal of Philosophy 110: 585–605. Sosa, E. 2015. Judgement and Agency. Oxford: Oxford University Press. Steup, M. 1999. A Defense of Internalism. In L. Pojman, ed., The Theory of Knowledge: Classical and Contemporary Readings, Third Edition. Belmont, CA: Wadsworth, 310–21. Sudduth, M. 2015. Defeaters in Epistemology. Internet Encyclopaedia of Philosophy. Retrieved from http://www.iep.utm.edu/ep-defea/ Turri, J. 2010. Epistemic Supervenience. In J. Dancy, E. Sosa, and M. Steup, eds., Blackwell Companion to Epistemology, Second Edition. Oxford: Blackwell, 340–3. Turri, J. 2011. Manifest Failure: The Gettier Problem Solved. Philosophers’ Imprint 11: 1–19. Wedgwood, R. 2002. Internalism Explained. Philosophy and Phenomenological Research 65: 349–69. Williamson, T. 1995. Is Knowing a State of Mind?. Mind 104: 533–65. Williamson, T. 2000. Knowledge and its Limits. Oxford: Oxford University Press. Williamson, T. 2009. The Philosophy of Philosophy. Oxford: Blackwell. Zagzebski, L. 1996. Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical Foundations of Knowledge. Cambridge: Cambridge University Press. Zagzebski, L. 1999. What is Knowledge? In J. Greco and E. Sosa, eds., The Blackwell Guide to Epistemology. Oxford: Blackwell, 92–116.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

11 Knowledge-First Virtue Epistemology Christoph Kelp

1. Introduction It was once widely agreed that knowledge is justified true belief. Enter Gettier (1963), who adduced two cases that showed, to the satisfaction of nearly all researchers in the field, that one can have a justified true belief and yet fail to know. Since then, fixing the problem with the ‘JTB account’ (or Justified True Belief) that Gettier identified has been one of the main points on the epistemological agenda. Unfortunately, this task has proved to be rather difficult. While the literature features a whole host of proposals for such fixes, there is an equally impressive number of Gettier-style cases that show that these fixes meet with the same fate as the original JTB account. In fact, it is fair to say that fifty years of research have not produced an account of knowledge that is even widely agreed among researchers in the field. That said, there are a number of live proposals in the literature. One of the most popular ones is virtue epistemology, which aims to offer accounts of knowledge and/ or justified belief in terms of epistemic abilities.1 On the other hand, some researchers take the continued failure of attempts to fix the JTB account to suggest that the JTB account is irredeemably flawed and have scrapped this point from their agendas. Instead, many of them have adopted a ‘knowledge-first’ approach to epistemology, which aims to explain various core phenomena in epistemology in terms of knowledge.2 This chapter connects these two developments in epistemology. Its core aims are (i)  to  develop knowledge-first virtue epistemological accounts of knowledge and justified belief and (ii) to show that these accounts compare favourably with their traditionalist cousins.3 1   Champions of virtue epistemology include John Greco (2010), Alan Millar (2010), Duncan Pritchard (2012), Wayne Riggs (2002), Ernest Sosa (2015), John Turri (2011a), and Linda Zagzebski (1996). 2  Champions of knowledge first epistemology include Alexander Bird (2007a), Clayton Littlejohn (2013), Alan Millar (2010), Jonathan Sutton (2007), and, most famously, Timothy Williamson (2000). 3   See Kelp (2016c) and Kelp and Ghijsen (2016) for arguments that knowledge first virtue epistemology compares favourably with alternative knowledge first accounts of justified belief.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

224  Christoph Kelp In order to achieve this aim, I first introduce virtue epistemology (VE) in its traditional guise and sketch an argument that VE does not succeed in offering a fully satisfactory solution to the Gettier problem (Section 1). The main part of the chapter is  concerned with developing the alternative knowledge-first version of virtue ­epistemology. Section 2 introduces knowledge first epistemology and describes the general project. In Section 3 I give a general account of the normativity of a certain species of performances, which I then (Section  4) apply to beliefs as epistemic ­performances, thus arriving at my knowledge first virtue epistemological accounts of  knowledge and justified belief. Section  5 argues that these accounts can offer a more satisfactory solution to the Gettier problem than its traditionalist rivals.

2.  Virtue Epistemology 2.1  Virtue epistemology’s accounts of justified belief and knowledge The kind of virtue epistemology (VE) I am interested in has most prominently been championed by Ernest Sosa (2015) and John Greco (2010). It starts with a general account of performance normativity according to which performances with a goal can be assessed along the following three dimensions: success, competence, and aptness. The idea is that a performance is successful iff it reaches its goal, competent iff it is ­produced by the exercise of an ability to reach the goal, and apt iff it is successful because competent. It may be worth noting that this account of performance normativity leaves at least a number of key notions undefined. It does not tell us what it takes to possess an ability, what it takes to exercise it, and what it takes for a success to be because of the exercise of an ability. Different champions of VE have offered different accounts of these key notions,4 thus arriving at different versions of VE. While there are disagreements between champions of VE, there are also a number of points that are widely agreed on by virtue epistemologists. First, they all take belief to be a kind of performance with a goal.5 As a result, the normative framework for performances can be applied directly to belief and yields an account of the normativity 4   For instance, Sosa and Turri unpack the because relation in terms of the manifestation of an ability, whereas Greco and Pritchard opt for an explanatory salience alternative, and Riggs and Zagzebski seem attracted to a primitive account. Millar takes the notion of exercise of an ability to be a success notion, in the sense that one cannot exercise an ability to φ unless one φs. In contrast, most other virtue epistemologists allow for unsuccessful exercises of abilities. There are also differences in just how abilities depend on conditions. Millar and Greco take abilities to be very strongly dependent on conditions in the sense that in sufficiently unfavourable conditions the ability is unavailable to the agent altogether. In contrast, most other virtue epistemologists opt for a slightly weaker version according to which the abilities are relative to conditions but abilities remain available to agents even in fairly unfavourable conditions. 5   For a recent criticism of this idea see (Chrisman 2012); for responses see (Sosa 2015, Broncano-Berrocal 2015).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  225 of belief. Secondly, virtue epistemologists have tended to take belief as type of epistemic performance that has truth as its aim. This gives us: Successful Belief.  A belief is successful iff true. Competent Belief.  A belief is competent iff it is produced by an exercise of an ability to form true beliefs. Apt Belief.  A belief is apt iff it is true because competent. Next, champions of VE identify knowledge and justified belief with different normative standings of beliefs as performances: VE-JB.  One justifiably believes that p iff one competently believes that p. VE-K.  One knows that p iff one aptly believes that p. VE carries a significant amount of promise. Besides offering an appealing account of the normativity of belief as well as nicely motivated accounts of knowledge and justified belief, the view promises to solve a number of venerable epistemological problems, including the value problem6 and, most importantly for present purposes, the notorious Gettier problem.

2.2  The Gettier problem To get a grip on the Gettier problem, consider first a run-of-the-mill Gettier case: Stopped Clock.  Having come down the stairs, Mr. White looks at the grandfather clock in the hallway, sees that it reads 8:22 and on that basis comes to believe that it is  8:22. The clock has an outstanding track record of functioning properly and Mr. White has no reason to think that it is currently not accurate. His belief is true. It is in fact 8:22. Unbeknownst to Mr. White, however, the clock has stopped exactly twelve hours ago. Intuitively, Mr. White’s belief that it is 8:22 is both justified and true, but does not qualify as knowledge. Cases like this one thus serve to show that justified true belief is not sufficient for knowledge. They also raise the question as to what condition must be satisfied in addition to justified true belief in order to get knowledge. The Gettier problem is the problem of providing a satisfactory answer to this question. Here is an appealing diagnosis of why agents in Gettier cases fall short of knowledge: it is pure luck that they end up with a true belief (Pritchard 2005). More specifically, their beliefs are afflicted by a stroke of bad epistemic luck that puts them at great risk of acquiring a false belief. However, thanks to an additional stroke of good epistemic luck, the agent ends up with a true belief after all (Zagzebski 1994). In Stopped Clock, the 6   Very roughly, the value problem is the problem of explaining just how knowledge is more valuable than mere true belief or perhaps belief that falls short of knowledge. For more on the value problem see (Kvanvig  2003, Pritchard 2007a). For virtue epistemological approaches to the value problem see e.g., (Riggs 2002, Greco 2010, Sosa 2015).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

226  Christoph Kelp stroke of bad epistemic luck consists in the fact that the clock Mr. White takes a reading from is stopped. The stroke of good epistemic luck is that the clock stopped exactly twelve hours earlier, thus displaying the time correctly. This diagnosis of Gettier cases motivates a constraint on solutions to the Gettier problem. Any condition that, in conjunction with justified true belief, is said to be sufficient for knowledge must rule out the kind of luck at issue in Gettier cases. With these points in play, we can now see why VE may, at first glance, appear to be particularly well positioned to solve the Gettier problem. According to VE, justified true belief is competent and successful belief, whereas knowledge is apt belief. But now recall that, in addition to competence and success, apt belief requires that the belief to be successful because competent. The reason why VE carries so much promise to solve the Gettier problem is that there is independent reason to believe that apt belief, i.e., belief that is successful because competent, is incompatible with the kind of luck at issue in Gettier cases. How so? The answer is that there is independent reason to think that apt performance in general is incompatible with the kind of luck at issue in Gettier cases. To see this consider the following non-epistemic analogue of a Gettier case: Diverted Arrow.  Mr. Blonde, a highly competent archer, fires a competent shot at the target before him. On its way to the target the arrow is brought off its trajectory by a gust of wind. However, a moment later, a second gust of wind brings the arrow back on target with the result that it hits the bulls-eye. In this case, Mr. Blonde’s shot is both successful and competent. However, it is intuitively clear that it is not successful because competent, and thus not apt. Notice also just how appealing the parallel diagnosis is: the reason why Mr. Blonde’s shot is not apt is that it is a matter pure luck that it hits the target. Moreover, we also find the familiar combination of a stroke of good luck and a stroke of bad luck. In Diverted Arrow, the stroke of bad luck consists in the arrow’s being brought off target by a gust of wind. Mr. Blonde’s shot is running a serious risk of missing the target. However, this risk is prevented from materializing by a stroke of good luck, which here takes the form of a second gust of wind that brings the arrow back on target. It comes to light that there are non-epistemic analogues of Gettier cases in which the agents’ performances are successful and competent, but not apt. What is more, these non-epistemic analogues exhibit exactly the combination of two strokes of luck that are widely agreed to be the driving force in Gettier cases. Given that this is so, it may now seem that VE has what it takes to offer a successful solution to the Gettier problem. After all, champions of VE may now claim that Gettier cases are simply an instantiation of a more general phenomenon. Since it is independently plausible that such cases in general involve performances that are successful and competent, but not apt, there is reason to believe that the same will hold true for its epistemic instantiations, belief. And since VE identifies knowledge with apt belief, there is independent reason to believe that beliefs in Gettier cases fall short of knowledge.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  227

2.3  Fake barn cases Unfortunately, there is a fly in the ointment. There is a second kind of Gettier case, besides the one exemplified by Stopped Clock. The most famous example was first stated by Goldman in 1976:7 Fake Barns.  Mr. Pink is driving through the countryside and is currently looking out of the window of his car. He sees what appears to be a barn in the field and forms a perceptual belief that there is a barn in the field. Unbeknownst to Mr. Pink, he is looking at one of the few real barns in an area peppered with barn façades that are so cleverly constructed as to be indistinguishable from real barns from Mr. Pink’s position on the road. Just as in Stopped Clock before, the agent’s (here: Mr. Pink’s) belief is justified and true, but falls short of knowledge. What is more, the same diagnosis as in Stopped Clock is plausible here: it is a matter of pure luck that Mr. Pink arrives at a true belief. In particular, Mr. Pink suffers from a stroke of bad luck in that he is in a part of the world in which fake barns prevail. This stroke of bad luck is cancelled out by a stroke of good luck: he is looking at one of the few real barns in the area. The difficulty with Fake Barns is that VE cannot plausibly explain Mr. Pink’s lack of knowledge in the same way as in Stopped Clock. The reason for this is that there are non-epistemic analogues of Fake Barns in which the agents’ performances are very plausibly apt. Consider the following case by Pritchard (Pritchard 2008): Sabotaged Range.  Mr. Orange, a skilled archer, is taking a shot a target at his shooting range. Unbeknownst to him, nearly all targets at the range are sabotaged to prevent any shot fired at them. The target he is shooting at is the only non-sabotaged target at the range. It is intuitively clear that Mr. Orange’s shot is not only successful and competent, but also successful because competent, and hence apt. If so, champions of VE cannot treat cases like Fake Barns as just another instantiation of a general phenomenon. VE’s attractive account of standard Gettier cases does not extend to fake barn cases. In fact, things are worse than this. It seems that the vast majority of non-epistemic analogues of Fake Barns are cases in which the agent attains aptness. Besides Sabotaged Range, think of a chef who successfully prepares a tasty omelette, but happened to take the only salt shaker in which the salt was not replaced by sugar; or think of an artist who successfully produces a beautiful monochrome after having taken the only can in which the colour was not replaced by acid. In all of these cases, the agents’ performances are intuitively apt. If so, the pressure on champions of VE to acknowledge that the Mr. Pink’s belief in Fake Barns is apt, also. 7   It may be worth noting that Goldman attributes this example to Carl Ginet. While Ginet does describe the case in print, to the best of my knowledge he doesn’t do so before his 1988 paper.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

228  Christoph Kelp Fake barn cases constitute a significant obstacle for VE. Champions of VE must explain the absence of knowledge in such cases. As a first observation, note that most virtue epistemologists aim to offer a separate account for fake barn cases, i.e., one that diverges from the account they give for standard Gettier cases such as Stopped Clock. Here, I point out that already the fact that fake barn cases are given a separate treatment constitutes a potential cost for VE. It will do so if there is an alternative account that offers a unified explanation of both standard Gettier cases and fake barn cases. After all, such an account will have the theoretical advantage of being simpler.8 What is worse, there is reason to believe that extant virtue epistemological accounts of Fake Barns remain ultimately unsuccessful. To see this, note first that these accounts venture to deal with fake barn cases by introducing a modal condition into the account of knowledge in one way or another.9 The problem now is that these modal conditions are too strong to accommodate intuitions of presence of knowledge in other cases. My own favourite argument to this effect involves epistemic Frankfurt cases, which are inspired by Frankfurt’s famous 1969 cases.10 8   One might worry that this argument begs the question against VE. After all, champions of VE often argue that fake barn cases are structurally importantly different from standard Gettier cases. If so, by the lights of champions of VE, it may seem to make perfect sense to give them a separate treatment. I have two points to make by way of response. First, the question remains as to whether this structural difference cuts at the epistemological joints. If the reason these cases are taken to be different is simply that VE’s treatment of Gettier cases doesn’t extend to fake barn cases, the above complaint can hardly be said to beg the question against the view. (Compare: a champion of a ‘no false lemma’ (NFL) account claims that there are important structural differences between inferential and non-inferential standard Gettier cases and that, as a result, non-inferential Gettier cases afford separate treatment. If the reason for this claim is simply that the NFL account doesn’t extend to non-inferential Gettier cases, this will do little to defuse a complaint of the kind made above.) Second, and more importantly, I don’t take the above considerations to constitute an argument against VE, at least not by themselves. Rather, the thought is that if, in addition, there is an alternative view that can offer a uniform account and if all else is equal between the alternative account and VE, then there is some reason to think (i) that the alternative view is preferable and (ii) that the structural difference in question does not cut at the epistemological joints. (Compare: there is an alternative to the NFL account (perhaps a version of VE) such that it offers a uniform treatment of inferential and non-inferential Gettier cases and all else is equal between it and the NFL account. If so, (i) the alternative account is preferable to the NFL account and (ii) there is reason to think that the difference between inferential and non-inferential Gettier cases does not cut at the epistemological joints.) While I will develop just such an account in what follows, I will do so only in Section 4. As a result, the above argument is presently still awaiting completion. 9   For instance, Pritchard (2012) adds a safety condition on knowledge to deal with fake barn cases. Greco (2010) uses a modal condition on abilities (which in essence is a safety condition) to deal with fake barn cases. Sosa (2015) uses a distinction between animal and reflective knowledge and argues for a modal condition (again, in essence a safety condition) on reflective knowledge. 10   Note, however, that Frankfurt cases are not essential to the argument. Any case in which the agent knows but doesn’t satisfy the proposed modal conditions on knowledge will do. Since these conditions all amount to something like a safety condition (see fn. 9), cases of unsafe knowledge are good candidates for doing the work that I take Frankfurt cases to do. Crucially, the literature features a wide variety of such cases (Baumann  2008, Bogardus  2013, Comesaña  2005, Neta and Rohrbaugh  2004). As a result, the argument may be successful even if Frankfurt cases eventually fail to do the trick (e.g., because they don’t generate the required intuition of knowledge in readers). Finally, suppose that there is no one case that works against all of the proposed modal conditions. Even so, the argument may be successful provided that as there is at least one case that can be put to work against each of them.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  229 Here is one example. Frankfurt Clock.  Mr. Blue’s arch-nemesis, a powerful demon, has an interest that Mr. Blue forms a belief that it is 8:22 by looking at the grandfather clock in the hallway when he comes down the stairs. In order to achieve this, Mr. Blue’s arch-nemesis is prepared to set the clock to 8:22 when Mr. Blue comes down the stairs. However, Mr. Blue’s arch-nemesis is also lazy. He will act only if Mr. Blue does not come down the stairs at 8:22 of his own accord. Suppose, as it so happens, Mr. Blue does come down the stairs at 8:22. Mr. Blue’s arch-nemesis remains inactive. Mr. Blue forms a belief that it is 8:22. It is 8:22. The grandfather clock is working reliably as always (Kelp 2009, Kelp 2016a). Intuitively, Mr. Blue knows that it is 8:22 in this case. After all, we may assume, he has the ability to read the clock and forms a belief via an exercise of this ability. Moreover, the clock is actually functioning properly and the reading is accurate. Recall that I pointed out that extant virtue epistemological accounts of knowledge introduce a modal condition to explain the absence of knowledge in fake barn cases. The problem they are now facing is that agents in epistemic Frankfurt cases, such as Mr. Blue, will not satisfy any modal condition on knowledge that serves to explain the absence of knowledge in fake barn cases. The reason for this is that fake barn cases and epistemic Frankfurt cases have exactly the same modal profile. Any modal condition on knowledge that successfully predicts absence of knowledge in fake barn cases will also predict absence of knowledge in Frankfurt cases. Correlatively, any modal condition on knowledge that allows for knowledge in Frankfurt cases will not be able to explain the absence of knowledge in fake barn cases. In any case, accounts of knowledge that venture to accommodate the intuition of absence of knowledge in fake barn cases by means of a modal condition are in trouble. That is why extant virtue epistemological accounts of fake barn cases remain unsatisfactory.11

3.  Knowledge-First Epistemology While, at first glance, VE appeared to offer promising accounts of justified belief and knowledge, it turns out that VE continues to struggle with the Gettier problem. In what follows, I develop a version of VE that avoids this difficulty. It differs from standard versions of the view in that it takes beliefs to be performances that aim not at truth, but at knowledge. Accordingly, it takes the epistemic abilities required for knowledge and justified belief to be abilities to know. It is easy to see that this version of VE will not yield a so-called ‘reductive analysis’ of knowledge. That is to say, it does not provide an analysis of knowledge in terms of a non-circular set of individually necessary and jointly sufficient conditions for knowledge. 11   Kelp (2016a) provides a more in-depth presentation of this argument. There I also argue in some detail the most prominent virtue theoretic accounts of knowledge in the literature, including Greco’s and Sosa’s, all succumb to this problem.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

230  Christoph Kelp After all, if the epistemic abilities required for knowledge are abilities to know, any account of knowledge in terms of epistemic abilities so understood will violate the non-circularity constraint. I am not the first to abandon the ambition of offering a reductive analysis of knowledge. Others have taken the persistent failure of proposals in recent literature as a reason to do so as well. The perhaps most prominent alternative to the ‘traditional analytical project’ is knowledge first epistemology. Its founder, Timothy Williamson, characterizes the view in the following passage: ‘Knowledge first’ is a slogan for epistemology that takes the distinction between knowledge and ignorance as the starting point from which to explain other cognitive matters [W1]. It reverses the direction dominant in much twentieth-century epistemology, which treated belief as explanatorily prior to knowledge, attempting to analyze knowledge as belief that meets further conditions, such as truth and justification [W2]. By contrast, a knowledge first epistemologist might treat believing something as treating it as if one knew it [W3].  (Williamson 2010: 208)

This characterization of knowledge first epistemology contains three claims, which I have labelled W1, W2, and W3, respectively. Williamson’s characterization already indicates just how ambitious his project is. In fact, it exceeds the boundaries of epistemology, at least as it is traditionally understood. By way of evidence, notice that W1 takes knowledge to be the starting point from which to explain other cognitive matters. Moreover, W3 anticipates one of Williamson’s famous theses, to wit, that knowledge is a mental state in its own right, and that belief is to be analysed in terms of knowledge. Since accounts of mental states are traditionally taken to fall within the domain of the philosophy of mind, and perhaps philosophical psychology, Williamson anticipates that he will go beyond the boundaries of epistemology as it is traditionally understood. Here I flag that I will not follow Williamson in his pursuit of this ambitious project. The reason for this is that I am interested in developing a viable virtue epistemology—one that is better than its traditionalist cousins. Since, for this purpose, the question as to whether knowledge is a state of mind in its own right is by-the-by, Williamson’s thesis concerning the philosophy of mind can safely be put to one side.12 At this stage, one might naturally ask the following question: in what sense exactly is my account going to be knowledge first epistemological? The answer is that I do adopt the core epistemological claim of Williamson’s characterization of knowledge first epistemology, that is, the following version of W2: 12   At the same time, knowledge first virtue epistemology is compatible with Williamson’s thesis that knowledge is a state of mind (henceforth also ‘KSM’). In fact, I recently discovered that Lisa Miracchi (2015) has developed a knowledge first virtue epistemology in which KSM takes centre stage. Like Miracchi’s, the account I develop below is compatible with KSM. However, my account neither entails KSM, nor does KSM play a key role in it. I take this to be a benefit of my account. After all, KSM remains a highly controversial thesis in the philosophy of mind. For that reason, it will be of some interest if it is possible to reap at least some of the epistemological benefits of knowledge first epistemology, without taking on the heavyweight commitments in the philosophy of mind that KSM brings with it. I’d also like to emphasize that I have developed my own version of knowledge first virtue epistemology independently of Miracchi’s.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  231 Knowledge-First Epistemology (KF).  Knowledge-first epistemology reverses the traditional direction of analysis in epistemology: rather than analysing knowledge in terms of justified belief, justified belief is analysed in terms of knowledge. In other words, rather than treating justified belief as explanatorily prior to knowledge, knowledge is taken to have explanatory priority over justified belief. I submit that KF retains enough of the spirit of Williamson’s project to qualify as a knowledge-first view. Since it reverses the direction of analysis, it is incompatible with the traditional analytical project. Since it takes knowledge to enjoy explanatory priority over justified belief, there is an important respect in which knowledge comes first. For that reason, a virtue epistemology that satisfies KF can plausibly be regarded as a knowledge-first virtue epistemology (KFVE). It is exactly this view that I will develop in more detail in the next sections.13

4.  Simple Goal-Directed Practices 4.1  A basic framework Let us begin with a framework for simple goal-directed practices (SGPs). For a practice to be goal-directed is for it to have a success condition, a condition under which the practice’s goal is attained. One very simple kind of goal-directed practice involves two types of particular, targets and moves, and a designated relation. The success condition of this kind of practice can be defined as obtaining iff a move stands in the designated relation to the target. In (a very simple version of) target archery, call it ‘ARCH’, the target is a disc with a set surface area, moves are shots taken from a set distance, and the designated relation is the hit relation. A success in ARCH is a shot that hits the target. Practitioners of SGPs are move-producers. They may attain success in a given SGP. They do so if and only if they produce a move that stands in the designated relation to the target. Practitioners of ARCH are shot-producers. A practitioner of ARCH attains a success in ARCH if and only if he produces a shot that hits the target.

