Instrumental Rationality: The Normativity of Means-Ends Coherence [Hardcover ed.] 0198746938, 9780198746935

Citation preview

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

Instrumental Rationality

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

Instrumental Rationality The Normativity of Means–Ends Coherence J O H N B RU N E R O

1

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

1 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 © John Brunero 2020 The moral rights of the author have been asserted First Edition published in 2020 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2019954576 ISBN 978–0–19–874693–5 Printed and bound in Great Britain by Clays Ltd, Elcograf S.p.A. Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

To my parents, John and Angela

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

Table of Contents Acknowledgements

ix

1. Introduction

1

1.1 1.2 1.3 1.4 1.5

The Roadmap Kant on Hypothetical Imperatives Contemporary Interest Formulating Means–Ends Coherence: The Instrumental Belief Conclusion

2. Bootstrapping

2.1 Clarifying the Question 2.2 The Bootstrapping Objection

2.2.1 What Is the Bootstrapping Objection? 2.2.2 Bootstrapping as Incoherent Advice

6 8 15 19 24

25

27 32

33 38

2.3 Rational Deliberation and Tie-Breaking Reasons

46

2.4 Promotion and Insufficient Means 2.5 Conclusion

55 58

2.3.1 Rational Deliberation and Underdetermination 2.3.2 From Rationality to Reasons

3. Scope

3.1 Rational Requirements 3.2 The Case for Wide-Scoping

3.2.1 The Argument 3.2.2 Challenges to the Argument

46 51

59

61 68

69 71

3.3 The Case for Narrow-Scoping

76

3.4 Conclusion

88

3.3.1 The Means–Ends Asymmetry 3.3.2 Escape and Compliance

79 82

4. Normativity I

90

4.2.1 4.2.2 4.2.3 4.2.4 4.2.5

95 97 101 105 108

4.1 Normative Disjunctivism 4.2 Is Normative Disjunctivism True?

“Ought” and Ability Advantageous Incoherence Practical and Theoretical Reason Mere Permissibility Normative Conjunctive Disjunctivism

4.3 Does It Support the Myth Theory? 4.4 Conclusion

93 95

111 117

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

viii  Table of Contents

5. Normativity II

5.1 Reasons to be Rational

5.1.1 Evaluating the Prichard Analogy 5.1.2 The Costs of Incoherence

118

119

122 124

5.2 Against Strong Normativity

130

5.3 Subjective Oughts 5.4 Conclusion

145 150

5.2.1 Advantageous Incoherence 5.2.2 Transmission to Necessary Means 5.2.3 Transmission to Sufficient Means

6. Belief

6.1 Cognitivism with the Strong Belief Thesis

6.1.1 The Case for the Strong Belief Thesis 6.1.2 The Case against the Strong Belief Thesis

131 137 141

152

154

156 164

6.2 Cognitivism without the Strong Belief Thesis

167

6.3 Problems for Cognitivism’s Explanatory Claim 6.4 Conclusion

173 176

6.2.1 Towards a Cognitivist Account without Strong Belief 6.2.2 Unknown Failures to Intend

7. Intention

7.1 The Aim of Intention

7.1.1 The Aim of Belief 7.1.2 Bratman’s Account 7.1.3 Controlled Action

168 172

177

180

182 189 195

7.2 Non-Normative Disjunctivism

197

7.3 Conclusion

214

7.2.1 7.2.2 7.2.3 7.2.4

Advantages Extension to Other Rational Requirements Explaining Pressure Social Practices, Normativity, and Myths

199 203 207 209

8. Conclusion

216

Bibliography Index

219 227

Acknowledgments I started thinking about instrumental rationality as a graduate student at Columbia, writing under Joseph Raz’s supervision. I’m grateful to Joseph for his encouragement and advice through the years. Work on this manuscript started in 2013–2014, when I was a Rockefeller Visiting Fellow at Princeton’s Center for Human Values, which is an ideal academic environment in which to think and write. Thanks to Peter Momtchiloff from Oxford University Press for his initial and continuing support of this project and valuable guidance along the way. Most of the writing was done during my first three years at Nebraska, and I’m very lucky to work in a Department full of philosophers who are both very smart and very kind. Special thanks go to Aaron Bronfman, Joe Mendola, and Mark van Roojen. They attended a graduate seminar on rationality I taught in Spring 2017. I learned a lot from that seminar, and from conversations with them. In what was the most enjoyable part of writing this book, Aaron, Joe, and I would meet regularly in The Coffee House to discuss one chapter at a time, often staying for several hours. I’d leave those sessions—sometimes walking home, having missed the last bus from downtown—exhausted, but full of ideas for revisions. (Thanks also for buying the coffee and pastries, Joe.) I was also fortunate to present some of these chapters at several places. I’m  grateful to those who arranged for these visits, and for the helpful ­questions from audiences at Princeton, Kansas State, McGill, Tennessee, and the Universidad de los Andes. Chapter 6 is a shortened and sharpened version of my ‘Cognitivism about Practical Rationality’ which was published in Oxford Studies in Metaethics, Vol. 9 (ed. R. Shafer-Landau), Oxford: Oxford University Press. I’m grateful for permission to reproduce that here. I’ve talked with many people about the ideas in this book over the years. However, my memory isn’t great, and I’m sure there will be several embarrassing omissions in any attempt to list them. But, anyway, many thanks to the following: Avery Archer, Sam Asarnow, Michael Bratman, Jonathan Dancy, Luca Ferrero, Daniel Fogal, Pete Graham, Bruno Guindon, Amelia Hicks, Niko Kolodny, Errol Lord, Hille Paakkunainen, Philip Pettit, Andrew Reisner, Jake Ross, Mark Schroeder, Kieran Setiya, Michael Smith, Daniel

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

x Acknowledgments Star, Dana Tulodziecki, Sarah Paul, Eric Wiland, Jonathan Way, Stephen White, and Alex Worsnip. I also owe a lot to two readers for Oxford University Press: Steve Finlay and Sergio Tenenbaum. Both carefully read through the manuscript and provided loads of tremendously helpful objections and suggestions for improvement. Lastly, I’m grateful for my wife, Anita, and son, Thomas, for being the ­wonderful people that they are, and to my parents, John and Angela, to whom this book is dedicated, for their unwavering support and encouragement since day one.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

1 Introduction Suppose that you intend to take the train from New York to Charleston, and believe that you’ll do so only if you buy a train ticket. Moreover, suppose you believe that you’ll buy a train ticket only if you intend to do so. (You know you won’t do so accidentally, or as a side-effect of your doing something else.) And suppose you don’t intend to buy a train ticket, nor trust that you will intend to do so later on. You’re instrumentally irrational. In particular, you exhibit a kind of instrumental irrationality called “means–ends incoherence,” since you fail to intend the means you believe are necessary for your ends. Means–ends incoherence is one type of attitudinal incoherence. There are others. You could have inconsistent intentions. Perhaps you both intend to be at home at 4:00 p.m. on Friday to take your son to soccer practice and intend to be at the faculty meeting at 4:00 p.m. on Friday, knowing you can’t be in two places at once—a natural mistake for those of us who use Google Calendar for work schedules and the kitchen whiteboard for family events without taking enough care to reconcile the two. Or you could have inconsistent beliefs. Perhaps you believe that your family won’t be visiting this weekend, but you believe they will visit you sometime in September, and you believe, since next weekend is the start of October, that if they visit you in September, it will be this weekend.1 Or you could be akratic in that you believe you ought not to do something but intend to do it anyway. Perhaps you know you shouldn’t make that sarcastic comment on the philosophy blog but can’t help yourself. Or you could have intransitive preferences, preferring apples to bananas, bananas to cherries, and cherries to apples. In all of these cases, there is incoherence among your attitudes. You are, borrowing an expression from Jonathan Dancy, at odds with yourself insofar as you have these combinations of attitudes.2 1  Perhaps, as some have argued (Worsnip, 2018), such combinations of beliefs are possible only if we understand the believer’s mind as being “fragmented” or “compartmentalized” or in some other way not fully transparent to the believer. (See Lewis 1982, p. 436, on compartmentalization.) It does seem extremely difficult, if not impossible, for someone to consciously attend to all three beliefs at the same time. 2  See Dancy 2009, p. 96. Another way to put the point is to say that all of these cases involve conflicts among your attitudes. (See Kolodny 2005, p. 516.) However, we should understand this phrase Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

2  Instrumental Rationality We’re inclined to call such attitudinal incoherence “irrational.” Yet we also sometimes apply that label without thinking there must be some attitudinal incoherence in order for the charge to stick. We might say that Amy is ir­ration­al in thinking that anthropogenic climate change is a Chinese hoax, and Brad is irrational in intending not to vaccinate his children, while we acknowledge that such beliefs and intentions could cohere perfectly well with the rest of the person’s attitudes. Like many philosophers, I find such criticism best expressed in the language of reasons: the available evidence provides Amy with conclusive reason to believe anthropogenic climate change is not a hoax, and health-related concerns provide Brad with conclusive reason to intend to vaccinate his children. And, although we do sometimes call people like Amy and Brad “irrational,” I’m going to be working primarily with a narrower conception of irrationality as attitudinal incoherence.3 There are rational requirements governing each of these examples of attitudinal incoherence. For instance, rationality requires that I do not both believe that I ought not make the sarcastic comment on the blog and intend to do so; rationality requires that I do not intend to be at home at 4:00 p.m. and intend to be at work at 4:00 p.m. while believing I can’t do both; rationality requires that I do not have the combination of believing that my family isn’t visiting this weekend, believing that they are visiting in September, and believing that if they are visiting in September, they are visiting this weekend; and, rationality requires that I do not intend to take the train to Charleston, believe I’ll do so only if I intend to buy a train ticket, and do not intend to buy a train t­icket.4 And so on. We can provide schematic formulations of the requirements, such as broadly so that it includes both conflicts between your having an attitude and having another (e.g., believing you ought not X and intending to X) and conflicts between your having an attitude and not having another (e.g., believing you ought to X and not intending to X). In cases of means–ends incoherence, the conflict involves the presence of two attitudes and the absence of another. 3  See Niko Kolodny’s (2005, pp. 509–10) related discussion of the distinction between “subjective” and “objective” rationality. 4  I’m going to accept the standard assumption that the various forms of irrationality prohibited by these requirements are possible. If it turns out that the standard assumption is incorrect, there would be a further question of whether it makes sense to have rational requirements that it is impossible to violate. (For discussion, see Lavin 2004.) Stephen Finlay (2009) has argued that means–ends incoherence is impossible, since, in his view, intending an end guarantees you intend the believed necessary means. Although I won’t take up his arguments here, this understanding of intentions strikes me as overly strict. If we allow for the possibility of inconsistent beliefs (or inconsistent intentions) in “fragmented” or “compartmentalized” minds—see note 1 in this chapter—it seems we should allow for the possibility of means–ends incoherence as well on similar grounds. Additionally, even if we accept such a strict view of intention, we would still have the ordinary phenomena of instrumental irrationality at hand—the familiar cases of those tripped up by distraction, depression, procrastination, fear, etc.— and we would have to discuss the phenomena using alternative concepts (perhaps aims instead of

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  3 Belief Consistency:  Rationality requires that [if one believes that P, and one believes that P→Q, then one doesn’t believe that ~Q]. and Means–Ends Coherence:  Rationality requires that [if one intends to X, and one believes one will X only if one intends to Y, then one intends to Y], which would cover these, and other, cases. According to Belief Consistency, the combination of believing that P, believing that P→Q, and believing that ~Q is rationally prohibited.5 And according to Means–Ends Coherence, the combination of intending to X, not intending to Y, and believing one will X only if one intends to Y, is rationally prohibited. There are controversies about how to formulate these and other rational requirements. (In Chapter  3, I’ll consider whether the requirements should have “wide-scope” or “narrow-scope” formulations. Means–End Coherence, as it appears here, is a “wide-scope” formulation in that “requires” has logical scope over the bracketed conditional. I’ll also discuss in Chapter 3 how these rational requirements could be read as either synchronic or diachronic requirements. And later in this introduction I’ll raise some issues relevant to the formulation of the requirement of means–ends coherence in particular.) But I don’t think that the views defended in this book will depend much upon disputes about the particular formulations. And so I’ll follow the practice in much of the literature on rationality of working with formulations ac­know­ ledged to perhaps be in need of refinement. And, to avoid confusion, I’ll use the capitalized “Means–Ends Coherence” to refer to the schematic wide-scope formulation above, and the lower-case “means–ends coherence” to refer to the requirement, however it’s formulated.

intentions, or perhaps even “schmintentions” where “schmintentions” involve all the features of intentions except those that necessitate intending believed necessary means). Even if Finlay is right that we shouldn’t understand such cases as involving a failure to intend means believed necessary for intended ends, there is clearly irrationality in such cases, and I suspect that many of the philosophical debates we’ll consider in this book about the formulation and normativity of the requirements governing such irrationality would re-emerge, albeit framed in a different vocabulary. 5  By calling this requirement “Belief Consistency,” I don’t mean to suggest that such combinations are the only way you could have inconsistent beliefs. Whenever the contents of your beliefs are logic­ al­ly inconsistent, you have inconsistent beliefs. There is a further question of whether it’s always ir­ration­al to have inconsistent beliefs. Perhaps preface paradox and lottery paradox cases show that it’s not always irrational.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

4  Instrumental Rationality As the title of this book indicates, I’m interested in understanding the normativity of means–ends coherence. However, there are at least two senses in which we might speak of a requirement as a “normative” one. First, there is the sense of “normative” frequently employed in recent philosophical debates about the normativity of rationality.6 For a requirement to be normative in this sense is for it to be such that we always ought to comply with it, where this is the “ought” of practical reason. (Or, on a weaker version of the thesis, it’s for it to be such that we always have a reason to comply.) A good bit of this book—the central part of Chapter 5—is devoted to arguing against the view that means–ends coherence is normative in this first, strong sense. I don’t think we always ought to be means–ends coherent. (I leave open the question of whether the weaker version of the thesis is true.) But, of course, I still think that means–ends coherence is normative in another, more mundane sense— the sense of “normative” that contrasts with “descriptive.” Means–Ends Coherence is a requirement of rationality—it presents some demand with which we might or might not comply—and so is clearly a norm. It’s a familiar fact that there can be norms that aren’t normative in the first sense. Plausibly, the requirements of golf etiquette, Nebraska state law, and English grammar all count as norms even though it’s not the case that we always ought to comply with them, and perhaps not the case that we always have a reason to do so. However, while these requirements are plausibly best explained in terms of contingent social practices or conventions, it seems doubtful that we could explain rational requirements in the same way.7 (What contingent social practice or convention would explain Belief Consistency, or Means–Ends Coherence?) After all, it’s commonly thought that rational requirements are necessary truths, and so they won’t be explained in the same way as these other requirements. But then what does explain them? In particular, why does rationality prohibit us from having means–ends incoherent combinations of attitudes? Here’s the answer I’ll give to this last question: if you’re means–ends incoherent, you’ve set yourself up so that you’ve guaranteed a failure in your attitudes. For instance, when you intend to take the train from New York to Charleston, and believe that you’ll do so only if you intend to buy a ticket, but don’t intend to buy a ticket, the logical structure of your attitudes has guaranteed that either your belief is false—you don’t actually need to intend to buy a

6  See, for instance, Broome 2008a. 7  I’ll say more about this in Chapter 7, Section 7.2.4. See also Kiesewetter 2018, p. 43; Worsnip 2018, pp. 186–8; and Way, 2018, pp. 490–1.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  5 ticket—or you won’t achieve your intended end of taking the train from New York to Charleston. So, either your belief is unsuccessful (because it is false) or your intention is unsuccessful (because it doesn’t issue in the intended action). In a slogan, you’re at odds with yourself because you’ve set yourself up to fail. This is a rough description of the main positive proposal of this book, which is developed in Chapter 7. It employs the strategy, originally suggested by Michael Bratman, of using the constitutive aim of intention to explain the consistency and coherence requirements governing intentions. This constitutivist strategy has been largely neglected in the literature on means–ends coherence, and Bratman himself provides only some brief remarks indicating how the explanation of means–ends coherence might go. And, as we’ll see in Chapter 7, I disagree with the particular way in which Bratman employs the strategy. However, I think there’s a way of employing the strategy that takes on board some of the components of the “disjunctivist” approach to means–ends coherence which I reject in Chapter  4. (The main reason for rejecting that approach is based on arguments given in Chapter  2.) To understand the advantages of the main positive proposal, we have to see how it avoids the pitfalls of other approaches to means–ends coherence, such as “cognitivist” approaches which explain means–ends coherence in terms of the requirements of theoretical rationality (discussed in Chapter  6), approaches which aim to show that we have conclusive reason, or conclusive “subjective” reason, to be means–ends coherent (discussed in Chapter 5), and others. So, we’ve got a lot of ground to cover before we get to the main positive proposal. As the previous paragraph indicates, there is a central argument running through this book, in which some of the results of earlier chapters support claims made in the later chapters. And so sometimes I’ll refer back to conclusions reached in earlier chapters. At the same time, however, I’ve tried to make each of the next six chapters relatively self-standing, so that they can be understood apart from the others. (The exception here is the two chapters on “Normativity,” Chapters 4 and 5, which are best read together.) Each chapter takes up some topic that has been central to the philosophical literature on means–ends coherence: the worry about allowing intentions to “bootstrap” reasons (Chapter 2); the debate over whether means–ends coherence is wideor narrow-scoped (Chapter 3); the relationship between means–ends coherence and normativity, in the first sense of “normativity” mentioned above (Chapters 4 and 5); the attempt to explain means–ends coherence by appealing to requirements of theoretical rationality (Chapter 6); and, finally, the use of the constitutive aims of intention and belief to explain means–ends

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

6  Instrumental Rationality coherence and related requirements (Chapter  7). My hope is that even if you’re not convinced by the overall project and the resulting view, you’ll find something of value in each individual chapter. Although I’ll maintain a focus on means–ends coherence throughout, many of the arguments and views discussed in the following chapters, such as those concerning the scope and normativity of means–ends coherence, will have implications for other rational requirements as well. In this introductory chapter, I’ll start by giving a brief roadmap (§1) of what’s to come in each of the next five chapters. I’ll then continue (§2–§4) to introduce our topic of means–ends coherence, explaining why the topic has generated the philosophical interest it has. I’ll briefly (§2) look at a wellknown historical precursor to contemporary philosophical work on means–ends coherence: Kant’s discussion of “hypothetical imperatives” in the Groundwork. And I’ll (§3) then turn to two more recent philosophical debates that have lead philosophers to be concerned with understanding means–ends coherence and its normativity. Lastly, I’ll (§4) discuss some issues relevant to the formulation of the rational requirement. Readers already familiar with this background should feel free to skip these sections and move straight ahead to Chapter 2.

1.1  The Roadmap Many philosophers have objected to certain views of the normative significance of intentions, such as the view that intentions provide reasons, or the view that narrow-scope requirements are normative (in the first sense above), by claiming that such views license implausible “bootstrapping.” However, it’s not exactly clear what bootstrapping is, nor why it’s problematic. There’s no shared understanding of the bootstrapping objection in the literature. In Chapter 2 (“Bootstrapping”), I present a novel interpretation of the bootstrapping objection—one that, unlike previous versions of the objection, won’t overextend by disallowing promises from providing reasons. Although there is controversy about whether intentions provide reasons, it’s less controversial to say that intentions are governed by rational requirements of consistency and means–ends coherence. (This isn’t entirely uncontroversial, however, since, as we’ll see, some have argued that such formal coherence requirements are a myth.) I’ll consider the structure of rational requirements in Chapter  3 (“Scope”), focusing in particular on the debate between “wide-scopers” and “narrow-scopers” about rational requirements.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  7 I’ll set out what I take to be the strongest argument for the wide-scope ­formulations, and respond to challenges raised against that argument. I’ll then consider how to defend the wide-scope formulations against the most pressing arguments put forth by narrow-scopers. In Chapters  4 and  5 (“Normativity”), I consider three different ways in which philosophers have appealed to normativity—specifically, to claims about the attitudes, or combinations of attitudes, one ought to have—to explain (or explain away) means–ends coherence. First, according to Normative Disjunctivism, being means–ends incoherent ensures that either you have some particular intention or belief you ought not have or you don’t have an intention you ought to have. Some, but not all, defenders of Normative Disjunctivism have used it in support of the view that the requirement of means–ends coherence is a myth. I argue that Normative Disjunctivism isn’t true, and wouldn’t support such a myth theory even if it were true. Second, I consider whether means–ends incoherent combinations of attitudes are forbidden by the “ought” of practical reason. I’m skeptical that they are. If we always ought not to have such combinations of attitudes, we should be able to state what the reasons are not to have them. I consider the Prichardian strategy of holding that rationality provides “its own” reasons, and so it’s a mistake—paralleling the mistake Prichard accused “moral phil­ oso­phy” of making—to request “external” reasons to comply with rational requirements. I don’t find this suggestion persuasive. And once we try to specify what the reasons to be means–ends coherent are, the task proves to be quite difficult. Moreover, I’ll discuss three independent arguments against the view that one always ought to be means–ends coherent. Third, some have argued that we can use the concepts of a subjective reason and a subjective ought to understand means–ends coherence. (Roughly, while objective oughts are determined by the relevant facts, subjective oughts are determined by your beliefs. For instance, if you falsely believe the hotel is on fire, you subjectively ought to leave, but objectively ought to stay put.8) I consider two different proposals for using the subjective ought to explain means–ends coherence, and argue that they won’t succeed. In Chapter 6 (“Belief ”), I consider several important “cognitivist” approaches to means–ends coherence. Cognitivists start by defending some thesis about the way in which intention involves belief. (Perhaps intending to X involves believing you will X. Or perhaps intending to X involves merely believing X is possible.) Cognitivists then argue that the requirements of 8  The example is from Parfit 1997, p. 99.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

8  Instrumental Rationality theoretical rationality governing those involved beliefs can explain some of the requirements of practical rationality governing those intentions, such as means–ends coherence. I consider several ways in which this approach could be developed, but ultimately conclude that the prospects for success are not great. In Chapter 7 (“Intention”), I suggest that we look in a new direction. Rather than trying to explain means–ends coherence by looking to the attitudes, or combination of attitudes, we ought to have, or by looking to the requirements of theoretical rationality governing involved beliefs, I suggest that we look to the constitutive aims of the very attitudes governed by means–ends coherence (intention and belief). I first spend some time trying to isolate the way in which there is an aim of intention analogous to the aim of belief, and I then consider how these constitutive aims can be used to explain several rational requirements, including means–ends coherence. This approach, in my view, is the most promising avenue available to understanding the normativity of means–ends coherence.