4.2 Abilities Practitioners of SGPs may have the ability to attain success in a given SGP. You, the reigning world champion in ARCH, have the ability to hit the target, while I, a blind man, do not. Let us take a closer look at these abilities (henceforth ‘SGP abilities’).

13   Or, to be more precise, I will develop a knowledge first virtue epistemological account of basic first-order knowledge and justified belief. That is to say, I want to set aside the issues of second-order knowledge/ justified belief as well as of non-basic knowledge/justified belief, i.e., knowledge/justified belief which depends for its status as knowledge/justified belief on the epistemic status of other beliefs. My main reason for this is to keep the complexity of the chapter manageable.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

232  Christoph Kelp SGP abilities involve ways of move production. Your ability to hit the target involves a way of shooting. More than one way of move production may be available to any one agent. For instance, you may shoot with your right hand or with your left hand. Of course, not all ways of move production will qualify as SGP abilities. While you may have the ability to hit the target in ARCH when shooting with your right hand, you may not have this ability when shooting with your left. SGP abilities are relative to conditions. You may have the ability to hit the target in ARCH when shooting with your left when sufficiently concentrated, sober, not being shoved while releasing the arrow, when shooting in normal winds, when there are no jokesters destroying the target at the last second, and so on. The conditions relative to which one may have an SGP ability may differ across different ways of move production. While you may have the ability to hit the target in ARCH when shooting with your left hand, but only when you are completely sober, you may have this ability when shooting with your right, even after a couple of beers. SGP abilities involve dispositions to attain success. Dispositions are associated with (i) trigger (T) and (ii) manifestation (M) conditions. In the case of SGP abilities, the manifestations are, of course, attainments of success. In addition, the trigger conditions are uses of ways of move production. Furthermore, dispositions are (iii) relative to conditions (C). In the case of SGP abilities these conditions are the ones to which the relevant ability is relative. To possess an ability to attain success in a given SGP, S, relative to conditions C, a practitioner must have a way of move production, W, such that he is disposed to attain S’s success (M) when using W (T) in C (C). For instance, to have the ability to hit the target in ARCH relative to C, you must have a way of shooting such that you are disposed to hit the target when using it in C. Finally, (iv) dispositions are widely regarded to be associated with counterfactual conditionals of the form were T to obtain in C, the possessor of the disposition would exhibit M. This means that practitioners of SGPs who have an ability to attain success in a given SGP, S, relative to conditions C will be such that were they to use the way of move production at issue in this ability (T) in C (C), they would attain S’s success (M). Those who have the ability to hit the target in ARCH relative to C will be such that, were they to use the way of shooting at issue in this ability in C, they would hit the target.14 A way of move production that qualifies as an ability to attain success in one SGP may or may not so qualify for another SGP. Consider another SGP, ARCH'. ARCH' is 14   Properties (i), (ii), and (iv) above are among what Jennifer McKitrick (2003: 157) calls ‘markers of dispositionality’, that is, properties of dispositions that are widely agreed upon among contributors to the debate on dispositions. For what I take to be convincing motivations of property (iii) see e.g., Mumford (1998), Sosa (2015). Finally, it might be better to construe dispositions as being associated with a probability of manifestation conditional on triggering in suitable conditions, rather than with the counterfactual conditionals at issue in (iv) above. One advantage that this kind of account of dispositions has for the application to abilities is that it can easily make sense of the fallibility of abilities, even in favourable conditions. After all, the relevant probability of manifestation may but need not be 1, even when the triggering conditions obtain in favourable conditions. See e.g., Healey (1991) and Suarez (2007) for more on probabilistic approaches to dispositions.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  233 just like ARCH, except that the target has a different shape than the target in ARCH. Instead of being disc-shaped, it is star-shaped. It may be that, in a certain set of conditions, C, a way of move production—shooting with your left hand, say—disposes you to hit the target in ARCH but not ARCH´. Your left-handed way of move production qualifies as an ability to hit the target in C relative to ARCH, but not ARCH´. You are not that good an archer when shooting with your left hand. At the same time, it may also happen that a(nother) way of move production qualifies as an ability to hit the target in C for both ARCH and ARCH´. It may be that, in C, shooting with your right hand disposes you to hit the target not only in ARCH, but also in ARCH´. Your right-handed way of move production qualifies as an ability to hit the target in C relative to ARCH and ARCH´. You are that good an archer when shooting with your right hand. For that reason a way of move production may also qualify as an ability to attain success across a range of SGPs. A way of move production qualifies as an ability to attain success across a given range, R, of SGPs only if it disposes one to attain success when used for all SGPs in R. Your right-handed way of move production may constitute an ability to hit the target across ARCH, ARCH´, ARCH″ (square-shaped), etc. Your left-handed way of move production may constitute an ability to hit the target across ARCH and ARCH″, but not ARCH´, etc. With these points in play, I would like to propose the following account of SGP abilities: SGP Ability.  One has an ability to attain success in a range, R, of SGPs and relative to conditions, C, if and only if one has a way of move production, W, such that, for any S ∈ R, using W in C disposes one to attain success in S.15

4.3  Exercises of abilities I suggest that exercises of SGP abilities are uses of ways of move production involved in SGP abilities. Or, more precisely, SGP Exercise.  One exercises an ability, A, to attain success for a range, R, of SGPs and relative to conditions, C, if and only if one has A and produces a move via the way of move production at issue in A. It is important to note that placing the agent in conditions relative to which he does not have an SGP ability can have different effects on an agent’s performances. Some such conditions will result in preventing an agent from using his way of move production. For instance, being too drunk, distracted, nervous, shoved while releasing the arrow, 15   See Kelp (forthcoming) for a more detailed account of SGP Abilities (as well as SGP Exercise and Competent SGP Move below). Note also that the full account of SGP abilities additionally features what I call a groundedness condition on abilities, which is a generalization of Millikan’s (2000: ch.4) idea that genuine abilities are distinguished from mere behavioural dispositions by being relative to conditions in which they were acquired via learning or natural selection. That said, the grounding condition does no substantive theoretical work in this chapter. Accordingly, for present purposes at least, it can safely be set aside.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

234  Christoph Kelp and so on, will prevent you from using the way of shooting that qualifies as an SGP ability relative to some (albeit different) conditions. I henceforth refer to conditions that, when not satisfied, prevent the agent from using his way of move production as conditions of shape (SH). According to SGP Exercise, then, exercising an SGP ability requires that SH be satisfied. In contrast, other such conditions do not prevent the agent from using his way of move production when not satisfied, and so allow him to exercise his ability anyway. Suppose, for instance, that you fire a shot that would have hit the target had it not been for a jokester who destroys the target when the arrow is about to hit it. Even though your shot misses the target, you do get to produce a move via the way of shooting that qualifies as an SGP ability relative to some (albeit different) conditions. I henceforth refer to conditions that, when not satisfied, do not prevent the agent from using his way of move production as situational conditions (SI). According to SGP Exercise exercising an SGP ability does not require that SI be satisfied.

4.4  Competent moves Competent moves in an SGP require the exercise of an SGP ability. When producing a shot in ARCH, your shot will be competent only if it is produced by an ability to hit the target. However, a competent move requires more than the exercise of an SGP ability. To see this, let us return to the case in which you are the reigning world champion of ARCH. Suppose that you are currently engaging in ARCHX in which the target changes its position discontinuously, randomly, and rapidly. Let us assume, as is plausible anyway, that you do not have the ability to hit the target in ARCHX. You have no way of shooting that disposes you to produce successful moves in ARCHX, no matter what conditions we may place you in. Suppose you take a shot using a way of move production that disposes you to hit the target in a range R of SGPs and relative to conditions C. Here you exercise your SGP ability to hit targets in range R and relative to C. However, that does not make your shot competent. The ability you exercise is the wrong ability for the SGP you are engaging in. (An even clearer example may be the following: suppose you have a way W of producing layups in basketball that qualifies as an ability to score relative to some C. Currently you are standing at the three-point line and have two seconds to score a basket to win the game. Suppose you produce a shot via W, which, of course, does not even get close to the basket. On the present view, you exercise an ability to score. However, your shot is not competent. Again, the reason for this is that you are exercising the wrong ability to score here.) For a move to be competent, then, it must be a move in an SGP, S, that is within the range R for which your way of move production qualifies as an ability. Contrast the situation described here with one in which you engage in an SGP that, we may assume, is within the range, R, of your SGP ability, but in which a jokester prevents the shot from being successful. Here, you not only exercise an ability to hit the target, but also your shot is competent.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  235 There is thus reason to believe that, in order to produce a competent move in a given SGP, the SGP must be within the range (of SGPs) of the SGP ability exercised. The considerations thus motivate the following account of competent moves: Competent SGP Moves.  A move in a given SGP, S, is competent iff it is produced by an exercise of an SGP ability to attain success in a range, R, of SGPs and relative to conditions, C, such that S ∈ R. This gives us a detailed account of what competence amounts to in SGP moves. Let us now look at aptness.

4.5 Aptness A SGP move is apt iff it is successful because competent. We already have precise accounts of success and competence. What is still needed is an account of the because condition on aptness. I will follow Sosa (Sosa 2015) in adopting a competence manifestation account of this condition. That is to say, an SGP move is apt iff its success manifests its competence. The success of an SGP move manifests its competence iff the competent move is successful in the SI of the ability exercised. For instance, a shot you take in ARCH is apt iff it is competent, i.e., produced by the exercise of an ability to hit the target such that ARCH is in the range of this ability, successful, i.e., the shot hits the target, and the ability’s SI are satisfied. If the shot is not competent, unsuccessful, or the SI are not in place, it will fall short of aptness. I would thus like to suggest the following account of aptness: SGP Aptness.  A move in a given SGP is apt iff it is (i) successful, (ii) competent, and (iii) the SI of the ability exercised are satisfied. Before moving on, I make one last point, which is of some importance in what follows. Note that aptness, competence, and success can come apart in various ways. In particular, a successful move in an SGP need not be competent. And moves that are both successful and competent can still fall short of aptness. Both of these points are nicely illustrated by the case of ARCH. Consider a case in which you take a shot in ARCH. Suppose you are completely drunk when taking the shot. As a result, your shot is not competent. But now suppose that, by an incredible stroke of luck, your shot hits the target anyway. Your shot is successful without being competent. Consider, next, a case in which, unbeknownst to you, there is a magnet at the shooting range that brings any shot fired at it off target. You take a competent shot at that shooting range which is brought off target by the magnet. Fortunately for you, a helper with a wind machine brings it back on target with the result that the shot is successful after all. Your move is both successful and competent, and yet falls short of aptness. While it is true that there are SGPs such that aptness and success come apart in this way, things are different for other SGPs. Consider the practice of joining consenting adults in lawful marriage (henceforth LM). LM can be understood as an SGP. Targets

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

236  Christoph Kelp here are couples of consenting adults, moves are tokens of ceremonies of a certain type, call it c, and the designated relation is effecting the lawful marriage of the couple. A success in LM is a token of c that effects the lawful marriage of the target couple. (The definitions of LM ability, exercise thereof, competent move, and apt move are as expected.) Crucially, there is reason to believe that one joins partners in lawful marriage only if one competently conducts a token of c. To see this, suppose you, the master of ceremony, fail to competently conduct a token of c. This may be for two reasons. You may fail to conduct a token of c altogether, say, because you are too drunk and so do not manage to say enough of the relevant words. Or else the SGP may not be in the range of c, say, because one of the partners is under age and so a different kind of ceremony is needed. In either case, the couple will not end up lawfully married. Since joining the couple in lawful marriage is the success condition of LM, this means that the success of LM will not be attained. If so, there is reason to believe that a move in LM will be successful only if competent. What is more, there is also reason to believe that one joins partners in lawful marriage only if the SI of the ability exercised are satisfied. To see this, suppose that the situational conditions are not satisfied. For instance, one key situational condition for the ability to join partners in lawful marriage is that any caveat lodged against the marriage has been discharged. So suppose that this SI is not in place. There remains an undischarged caveat against the marriage. If so, again, the couple will not end up lawfully married, even if you have competently conducted a token of c. Since joining the couple in lawful marriage is the success condition of LM, this means that the success of LM will not be attained. As a result, there is also reason to believe that a move in LM will be successful only if the SI of the ability exercised are satisfied. These considerations motivate the following condition on success in LM: LM Success.  A move in LM is successful only if it is competent and the SI of the ability exercised are satisfied.16 LM Success is an informative condition on success in LM. After all, it does not hold for successes in other SGPs such as ARCH and so captures relevant information about success in LM. At the same time, LM Success is unfit to constitute part of a reductive analysis of success in LM/joining consenting couples in lawful marriage. After all, it features the notion of competence, which, in turn, is unpacked in terms of an ability to attain success in LM/join consenting couples in lawful marriage. Now, let Δ be any SGP such that a move in Δ is successful only if it is competent and the SI of the ability exercised are satisfied. Since, trivially, a move in Δ will be successful only if it is successful, it follows that a move in Δ is successful only if successful, 16   Other plausible candidates for types of performances such that success entails a competence-in-SI condition are certain types of intentional action, such as winking, waving, applauding, and reading, as well as Zagzebski’s (1996) acts of virtue.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  237 competent, and the SI of the ability exercised are satisfied. However, by the relevant instance of SGP Aptness, we get the result that a move in Δ is successful iff apt. Since, by LM Success, LM is an SGP that features the relevant condition on success, we get: LM Aptness.  A move in LM is successful iff it is apt. It comes to light, then, that while there are SGPs in which aptness and success come apart in the sense that one can attain success while falling short of aptness, in others, aptness not only entails success, but success also entails aptness.

5.  Knowledge and Justified Belief I now apply this account of SGPs, SGP abilities, exercises thereof, and competent SGP moves to the epistemic case. To begin with, I suggest that a relevant fragment of epistemic activity—viz. inquiry into specific whether questions (henceforth simply ‘inquiry’)—can be understood as an SGP. Or, to be more precise, it can be understood as a collection of SGPs, one for each question. More specifically, my suggestion is that the targets of inquiry are true answers. For instance, the target of an inquiry into whether p is the true member of the set including the proposition that p and the proposition that not-p. Moves in inquiry are beliefs.17 For instance, believing p constitutes a move in an inquiry into whether p, as does believing not-p. The designated relation in inquiry is the knowledge relation.18, 19 A success in inquiry, then, is a belief that qualifies as knowledge (henceforth also ‘knowledgeable belief ’ for short). In other words, Successful Belief.  A belief is successful iff it qualifies as knowledge. What is more, we can apply the accounts of SGP abilities, their exercises, competent belief, and apt belief from the previous section to the epistemic case. This gives us first the following accounts of abilities to know and their exercises: Ability to Know.  One has an ability to know propositions in a range, R, and relative to conditions, C, iff one has a way of belief formation, W, such that, for any p ∈ R, using W in C disposes one to form knowledgeable beliefs that p. 17   A complete account would also countenance suspension of judgement as a type of move. Notice, however, that suspension of judgement is a second-order attitude. Given my aim of providing an account of first-order knowledge and justified belief (fn. 13), the issue of suspension of judgement can safely be set aside. 18   Given that in inquiry targets are true answers and moves are beliefs, doesn’t this commit me to saying that true beliefs constitute successes in inquiry? No. To see this, imagine a variation of eight-ball in which pockets must be designated and balls must be potted intentionally—flukes that find the designated pocket via a set of cushions do not count. Here, there is a clear sense in which the pocket is the target, but not any move that finds the pocket qualifies as a success. 19   I argue in some detail that knowledge rather than weaker epistemic standings (such as justified and/ or true belief) is the goal of inquiry into particular whether questions in Kelp (2014b). For an argument that understanding why is not the goal of inquiry into such questions see Kelp (2014a). Compatibly with this, understanding may be the goal of inquiry into various phenomena and subject matters. The reason for this is that understanding is a form of knowledge (see Kelp 2015 for an argument).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

238  Christoph Kelp Exercises of Abilities to Know.  One exercises an ability, A, to know propositions in range R and relative to conditions C iff one has A and forms a belief via the way of belief formation at issue in A. Competent Belief.  One competently believes that p iff one’s belief that p is formed by an exercise of an ability to know propositions in range R and relative to conditions C such that p∈R. Apt Belief.  A belief is apt iff it is (i) successful, (ii) competent, and (iii) the SI of the ability to know exercised are satisfied. Recall that traditional virtue epistemology identifies justified belief and knowledge with different normative standings of beliefs as performances. In particular, they identify justified belief with competent belief and knowledge with apt belief. Here, I follow traditional virtue epistemologists on this front. That is to say, I offer the following accounts of justified belief and knowledge: KFVE-JB.  One justifiably believes that p iff one competently believes that p. KFVE-K.   One knows that p iff one aptly believes that p. Let me say a few words to explain these accounts of justified belief and knowledge. They are superficially indistinguishable from their traditionalist rivals. At the same time, once they are properly unpacked, there are substantial differences between the two. Starting with KFVE-JB, notice that, on my account, competent belief is analysed in terms of abilities to know. In true knowledge first style, then, KFVE-JB reverses the traditional direction of analysis and analyses justified belief in terms of knowledge, where knowledge has explanatory priority over justified belief. What is more, while the accounts of abilities to know and their exercises do not figure directly in KFVE-JB they are of crucial importance to it. After all, they contribute to making the core notions in terms of which justified belief is defined more precise. Let us move on to KFVE-K, then. First, it is of crucial importance to keep in mind that, unlike its traditionalist cousin, KFVE-K does not (and actually could not) offer a reductive analysis of knowledge. The easiest way to appreciate this point is by taking note of the fact that apt belief requires successful belief and successful belief is knowledgeable belief. According to KFVE-K, then, one key condition a belief must satisfy in order to qualify as knowledge is that it qualifies as knowledge. KFVE-K is blatantly circular and thus unfit for the purposes of reductive analysis. Second, notice that it follows from Successful Belief and KFVE-K that a belief is successful iff it is apt. After all, by Successful Belief, a belief is successful iff it qualifies as knowledge and, by KFVE-K, a belief qualifies as knowledge iff it is apt. It follows that a belief is successful iff it is apt. As shown in the previous section, in and of itself, this is no cause for concern. There are other cases of SGPs in which a move in the relevant SGP is successful iff apt. The practice of joining consenting adults in lawful marriage (LM) is a case in point.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  239 What is more, on reflection, it is independently plausible that in the case of inquiry, moves are successful iff apt. In fact, an argument parallel to the one that made this point for LM will do the trick here too. Here goes. To begin with, there is reason to believe that knowledge requires competent belief. To see this, suppose you, an epistemic agent, form a belief that you are facing a barn, say, that falls short of being competent. This may be for two reasons. You may fail to exercise an ability to know, say, because you form your belief on the basis of a coin toss. Or else the proposition may not be in the range of your ability, say, because it has been produced by an ability to recognize colours. In either case, you end up not knowing that you are facing a barn. Since knowledge is success in inquiry, this means that success in inquiry will not be attained. If so, there is reason to believe that a move in inquiry will be successful only if competent. What is more, there is also reason to believe that knowledge requires that the SI of the ability exercised are satisfied. To see this, suppose that the situational conditions are not satisfied: you are in fake barn county, are currently looking at a fake barn, etc. If so, again, you end up not knowing that you are facing a barn. Since knowledge is the success condition of inquiry, this means that success in inquiry will not be attained. As a result, there is also reason to believe that a move in inquiry will be successful only if the SI of the ability exercised are satisfied. Just as in the case of LM before, these considerations motivate the following condition on success in inquiry: Inquiry Success.  One’s move in inquiry is successful only if it is competent and the SI of ability exercised are satisfied. But now recall that we saw in the last section that, for any SGP such that this condition holds, it follows that a move in that SGP is successful iff apt. Since we have seen that there is independent reason to think that Inquiry Success is true, this means that it is independently plausible that a belief is successful iff it is apt. Finally, it may be worth noting that, just as in the case of LM Success, Inquiry Success is an informative condition on success in inquiry and yet is not suited to constitute part of a reductive analysis of success in inquiry/knowledge. It is informative because the parallel condition does not hold for successes in other SGPs such as ARCH, and so captures relevant information about success in inquiry. It is unfit to constitute part of a reductive analysis of success in inquiry/knowledge because it features the notion of competence, which, in turn, is unpacked in terms of an ability to attain success in inquiry/knowledge.

6.  The Gettier Problem With KFVE in play, I now take a closer look at how the view deals with the Gettier problem. One might think that, once the traditional analytical project is abandoned in

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

240  Christoph Kelp favour of a knowledge-first approach to epistemology, the Gettier problem does not arise in the first place. While this is true in a sense, there is a variation of the problem that arises for knowledge first epistemology. To see this, let us have another look at our two Gettier cases: Stopped Clock.  Having come down the stairs, Mr. White looks at the grandfather clock in the hallway, sees that it reads 8:22 and on that basis comes to believe that it is 8:22. The clock has an outstanding track record of functioning properly and Mr. White has no reason to think that it is currently not accurate. His belief is true. It is in fact 8:22. Unbeknownst to Mr. White, however, the clock has stopped exactly twelve hours ago. Fake Barns.  Mr. Pink is driving through the countryside and is looking out of the window of his car. He sees what appears to be a barn in the field and forms a perceptual belief that there is a barn in the field. Unbeknownst to Mr. Pink, he is looking at one of the few real barns in an area peppered with barn façades that are so cleverly constructed as to be indistinguishable from real barns from Mr. Pink’s position on the road. Recall that, intuitively, Mr. White’s and Mr. Pink’s beliefs do not qualify as knowledge. Crucially, it is intuitively equally plausible that their beliefs are justified. Knowledgefirst epistemological accounts of justified belief will have to accommodate the intuition that beliefs of agents in Gettier cases are justified. While the original version of the Gettier problem does not arise within knowledge-first epistemology, a variation of it does. What is more, even though knowledge-first epistemological accounts do not offer a reductive analysis of knowledge, they may offer informative necessary conditions on knowledge. If so, the question remains whether the necessary conditions countenanced are satisfied in Gettier cases. If not, knowledge-first accounts may even be able to predict the absence of knowledge here. In what follows, I look at how KFVE deals with each kind of Gettier case. In particular, I argue that KFVE successfully predicts both the presence of justification and the absence of knowledge in both kinds of case. I start once more with standard Gettier cases.

6.1  Standard Gettier cases First, in Stopped Clock, Mr. White’s belief that it is 8:22 is intuitively justified. Can KFVE accommodate this intuition? To answer this question, notice first that Mr. White acquires his belief that it is 8:22 in a way that qualifies as an ability to know propositions about the time. After all, forming his belief via this way disposes Mr. White to acquire knowledgeable beliefs about the time across a range of conditions. For instance, it does so in cases just like Stopped Clock with the exception that the clock he is taking a reading from is functioning properly. Given that Mr. White acquires his belief that it is 8:22 in a way that qualifies as an ability to know propositions about the time, his belief

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  241 is the product of an exercise of an ability to know propositions about the time. Since the belief Mr. White forms is a belief about the time, the target proposition is within the range of the ability to know that produced his belief. By Competent Belief, Mr. White’s belief that it is 8:22 is competent and, by KFVE-JB, justified.20 Second, Mr. White’s belief intuitively falls short of knowledge. Recall that KFVE does provide an informative virtue condition on knowledge. As a result, we may ask whether this condition is satisfied here. If it is not, then KFVE has the additional benefit of being able to explain the intuition of absence of knowledge, besides the intuition of presence of justified belief in this kind of case. Let us return to Stopped Clock once more, then. Notice that the clock Mr. White is taking a reading from is stopped. As a result, he is not in conditions C relative to which he has the ability to acquire knowledge about the time. After all, Mr. White’s way of belief formation does not dispose him to form knowledgeable beliefs about the time in conditions in which the clock he is taking a reading from is stopped. At the same time, the conditions Mr. White finds himself in do not prevent him from forming his belief via the way that constitutes an ability to know relative to some (but different) conditions. This means that Mr. White is in unsuitable SI for his ability to acquire knowledge about the time. By Apt Belief, his belief that it is 8:22 is not apt and, by KFVE-K, it falls short of knowledge. In this way, KFVE predicts not only presence of justified belief, but also the absence of knowledge in Stopped Clock. There is thus reason to believe that KFVE can successfully handle standard Gettier cases.

6.2  Fake barn cases Arguably, standard Gettier cases have never posed much of a difficulty for VE. The hard nut to crack are fake barn cases. Let us return to Fake Barns then. Here, too, Mr. Pink’s belief that he is facing a barn is intuitively justified, but falls short of knowledge. Fortunately, it is easy enough to see that KFVE can explain the presence of justified belief and the absence of knowledge here in much the same way as in standard Gettier cases. First, concerning the presence of justified belief, the way in which Mr. Pink forms his belief qualifies as an ability to recognize barns, for the same reasons as Mr. White’s way of belief formation in Stopped Clock qualifies as an ability to know propositions about the time. If so, Mr. Pink’s belief that he is facing a barn is the product of an exercise of an ability to recognize barns. Since the belief Mr. Pink forms is a belief about the presence of a barn, the target proposition is within the range of the ability to know that 20   It may be worth noting that, as a result, KFVE can also allow for justified false beliefs. If this isn’t immediately obvious, consider a variation of Stopped Clock in which Mr. White acquires his belief about the time a minute earlier or later. In that case, his belief that it is 8:22 will be false. At the same time, according to KFVE-JB it will still be justified. In fact, as I argue elsewhere (Kelp forthcoming) in more detail the virtue epistemology at issue in KFVE can solve or defuse a number of long-standing problems for broadly reliabilist accounts of justified belief, including the new evil demon problem, the problem of clairvoyance cases and the generality problem.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

242  Christoph Kelp produced his belief. By Competent Belief, Mr. Pink’s belief that he is facing a barn is competent and, by KFVE-JB, justified. Second, concerning the absence of knowledge, the fact that Mr. Pink is in a part of the country in which fake barns predominate means that he is not in conditions C relative to which he has the ability to recognize barns. At the same time, the conditions Mr. Pink finds himself in do not prevent him from forming his belief via the way that constitutes an ability to know relative to some (but different) conditions. This means that Mr. Pink is in unsuitable SI for his ability to recognize barns. By Apt Belief, Mr. Pink’s belief that he is facing a barn is not apt. By KFVE-K, it falls short of knowledge. Of course, the real problem that fake barn cases pose for VE is not just to accommodate the intuition of absence of knowledge in fake barn cases. Traditional versions of VE can achieve this much. Rather, the challenge is to offer a satisfactory account of fake barn cases, without, at the same time, ending up making incorrect predictions in Frankfurt cases. Let us see whether KFVE can rise to this challenge as well. Recall our toy Frankfurt case: Frankfurt Clock.  Mr. Blue’s arch-nemesis, a powerful demon, has an interest that Mr. Blue forms a belief that it is 8:22 by looking at the grandfather clock in the hallway when he comes down the stairs. In order to achieve this, Mr. Blue’s archnemesis is prepared to set the clock to 8:22 when Mr. Blue comes down the stairs. However, Mr. Blue’s arch-nemesis is also lazy. He will act only if Mr. Blue does not come down the stairs at 8:22 of his own accord. Suppose, as it so happens, Mr. Blue does come down the stairs at 8:22. Mr. Blue’s arch-nemesis remains inactive. Mr. Blue forms a belief that it is 8:22. It is 8:22. The grandfather clock is working ­reliably as always. Intuitively, Mr. Blue knows that it is 8:22. What does KFVE predict about this case? And does KFVE’s treatment of Fake Barns entail that Mr. Blue does not know that it is 8:22? Let us start with the second question. Fortunately, the answer here is no. In a nutshell, the reason for this is that KFVE does not offer a reductive analysis of knowledge and hence is free to treat the difference between Fake Barns and Frankfurt Clock as a fundamentally epistemic difference, that is, roughly, a difference in knowledge. More specifically, while both Mr. Pink and Mr. Blue form competent beliefs, only Mr. Blue is in suitable SI for the ability to know he exercises and so satisfies KFVE’s necessary condition on knowledge. Why is that? The answer is that only Mr. Blue is in SI such that his way of belief formation disposes him to form the relevant knowledgeable beliefs. In this way, KFVE can make a difference between Mr. Pink and Mr. Blue, albeit one that is fundamentally epistemic. As a result, KFVE’s treatment of Fake Barns does not lead to a corresponding incorrect prediction in Frankfurt Clock.21 21   Note that while KFVE can allow that Mr. Blue knows while Mr. Pink doesn’t, it is not committed to these verdicts. For instance, KFVE can also accommodate the intuitions of those who think that Mr. Pink knows, viz. by allowing that the SI in Fake Barns are suitable (as Mr. Pink’s way of belief formation does dispose him to form the relevant knowledgeable beliefs in those SI). Similarly, KFVE can accommodate the

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  243 This leaves the question as to what KFVE predicts about Frankfurt Clock: presence or absence of knowledge? The answer to this question is neither. After all, KFVE does not offer a sufficient condition for knowledge. What is more, it is in the spirit of knowledge-first epistemology to consider knowledge as basic in the sense that we need not seek to offer an account of knowledge that explains the presence of knowledge in particular cases. But then we need not expect KFVE to make predictions about the presence of knowledge in Frankfurt Clock to begin with. Be that as it may, the important point for present purposes is that KFVE’s treatment of fake barn cases does not lead the view into trouble with Frankfurt cases.

7. Conclusion This chapter has developed a novel virtue epistemological account of knowledge and justified belief, one that gives VE a knowledge-first twist. In order to achieve this, I first outlined a framework for simple goal-directed practices, i.e., practices that can be characterized in terms of targets, moves, and a designated relation, and defined the notions of ability, exercise of ability, and competent and apt moves for such practices. I  then suggested that inquiry into whether questions can be viewed as a collection of simple goal-direct practices in which targets are true answers, moves are beliefs and the designated relation is the knowledge relation. I derived the corresponding accounts of abilities to know, their exercises, competent and apt belief. With these accounts in play, I went on to embrace the familiar virtue epistemological accounts of knowledge as apt belief and justified belief as competent belief. The crucial difference between my account and its traditionalist cousins is that abilities are unpacked as abilities to know. In this way, the account qualifies as a knowledge first version of virtue epistemology. Finally, I showed that KFVE has an edge over its traditionalist cousins. Not only can it handle fake barn cases without further difficulty, it can also offer a uniform account of all Gettier cases, including the fake barn and the standard variety. In view of all this, I submit, knowledge-first virtue epistemology is a promising view that deserves to be taken seriously.22 intuitions of those who think that Mr. Blue doesn’t know, viz. by allowing that the SI in Frankfurt Clock are unsuitable (as Mr. Blue’s way of belief formation does not dispose him to form the relevant knowledgeable beliefs in those SI). Of course, if either one of these intuitions turns out correct, my argument from Frankfurt cases will not serve to establish an advantage for KFVE over traditional VE after all. For what it’s worth, however, my own intuitions are that Mr. Blue knows, whereas Mr. Pink doesn’t, which is why I think the argument does go through. 22   Acknowledgements. Thanks to the audiences of the following conferences for helpful feedback on this paper: 2012 EEN Meeting, Universities of Bologna and Modena (2012); CCPEA, Academia Sinica, Taipei (2012); 2013 Bled Philosophical Conferences, Bled (2013); Epistemic Justification and Reasons, University of Luxembourg (2013); Neue Perspektiven der Epistemischen Rechtfertigung, University of Dresden (2013). Thanks also to the audiences of the following conferences for helpful feedback on other papers on knowledge first virtue epistemology, from which this chapter has benefited: Saving Safety, Bonn (2013), Yonsei Philosophy Summer Conference, Yonsei University (2014); Normative Epistemic Reasons, University of Luxembourg (2014), The Virtue Turn, University of Taipei (2014); 2015 Bled Philosophical

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

244  Christoph Kelp

References Baumann, P. 2008. Is knowledge safe? American Philosophical Quarterly 45: 19–30. Bird, A. 2007. Justified judging. Philosophy and Phenomenological Research 74: 81–110. Bogardus, T. 2013. Knowledge under threat. Philosophy and Phenomenological Research 86: 289–313. Broncano-Berrocal, F. 2015. In defence of performance epistemology. Unpublished Manuscript. Chrisman, M. 2012. The normative evaluation of belief and the aspectual classification of belief and knowledge attributions. Journal of Philosophy 109: 588–612. Comesaña, J. 2005. Unsafe knowledge. Synthese 146: 395–404. Frankfurt, H. 1969. Alternate possibilities and moral responsibility. The Journal of Philosophy 66: 829–39. Gettier, E. 1963. Is justified true belief knowledge? Analysis 23: 121–3. Ginet, C. 1988. The fourth condition. In Austin, D., ed., Philosophical Analysis. A Defense by Example. Springer, Dordrecht, pp. 105–17. Goldman, A. 1976. Discrimination and perceptual knowledge. Journal of Philosophy 73: 771–91. Greco, J. 2010. Achieving Knowledge. Cambridge University Press, Cambridge. Healey, R. 1991. The Philosophy of Quantum Mechanics: An Interactive Interpretation. Cambridge University Press, Cambridge. Kelp, C. 2009. Knowledge and safety. Journal of Philosophical Research 34: 21–31. Kelp, C. 2014a. Knowledge, understanding and virtue. In Fairweather, A., ed., Virtue Epistemology Naturalized: Bridges between Virtue Epistemology and Philosophy of Science. Springer, Dordrecht, pp. 347–60. Kelp, C. 2014b. Two for the knowledge goal of inquiry. American Philosophical Quarterly 51: 227–32. Kelp, C. 2015. Understanding phenomena. Synthese 192: 3799–816. Kelp, C. 2016a. Epistemic Frankfurt cases revisited. American Philosophical Quarterly 53: 27–37. Kelp, C. 2016c. Justified belief: knowledge first-style. Philosophy and Phenomenological Research 93: 79–100. Kelp, C. Forthcoming. How to be a reliabilist. Philosophy and Phenomenological Research. Kelp, C. and H. Ghijsen. 2016. Perceptual justification: factive reasons and fallible virtues. In C. Mi, M. Slote, and E. Sosa, eds., Moral and Intellectual Virtues in Western and Chinese Philosophy. Routledge, London, 164–83. Kvanvig, J. 2003. The Value of Knowledge and the Pursuit of Understanding. Cambridge: Cambridge University Press. Littlejohn, C. 2013. The Russellian retreat. Proceedings of the Aristotelian Society 113: 293–320. McKitrick, J. 2003. A case for extrinsic dispositions. Australasian Journal of Philosophy 81: 155–74. Millar, A. 2010. Perceptual knowledge and recognitional abilities. In D. Pritchard, A. Millar, and A. Haddock, eds., The Nature and Value of Knowledge. Oxford: Oxford University Press, 120–43. Conferences, Bled (2015). Thanks also to Conor McHugh, Daniel Whiting, Chris Tucker, Adam Carter, Emma Gordon, Benjamin Jarvis, Jan Heylen, Fernando Broncano-Berrocal, Ernest Sosa, anonymous referees of various journals and OUP, as well as the Leuven Epistemology Group for detailed feedback on the paper. Special thanks to Harmen Ghijsen and Mona Simion for commenting on various versions of the chapter. This work was funded by grants from KU Leuven’s Special Research Fund (BOF) and Research Foundation Flanders (FWO).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Knowledge-First Virtue Epistemology  245 Millikan, R. 2000. On Clear and Confused Ideas. Cambridge: Cambridge University Press. Miracchi, L. 2015. Competence to know. Philosophical Studies 172: 29–56. Mumford, S. 1998. Dispositions. Oxford: Oxford University Press. Neta, R. and G. Rohrbaugh. 2004. Luminosity and the safety of knowledge. Pacific Philosophical Quarterly 85: 396–406. Pritchard, D. 2005. Epistemic Luck. Oxford: Oxford University Press. Pritchard, D. 2007. Recent work on epistemic value. American Philosophical Quarterly 44: 85–110. Pritchard, D. 2008. Greco on knowledge: Virtues, contexts, achievements. The Philosophical Quarterly 58: 437–47. Pritchard, D. 2012. Anti-luck virtue epistemology. The Journal of Philosophy 109: 247–79. Pritchard, D., A. Millar, and A. Haddock. 2010. The Nature and Value of Knowledge. Oxford: Oxford University Press. Riggs, W. 2002. Reliability and the value of knowledge. Philosophy and Phenomenological Research 64: 79–96. Sosa, E. 2015. Judgment and Agency. Oxford: Oxford University Press. Suarez, M. 2007. Quantum propensities. Studies in History and Philosophy of Modern Physics 38: 418–38. Sutton, J. 2007. Without Justification. Cambridge/MA: MIT Press. Turri, J. 2011. Manifest failure: the Gettier problem solved. Philosopher’s Imprint 11: 1–11. Williamson, T. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Williamson, T. 2010. Knowledge first epistemology. In S. Bernecker and D. Pritchard, eds., The Routledge Companion to Epistemology. New York: Routledge, 208–18. Zagzebski, L. 1994. The inescapability of Gettier problems. The Philosophical Quarterly 44: 65–73. Zagzebski, L. 1996. Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical Foundations of Knowledge. Cambridge: Cambridge University Press.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

12 In Support of the Knowledge-First Conception of the Normativity of Justification Anne Meylan

1.  Delineating the Question It is commonplace to claim that justification is a normative concept. In this article, “normative concept” refers to the general class that has deontic concepts (e.g., being obligatory, permitted, etc.), axiological concepts (e.g., being good, bad, elegant, etc.), and aretic concepts (being virtuous, vicious, wise, etc.) as its members.1 The normativity of justification implies that every time an entity is justified, it also possesses specific (non-normative) characteristics that make it justified. The same is true for all normative concepts. For instance, when a person’s behaviour is nice, this behaviour also has characteristics that make it nice. When someone is courageous, this person also possesses some courageous-making characteristics. To reiterate, then, a justified entity: 1. Has the characteristic of being justified; 2. Also has other characteristics that make it justified. For this reason, when we ask—as epistemologists often do—“what is it for x to be justified?” (where the x in question can be a belief or an action), we can understand the question in two ways depending on whether one takes this question to be directed at 1) the normative characteristic of being justified, or 2) the non-normative characteristics that make x a justified x. 1. When “what is it for x to be justified?” is understood as a question pertaining to the characteristic of being justified, it amounts to asking: “what is justification itself?”

1   See Mulligan (2009) for a similar use of the term “normative”. I say more on these distinct kinds of normative concepts later on.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

In Support of the Knowledge-First Conception  247 2. When “what is it for x to be justified?” is understood as a question pertaining to the characteristics that make x a justified x, it amounts to asking: “what are the characteristics in virtue of which this entity possesses what we have already identified as justification?” I will dub these two questions as the “primary” question of justification and the “grounding” question of justification, respectively. While the grounding question has been the focus of wide attention in epistemology (after all, reliabilism, the various forms of evidentialism, etc., are all answers to the grounding question), the primary question of justification has not received as much consideration (with some notable exceptions, as we will see).2 It is not the grounding question that interests us here. This contribution is not dedicated to understanding “the characteristics that make an x something justified”. Rather, it is concerned with— let me say somewhat loosely—justification “itself ”. More precisely, taking for granted that justification is a normative property, this chapter is concerned with the nature of this normativity. An analogy might help to understand what I mean by this. As we can try to find out to which kind of normativity characterizes the normative property of being beautiful (and answer that beautifulness is a value or axiological property), we can try to discover which kind of normativity characterizes the normative property of being justified. Now, the latter is the general issue that this contribution addresses. As we will see, the knowledge-first solution to the New Evil Demon Problem (NEDP) relies on a particular view of the normativity of justification that I shall simply call, “the knowledge-first conception”. What, to my knowledge, has not been emphasized yet is that the knowledge-first conception of the normativity of justification is incompatible with another, well-accepted conception of its normativity. The purpose of this chapter is to settle the debate between these two irreconcilable views about the normativity of justification. The upshot is that the knowledge-first conception—viz. the one on which the knowledge-first solution to the NEDP relies—seems superior.

2.  The Optional Conception of the Normativity of Justification Even if, as just said, the primary question of justification has not received as much ­attention as the grounding question, it has not been completely neglected by epistemologists.3 William Alston is certainly the most prominent of those who address   See Plantinga (1988, 1) for agreement.   Beside Alston, other epistemologists who have addressed the primary question are, e.g., Bedford Naylor (1988); Brandt (1985); Chisholm (1977, Ch. 1); Ginet (1975); Plantinga (1988). Note that the question: “what is justification itself ”? has been far more widely discussed in the philosophy of law than in epistemology. More on this can be found further on in the chapter. 2 3

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

248  Anne Meylan it. Alston addresses the primary question of justification; specifically, the two following excerpts pertain to the very problem that interests us in this chapter, namely, the kind of normativity that characterizes justification: The terms ‘justified’, ‘justification’ and their cognates are most naturally understood in what we may term a ‘deontological’ way, as having to do with obligation, permission, blame, and the like. . . . More specifically, when we consider the justification of actions, something on which we have a firmer grip than the justification of beliefs, it is clear that to be justified in having done something is for that action not to be in violation of any relevant rules, regulations, . . . , obligations, duties, . . . To say that the action was justified does not imply that it was required or ­obligatory, only that its negation was not required or obligatory. . . . The most natural way of construing the justification of beliefs is in parallel fashion. To say that S is justified in believing that p at time t is to say that the relevant rules or principles do not forbid S’s believing that p at t. In believing that p at t, S is not in contravention of any relevant requirements. Again, it is not to say that S is required or obligated to believe that p at t.  (Alston 1989, 115–16)

In another paper, Alston asserts the same view: It is natural to set out a deontological concept on the model of the justification of behavior. To say that my expenditures on the trip were justified is not to say that I was obliged to make those expenditures (e.g., for taxis), but only that is was all right for me to do so, that in doing so I was not in violation of any relevant rules or regulations. . . . Similarly, to say that a belief was deontologically justified is not to say that the subject was obligated to believe this, but only that he was permitted to do so, that believing this did not involve any violation of relevant obligations. (Alston 1989, 85)4

Before we summarize the points made here, we need to say more about the variety of normative properties. The claim that there are various kinds of normative properties is a familiar one.5 I content myself with presenting three kinds. 1. The first is the deontic kind of normative properties. It is sometimes said to encompass two sub-kinds: 1a.  The first deontic sub-kind, which entails the properties that are traditionally located on the deontic square of opposition—the obligatory, the permissible, the forbidden, and the non-obligatory—as well as the optional, i.e., what is neither obligatory nor forbidden.6 1b. The second deontic sub-kind, which entails the properties of being right and wrong.   See also Alston (1993; 2006).   See Mulligan (2009); Ogien and Tappolet (2009); Ross (1930); Thomson (2008); Timmons (2002); Von Wright (1963). Sometimes the expression “normative concepts” is used to denote what I call “deontic concepts”. As said already, “normative concepts” refers here to the general class that has deontic concepts as its members. 6   See Timmons (2002, 9). This list of deontic concepts is not intended to be exhaustive. Some philosophers will probably add praiseworthiness, blameworthiness, and the supererogatory, for instance. 4 5

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

In Support of the Knowledge-First Conception  249 2. The axiological kind, to which all the value properties belong, that is, the valuable and the disvaluable, but also the elegant, the ugly, the disgusting, the morally good, the prudentially bad, etc. 3. The aretic kind, which entail the properties of being vicious and virtuous, but also the properties of being wise, courageous, etc. We are now in a better position to formulate Alston’s conception of the normativity of justification. According to Alston: • Justification is a deontic property (and not, e.g., an axiological one). More precisely: • A justified entity is neither an obligatory nor a forbidden one,7 that is, a justified entity is one that is optional.8 I shall dub this conception of the normativity of justification the optional conception. Note that the endorsement of the optional conception is not confined to Alston.9 Carl Ginet, for instance, has something very similar in mind when he claims that: One is justified in being confident that p if and only if it is not the case that one ought not to be confident that p: one could not be justly reproached for being confident that p. (Ginet 1975, 27)

The optional conception undeniably has something going for it. As Alston emphasizes, it seems to be supported by the following intuition (an explanation of why it is not, in fact, the case is given here). When one says that an action, e.g., the breaking of diplomatic relations, is justified, one does not say that this action is obligatory. One rather says that this action is permitted.

Given that this intuition pertains to what we ordinarily mean when we say that an action is justified, I shall, accordingly, call it the semantic intuition of ordinary justification.

7   That a justified belief is not a forbidden one is obvious. This is certainly why Alston does not even mention this possibility. 8   According to another possible way of understanding these quotations, the second clause could rather be the following: a justified entity is a permissible one (and not an optional one), that is, when an entity is justified, it is not necessarily, though it is possibly, obligatory. The choice of how to understand Alston’s conception of the normativity of justification does not affect the reasoning laid out here. According to the alternative interpretation, indeed, justified entities are not entities that comply with a norm either. Thus, even if we interpret Alston’s view in this alternative way, his idea still clashes with the claim that to be justified is to comply with a norm, and the forthcoming discussion keeps its relevance. 9   See also Naylor (1988, 51) for what seems to be another statement of the optional conception.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

250  Anne Meylan

3.  The New Evil Demon Problem and the Knowledge-First Conception of the Normativity of Justification I have said that the purpose of this chapter is to settle the debate between two irreconcilable conceptions of the normativity of justification. The first of these two incompatible views, the optional conception, has just been presented. The second of these two conflicting conceptions of the normativity of justification can be found in the specific solution that some advocates of a knowledge-first epistemology offer to the NEDP. I  shall now briefly remind you what the NEDP is and present the knowledge-first ­solution to this problem.10 Consider two epistemic counterparts, that is, two subjects who are the same as far as their non-factive mental states are concerned. The world appears to both of them in exactly the same way. What seems, on the basis of the deliverances of visual perception, true to the first subject also seems true to the second on the basis of the same kind of deliverances; what appears true to the first subject given what she seems to remember also seems true to the second, on the basis of what she seems to remember, etc. Suppose, furthermore, that these two subjects are, in fact, involved in two kinds of situation: a good one and a bad one. The difference is as follows. What appears to be true to the subject involved in the good situation is the result of her reliable cognitive capacities. The beliefs that are based on the appearances in the good case (below “good-case beliefs”) are, as such, generally true. But what appears to be true to the subject in the bad situation is the result of a malevolent demon’s manipulations. The subject’s beliefs based on the appearances in the bad case (below “bad-case beliefs”) are, as such, generally false. This is the thought experiment that lies at the heart of the so-called New Evil Demon Problem. The NEDP is an objection to many externalist conceptions of justification, including the knowledge-first one. According to an externalist conception of ­justification, indeed, only the good-case beliefs are justified. This result clashes with the following appealing internalist intuition: Both the good-case and the bad-case beliefs are equally justified. The NEDP is not a superficial difficulty for externalism. The new evil demon objection is, to my mind, the single most important objection to externalism in the literature. If . . . Cohen’s intuitions are correct and they have the significance he takes them to, his objection to externalism is conclusive.  (Littlejohn, forthcoming)

It is not surprising, then, that in the externalist literature we see numerous attempts to resist the NEDP. The advocates of the knowledge-first epistemology (e.g. Littlejohn 2009, forthcoming; Williamson  forthcoming) have offered what seems to be  a   The paternity of the NEDP is generally attributed to Cohen (1984).