1.2  Kant on Hypothetical Imperatives As a way of introducing our topic to readers who aren’t already specialists working within the narrow literature on rationality, it will help to start off with Kant’s famous discussion of hypothetical imperatives in the Groundwork. Kant tells us that “whoever wills the end also wills (insofar as reason has de­cisive influence on his actions) the indispensably necessary means to it that are within his power.”9 Although, as we’ll see, there are important differences between Kantian hypothetical imperatives and rational requirements, there is a key insight in Kant’s discussion which is important for thinking about means–ends coherence. That key insight is the idea that agents are subject to an instrumental “ought” that won’t be explained by appeal to the moral or rational status of their ends—an “ought” that applies equally well to both the

9  Kant  1785/1997, p. 28. [4:417]. The standard reading of this passage is that the parenthetical clause indicates that Kant thinks “reason” requires that one not will the end without willing the indispensably necessary means in his power. See, for instance, Korsgaard 2008, §3. (Lee 2018 presents a case for an alternative “descriptive” reading of it, such that Kant isn’t making a normative claim here.). If the standard reading is correct, there is a further question of whether this requirement of reason is wide- or narrow-scoped. Schroeder 2005b defends the narrow-scope interpretation, whereas Hill 1973 and Rippon 2014 defend the wide-scope interpretation. Since our aim here is simply to use Kant as a way of introducing some important aspects of our topic, it won’t help to go into the finer points of Kant exegesis.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  9 physician aiming to cure his patient and the murderer aiming to poison his victim, to use Kant’s own examples. I’ll first explain Kant’s distinction between hypothetical and categorical imperatives, then explain why I think this key insight is important, and then turn to some differences between hypothetical imperatives and rational requirements. Kant distinguishes two kinds of imperatives, or commands: hypothetical and categorical. Here’s his explanation of the distinction: Now, all imperatives command either hypothetically or categorically. The former represent the practical necessity of a possible action as a means to achieving something else that one wills (or that it is at least possible for one to will). The categorical imperative would be that which represented an action as objectively necessary of itself, without reference to another end.10

Kant notes that each kind of imperative can be expressed using an “ought.”11 For instance, a hypothetical imperative, such as “To go to Sugar Hill way up in Harlem, take the ‘A’ train,” could be expressed as “To go to Sugar Hill way up in Harlem, you ought to take the ‘A’ train.”12 The “ought” here depends upon a certain instrumental relation holding between your taking the “A” train and your going to Sugar Hill: were your taking the “A” train not a means to going to Sugar Hill—suppose the tracks are under repair and so the “A” train doesn’t service that neighborhood—then the “ought” would be withdrawn. The person making this hypothetical “ought” claim isn’t committed to thinking there’s anything good in itself about your taking the “A” train, nor is she committed to thinking that your taking the “A” train is something you have conclusive reason to do, regardless of whether it brings you to where you want to go. This marks a contrast with categorical imperatives, in which “the action is represented as in itself good, hence as necessary in a will in itself conforming to reason.”13 Kant’s distinction between hypothetical and categorical imperatives shouldn’t be construed as a grammatical distinction between conditional and unconditional imperatives. After all, the conditional imperative “If someone is drowning, you should save them” isn’t thereby hypothetical. Saving a drowning 10  Kant 1785/1997, p. 25. [4:414]. 11  “All imperatives can be expressed by an ought and indicate by this the relation of an objective law of reason to a will that by its subjective constitution is not necessarily determined by it (a necessitation).” Kant 1785/1997, p. 24. [4:413]. 12  The example is a familiar one in the literature on anankastic conditionals, the exploration of which would be distracting here. 13  Kant 1785/1997, p. 25. [4:414].

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

10  Instrumental Rationality person isn’t here presented as necessary for some further goal in the way that taking the “A” train was presented as necessary for the further goal of going to Sugar Hill. (Rather, it seems in itself good to save people from drowning.) And hypothetical imperatives can sometimes be formulated without an antecedent: a speaker might take the relevant goals as shared background knowledge, and simply tell you to take the “A” train, without bothering to state that the command depends upon your having the goal of going to Sugar Hill. Nor would it be correct to say that hypothetical imperatives are those imperatives whose applicability depends on your ends, while categorical imperatives are those imperatives whose applicability doesn’t depend on your ends. Consider the following imperative: “if you’re intending to send an insulting email to all your colleagues, sleep on it and talk to your Chair over coffee before hitting ‘send’. ” The imperative here does depend on your end in that the command to sleep on it and talk to your Chair is applicable only because you have the intention of sending the insulting email. But it’s clearly not a hypothetical imperative. The command doesn’t, as Kant puts it, “represent the practical necessity of a possible action as a means to achieving something else that one wills.”14 Unlike the case of the command to take the “A” train, the commanded action in this case doesn’t bear an instrumental relation to your end, and so the command doesn’t count as a hypothetical imperative. With our hypothetical imperative, “To go to Sugar Hill way up in Harlem, you ought to take the ‘A’ train,” the “ought” which attaches to your taking the “A” train depends on whether your taking the “A” train promotes your goal of getting to Sugar Hill, in the sense that we would withdraw the “ought” if taking the “A” train didn’t promote your goals. It’s clear that many moral imperatives will be non-hypothetical, since they will lack this feature. For instance, you ought to refrain from making insincere promises, regardless of whether doing so promotes any of your goals. But, as Philippa Foot famously observed, we shouldn’t infer from an imperative’s being non-hypothetical in this sense that conformity with it is, as Kant puts it, “necessary in a will in itself conforming to reason.”15 After all, the requirements of etiquette, or of a social club, might also be non-hypothetical in this sense. These requirements could apply to you regardless of whether your doing what is required promotes any of your ends. But compliance with these requirements is not therefore ne­ces­sary in a will conforming to reason; one could reasonably dismiss some of these rules as pieces of nonsense, as Foot puts it.16 So, some further argument would have to 14  Kant 1785/1997, p. 25. [4:414]. Emphasis added. 15 Foot 1972. 16  Foot 1972, p. 308.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  11 be given for the conclusion that the “oughts” of morality are, as Kant thinks, necessary in a will conforming to reason. It’s not enough to note that moral “oughts” apply regardless of whether they promote your ends, since the same could be said of the “oughts” of etiquette and club rules. However, we need not concern ourselves here with the status of the “oughts” of morality. We’re interested in how Kant’s discussion of hypothetical imperatives relates to instrumental rationality, and means–ends coherence in particular. I’ll first outline an important lesson we can draw from Kant’s discussion, and then highlight three differences between hypothetical imperatives and means–ends coherence. Kant recognizes that there is an instrumental “ought” that applies to agents regardless of the moral or rational status of their ends. Concerning imperatives of skill, which are a kind of hypothetical imperative (which, again, can all be expressed using “ought”), Kant writes: Whether the end is rational and good is not at all the question here, but only what one must do in order to attain it. The precepts for a physician to make his man healthy in a well-grounded way, and for a poisoner to be sure of killing, are of equal worth insofar as each serves to bring about his purpose.17

Just as we can say there’s a way in which the physician who doesn’t intend the necessary means to making his patient healthy is not doing something he ought to be doing, we can say that there’s a way in which the poisoner who doesn’t intend the necessary means to killing his intended victim is not doing something he ought to be doing. But it’s apparent that such “oughts” won’t be explained by appealing to the ends the agent ought to have, since in only one of these two cases is the end a permissible one. We might have initially thought, had we just focused on the case of the physician, that we could explain the relevant “ought” in the following way: the physician ought to heal his patient, and his reasons to heal the patient will “transmit” from ends to means, and so he ought to do what’s necessary for healing the patient. As Thomas Nagel puts it: We might say that if being thirsty provides a reason to drink, then it also provides a reason for what enables one to drink. That can be regarded as the consequence of a perfectly general property of reasons for action: that they transmit their influence over the relation between ends and means.18 17  Kant 1785/1997, p. 26 [4:415].

18  Nagel 1970, p. 34.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

12  Instrumental Rationality (To fill out this explanation, we’d need precise specifications of the relevant principles governing the transmission of “oughts” and reasons from ends to means.19) And this explains why the physician who doesn’t intend the means to healing his patient isn’t doing something he ought to be doing. However, it is obvious that we can’t give a parallel explanation for the case of the poisoner. In that example, it’s not the case that he ought to kill his intended victim, and so there would be no reason—at least no reason of significant weight—that would transmit from ends to means. And so the parallel explanation is a non-starter. But, as Kant observes, both the physician and the poisoner are subject to the same kind of imperatives—these “imperatives of skill.” When the phys­ ician doesn’t intend the means necessary for making the patient healthy, he goes wrong in the same way as does the poisoner who doesn’t intend the means necessary for killing his intended victim. The same kind of “ought” applies in both cases. So, while the above transmission-based explanation might explain one way in which the physician is not doing what he ought to be doing when he doesn’t intend the means to heal his patient, it leaves unexplained the “ought” to which Kant was referring, since an explanation of that “ought” would also extend to the case of the poisoner. Kant’s insight is important for thinking about instrumental rationality. It shows us that we can’t explain the requirement of means–ends coherence simply by looking at how reasons for intended ends transmit to reasons to intend means. And it illustrates that any adequate explanation of means–ends coherence must allow that the requirement is applicable regardless of the normative status of the agent’s intentions; it applies equally well to intentions to heal and intentions to kill.20 (It might help to observe that the physician and the poisoner can exhibit the same kind of rational defects in the pursuit of their ends. Perhaps they are both procrastinators, and keep putting off buying the needed drugs—medicine in one case, poison in the other—until it’s too late. Or perhaps they are both distracted, and because of this forget to intend some means they know they must intend to carry out their projects.

19  For further discussion of instrumental transmission, see Kiesewetter (2015) and Kolodny (2018). 20  We might also add that just as we can display instrumental rationality (or irrationality) with respect to ends which vary in their normative status, we can also display instrumental rationality (or irrationality) with respect to ends that vary in their believed normative status. I might think I ought not buy the expensive bottle of bourbon, but be instrumentally rational in carrying out my plan to do so. (We might say, with R. Jay Wallace (2001, p.1), following Aristotle, that I here display a kind of “cleverness” in the pursuit of this end I believe I ought not have.) And I also could have been instrumentally irrational with respect to this end.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  13 This irrationality might be beneficial in the case of the poisoner. But it’s no less irrational.) However, we should also keep in mind that there are several important differences between Kantian hypothetical imperatives and the rational requirement to be means–ends coherent. I’ll mention three.21 First, Kant notes that hypothetical imperatives can be formulated for possible or actual ends. (The former involve what Kant calls a “problematically practical principle” while the latter involve what he calls an “assertorically practical principle.”22) But means–ends coherence applies only to those ends intended by an agent.23 Suppose I don’t intend to buy a coffee right now. In that case, I’m not rationally required to intend to go to one of the campus coffee shops, even if I know that’s the only way to get coffee right now. But one could formulate a hypothetical imperative (e.g., “If you intend to buy a coffee right now, go to one of the campus coffee shops”) with respect to that possible end.24 Second, hypothetical imperatives are commands to take what are actually means, not what are believed to be means, to our ends. (For instance, Kant writes that “the hypothetical imperative says only that the action is good for  some possible or actual purpose” and, in the passage quoted earlier, “[w]hether the end is rational and good is not at all the question here, but only what one must do to attain it.” He doesn’t say “the action is believed to be good for” nor “what one thinks one must do to attain it.”) But it’s our beliefs about the means, not the facts about the means, that’s relevant to our being rational or irrational. Suppose I intend to buy a coffee on campus and believe that Starbucks is the only place open, but this isn’t so: the Starbucks is closed and Dunkin’ Donuts is open. There are no other places to get coffee. In this case, I wouldn’t display any irrationality in not intending to go to Dunkin’ Donuts. And I would display irrationality in not intending to go to Starbucks. So, whether I’m instrumentally rational or irrational doesn’t depend upon 21  The following points, although not always fully appreciated in the literature, are familiar ones made by many others. (See, for instance, Korsgaard 2008, and Finlay 2009, 2010.) 22  Kant 1785/1997, p. 26 [4:415]. 23  It’s also worth noting that there need be no irrationality in failing to intend means believed ne­ces­sary to merely desired ends. I may, without irrationality, want to be a world-class athlete without intending to do what I know it takes. While there would be irrationality in my intending to be a worldclass athlete, there’s no irrationality in my merely continuing to have this desire. 24  It might appear that the wide-scope formulation of means–ends coherence commits us to thinking there’s a requirement that applies to us regardless of whether we have any of the relevant attitudes. For instance, Means–Ends Coherence prohibits me from intending to go to Mars, believing I must intend to build a red spaceship to do so, and not intending to build a red spaceship. And this pro­hib­ ition applies to me, despite my having neither this intention nor this belief. However, as I’ll argue in Chapter 3, there is a way to be a wide-scoper while limiting the applicability of rational requirements to those agents who already have the relevant attitudes.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

14  Instrumental Rationality how things actually are, but on how I believe them to be. (Additionally, we could note here a further difference: means–ends coherence is limited to beliefs about necessary means, while hypothetical imperatives could command one to take non-necessary means as well.) Third, hypothetical imperatives are commands to perform some action that would be good as a means to achieving some end. (Look again at the two quotations from Kant in the previous paragraph. He speaks of “the action” being good for some purpose, and says that hypothetical imperatives tell us what one “must do”—not intend to do—to achieve the end.) But means–ends coherence requires that we intend the means to our ends, not that we take the means to our ends. And that’s a good thing, since our failure to take the means to our ends might not involve a failure of rationality. Suppose I intend to buy a coffee right now, believe that to do so I must go to a campus coffee shop, and intend to go to a campus coffee shop. But you prevent me from carrying out this intention, perhaps by bribing the owners of the campus coffee shops to close their doors early. Surely I don’t violate a requirement of rationality when you prevent me from carrying out my intention in this way. If you could somehow prevent me from intending to go—perhaps you have some means of messing with my mind—then I would be means–ends incoherent (though, since it would be through no fault of my own, perhaps I shouldn’t be criticized for that). But, intuitively, unlike the case in which you prevent me from carrying out my intention by closing up the coffee shops, your interference with my mind in this case has prevented me from being fully rational. And that intuitive result follows from our allowing that there is a rational requirement I violate in this case, but not in the case in which you prevent me from carrying out my intention by causing the coffee shops to close up early.25 In summary, for each of the three attitudes relevant to the requirement of means–ends coherence (intending an end, believing that intending some means is necessary for achieving that end, intending the means), there is a way in which Kantian hypothetical imperatives aren’t relevant: they apply to both intended and unintended ends; they concern actual, as opposed to believed, means (including non-necessary means); and, they require that we take, rather than intend to take, the means. But Kant’s observation that the physician and poisoner are both subject to hypothetical imperatives, though they differ in the moral and rational status of their ends, is an important starting point for thinking about instrumental rationality. Our account of instrumental rationality should explain how the means–ends incoherent 25  Broome (2013, p. 89) makes a point along these lines.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  15 physician and the means–ends incoherent poisoner go wrong in the same way, despite the variation in the normative status of their ends.

1.3  Contemporary Interest There are two main contemporary sources of philosophical interest in instrumental rationality. One is an older debate between Kantians and Humeans about the scope of practical rationality. Another is a more recent debate about the normativity of rational requirements. As we’ll see, the background assumptions in these two debates differ dramatically. Let’s start with the first debate about the scope of practical rationality. Much of the initial contemporary philosophical interest in instrumental rationality arose from the debate about whether instrumental rationality is, as it is sometimes put, all there is to practical rationality. Kant didn’t think it was. He held that, in addition to instrumental rationality, the domain of practical rationality extended to include rational requirements of prudence and morality as well. However, many philosophers have been skeptical about this more expansive conception of practical rationality. As Christine Korsgaard observes: But this [Kantian] approach has not usually been followed in the AngloAmerican tradition. Empiricist moral philosophers, as well as the social scientists who have followed in their footsteps, have characteristically assumed that hypothetical imperatives do not require any philosophical justification, while categorical imperatives are mysterious and apparently external constraints on our conduct.26

For an example of a view in the tradition Korsgaard has in mind, we might look to John Mackie’s well-known denial of the objectivity of values. In explaining his thesis that there are no objective values, Mackie first distinguishes hypothetical from categorical imperatives and then observes that many moral judgments “contain a categorically imperative element.” He then tells us that “so far as ethics is concerned, my thesis that there are no objective values is specifically the denial that any such categorically imperative element is objectively valid.”27 He has no quarrel with hypothetical imperatives.

26  Korsgaard 2008, p. 29.

27  Mackie 1977, p. 29.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

16  Instrumental Rationality In Korsgaard’s view, it’s common for philosophers to simply assume that instrumental rationality is not in need of any philosophical justification, while claiming that the more inclusive conception of practical rationality stands in need of support. But it’s worth adding that some in the Humean camp have argued for the view that instrumental rationality has some special status not shared by the alleged rational requirements of prudence and morality. For instance, James Dreier has appealed to a practical analogue of Lewis Carroll’s observations about modus ponens in “What the Tortoise Said to Achilles” in order to show that instrumental rationality has a “ground-level normative status.”28 In Dreier’s view, this status justifies a Humean skepticism about the scope of practical rationality, as well as the conclusion that “the only ultimate sort of reasons are instrumental reasons.”29 In response to such skepticism about the scope of practical rationality, other philosophers—mostly notably, Korsgaard and Jean Hampton30—have argued, in different ways, that a proper understanding of the normativity of instrumental rationality could make room for expanding the scope of prac­ tical rationality to include prudential and moral requirements as well. Hampton has argued that the kind of naturalistic skepticism about categorical imperatives espoused by Mackie and others would also extend to target hypothetical imperatives. Specifically, she aims to show that “hypothetical imperatives . . . have what I call objective normative authority, and that this authority is just the same as the authority that is supposed to make objective moral reasons scientifically unacceptable.”31 So, if you find the “ought” of categorical imperatives puzzling, you should find the “ought” of hypothetical imperatives puzzling as well.32 And, more ambitiously, Korsgaard has argued against the very coherence of thinking instrumental rationality is all there is to practical rationality. As she puts it, “unless there are normative principles directing us to the adoption of certain ends, there can be no requirement to take the means to our ends.”33 28  See Dreier 2001, p. 42, and Carroll 1895. 29  Dreier 2001, p. 43. One could challenge whether this conclusion follows. See Jollimore 2005 for discussion. 30  See Korsgaard 2008 and Hampton 1998, especially Part II. 31  Hampton 1998, p. 126. 32  Hampton 1998, p. 127. Along similar lines, Russ Shafer-Landau (2006, p. 227) has argued that Gilbert Harman’s causal test for the reality of moral facts, according to which moral facts, if they are to be construed realistically, must have independent causal powers, is too strong, since that test would also rule out the reality of the entire class of normative facts, including facts about our “instrumental duties.” 33  Korsgaard  2008, p. 220. I don’t think that it’s correct to think of instrumental rationality as involving a requirement to take the means to our ends. As I said earlier, there need be no irrationality in my not taking the means, though there would be in my not intending to take the (believed ne­ces­ sary) means. Also, it’s better to understand “requires” as having logical scope over a conditional, so as

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  17 In my view, there is some confusion involved in this debate on both sides. But it’s not worth going into, since my only aim here is to set forth one source of philosophical interest in instrumental rationality. In Chapter 7, I’m going to argue that means–ends coherence can be explained (in part) in terms of the constitutive aims of the attitudes governed by the requirement. And that might remove metaethical doubts about means–ends coherence that might remain for moral and prudential requirements, assuming that a similar ex­plan­ation isn’t available for those requirements. And so there is a sense in which means–ends coherence, and related norms of rationality, might have a “special status.” But that will have to wait until later. The second, more recent source of philosophical interest in instrumental rationality is due to the debate over the normativity of rationality—that is, the debate over whether the requirements are normative in the first sense of “normative” introduced earlier. In other words, are the requirements of rationality such that we always ought to (or, more weakly, have a reason to) comply with them? (We could pose such a question for any set of requirements, such as the requirements of etiquette, Catholicism, law, prudence, morality, or rationality.) As a central, relatively uncontroversial, example of a requirement of rationality, means–ends coherence has figured prominently in the literature on the normativity of rationality.34 Niko Kolodny, in his influential paper “Why be Rational?,” presents several challenges to the idea that rationality is normative. According to one challenge, if we say that one always has a reason to be rational, we should be able to state what that reason is.35 But it’s not clear what the reason is. Although there may in some cases be an instrumental reason to be rational—perhaps doing so would on occasion promote some good—there don’t seem to be such instrumental reasons in general. And it’s doubtful that there’s a noninstrumental reason to be rational.36 Kolodny puts the point rather forcefully in another paper, claiming that it’s “outlandish” to think that the kind of “psychic tidiness” demanded by coherence constraints should be “set alongside such final ends as pleasure, friendship, and knowledge.”37 Other philosophers

to allow for the rational permissibility of giving up an intention instead of intending the means. We’ll return to this in Chapter 3. Korsgaard herself notes that it is sometimes rationally permissible to give up an intention instead of taking the means (p. 49), and so she wouldn’t endorse any “narrow-scope” formulation that would rule out this possibility. 34  See Broome  2000, Broome  2008a, Kolodny  2005, Raz  2005, Setiya  2007, Bratman  2009c, and Ross 2012. 35  Kolodny 2005, pp. 542–7. 36  Kolodny 2005, pp. 543–5. 37  Kolodny 2007b, p. 241.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

18  Instrumental Rationality have responded to Kolodny in various ways, attempting to answer the challenge of specifying why we should be rational.38 If they succeed in meeting this and other challenges, we’ll be able to hold that the requirements of rationality can be included, perhaps alongside the requirements of prudence and morality, in the class of “normative” requirements. The assumptions in this more recent debate differ rather dramatically from the assumptions of the earlier debate about instrumental rationality. Very roughly, while the earlier debate assumes hypothetical imperatives are unproblematic and categorical imperatives questionable, this more recent debate assumes categorical imperatives are unproblematic and hypothetical imperatives questionable. Less roughly, it’s here assumed that the requirements of prudence and morality are normative (in the reason-implying sense), and the debate is about whether we can say the same for rational requirements like means–ends coherence. But the earlier debate was over whether requirements of prudence and morality were normative (in the reasonimplying sense), and it took requirements of rationality to be largely unproblematic. (Consider, for instance, the following remark from Jean Hampton: “Most theorists believe that whereas the question ‘Why be moral?’ is deeply troub­ling, the question ‘Why be rational?’ is not.”39 But this remark would be a poor description of participants in the more recent debate sparked by Kolodny’s “Why Be Rational?” paper. In this debate, there’s nothing deeply troubling about thinking we have non-instrumental reason to care about pleasure, friendship, and knowledge. What needs to be explained here is why “psychic tidiness” gets on that list.) So, we’ve identified a second source of philosophical interest in instrumental rationality: the debate over whether rational requirements are normative. We won’t, at least directly, take up this debate in this book. However, in Chapters 4 and 5, I’ll consider several possible ways of explaining (or explaining away) the rational requirement of means–ends coherence in terms of the attitudes, or combinations of attitudes, you ought to have. Many of the arguments in the debate over rationality’s normativity will figure prominently in those chapters. One of the main theses of Chapter 5 is that it’s not true that we always ought to be means–ends coherent. (I’ll give three independent arguments in support of this thesis.) Since means–ends coherence is a requirement of rationality, it follows that it’s not true that we always ought to be 38  We’ll consider some answers in Chapter  5, including Nic Southwood’s Prichardian claim that rationality provides “its own” reasons, and Michael Bratman’s Kantian claim that there are noninstrumental reasons of self-governance to be practically rational. 39  Hampton 1998, p. 126.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  19 rational. Thus, rationality isn’t, as it often put, “strongly normative.” (But these arguments will leave open the possibility of a weaker view—that one always has a reason to be rational—being true.) In summary, I’ve briefly looked at two philosophical debates with different background assumptions—an older debate about the scope of practical rationality, and a more recent debate about the normativity of rationality— that explain why philosophers have been interested in instrumental rationality in particular. (And I’ve given some indications of how the views defended in the book will relate to those debates.) But I also think (and hope you will agree) that the questions about means–ends coherence that we’ll explore in the next six chapters are interesting in their own right, independently of their implications for these two debates.