10

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

In Support of the Knowledge-First Conception  251 promising solution to this problem. The general format of the solution proposed by the knowledge-firster is as follows: the norm fulfilled by the good-case beliefs is the norm of justification. But the norm satisfied by the bad-case beliefs is not the norm of justification. In the bad case, the subject’s beliefs, in fact, comply with another norm that makes her excusable for what she believes. Williamson’s recent way of describing the distinction between the norm of justification and the norm of excusability runs along the following lines: • A justified belief is a belief that complies with a “primary truth-related norm of justification” (Williamson forthcoming, 12). • An unjustified but excusable belief is not one that complies with the primary norm of justification but with another (derivative) norm: the norm of being a belief that a person who tends to comply with the norm of justification would hold. If this is right, then, only the good-case beliefs are justified (as the knowledge-firster wishes to say). Against the aforementioned internalist intuition, two epistemic counterparts are not, in fact, equally justified in what they believe. In law, justification and excuses constitute two forms of defence, and their precise difference is a matter of important disagreement among legal philosophers. Confronting the aforementioned knowledge-first solution to the NEDP with the numerous distinctions that the legal literature has put forward would lead me too far afield.11 Let me say in passing that the difference between excusability and justification is often taken to hinge on the absence or presence of responsibility. Following Austin (1956), it is widely held that an excusable action is, in contrast to a justified one, an action for which the responsibility of the agent is removed, or at least mitigated.12 Now, an unjustified belief that complies with the aforementioned derivative norm— the one governing excusability—is typically one whose absence of justification is not something for which the believer is responsible. In this sense, Williamson’s way of differentiating between justification and excusability is in line with the dominant view in legal philosophy. What mainly interests us here, however, is the conception of the normativity of ­justification on which the proposed knowledge-first solution to the NEDP relies. According to this solution, a justified belief is a belief that complies with a norm, that   See e.g., Baron (2005; 2006); Bickenbach (1983); Gardner (2007).   Gardner (2007, 83) takes the characterization of excuses as (full or partial) denials of responsibility to be seriously misleading. Very briefly, his conviction is that we do not deny responsibility when we excuse people, for the reason that excusing someone for her wrongdoing requires her to be answerable for this wrong. We do not excuse people who cannot even be punished for their wrongdoings. One way of making Gardner’s claim compatible with the widespread view, according to which an excusable action is an action for which the responsibility of the agent is at least mitigated, is to differentiate between two levels of responsibility in the way suggested by Feinberg (1970, 130–9). We could then say that excuses require, on one hand, first-level responsibility (what Feinberg calls “authorship” perhaps) while implying, on the other hand, (full or partial) denial of higher-level responsibility (what Feinberg calls “liability”). 11 12

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

252  Anne Meylan is, a belief, as Williamson claims,13 that satisfies some sort of “ought” or “should”. Crucially, this conception of the normativity of justification—viz., the one on which the knowledge-first solution to the NEDP relies—and the so-called optional conception seem irreconcilable. The claim that: • A justified belief is something that satisfies some sort of “ought” is not compatible with the claim that: • A justified belief is something optional, that is, something that neither satisfies nor violates any specific norm or requirement. To clearly distinguish it from Alston’s view, I shall occasionally call the view about the normativity of justification included in the knowledge-first reply to the NEDP: the non-optional conception. The knowledge-first conception of the normativity of justification is the non-optional conception.

4.  Ordinary Justification: Why it Matters in the NEDP I have now presented two existing conceptions of the normativity of justification that I consider incompatible. The statement that they are opposing conceptions might—I anticipate—raise the following worry: The two conceptions of the normativity of justification in question do not in fact conflict, because they are not intended to capture the same thing. The optional conception is intended to capture the normativity of ordinary justification, while the knowledge-first conception —i.e., the non-optional conception— is intended to capture the normativity of some specifically t­echnical property also called “justification”. In this section, I reply to the worry stated by showing that the knowledge-first conception intends to capture ordinary justification as well (contrary to what this expected objection states). Before doing this, it is necessary to make a brief terminological ­clarification as regards the use of the expression “ordinary/technical justification” in this chapter. There is, as Alston claims, as well as in the quotations earlier, a natural or ordinary meaning of the term “justified” (and its cognates). This is what we take “justified” to mean when the term is included in sentences of ordinary language such as, for instance, in the sentence: “The breaking of diplomatic relations is justified in these ­circumstances”. Now, ordinary justification is what is denoted by “justified” when it is used in this ordinary way. In contrast, technical justification is any property (whatsoever) that is denoted by “justified” when the latter term is not used according to this ordinary meaning.   See Williamson (Forthcoming).

13

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

In Support of the Knowledge-First Conception  253 It is important to note that philosophers, more often than not, do not have ordinary justification in mind when they speak of “justified belief(s)”. This is, for instance, true of many epistemologists who endorse the traditional definition of knowledge.14 Many epistemologists who take knowledge to be “justified true belief ” do not take “justified” to denote ordinary justification. They do not take the predicate “justified” to mean the same thing that it denotes in sentences like “The breaking of diplomatic relations is justified in these circumstances”.15 As BonJour states explicitly, epistemologists, more often than not, make use of the term “justified” (and its cognates) in order to express a technical notion. The specifically epistemological notion of justification is to a significant extent a technical ­philosophical notion, one that is not clearly and unquestionably present in common sense. (BonJour 2007, 36)

In many epistemological discussions, “justified” refers to a technical property, one that does not necessarily have much to do with what is denoted by sentences of ordinary language such as “The breaking of diplomatic relations is justified in these circumstances”. What about the discussion surrounding the NEDP? Is the notion of justification that the knowledge-firster and her (internalist) opponent have in mind in this particular epistemological discussion also a technical notion? The fact that many epistemologists who endorse the traditional definition of ­knowledge have a technical notion of justification in mind is not problematic,16 because they share, at least, a minimal view of what this technical notion of justification is. Obviously, all the epistemologists who take knowledge to consist in justified true belief commonly conceive of justification as: “a necessary component of knowledge”. But, crucially, this minimal agreement over a technical notion of justification is not reached in the debate opposing the knowledge-firster and her (internalist) opponent as regards the NEDP. First, indeed, the advocates of the knowledge-first epistemology do not conceive of justification as “a necessary component of knowledge”. Famously enough, justification is not a component of knowledge according to them. Second, there is no reason to think that the epistemologists who are opposed to each other in the discussion of the NEDP—viz., the knowledge-firster and her (internalist) opponent—have any other common technical concept in mind. What, indeed, would it be? What other technical notion of justification could these opposing parties share? I cannot conceive of any myself, and I challenge anyone to find one. However, when disagreeing on whether only good-case beliefs or both good- and bad-case beliefs are justified, the   For the sake of brevity, I leave the “unGettiered” feature aside.   It is relatively simple to consolidate the claim that these epistemologists have a technical notion in mind with an argument (see Meylan (Forthcoming) for such an argument). Moreover, this claim does not generally raise eyebrows. Even though few epistemologists explicitly recognize this technical use, many of them would—I suspect—acknowledge it if they were asked. 16   I do not deny that some problems might remain as Alston (1993; 2006) emphasizes. 14 15

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

254  Anne Meylan knowledge-firster and her (internalist) opponent need to share some minimal view of what they are discussing, i.e., justification. Otherwise, of course, the two camps could simply be m ­ isunderstanding each other. And nothing would prevent each camp from being right with regards to the particular technical notion of justification it has in mind. This is why ordinary justification really matters in the debate surrounding the NEDP. Ordinary justification, viz. what is denoted by the term “justified” in sentences of ordinary language, is what is commonly discussed by the knowledge-firster and her (internalist) opponent in the NEDP. This warrants that they are not simply misunderstanding each other. For sake of clarity, here is a brief reformulation of the reasoning from which this conclusion results. 1. Either the opposing camps—viz., the knowledge-firster and her (internalist) opponent—involved in the discussion of the NEDP share a minimal notion of what they are discussing, i.e., justification, or they do not share such a minimal notion and they are, in fact, misunderstanding each other. 2. The camps in question are not misunderstanding each other (assumption).17 3. When discussing the NEDP, the minimal notion of justification that the ­knowledge-firster and her (internalist) opponent share is either a specifically technical notion of justification or the ordinary notion of justification. 4. The minimal notion of justification that the knowledge-firster and her (internalist) opponent share is not a specifically technical notion of justification (which one would it be?). 5. The minimal notion of justification that the knowledge-firster and her (internalist) opponent share is the ordinary notion of justification. That the optional conception is supposed to capture ordinary justification is indisputable, since this conception, as we saw, is explicitly taken to reflect the semantic ­intuition of ordinary justification. What the reasoning that I have just conducted is supposed to show is that this is also the case for the non-optional or knowledge-first conception, viz., the one that is put forward by the knowledge-firster as part of her solution to the NEDP. Thus, against the worry introduced at the beginning of this section, the optional and the non-optional or knowledge-first conception really are two incompatible answers to the following question: what is the normativity of ordinary justification? In what remains of this chapter, I set out to show that the non-optional or knowledge-first ­conception seems to be a better answer to the latter question.

17   This is assumed to be true in this chapter. Note that several philosophers have defended the opposite view, according to which epistemologists—mainly internalists and externalists—simply have distinct and possibly compatible notions of justification in mind. See Alston (2006); BonJour (2007); Foley (2002; 2004).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

In Support of the Knowledge-First Conception  255

5.  The Mistake of the Optional Conception The optional conception is supposed to be strongly supported by the semantic ­intuition of ordinary justification, that is, to recall, the intuition according to which: When one says that an action, e.g., the breaking of diplomatic relations, is justified, one does not say that this action is obligatory. One rather says that this action is permitted. The misleading tendency to think that the semantic intuition of ordinary justification supports the optional conception is due to the ambiguity of the term “permitted”. To see this, let me start with two examples in which the term “permitted” is used in a ­perfectly ordinary way: Case 1 While Sam is walking along the lakeshore, he sees a sign that says: “Swimming is ­forbidden is this lake section, but permitted 200 yards away.” Case 2 Learning that her son has stolen a pair of trainers from the supermarket, Carla promptly decides to take them back to the shop discreetly. On her way there, her neighbour, Felicity, who is well known for being a gossip, stops Carla and asks her what she is doing with this brand-new pair of shoes in her hands. Annoyed, Carla decides to lie. While Carla feels guilty because of her lie, her friend Dan tells her: “Don’t worry, Carla, it is permitted to lie in such circumstances”. The term “permitted” is ambiguous and has (at least, but probably more) two distinct meanings that are respectively at work in cases 1 and 2. In case 1, what the sign means is that it is neither obligatory nor forbidden to swim 200 yards away. To say that something is permitted, according to the first meaning of this term, is to say that a thing is optional. But this is not what “permitted” means in the second case. What Dan says to Carla in order to comfort her is not that lying is optional in these circumstances. This would amount to say that Carla’s behaviour would have been “OK” no matter what she did, or that she had as many reasons to lie as reasons not to. This would be saying that lying is not even prima facie wrong. This is certainly incorrect and such a dubious claim would not help Carla feel better. Rather, what Carla needs to be convinced of is the following: “even though lying is prima facie wrong, you really had overriding reasons to lie in these circumstances”. Put differently, the most natural way of understanding Dan’s statement: “it is permitted to lie in these circumstances” is as: “lying is the lesser evil in this case”. Given the fact that “permitted” does not necessarily mean “optional”, there is no ­reason anymore to think that the semantic intuition of ordinary justification supports the optional view specifically.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

256  Anne Meylan What is more, there is reason to think that the semantic intuition of ordinary justification rather supports the view according to which, for an entity to be justified (in the ordinary sense) is for it to be the lesser evil. Indeed, the claim that a justified action is one that constitutes the lesser evil, in a particular situation, is generally accepted by lawyers and philosophers of law (see e.g., Bickenbach  1983, Alexander  2005). According to the Model Penal Code (section 3.02), for instance: A conduct which the actor believes to be necessary to avoid a harm or evil to himself or to another is justifiable, provided that: (a) The harm or evil sought to be avoided by such conduct is greater than that sought to be prevented by the law defining the offense charged.18

To reiterate, against Alston, the semantic intuition of ordinary justification rather ­supports the view according to which: • For an entity to be justified is for it to be the lesser evil. Clearly, drawing this conclusion is still insufficient to prove, as I intend to do, that the non-optional conception—which is, to recall, the knowledge-first conception—better captures the normativity of justification. A further point has to be made: I need to show that a belief that is the lesser evil is, in fact, one: that complies with a norm, that is to say, one that satisfies some sort of “ought” or “should”, as the knowledge-firster claims. I will make this additional point by relying on Ross’ notion of all things considered rightness. According to Ross: The all things considered—in contrast to prima facie—right action to perform, in a particular set of circumstances, is the action which, of all those possible for the agent in these circumstances, has the greatest balance of prima facie rightness in those respects in which it is prima right, over its prima facie wrongness in those respects in which it is prima facie wrong.19

According to Ross, actions that constitute the lesser evil, in a given set of circumstances, are (all things considered) right actions.20 Importantly for our purpose, Ross (1930, 3–6) also explicitly identifies right actions with actions that ought to be done. The claim that all things considered right actions are actions that ought to be done results indeed from the following apparently compelling reasoning.

  The MPC stipulates two additional conditions, but these are irrelevant for our purposes here.   This is a slightly modified version of Ross’s formulation. See Ross (1930, 41). 20   Sometimes philosophers take “right” to be synonymous with “permissible”. That is, they take “right” to denote all actions that are either obligatory or optional. See, e.g., Timmons (2002, 9). This conception is broader than Ross’s. As I argue, when “right” is understood along Rossian lines, right actions do not include optional actions. 18 19

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

In Support of the Knowledge-First Conception  257 1. Actions are either optional or forbidden or they ought to be performed (i.e., obligatory). 2. An optional action is neither something that ought to be performed nor something forbidden. 3. Whenever an action A is optional, non-A is also optional.21 4. Whenever an action A is optional, non-A equals A with respect to their all things considered rightness. (3)22 5. Whenever an action A is optional, it does not qualify as the action that has the greatest balance of prima facie rightness over prima facie wrongness with respect to all the actions that the agent can perform in the circumstances. (4) 6. Whenever an action A is optional, it is not an “all things considered” right action. (5 and Rossian conception of “all things considered” rightness presented here). 7. An “all things considered” right action is not a forbidden action. 8. Then, an “all things considered” right action is an obligatory action, one that ought to be performed. (6, 7) Let me briefly marshal the last considerations. First, I have tried to show why—against Alston and, more generally, against those who take justification to be simply a matter of permissibility—the semantic intuition of ordinary justification does not support the optional conception. The semantic ­intuition of ordinary justification rather supports the view according to which to be justified is to be the lesser evil. Second, relying on Ross’s discussion of all things considered rightness, I have explained why an action or belief that is the lesser evil—that is, all things considered right in Rossian terminology—is, in fact, an action or belief that ought to be performed or held. When put together, these two elements provide an answer to our initial question: which of the two conceptions better captures the normativity of justification, namely: 1. the optional conception; or 2. the knowledge-first or non-optional conception? Against, perhaps, all expectations, the knowledge-firster’s claim according to which a justified belief is one that satisfies some sort of ought or should matches perfectly with what we ordinarily conceive to be a justified action.

21  This directly results from the fact that an optional action is one that is neither obligatory nor forbidden. 22   Indeed, an optional action A cannot be optional if it would be either better or worse to perform non-A.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

258  Anne Meylan

References Alexander L. 2005. Lesser Evils: A Closer Look at the Paradigmatic Justification. Law and Philosophy 24: 611–43. Alston W. 1989. Epistemic Justification. Essays in the Theory of Knowledge. Ithaca, NY: Cornell University Press. Alston W. 1993. Epistemic Desiderata. Philosophy and Phenomenological Research 53(3): 527–51. Alston W. 2006. Beyond “Justification”: Dimensions of Epistemic Evaluation. Ithaca, NY: Cornell University Press. Austin J. L. 1956. A Plea for Excuses. Proceedings of the Aristotelian Society 57: 1–30. Baron M. 2005. Justifications and excuses. Ohio State Journal of Criminal Law, 2: 387–406. Baron M. 2006. Excuses, excuses. Criminal Law and Philosophy 1: 21–39. Bedford Naylor M. 1988. Epistemic Justification. American Philosophical Quarterly 25(1): 49–58. Bickenbach J. E. 1983. The Defence of Necessity. Canadian Journal of Philosophy 13(1): 79–100. BonJour L. and E. Sosa. 2007. Epistemic Justification. Internalism vs. Externalism. Foundations vs. Virtues. Oxford: Blackwell. Brandt R. B. 1985. The Concept of Rational Belief. The Monist 68(1): 3–23. Chisholm R. 1977. Theory of Knowledge, Second Edition. Englewood Cliffs, NJ: Prentice-Hall. Cohen S. 1984. Justification and Truth. Philosophical Studies 46(3): 279–95. Feinberg J. 1970. Doing and Deserving. Essays in the Theory of Responsibility. Princeton: Princeton University Press. Foley R. 2002. Conceptual Diversity in Epistemology. In P. Moser, ed., The Oxford Handbook of Epistemology. New York: Oxford University Press, 177–203. Foley R. 2004. A Trial Separation between the Theory of Knowledge and the Theory of Justified Belief. In J. Greco, ed., Ernest Sosa and His Critics. New York: Wiley-Blackwell, 59–71. Gardner J. 2007. Offences and Defences: Selected Essays in the Philosophy of Criminal Law. Oxford: Oxford University Press. Ginet C. 1975. Knowledge, Perception and Memory. Dordrecht: D. Reidel. LittleJohn C. 2009. The Externalist’s Demon. Canadian Journal of Philosophy 39(3): 399–434. Littlejohn C. Forthcoming. A Plea for Epistemic Excuses. In F. Dorsch and J. Dutant, eds., The New Evil Demon. Oxford: Oxford University Press. Meylan A. Forthcoming. The Pluralism of Justification. In A. Coliva A. and N. Pedersen. eds., Pluralism, London: Palgrave Macmillan. Mulligan K. 2009. Values. In R. Poidevin, P. Simons, A. McGonigal, and R. Cameron. eds., The Routledge Companion to Metaphysics. London: Routledge, 401–11. Ogien R. and C. Tappolet. 2009. Les concepts de l’éthique. Faut-il être conséquentialiste? Paris: Hermann. Plantinga A. 1988. Positive Epistemic Status and Proper Function. Philosophical Perspectives 2: 1–50. Ross W. D. 1930. The Right and the Good. Oxford: Oxford University Press. Thomson J. J. 2008. Normativity. Chicago: Open Court. Timmons M. 2002. Moral Theory. An Introduction. Oxford: Rowman and Littlefield. Von Wright G. H. 1963. The Varieties of Goodness. London: Routledge and Kegan. Williamson T. Forthcoming. Justifications, Excuses, and Sceptical Scenarios. In F. Dorsch and J. Dutant, eds., The New Evil Demon. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

13 Sustaining Rules A Model and Application John Turri

1. Introduction We value many things. Some are easily obtained, others are not. We also disvalue many things. Some are easily avoided, others are not. To obtain or avoid any of these things, we need ways of doing so, ways that are good enough to be worth the effort. Worthy ways persist and unworthy ways perish. Worthy ways of the right sort proliferate, thereby instituting and sustaining practices. These observations can seem uncontroversial, even bordering on platitudinous. But they provide most of the materials needed to develop a robust theory of the rules that normatively sustain a practice. In short, they point toward a theory of what I call ­sustaining rules. Before proceeding I should clarify what I mean by ‘rule’ and ‘practice’. A rule is a principle or standard of conduct. Conduct can be assessed according to a rule. If you conform to the rule, then your conduct is correct, according to the rule. If you break the rule, then your conduct is incorrect, according to the rule. A practice is an ­ongoing way of doing things. Practices are instituted, can persist, change, and expire. I am thinking of practices as concrete patterns of actions, held together as patterns by the rules that their agents follow. The rules ‘hold the practice together,’ by analogy to the way that water binds together isolated bits of powdered milk into liquid milk. A practice could also be thought of as an abstract type, a description of a series of actions done according to a rule. For convenience, I will continue speaking of practices as concrete patterns rather than abstract types, and of rules sustaining practices thus construed. I will have more to say about how I am thinking of practices as the discussion unfolds.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

260  John Turri

2.  The Model: a Simple and Intuitive Presentation There are six basic elements of the model:1 agent, goal, rule, achievement, persistent pattern, and explanation.2 A rule sustains a practice when the agent follows the rule and achieves what it prescribes, the value of which, in turn, explains why the agent continues following the rule, thereby establishing a persistent pattern. In a normatively sustained practice, the value achieved through earlier instances of rule-following breeds subsequent rule-following. In a word, there is reproduction via value produced. At the risk of oversimplification, it might be helpful to visualize the model like so: Agent + Goal + Rule

Achievement

Persistent Pattern

Figure 13.1  A model for sustaining rules.