1.4  Formulating Means–Ends Coherence: The Instrumental Belief We’ve already noted a couple of important points relevant to the formulation of means–ends coherence: the requirement should be made applicable by our intentions, not our desires, and it should require us to intend, as opposed to take, the means. But we haven’t yet discussed the instrumental belief specified in the requirement: the belief that one will X only if one intends to Y. It’s worth noting that it’s possible for someone to have this belief while at the same time acknowledging that she could be wrong about it. In other words, one could be a fallibilist with respect to this belief. I might, for instance, believe that I will ride the roller coaster only if I intend to buy a ticket, but also acknowledge I could be mistaken about this.40 (Perhaps I will, surprisingly, be given a free ticket from the ride operator.) In the same way, I could acknowledge that I could be mistaken in my belief that the amusement park is open—perhaps it has unexpectedly closed due to an emergency—or my belief that the roller coaster is running—perhaps it has shut down for repairs. Indeed, this seems to be the ordinary approach most of us have toward our beliefs. We acknowledge they might be wrong. And in the same way that one doesn’t have to be certain that the park is open in order to believe the park is open, one doesn’t have to be certain that one won’t ride without intending to buy a ticket in order to believe that one won’t ride without intending to buy a ticket. 40  The roller coaster example comes from Korsgaard 2008, pp. 40–1.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

20  Instrumental Rationality You might wonder why the belief that makes the requirement applicable is the belief that I will ride the roller coaster only if I intend to buy a ticket and not the belief that I will ride the roller coaster only if I buy a ticket. After all, an agent engaged in a process of instrumental reasoning, proceeding from an intention to ride the roller coaster to the formation of an intention to buy a ticket, isn’t likely to reflect inwardly on her own intentions. She’s more likely to think, “I won’t ride unless I buy a ticket”—reflecting instead on what she must do, as opposed to intend to do, in order to ride. In reply, we can note two things. First, our formulation, Means–Ends Coherence, doesn’t commit us to the view that the instrumental belief which makes the requirement applicable should also figure in an agent’s instrumental reasoning. The requirements of attitudinal coherence we’re considering prohibit you from having certain combinations of attitudes. But they don’t commit us to any view about the nature of good reasoning. Specifically, Means–Ends Coherence doesn’t commit us to the view that this belief should figure into good instrumental reasoning. (Moreover, Means–Ends Coherence holds that a rational requirement is put in place so long as one has this belief; it need not be the case that the belief is occurrent or explicit—as it normally would be when it figures in reasoning—in order for the requirement to be applicable.41) Second, we can note that formulating means–ends coherence with the belief that one will X only if one will Y will yield problems with cases involving foreseen side-effects—problems that our formulation neatly avoids. Consider: X→Y Formulation:  Rationality requires that [if one intends to X, and one believes one will X only if one will Y, then one intends to Y]. Suppose I intend to grade the exams fairly, and I believe that I will do so only if I make some students upset. I don’t intend to make them upset, but I know that this will be a side-effect of my grading fairly. But we don’t want to say that rationality requires me to also intend to make them upset (or revise my other attitudes).42 I would display no irrationality in not forming this intention. Since X→Y Formulation would convict me of irrationality here, it should be

41 On the distinction between occurrent and dispositional belief, and the distinction between explicit and implicit belief, see Schwitzgebel 2015. 42  The parenthetical is needed since Means–Ends Coherence is a wide-scope requirement. We’ll discuss the rationale for this in Chapter 3.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  21 rejected.43 But this example is no counterexample to Means–Ends Coherence, since I don’t believe I will grade fairly only if I intend to make some students upset. (I don’t think that’s necessary for grading fairly.) So, Means–Ends Coherence simply doesn’t apply in this case. But it does apply to those intentions I believe I must form to grade fairly. For instance, suppose I think I’ll grade fairly only if I intend to anonymize the exams. (After all, they won’t anonymize themselves.) According to Means–Ends Coherence, I’m required to intend to anonymize them (or revise my other attitudes). One tempting quick solution to this problem of foreseen side-effects would be to amend the requirement so that it applies to believed necessary means: X→Y Formulation – Amended:  Rationality requires that [if one intends to X, and one believes that one will X only if one will Y and one believes that X-ing is a means to Y-ing, then one intends to Y]. Since making the students upset isn’t believed to be a means to grading fairly, the amended formulation will not entail that there is irrationality in my failing to intend to make them upset. But it would allow that there is irrationality in my failing to intend to anonymize them, since doing so is believed to be a means to grading fairly. In developing this thought, one would have to say more about what’s involved in believing of some action that it is a means to performing another one. But I don’t think it’s worth pursuing that project, since it seems that foreseen side-effects will threaten this proposal in another way: suppose that I intend to grade fairly, believe that I will do so only if I anonymize the exams, but don’t intend to anonymize the exams because I believe I’ll anonymize them as a side-effect of my doing something else I intend to do. Perhaps I intend to use the campus online testing center for exams and know that this will guarantee the anonymization of the exams. Things will take care of themselves. So, there’s no need for me to intend to anonymize them, nor would there be any irrationality in my failing to intend to anonymize them.44 But our amended proposal would entail that I  would be irrational in not intending to anonymize them. Means–Ends Coherence, however, avoids this undesirable result, since I don’t believe that

43  See Kamm 2000, pp. 32–6, and Broome 2013, pp. 162–3. (For a discussion of the broader significance of the distinction between intended ends and foreseen side-effects for our understanding of intentional action, see Paul 2011.) 44  See also Setiya 2007, pp. 667–8, on “automatic means.” In his example, one intends to make a fire and believes that one will do so only if one causes a chemical reaction in the wood and flexes the muscles in one’s hand. One would display no irrationality in not intending these means.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

22  Instrumental Rationality I must intend to anonymize them in order to grade fairly. And so there need not be any irrationality in this case. Another issue important to the formulation of means–ends coherence is the phenomenon of self-trust. Suppose I don’t now intend the means but trust I will at a later time. Kieran Setiya provides an example: I know that I will need to strike a match in order to light the kindling in order to make a fire. But I am sure that I will decide to [do] so when the materials for the fire are ready and that it is not necessary to settle this ahead of time. A proper formulation of the instrumental principle would deny that I must form the intention to strike a match, instead of the belief that I will – although forming it remains a possibility.45

In this case, I intend to make a fire, believe I must intend to strike the match to make a fire, but don’t intend to strike the match, since I believe that I will intend to do so at a later point. I have a combination of attitudes prohibited by Means–Ends Coherence, and yet there’s no irrationality involved in this case. Setiya and others accommodate such cases by formulating the requirement of means–ends coherence so that it applies only when one believes one will X only if one now intends to Y.46 As he puts it, “it is only when you believe a decision is necessary that the instrumental principle gets a grip.”47 And so the match striker doesn’t come out to be means–ends incoherent. However, I worry that this fix is too restrictive. Intuitively, one can display instrumental irrationality even before one believes a decision is necessary. Consider the following example from Korsgaard: You want to ride on this immense roller coaster but you are prevented by terror. Every night of the carnival you go and look at it, get in line for a ticket, and then lose your nerve and shuffle meekly away. You don’t think riding the coaster is essential to your overall good. Maybe you even think it’s 45  Setiya 2007, p. 668. 46  Here’s the formulation: “You should [if you intend to do E and believe that you will do E only if you do-M-because-you-now-intend-to-do-M, intend to do M]” (p. 668). The “now-intend” ensures that the requirement doesn’t apply to cases of self-trust. John Broome also mentions such cases (2013, p. 162) and, along the same lines, formulates his synchronic “Instrumental Requirement” to avoid such counterexamples: “Rationality requires of N that if (1) N intends at t that e, and if (2) N believes at t that, if m were not so, because of that e would not be so, and if (3) N believes at t that, if she herself were not then to intend m, because of that m would not be so, then (4) N intends at t that m. (2013, p.  159) The addition of “then” in (3) ensures that the requirement doesn’t apply to such cases of self-trust. 47  Setiya 2007, p. 668.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Introduction  23 risky and a little foolish. But you’ve made up your mind to do it. And all you have to do is buy a ticket and get on – only you can’t bring yourself to.48

The correct thing to say about this case—and what Korsgaard herself says—is that your fear is making you irrational, preventing you from intending the means thought necessary for your end. But if we formulate the instrumental principle so that you aren’t means–ends incoherent until the last moment— the moment you believe a decision is now necessary—perhaps as the gate is closing on the final ride on the last day of the carnival, then we’re unable to deliver that verdict. Indeed, it seems that on Setiya’s formulation the requirement wouldn’t even kick in until, by your lights, it’s too late for you to escape your instrumental irrationality by coming to intend the means. If you’re irrational only when you believe that you must now intend the means to achieve you end, and you don’t now intend the means, then by your lights it’s too late: intending now is necessary and you don’t intend now. The only rational way for you to escape your incoherence at that point would be to give up your end. But that’s at odds with the intuitive thought—we’ll return to this intuitive thought in discussing the motivations for the “wide-scope” formulation of means–ends coherence in Chapter  3—that there’s more than one way to remove oneself from a state of instrumental incoherence. For instance, in Korsgaard’s ex­ample, you could remove yourself from instrumental incoherence either by changing your mind about riding the coaster or by overcoming your fear, buying the ticket, and getting on board. There might be other less restrictive ways to accommodate self-trust cases. Here’s one possibility: Revised Means–Ends Coherence:  Rationality requires that [if one intends to X, and one believes one will X only if one intends to Y, then one intends to Y or trusts that one will].49 In this formulation, we leave the instrumental belief alone, but the added disjunct in the consequent of the bracketed conditional ensures that the 48  Korsgaard 2008, pp. 40–1. 49  We should understand “trusts that one will” to allow for one to trust that the intention will be formed at the appropriate time, under the appropriate circumstances. I may intend to ride the roller coaster, believe that buying a ticket is necessary, but look around for free tickets, since, even though it’s highly improbable that there are free tickets, there’s no cost in my looking around. There’s no ir­ration­ ality here. I here trust that I’ll intend to buy the ticket when my search turns up empty. Thanks to a reader for this case.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

24  Instrumental Rationality self-trusting match striker does not violate the rational requirement.50 But we can declare irrational those who (perhaps because of fear, idleness, or depression) don’t intend the means nor trust that they will. And since the instrumental requirement gets a grip earlier on, we can allow for the possibility of rationally exiting such states of instrumental incoherence by intending the means. There may be other ways to provided the needed fix—that is, the fix of allowing for the possibility of rational self-trust without preventing the requirement from kicking in only when there’s just one way out remaining— but I think I’ve shown that we do not need to, and should not, restrict the instrumental belief relevant to means–ends coherence in the way Setiya suggests. In summary, I’ve provided some initial defense of the instrumental belief relevant to Means–Ends Coherence. First, we noted that one could hold this belief while at the same time believing one could be mistaken about it. Second, we noted that this belief need not figure into the agent’s reasoning. Third, we noted that an alternative, the X→Y Formulation, falters when it comes to foreseen side effects, while our formulation does not. Fourth, we noted that there’s another way of dealing with cases of self-trust that wouldn’t involve restricting the relevant instrumental belief in the way Setiya suggests. In what follows, I’ll continue to work with our initial formulation, Means–Ends Coherence, since it is simpler and the main arguments in this book won’t depend on the particular details of how we formulate the requirement. But at least we’ve seen some initial motivation for working with the formulation with which we’re working, and some of the issues that would be relevant to the task of arriving at a more precise formulation.

1.5 Conclusion In this introductory chapter, I’ve set out the plan for what follows, and explained some of the historical and contemporary bases of philosophical interest in means–ends coherence. I’ve also made several points relevant to the formulation of a requirement of means–ends coherence, especially regarding the instrumental belief in the requirement.

50  We can also assess the rationality and reasonableness of one’s self-trust. One could satisfy this requirement by trusting oneself in a way that violates some other requirement of theoretical rationality.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

2 Bootstrapping Let’s start by considering how intentions might be normatively significant. Many philosophers agree that in forming an intention, one makes applicable certain requirements of rationality, such as a requirement to avoid having inconsistent intentions, and a requirement to intend the means believed to be necessary for doing what one intends to do. As we noted, there is a debate about whether these requirements of rationality are normative—that is, whether they are such that we ought (or, more weakly, have a reason) to comply with them. If these requirements are normative, intentions would then have an indirect normative significance in that they would put in place rational requirements with which we ought, or have a reason, to comply. I’ll set that issue aside for now. (We’ll return to it in Chapter  5.) But there might be a more direct way in which intentions have normative significance: perhaps intending to X provides one with a reason to X, or a reason to take, or intend to take, steps that promote one’s X-ing. This chapter will consider whether intentions provide such reasons. The question of whether intentions provide reasons, while important in itself as a topic within the literature on practical reason, is also important for the overall argument of this book, since one of the main approaches to means–ends coherence (Normative Disjunctivism, which we’ll discuss in Chapter 4) would be much more plausible if we allowed that intentions can “bootstrap” reasons into existence. But if we reject such bootstrapping as il­legit­im­ate, the case against Normative Disjunctivism is rather strong.1 Let’s start by considering some views about the relationship between intentions and reasons. Some philosophers, including Mark Schroeder and Christine Korsgaard, have argued that intending an end provides one with a reason to take the means. Schroeder defends “Schema Detach: For all agents x and act-types m, if x has end E and m is a necessary means for x to accomplish E, 1  Additionally, it’s commonly thought that a narrow-scope interpretation of means–ends coherence combined with the thesis that rationality is normative results in im­plaus­ible bootstrapping, forcing either the rejection of the narrow-scope interpretation or the rejection of the thesis that rationality is normative. For the purposes of that objection, it’s important that we arrive at an understanding of why bootstrapping is problematic. Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

26  Instrumental Rationality then there is a reason for x to do m.”2 Korsgaard writes, “If doing a certain action is necessary for or even just promotes a person’s aims, the person obviously has at least a prima facie reason to do it.”3 And Korsgaard goes on to claim that you must think the fact that you will the end is what gives you a reason to do what promotes that end, and she explains this idea as follows: “It means your willing the end gives it a normative status for you, that your willing the end in a sense makes it good.”4 These are versions of what I’ll call the “intentions-provide-reasons view.”5 Others, including Ruth Chang and T. M. Scanlon, have held that intentions provide reasons only when the other applicable reasons fail to issue some ­specific, non-disjunctive, verdict.6 Perhaps you need to buy just one can of tomatoes, and there are two equally good cans directly in front of you on the supermarket shelf. (You have sufficient reason to pick up each one, and conclusive reason not to pick up both or neither.) Or perhaps, to borrow Chang’s example, you are choosing between a career in philosophy and a career in the circus, and you would be happy and successful in either profession, and so you don’t have more reason to choose one over the other.7 Perhaps intentions here make a normative difference in that your choosing the can on the left gives you a reason to move your arm in that direction rather than toward the can on the right, and in that your choosing the career in philosophy gives you a reason to buy philosophy books rather than trapeze equipment. Your intentions here serve as “tie-breakers,” though that terminology is misleading, since if the two options were incommensurable—as perhaps they are in the example involving choice of careers—there wouldn’t, strictly speaking, be a tie to be broken.8 Keeping that in mind, I’ll nonetheless call this view the “tiebreaker view.” This view isn’t committed to every intention providing a reason to do what promotes it. And yet others, including Michael Bratman and John Broome, reject both the intentions-provide-reasons view and the tie-breaker view. As they see it, these views license the implausible “bootstrapping” of reasons into existence.9 2  Schroeder 2005, pp. 1, 6–11. For Schroeder, “[i]f Schema Detach is valid, then in virtue of having an end, you have reasons to take the means to that end” (2005, p. 1). See also Schroeder  2004, pp. 344–5. 3  Korsgaard 2008, p. 27. 4  Korsgaard 2008, p. 58. 5  See also Raz 1975, p. 490, and Verbeek 2014, p. 98, for versions of the intentions-provide-reasons view. 6  See Scanlon 2004, p. 236. See Chang 2009, 2013a, 2013b. While Scanlon defends a straightforward version of the tie-breaker view, the way in which the will generates reasons on Chang’s view is, as we’ll see, more complicated. We’ll discuss the views in more detail in Section 2.3. 7  Chang 2009b, pp. 249–51. 8  See Hsieh 2008 for a general discussion of incommensurability. 9  Bratman 1987, pp. 24–7, and Broome 2001, p. 98.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  27 However, it’s not clear what exactly the bootstrapping objection is—there’s no shared understanding of it in the literature—and doubts have been raised about whether it is a good objection to either of the views above.10 In this chapter, I’m going to argue against the intentions-provide-reasons view and the tie-breaker view. After some preliminaries (Section 2.1), I’ll argue (Section 2.2) that both views are vulnerable to the bootstrapping objection. (In Section 2.2.1, I’ll consider several inadequate versions of the bootstrapping objection. But, in Section 2.2.2, I’ll present what I take to be the best version of the objection—one which calls attention to how these views result in a normative theory which issues incoherent advice.) I then consider (Section 2.3) some arguments related to rational deliberation that Chang (Section  2.3.1) and Scanlon (Section  2.3.2) have advanced for versions of the tie-breaker view, and conclude that they do not succeed. Despite my rejection of these two views—the intentions-provide-reasons view and the tie-breaker view—I do think there is a way in which intentions are relevant to what one has reason to do. I outline this view, which I’ll call the “Promotional Significance View,” in Section 2.4. This view holds that sometimes whether you have a reason to take some particular insufficient means depends upon what good will come of it, and what good will come of it depends upon what else you will do besides taking that particular insufficient means, and what else you will do depends upon what you now intend to do. (There may be a temptation to conclude from intentions being relevant in this way that either the intentionsprovide-reasons or the tie-breaker view is true. But that would be a mistake.) On this view, the will isn’t a “source” of normativity, but may be relevant—in a fairly ordinary way—to whether normativity transmits from a source to a particular insufficient means.

2.1  Clarifying the Question We’re interested in whether intentions provide reasons. So, I’ll start by saying a bit about what intentions are, what reasons are, and what it is for an intention to provide a reason. In doing so, I hope to clarify how I’m understanding the question of whether intentions provide reasons. As we’ll see, the arguments that follow in this chapter won’t, despite the various philosophical controversies 10 For instance, both Christian Pillar  2013 and Matthew Noah Smith  2016 distinguish several v­ ersions of the objection and conclude that none of them work. (Neither considers the version of the objection I offer in Section 2.2.2.) Bruno Verbeek 2014 also argues that the bootstrapping objection isn’t as forceful as it appears.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

28  Instrumental Rationality about reasons and intentions, depend on controversial assumptions made at the outset. Let’s start with reasons. Our question concerns normative reasons—that is, the considerations that count in favor, or count against, performing some action, or having some attitude (belief, intention, fear, hope, etc.). We aren’t concerned here with motivating reasons—that is, the reasons for which one acts.11 We sometimes talk about intentions providing motivating reasons (e.g., “John’s reason for going to the convenience store was that he intended to buy some milk”), though of course not all intentions motivate one to action. But we should take care not to confuse motivating reasons with normative reasons. You could have a motivating reason that is not a normative reason: it could be that the reasons for which you act aren’t such that we would put them in the “pro” column when listing the “pros” and “cons” of your performing that action. And you could have a normative reason that isn’t a motivating reason: there could be some consideration in the “pro” column of which you are unaware, and so that consideration doesn’t motivate you to act. Can we say anything more informative about what normative reasons are? Many philosophers, including T.  M.  Scanlon and Derek Parfit, doubt that we can: I will take the idea of a reason as primitive. Any attempt to explain what it is to be a reason for something seems to me to lead back to the same idea: a consideration that counts in favor of it. “Counts in favor how?” one might ask. “By providing a reason for it” seems to be the only answer.12 It is hard to explain the concept of a reason, or what the phrase ‘a reason’ means. Facts give us reasons, we might say, when they count in favour of our having some attitude, or our acting in some way. But ‘counts in favor of ’ means roughly ‘gives a reason for’. Like some other fundamental concepts, such as those involved in our thought about time, consciousness, and possibility, the concept of a reason is indefinable in the sense that it cannot be helpfully explained merely by using words.13

11  Nor are we concerned with explanatory reasons more generally. We might say, “The reason the store is out of milk is that the shipment from the dairy farm was delayed.” Here, the delayed shipment is the reason the store is out of milk in the sense that the delayed shipment explains why the store is out of milk. It’s an explanatory reason. But it is not a reason for which anyone acts, and so it’s not a motivating reason, and it does not count in favor of the store being out of milk, and so it’s not a normative reason. 12  Scanlon 1998, p. 17. 13  Parfit 2011, p. 31. See also Parfit 2001, p. 18.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  29 Following Scanlon and Parfit, I’m going to here take the idea of a normative reason as primitive.14 However, we should take care to distinguish normative reasons (henceforth, “reasons”) from rational requirements. As I mentioned above, it’s widely accepted that there are rational requirements relevant to intentions, such as: Intention Consistency:  Rationality requires that [if one intends to X, then one doesn’t intend not to X] and Means–Ends Coherence:  Rationality requires that [if one intends to X, and one believes one will X only if one intends to Y, then one to intends to Y]. If one has the combination of attitudes prohibited by these requirements, one would be irrational. But it doesn’t follow from these requirements that an intention provides a reason for intending or taking the means. First of all, as we’ve noted, there is a controversy about whether the requirements of rationality are such that we have a reason to comply with them.15 For any system of rules—whether of morality, prudence, law, etiquette, chivalry, or rationality, we can ask whether the requirements are such that we have a reason to comply with them in general. And it may come out that we do not. Second, the requirements are “wide-scope” requirements in that the logical scope of “requires” ranges over a conditional.16 Consider Means–Ends Coherence. It doesn’t require that the means–ends incoherent agent have some specific attitude. Rather, what it requires is that she either not intend to X or not believe she will X only if she intends to Y or intend to Y. But, on the intentions-provide-reasons view, “reason” has narrow scope: we’re considering whether intending an end gives someone a reason to take some means, not

14  This might not be an insignificant choice on my part. Stephen Kearns and Daniel Star (2008, 2009) have analyzed the concept of a reason in terms of the concepts of evidence and ought, endorsing the following biconditional, which I’ll label “RAE”: “Necessarily, a fact X is a reason for an agent N to F if and only if X is evidence that N ought to F.” (2008, p. 37). If we also assume that people by and large intend to do what they ought to do, we could argue that someone’s intending to X is (defeasible) evidence that they ought to X, and so, by RAE, a reason for him to X. However, it’s not clear that someone’s intending to X would always be evidence he ought to X. (For instance, we might know someone to be such that he reliably intends to X when he shouldn’t X, in which case his intending to X is no evidence at all that he ought to X.) Additionally, it’s not clear that RAE is the correct analysis of ­reasons. (See Broome 2008b and Brunero 2009a.) 15  See Broome 2008a, Kolodny 2005, Setiya 2007, Bratman 2009c, Southwood 2008. 16  See Broome 2000. I’ll defend these wide-scope formulations in Chapter 3.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

30  Instrumental Rationality a reason to either abandon the end or abandon the instrumental belief or intend the means. Third, for requirements of rationality, like Means–Ends Coherence, it matters what the agent believes about the means to her ends. If one is unaware that a certain means is necessary for achieving one’s ends, one need not be irrational in not intending that means. But defenders of the intentions-providereasons view, and the tie-breaker view, think that we have reason to do what actually promotes—not what we believe promotes—our ends. Recall the passages from Schroeder and Korsgaard at the start of this chapter. On their views, one would have a reason to do what in fact promotes one’s ends, regardless of whether one is aware of it. More could be said on how to distinguish rational requirements from ­reasons, but it should be clear that it doesn’t follow from the fact that an intention makes applicable a rational requirement, like Means–Ends Coherence, that it also provides a reason to intend or take the means. Let’s now consider “provide.” In asking whether intentions provide reasons, we can remain neutral on the question of whether the intention itself is the reason, or instead something else is the reason, but that reason is put in place by one’s intention. For instance, one might hold that the fact that attending law school promotes Jack’s becoming a lawyer is a reason for Jack to attend law school, but that fact is a reason only because Jack intends to be a lawyer. In this case, Jack’s intention to be a lawyer isn’t itself a reason, but it explains why some other fact is a reason. Additionally, it’s worth noting that there is much controversy about whether reasons are psychological states, facts, propositions, states of affairs, or some combination of these options.17 These worries about the metaphysics of reasons would distract us from our question concerning the normative significance of intentions. In asking whether intentions “provide” reasons, we can avoid committing ourselves to any views about what the reasons are. What we’re interested in is whether it’s true in general that intending to do something makes it the case that there’s a reason—whatever that reason is—to, say, take the means. In other words, we’re interested in whether in general there’s a reason that exists because of the agent’s intention. This suggests a way to formulate the intentions-provide-reasons view, although I’ll leave the formulation incomplete for now, since there are several ways to complete it: Intentions Provide Reasons:  If an agent A intends to X, then A has a reason R to [. . .] because of her so intending. 17  See Dancy 2000 and Wiland 2012.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  31 (One could fill in the blank in various ways: “X” or “intend to take means that promote X-ing” or “take means that promote X-ing” or something similar.) What the formulation captures is that R—whatever R is, and whatever R is a reason for—is explained by one’s intending to X. We could do the same for the tie-breaker view: Intentions Provide Tie-Breaking Reasons:  If an agent A has sufficient reason to X, sufficient reason to Z, conclusive reason to not both X and Z, then if A intends to X, then A has reason R to [. . .] because of her so intending. That last clause is important. If there’s sufficient reason to X, and an agent intends to X, then of course there is a reason for her to X. But it need not be true that there is a reason for her to X because of her so intending. So much for “reasons” and “provide.” Let’s turn to “intentions.” The arguments in this chapter do not turn on any particular account of what an intention is.18 But it might be worth calling attention to those (hopefully uncontroversial) features of intentions that will play a role in those arguments. Two features of intentions will figure into the main argument in Section 2.2.2. First, intentions are attitudes for which there are reasons, and usually, though not always, the reasons to intend to act just are the reasons to act. For instance, the fact that it’s pleasant outside is both a reason for me to intend to go for a walk and a reason for me to go for a walk. And the fact that I’m tired is a reason for me to intend to stay in bed and a reason for me to stay in bed.19 Second, intentions are revisable in that one can change one’s mind. This is compatible with intentions having “stability” in the sense that we don’t, at least insofar as we’re rational, frequently re-open deliberation once we’ve decided upon a course of action. After all, one of the functions of intentions is to save deliberative resources—by, for instance, allowing me to decide on Monday what I’ll do on Friday, thereby freeing me up to deliberate about other things during the week—and intentions wouldn’t allow for this if they didn’t resist reconsideration to some extent.20 But that’s compatible with it being possible for us to change our minds.

18  See Setiya 2014 for an excellent overview of some philosophical issues related to intentions. It’s worth noting that we are primarily concerned here with understanding what is often call “prospective” or “future-oriented” intentions, rather than with understanding the nature of intentional (as opposed to unintentional) action, or the intention with which one acts. 19  These are examples of what are often called “object-given” reasons for intending. In contrast, “state-given” reasons for intending to X wouldn’t also be reasons to X. See Parfit 1997. 20  Bratman 1987, especially pp. 15–18.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

32  Instrumental Rationality And there’s an additional uncontroversial feature of intentions that’s ­relevant to the account presented in Section  2.4: intending to X normally involves being disposed to engage in further sub-planning, at the appropriate times, to facilitate one’s X-ing.21 For instance, someone who intends, as opposed to merely desires, to attend some conference will normally be disposed to plan to register for the conference, arrange for travel, book a hotel, and so forth. Beyond these uncontroversial assumptions, we need not here commit ourselves to any further claims about the nature of intentions.

2.2  The Bootstrapping Objection As an initial approach to our question of whether intentions provide reasons, we might appeal to our intuitions about examples. It’s natural to consider examples in which one intends to do something crazy, imprudent, or immoral. Suppose Kant’s poisoner—to be friendly, we’ll call him “Bob”—intends to kill his all of his colleagues and knows that to do so he must poison them. Intuitively, Bob’s intention provides him with no reason to poison his colleagues, or do anything else that would bring about their deaths. (Some authors have taken this objection to be a version of the bootstrapping objection.22 I would categorize it as a distinct objection. But the question of what label to give it isn’t very interesting.) There are some limitations to this objection. One limitation is that it has no force against Intentions Provide TieBreaking Reasons. According to that view, Bob’s intention does not provide him with a reason, since there must already be sufficient reason to pursue an end in order for one’s intention to make a normative difference, and there isn’t sufficient reason for him to kill his colleagues. A second limitation is that not everyone shares this intuition about Bob’s case. Those who defend the intentions-provide-reasons view, such as Schroeder and Korsgaard, would claim that there is a reason for Bob to poison his colleagues, but this reason is heavily outweighed by the reasons not to do so. Moreover, as Schroeder has argued, we can appeal to conversational pragmatics to explain why we might be misled into denying that Bob’s intention provides a reason for him to poison his colleagues.23 When someone says that there’s a reason to do something, we assume they’re talking about a rea-

21  Bratman 1987, especially pp. 15–18. 22  For instance, see Verbeek 2014, p. 88, and Smith 2016, pp. 2252–4. 23  See Schroeder 2005, pp. 7–8, and Schroeder 2007, pp. 84–102.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  33 son of significant weight—one worth thinking about and taking into account in deciding what to do. (Why bother to mention the lightweight ones?) This is why we’re inclined to think someone is speaking falsely when they say that there’s a reason for Bob to poison his colleagues: we assume they must be speaking of a weighty reason, since lightweight reasons wouldn’t be worth mentioning, and there is no such weighty reason. So, it would be better if we could find an objection which also targets the ­tie-breaker view and doesn’t rely upon controversial intuitions about whether immoral intentions, like Bob’s intention to kill his colleagues, generate reasons.

2.2.1  What Is the Bootstrapping Objection? The bootstrapping objection originates in Michael Bratman’s Intentions, Plans, and Practical Reason, but has been endorsed by many authors since then. Here is John Broome’s statement of it: The view that intentions are reasons is implausible. If you have no reason to do something, it is implausible that you can give yourself a reason, just by forming the intention of doing it. How could you create a reason for yourself out of nothing? Suppose, say, that you have no reason either for or against doing some act, and you happen to decide to do it. Now you intend to do it. So now, if intentions are reasons, you have a reason to do it. Since you have no contrary reason not to do it, the balance of reasons is in favour of your doing it. You now actually ought to do it, therefore. But this is implausible. It is implausible that just deciding to do something can make it the case that you ought to do it, when previously that was not the case. I shall call this ‘the bootstrapping objection’ . . .24

There are perhaps two points being made here: first, that it’s implausible to think you can “create a reason for yourself out of nothing,” and, second, that it’s implausible to think deciding to do something can make it the case that you ought to do it, when this previously was not the case. Start with the first point. If by “creating a reason for yourself out of nothing” we mean “bringing into existence through one’s voluntary activity some reason that didn’t exist before,” then it’s not clear that this is in any way 24  Broome 2001, p. 98. Broome presents the same objection in his book (2013, p.184). He traces the objection back to Bratman 1987, pp. 24–7. We’ll consider Bratman’s formulation of it in a moment.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

34  Instrumental Rationality problematic. Suppose you promise to read a draft of my latest paper. Your voluntary act has here made it the case that you have a reason to read my paper, where this reason didn’t exist before.25 (Moreover, we could imagine there was no reason for you to read my paper before, since I’m a terrible scholar and a bad colleague who will rudely dismiss your feedback on it.) If this version of the bootstrapping objection would also rule out the reasons generated by such promises, it’s surely too strong.26 Perhaps, one might reply, the problem with intentions providing reasons is that an intention is an “internal” mental state, and it’s implausible to think that such mental states would generate reasons. (This allows us to distinguish intentions from promises, since your promising to do something is a speech act, one which most often creates expectations in others. It makes some external difference.) But this will be of no help, since internal mental states can generate reasons. If I believe the CIA is recruiting my students to conspire against me, or if I intend to build a tin-foil hat to shield myself from mind control, these attitudes provide me with a reason to seek professional psychiatric help.27 Additionally, many philosophers hold that one’s desiring to do something can, at least in certain circumstances, bring a new reason into existence: a reason to do what would promote the satisfaction of that desire. (And Humeans go even further in thinking all practical reasons are to be explained in terms of the agent’s desires.28) But if we were to insist that in­tern­al mental states do not give rise to reasons, we would have to rule out these reasons as well. And that would be too strong. Perhaps there’s yet another reading of “creating a reason for yourself out of nothing” that wouldn’t prove too strong. But it’s not clear to me what it is. Let’s turn to Broome’s second point. Broome claims that it’s implausible to think that deciding to do something can make it the case that you ought to do it. But this seems to simply beg the question against the intentions-providereasons theorist, who thinks it is plausible that intentions can make a difference in this way. 25  It isn’t entirely your doing, since it’s standardly thought that there must be some appropriate “uptake” on my part for you to have this promise-based reason. But assuming there would be such uptake, your voluntary act creates a reason for you. 26  Christian Piller (2013, p. 622) also draws attention to promising in the course of arguing against a version of the bootstrapping objection due to Richard Holton (2004). Promising isn’t the only way to voluntarily alter one’s normative situation. Holly Smith (1997, p. 163) provides several other examples: a person may waive a right to something, do something which injures another person, release someone from a promise, move herself into a position where she couldn’t perform some act that she might otherwise be obligated to perform. 27  Broome himself acknowledges such reasons. See Broome 2013, pp. 81–2, for an example along these lines. 28  See Schroeder 2007 for an articulation and defense of the Humean theory.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  35 On another understanding of the bootstrapping objection, articulated by Garret Cullity, the problem isn’t that reasons are created “out of nothing” but that reasons are created out of a faulty response to reasons. Suppose there’s a single set of reasons that determines both how you ought to intend to act and how you ought to act. But how can intending contrary to what those reasons require you to intend thereby change what the reasons require when it comes to action? As Cullity puts it, “[t]he problem is rather that it is hard to see how a decision which is one, incorrect, response to a set of reasons could itself provide a further reason to add to that set, thereby transforming another kind of response—an action—into something that ceases to be incorrect.”29 But our example of promising suggests that this version of the objection is also too strong. Since I’m a terrible scholar and a rude and dismissive colleague, you have reasons not to read my paper. These reasons are also reasons for you not to promise to read my paper. But here we do think that a bad response to those reasons—your promising to read my paper—can add a reason that could potentially shift the balance of reasons in favor of your reading my paper. Yet another way to develop the bootstrapping objection, which can be found in Bratman’s original presentation of it, would be to consider side-by-side two examples of decision-making over time, but have one case involve an additional intention, and then note how this addition would arbitrarily change the reasonableness of a later decision, if intentions provided reasons. Here’s an example: Suppose I shouldn’t order any more food, but I go on to look over the dessert menu and decide, contrary to the balance of reasons, to order the chocolate cake, and do so. Now consider a slight variation: suppose again I shouldn’t order any more food, but I form the intention to order dessert, and I go on to look over the dessert menu and decide to order the choc­olate cake, and do so. In the first example, where I proceed in one step—by intending to  order the chocolate cake—my intending to order the chocolate cake is contrary to the balance of reasons, whereas in the second case, where I proceed in two steps—by intending to order dessert and then intending to order chocolate cake—my intending to order the chocolate cake isn’t contrary to the balance of reasons, on the assumption that my intention to order dessert provided a reason which tipped the scales in its favor. It seems odd to think that this extra step can make my decision to order the chocolate cake (and my ordering it) come out to be reasonable. As Bratman puts it, in discussing a

29  Cullity 2008, p. 63.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

36  Instrumental Rationality structurally similar example, it seems to “too easily let this decision and action off the hook of criticizable irrationality.”30 But there are two problems with this version of the bootstrapping objection. First, even though in the second case the decision to order chocolate cake comes out to be supported by the balance of reasons, we can still criticize the intention to order dessert. So, it’s not clear how big a cost we face in having to shift our criticism in this way. We are still able to criticize the agent—I’m not let off the hook—we just need to locate the criticism further upstream in my decision-making. Second, it’s open to the defender of Intentions Provide Reasons to simply accept that the extra step does in fact make a normative difference. After all, if the extra step involved a promise— suppose I promised you I’d order dessert—then we wouldn’t hesitate to think that the extra step made a difference. We’d just shift our criticism upstream, and say I shouldn’t have made that promise. The defender of Intentions Provide Reasons can insist that intentions provide reasons in the same way that promises provide reasons; in both cases, a later decision can be let off the hook. If there’s something problematic about letting a later decision off the hook in this way, it would also speak against promises providing reasons. At this point, it’s clear that what we need, and what I hope to provide in Section 2.2.2, is a version of the bootstrapping objection that wouldn’t also disallow promises from providing reasons. The problem with the intentions-provide-reasons view isn’t that it allows us to create reasons out of nothing, nor that it allows bad responses to reasons to provide reasons, nor that it allows later decisions to be let off the hook. If these really were the problem with the intentions-provide-reasons view, we’d have to say that promises don’t provide reasons either. But surely they do. So, we haven’t yet explained why there would be objectionable bootstrapping if intentions provided reasons. Another approach to the bootstrapping objection looks at an analogy with belief. John Broome has argued that in the same way that believing that P doesn’t provide you with a reason to believe the obvious logical consequences of P, intending an end doesn’t provide you with a reason to intend or take the necessary means.31 Let’s first consider belief. Since P is an obvious logical consequence of itself—that is, obviously, if P, then P—then if believing that P gives you a reason to believe the obvious logical consequences of P, then believing that P gives you a reason to believe that P.  And surely that is

30  Bratman 1987, p. 26.

31  Broome 2003, pp. 92–7.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  37 im­plaus­ible self-justification.32 So, we should reject the thought that believing that P provides you with a reason to believe the obvious logical consequences of P.33 In the same way, we should reject the thought that intentions provide reasons. However, it’s not clear how much traction such an argument will have against the intentions-provide-reasons theorist. Those who think that intentions provide reasons need not be committed to thinking that intentions provide reasons for themselves. They could hold instead that intending to X provides a reason to X, or to intend means that promote X-ing, or to take means that promote X-ing. (Recall again the views expressed by Schroeder and Korsgaard at the start of this chapter. They hold that intending to X provides a reason for one to take means to X-ing, without holding that intentions 32  The claim Broome is rejecting here is a claim about what reasons there are. Broome is rejecting the idea that in listing out the pros and cons of S’s believing that P, we should include on that list, in the pro column, that S believes that P. The rejection of this claim is compatible with a number of different claims about how people should track their reasons for belief. Gil Harman (1986, pp. 29–42) critiques views which require people to be aware of their reasons for believing as they do in order to be justified in holding their beliefs. For instance, the “foundations theory” holds that “people should keep track of their reasons for believing as they do and should stop believing anything that is not associated with adequate evidence” (p. 38). Harman argues, plausibly, that people very seldom keep track of their reasons—after all, doing so would involve excessive clutter in one’s mind—and it would be wrong to think that people are unjustified in almost all of their beliefs. It’s open to Broome to endorse such a critique, and also endorse the view that S could be justified in believing that P even though he’s lost track of his reasons to believe P. All of this is compatible with denying that S’s believing that P is a reason for itself. What is less clear is whether Broome could endorse Harman’s “negative coherence theory” or “general foundationalism,” according to which all of a person’s beliefs at a time are such that the person is “prima facie justified” in having them. (Harman 2001, p. 657). It depends on how we understand the relationship between prima facie justification and reasons. One possibility is to understand Harman as holding that reasons for belief aren’t necessary for prima facie justification. Pollock and Cruz (1999, p. 81) interpret Harman this way: “By definition, negative coherence theories hold that reasons are not required for the justification of a belief – beliefs are automatically justified unless you have a reason for rejecting them.” A second possibility is to understand Harman as embracing self-justification: your believing that P provides you with a reason to believe that P, one which is sufficient for prima facie justification. (Broome could endorse the first, but not the second, interpretation.) I won’t consider the formulation or merits of negative coherence theory in any detail here. (See Pollock and Cruz 1999, esp. 80–4, for further discussion.) But I think it’s false. Suppose I randomly come to believe that the number of primes between 123,456 and 654,321 is even. It seems right to say that I have no justification for this belief. But Harman’s theory says otherwise. Harman offers a reply: “If one believes not only P but also that one randomly came to believe P, the two beliefs are in tension and one has a reason to abandon at a least one of them” (2001, p. 658). But suppose I abandon, or never even had, the second belief. Suppose I have no view on whether I came by that belief randomly, perhaps because I’ve never thought about that. Still, it seems right to say that I have no justification for this belief. Moreover, the negative coherence theory itself holds that I have as much justification for holding this belief as I would have for randomly believing its negation instead—which, to my mind, is to say I have no justification at all. (On the theory, had I instead randomly believed that the number of primes between 123,456 and 654,321 is odd, I would be as justified in believing that as I am in believing the number is even.) 33  One way to respond to Broome here would be to develop a view according to which beliefs don’t self-justify, but are nonetheless reasons for having other beliefs. Christian Piller (2013, pp. 618–22) develops a Bayesian account along these lines in response to Broome. I won’t pursue this here.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

38  Instrumental Rationality provide reasons for themselves.) So, it’s open to the intentions-provide-reasons theorist to accept the point that intentions, like beliefs, don’t self-justify, but then deny that this is any objection to their view.34 However, this might give us a point to rely on in developing our argument. If everyone agrees that intentions don’t self-justify—that is, that intending to X isn’t a reason to intend to X—we could use this as a premise in an argument against the intentions-provide-reasons view and the tie-breaker view.35 I’ll now turn to that argument. My aim is to provide a version of the bootstrapping objection that targets both Intentions Provide Reasons and Intentions Provide Tie-Breaking Reasons without overextending to disallow promises from providing reasons. This new version of the bootstrapping objection alleges that both of these views are flawed in that, given certain assumptions, they provide us with a normative theory that issues incoherent advice.