The arrows in Figure  13.1 represent explanatory relations. The emergence of a ­persistent pattern in this way just is the emergence of a practice: it makes the practice. Generation is the initial, limiting case of sustainment. Strictly speaking, it is the activity of agents that generates and sustains practices, what we might call sustaining agency. Sustaining agency and sustaining rules are but two sides of the same coin.3 I aim to illuminate both sides, although it has proven easier for me to conduct the discussion directly in terms of rules and allow the points about agency to enter at the appropriate junctures. Our powers and abilities equip us to exercise some measure of control over ­outcomes, such that successful outcomes can manifest our powers and abilities.4 (For convenience, I will often just say either ‘powers’ or ‘abilities’, instead of mentioning both.) An outcome is our achievement if and only if it manifests our abilities. Not every successful outcome is an achievement. For example, some are due to luck.5 A feasible outcome is one that we can achieve as we are actually constituted, which requires that we actually have some ability to produce it. An infeasible outcome is one that we have no ability to produce as we are actually constituted. (An infeasible outcome might still be possible, even if we actually have no ability to produce it.) 1   The view is partly inspired by Millikan’s (2005) ‘biological model’ of public linguistic meanings as ‘stabilizing functions.’ It is also broadly related to the theory of ‘domain’ normativity and performance assessment discussed in Sosa (2007: Chs. 2 and 4), and the project of the primitive enquirer (‘A’) discussed in Williams (1978: 23ff). There are also superficial similarities to Alston’s (1991) ‘doxastic practice approach’ to epistemology, though deep disagreements divide me and Alston. My thinking on this subject was influenced by reading in the summer of 2010 a paper by Benjamin Jarvis, entitled ‘Assertion and Advocation’ (unpublished MS). 2   Several of these elements could be plural. More on this further on. 3   MacIntyre (1999) has influenced my thinking here. 4   For a partial theory of this relation, see Turri (2015a). 5   For more on this way of thinking about powers and outcomes, see Sosa (2007, 2011), Greco (2010), Turri (2011a, 2011b).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  261 It is trivially true that sustaining rules prescribe a feasible outcome. For if we could not achieve the prescribed outcome, then subsequent rule-following could not be explained by our previously achieving the outcome; and if subsequent rule-following could not be explained by our previously achieving the outcome, then we could not sustain the practice by following the rule. Of course, the rule-following behavior might somehow persist even if we achieve nothing valuable through it. We might continue on in the same hopeless way. And our persistence in such hopeless ways might even incidentally cause something valuable to happen to us, perhaps through blind luck or the intervention of a benevolent and merciful higher power. But this would not amount to us sustaining the practice in the sense I am interested in. My view about sustaining rules and agency does not commit me to any particular view about the rules that constitute the actions that occur within a practice. It does not even commit me to the view that the actions that occur within a practice are constituted by rules.6 From the fact that a practice is sustained by a rule, it does not follow that actions occurring within that practice are made possible by the rule, or indeed by any rule. For example, the practice of horse racing involves mounting and riding a horse. But intuitively the actions mounting and riding are not constituted or made possible by the rules of horse racing. Rather, they are antecedently given actions that the practice of horse racing incorporates into more complex patterns of activity.

3.  Refinement and Elaboration Having presented the basic model, in this section I include some finer points and ­qualifications to further clarify the view. I include these refinements and qualifications at this point in order to forestall common questions and misconceptions. Those ­uninterested in such refinements can skip to Section 4. Not all ways of sustaining a practice are ways of normatively sustaining it. My theory applies to both ways of sustaining a practice, but I am really interested in the latter. Normatively sustained practices have genuine reason-giving force, at least for their practitioners: practitioners have good reason to achieve what the rules prescribe. But what distinguishes normatively sustained practices in this way? Every practice has a point, which can be identified with that goal, achievement of which explains why 6   Searle (1969: Ch. 2.5) and Williamson (2000: Ch. 11) both take a different approach to constitutive rules; Rawls (1955: 25) presents a related theory, what he calls ‘the practice conception’ of rules, according to which rules are ‘logically prior’ to certain actions. One main reason that I do not offer my view as a ­theory of ‘constitutive rules’ is that the phrase ‘constitutive rules’ carries considerable baggage, and I wish to avoid verbal squabbles. More importantly, sustaining rules are not ‘general’ or ‘conventional’ rules (Williamson 2000: 238–9), ‘regulative rules’ (Searle 1969: 33), or ‘rules of thumb’ either (Rawls 1955: 23). Sustaining rules, as I define them, differ from all these widely discussed categories. In particular, I am not offering a theory of rules that necessarily ‘create the possibility of new forms of behavior’ (Searle 1969: 33) or are ‘logically prior’ to any actions that occur within the practice. My present purpose is to identify and explain a theory of sustaining rules and aside from ensuring that what I am doing is not confused with what Rawls, Searle, Williamson, and others have tried to do, I lack the space to compare the approaches.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

262  John Turri ­ ractitioners persist in following the rule. (I do not intend to mark an important p ­distinction between a point and a goal. They are both just objectives or ends. I often use ‘point’ for a practice and ‘goal’ for agents, to help keep straight the different elements of the model.) But not all goals are good. A normatively sustained practice has a good point, which explains its reason-giving force. The practice’s point need not be intrinsically or necessarily good in order to imbue it with reason-giving force. It is enough, I think, that the point is usually good, all things considered, in some way or another under normal circumstances. In a word, the point need only be ordinarily good. Consider preventing tooth decay, which is the point of the practice of brushing your teeth daily. Preventing tooth decay is instrumentally good in normal circumstances because in those circumstances it tends to promote better health, which tends to promote greater happiness. This explains why we have a reason to follow the rule brush your teeth daily. Preventing tooth decay is only contingently instrumentally good. In some strange circumstances, it would be instrumentally bad to prevent tooth decay—e.g., a situation in which all people without rotten teeth are tortured to death, or in which our teeth rot as part of a natural and healthy exfoliation process. Nevertheless, although I think that a liberal approach on this point is advisable, it is consistent with the basic model to impose stricter requirements on the point of a normatively sustained practice. In order for a rule to sustain a practice, practitioners need not succeed every time that they engage in the practice or attempt to follow the rule. They need not even succeed most of the time. They need only succeed enough of the time for the practice to sustain itself through the value it brings. Depending on the point of the practice, very infrequent success could be enough. The practice of diamond hunting and other forms of prospecting are like this. And as examples of such practices show, the involvement of later parties in a practice might be explained by the benefits achieved through the rulefollowing of previous parties long ago. For example, a family might engage in the practice of diamond prospecting over many generations and succeed only once in a century. Nevertheless, the benefits earned are enough to sustain the pattern of rulefollowing of family members in between successes. Not all rule-following sustains a practice in the sense I am interested in. In particular, repeated but isolated rule-following that fails to explain further instances of similar behavior does not sustain a practice. For example, suppose that we are consistently convinced by a different person with a different reason each day to hang our laundry out to dry, and the benefit of doing this previously does not explain why we do it subsequently. Even though we frequently follow the rule hang the laundry out to dry, and we benefit from doing so, this is not a normatively sustained laundering practice or even, in the sense I am interested in, a practice at all. There is no ‘reproduction value’ here. Someone observing our daily actions might reasonably infer that we had instituted a laundering practice, but that is because, in this case, outward appearances are misleading. By contrast, if we had as our goal saving energy, and we followed the rule hang the laundry out to dry and thereby achieved energy savings, and this benefit explained

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  263 why  we continued following the rule and reaping the benefit, then it would be a ­normatively sustained laundering practice. Here we have reproduction through the value produced. Closely related to this last point, continued rule-following behavior must have an appropriate etiology in order for it to constitute a practice. Even if a set of activities follows the exact same rule in exactly the same way, it does not follow that they are instances of a single practice. Practices are like biological species in this respect. Specimens must share a proper evolutionary etiology to be conspecifics. Even if a group of specimens share all their gross morphological features, it does not follow that they form a species. They might still be a cryptic species. Moreover, practices that ­prescribe outwardly similar behavior in a wide range of circumstances might still be very different due to differences in their rules. Again, the comparison to biological species is helpful: two species might be very different genetically (follow different ‘genetic rules’) despite being outwardly similar due to convergent evolution. In light of the points made in the last two paragraphs, I find it very plausible that even though practices are uncontroversially artificial in an important sense of that term, they, nevertheless, have an underlying nature that is partly determined by their history. In this respect, they are importantly similar to natural kinds.7 Some practices are intentionally designed, but they need not be. Many important practices are not intentionally designed. For example, the practice of assertion was not intentionally designed. Neither were the chimpanzee practices of smashing open nuts with rocks or using straws to extract termites from crevices. Again, we might compare practices to species, which can be intentionally designed through husbandry or genetic engineering, even though most have evolved without intentional design through ­natural selection. Goals can be complex and difficult to identify precisely. Often we do not just want some particular kind of thing, but rather that kind of thing without any admixture of another particular type of disvaluable thing, both in this specific case and also in the long run. The importance of avoiding disvaluable admixture will partly determine the shape of any sustainable practice organized around the goal. Powers and abilities need not be infallible. Indeed, virtually none is infallible. A power or ability enables you to achieve a certain result in a certain environment. It might enable you to achieve the result on fifty percent of attempts in a normal environment, on ten percent of attempts in an especially hostile environment, and on ninety percent of attempts in an especially hospitable environment. From the fact that you sometimes fail to, say, flip the switch when you try, it does not follow that you lack the ability to flip the switch. Neither does it follow that you have power only over whether you try to flip the switch. You still have the power to flip the switch. It is just not an infallible power. 7   This is in keeping with the most influential contemporary scientific work on human norms, culture, and evolution (e.g. Richerson and Boyd 2005, Tomasello 2008, 2009).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

264  John Turri Getting lucky is a way of getting what we want and it is, trivially, good to have good luck. But, equally trivially, getting lucky is not a way of achieving success, and so not a  way of sustaining a practice. Achievement is the antithesis of lucky success. Achievement is success manifesting ability, and the extent to which success manifests our ability is inversely proportional to how lucky we get. Consistent with this, there could be cases where some combination of luck and ability conspire to bring success. Indeed, most actual cases of success are probably like this. Success is often a matter of degree, as can be the extent to which it manifests our abilities. I will abstract away from these complications and speak as if these were dichotomous matters. Not all rule-following is conscious or explicitly intentional. In fact, I think that most rule-following is instead habitual and automatic, in the sort of way that North American drivers drive on the right side of the road. North Americans habitually, mechanically follow the rule drive on the right side. They do not consciously reflect on it or wake up in the morning and form an explicit intention to drive on the right side today. Closely connected to the previous point, it is worth noting that we typically practice ways of doing things so that they become ‘second nature’—that is, so that we conduct ourselves that way ‘without thinking,’ and especially without consciously deliberating or explicitly formulating and executing an intention. ‘Practice’ in this sense, a synonym of ‘rehearse’, is clearly closely related to the sense of ‘practice’ relevant to my view of how rules sustain practices, a synonym of ‘customary procedure’ or ‘routine’. It is through rehearsal that things become routine: by practice is practice established.8 Practitioners must be implicitly sensitive to the rules of the practice, otherwise they could not follow the rule as opposed to merely conform with it. I take no positive stand on what form such sensitivity must take or what makes it true that such sensitivity is present. But it is clear that sensitivity does not require being able to explicitly formulate the rules. Practitioners need not be especially good judges of what the rules are, even if we expect patterns in their behavior to offer helpful clues to what the rules are. (Actions speak louder than words.) Practitioners might not be explicitly aware of the rules that they are following, or that they are engaging in the practice that they are engaged in. They might even sincerely judge themselves to not be engaged in a practice that they regularly engage in. Plausibly, at least some people engaged in contemporary racist or sexist practices are like this. It is possible that observers of a practice are better positioned than the participants to discover what the rules are. I am not offering a theory of rule-following. I assume that people do follow rules and I take this assumption to be uncontroversial.

8   There are potentially interesting points of convergence here with how Aristotle characterizes moral virtue in the Nicomachean Ethics. Writes Aristotle, ‘Moral virtue comes about as a result of habit’ (Aristotle 350 BCE, Book 2.1, 16–17).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  265

4.  An Example: Robin Hood and his Merry Men This section presents, and the following three sections discuss, a toy example illustrating the model. Robin Hood and his Merry Men: When Robin was a young lad, he witnessed a curious event that left a lasting impression. Just outside a poor village at the edge of the forest, a bold group of criminals robbed a very wealthy nobleman who was transporting his treasures. As they made their getaway, out of exuberance and without design, the criminals tossed some of the stolen treasure along the road in the ­village. The peasants retrieved it and felt very obliged. When the Sheriff arrived to investigate, none of the peasants revealed what they knew, namely, who the criminals were and which direction they fled. Instinctive gratitude led them to remain silent. The criminals robbed from the rich and ‘gave’ to the poor. But this was not their practice. It was an isolated event and they were not in any way aiming at what turned out to be, from their perspective, a very fortunate outcome. Years later when Robin was a man, he became disillusioned with the unjust distribution of wealth in the English feudal system. He desired greater distributive justice in the local economy. Robin recalled the successful robbery he witnessed as a young lad. The lesson of that robbery stuck with him and inspired his incipient plan to enhance distributive justice: rob from the rich and give to the poor. Robin recruited a group of like-minded merry men and set up in Sherwood Forest. They robbed rich people traveling through the forest and gave the windfall to the poor, retaining enough to pay their expenses and sustain their redistributive enterprise. The local population benefitted greatly from this and adored Robin and his merry men because of it. Consequently, the locals refused to cooperate with the Sheriff of Nottingham’s effort to imprison Robin and his merry men. Things continued this way for a long time. Here we have a normatively sustained practice: the practice of Robin Hoodery. An ­initial application of the model to it might go as follows. The agents are Robin and his gang. Their goal is greater distributive justice, which is a good thing. The relevant abilities include their thieving skills.9 The rule is, roughly, rob from the rich and give to the poor. They follow this rule and achieve the goal, which in turn explains why they persist in this pattern of behavior, thus instituting and sustaining the practice. If Robin Hoodery is to be sustained, then its rules—call them Robin’s Rules—will be more complicated than the initial formulation just suggested. The merry men must 9   I am eliding some real but minor complications for the sake of clarity and simplicity. The peasants also play a role in the practice, even if their participation is mainly passive, and their behavior is prescribed by rules of the practice pertaining to gratitude and complicity. The number of abilities is actually surprisingly many, as is fitting for a complicated activity like organized thievery.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

266  John Turri also not undermine their support in the community by, for example, also robbing from poor people. If they rob everybody they encounter, rich and poor alike, then their ­support in the community will be undermined, which will very shortly lead to their demise. Robin’s Rules will also instruct them to avoid this disvaluable admixture of results. At the same time, it is implausible that the merry men would be engaging in the practice incorrectly by not constantly robbing, and presumably the rules should reflect this. Here, then, is one plausible suggestion for a set of rules for Robin Hoodery: Robin’s Rules Robin’s Imperfect Duty:  Not infrequently, rob a traveler. Robin’s Rule of Permission:  You may rob a traveler only if the traveler is rich.10 Robin’s Perfect Duty:  You must distribute the plunder to the local poor, less the ­necessary and appropriate overhead costs of the redistributive activity. Prevailing conditions of human existence demand that a successful plan have some flexibility. Lack of precision is the cost of flexibility. Robin’s Rules are not perfectly precise. What counts as not infrequently will depend on how much redistribution is needed to keep the local population on Robin’s side. What counts as necessary and appropriate overhead will likewise vary and be determined by such things as the costs associated with robbing travelers (e.g., the right kinds of weaponry, disguises, and other equipment) and maintaining the merry men’s living quarters and health. Attitudes of the local poor will also influence what is appropriate. For example, if Robin and his merry men start retaining enough to live a lavish lifestyle, then it might alienate their allies and lead to their demise. Plausibly there is no unique right way to set values for these variables in the plan. Vigilance will be required to keep the values within appropriate ranges. Appropriate attention to environmental factors will shape how Robin and his merry men adjust the values as their practice unfolds. Otherwise put, the relevant environmental factors remotely condition an effective interpretation of the rules and thereby the practice.11 Robin’s Rules prescribe certain outcomes. The merry men have no power to make these outcomes just happen, or happen by luck, or happen by magic. They can achieve these outcomes only by exercising their ability to produce the outcomes. There simply is no other way for them to do it. And it is by achieving these things that the merry men achieve what the rules prescribe, and thereby achieve the goal of the practice, namely, 10   Alternatively: Avoid robbing non-rich travelers. Avoidance could be understood as a matter of degree, in which case the rule need not forbid the robbing of any non-rich traveler; instead, it might require only that non-rich travelers not exceed a certain proportion of overall travelers robbed. As will be noted several times in subsequent footnotes, I am sympathetic to rules requiring only that participants’ activity be marked by certain central tendencies. I include these points in footnotes because it is important to flag that the model is flexible in this way. At the same time, I relegate these points to footnotes because I want to keep the proposals in the main text uncluttered. 11   This idea is inspired by Ben Jarvis’s work on ‘distal shaping’ (unpublished MS). Jarvis expressed some concern about my use of ‘distal shaping’ in the present context, so I switched to ‘remote conditioning’ to pick out the related phenomenon described in the main text.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  267 greater distributive justice. Greater distributive justice is their achievement, an outcome that manifests their skill, effort, and agency. And the value of their achievements breeds further activity of the same sort.

5.  Success and Evidence I am often asked why Robin’s Rules are not framed in terms of evidence, such as the following alternative for a permission rule: Robin’s Evidential Rule of Permission:  You may rob a traveler only if you have good evidence to believe that the traveler is rich. Here good evidence does not guarantee that the traveler is rich, but it does make it probable.12 This is an important suggestion providing an opportunity to explain two other features of sustaining rules, namely, what sort of rules are most eligible to be sustaining rules, and how sustaining rules give rise to derivative rules of a practice. In the remainder of this section, I discuss the former feature; in the following two sections, I discuss the latter. Evidential rules can have an important derivative place in a practice, but they are inferior to what I call direct success rules as candidates for sustaining rules. But before getting to that, I would like to first point out that although there is a fairly natural evidential substitute for Robin’s Rule of Permission, there is no natural evidential substitute for the others. Consider: Robin’s Imperfect Evidential Duty:  Not infrequently, do something that you have good evidence to believe is robbing a traveler. Robin’s Perfect Evidential Duty:  You must do something that you have good ­evidence to believe is distributing the plunder to the local poor, less the necessary and appropriate overhead costs of the redistributive activity. Given how unnatural these are, it seems unlikely that a thoroughgoing commitment to evidential substitutes will be attractive. In light of this, and in order to give the alternative evidential approach the best run for its money, I will focus on the more natural evidential permission rule. Robin’s Rule of Permission is a better candidate for a sustaining rule because it is a direct success rule. Direct success rules directly prescribe the goal. Evidential rules do not. Instead, evidential rules prescribe something that you have evidence to believe will count as satisfying the goal. The difference is critical. Consistently following Robin’s Evidential Rule of Permission could directly undermine the practice. Despite following 12   In my experience, this sort of alternative gets posed by those who favor a non-factive conception of good evidence, so I entertain it accordingly. For work investigating how closely this tracks the ordinary way of evaluating evidence, see Turri (2015b, 2016b).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

268  John Turri the rule, the merry men might still rob many poor people and thereby alienate their allies, which, under prevailing conditions, will effectively end the practice of Robin Hoodery. This would happen if the merry men’s evidence was consistently enough misleading. However, consistently following a sustaining rule would not undermine a practice. Indeed, just the opposite. In line with this, consistently following Robin’s Rule of Permission would not undermine Robin Hoodery, but would instead help sustain it. This is because it is a direct success rule. When I say that direct success rules directly prescribe the goal, this is an oversimplification due largely to the potential complexities of goals. Strictly speaking, it is only a practice’s entire set of sustaining rules that must directly prescribe the goal. That is, the set of sustaining rules, taken as a whole, will have enough content to prescribe the goal. Robin’s Rules form such a set for Robin Hoodery. Individual rules within the set might merely prescribe things that are, at least given prevailing conditions, partly constitutive of or instrumental to the goal of the practice. For example, Robin’s Rule of Permission by itself is not enough to specify the goal of Robin Hoodery, and following it is not enough to sustain the practice.

6.  Heuristics and Evidence What role do evidential rules play in a practice, if not as sustaining rules? They function derivatively as heuristics and as standards of criticism. I discuss heuristics in this section and standards of criticism in the next section. Heuristics are important and useful. A heuristic is intended as a convenient substitute for its correlative direct success rule. If the direct success rule prescribes S, then a correlative heuristic would prescribe H, where achieving H (1) is at least typically easier than achieving S, and (2) correlates well enough with achieving S. (For convenience, I will call a heuristic’s correlative direct success rule its superior rule.) The strength of the correlation determines the quality of the heuristic. Following a good heuristic is a good way of approximating the results of following the superior rule. If the heuristic is good enough, then following it might even be a way of following its superior rule. In light of this, we might call heuristics indirect success rules for a practice. Nevertheless, their usefulness and importance cannot conceal the fact that even good heuristics are parasites. A heuristic’s place and importance in a practice derives entirely from that of its superior rule. If a once good and popular heuristic becomes bad due to some change in the environment, then we would expect its popularity to diminish over time among practitioners committed to the superior rule. And if a better heuristic was discovered, over time committed practitioners would abandon the old heuristic and adopt the new one. The change of preference for one heuristic over the other is explained by the primary importance of the underlying superior rule. And it is the enduring commitment to the underlying superior rule that explains why we have continuity of practice across the more superficial change of heuristics. This is similar to

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  269 the way that, in corporate or military planning, an improvement in tactics is understood by reference to continuity in the underlying strategy that the tactics are intended to implement. Notice something important but easily overlooked. Heuristics are also direct ­success rules in their own right. This might sound paradoxical initially, but a moment’s reflection reveals that it is trivially true. A heuristic tells you directly to do something, too. If you do not do it, then you break the rule and your conduct is incorrect according to the heuristic. Even if it is easier to follow a heuristic than its superior rule, it still might not be easy to follow the heuristic. There can also be heuristics for heuristics, so there might be a temptation to keep finding heuristics that are easier and easier to follow, until we reach rules that we can infallibly follow whenever we choose. However, the relation x is a good heuristic for y is not transitive, because the relation x correlates well with y is not transitive. From the fact that H* correlates well with H, and H correlates well with S, it does not follow that H* correlates well with S. Returning now to Robin’s Evidential Rule of Permission, I think that it functions as a heuristic in the practice of Robin Hoodery. (A better name for it would be ‘Robin’s Permission Heuristic’.) Robin and his men might use evidence of wealth as a proxy for wealth, but the evidential rules are still derivative and parasitic. This fits the pattern of a heuristic in subordinate position to its correlative direct success rule. To illustrate the point, suppose that the prevailing fashion around Sherwood is as follows: all and only rich people wear purple robes, whereas all and only poor people wear green rags. Further suppose that this is common knowledge. Since the color of someone’s robe is typically much easier to detect than their net worth, this provides Robin and his men with a convenient heuristic for selecting eligible travelers to rob: rob a traveler only if the traveler wears a purple robe. Now suppose that fashions change without Robin and his merry men hearing about it. The local rich people suspect that they are being marked for robbery because of their purple robes. So they decide to give away their vast quantities of purple robes to the local poor, who are more than happy to wear them; the rich people then begin wearing green rags instead. Now when Robin and his merry men rob purple-robed travelers, they are robbing poor people—and, perhaps, then giving to the rich in green rags! They are breaking Robin’s Rule of Permission, but they are following Robin’s Permission Heuristic. Will this conduct sustain Robin Hoodery? No, it obviously will not. Their actions destabilize the practice and even threaten to undermine it entirely. If Robin Hoodery is to persist, then Robin and his merry men will have to alter their behavior, bringing it back in line with the Rule of Permission. In the process, if they are wise, they will also adjust their views about what constitutes evidence of wealth. They will interpret the Permission Heuristic in light of what counts as correct observance of the Rule of Permission. Moreover, if given a choice between having their behavior conform to the Rule of Permission or the Permission Heuristic, committed practitioners will choose the former. In light of all this, I submit that the

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

270  John Turri Rule of Permission helps to sustain Robin Hoodery, whereas the Permission Heuristic is just that: a heuristic.