2.2.2  Bootstrapping as Incoherent Advice We’re looking for a version of the bootstrapping objection that wouldn’t rule out promises providing reasons. So, a good place to start would be to think about some differences between promises and intentions. Here’s one: promises can’t be undone simply by changing your mind. If I’ve formed an intention to have lunch at Joe’s, I can change my mind and intend to eat at Ray’s instead. But if I’ve promised you that I’ll meet you at Joe’s for lunch, a change of mind won’t suffice to undo the promise. It’s already “out there.” To remove the reason provided by the promise, you’d have to release me from the promise, which I could ask you to do. But merely changing my mind won’t suffice.36 So, a rational agent deliberating about what to do will treat promises differently from intentions. Revising an intention will be seen as an available option in a way in which revising a promise is not. Now, recall a point mentioned earlier: the reasons for intending to act are usually also reasons for acting. Suppose you’re deliberating about whether to

34  Such reasons would be unusual, in that you would have a reason to X that isn’t a reason to intend to X. But we can concede such reasons to the intentions-provide-reasons theorist for the sake of argument. 35  I’ll say more in defense of this premise later on. 36  This feature isn’t present for self-promises. But that, in part, is why self-promising is so controversial. There are questions about whether self-promises are conceptually possible, and whether, if they are, they generate reasons. (See Rosati  2011 for further discussion.) In any case, I’m still in a position to hold that my version of the bootstrapping objection has the advantage of not ruling out the reasons provided by ordinary, uncontroversial cases of promising.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  39 book your flight to Chicago with Delta or with United. (To keep it short, I’ll say that you’re deliberating about whether to “fly Delta” or “fly United.”) The cheap ticket is a reason to intend to fly United and a reason to fly United. The comfortable seats are a reason to intend to fly Delta and a reason to fly Delta. Ordinarily, the reasons for intending to act and the reasons for acting will line up in this way. There are perhaps some exceptions to the rule. It’s sometimes said that in Kavka’s toxin puzzle, you have a reason to intend to drink a toxin at noon tomorrow (the eccentric billionaire will deposit a million dollars in your bank account if he detects this intention) but no reason to drink the toxin, since the money is already in, or not in, your bank account by the time you have the opportunity to drink it.37 However, this is one of those rare moments in phil­oso­phy where it is better to focus on the rule than the possible exceptions. Let’s work with our ordinary example involving the two airlines flying to Chicago. Delta has comfortable seats, but United is cheaper. Let’s suppose it’s a close call, but the financial considerations win out, and you ought to fly United. Since the reasons involved here are both reasons to intend to act, and reasons to act, we can conclude that you ought to intend to fly United, and you ought to fly United. So far, so good. Suppose that you intend to fly Delta. Well, you ought to change your mind and intend to fly United instead. But now suppose Intentions Provide Reasons is true. And let’s work with the following version of the view: Intentions Provide Reasons to Do:  If an agent A intends to X, then A has a reason R to X because of her so intending. (Note that Intentions Provide Reasons to Do isn’t committed to any kind of objectionable self-justification; it doesn’t hold that intending to X provides one with a reason to intend to X.) And let’s suppose that the reason provided by your intention to fly Delta has sufficient weight to tip the scales in favor of flying Delta. (If you think that intentions provide reasons of little weight, then suppose it’s a very close call, and that the reasons to fly United only very slightly outweigh the reasons to fly Delta.) Since your intention tips the scales in favor of flying Delta, you ought to fly Delta. But if we put this together with what we said earlier we get: you ought to change your mind and intend to fly United, and you ought to fly Delta. This is incoherent advice. Of course, you could do what you’re advised to do: you could become disoriented or 37 Kavka 1983.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

40  Instrumental Rationality confused, and then intend to fly United, but unintentionally fly Delta. But this still strikes us as incoherent advice for a normative theory to offer. If you ought to fly Delta, why ought you intend to fly United? One way to bring out the incoherence would be to think about the transmission of oughts to necessary means: if I ought to E, and M-ing is a necessary means for E-ing, then I ought to M.38 Suppose I ought to fly Delta. Plausibly, my not changing my mind is a necessary means to flying Delta. (Sure, as I noted, there’s still a very remote possibility of my flying Delta unintentionally if I change my mind. But usually our understanding of “necessary means” is compatible with such very remote possibilities. For instance, in saying that buying an airplane ticket is a necessary means to flying to New York, I can allow for the remote possibility of being kidnapped and taken there by a military jet overnight.) If that’s right, then the ought would transmit: I ought not change my mind. But then how can it be that I ought to intend to fly United instead?39 This incoherent advice is a direct result of our applying Intentions Provide Reasons To Do to a fairly ordinary example. So, that view should be rejected. One nice feature of this new version of the bootstrapping objection is that it  doesn’t overextend to rule out promises providing reasons. And that is because revising a promise, unlike revising an intention, isn’t an option available to an agent. Unlike my intention to fly Delta, my promise to meet you for lunch can’t be undone. I can deliberate about whether to keep the promise, to break it, to request release from it, etc., but not whether to undo it. So, unlike the intentions-provide-reasons view, the promises-provide-reasons view will not generate any such incoherent advice. So, we finally have a version of the bootstrapping objection that, unlike the versions considered in Section 2.2.1, doesn’t overextend to rule out promises as a source of reasons. It’s worth noting that the defender of the intentions-provide-reasons view could avoid giving incoherent advice by also endorsing the view, which Broome rightly thinks of as implausible, that attitudes self-justify—spe­cif­ic­al­ly, that intending to X provides one with a reason to intend to X. If, for instance,

38  See Setiya 2007, p. 656, Schroeder 2009, p. 234, Bratman 2009c, p. 424, and Kiesewetter 2015, p. 922, for similar versions of this principle. 39  In Chapter 5, we’ll consider some challenges to the transmission principle I’m relying on here having to do with cases in which there would be little point, and perhaps some cost, in one’s M-ing because one won’t take other necessary means to E-ing. But I think a suitably revised version of the transmission principle would do the trick, perhaps along the lines of the following: “if I ought to E, and M-ing is a necessary means for E-ing, and if I M, then I will also take other necessary means to E-ing, then I ought to M.” Such a principle would avoid the challenges, but still be strong enough for the purposes of this argument.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  41 the intention to fly Delta also provides a reason to intend to fly Delta, thereby tipping the scales, then Intentions Provide Reasons to Do wouldn’t issue incoherent advice here. We’d instead get the result that you ought to intend to fly Delta and you ought to fly Delta, which isn’t incoherent. So, we could present this bootstrapping objection as a dilemma for the intentions-provide-reasons theorist: either intentions implausibly self-justify or we get a normative theory that issues incoherent advice. Either way, we should reject the intentionsprovide-reasons view. I’ll now show how this new version of the bootstrapping objection applies to various possible versions of Intentions Provide Reasons and Intentions Provide Tie-Breaking Reasons. (If you can already see how this will play out, please skip this and the next three paragraphs.) Let’s start with: Intentions Provide Reasons to Intend Means:  If an agent A intends to X, and Y-ing is a means to X-ing, then A has a reason R to intend to Y because of her intending to X. But now consider some means that is sufficient for achieving the end. Suppose that you’re about to purchase tickets online, and once you click the red square, you’re flying Delta.40 And let’s suppose you’ve got the United website up in another window, and once you click the blue square, you’re flying United. Again, you intend to fly Delta. Well, you ought to change your mind and intend to fly United instead. But suppose Intentions Provide Reasons to Intend Means is true. Your intending to fly Delta provides you with a reason to intend to click the red square. Suppose again that this reason has sufficient weight to make a difference, tipping the scales in favor of intending to click the red square rather than the blue one, and so you ought to intend to click the red square. Again we get incoherent advice: you ought to intend to fly United and you ought to intend to click the red square. And it won’t help to instead hold: Intentions Provide Reasons to Take Means:  If an agent A intends to X, and Y-ing is a means to X-ing, then A has a reason R to Y because of her intending to X. On this view, your intention to fly Delta doesn’t provide a reason to intend to click the red square, but does provide a reason to click the red square. Assuming 40  For both Delta and United, the tickets are non-refundable. Once you click the square, you’ve paid for the ticket and can’t get your money back.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

42  Instrumental Rationality again that the reason has enough weight to make a difference, tipping the scales in favor of clicking the red square instead of the blue one, we’d again get incoherent advice: you ought to intend to fly United, ought to fly United, ought to intend to click the blue square, but ought to click the red square (since there is now an additional reason to click, but not to intend to click, the red square, provided by your intention to fly Delta). But, again, this is incoherent. You might reasonably ask, “If I ought to click the red square, why ought I intend to click the blue square?” Another virtue of this new version of the bootstrapping objection is that it applies also to the tie-breaker view. But we need to vary our example slightly to see this. Suppose you have changed your mind and now intend to fly United. But Southwest has just announced a deal on flights to Chicago. Southwest is superior to Delta, but, taking into account all of the relevant considerations (price, baggage fees, comfort, customer service, etc.), Southwest and United are equally good. In this example, you have sufficient reason to fly United and sufficient reason to fly Southwest. Of course, you don’t want to book more than one flight. Suppose that you’ve also pulled up the Southwest flight on your browser, and to book it, all you need to do is click the orange square. Let’s now consider some versions of the tie-breaker view that run parallel to the versions of the intentions-provide-reasons view discussed above, starting with: Intentions Provide Tie-Breaking Reasons to Do:  If an agent A has sufficient reason to X, sufficient reason to Z, conclusive reason to not both X and Z, then if A intends to X, then A has reason R to X because of her so intending. In our example, you are permitted to intend to fly United and permitted to intend to fly Southwest (but not permitted to do both). Suppose that you intend to fly United. That intention doesn’t provide a reason for itself that makes it such that you ought to intend to fly United. Rather, you are still permitted, but not required, to intend to fly United. Now, if Intentions Provide Tie-Breaking Reasons to Do is correct, then your intention to fly United provides you with a reason to fly United. Suppose that this reason has sufficient weight to tip the balance of reasons, so that it now comes out that you ought to fly United. But now we again have incoherent advice: you ought to fly United, but you’re permitted to intend to fly Southwest. This is puzzling. You might reasonably ask, “If I ought to fly United, then shouldn’t I intend to fly United? And doesn’t that mean it’s not permissible to intend to fly Southwest

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  43 instead?” or “How can it be that I’m permitted to intend to fly Southwest, but not permitted to fly Southwest (since I must fly United instead)?” These questions are different ways of formulating the thought that Intentions Provide Tie-Breaking Reasons To Do leads to incoherent advice. Similar problems will arise for versions of the tie-breaker view which have the intention provide a reason to intend or take some means. Consider: Intentions Provide Tie-Breaking Reasons to Intend Means:  If an agent A has sufficient reason to X, sufficient reason to Z, conclusive reason to not both X and Z, then if A intends to X, and Y-ing is a means to X-ing, then A has a reason R to intend to Y because of her intending to X. Here, your intention to fly United provides a reason to intend to click the blue square. Suppose again that it tips the scales so that you ought to intend to click the blue square. Again, we get incoherent advice: you ought to intend to click the blue square, but you’re permitted to intend to fly Southwest instead. You might reasonably ask, “If I ought to intend to click the blue square, which would make it such that I fly United, how can it be that I’m permitted to intend to fly Southwest instead?” Similar problems would arise for: Intentions Provide Tie-Breaking Reasons to Take Means:  If an agent A has sufficient reason to X, sufficient reason to Z, conclusive reason to not both X and Z, then if A intends to X, and Y-ing is a means to X-ing, then A has a reason R to Y because of her intending to X. Here, your intending to fly United doesn’t provide a reason to intend to click the blue square, but does provide a reason to click the blue square. And so it would come out that you are permitted to intend to fly Southwest, permitted to intend to click the orange square, but not permitted to click the orange square, since you ought to click the blue square instead. (We’re here assuming that the intention to fly United provides a reason to click, but not to intend to click, the blue square, and that reason shifts the balance of reasons in favor of clicking the blue square.) Again, we have incoherent advice. You might reasonably ask, “If I ought to click the blue square, then shouldn’t I also intend to do so? But that would mean I’m not permitted to intend to click the orange square!” There’s one final point to make about the main argument in this section. As I’ve noted, the way to avoid issuing incoherent advice is to explicitly embrace the idea that intentions self-justify. On that view, returning to our original

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

44  Instrumental Rationality case, we could say that once you intend to fly Delta, your intention provides a  reason for itself which tips the scales so that you ought to intend to fly Delta and ought to fly Delta, and there’s no incoherence there. And I assumed that both proponents and opponents of the intentions-provide-reasons view would agree that such self-justification is implausible. It’s implausible to think that by intending to X one provides a reason to intend to X.41 However, one might wonder whether this allegedly implausible view might receive some support from the observation that once you’ve made up your mind, there are often costs involved in changing it. For instance, once you intend some option, you might invest time and resources in that option which would all be for nothing if you then abandon the intention. Additionally, abandoning the intention might require you to reopen deliberation on the question, which takes up time and energy that could instead be devoted to other projects. So, we need to consider whether such “deliberative costs” provide support for the idea that intentions self-justify. I don’t think that they do. First, note that such costs won’t always be present. I could form an intention to do something without investing any further resources, including mental resources, in that project.42 Likewise, I can simply change my mind without having to reopen deliberation on a question. And even if I had to reopen deliberation, I might not have anything better to think about. Perhaps I would have instead spent the time thinking, as Charles Peirce did, about the advantages and disadvantages of various train routes I don’t plan on taking, just to pass the time.43 Moreover, in cases in which the deliberative costs are already sunk—suppose I just happen to start thinking again about the comfortable seats and the expensive airplane ticket while waiting in line for coffee—we’d be unable to appeal to such costs to explain why my intention provides a reason for itself in those cases. And, lastly, even if there are deliberative costs in abandoning an intention, such costs aren’t standardly thought to provide reasons to maintain the intention. For instance, consider a variation on Williams’s famous gin and gasoline case in which the only way to discover that the glass which is reasonably believed to contain gin actually contains some odorless, colorless poison is by conducting some very expensive and time-consuming test.44 Suppose you intend to drink it. We’re

41  There might be very rare cases in which your having some specific attitude is a reason for your having that attitude. Perhaps your believing that you have beliefs is a reason for you to believe you have beliefs. But it’s implausible to think that in general believing that P provides a reason for believing that P, and intending to X provides a reason for intending to X. 42  Broome (2013, p. 184) makes a similar point. 43  The example is from Peirce 1878, p. 290. 44  Williams 1981, p. 102.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  45 comfortable saying that you ought to change your mind and not drink it, even though the only way for you to do so rationally is by taking on the costs of conducting the test. In the same way, we should be comfortable saying that someone who intends against the balance of reasons ought to change his mind, even when the only way for him to do so rationally is by taking on significant deliberative costs. Such costs don’t affect the balance of reasons. (The point here doesn’t depend on there being a significant health risk in drinking the contents of the glass. Even if doing so resulted only in some very mild displeasure, we’d say that he ought not intend to drink the contents of the glass.) There’s yet another, more controversial, way to argue that intentions ­self-justify. One could argue that there’s a diachronic rational requirement of intention persistence,45 and, since rationality is normative, your intention (perhaps along with this rational requirement) provides you with a reason to persist in your intention. But both of the central premises in this sketch of an argument are highly controversial. First, it’s controversial whether there’s a rational requirement for you to persist with your intentions. I do not think there is one. Suppose I intend to order the tofu ramen noodle bowl, but, as the waiter approaches, I change my mind and order the chicken ramen noodle bowl instead. I just happen to change my mind. That’s not irrational. (Perhaps I would be irrational if I continued to flip-flop between the chicken and the tofu to the point where I’m unable place an order, or unable to pay any attention to the dinner conversation. But there’s nothing irrational about a onetime change of mind.) Nor do I think such changes in mind are acceptable only if there’s new information, or only if one has reopened deliberation. We could suppose that I had all the relevant information at the start, and de­lib­er­ ation leads me to think the two dishes are equally good. I then initially decide on the tofu. As the waiter approaches, without reopening deliberation, I change my mind and order the chicken. Again, there’s no irrationality here. I’ve simply changed my mind, swapping out one permissible (as I see it) option for another. Second, the normativity of rationality is controversial, as we noted earlier. But perhaps we can make the argument more directly, instead of proceeding via the normativity of diachronic rational requirements.46 Perhaps we could simply insist that intending to X provides one with a reason to persist in 45  See Broome 2013, Ch. 10, for a formulation of a requirement of intention persistence and relevant discussion. For a more general discussion defending the existence of diachronic rational requirements, see Abelard Podgorski’s (2016) reply to Brian Hedden’s (2015) “time-slice rationality.” 46  Thanks to a reader for this suggestion.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

46  Instrumental Rationality that intention to X. However, this view is in tension with the plausible idea that it’s permissible to change one’s mind in tie-breaker cases. Let’s consider an example. On the proposed suggestion, my intending to order the tofu provides a reason to stick with that intention. Prior to forming any intention (at t1), the two possible intentions I could go on to form (an intention to order the tofu and an intention to order the chicken) are equally reasonable. After I form the intention (at t2) to order the tofu, there’s an additional reason to intend to order the tofu (at t3), tipping the scales in favor of continuing to intend to order the tofu (at t3), and so that’s what I ought to do (at t3). And so I proceed impermissibly when I change my mind (at t3) and instead intend to order the chicken that, to my mind, is just as good as the tofu in every respect. That’s an implausible result. There need not be anything wrong with such changes of mind. (Of course, this thought is compatible with thinking there could be something wrong with excessive flip-flopping.) Let’s sum up. In this section, I’ve proposed a new version of the bootstrapping objection. This version doesn’t rest on controversial intuitions about the non-existence of reasons, nor does it overextend to disallow promises from providing reasons. I’ve argued that it poses a serious problem for both the intentions-provide-reasons view and the tie-breaker view. Such views either will allow for intentions to implausibly self-justify or will issue incoherent advice.

2.3  Rational Deliberation and Tie-Breaking Reasons I’ve given an argument against the tie-breaker view. But we should also ­consider what’s to be said in favor of the view. One promising strategy is to look at plausible claims about the rationality of deliberation in tie-breaker situ­ations, and try to build an argument from those claims to the conclusion that intentions provide reasons in tie-breaker situations. I’ll here consider two such arguments, one from Ruth Chang and one from T. M. Scanlon. Despite the promise of this strategy, I don’t find either argument to be persuasive.

2.3.1  Rational Deliberation and Underdetermination To find support for the tie-breaker view, one might look toward the nature of rational deliberation in cases where an agent knows that what she should do is underdetermined by reasons, as in the cases of career choice mentioned

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  47 earlier. Chang presents an argument along these lines. But before getting to that, we should consider her unique version of the tie-breaker view, which she labels “hybrid voluntarism.” Voluntarists think that the will is a source of normativity. Chang contrasts voluntarism both with externalism, which holds that there are irreducible “external” normative facts about which non-normative facts are reasons for which attitudes and actions, and with internalism, which holds that “internal” facts about our (passive) desires and dispositions provide reasons for actions. Here she explains the contrast: normative voluntarists locate the source of normativity in us, but not in our passive states. Rather, normativity has its source in something we do, and, in particular, in our active attitudes of willing or reflective endorsement. By willing something, that is, by actively engaging our volition, we can give a consideration the normativity of a reason. So, for example, if we will a law or principle according to which we avoid pain, the fact that an experience is painful can thereby be action-guiding.47

However, Chang is worried about versions of voluntarism, including Korsgaard’s, that would allow for crazy, imprudent, or immoral intentions to generate reasons.48 But hybrid voluntarism (also called “hierarchical vol­un­ tar­ism”) doesn’t have this problem. Hybrid voluntarists distinguish two classes of reasons: “given” reasons and “voluntarist” reasons, which are ordered hierarchically, so that it’s only when our given reasons have “run out”—that is, when the given reasons “fail fully to determine what you have reason to do”—that one can create voluntarist reasons through willing.49 For instance, suppose Bethany is choosing between careers in law and academia, but the given reasons (reasons related to likely job satisfaction, flexibility, security, compensation, happiness, etc.) do not determine which career she should pursue. Since her given reasons have “run out,” she can now create reasons to pursue one career over the other through willing.

47  Chang 2009, pp. 244–5. See also 2013a, pp. 74–6, and 2013b, pp. 164–9, for further discussion of the contrast between externalism, internalism, and voluntarism. For another defense of the hybrid view, see Behrends 2015. 48  See Chang’s discussion of Gerald Cohen’s Mafioso objection to Korsgaard. Surely a Mafioso who wills the death of his enemies does not generate a reason to bring about those deaths. See 2009, p. 247; 2013a, p. 108; and 2013b, p. 175. 49  2013a, p. 104. See also 2013b, p. 177, and 2009, pp. 245–7. I understand Chang, in the quoted passage, to be referring to a situation in which there is no specific (non-disjunctive) action that one has most reason to perform.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

48  Instrumental Rationality Chang offers a unique view about the way in which someone creates ­reasons through willing. She argues that by “taking” some fact to constitute a reason, we make it so. She provides an example: “Suppose you take the dulcimer tones of the harp as a reason to play the harp. By taking the dulcimer tones to be a reason you can—under suitable conditions—make this consideration a reason for you to play. You can create new voluntarist reasons to play the harp through an act of will.”50 On her view, taking something to be a reason is an act of will; it should not be confused with believing of some consideration that it is a reason. As she puts it, while we discover our given ­reasons, we create our voluntarist reasons. Chang’s argument from rational deliberation starts with a puzzle: let’s suppose that Bethany has deliberated thoroughly about whether to pursue the legal or the academic career, and she has considered all the relevant reasons and their weights. And she is “practically certain” that the reasons have run out. Now, note that it would be odd for Bethany to then just pick one or the other career, in the way one might pick one of the two identical cans of tomatoes from the shelf. Instead of simply picking, Bethany is likely to continue to deliberate, and, moreover, it’s “perfectly appropriate” for her to do so. But how can that be? How can it be appropriate for her to continue to deliberate when she is certain that the reasons have run out?51 That’s the puzzle.52 Chang’s view offers us an explanation. On Chang’s view, there are two distinct stages of rational deliberation. In the first stage, one assesses the given reasons. If those reasons fail to determine which career to choose, “[t]here is a second stage of deliberation in which we can create new voluntarist reasons which, in conjunction with our given reasons, may deliver an alternative we have most reason to choose.”53 And so the two-stage model of deliberation offers an answer to our puzzle. It can explain why it’s appropriate for Bethany to continue to deliberate even when she is “practically certain” the reasons have run out: although she knows the “given” reasons have run out, de­lib­er­ ation also involves the creation of new reasons through willing, and that’s what is involved when she continues to deliberate. (It’s worth noting that

50  Chang 2009b, p. 247. 51  Chang 2009b, p. 250. 52  At times, Chang seems concerned with a different puzzle, in which we must “explain how we can have most reason to choose an alternative when our (given) reasons have run out” (2009, p. 263). However, if that’s the puzzle, it’s hard to see how it works as an argument for hybrid voluntarism, since it seems to already assume its truth. Those who are not already convinced by hybrid voluntarism will deny that this needs to be explained, because they will deny that it’s true. On the alternative picture, if the given reasons have run out, it’s not the case that we have most reason to choose a particular alternative; rather, we have sufficient reason to choose each one. 53  Chang 2009, p. 256.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  49 Chang’s view of deliberation differs from what Joseph Raz calls the “classical conception of agency” where given reasons determine which options we have sufficient reason to pursue, and then the role of the will is to select among those “rationally eligible” options. In the second stage of Raz’s view, the will doesn’t give rise to any further reasons, but just selects among those options one has sufficient reason to pursue.54) However, there’s a competing explanation of Bethany’s continued de­lib­er­ ation that we should consider. Unlike decisions about canned tomatoes, it can be very costly to make a mistake when choosing a career, buying a house, deciding who you want to spend your life with, and so forth. But deliberation is cheap. There’s likely to be available time in the day—even if it’s just when one is stuck in traffic or procrastinating—to revisit a question and think through the reasons once again. So, even with a high credence that the r­ easons have run out, it could make sense to think it over one more time, given that deliberation comes at little or no cost, and the stakes are high.55 On this model, Bethany never enters Chang’s second stage of deliberation in which she creates reasons through willing. Rather, everything proceeds at the first stage, where she again reviews the “given” reasons and their weights, despite her confidence that they have indeed run out. And, of course, it could be that she is surprised to discover some new reason, or to come to see that some reason has greater weight than she initially thought. When it comes to explaining Bethany’s continued rational deliberation, I  think we should prefer our explanation. One advantage is that it explains why such further rational deliberation isn’t limited to cases of under­deter­min­ ation. Suppose Claire is practically certain the balance of reasons supports her marrying Bethany, but continues to go through the pros and cons. We can

54  Raz 1999a, pp. 47–8. 55  It’s tough to imagine someone having a very high credence that the reasons have run out on some practical question. After all, it seems very easy to be mistaken about such a thing. But one might worry that on the theory offered here, we wouldn’t be able to make sense of Chang’s intuition that it’s appropriate for Bethany to continue to deliberate if we suppose Bethany has a credence of one that the reasons have run out, and there’s some cost to continued deliberation. (As an aside, I think it’s appropriate to see such deliberation as coming at a cost only if it prevents one from engaging in other valuable deliberation. But lots of us don’t have our mental lives so tightly scheduled so that deliberation about one matter comes at the expense of deliberation about another.) On this objection, we’d have to concede that on our theory it would be inappropriate for Bethany to continue to deliberate, even though the stakes are high. And so Chang’s intuition would here be unexplained by our theory. In reply, I think that what matters for the question of whether it’s appropriate for Bethany to continue to deliberate is not the credences she has, but the credences she should have, given the evidence. And Bethany, in this variation on the example, is overconfident that the reasons have run out. And so it would be appropriate for her to have less confidence they have run out, and, given that the stakes are high and deliberation is cheap, to continue to deliberate. And so we can explain Chang’s intuition. Thanks to an anonymous referee for raising this concern.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

50  Instrumental Rationality explain this: since the stakes are high, and there’s little or no cost to further deliberation, it’s not inappropriate for Claire to continue to deliberate, despite her high level of confidence that she ought to marry Bethany. But Chang’s view cannot explain why it would be reasonable for Claire to continue to deliberate, since there’s no “second stage” of deliberation available, since the reasons at the first stage haven’t “run out.”56 Additionally, one might worry about whether the two-stage explanation gets the phenomenology of deliberation right. As you become more confident that the reasons have run out, but continue to deliberate, does the nature of that deliberation change so drastically? It certainly doesn’t feel as though one moves from discovering and weighing reasons to creating reasons by willing.57 Lastly, the view that we can create reasons by willing some consideration to be a reason is subject to versions of Euthyphro-style objections similar to those leveled against voluntarist views which have normativity spring not from our wills, but from God’s. It would help to work with a concrete example here. Chang tells us that in an underdetermined career-choice case, “you might take the challenge of understanding a difficult philosophical problem as normative for yourself, thereby giving yourself a voluntarist reason to choose the philosophical career.”58 Let’s suppose you’ve willed some specific philosophical problem to be a reason to become a philosopher. But now you 56  In the same paper, Chang (2009, pp. 263–7) offers a second argument for the tie-breaker view. We are asked to imagine a person and his doppelganger in a nearby possible world. They both share the same given reasons. (If given reasons include desire-based reasons, then assume they have the same relevant desires.) There are at least two hobbies they have sufficient reason to pursue: collecting stamps and watching trains. (We are also to assume that one hasn’t made any prior investments, like purchasing a stamp album, researching trains, or changing the circumstances in any way that would favor one hobby over the other.) Chang argues that, without changing given reasons, one person can make it such that she has a “rational identity” such that she has most reason to collect stamps as a hobby, while the doppelganger can create a rational identity such that she has most reason to watch trains instead. She then argues that hybrid voluntarism best explains this. But this description of the case is question-begging. Her opponent won’t accept that two people with the same given reasons nonetheless have reason to pursue different hobbies. Of course, it could be that these given reasons make it such that the person and doppelganger each have sufficient reason to collect stamps and sufficient reason to watch trains, and one person chooses to collect stamps and the other chooses to watch trains. But one needs to argue for the possibility that such a person has more reason to collect stamps and the other has more reason to watch trains. That cannot be assumed. 57  Chang disagrees. In her view, when you are practically certain the reasons have run out, de­lib­er­ ation involves agonizing over the considerations, which “can be understood as deliberating over whether and which such considerations to create as new, voluntarist reasons for yourself.” (On her view, a given reason could also be created as a voluntarist reason.) See Chang 2009, p. 258. However, it’s not clear that we have to understand it in this way. Following our alternative explanation, one might be “agonizing” over the considerations in that one wants to be extra careful not to make a mistake since the stakes are high; nothing in that explanation appeals to our deciding which voluntarist reasons to create. 58  Chang 2009, p. 265.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  51 face a tricky question: “Did you do so for a reason or for no reason?” If for no reason—if it was arbitrary, or done on a whim—it’s not clear why anyone should take it seriously. If you had instead willed tomorrow’s weather forecast as a reason to pursue a philosophical career, should we take that seriously? Surely something so arbitrary would not be a source of normativity. But if for a reason—perhaps because challenging problems are worth exploring, or because you think you can make a contribution to the solution—it seems you are responding to reasons that already exist independently of your will. In short, in order to avoid the arbitrariness that comes along with “creating” reasons out of nothing, we’re forced to see the will as responsive to reasons that are already there, in which case, it’s not clear that we’ve left that first stage of deliberation after all.59 In summary, I don’t think the nature of deliberation in cases of under­deter­ min­ation will provide support for Chang’s version of the tie-breaker view. We’ve given an alternative explanation for why it would be rational to continue to deliberate when one is practically certain the reasons have run out. And our alternative explanation also extends to cases, like Claire’s, where one continues to deliberate when one is practically certain the reasons support some specific option. Additionally, it’s not clear that agents see themselves as transitioning from finding and weighing existing reasons to creating reasons through the will as they become more confident that the reasons have run out. And the idea of creating reasons through willing inherits a dilemma analogous to a dilemma facing theological voluntarism: the will is either arbitrary or it’s tracking reasons that are already there.60

2.3.2  From Rationality to Reasons We noted earlier (Section 2.1) that it doesn’t follow from the fact that rational requirements apply to your intentions that you have a reason to intend the means. As we noted there, the rationality of intending a means depends upon 59  It won’t help to concede that the will doesn’t create reasons, but maintain that it can add to or subtract from the weight of these already existing reasons. Suppose the challenging philosophical problem is a reason to become a philosopher, but you can make that reason weightier through an act of will. Suppose you do this. Did you increase its weight for a reason or for no reason? If for no reason, why take such an arbitrary whim seriously? (What if you had decreased its weight instead, or multiplied its weight by a thousand?) If it’s for a reason, it seems that you’d now be tracking independent facts about the relative significance of the reasons. The same dilemma reemerges here. 60  There are other arguments for hybrid voluntarism, including those which point toward its metaethical advantages, that we won’t have space to consider here. See Chang 2013a and 2013b, as well as Behrends 2015.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

52  Instrumental Rationality whether you believe it promotes your end, regardless of whether it actually does. But whether you have a reason to intend some apparent means depends upon whether it actually promotes your end, regardless of whether you believe it to do so. Rationality depends on what you believe; reasons depend on what is the case.61 However, rationality and reasons might be closely related. Consider this slight variation on an example from Parfit. Suppose you believe your hotel is on fire, and all the exits, except for the fire escape, are blocked. That makes it rational for you to leave by the fire escape. This is so even if your belief is false. In this case, it is rational for you to leave by the fire escape, but you have no reason to leave by the fire escape. But we can say this: if what you believed were true—that is, were the hotel actually on fire—you would have a reason to leave by the fire escape.62 In the course of an argument for the tie-breaker view, Scanlon discusses a similar example.63 If you believe tomatoes are a heath risk, it’s rational to stay away from tomatoes. (This holds even if you’re wrong about tomatoes being a health risk.) But what you believe, if true, would provide a reason to stay away from food containing tomatoes; if tomatoes were a health risk, you’d have a reason to stay away from them. This rationality–reasons connection suggests an argumentative strategy. If we can show first how some belief that P makes it rational for one to X, then all we have to add is that P is true in order for P to provide a reason for one to X. As Scanlon puts it, claims about rationality “turn into” claims about ­reasons when we add that the beliefs that ground the claims about rationality are true.64 He starts off his argument with a claim about rationality: Suppose Jones reasonably believes the following: 1.  There are a number of ends, call them G, H, and K, which he has sufficient reason to adopt (they are all eligible, to use Raz’s phrase). 2.  He has good reason not to adopt more than one of these goals. If Jones, believing (1) and (2), has adopted G as a goal, then he must, if he is not irrational, take the fact that x-ing would promote G as a reason for him to x.65

61  It’s assumed here, following Scanlon, that reasons are objective reasons. But one might reasonably think that the concept of a subjective reason is crucial to explaining instrumental rationality. We’ll return to this important idea in the third section of Chapter 5. 62  For this view about the relationship between rationality and reasons, see Parfit 1997, p. 99, and 2001, pp. 33–6. 63  Scanlon 2004, p. 233. 64  Scanlon 2004, p. 234. 65  Scanlon 2004, p. 236.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  53 Jones need not similarly take the fact that Y-ing would promote H as a reason for him to Y. But this claim about rationality could turn into a claim about reasons, if we add that the relevant beliefs are true: So far, this is just a claim about what Jones must, in so far as he is not irrational, take to be a reason. But if we add that Jones not only reasonably believes (1) and (2) but that these claims about the reasons he has are correct, then we are committed to the stronger claim that the things he must, in so far as he is not irrational, take to be reasons are in fact reasons.66

In this way, the truth of the relevant beliefs allows us to go from a claim about rationality to a claim about reasons. But one might wonder whether Jones’s believing (1) and (2) are at all relevant to his being required by rationality to take the fact that X-ing would promote G as a reason for him to X. Isn’t he required to do this simply because he adopts G as a goal, regardless of what he happens to believe about his options? Scanlon himself thinks so: adopting a goal gives rise to a difference in what an agent must, in so far as he has that goal and is not irrational, see as reasons. Someone who has adopted G as a goal, believes that x-ing is a necessary means to G and (while continuing to have G as a goal) does not count this as a reason for x-ing, is being irrational. This difference follows simply from principles that are constitutive of practical reasoning. Since adopting a goal involves taking oneself to have a reason to do what advances it, a failure to take oneself to have a reason to advance a goal one has adopted, and continues to hold, is irrational.67

But if the beliefs (1) and (2) aren’t relevant to Jones’s being rationally required to take the fact that X-ing would promote G as a reason for him to X—if all the work is being done by the fact that Jones adopted G as a goal—then can we go on to conclude that if those beliefs are true, then what Jones must, to be rational, take to be reasons are in fact reasons?

66  Scanlon 2004, p. 236. There are some nuances to Scanlon’s view that I won’t have space to discuss here. For instance, on his view, the claim that it’s rational for him to take himself to have a reason to promote G turns into a claim about second-order reasons: he has a reason to take himself to have a reason to promote G. See Brunero 2007 for further discussion. 67  Scanlon 2004, p. 235.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

54  Instrumental Rationality I don’t think so. Let’s go back to Parfit’s hotel example. Suppose you believe both that your hotel has a three-star rating and is on fire. The latter belief makes it rational for you to leave by the fire escape. The former belief does not. It’s irrelevant. And so it would be absurd to continue on to argue that if what you believed were true—that is, were the hotel to actually have a three-star rating—then that fact would be a reason to leave by the fire escape. That’s absurd because the belief has to be relevant to the rationality of your leaving by the fire escape for it to be plausible that the believed proposition, if true, would constitute a reason. But the problem with Scanlon’s argument is that the beliefs (1) and (2), like the belief that it’s a three-star hotel, are irrelevant to the claims about what it’s rational for the agent to do. In short, to go from rationality to reasons, the beliefs whose truth would provide reasons must be relevant to the rationality of one’s actions, and it’s not clear here that they are. There’s a second worry about Scanlon’s argument, having to do with his claim about what rationality requires—specifically, his claim that rationality requires Jones, who believes (1) and (2) and adopts G as a goal, to see the fact that X-ing would promote G as a reason to X (where there’s no similar requirement for Jones, who does not adopt H as a goal, to see the fact that Y-ing would promote H as a reason to Y). While many philosophers would agree that Jones’s adoption of G puts in place the usual consistency and coherence requirements with respect to G—for instance, a requirement to not also intend ~G, and a requirement to intend means believed necessary for G-ing— the claim that Jones is required to see this additional reason to X is, I think, incorrect. Return to the Southwest/United tie-breaking case, and suppose that I intend to fly Southwest, and don’t violate any of the usual consistency or coherence requirements with respect to this intention. For instance, I also intend to click the orange square which I know to be a necessary (and sufficient) means to flying Southwest. However, suppose further that I am aware of all the reasons related to the question of whether to fly Southwest or fly United (the considerations having to do with price, comfort, etc.) and I don’t think my intentions have added to my stock of reasons in any way. Specifically, as I see it, my intending to fly Southwest hasn’t put in place any additional reason to click the orange square—that is, any reason which wouldn’t be in place had I not formed any intention at all, or intended to fly United instead. As I see it, I still have no more reason to click the orange square than the blue square, just as it was before I intended to fly Southwest. It’s not clear to me that there’s any irrationality involved in this example.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  55

2.4  Promotion and Insufficient Means I’ll now explain one way in which intentions may be relevant to what one has reason to do. Let’s start with the familiar observation that a set of actions could be such that undertaking them would promote a valuable outcome without it being the case that an action within that set, undertaken in isolation from other actions in that set, would promote a valuable outcome. In the case of Professor Procrastinate, introduced by Frank Jackson and Robert Pargetter, it would be very good if Procrastinate were to accept a review invitation and write the review.68 But it doesn’t follow that there would be any value in his merely accepting the invitation and, say, procrastinating to the point that the review never gets done. Indeed, it would have been much better for Procrastinate to decline the invitation, allowing someone else to write the review, even if that person isn’t as well qualified to do so. At least then the book, which deserves a review, would get one. Whether anything good will come from Procrastinate’s accepting a book review invitation depends on what else he will do. And if he lives up to his name, it would be better for him to decline.69 If we keep in mind this familiar point about value, then we can understand one way in which intentions can have normative significance. Here are the basic components of the Promotional Significance View: it’s sometimes the case that whether you have a reason to take some particular insufficient means depends upon what good will come of it, and what good will come of it depends upon what else you will do besides taking that particular insufficient means—this is an application of the familiar point about value above—and what else you will do depends upon what you now intend to do.70 In this way, intentions are relevant to what one has reason to do. Let’s consider an example. Suppose David is applying to Prestigious Law School (PLS). There are several things he must do by March 1 to apply: take 68  Jackson and Pargetter 1986, p. 235. 69  See also the “problem of the second best” from Lipsey and Lancaster (1956–1957), which is discussed by David Estlund (2011, p. 398), who concisely summarizes their point as follows: “When there are several desiderata that are valuable as a package, if one of them is not satisfied, the value of the rest of them is thrown back into question.” 70  I proposed and defended an account along these lines in Brunero 2007, although I’m not entirely happy with how I presented the point there, for reasons that aren’t worth getting into. Niko Kolodny arrives independently at a similar account in his excellent paper “Aims as Reasons” (2011). Scanlon 2008 (see, for instance, pp. 13, 63–4) refers to the “predictive significance” of intention. I here adopt the terminology of “promotional significance” to highlight the particular predictive significance of intentions that’s most relevant in this context, where we are concerned with predictions regarding what good would be promoted by one’s taking some particular insufficient means.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

56  Instrumental Rationality the LSAT, register for the credential assembly service, submit transcripts, ­submit two letters of recommendation, and send in an application form, personal statement, and resume. He also must pay an $85 application fee. If, and only if, he does all of these things, will he have completed an application to PLS. I’ll note three points about the case of David’s application. First, although there is a set of actions that would suffice for applying to PLS, there is no single member in this set that suffices for applying to PLS. Each member of the set is an insufficient means. We might say that these insufficient means each “facilitate” one’s applying to PLS.71 But whatever terminology we use, what’s important to note here is that it’s not the case that if one takes some particular means, then one applies to PLS. Second, it seems that whether one should take any particular insufficient means in this set depends upon what good would come of it. Consider the question, “Should David write an $85 check to PLS?” A natural answer is: “Well, that depends. If the rest of the application isn’t going to be submitted, he’ll just be throwing away $85 by sending in the check. But, if he’ll be submitting the rest of his materials, then he should send in the $85 too, for only then will his application be considered.” Here, nothing good would come of sending in the $85 without the rest of the application, while something good would come of sending in the $85 with the rest of the application. Third, how likely David is to send in the other components of the application depends upon whether he intends to apply to PLS. If he has no intention of applying, or has made up his mind not to apply, it’s highly unlikely that he’ll send in the transcripts, letters, and so forth. But if he intends to apply to PLS, then it’s significantly more likely that he’ll send in the transcripts, letters, and so forth. Of course, there’s no guarantee that he will. But intending to do something normally disposes one to engage in the relevant sub-planning at the appropriate times, and such plans normally result in one’s performing the relevant actions that need performing.72 And so we can now see how intentions may be relevant to reasons. In this case, whether David has a reason to send in the check depends on what good would come from his doing so, and what good would come from his doing so depends upon what else he will do—this is the familiar point about value—and what else he will do depends upon what he now intends to do. On this view, the will isn’t a source of normativity—if you like such metaphors—but is relevant to how normativity flows from its source to some particular insufficient means. 71  The terminology of “facilitating steps” comes from Raz 2005. 72  Bratman 1987, pp. 2–3.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bootstrapping  57 (The “source” of normativity is the value of the outcomes—or, if you prefer a desire-based theory, the extent to which the outcomes satisfy intrinsic desires— and the will is relevant only insofar as it affects the prob­abil­ity of these outcomes obtaining.) The Promotional Significance View might also be able to explain many of the intuitions that lead philosophers (mistakenly, in my view) to endorse the tie-breaker view. After all, the same considerations that are at work in David’s case could very well be at work in Bethany’s case. Bethany is deciding between the career in law and the career in academia. Intuitively, once she decides to pursue the legal career she has a strong reason to take the LSAT—a reason she wouldn’t have had, had she instead remained undecided or opted for the academic career instead. The Promotional Significance View provides a way to explain this intuitive datum: whether anything good will come from Bethany’s taking the LSAT depends upon what else she will do—will she also do those other things necessary for becoming a lawyer, or will taking the test just be a waste of time, money, and effort?—and her intention to be a lawyer is relevant to whether she will do those others things. This explanation of the normative significance of Bethany’s intention doesn’t require that we endorse the tie-breaker view, and so isn’t vulnerable to the bootstrapping objection. Recall the problem with the versions of the tie-breaker view discussed in Section  2.2.2. The problem was that when intending an end provided a reason to intend, or take, some sufficient means (such as clicking the blue square, which was sufficient for flying United) we’d get incoherent advice. But the Promotional Significance View doesn’t hold that intending an end provides you with a reason to take sufficient means; it speaks only of particular insufficient means. This isn’t because we can’t specify what the sufficient means are in Bethany’s case. We can. It would just be a long conjunction of actions. We could specify a set of actions that together suffice for Bethany’s being a lawyer (SL), and a set of actions that together suffice for her being a professor (SP). The Promotional Significance View allows us to accept that Bethany has no more reason to pursue SL than SP. We aren’t therefore forced into saying anything like, “It’s permissible for her to change her mind and pursue the academic career instead, but she ought to take SL.” (And so we avoid the incoherent advice that, I argued in Section 2.2.2, is at the core of the bootstrapping objection.) But we do recognize that it doesn’t follow from the fact that SL promotes some valuable outcome that taking any particular member of SL promotes some valuable outcome, just as it doesn’t follow from the fact that Professor Procrastinate’s accepting and writing promotes some valuable outcome that his merely accepting promotes some

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

58  Instrumental Rationality valuable outcome. It matters what else he will do. And this fact provides an op­por­tun­ity for Bethany’s intention to be normatively significant, since it’s relevant to what else she will do. It’s also worth noting that, although I discussed the example of Professor Procrastinate above, which Jackson and Pargetter use in the course of their defense of actualism, I am not here proposing or defending a version of actualism. The thought is simply that if we consider the intuitions at work in the ordinary cases used to motivate actualism, like the case of Professor Procrastinate, we might see those same intuitions at work in the examples of David and Bethany, with intentions merely playing the role, in the latter cases, of increasing the likelihood of the relevant agents taking other insufficient means. If that’s right, then when we think intentions are normatively significant in these cases, it might not be in virtue of their being a “source” of normativity, as the volunarist would have it, but instead in virtue of their being relevant to the question of what good would come from an agent’s taking some particular insufficient means.