7.  Criticism and Evidence Now let us turn to standards of criticism. Here I mean ‘criticism’ in the negative sense of  ‘expressing disapproval’ or ‘fault finding’, rather than the more neutral sense of ‘evaluation’. Standards of criticism, in the relevant sense, are actually standards of ­punishment, mild though it may be. Telling someone that they did something bad or wrong is, in effect, to impose a mild verbal penalty for their behavior. Many practices involve cooperation among people in different roles. When people’s fates are intertwined, it makes sense to have ways of influencing others to avoid incorrect behavior. Criticism is one way of influencing others. Thus it makes sense that social practices will involve standards of criticism that enable participants to effectively influence associates by negatively reinforcing correct behavior. From the fact that someone breaks a rule, it does not follow that criticizing them or their action is appropriate. If criticism will not improve behavior—or worse, if it will lead to more rule-breaking, or cause some other serious damage—then it would be dysfunctional for a practice to condone such criticism. A well-functioning practice will tend to condone criticism only if, and to the extent that, it is instrumental in promoting the goals of the practice. What will promote the goals of the practice is typically a contingent matter. Humans tend to resent criticism of incorrect behavior that the agent had good reason to think was correct. Thus we would expect successful human practices to reflect this fact by incorporating standards of criticism that place a premium on the agent’s evidence at the time of action. We typically excuse people from criticism when they had good reason to believe that their action was correct, even if it was not.13 Standards of criminal liability often display just this pattern. For example, ­common law typically requires an objective violation of a rule (actus reus) and an understanding that the conduct would violate a rule (mens rea) in order for someone to be legally guilty and thus subject to punishment. Proving that you had good ­evidence that your conduct was legal is enough to avoid guilt, because it is enough to invalidate a claim of mens rea. There are many exceptions to this and a host of other technical legal issues threaten to crowd in almost immediately. But for my purposes it is enough to note this central tendency in the law as a way of illustrating the ­phenomenon I am describing. Breaking the law is considered necessary but (usually) insufficient to warrant a penalty. Of course, following some rules, such as those forbidding the killing of innocents, might be so important to us that we are willing impose punishment for breaking them   For experimental verification of this empirical claim, see Turri (2013); Turri and Blouw (2014).

13

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  271 at all (for further discussion, see the papers collected in Simester 2005). We hold people strictly accountable for breaking such rules and do not excuse them, regardless of their evidence. But strict accountability is the exception in human practice, not the rule. Standards of criticism clearly play an important role in human practices. But standards of criticism are not suitable to hold a practice together, so they are not good candidates for sustaining rules. Escaping criticism can be a good thing but it is not the sort of thing that explains why a practice persists. As an analogy, consider a family consisting of a mother, father, and children. The parents have a duty to provide food for the ­children: the First Family Rule is parents must give their children daily bread. If the ­parents consistently provide bread to the children (‘sustenance’ in one sense), then they are following the rules and this contributes to the family’s continued existence (‘sustenance’ in another sense). By contrast, consider the evidential substitute, the First Family Evidential Rule: parents must do something that they have good reason to believe constitutes giving their children daily bread. If the parents consistently escape criticism for not providing bread to the children—because, say, most things that appeared to be bread were just a bunch of nutritionless lookalikes made from ash, water, and artificial flavoring—then their conduct will not contribute to the family’s continued existence, even though they are following the evidential rule. Indeed, before long their conduct will ensure the family’s demise. Only objective success in providing sustenance can sustain the family. Returning now to the evidential substitutes for Robin’s Rules, I propose that in ­addition to functioning as heuristics, they also help set standards of criticism. Following Robin’s Permissibility Heuristic (i.e. the Evidential Rule of Permission) exempts the merry men from criticism for their actions, even though it does not set the standard for permissible conduct. Recall the imagined situation in which it was common knowledge that all and only rich people wore purple robes. Robin and his men know this, so they have good ­evidence that, by robbing a purple-robed traveler, they are acting permissibly. But then the rich secretly decided to give away their purple robes to the poor and wear green rags instead. Now if Robin robs a purple-robed traveler, he acts impermissibly. But others would be reluctant to criticize him for this mistake (at least initially). That is because he was following the Permissibility Heuristic, which puts his action beyond appropriate criticism. In general, where there is a human social practice with sustaining rule R, and H is commonly employed by practitioners as a good heuristic for R, then we would expect there to be a derivative standard of criticism, C, to the effect that agents are generally exempt from criticism if they follow H but thereby violate R.14 14   My views on this are partly inspired by Williamson’s (2000, 256) distinction between the impermissible and the reasonable, and DeRose’s (2002, 180) related distinction between primary and secondary propriety.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

272  John Turri

8.  Belief Management We have ongoing ways of managing our beliefs. Supposing that belief management is a normatively sustained practice, what can the present model tell us about it? It is a familiar idea that the point of belief management involves believing truths and not believing falsehoods.15 Either goal on its own is easy enough to accomplish, but their combination can make belief management challenging. If the only goal were to believe truths, then believe everything would be a good rule to follow. Likewise, if the only goal were to avoid falsehoods, then believe nothing would be a good rule to follow. But following either of these rules would do us no good; nor would it sustain the practice of belief management. Indeed, it would be a total disaster! We want true beliefs, but without a harmful admixture of falsehood. In the abstract, it is difficult to say precisely how to correctly balance these twin goals, but no one doubts that belief management aims at some mix of acquiring true beliefs and avoiding falsehoods. Belief management might also aim at acquiring beliefs that serve other purposes. These might not be prototypical cases of beneficial belief management, but not only are they possible, they almost certainly actually happen. Athletes work themselves into the false belief that they are better than their opponent, so that they perform better; scientists focus intently on the virtues of their inductive theories and ignore the problem of the pessimistic meta-induction from the history of science; religious fanatics find ways to convince themselves that the evidence for evolution is an elaborate hoax, so that they can sleep better; free-market fundamentalists manage to continue believing that no government regulation of commerce has good consequences, so they do not have to admit to themselves the error of their ways; people who overestimate their competence tend to be more confident and better performers as a result; etc. I accept that belief management could also aim at these outcomes and that such outcomes can benefit us and thereby help to sustain the practice of belief management. Although I remain open to the possibility that such outcomes should not be counted as benefits of belief management per se, but of some other practice instead, I do not find that suggestion very plausible. More generally, I find it implausible that the sustaining rules of belief management will necessarily disallow or discourage such outcomes. Consequently, I will propose an application of the model to the practice of belief management, assuming that these non-truth-directed outcomes promote the goal of belief management and bring value that helps sustain the practice. One way to handle all of this is to posit distinct practices of belief management. One central practice is what I call position management. The goal of this practice is to equip us with beliefs needed to accomplish our goals and continue in a suitable way of life— that is, to help position us for success. We might need to accept or reject certain things just to get our projects off the ground or keep them going. If science would stall unless we believed that the pessimistic meta-induction from the history of science is fatally flawed, then we are well served to get ourselves to believe as much. More generally, if   E.g., James (1897), DePaul (2001), and David (2005).

15

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  273 cognition would stall unless we believed that we know that our fundamental beliefforming processes are reliable, then we have an interest in getting ourselves to believe that. For those who would fall into despair and ruin without the crutch of supernatural religious beliefs, they are prudent to keep themselves believing such things. Again, the point of believing any of these things would be to help position us for success. Such positioning can pertain to global matters, as with the reliability of our basic beliefforming processes, or to exceedingly local matters, as with a cancer patient who needs to believe, ‘The odds are good that I will survive this illness,’ in order to keep things together and have a chance at survival. Position management is a purely strategic belief management practice; whether the beliefs are true is beside the point. The point is effective positioning. Another central belief management practice is what I call inquiry management, which itself seems to divide into at least two further practices. One side of inquiry is attention management. Which questions should we consider? And which of these deserve more of our time and energy? There is no point in considering questions ­useless to us, or in devoting more time and energy to a question than it deserves. We could be the best truth-detectors in the world, but that would not matter if we never considered worthy questions. Acquiring worthless things will not sustain a practice. Attention management is also a strategic belief management practice, concerned with aiming cognition in the right direction. Its point is to set an appropriate cognitive agenda. The other side of inquiry is truth-detection, or cognition proper. The goals and rules of position management do not apply here. And once a question has been considered, attention management has done its work, for better or worse. The only concern at this point is whether P is true. The aim of cognition is twofold: get the correct answer and avoid the incorrect answer. Getting the correct answer means believing P, if P is true, and believing not-P, if P is false. Avoiding the incorrect answer means not believing P, if P is false, and not believing not-P, if P is true. What are the sustaining rules of belief management? The point of belief management in general is to acquire useful beliefs. This is a very broad goal, but that is appropriate because belief management is a very broad practice serving our interests across the whole range of life’s activities. Call any belief, true or false, that benefits us a useful belief. Just as robbing some travelers was required to sustain the practice of Robin Hoodery, the formation of some beliefs will be required to sustain the practice of belief management. So at least one of the rules of belief management will be: Belief ’s Imperfect Duty:  Not infrequently, form a belief. Forming useless beliefs will not sustain the practice, so we should also expect a rule of this form: Belief ’s Rule of Permission:  You may believe P only if believing P is useful.16 16   Perhaps strict permission is the wrong deontic category. If so, a natural alternative would be Belief ’s Rule of Discouragement: You ought to believe P only if believing P is useful. Another weaker alternative is Belief ’s Rule of Recommendation: It is recommended that you believe P only if believing P is useful.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

274  John Turri But as we have seen, beliefs can be useful in different ways. They can be useful because they position us to proceed and perform well enough. The practice of position management looks after that. And they can be useful because they are correct answers to questions that matter. The practice of inquiry management looks after that. What are the rules of position management? Whether the belief is true is beside the point for this practice. Focusing for the moment on the most serious cases where the presence or absence of certain beliefs would immobilize us, two rules of this practice might be: Positional Rule of Permission 1:  You may believe P only if believing P will not immobilize you. Positional Rule of Permission 2:  You may refrain from believing P only if doing so will not immobilize you. The practice might also incorporate standing recommendations: Positional Recommendation 1:  To the extent that believing P will help you, it is ­recommended that you believe P. Positional Recommendation 2:  To the extent that believing P will harm you, it is ­recommended that you not believe P. What are that rules of inquiry, beginning with attention management? Attention ­management is presumably characterized by at least a pair of rules: Attention’s Imperfect Duty: Not infrequently, pose a question. Attention’s Rule of Permission: You may pose a question only if its answer matters.17 Once attention management has played its role, cognition begins. What are the rules of cognition? It is doubtful that we break the rules of cognition by not answering every question posed. Nevertheless, there is an expectation that questions will be answered most of the time. This suggests that the duty is imperfect but still stringent, such that you are required to usually answer. Add to this that cognition aims at providing true answers and avoiding false ones, and a plausible pair of rules suggests itself: Cognition’s Imperfect Duty: Usually when the question whether P is posed, form a belief that P, or form a belief that not-P. Cognition’s Rule of Permission: You may form a belief only if it is true. Just as the merry men achieved the goal of Robin Hoodery by achieving what Robin’s Rules prescribe, so too will we achieve the goal of belief management by achieving 17   Again, I am open to the possibility of replacing ‘may’ with ‘ought’ or ‘should’, and also to the suggestion that the rule is better understood as a stringent imperfect duty, placing restrictions only on the overall ratio of useful to non-useful questions asked.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  275 what the rules prescribe. In the case of cognition, true beliefs are prescribed, so we achieve true beliefs. We are naturally equipped with powers of perception, are disposed to trust the word of others, and are habituated to make certain inferences. These are powers we have for achieving true beliefs: powers of detection and discovery. Some truths are harder won than others, but in a wide range of typical cases we can more or less directly achieve true beliefs through the exercise of these powers. I am offering a view about the rules that normatively sustain practices. This does not commit me to the view that actions that occur within a practice are themselves made possible by the practice’s rules, or indeed by any rules at all. Even if the rules of belief management pertain to the activities of questioning and believing, it does not follow that either questioning or believing is made possible by these rules, or that they could occur only in the context of such a practice. To the contrary, one might argue that inquiry management aims to manage the antecedently given activities of questioning and believing. In any event, here I am not taking a definite stand on whether questioning or believing can or must be metaphysically grounded in a prior practice.

9.  Knowledge and Belief Management If what I have said thus far is on the right track, then in light of an important result from  recent epistemology, we learn something important about the relationship between knowledge and belief management. First I will explain the result from recent epistemology. Then I will explain what this entails about knowledge’s relationship to belief management. It has recently been persuasively argued that to know just is to achieve true belief.18 This thesis has been defended on a wide range of grounds, including the following: it provides a simple, elegant, and extensionally adequate definition of knowledge, it solves the Gettier problem,19 it underwrites a compelling account of epistemic value, and it helps explain why knowledge is the norm of assertion. This is not the place to review these arguments. I am convinced that they succeed and for present purposes I will assume that they do. Now suppose that the rules of cognition prescribe acquiring true beliefs. And ­suppose that it is by achieving what the rules prescribe that we sustain a practice. Thus it is by achieving true belief that we achieve the goals of cognition and thereby sustain the practice. So since achieving true belief just is knowing, it is by knowing that we sustain the practice of cognition and, in turn, inquiry. Knowing is the only way to do this. It is by knowing, and by knowing alone, that we hold the practice together. 18   Sosa (2007), Greco( 2010), Turri (2011a, 2015a). Important historical antecedents include Aristotle, Descartes, and Reid. 19   This includes so-called ‘fake barn’ cases, which many philosophers categorize as Gettier cases. For further discussion and review of relevant literature, see Turri (2011a); Turri, Buckwalter, and Blouw (2015); Blouw, Buckwalter, and Turri (2015a; 2016a).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

276  John Turri Knowledge is not the goal of belief management. The sustaining rules of belief management do not mention knowledge. But knowledge still plays a unique and essential role: it is the constitutive means and sustenance of cognition and inquiry.20 It is widely believed that there is an important normative connection between belief and knowledge.21 I have proposed one special way that knowledge sets the normative standard for belief: knowing is essentially the unique way of normatively sustaining cognition and, thereby, inquiry. Knowing is the constitutive means and sustenance of these central practices of belief management. Of course, there are other belief management practices, such as position management, where knowledge does not play this role, and I have said nothing about whether knowledge is a norm of belief itself, independently of any practice of managing beliefs. But that is consistent with knowledge playing a special and unique normative role in our practices of cognition and inquiry.22

References Alston, William P. 1991. Perceiving god: the epistemology of religious experience. Ithaca, NY: Cornell University Press. Aristotle. (350 bce/1941). Nichomachean Ethics. Trans W. D. Ross. In The basic works of Aristotle. Ed. Richard McKeon. New York: Random House. Blouw, Peter, Wesley Buckwalter, and John Turri. Forthcoming. Gettier Cases: a Taxonomy. In Rodgrigo Borges, Claudio de Almeida, and Peter Klein, eds., Explaining Knowledge: New Essays on the Gettier Problem. Oxford: Oxford University Press. David, Marian. 2005. Truth as the primary epistemic goal: a working hypothesis. In Matthias Steup and Ernest Sosa, eds., Contemporary Debates in Epistemology. Malden, MA: Blackwell, 296–311. DePaul, Michael. 2001. Value monism in epistemology. In Matthias Steup, ed., Knowledge, Truth, and Duty. Oxford: Oxford University Press, 170–183. DeRose, Keith. 2002. Assertion, knowledge, and context. Philosophical Review 111(2): 167–203. Greco, John. 2010. Achieving Knowledge: a Virtue-Theoretic Model of Epistemic Normativity. Cambridge: Cambridge University Press.

20  Again, I am open to the possibility of replacing Cognition’s Rule of Permission with a Rule of Discouragement involving ‘ought’ or ‘should’. I am even open to the possibility that the rule states only an imperfect duty, such as: For the most part, form a belief only if it is true. If the relevant deontic rule were such an imperfect duty, then we could still say that knowledge is essentially the central sustaining tendency of cognition: in order to normatively sustain the practice, the central tendency of cognition must be knowledge. Finally, I am open to the possibility that knowledge does not, strictly speaking, require truth, but only approximate truth (see Turri 2011c). 21   E.g. Williamson (2000), Sutton (2007), Huemer (2011). 22   Acknowledgments—For helpful feedback, I thank Matt Benton, Wesley Buckwalter, Tim Kenyon, Rachel McKinnon, Adam Morton, Ernest Sosa, and Angelo Turri. Special thanks go to Ben Jarvis. This research was supported by the Social Sciences and Humanities Research Council of Canada, the Association of Commonwealth Universities, the British Academy, The Character Project at Wake Forest University and the John Templeton Foundation, the National Endowment for the Humanities, and an Early Researcher Award from the Ontario Ministry of Economic Development and Innovation.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

Sustaining Rules: A Model and Application  277 Huemer, Michael. 2011. The puzzle of metacoherence. Philosophy and Phenomenological Research 82(1) (January 2011): 1–21. James, William. 1897 [1948]. The will to believe. In Alburey Castell, ed., Essays in Pragmatism. New York: Hafner Press, 88–109. MacIntyre, Alasdair. 1999. Dependent Rational Animals: Why Human Beings Need the Virtues. Chicago: Open Court. Millikan, Ruth. 2005. Language: a Biological Model. Oxford: Oxford University Press. Rawls, John. 1955. Two conceptions of rules. Philosophical Review 64: 3–32. Richerson, Peter J. and Robert Boyd. 2005. Not by Genes Alone: How Culture Transformed Human Evolution. Chicago: University of Chicago Press. Searle, John. 1969. Speech Acts. Cambridge: Cambridge University Press. Simester, A. P., ed. 2005. Appraising Strict Liability. Oxford: Oxford University Press. Sosa, Ernest. 2007. A Virtue Epistemology. Oxford: Oxford University Press. Sosa, Ernest. 2011. Knowing Full Well. Princeton: Princeton University Press. Sutton, Jonathan. 2007. Without Justification. Cambridge, MA: MIT Press. Tomasello, Michael. 2008. Origins of Human Communication. London: MIT Press. Tomasello, Michael. 2009. Why We Cooperate. London: MIT Press. Turri, John. 2011a. Manifest failure: the Gettier problem solved. Philosophers’ Imprint 11(8) (April 2011): 1–11. Turri, John. 2011b. Believing for a reason. Erkenntnis 74(3): 383–97. Turri, John. 2011c. Mythology of the factive. Logos & Episteme 2(1): 143–52. Turri, John. 2013. The test of truth: an experimental investigation of the norm of assertion. Cognition 129(2): 279–91. Turri, John. 2015a. From Virtue Epistemology to Abilism: Theoretical and Empirical Developments. In C. B. Miller, M. R. Furr, A. Knobel, and W. Fleeson, eds., Character: New Directions from Philosophy, Psychology, and Theology. Oxford: Oxford University Press, 315–30. Turri, John. 2015b. Evidence of factive norms of belief and decision. Synthese 192: 1–22. Turri, John. 2016a. Knowledge Judgments in “Gettier” Cases. In Justin Sytsma and Wesley Buckwalter, eds., A Companion to Experimental Philosophy. Oxford: Wiley-Blackwell, 337–348. Turri, John. 2016b. The radicalism of truth-insensitive epistemology: truth’s profound effect on the evaluation of belief. Philosophy and Phenomenological Research 93(2): 348–67. Turri, John and Peter Blouw. 2014. Excuse validation: a study in rule-breaking. Philosophical Studies 172: 615–43. Turri, John, Wesley Buckwalter, and Peter Blouw. 2015. Knowledge and luck. Psychonomic Bulletin and Review 22(2), 378–90. Williams, Bernard. 1978. Descartes: the Project of Pure Enquiry. New York: Routledge. Williamson, Timothy. 2000. Knowledge and its Limits. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

14 ‘More Likely Than Not’ Knowledge First and the Role of Bare Statistical Evidence in Courts of Law Michael Blome-Tillmann

1.  The Puzzle of Bare Statistical Evidence Proceedings in courts of law are governed by rules of evidence.* Criminal proceedings in the USA and the UK, for instance, are governed the by rule of Beyond Reasonable Doubt, while the evidentially far less demanding rule of Probable Cause is applied to order searches, arrests, or to issue indictments in the mentioned jurisdictions. The ­evidential standard that I am interested in in this chapter, however, governs civil procedure or tort suits—namely, the standard of the Preponderance of the Evidence.1 This standard is usually paraphrased in legal textbooks as more likely than not or as greater than 50% chance. But we can be more precise. Since the mentioned paraphrases are meant to govern evidential probabilities, we may construe them as constraints on the probability of a proposition p conditional on the evidence available to and admissible in court. Let us therefore define the rule as follows: Preponderance of the Evidence (PE): p meets the standard of proof of PE iff P(p|e) > 0.5 On this understanding, a proposition p has been proven to the standard of the preponderance of the evidence just in case the probability of p given the admissible evidence e is greater than 0.5. Thus, in order to find a defendant liable in a civil tort suit, the probability that the defendant is at fault must be greater than 0.5 given the evidence. *  I am indebted to Brian Ball, Philip Ebert, Peter Graham, Bruno Guindon, Andrew Higgins, Alicia Hinarejos, Iwao Hirose, Jesper Kallestrup, Aidan McGlynn, Duncan Pritchard, Eric Schwitzgebel, Paulina Sliwa, Levi Spectre, Sarah Stroud, and Han-Ru Zhou. Special thanks go to Arif Ahmed, Matthew Kramer, and Martin Smith for extensive comments on an earlier version of this chapter and to three anonymous referees for OUP. This work was supported by Marie Curie Actions (PIIF-GA-2012-328969 ‘Epistemic Vocabulary’). 1   See Ingram (2015) for an overview of the mentioned standards.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  279 Consider an example for illustration:2 Red Cab Case—Version A: Mr. Smith is driving home late one night when his car is sideswiped and run off the road by a taxi. In the crash, Mr. Smith suffers two broken legs. He decides to sue Red Cab Company for damages. His evidence is as follows: there are only two taxi ­companies in town, Red Cab Company (all of whose taxis are red) and Blue Cab Company (all of whose taxis are blue). An independent bystander who witnessed the accident testifies that the taxi in question was red. Given standard assumptions about the reliability of eyewitness testimony, the ­probability that the taxi was red, and thus that Red Cab Company is at fault, given her testimony, is well above the threshold of 0.5—namely, at roughly 0.7.3 Conditional on the evidence presented and admitted in court, it is therefore more likely than not that the car that caused Mr. Smith’s injuries belonged to Red Cab Company. And since the standard of proof is met in the above fictitious case, Red Cab Company would, in the example, be found liable to pay damages to Mr. Smith. The example illustrates a clear and uncontroversial application of the rule of PE. However, consider a slight variation of the above example: Red Cab Case—Version B: Mr. Smith is driving home late one night when his car is sideswiped and run off the road by a taxi. In the crash Mr. Smith suffers two broken legs. He decides to sue Red Cab Company for damages. His evidence is as follows: there are only two taxi companies in town, Red Cab Company (all of whose taxis are red) and Blue Cab Company (all of whose taxis are blue), and of the taxis in town on the night of the accident, 70% were operated by Red Cab Company and 30% by Blue Cab Company. This is the only evidence Mr. Smith can produce against Red Cab Company, for even though he could see that it was a taxi that caused the accident, he could not see its colour. No further eyewitnesses have come forward. On evaluating the evidence in Version B of the red cab case, it is again obvious that the evidential probability that Red Cab Company is at fault is well above the threshold of 0.5. Given the evidence—that is, given the fact that Red Cab Company’s market share on the night of the accident was 70%—the probability that the taxi that caused the accident was one of Red Cab Company’s is 0.7. However, despite the fact that PE—the standard of proof for civil cases—has been met in Version B of our example, courts 2   See Tribe (1971) for the original version of the example. The present formulation of the case is largely borrowed from Thomson (1986, 199–200). 3   I assume a prior probability that the defendant is at fault of 0.5, which results—according to Fields (2013, 1799)—in a posterior probability of even 0.77. As Fields points out, the probability that the defendant is at fault given positive identification falls below 0.5 only if the prior probability that the defendant is at fault is below roughly 0.3. See Schauer (2003, 317, n. 15), for further references on the reliability of eyewitness testimony.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

280  Michael Blome-Tillmann routinely find, with overwhelming consistency, for the defendant in such cases. Moreover, I take it that such decision-making accords rather well with our intuitions about the case, for most people are hesitant to impose liability on the basis of the market share evidence in Version B, citing the intuition that the evidence presented in court is insufficient for a verdict for the plaintiff.4 We thus face a puzzle. According to our intuitions, the court should find Red Cab Company liable in Version A of the Red Cab Case, but it should not do so in Version B—despite the fact that the standard of the Preponderance of the Evidence is clearly met in either case. What is more, the evidential probability that the defendant is at fault is identical in the two cases. Why, then, and on what basis, are civil courts willing to violate PE so blatantly? The problem generalizes, as the following example, inspired by Jonathan Cohen’s (1977, 74–81) Paradox of the Gatecrasher, illustrates. The Gatecrasher—Version A: The organizers of the local rodeo decide to sue John for gatecrashing their Sunday afternoon event. Their evidence is as follows: John attended the Sunday afternoon event—he was seen and photographed on the main ranks. No tickets were issued, so John cannot be expected to prove that he bought a ticket with a ticket stub. However, a local police officer observed John climbing the fence and taking a seat. The officer is willing to testify in court. In this case, just as in the previous A-version of the Red Cab Case, courts will, without much hesitation, find for the plaintiff. The police officer’s eyewitness testimony establishes, in the absence of countervailing evidence, with a high-enough probability that John gatecrashed. To illustrate this further, let us again assume that the police officer is, given the evidence presented in court, ordinarily reliable. Then, the probability, given the evidence, that John gatecrashed is 0.7—well above the threshold required by PE.5 However, there is again a very similar case in which courts would find differently. Consider what I shall call Version B of the gatecrasher example: The Gatecrasher—Version B: The organizers of the local rodeo decide to sue John for gatecrashing their Sunday afternoon event. Their evidence is as follows: John attended the Sunday afternoon event—he was seen and photographed on the main ranks during the event. No ­tickets were issued, so John cannot be expected to prove that he bought a ticket with a ticket stub. However, while 1,000 people were counted in the seats, only 300 paid for admission. In this revised version of the case, courts will clearly find for the defendant. And again, I take it that this accords well with our intuitions about fairness and justice. It does not   See Wells (1992) for empirical evidence to this effect.   See Fields (2013, 1799).

4 5

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  281 seem right to find John, who was randomly picked out in the arena, liable just because 70% of attendees at the rodeo gatecrashed. If such a case was allowed to succeed, the organizers of the rodeo could, after all, in principle win similar cases for every person in attendance at the rodeo, including the 300 people that paid the entrance fee.6 We are again facing a puzzle. What justifies courts’ finding differently in the A- and the B-version of the Gatecrasher, given that the evidential probability that the defendant is at fault is in either case well above the threshold required by the rule of the Preponderance of the Evidence? Given that all evidence is probabilistic and that the evidential probabilities are identical with respect to the A- and B-versions of our cases, courts should find for the plaintiff in the A-versions iff they do so in the B-versions. But the imposition of liability on the basis of bare statistical evidence in the B-versions is intuitively unwarranted and unjust. But, then, what is, in the light of the rule of the Preponderance of the Evidence, the decisive difference between the two different types of cases?

2.  Individual vs. Bare Statistical Evidence Cases such as the Red Cab/Blue Cab example and the Gatecrasher have puzzled ­lawyers, judges, and jurists since at least the 1940s. A number of explanations have been proposed to resolve the puzzle.7 To begin with, let us distinguish two different broad approaches to the puzzle, what I shall call the conservative and the revisionist approaches. According to the revisionists, our willingness to violate PE in the B-versions of the above cases is due to a flawed underestimation of the evidential standing of statistical evidence. On this view we are mistaken in ‘excluding morereliable statistical evidence and admitting less-reliable personal testimony, [a practice that] is explained by the widespread but empirically unsupported faith in eyewitness identification’ (Schauer 2003, 94). I shall not discuss such revisionist approaches in this chapter. Rather, I take the intuition that courts would be at fault in finding the defendant liable in the above B-cases to be far too strong a datum to be simply ignored in favour of revisionism. Instead, I shall here explore a conservative approach to the puzzle that aims to explain why courts are right in finding the way they actually do. A large number of jurists have argued along conservative lines that the puzzle of bare statistical evidence can be resolved by distinguishing between two fundamentally different types of evidence. We can find such distinctions, drawn with more or less ­rigour, in a large number of court judgments, but the distinction is also widely ­discussed in the academic literature on evidence law.8 The distinction invoked is often 6   To soften the blow, one might restrict compensation to actual losses and share liability among all attendees at the rodeo. While slightly less unfair to those individuals who paid the entrance fee, I take it that such an approach would still place unreasonable demands on fee-paying attendees. Thanks to an anonymous referee for OUP here. 7   See, for instance, Smith v Rapid Transit, Inc. 8   See Thomson (1986, 200) for references.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

282  Michael Blome-Tillmann referred to as that between individual and bare statistical evidence.9 JJ Thomson (1986), for instance, observes that: what is at work in the friends of individualized evidence is precisely the feeling that just ­imposition of liability requires that [a] stronger requirement be met. They believe [. . .] that “mathematical chances” or “quantitative probability” is not by itself enough; on my view of them, that is because they feel, rightly, that if a jury declares a defendant guilty on the ground of nonindividualized evidence alone, then it is just luck for the jury if what it declares true is true—and they feel, not without reason, that it is unjust to impose liability where that is the case  (Thomson 1986, 214).

As Thomson points out, bare statistical evidence does not seem to eliminate luck. Intuitively, if one’s judgement is based on bare statistical evidence, then one’s judgement is, if correct, correct as matter of luck. Judgements that are based on individualized evidence, such as eyewitness testimony, however, are intuitively correct not merely as a matter of luck. Individual evidence, in short, seems to eliminate an element of luck that bare statistical evidence fails to rule out. The issue can be illustrated by an analogy to lottery examples. Even though the ­evidential probability that one’s lottery ticket is a loser may be exceedingly high, one’s belief that it is a loser is, if true, true as a matter of luck: one’s belief that one’s ticket is a  loser could, after all, very easily have been false (the balls deciding the winning ­numbers could very easily have fallen in a different order, making one’s ticket a ­winner). Thus, for all one knows, one’s ticket might win the jackpot, and even though the probabilities speak strongly against such a positive result, one’s belief that one will not win is, in an intuitive sense, true as a matter of luck. What the lottery example and the B-versions of the earlier examples have in common is thus that a high evidential probability alone cannot eliminate luck. The elimination of the relevant type of luck, however, seems to be what is necessary for a proposition to meet the evidential standards that are in fact applied in civil tort suits. Before taking a closer look at how to illuminate the distinction between individual and bare statistical evidence in Section 3, note that the conservative strategy under consideration leads to a reformulation of PE. Here is Preponderance of the Individual Evidence (PEI), where ‘ei’ refers to the court’s individual evidence: Preponderance of the Individual Evidence (PEI): p meets the standard of proof in civil cases iff P(p|ei) > 0.5. While this strategy—and the distinction underlying it—is intuitively appealing, it is in need of further clarification. We do not yet have a very clear understanding of the difference between individual and bare statistical evidence. There have been several attempts to elucidate the distinction in the recent literature. Cohen (1977), for instance, 9   Some theorists and judges use the terms ‘direct evidence’ and ‘general evidence’ (see Smith v Rapid Transit, Inc.).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  283 has argued that the notion of probability at issue is not that of ordinary evidential probability, but rather that of inductive probability. JJ Thomson (1986) has offered an account of individual evidence in causal terms, and, more recently, Enoch et al. (2012) have argued that individual evidence can be accounted for in terms of modal sensitivity. Discussion of these approaches, however, is beyond the scope of this chapter.10 Instead, I shall here propose a novel approach to the distinction between individual and bare statistical evidence that aims to resolve the puzzle of bare statistical evidence by resorting to the conceptual framework of a knowledge-first epistemology.

3.  The Knowledge Account A recent revolution in epistemology—the knowledge-first movement—favours knowledge as the basic notion of epistemological theory. Knowledge is, on this view, taken as a primitive, which is then used to elucidate and explain a number of other philosophically interesting concepts and phenomena. Moreover, a number of knowledge-firsters also take the view that knowledge plays an important normative role. On such views knowledge is, for instance, the norm of practical reasoning, assertion, or belief. Consider, for illustration, the knowledge rule of assertion as proposed by Timothy Williamson (2000, 243): (TW)  It is permissible for x to assert p only if x knows p. While (TW) is the principle most widely associated with the label ‘Knowledge Rule of Assertion’, other authors—notably Hawthorne (2004, 23, n. 58) and DeRose (2002, 167, 187)—have been tempted by a logically stronger, biconditional version of Williamson’s knowledge rule:11 (A)  It is permissible for x to assert p iff x knows p. Along similar lines, Hawthorne (2004: 30) and Hawthorne and Stanley (2008) have defended the view that knowledge is the norm of practical reasoning. Here is the Knowledge Rule of Practical Reasoning: (PR)  It is permissible for x to use p as a premise in her practical reasoning iff x knows p. Drawing inspiration from Williamson (2005, 108) and McGrath (2010, 396), we may add the Knowledge Norm of Belief to the mix: (B)  It is permissible for x to outright believe p iff x knows p.12   But see Blome-Tillmann (2015b) for criticism of the mentioned views.   Cp. Brown (2010, 249–50). 12   As Williamson (2005, 108) puts it, ‘if one knows [p], then one can hardly be wrong to believe [p]; conversely, given that one does not know [p], it arguably is wrong to believe [p].’ McGrath (2010, 396) only assumes the sufficiency of knowledge for permissible belief. 10 11

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

284  Michael Blome-Tillmann Several authors have objected to one or another of these principles on different grounds.13 However, the knowledge-first approach is growing in popularity and promises hope for a slow but steady paradigm shift in epistemology. What is the relevance of the above views about the normative role of knowledge for the puzzle of bare statistical evidence? Assuming that we ought to act on p only if we know p, the ideal and strongest epistemic state a court could be in, from a normative point of view, is one in which the court knows whether the defendant is at fault. It is, for instance, only if the court knows that the defendant is at fault that the court should assert, outright believe, or use as a premise in its judicial decision making the proposition that the defendant is at fault. However, we certainly cannot require courts to reach the epistemically normative ideal state of knowledge in the sense that a judgment for the plaintiff can only be delivered, if the court knows that the defendant is at fault. Courts need to be able to make decisions in the absence of the epistemically normative ideal— that is, in the absence of knowledge that the defendant is at fault. What is needed is a measure of gradual approximation to the normative ideal. One way to measure such approximation is in terms of evidential probabilities that one knows that the defendant is at fault. Knowledge then remains the normatively ideal state, but epistemic success in courts of law can be measured and understood in terms of something less than knowledge—namely, the evidential probability that knowledge has been achieved. On the view proposed here, courts are thus to measure the evidential probability that they are in the epistemically ideal state with respect to the proposition that the defendant is at fault rather than simply the evidential probability that the defendant is at fault.14 What matters, in civil cases, is thus P(Kp|e) rather than P(p|e). The view at issue is a natural extension of the knowledge first approach in epistemology to legal theory and evidence law. Consider a second argument for the knowledge approach to evidence in judicial proceedings. In courts of law, we want to get things right, and—as we have seen ­previously—we do not just want to get things right by luck. In other words, if a court takes p to be established according the relevant rules of evidence, then, ideally, we want it to be the case that p is, if true, not just true as a matter of luck. Rather, we want to be sure, to varying degrees, that we are not lucky in getting things right, if we are getting them right. What is needed is not just evidence that we got it right, but also evidence that we are not merely lucky in getting it right. Next, note that it is a familiar fact that if one knows p, then one is not just lucky in getting it right with respect to p—knowledge precludes epistemic luck (as is familiar from the lottery cases briefly mentioned 13   See, for instance, Brown (2008, 2010), Goldman (2008), Kvanvig (2007), Lackey (2007), Neta (2009). For an alternative view to the knowledge norm of belief, see Wedgwood (2002, 2013). 14   Note also that the view proposed is not at odds with the above norms and, in particular, not with PR. To see this, note that the idea that it is permissible to rely on p in our practical reasoning iff we know p remains satisfied. The court should only rely on a proposition p if it knows p—which means, in the particular cases envisaged, that the court should rely on propositions about evidential probabilities iff it knows those propositions. These constraints are satisfied in proceedings that are not in violation of due process.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  285 above).15 So, what is needed in court is evidence that one knows that the defendant is at fault and not just evidence that the defendant is at fault.16 Let us be more precise about the view proposed here. There are at least two different ways to implement the main idea underlying the above considerations. On the first approach, we replace the traditional rule of the Preponderance of the Evidence with what I shall call the Knowledge Rule of the Preponderance of the Evidence (KPE): Knowledge Rule of the Preponderance of the Individual Evidence (KPE): p meets the standard of proof in civil cases iff P(Kp|e) > 0.5 According to KPE, the standard of proof is met in a civil case iff the evidential probability that the court knows that the defendant is at fault is greater than 0.5. This view is revisionary to some degree, as it replaces PE with KPE, but it is—as we shall see below—in harmony with our intuitions about the just imposition of liability in courts. A second, slightly different approach leaves the original PE in place, but formulates a novel constraint on what type of evidence is admissible in court. Consider the following principle: Evidence Constraint (EC): e is admissible in court as evidence for p only if it raises the probability that someone in possession of e knows p. Evidence Constraint allows us to achieve the same results as KPE by disqualifying what we have previously called ‘bare statistical evidence’ from civil proceedings—­ without amending the familiar rule of the Preponderance of the Evidence PE. As far as I can see, both of these approaches are promising. Next, consider how the above approaches resolve the puzzle of bare statistical ­evidence. Consider first the Gatecrasher cases. In the A-version of the example, the probability that the court knows that John is at fault is reasonably high given the court’s evidence. Given that the average reliability of witnesses in court, the probability that the witness in this case has correctly identified a gatecrasher is, given the court’s evidence, roughly 0.7.17 Since there is no countervailing evidence to the effect that the court has been led astray, the probability that the court knows that John is at fault is high: something unusual would have to have happened for the court not to know that   See also Pritchard (2005) for the connection between knowledge and luck.  Cp. Day v Boston & Maine Railroad, 96 Me. at 207, 52 A. at 774, quoted in Thomson, Liability and Individualized Evidence, 234, n. 13: ‘Quantitative probability, however, is only the greater chance. It is not proof, nor even probative evidence, of the proposition to be proved. That in one throw of dice there is a quantitative probability, or greater chance, that a less number of spots than six will fall is no evidence whatever that in a given throw such was the actual result. Without something more, the actual result of the throw would still be utterly unknown. The slightest real evidence that sixes did in fact fall uppermost would outweigh all the probability otherwise’. As Thomson (1986, 206, n. 14) points out, the passage is somewhat obscure, as it suggests that statistical evidence does not probabilify. However, it also nicely brings to light the legal relevance of the intuition that knowledge is incompatible with epistemic luck. 17   See Fields (2013, 1799). 15 16

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

286  Michael Blome-Tillmann the defendant is at fault.18 Of course, the court fails to know that they know that the defendant is at fault, but the evidential probability that the court knows is, given the court’s evidence, fairly high. Therefore, the court ought to find for the plaintiff in the A-version of the Gatecrasher. In the B-version of the Gatecrasher, however, things are rather different. In this case, the court’s belief that John gatecrashed is, if true, true as a matter of luck. For all the court knows, John could have easily been one of the 300 attendees at the rodeo who paid the entrance fee and there is no evidence whatsoever to eliminate that hypothesis. Thus, the court knows (or is at least in a position to know) that it does not know that John is at fault. In other words, given the evidence available to the court, the probability that the court knows that John is at fault is zero—hence the strict and categorical difference to our intuitions in the A-version of the Gatecrasher. Imposing liability in the B-version of the case would be gravely inappropriate, and the strength of that intuition is accounted for by the fact that the probability that the court knows that the defendant is at fault is zero in the B-version of the Gatecrasher. Next, consider the Red Cab cases. In the A-version of the example, the probability that the court knows that Red Cab Company is at fault is reasonably high, given the evidence. Given the average reliability of the type of witness at issue, the probability that the witness has correctly identified the colour of the taxi is, given the court’s ­evidence, 0.7. Since there are no reasons to assume that the court has been led astray, the probability that the court knows that the taxi was red is high: something unusual would have had to happen for the court not to have such knowledge. The probability that nothing unusual has happened, however, is measured by the reliability figures of witnesses in court. Thus, the evidential probability that the court knows that the defendant is at fault is roughly 0.7 and thus above the threshold required by the ­knowledge rule of the preponderance of the evidence.19 In the B-version of the case, however, things are again rather different. To see this, note again that, in the B-version of the Red Cab case, the court knows (or is in a ­position to know) that it does not know that the defendant is at fault. Thus, the probability that the court knows that the defendant is at fault given the court’s evidence is zero— analogously to the B-version of the Gatecrasher example. Again, the knowledge account explains rather elegantly our intuition that it would be grossly unjust for the court to find for the plaintiff in the B-version of the Red Cab case. In summary, the knowledge approach to individual evidence does not only cohere rather well with a knowledge-first approach in epistemology, but it also explains our reluctance to impose liability in cases involving bare statistical evidence. Individual evidence, after all, is crucially not more potent in virtue of being more highly 18   The probability that nothing unusual has happened is, in the mentioned cases, measured by the r­ eliability figures of witness testimony in court. Thus, the evidential probability that nothing unusual has happened and that the defendant is in fact at fault, as testified by the witness, is roughly 0.7. Consequently, the probability that the court knows is roughly 0.7. 19   See Fields (2013, 1799).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  287 ­ robabilifying than bare statistical evidence. Rather, individual evidence is perceived p to be more potent in courts of law because it has the power to raise the probability that one knows. Evidential probability, in the broad sense, that the defendant is at fault is therefore not really what matters in civil cases. Rather, it is the evidential probability that the court knows the defendant to be at fault. It is the latter type of probability that tracks our intuitions concerning the just imposition of liability.

4.  Some Objections 4.1  Knowledge and safety It is worthwhile illustrating the above considerations further by means of the Safety principle. Why is the evidential probability that the court knows that the defendant is at fault zero in the B-versions of the above cases? To illustrate this further, consider Timothy Williamson’s (2000, 147) preferred formulation of Safety: (WS)  If one knows, one could not easily have been wrong in a similar case.20 WS is independently motivated, but I shall not go into the details of the literature on safety in this chapter.21 Instead, note that WS allows us to explain why P(Kp|e) = 0 in the B-versions of the above cases. To see this, note that in those cases the court’s belief that the defendant is at fault, if it forms such a belief, is clearly not safe. For the court could, in the B-versions of our examples, easily have been wrong in a similar case—namely, in all those similar cases in which the taxi at issue was blue (in the Red Cab example) or in which John paid the entrance fee to the rodeo (in the Gatecrasher example). Those possibilities are very similar to the ones described in the relevant B-versions, but it is important to point out that the court is mistaken in believing, in those rather similar cases, that the defendant is at fault. Thus, WS explains elegantly why the court does not know that the defendant is at fault in the B-versions. What is more, WS also allows us to account for the fact that the court knows (or is in a position to know) that they do not know that the defendant is at fault. That is so because it is rather obvious that the court could very easily have been wrong in a similar case. It is obvious, in other words, that the court’s belief that the defendant is at fault, if it has formed such a belief, is not safe. But if the court knows that they do not know that the defendant is at fault because their corresponding belief is not safe, then the proposition that they do not know is part of the court’s evidence, and P(Kp|e) must, therefore, equal zero.22 In summary, WS explains why courts cannot acquire knowledge 20   Williamson (2009, 23) illustrates this view further by arguing that ‘knowing p requires safety from the falsity of p and of its epistemic counterparts.’ 21   But see Pritchard (2005, 2009), and Williamson (2000, 2009). 22   I here assume that if one knows p, then p is part of one’s evidence. This principle is entailed by, but not  equivalent, to Timothy Williamson’s (2000) thesis that E=K. See also Blome-Tillmann (2015a) for discussion.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

288  Michael Blome-Tillmann on the basis of bare statistical evidence, and thus helps motivating the view that P(Kp|e) = 0. Thus, WS provides us, in conjunction with KPE or EC, with a good ­explanation of why we cannot use bare statistical evidence in courts of law.23 It might be objected at this point that it would be methodologically simpler to resolve the puzzle of bare statistical evidence by appeal to safety rather than knowledge.24 Is it not so that we should measure the evidential probability that the court’s belief is safe, rather than the evidential probability that it knows when contemplating the just imposition of liability? Nothing much hangs on the answer to this question. However, I take it that, for independent (and largely methodological) reasons, our only grasp of the notion of safety is via the notion of knowledge: safety, or the notion of similarity between cases employed in WS, cannot be defined in terms other than knowledge.25 It  is, therefore, knowledge—not safety—that comes explanatorily first. On such an approach, what is really doing the explanatory work is the notion of knowledge—even if we were to reformulate the rule of the Preponderance of the Evidence in terms of Safety. Thus, on the knowledge-first view, the methodologically simpler approach is to measure the probability that the court knows rather than the probability that the court’s beliefs are safe. It is knowledge—not safety—that is the ultimate anti-luck ­condition in epistemology.26

4.2  The role of knowledge in legal theorizing It might be argued that knowledge is a confusing or somewhat slippery notion and that it therefore ought not to play a role in courts of law or our theories of evidence law.27 But it is important to note that the defenders of such a view still owe us an account of evidence—a prima facie equally confusing and slippery notion. They must, in other words, provide us with an answer to the question: under what conditions is a proposition p part of the court’s evidence? One simple and explanatorily powerful response to this challenge derives from Timothy Williamson’s view that E=K—that is, from the view that all and only those propositions that one knows are part of one’s evidence. On such a view, a proposition p is part of the court’s evidence iff the court knows p. Now, if we accept that E=K, then we are stuck with the notion of knowledge in our theorizing about evidence law independently of our acceptance of KPE or EC. If we reject E=K, on the other hand, then we must offer an alternative account of evidence—an account that is not only independently well motivated (as E=K is), but is also workable within judicial contexts. While it surely is not impossible to develop such a theory of evidence, rejecting a knowledge-based account of evidence in legal contexts on the worry that 23   In the A-versions of our cases, on the other hand, the evidence can be interpreted as raising the probability that the court’s relevant beliefs are safe, as outlined for knowledge in the previous section. 24   See Pritchard MS for an account of legal risk in terms of safety. 25   Cp. Blome-Tillmann (2014, Ch. 5). 26   On the view proposed here, safety is eventually nothing but a heuristic device that illustrates a property of knowledge—a property that cannot be understood or defined independently of knowledge. 27   Enoch et al. (2012) seem to be attracted by such a view.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  289 the notion of knowledge may appear somewhat slippery will seem to many to amount to throwing the baby out with the bathwater.