2.5 Conclusion Let’s sum up the main conclusions reached in this chapter. At the start, I considered some ways to understand the normative significance of intentions in terms of the provision of reasons, including views according to which the provision of reasons is restricted to tie-breaker situations. Following Bratman, Broome, and others, I think these views are committed to objectionable bootstrapping. But I hope to have offered a novel interpretation of the bootstrapping objection, one which avoids any reliance on potentially unreliable intuitions about the non-existence of reasons, extends to the tie-breaker view, and isn’t so strong as to rule out promises as a source of reasons. I’ve also considered two arguments that attempt to proceed from the nature of rational deliberation to claims about reasons in tie-breaker scenarios. I’ve argued that neither of these arguments succeeds. And I’ve tried, with the Promotional Significance View, to suggest an alternative, unproblematic account of one way in which intentions can be normatively significant.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

3 Scope In the previous chapter, the requirements governing intentions were given wide-scope formulations, in that “requires” had logical scope over a conditional. Consider: Means–Ends Coherence:  Rationality requires that [if one intends to X, and one believes one will X only if one intends to Y, then one intends to Y]. This formulation doesn’t require of the means–ends incoherent agent that she have any particular attitude. It doesn’t require that she, for instance, intend the means. Rather, it prohibits a combination of attitudes. Specifically, it prohibits someone from having the combination of intending to X, not intending to Y, and believing one will X only if one intends to Y. There are other ways one could formulate the requirement. Consider the following requirement, in which “requires” has narrow-scope: Narrow Means–Ends Coherence: If one intends to X, and one believes one will X only if one intends to Y, then rationality requires that one intend to Y.

This formulation, in contrast, does require the means–ends incoherent agent have some specific attitude. Someone who intends to X and believes she will X only if she intends to Y, but does not intend to Y, is required by Narrow Means–Ends Coherence to intend to Y. Yet another option would be to have the instrumental belief be outside the scope of “requires” while still having “requires” have logical scope over a conditional: Medium Means–Ends Coherence:  If one believes one will X only if one intends to Y, then rationality requires that [if one intends to X, then one intends to Y].1

1  See Way 2010, p. 223, and 2012, pp. 495–6. Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

60  Instrumental Rationality On this formulation, the means–ends incoherent agent is required to either not intend to X or intend to Y.2 Similar options are available for formulating other requirements of rationality. Consider, for instance, the enkratic requirement, which we could formulate in two ways: Wide Enkratic Requirement:  Rationality requires that [if one believes one ought to X, one intends to X.]3 Narrow Enkratic Requirement:  If one believes one ought to X, rationality requires that one intend to X. Likewise, we could consider requirements of theoretical rationality, like the requirement against believing what one believes to be insufficiently supported by the evidence. Niko Kolodny4 presents the following two possible formulations, and defends the narrow scope formulation: Wide Believed Insufficient Evidence:  Rationality requires that [if one believes that there is insufficient evidence that P, then one not believe that P]. Narrow Believed Insufficient Evidence:  If one believes that there is insufficient evidence that P, then one is rationally required not to believe that P.

2  And one could formulate requirements with the same logical structure as Medium Means–Ends Coherence by swapping out an attitude within the scope of “requires” for the one outside of it: First Variation on Medium Means–Ends Coherence:  If one intends to X, then rationality requires [if one believes one will X only if one intends to Y, then one intends to Y]. Second Variation on Medium Means–Ends Coherence:  If one does not intend to Y, then rationality requires [if one believes one will X only if one intends to Y, one does not intend to X]. No one, to my knowledge, has defended these variations, so I’ll set them aside. 3  These are rough formulations. A better wide-scope formulation is found in Broome 2013, p. 170: Enkrasia: Rationality requires of N that, if (1)  N believes at t that she herself ought that p, and if (2)  N believes at t that, if she herself were then to intend that p, because of that, p would be so, and if (3)  N believes at t that, if she herself were not then to intend that p, because of that, p would not be so, then (4)  N intends at t that p. But even here there might still be room for improvement. One worry is that (2) renders Enkrasia incapable of covering certain cases of akratic irrationality that intuitively should be covered by it. Consider Michael Bratman’s bicyclist who intends to stop by the bookstore on the way home, but is agnostic about whether he will, since he knows of his tendency to go on “autopilot” once on his bi­cycle. (See Bratman 1987, p. 32.) He doesn’t believe the relevant version of (2). And yet were he not to intend to stop by the bookstore, while believing he ought to stop by the bookstore, he would be irrationally akratic. Broome’s Enkrasia does not cover this case. 4  Kolodny 2005. p. 521. This is the requirement he labels “B-.”

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  61 And similar formulations are available for related requirements of rationality. In the debates between wide-scopers and narrow-scopers, many of these requirements are considered alongside one another. And, to some extent, I’ll follow suit in this chapter. But my main concern is with instrumental rationality in particular. And the aim of this chapter is to defend our wide-scope formulation, Means–Ends Coherence. (However, the arguments considered here might provide the added benefit of settling the question of how to formulate those other requirements as well.) In Section  3.1, I begin with some remarks about the nature of rational requirements, which I take to be common ground between the wide-scoper and narrow-scoper. Specifically, I note that rational requirements are taken to govern conflicts among our attitudes, can be interpreted either as s­ ynchronic or diachronic requirements, and are “local” requirements in that they remain in place even if one of the conflicting attitudes is party to another conflict. In Section  3.2, I present what I take to be the strongest argument for the wide-scope interpretations of rational requirements, and respond to two strategies of reply that narrow-scopers have employed against that argument. In Section 3.3, I consider what I take to be the strongest argument against the wide-scope interpretation—the so-called “asymmetry objection”—and consider the resources available to the wide-scoper to respond to various versions of this objection. I conclude, in Section 3.4, that we should adopt the wide-scope formulation of means–ends coherence.

3.1  Rational Requirements Requirements of rationality govern conflicts among our attitudes. For instance, a consistency requirement governs the conflict between one’s intending to X and one’s intending not to X. But, as we noted earlier, in add­ition to conflicts between the presence of one attitude and the presence of another, there can be conflicts between the presence of one attitude and the absence of another. The  enkratic requirement, for instance, governs the conflict between one’s believing one ought to X and one’s not intending to X. Both wide-scopers and narrow-scopers agree that if you have such a conflict among your attitudes, rationality requires something of you. But they disagree about what it requires of you. Narrow-scopers think you are required to intend to X, while widescopers think you are required to either intend to X or not believe that you ought to X.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

62  Instrumental Rationality There is a question of whether we should take rational requirements, like Means–Ends Coherence, to specify synchronic requirements—that is, requirements which tell us what is required at a particular time—or diachronic requirements—that is, requirements which tell us how we are required to proceed over time.5 The requirements formulated above could be interpreted either as synchronic or diachronic. For instance, the wide-scope requirement Means–Ends Coherence could be thought to prohibit one from holding a certain combination of attitudes at some time. But it could also be thought to specify how one should proceed over time: roughly, if you have the prohibited combination of attitudes, you should proceed by revising one of them (giving up your end, giving up your belief, or coming to intend the means), and if you don’t yet have the prohibited combination of attitudes, you should avoid proceeding so as to acquire it. More precise formulations of both the synchronic and diachronic in­ter­ pret­ations of Means–Ends Coherence appear below. But with just these rough formulations in mind, we can consider how the wide-scope synchronic and diachronic requirements are related. Here’s a fairly natural thought: the synchronic requirement explains the diachronic requirements. In other words, it’s because rationality prohibits us from having that combination of attitudes (intending an end, believing intending the means is necessary, and not intending the means) at a certain time that there exist diachronic requirements to escape from being in that prohibited state, if you are in it, and to avoid entering that prohibited state, if you are not yet in it. It’s natural to think that if some state is rationally prohibited, because of this, you should avoid, or escape from, being in that state. And so we can take Means–Ends Coherence to specify both synchronic and diachronic requirements, with the former explaining the latter.6 There’s room for the narrow-scoper to take issue with the wider-scoper when it comes to both synchronic and diachronic requirements. Let’s first consider the debate over synchronic requirements. Here are the relevant formulations: Wide Synchronic MEC:  Rationality requires that [if one intends, at t, to X, believes, at t, one will X only if one intends to Y, then one intends, at t, to Y]. 5  Some have suggested that this question matters a great deal for the scope debate. See, for instance, Schroeder (2004) and Kolodny (2005). 6  It’s worth emphasizing the limited ambitions of this view. I’m not here claiming that all diachronic requirements can be explained by synchronic requirements in this way. (Perhaps requirements governing processes of reasoning cannot be so explained.) Nor I am endorsing any sort of skepticism about diachronic requirements, like that defended by Hedden 2015, and critiqued by Podgorski 2016.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  63 Narrow Synchronic MEC:  If one intends, at t, to X, believes, at t, one will X only if one intends to Y, then rationality requires that one intends, at t, to Y. The dispute over which of these is correct will not concern their predictions for when an agent violates the requirement, since an agent violates Wide Synchronic MEC if and only if she violates Narrow Synchronic MEC. To see this, suppose I violate Wide Synchronic MEC. That means I have the following combination at t: I intend to X, believe I will X only if I intend to Y, and don’t intend to Y. But if I have that combination then I also violate Narrow Synchronic MEC, since I have the two attitudes in the antecedent and lack the attitude required by the consequent. Likewise, suppose I violate Narrow Synchronic MEC. That means that I have the two attitudes in the antecedent, and lack the attitude required by the consequent. But that combination also violates Wide Synchronic MEC.7 However, there is still a difference between the two formulations, although it’s a relatively insignificant one: it’s possible to comply with Wide Synchronic MEC without complying with Narrow Synchronic MEC.8 The requirement specified in the consequent of Narrow Synchronic MEC applies only to those agents who have the attitudes specified in the antecedent. For agents who have those attitudes, if they fail to intend to Y, they violate the requirement, and, if they intend to Y, they comply with the requirement. But if they lack the attitudes specified in the antecedent, the requirement fails to apply to them. However, the wide-scoper is committed to thinking that the requirement still applies in such a case. For the wide-scoper, there are prohibitions on com­bin­ ations of attitudes, and these prohibitions apply to all rational agents, including those who lack the relevant attitudes.9 7  Related to this, Broome provides a proof (2013, p. 148) showing that one will have the property of being (fully) rational in exactly the same worlds on both the wide-scope or narrow-scope formulations. 8  As we’ll see momentarily, there are more significant differences when we turn to diachronic requirements. 9  There is thus already some restriction on the domain of application for the wide-scoper, since the requirement applies to rational agents, but not to trees, skyscrapers, cats, etc. We could make this explicit in the formulation of requirements like Wide Synchronic MEC, perhaps by prefacing them with something like “For all rational agents, rationality requires…” (See Broome 2013, p. 135.) One reader has suggested that doing so makes it inappropriate to classify the requirement as a wide-scope requirement; instead, it should count as narrow-scope. And, likewise, the requirement I’ll introduce in the next paragraph, Revised Wide Synchronic MEC, should also count as a narrow-scope requirement. I hesitate to divide things up this way, since the requirement wouldn’t possess the traditional virtues attributed to narrow-scope requirements (asymmetry, guidance, etc.) and so it’s unlikely that narrow-scopers would want to defend such formulations. But the issue here seems to be merely ter­ mino­logic­al. If we decide to call such requirements narrow-scope requirements, then I’m not opposed to all narrow-scope formulations of means–ends coherence.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

64  Instrumental Rationality Is this a problem for the wide-scoper? Some might find it counterintuitive to think that there are very many requirements with which one is complying at this very moment. For instance, on the wide-scope view, rationality prohibits you from having this combination: intending to do cartwheels across the quad, believing that intending to train is necessary for doing cartwheels across the quad, and not intending to train. Even if something as outrageous as doing cartwheels across the quad has never occurred to you, and will never occur to you, there was all along an applicable requirement prohibiting you from having that combination—a requirement with which you complied. But some find this to be a bizarre thing to say.10 We aren’t forced to say this on a narrow-scope view. If we apply Narrow Synchronic MEC to this case, we generate the following requirement: if you intend to do cartwheels across the quad, and believe intending to train is necessary, then rationality requires that you intend to train. Since you never intended to do cartwheels across the quad, this requirement never applied to you. But there’s no obstacle to revising the wide-scope view so as to accommodate the point. If we want to deliver the result that the requirement applies only to those who already have the relevant attitudes, we could adopt: Revised Wide Synchronic MEC:  If one intends, at t, to X, believes, at t, one will X only if one intends to Y, then rationality requires that [if one intends, at t, to X, believes, at t, one will X only if one intends to Y, then one intends, at t, to Y].11 The requirement is still a wide-scope requirement, since “requires” has logical scope over a conditional. Yet “requires” occurs in the consequent of a conditional whose antecedent provides the desired jurisdictional restriction, limiting the applicability of the requirement to those who already have 10  In discussing a similar case applied to a wide-scope diachronic instrumental requirement, Errol Lord (2014b, p. 460) writes that it “seems a bit far fetched, to say the least.” Schroeder (2004, p. 347) raises a related worry about the proliferation of unexplained, eternal, agent-neutral requirements on the wide-scope view. 11  I propose this kind of strategy in Brunero 2015, pp. 243–4. Daniel Fogal (2016, p. 77) also arrives at a view along these lines. See his example of the requirement (Grandma-Wide*), and the related discussion of the distinction between conditions of application and jurisdiction. Likewise, Benjamin Kiesewetter (2018, p. 50) develops a view along these lines. (See his example of Enkratic Requirement, Conditional Wide-Scope.) As Kiesewetter also correctly observes, we could accommodate the “main point” (p. 138) of the wide-scope account without having “requires” have scope over a conditional. We could just as easily have the requirement relate directly to a disjunction of responses. For instance, in the case of the Enkratic Requirement, we would have it be a requirement to either not believe one ought to X or intend to X. (See Enkratic Requirement, Conditional + Disjunctive on p. 61.) Since I agree that such a view captures the main point of the wide-scope account, I have no objection to it.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  65 the relevant intention and belief at t.12 With such a revision, we’ve effectively eliminated any significant difference between the wide-scope and narrow-scope synchronic requirements. For now not only does an agent violate Revised Wide Synchronic MEC if and only if she violates Narrow Synchronic MEC, but Revised Wide Synchronic MEC applies if and only if Narrow Synchronic MEC applies (given that the antecedents of the two are the same), and an agent complies with Revised Wide Synchronic MEC if and only if she complies with Narrow Synchronic MEC (since the only way to comply with these requirements would be to have all three attitudes at the same time—the intention to X, the belief that one will X only if one intends to Y, and the intention to Y). There are thus no differences with respect to violation, applicability, and compliance. If there are important differences between the wide-scope and narrow-scope interpretations, they aren’t to be found here. Let’s now turn to diachronic requirements, which have received more attention in literature on the scope of rational requirements, since many have suspected that this is where the important differences between the wide-scope and narrow-scope interpretations are to be found. Some narrowscopers, including Niko Kolodny and Mark Schroeder, have argued that the wide-scope interpretations lose their plausibility once it’s made clear that we are dealing with diachronic, not synchronic, requirements.13 However, as I’ll argue in the next section, I think it’s the other way around: the narrowscope interpretations lose their plausibility. In any case, since this is where the important differences are to be found, we’ll now turn our attention to the formulations of diachronic requirements. These are the requirements which will be relevant to the arguments I’ll present and discuss in the remainder of this chapter.14

12  Mark Schroeder (2014a, esp. pp. 222–5; 2014b, esp. p. 233) has argued that the conditionality of narrow-scope requirements provides the narrow-scoper with “richer explanatory resources” to explain rational requirements (2014a, p. 225). I won’t explore this claim here. But it’s worth noting that Revised Wide Synchronic MEC would allow for the wide-scoper to share those alleged explanatory resources, since it also makes the requirement conditional on one’s having an end and an instrumental belief. 13  Schroeder writes, “Wide-scope principles are good at predicting what is wrong with an agent at a time. But they are not good at predicting the rational ways for an agent to change her situation” (2004, p. 346). Niko Kolodny (2005, 2007) distinguishes “state requirements,” which tell us how to be at a certain time, from “process requirements,” which tell us how we should revise our attitudes over time, and argues that once it’s made clear that the debate is about process requirements—which is where the interesting differences between the wide-scope and narrow-scope views are to be found— we should favor the narrow-scope interpretations (2007, p. 371). 14 I’ll avoid the tedium of providing the synchronic and diachronic interpretations of all the requirements mentioned at the start of this chapter. But it will be clear that the arguments in the following sections are concerned with the diachronic interpretations. Again, this is where the interesting differences are to be found.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

66  Instrumental Rationality As I mentioned above, I take the wide-scope view to involve a package of claims: there’s a synchronic prohibition on certain combinations of attitudes, and, because of this, there are diachronic requirements to avoid or escape being in those prohibited states. Here’s a formulation that would require agents to avoid or escape being in the prohibited state: Wide Diachronic MEC:  Rationality requires that you not proceed, from t to t1, in a way that would constitute your coming to have (or continuing to have), at t1, the combination [intending to X, believing you’ll X only if you intend to Y, not intending to Y].15 Let’s suppose someone already has, at t, the prohibited combination. For instance, suppose that, at t, Aaron intends to pass the test, believes that to do so he must intend to study, but doesn’t intend to study. Wide Diachronic MEC requires, going forward, that he not continue to have the combination, which he could do either by not intending to pass, or by intending to study, or by revising his instrumental belief. Likewise, even if Aaron doesn’t already have that prohibited combination of attitudes at t, Wide Diachronic MEC prohibits him from coming to having it at t1.16 We can contrast this formulation with the narrow- and medium-scoped formulations of diachronic means–ends coherence: Narrow Diachronic MEC:  If, at t, you intend to X and believe that you will X only if you intend to Y, then rationality requires that you proceed, from t to t1, by intending to Y. Medium Diachronic MEC:  If, at t, you believe you will X only if you intend to Y, then rationality requires that you proceed, from t to t1, by either not intending to X or intending to Y.