4.3  Negligence, racial discrimination, and statistical evidence as knowledge-generating Sometimes, but only sometimes, we can know on the basis of bare statistical evidence. Some seemingly bare statistical evidence leads to knowledge.28 Thomson (1986, 205) points out, for instance, that in racial discrimination suits, evidence about the racial composition of both an organization’s work force and the local population may be admissible and lead to a verdict for the plaintiff. It is noteworthy, however, that in such cases we also intuit that—if the statistical evidence is strong enough—we know that the organization discriminated racially (or at least that the probability that we know given our evidence is significantly higher than 0.5).29 Thus, as long as our intuitions about whether we know correlate with our intuitions about whether statistical evidence is sufficient for the just imposition of liability, the mere fact that statistical evidence is sometimes sufficient for the just imposition of liability does not threaten the view defended here.30 Further cases in which bare statistical evidence appears sufficient for the just ­imposition of liability comprise cases of so-called market share liability.31 In certain pharmaceutical or tobacco litigation, liability is assigned according to the market share a particular company had—even if the company’s market share is below the threshold of 50% and thus reduces the evidential probability that the company is responsible for the plaintiff ’s impairments to below 0.5. However, it is important to emphasize that, in successful cases of market share liability, the negligence of the defendant companies has usually been established independently. Thus, liability is assigned according to market share only once the defendant has been independently shown, by the standard of the preponderance of the evidence, to have acted negligently. Consequently, cases of market share liability provide harmless exceptions to the principles of KPE and EC discussed earlier. The relevant examples can easily be accounted for by restricting KPE and EC to cases in which negligence has not been established independently.

4.4  Causation as an epistemic proxy In an important paper on the puzzle of bare statistical evidence JJ Thomson (1986) argues that individual evidence ought to be understood in causal terms. According to Thomson,   See Blome-Tillmann (2014, Ch. 5) for discussion.   Thomson’s view on the matter is that the statistical evidence is in this case causally explained by the fact that there is actual racial discrimination. 30   Further cases of the mentioned type are, presumably, some cases involving DNA evidence. I shall leave the treatment of issues arising from the admissibility of DNA evidence in courts of law for another occasion. 31  See Sindell v Abbott Laboratories (1980) 26 C3d 588. 28 29

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

290  Michael Blome-Tillmann (JJT) e is individual evidence in support of p iff the fact that p caused e. I have argued elsewhere against Thomson’s view, but its usefulness must be emphasized here.32 To see what I have in mind, note the following connection between knowledge and causation: (KC) If x knows p, then x’s belief that p was caused by the fact that p. KC is true with respect to most, if not all, cases of empirical knowledge. If KC is true, however, then we can use causation as an indicator of knowledge. In other words, from a practical point of view, courts may aim to determine whether the above causal ­constraint is met with sufficient evidential probability, instead of aiming to settle the potentially harder question of whether KPE is satisfied. While the theoretical explanation of the puzzle must rest on the knowledge account outlined above, the practical endeavour of decision making in courts of law can safely be guided by the causal account outlined by Thomson. While causation by itself is not sufficient to eliminate epistemic luck, it is, in ordinary cases (including the cases discussed in Section 1 of this chapter), a reliable indicator of its elimination and thus of knowledge.

5. Conclusion Knowledge matters. We value knowledge that p higher than the mere high evidential probability that p. Moreover, in many cases of legal relevance—such as the ones discussed in this chapter—we value a high evidential probability that we know that p significantly more highly than the mere high evidential probability that p. The actual behaviour of courts and their dismissal of bare statistical evidence is a manifestation of this fact: our intuitions about the just imposition of liability in civil suits can be accounted for in a rather straightforward manner within a knowledge first approach to legal standards of proof. It is, therefore, knowledge, along with our appreciation of the likelihood that we have it, that guides our actions in courts.

References Blome-Tillmann, M. 2014. Knowledge and Presuppositions. Oxford: Oxford University Press. Blome-Tillmann, M. 2015a. Solving the Moorean Puzzle. Philosophical Studies 172(2): 493–514. Blome-Tillmann, M. 2015b. Sensitivity, Causality, and Statistical Evidence in Courts of Law. Thought 4(2): 102–12. Brown, J. 2008. The Knowledge Norm of Assertion. Philosophical Issues 18: 89–103. 32   One obvious problem with JJT is that causation is factive. Thus, given JJT a judgment meets the s­ tandard of proof of the preponderance of the evidence only if the defendant is in fact at fault. This seems an unintuitive and problematic consequence of JJT. For a more detailed discussion of Thomson’s view, see Blome-Tillmann (2015b).

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

‘More Likely Than Not’  291 Brown, J. 2010. Knowledge and Assertion. Philosophy and Phenomenological Research 81(3): 549–66. Cohen, L. J. 1977. The Probable and the Provable. Oxford: Clarendon Press. DeRose, K. 2002. Assertion, Knowledge, and Context. The Philosophical Review 111(2): 167–203. Enoch, D., L. Spectre, and T. Fisher. 2012. Statistical Evidence, Sensitivity, and the Legal Value of Knowledge. Philosophy and Public Affairs 40(3): 197–224. Fields, K. L. 2013. Toward a Bayesian Analysis of Recanted Eyewitness Identification Testimony. New York University Law Review 88(1769): 1769–801. Goldman, A. 2008. Knowledge, Explanation, and Lotteries. Noûs 42(3): 466–81. Hawthorne, J. 2004. Knowledge and Lotteries. Oxford: Oxford University Press. Hawthorne, J. and J. Stanley. 2008. Knowledge and Action. Journal of Philosophy 105(10): 571–90. Ingram, J. L. 2015. Criminal Evidence. Waltham, MA: Anderson Publishing. Kvanvig, J. L. 2007. Contextualism, Contrastivism, Relevant Alternatives, and Closure. Philosophical Studies 134: 131–40. Lackey, J. 2007. Norms of Assertion. Noûs 41(4): 594–626. McGrath, M. 2010. Contextualism and Intellectualism. Philosophical Perspectives 24(1): 383–405. Neta, R. 2009. Treating Something as a Reason for Action. Noûs 43(4): 684–99. Pritchard, D. 2005. Epistemic Luck. Oxford: Clarendon. Pritchard, D. 2009. Safety-Based Epistemology: Whither Now? Journal of Philosophical Research 34: 33–45. Pritchard, D. MS. Risk. Unpublished manuscript. Schauer, F. 2003. Profiles, Probabilities, and Stereotypes. Cambridge, MA: Harvard University Press. Thomson, J. J. 1986. Liability and Individualized Evidence. Law and Contemporary Problems 49(3): 199–219. Tribe, L. H. 1971. Trial by mathematics: precision and ritual in the legal process. Harvard Law Review 94(6): 1329–93. Wedgwood, R. 2002. The Aim of Belief. Noûs 36: 267–97. Wedgwood, R. 2013. The Right Thing to Believe. In T. Chan, ed., The Aim of Belief. Oxford: Oxford University Press: 123–39. Wells, G. L. 1992. Naked Statistical Evidence of Liability: Is Subjective Probability Enough? Journal of Personality and Social Psychology 62(5): 739–52. Williamson, T. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Williamson, T. 2005. Knowledge, Context and the Agent’s Point of View. In G. Preyer and G. Peter, eds., Contextualism in Philosophy: On Epistemology, Language and Truth. Oxford: Oxford University Press: 91–114. Williamson, T. 2009. Probability and Danger. The Amherst Lecture in Philosophy 4: 1–35.

OUP CORRECTED PROOF – FINAL, 10/05/2017, SPi

OUP CORRECTED PROOF – FINAL, 10/12/2017, SPi

Index ability  26–8, 38, 42, 42 n. 43, 56, 73, 75–6, 82, 88 n. 26, 91, 176 n. 17, 179, 192, 194–5, 203, 205, 224, 224 n. 4, 225, 229, 231–43, 260, 263–4, 266, 288 n. 23 action  1, 3–4, 8, 11, 20 n. 6, 25, 46, 53, 55 n. 12, 57, 60, 61–2, 64, 72, 72 n. 2, 77, 85–7, 103, 107, 109–11, 121, 123, 128, 134, 136, 144, 146–7, 147 n. 38, 154, 155, 155 n. 58, 163, 163 n. 2, 164–6, 167 nn. 7&8, 168–72, 172 n. 14, 173, 173 n. 16, 174–6, 176 n. 17, 177, 178, 179, 179–80, 188–90, 196, 198, 236 n. 16, 246, 248–9, 251, 255–7, 259, 261–2, 264, 269–71, 275, 290 Anscombe, Elizabeth  163–4 analysis conceptual  48, 52, 64, 66, 122, 123 n. 22, 126, 129, 133, 134 n. 5, 183 linguistic 64 metaphysical  191, 194, 197 anti-luminosity see luminosity Aristotle  264 n. 8, 275 n. 18 assertion  1 n. 2, 3–4, 5, 5 nn. 19&21, 6, 6 n. 22, 8–9, 9 n. 27, 10–11, 42 n. 44, 46–7, 50, 53–5, 55 nn. 12&13, 56–62, 64, 68, 72 n. 2, 102, 128, 132–7, 142, 142 n. 25, 143, 143 n. 28, 144, 144 n. 29, 145–6, 146 nn. 33&34, 147–55, 155 nn. 59&59, 156, 156 nn, 162–4, 189, 260 n. 1, 263, 275, 283 Ayer, A. J.  114, 170 n. 11 barn facades see fake barns basing relation  202, 202 n. 9, 209 belief  1, 1 nn. 1&2, 2–4, 4 n. 14, 5–6, 6 n. 22, 10–12, 12 n. 35, 13, 13 n. 38, 19–20, 20 nn. 5–8, 21–6, 26 nn. 18&19, 27–33, 33 n. 31, 34, 36–38, 40–1, 41 n. 41, 42, 42 nn. 44&45, 48–9, 50 n. 5, 51–2, 54–6, 59–60, 63–5, 67–8, 72, 72 nn. 1&2, 73–6, 76 n. 8, 77–9, 79 n. 11, 80, 80 n. 12, 81, 81 n. 13, 82, 82 n. 15, 83, 83 nn. 16&17, 84, 84 nn. 18&19, 85, 85 n. 21, 86–8, 88 n. 26, 89, 89 n. 28, 90–1, 91 n. 31, 92, 95–9, 99 n. 4, 100–4, 106, 108–10, 113–15, 115 n. 4, 116–19, 119 n. 11, 120–2, 122 n. 20, 123, 123 n. 22, 124–5, 128–9, 132–3, 133 n. 2, 134–8, 138 nn. 15&16, 139–40, 140 n. 22, 141–2, 145–6, 146 nn. 33&35, 147–50, 152 n. 54, 153–4, 154 n. 57, 155, 155 nn. 58&59, 156, 163–4, 164 n. 3, 165, 165 nn. 4–6, 166–9, 169 n. 11, 170,170 nn. 11&12, 171–2, 172 n. 14, 173,

173 n. 16, 174–7, 177 n. 19, 178, 179 n. 20, 182–3, 185–6, 189 n. 14, 190, 192, 192 nn. 22&24, 193, 195 n. 31, 196–7, 198 n. 38, 201, 201 n. 6, 202–3, 203 n. 14, 204, 205–6, 206 nn. 19&20, 207–8, 208 n. 26, 209–10, 210 n. 32, 211, 211 nn. 33&34, 212, 212 n. 35, 213–17, 217 n. 42, 223, 223 n. 3, 224–5, 225 n. 6, 226–7, 229–31, 231 n. 13, 237, 237 nn. 17–19, 238, 238–41, 241 n. 20, 242, 242 n. 21, 243, 246, 248, 249 n. 7, 250–3, 256–7, 272–3, 273 n. 16, 274–6, 276 n. 20, 282–3, 283 n. 12, 284 n. 13, 286–8, 288 n. 23, 290 Blome-Tillmann, Michael  13, 13 n. 39, 278–91 Burge, Tyler  2, 2 n. 6, 29 n. 24, 30 n. 24, 46 n. 1, 55, 62, 73, 74 n. 6, 98–101, 173 Chisholm, Roderick  114, 203 n. 12, 207 n. 22, 247 n. 3 Conee, Earl see evidentialism criminal liability see liability disjunctivism  30 n. 24, 104–6, 108, 109, 140 n. 20, 196, 214 n. 38 evidence  1, 5, 9, 10, 13, 19 nn. 2&3, 20, 20 nn. 4, 5&7, 21, 21 n. 9, 22, 22 n. 10, 23, 23 n. 13, 26 n. 19, 27, 27 n. 19, 34, 40, 42, 48, 48 n. 2, 50–1, 54–5, 55 n. 13, 58–60, 64–5, 73–4, 74 n. 6, 75–6, 76 n. 8, 77–81, 81 nn. 13&14, 82, 83 n. 16, 85–6, 88–9, 89 n. 28, 90, 90 n. 28, 91, 91 n. 31, 115–16, 122–3, 123 n. 22, 128, 133, 135, 137–8, 145, 148, 165, 167, 172–3, 176 n. 17, 179, 189 n. 14, 192 n. 23, 193, 202, 208, 208 n. 26, 209–11, 211 n. 34, 212, 214–15, 217–18, 230, 267, 267 n. 12, 268–72, 278–80, 280 n. 4, 281, 282, 282 n. 9, 283–5, 285 n. 16, 286–8, 287 n. 22, 288, 288 n. 23, 289, 289 nn. 29&30, 290 evidentialism  12 n. 34, 19 n. 2, 21 n. 9, 22 n. 12, 202, 206 n. 21, 207 n. 24, 208, 208 nn. 26&27, 209, 209 n. 28, 211, 213, 213 n. 36, 214, 215, 215 n. 40, 247 factive mental state operator (FMSO)  2 n. 5, 117–18, 216 fake barns  31–3, 115, 204, 204 n. 17, 205, 227–8, 228 nn. 8&9, 229, 239–42, 242 n. 21, 243, 275 n. 19 Feldman, Richard see evidentialism

OUP CORRECTED PROOF – FINAL, 10/12/2017, SPi

294  index Gerken, Mikkel  9, 46–71, 74, 145 n. 30, 148, 148 n. 41 Gettier  11, 11 n. 31, 20, 27, 51, 59, 61 n. 20, 73, 114, 114 n. 2, 115–17, 122, 125, 129 n. 26, 133, 133 n. 2, 137, 137 n. 13, 139, 145, 148–50, 155, 172 n. 14, 182, 192, 192 n. 22, 203, 203 n. 12, 204, 211 n. 34, 223–7, 228 n. 8, 229, 239–41, 243, 253 n. 14, 275, 275 n. 18 Gettier cases see Gettier Gettier problem see Gettier Goldman, Alvin I.  62, 75 n. 7, 115 nn. 3&4, 133 n. 3, 171 n. 13, 182, 188, 193, 195 n. 31 204 n. 17, 206, 209, 227, 227 n. 7, 284 n. 13 grounding, metaphysical  21, 21 n. 9, 23, 37, 37 n. 36, 38, 117, 119, 144, 183–4, 185 n. 6, 187, 195–6, 196 n. 36, 198, 218, 233 n. 15, 247, 275 Hawthorne, John  9 n. 27, 10 n. 30, 19 n. 1, 46, 53, 53 n. 8, 57, 58 n. 16, 59, 60 n. 18, 64, 66 n. 30, 128, 132 n. 1, 142 n. 25, 147 n. 36, 170 n. 11, 177, 179 n. 21, 283 Ichikawa, Jonathan Jenkins  1 n. 1, 7 n. 25, 10, 58 n. 16, 74, 113–31, 183 illusion  102, 104–5, 138, 165 n. 6, 184–5, 187, 188, 195 n. 34 Jenkins, C. S. I  1 n. 1, 10, 74, 113–31, 183 JTB  1, 4, 7, 73, 114–15, 115 n. 4, 122, 123 n. 22, 125, 128, 133, 133 n. 2, 137, 140, 146–7, 182, 192, 192 n. 22, 193, 223, 225–6, 253 JTB analysis of knowledge see JTB justification  12, 19 n. 1, 20 nn. 6&7, 21, 22 n. 10, 23, 40–2, 42 n. 43, 46–7, 49, 49 n. 4, 50, 55–6, 61 n. 20, 64, 66, 113, 114 n. 2, 115 n. 4, 116, 119–20, 122, 124, 127–8, 133, 135, 137, 141, 147 n. 39, 155, 176, 183–6, 189 n. 14, 192, 192 n. 23, 195 n. 31, 197, 201–2, 206, 206 n. 21, 207–8, 208 n. 27, 209–10, 210 nn. 31&32, 211, 211 n. 33, 212, 212 n. 35, 213–15, 215 n. 40, 217–18, 230, 240, 246–7, 247 n. 3, 248–9, 249 n. 8, 250–4, 254 n. 17, 255–7 Kallestrup, Jesper  12, 12 n. 33, 200–22 Kelp, Christoph  3 n. 13, 12, 12 n. 35, 50, 141 n. 24, 223–45 liability  251, 270, 280–1, 281 n. 6, 282, 285, 285 n. 16, 286, 286 n. 18, 287–90 liability in courts see liability Littlejohn, Clayton  9, 12 n. 36, 13 n. 38, 19–45, 49, 59, 132 n. 1, 147 nn. 36&39, 148 n. 43, 151 n. 51, 214 n. 38, 223 n. 2, 250 Logue, Heather  11, 19 n. 1, 29 n. 22, 140 n. 20, 182–99

luck  4, 31–3, 33 nn. 30–1, 34, 64, 67, 86, 87, 115, 138, 138 n. 16, 182, 203–4, 205 n. 18, 206 n. 19, 219 n. 47, 225–7, 235, 260, 264, 266, 282, 284, 285 nn. 15&16, 286, 288, 290 luminosity  5, 15, 67–8, 112, 139, 140, 177 n. 18, 181, 197, 245, 260, 282 McDowell, John  19 n. 1, 20 n. 8, 23 n. 13, 27 n. 20, 29, 29 nn. 22–4, 30, 30 n. 26, 31, 31 n. 27, 32, 32 n. 28, 36, 36 nn. 34&35, 38 n. 39, 102, 104, 140 n. 20, 184, 214 n. 38 McGlynn, Aidan  9, 9 n. 28, 10, 20 n. 4, 50, 51 n. 7, 52, 53–4, 59, 61, 65, 72–94, 97, 132 n. 1, 134 n. 8, 142 n. 25, 145 n. 30, 148 n. 43, 218 n. 46 memory see remembering Meylan, Anne  12, 246–58 non-factive attitudes and states  2, 22, 27 n. 19, 46, 55, 55 n. 13, 72, 125, 169 n. 10, 215, 215 n. 40, 216, 250, 267 n. 12 norms  1 n. 2, 5–6, 11, 42, 46, 49, 53–5, 55 n. 13, 56–7, 59, 59 n. 17, 60–2, 64, 66, 68, 72 n. 2, 128, 144, 146, 146 n. 34, 147, 147 nn. 38&39, 151–6, 176–7, 177 n. 19, 263 n. 7, 284 n. 14 normativity  8, 12, 143–4, 150 n. 46, 177, 224–5, 246–9, 249 n. 8, 250–2, 254, 256–7, 260 n. 1 ontic vagueness see vagueness perception  8, 11–12, 21, 28–9, 30 n. 24, 32, 35–7, 40, 42, 57 n. 15, 86, 104–5, 140, 174, 178–9, 179 n. 20, 182–91, 194–8, 250, 275 perceptual illusion see illusion Plato 114 practical reasoning  11, 137, 164, 165 n. 6, 166, 173–4, 176–8, 217, 283, 284 n. 14 Pritchard, Duncan  4 nn. 14&15, 12, 12 n. 33, 20 n. 5, 29 n. 23, 32 n. 28, 34 n. 32, 38 n. 39, 109, 184, 200–22, 223 n. 1, 224 n. 4, 225, 225 n. 6, 227, 228 n. 9, 285 n. 15, 287 n. 21, 288 n. 24 Putnam, Hilary  2, 2 n. 6, 73, 97–8, 100, 173 reasoning, practical see practical reasoning relation, basing see basing relation reliabilism  210–11, 247 remembering  57, 99, 135, 197, 211 safety  34, 65, 65 n. 29, 67 n. 32, 109–10, 134 n. 5, 206 n. 20, 228 nn. 9&10, 243 n. 22, 287, 287 n. 20, 288, 288 nn. 24&26 Schechter, Joshua  10, 132–58 Smith, Martin  10, 73 nn. 3&4, 95–131, 197 n. 37

OUP CORRECTED PROOF – FINAL, 10/12/2017, SPi

index  295 Sosa, Ernest  2 n. 8, 7 n. 23, 12 nn. 32, 21 n. 9, 32, 57 n. 14, 62, 109, 156 n. 64, 163 n. 2, 172 n. 14, 201–3, 203 nn. 13–15, 204, 204 n. 16, 205, 206, 206 nn. 19&20, 217, 223 n. 1, 224, 224 nn. 4&5, 225 n. 6, 228 n. 9, 229 n. 11, 232 n. 14, 235, 260 nn. 1&5, 275 n. 18 Stanley, Jason  3 nn. 11&12, 9 n. 27, 10 n. 30, 46, 53, 53 n. 8, 57, 58 n. 16, 59, 64, 128, 132, 142 n. 25, 147, 177, 283 testimony  4, 49 n. 4, 57 n. 15, 69, 104, 140, 140 n. 22, 189–90, 279, 279 n. 3, 280–2, 286 n. 18 tripartite analysis of knowledge see JTB Turri, John  4 n. 15, 13, 22 n. 12, 32, 32 n. 28, 50, 53, 58 n. 16, 59, 142 n. 25, 201 n. 3, 206 n. 19, 223 n. 1, 224 n. 4, 259–77 vagueness  194, 195, 195 n. 32 virtue epistemology  7 n. 23, 12, 12 n. 32, 202–4, 206, 219, 223, 223 nn. 1&3, 224, 230 n. 12, 231, 238, 241 n. 20, 243, 243 n. 22 warrant  12, 27, 46–7, 49, 54, 55, 55 n. 12, 56–7, 59–64, 75, 201, 201 nn. 5&6, 202–3, 203 n. 15,

204, 204 n. 16, 206, 211 n. 33, 217, 254, 270, 281 Williamson, Timothy  1, 1 n. 3, 2, 2 nn. 5, 7 & 9, 3, 5, 5 n. 17, 6, 6 n. 22, 7, 7 n. 26, 8, 9, 9 n. 27, 10 n. 30, 11–12, 19, 20 n. 5, 22 n. 12, 28 n. 21, 42 nn. 43&45, 46–8, 48 n. 3, 49, 49 n. 4, 50, 50 n. 5, 51–4, 57, 58 n. 16, 60, 60 n. 18, 62–3, 63 n. 22, 64, 65 n. 29, 67, 67 n. 33, 68 n. 38, 72, 72 n. 2, 73, 73 n. 4, 75–6, 79, 95–6, 96 n. 3, 97, 105–6, 109, 113, 116, 116 nn. 6&7, 117–18, 119 nn. 11&12, 121, 121 n. 16, 122, 122 nn. 19&20, 123 nn. 21–3, 124, 125, 128, 129 n. 26, 132, 132 n. 1, 133, 133 n. 4, 134 nn. 5, 6&8, 136 n. 11, 139 n. 19, 142, 142 n. 25, 143 nn. 26&27, 144, 146 n. 33, 147 n. 36, 151, 151 nn. 50&51, 152, 152 n. 53, 153, 153 n. 56, 154, 154 n. 57, 155, 156 nn. 61&66, 163, 163 n. 2, 163–80, 177 nn. 18&19, 178, 182, 182 n. 3, 183–4, 189 n. 14, 192 nn. 25&27, 194 n. 29, 196 n. 36, 197–8, 202, 206 n. 20, 214–15, 215 n. 39, 216, 216 n. 41, 217, 217 n. 42, 218–19, 223 n. 2, 230, 230 n. 12, 231, 250, 251–2, 252 n. 13, 261 n. 6, 271 n. 14, 283, 283 n. 12, 287, 287 nn. 20–2, 288