15  We don’t want to say that any way of proceeding that would predictably result in my having this combination would be irrational. I may know that my deciding to travel to New York will lead to my being irrational with respect to some tempting cheesecake, without it being the case that my so deciding is irrational. (Thanks to a referee for this nice point and example.) 16  Andrew Reisner (2009, p. 249) has proposed the following diachronic wide-scope requirement of means–ends coherence: “You are diachronically rationally required that [(you intend q at time t1 or do not at t1 both intend p and believe if p then q) if you intend p at time t and also believe if p then q at time t, but you do not intend q at time t].” However, as Reisner notes, since at time t you are unable to change the past—and assuming some plausible transmission principles (in particular, “necessary detachment,” according to which, where “R” is “rationality requires,” if R(p→q), and p is necessary, then R(q))—it follows that rationality requires that (you intend q at time t1 or do not at t1 both intend p and believe if p then q). And this derived principle has a structure similar to Wide Diachronic MEC. See Reisner 2009, p. 250.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  67 According to Narrow Diachronic MEC, Aaron is rationally required to proceed by intending to study. And according to Medium Diachronic MEC, Aaron is rationally required to proceed by either not intending to pass or intending to study. There is a clear difference here between the wide-scoped, medium-scoped, and narrow-scoped diachronic requirements. They issue different prescriptions for how Aaron is to proceed.17 There’s one final preliminary point about the nature of rational requirements we should note: rational requirements are local in the sense that they govern particular conflicts among one’s attitudes, regardless of the other conflicts one might have. Niko Kolodny explains locality as follows: Rational requirements, accordingly, ought to be local. In each instance in which one is under a rational requirement, what it ought to require of one is to avoid or resolve some specific conflict among one’s attitudes – as opposed to, say, to satisfy some global constraint on all of one’s attitudes. One might liken each application of a rational requirement to a referee with authority over a different part of a playing field, or to an inspector with authority over a different stage in a production process.18

As Kolodny notes, it’s possible for a single attitude to be governed by two local requirements, if that attitude is part of two separate conflicts. Suppose I believe P and believe ~P, and also believe I have insufficient evidence for ~P. My believing ~P is part of the conflict between my believing P and my believing ~P (which is governed by a consistency requirement on belief), and part of the conflict between my believing ~P and believing I have insufficient evidence for ~P (which is governed by the “believed insufficient evidence” requirement). In this case, the presence of a conflict suffices to put in place a rational requirement governing that conflict, even though one of the attitudes generating that conflict is itself part of another conflict governed by another requirement. Since requirements are “local,” we don’t need to look beyond the conflicting attitudes, such as to the agent’s psychology as a whole or even to other parts of the agent’s psychology, to see whether a requirement applies. It’s enough to note the presence of the conflict governed by the requirement.19 17  More precise formulations of any of these diachronic requirements would have to accommodate the fact that updating attitudes take time. We don’t want to have it come out that those who don’t update in the blink of an eye are irrational. This is what Abelard Podgorski (2017) refers to as “rational delay.” (On the other hand, as Podgorski notes, we don’t want to permit sluggishness either.) 18  Kolodny 2005, p. 516. 19  In the next section, I argue that these local requirements are wide-scope. It’s compatible with this view to hold that there are non-local rational requirements that are narrow-scope. For instance, one

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

68  Instrumental Rationality To sum up, rational requirements govern conflicts among our attitudes. They can be formulated as either synchronic or diachronic requirements. The wide-scope view is most plausibly understood to involve both synchronic and diachronic requirements, with the former explaining the latter: there are prohibited combinations of attitudes at a time, and, because of this, rationality requires one to avoid and escape one’s having a prohibited combination of attitudes at a time. And the requirements are “local” in that they govern particular conflicts among our attitudes, even when some of the attitudes involved in the relevant conflict are also involved in other conflicts.

3.2  The Case for Wide-Scoping In this section, I’ll develop a general argument for preferring the wide-scope formulations of rational requirements.20 Wide-scopers, in defending their view, often consider examples in which agents adopt attitudes they shouldn’t adopt, such as immoral or imprudent intentions.21 For instance, suppose that Bob intends to kill his colleagues and believes that he’ll do so only if he intends to poison them. Narrow Diachronic MEC would require that Bob then intend to poison them. But that seems counterintuitive. It seems as though it would be permissible for him to instead revise his intention to kill his colleagues. I think that there is something basically right about this common strategy of argument. However, it needs to be refined. As it stands, the narrow-scoper has an easy reply to make. She could insist that Bob is indeed rationally required to intend to poison his colleagues, but morally required not to. And, she might continue, we think of Bob’s case as an objection to the narrowscope view only because we confuse two distinct sources of requirements: morality and rationality. Moreover, as we noted earlier, there is a debate about might hold that, taking all of one’s attitudes into account, rationally requires you to proceed by revising or forming some particular attitude. See Brunero 2010a. (And this seems to be enough to run an argument along the lines of Kolodny 2005, §1, against the normativity of rationality.) However, we’re here limiting ourselves to consideration of local requirements, like means–ends coherence. 20  More precisely, the argument will show that the rational requirements often discussed within the literature on rationality—such as the enkratic requirement, the consistency and coherence constraints governing intentions, as well as those governing beliefs—should be given wide-scope formulations. The argument does not aim to show that all rational requirements are wide-scope requirements. I don’t think this is so. Indeed, it might be that narrow-scope requirements can be derived from the wide-scope ones. (See Titelbaum 2015 and Brunero 2013, pp. 563–5.) 21 See, for instance, Hill  1973, p. 435–6, Broome 2000, pp. 83-4; Dancy  2000, pp. 59–60, 72; Wallace 2001, p. 17.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  69 whether rational requirements are normative in the sense that we ought to, or have a reason to, comply with them. If we hold that rational requirements are not normative in this sense, we could maintain that Bob is rationally required to intend to poison his colleagues, but it’s not the case that he ought to intend to do this, nor the case that he has any reason to intend to do this.22 However, there’s a way to develop the argument against narrow-scope interpretations that makes it clear that the argument doesn’t trade on our confusing one source of requirements with another. Even when we hold firmly in mind that rational requirements are distinct from moral ones (and perhaps also not normative), the narrow-scope view makes intolerable predictions.23 In Section  3.2.1, I will introduce this way of developing the argument, focusing in particular on the narrow-scope interpretations of the enkratic requirement and means–ends coherence, though versions of the same argument could be used to target other narrow-scope formulations as well.24 In Section 3.2.2, I will consider two important responses to this argument that have been advanced by narrow-scopers, and I’ll argue that neither one succeeds.

3.2.1  The Argument Let’s start with an argument against the narrow-scope interpretation of the enkratic requirement. It involves a case in which a single attitude is part of two conflicts. Suppose Annie believes that she ought to attend a lecture, but she doesn’t intend to attend it. She’s akratic. And suppose that Annie also believes that she lacks sufficient evidence for her belief that she ought to attend the lecture, but she believes this anyway. She’s epistemically akratic. Given the combinations of attitudes Annie has, it’s clear that both the enkratic 22 Kolodny  2005 is an example of a view that holds onto the narrow-scope interpretations of rational requirements while denying the normativity of those requirements. 23  The argument will thus differ from what Errol Lord has correctly described as “the argument that often motivates the wide scope view,” which goes as follows: (1)  Either the wide-scope view is true or the narrow-scope view is true. (2)  If the narrow-scope view is true, then rationality is not deontically significant. (3)  Rationality is deontically significant. (4)  Thus, the narrow-scope view is false. (C)  Therefore, the wide-scope view is true. (2017, p. 1116) In saying that some requirement is “deontically significant,” Lord means that the requirement is ­significant in that one ought, or has a reason, to comply with it—“normative” in our terminology. It will be clear that my argument below in no way depends upon the controversial premise (3) above. 24 I present this argument in greater detail in Brunero 2010a, pp. 34–8, and Brunero  2012, pp. 132–3, 135.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

70  Instrumental Rationality requirement and the “believed insufficient evidence” requirement are relevant. (The combination of believing she ought to attend and not intending to do so makes the former relevant, while the combination of believing she ought to attend and believing there’s insufficient evidence for this belief makes the latter relevant.) Now, if the narrow-scope readings of these requirements, Narrow Enkratic Requirement and Narrow Believed Insufficient Evidence, are correct, then Annie is required to do two things: drop her belief that she ought to attend the lecture, and intend to attend the lecture. (Of course, after she drops the belief, she is no longer required by the Narrow Enkratic Requirement to intend to attend. But before she drops it, she is rationality required, going forward, to intend to attend. Failure to do so constitutes a violation of a rational requirement, and hence is irrational.) But that’s a counterintuitive result. It’s not irrational for her to just drop her belief while not also intending to attend the lecture. For the purposes of this argument, the formulation of the “believed insufficient evidence” requirement, whether wide-scoped or narrow-scoped, doesn’t matter. On either formulation, we still get the counter-intuitive result that Annie is rationally required to intend to attend the lecture. That’s a counter-intuitive result because it seems as though she is rationally permitted to revise her belief without also intending to attend. If we use the wide-scope version of the enkratic requirement, we’ll be able to avoid this counterintuitive result. According to the Wide Enkratic Requirement, Annie is required to either drop her belief that she ought to attend or intend to attend. So, it need not be irrational for her to just drop her belief that she ought to attend the lecture and not also intend to attend it. If we use the Wide Enkratic Requirement, she violates no rational requirement in proceeding this way. A similar argument can be deployed against the medium-scoped and narrow-scoped versions of means–ends coherence. Suppose that Aaron intends to pass the test, believes that he must intend to study to pass, but doesn’t intend to study. According to Medium Means–Ends Coherence, since he believes he must intend to study to pass, rationality requires him to either drop his intention to pass or form an intention to study. And according to Narrow Means–Ends Coherence, since he intends to pass and believes he must intend to study to pass, rationality requires him to form an intention to study. But suppose that Aaron acquires good evidence that studying isn’t ne­ces­sary for passing—perhaps he learns that the English Department faculty are so obsessed about student retention that they’ve instituted a policy according to which no student can fail—and so Aaron concludes that he doesn’t have sufficient evidence for his instrumental belief. It seems that we should

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  71 say that rationality doesn’t require him to either drop his intention to pass or form the intention to study. Nor does it require him to form the intention to study. It would be permissible for him to drop his belief that studying is ne­ces­ sary for passing, and not make any other revisions in his attitudes. So, the medium-scoped and narrow-scope interpretations cannot be correct. If we instead adopt the wide-scoped Means Ends Coherence, we aren’t committed to thinking Aaron would violate any requirement of rationality by dropping his belief that studying is necessary for passing without also making any other revisions in his attitudes. In conclusion, the narrow-scoped and medium-scoped interpretations of these requirements of rationality yield false predictions about what Annie and Aaron are rationally required to do. And so they should be rejected. It’s clear that this argument doesn’t involve any confusion between rational requirements and moral requirements, since moral requirements never enter the picture in these examples. It’s also clear that the wide-scoped interpretations of these requirements avoid issuing these false predictions. Hence, we should prefer the wide-scoped formulations to their narrow-scoped and mediumscoped alternatives.

3.2.2  Challenges to the Argument One way to deflect the force of the argument would be to develop a version of the narrow-scope view that avoids the counterintuitive result that Annie violates a rational requirement when she proceeds by dropping her belief instead of by coming to intend to attend the lecture. There are two proposals for how to do this, one advanced by Alida Liberman and Mark Schroeder (2016), and another advanced by Errol Lord (2014b). I’ll discuss each in turn, noting how the narrow-scoper would incur additional costs in adopting them. Liberman and Schroeder advance a theory of how to weigh up rational requirements, which they label “Sophisticated Weighing.” The details of the proposal, which is framed in terms of the concept of a commitment, need not concern us.25 We need only note that they hold that any version of Sophisticated Weighing will have the following two features: First, it must give answers about each of the things that an agent ought to do, believe or intend, in a way that is determined, inter alia, by some among the 25  See also Shpall 2014 on commitments.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

72  Instrumental Rationality agent’s commitments. But second, commitments that the view predicts the agent should give up must not be among the ones that do this determining. So, for example, if a commitment is grounded in an irrational belief, then any version of Sophisticated Weighing should ignore that commitment, because that is a belief that the agent should give up.26

The case of Annie would be covered by this second feature. Since Annie’s belief that she ought to attend the lecture is irrational—one which she should give up given that she thinks there’s insufficient evidence for it—we should ignore the commitment it grounds to intend to attend the lecture. The result is that Annie violates no rational requirement when she fails to proceed by intending to attend the lecture. However, the worry is that if we develop a view that allows us to say that the Narrow Enkratic Requirement can be ignored in specifying what Annie rationally ought to do because of the other attitudes Annie has, we lose out on the advantages of thinking of rational requirements as local requirements. As I mentioned in Section 3.1, I am here following Kolodny and others in taking rational requirements to be local in the sense that they govern particular conflicts of attitudes and are in place regardless of the attitudes one has besides those involved in the conflict. But it’s clear that Liberman and Schroeder are giving up on the locality of rational requirements. On their view, one’s other attitudes—that is, attitudes besides the ones involved in the conflict governed by the requirement—are relevant to whether the requirement is in place, or, at least, to whether the requirement deserves consideration in determining what one rationally ought to do. For instance, on their view, the enkratic requirement, which governs the conflict between Annie’s believing she ought to attend and her not intending to attend, can be ignored in determining what she rationally ought to do, since she believes there’s insufficient evidence for this belief. There are costs to giving up on the locality of requirements. Suppose that instead of dropping her normative belief—the belief which she thinks is insufficiently supported by the evidence—Annie holds onto it, but forms an intention to attend the lecture and so is no longer practically akratic. She has resolved some of conflicts among her attitudes, but not others. So, we should say that she has done well on some fronts but not on others. The wide-scoper can say this. The wide-scoper can maintain that Annie has come to comply with one rational requirement (Wide Enkratic) though she violates another 26  Liberman and Schroeder 2016, pp. 112–13.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  73 (Wide Believed Insufficient Evidence). Additionally, the wider-scoper can say that had Annie instead remained both epistemically and practically akratic, she would have violated two rational requirements, thereby coming in for criticism on two fronts. But the Liberman–Schroeder narrow-scoper has trouble here. In the case in which Annie holds onto her normative belief but is no longer practically akratic, there is no longer a requirement in view with which she has come to comply. And in the case in which she is both epi­stem­ ic­al­ly and practically akratic, she comes in for criticism on just one front, since there’s no longer a requirement in view to ground criticism of her practical akrasia. Indeed, on the Liberman–Schroeder view, the two cases— the case in which she is epistemically but not practically akratic and the case in which she is both epistemically and practically akratic—are on a par as far as rational requirements go. In both cases, Annie violates Narrow Believed Insufficient Evidence and there are no other requirements we need to consider. But this strikes me as a significant cost. It’s a cost to say that practically strong-willed Annie and practically akratic Annie are on a par as far as rational requirements go. Additionally, the kind of rational error that Annie makes when she believes she ought to attend the lecture but doesn’t intend to do so may be the same kind of error as in ordinary cases of akrasia. Consider an ordinary case. Suppose Bonnie believes she ought to attend the lecture, but unlike Annie doesn’t have any beliefs further upstream about the sufficiency of the evidence for her normative belief. Presumably, Liberman and Schroeder would say that Bonnie violates a rational requirement by being akratic. However, when Annie fails to intend to attend the lecture, she is most likely making the very same kind of mistake as Bonnie.27 And it would be nice to have the resources to be able to deliver the verdict that she is irrationally akratic here as well. However, the problem is that Sophisticated Weighing tells us that the requirement that would be the basis for delivering that verdict can be ignored. Errol Lord (2014b) provides another way to avoid the counterintuitive result that Annie violates a rational requirement when she proceeds by dropping her belief instead of by coming to intend to attend the lecture. The basic 27  One might wonder about cases in which Annie’s motivational failure is explained in some way by her upstream belief that there’s insufficient evidence for her belief that she ought to attend the lecture. In such cases, Annie’s being akratic, while yielding a local incoherence with her normative belief, might be thought to be rational given that it involves a sensitivity to other relevant parts of her psych­ology. Some philosophers have used cases with this structure argue for the possibility of “rational akrasia.” (See Audi 1990, especially p. 280.) I have argued that such a conclusion does not follow. (See Brunero 2013, pp. 548–53.) In any case, I’m here assuming that this is an ordinary case, where Annie’s akrasia isn’t explained by the influence of the beliefs further upstream.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

74  Instrumental Rationality idea is to appeal to the concept of a cancelling condition. Suppose I promise my wife that I’ll pick up my son after school, but my wife releases me from the promise. In this case, it’s not that I’ve complied with the obligation generated by my promise. Rather, the obligation is cancelled by my being released. As Lord puts it, “[g]etting released cancels the obligation because it makes it the case that the relevant requirement doesn’t apply to me anymore.”28 In the same way, we can think that Annie’s dropping her normative belief is a cancelling condition: it makes it the case that the narrow-scope requirement doesn’t apply to her anymore.29 If that’s correct, then we can say that Annie proceeds in a perfectly rational way when she gives up her normative belief. In doing so, she complies with Narrow Believed Insufficient Evidence, and she makes inapplicable the enkratic requirement.30 One worry about Lord’s proposal is that it involves giving up on a claim that was previously thought to be common ground between wide-scopers and narrow-scopers. As we set up the debate at the start of the chapter, we noted that it was common ground between wide-scopers and narrow-scopers that if you have a certain combination of attitudes, rationality requires something of you. For instance, if you believe you ought to do something but don’t intend to do it, rationality requires something of you. And the debate between wide-scopers and narrow-scopers is about what rationality requires of you— whether, in this case, it requires you to intend to do it or, instead, to either intend to do it or give up your belief that you ought to do it. But Lord’s view gives up on that common ground. He doesn’t think that if you have this com­ bin­ation of attitudes, rationality requires something of you. He does think that if you have this combination, a certain conditional is true of you, namely, that if you don’t cancel the requirement, then rationality requires something of you.31 But that’s different from saying that if you have this combination of attitudes, rationality requires something of you. 28  Lord 2014b, p. 454. 29  One difference in the promising case is that while it’s up to Annie to revise her belief, it’s not up to me to be released from the promise. I can ask to be released, but whether I am released depends on what my wife says. 30  Lord’s Ought ND+ holds that “If, at t1, you believe you yourself ought to ϕ and no canceling condition obtains in the inclusive interval between t1 and t2, then, at t2, you are rationally required to intend to ϕ.” (2014, p. 455). The requirement in the consequent is made inapplicable by Annie’s giving up her normative belief, which is a cancelling condition. 31  This strategy seems to involve forfeiting one of the alleged advantages of narrow-scope rational requirements. Some have argued that a theory of rationality, if it’s to be useful, should provide guidance to an agent, and that disjunctive requirements by themselves are unable to do this, since they leave open the question of which of the two (or more) options to pursue. See Reisner 2009, esp. pp. 245 and 249–50 for further discussion of this idea. Traditional narrow-scope views provide the required non-disjunctive guidance. Yet Lord’s narrow-scope requirement does not provide non-disjunctive guidance. For instance, Annie is here told that, going forward, either she escapes the requirement by

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  75 There’s a second, more serious, worry about the proposal. On Lord’s view, when Annie drops her normative belief, she cancels the rational requirement, and so the enkratic requirement no longer applies. But, intuitively, we want it to come out that there is an enkratic requirement that still applies to Annie even when she doesn’t have the normative belief—specifically, we want it to come out that she is prohibited from coming to believe that she ought to X without also coming to intend to X. The wide-scope view has no trouble here, since, as we saw, there are diachronic requirements to avoid, as well as escape, prohibited states. But, on the narrow-scope view, the enkratic requirement applies only after Annie believes she ought to X, and so it cannot ground a prohibition on her coming to believe she ought to X without also intending to X. Of course, it’s true that after she comes to believe she ought to X, she is required to intend to X provided that she doesn’t cancel this requirement. But that’s not the requirement we’re looking for. What we want is the rational equivalent of a “No Trespassing” sign, not one that reads, “If you trespass, and don’t remove yourself, you’ll be required to leave.”32 Let’s sum up the main conclusions of the section. Cases like those of Annie and Aaron provide a strong argument in favor of the wide-scope in­ter­ pret­ations of rational requirements—an argument whose force cannot be deflected by distinguishing rational requirements from moral ones. However, there are resources available to the narrow-scoper to respond to this argument. I’ve here considered two proposals, one involving a view about how to weigh up rational requirements, and another involving the introduction of cancelling conditions on requirements. However, there are further theoretical costs involved in adopting either of these proposals. In light of these costs, we would do better to adopt the wide-scope formulations of rational dropping her normative belief or she will be required to intend to attend the lecture. Kiesewetter (2018, p. 137) also presses a version of this point against Lord’s view, as well as providing a general discussion of the guidance objection to the wide-scope view (pp. 148–56). Since discussion of whether and how rational requirements should provide guidance would take us too far afield, I’ll leave this issue aside. 32  The wide-scoper need not be committed to claiming that the two attitudes—the belief that one ought to X and the intention to X—must be formed at precisely the same time. The formation of attitudes take some time, and presumably there’s nothing rationally impermissible about forming an intention to X a millisecond after forming your belief that you ought to X. As I mentioned earlier, more precise formulations of both wide- and narrow-scope requirements will want to allow for a permissible “rational delay” (Podgorski 2017). However, in a normal case of reasoning well to a revision in attitudes, when one comes to understand the considerations that conclusively favoring X-ing, one comes both to intend to X and to believe that one ought to X at roughly the same time. And we want our rational requirements to capture the idea that something has gone wrong, so far as one’s rationality goes, if one makes the theoretical revisions without making the practical ones—again, allowing for a permissible range of rational delay. And the wide-scope formulation does well on this front.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

76  Instrumental Rationality requirements, including our wide-scope formulation of instrumental rationality, Means–Ends Coherence.

3.3  The Case for Narrow-Scoping Many of the objections to the wide-scope interpretations of rational requirements concern the allegedly problematic “symmetries” predicted by those interpretations.33 For instance, Means–Ends Coherence predicts that, so far as this requirement goes, there are three equally good ways of complying with it: intending the means, abandoning the end, and abandoning the instrumental belief. But, intuitively, intending the means is a more rational way of complying than dropping the end, and both of these are more rational ways of complying than giving up one’s instrumental belief.34 These intuitive asymmetries aren’t explained by the wide-scope interpretation of the requirement. One might also note other asymmetries related to the direction of reasoning: while it’s perfectly fine for Aaron to reason from his intention to pass and his belief that intending to study is necessary for passing to intending to study, it’s not acceptable for him to reason from his not intending to study and intending to pass to not believing that intending to study is necessary for passing. That would be unacceptable wishful thinking.35 The wide-scoper can issue a two-part reply to such challenges. First, the wide-scoper will remind us that Means–Ends Coherence predicts that so far as this requirement goes these three ways of proceeding are equally rational ways of complying with this requirement. It doesn’t follow from this that they are equally rational ways of proceeding, taking everything into account. For instance, suppose Aaron thinks there’s conclusive evidence that he’ll pass only if he intends to study. In this case, his dropping his instrumental belief (without making any other changes in attitude) would violate a requirement of theoretical rationality.36 So, even though, so far as the requirement of 33 See Schroeder  2004,  2009,  2014a, Kolodny  2005, 2007, Bedke  2009, Lord 2014b, and Kiesewetter  2018. For some replies in defense of the wide-scope interpretations, see Way  2011, Brunero 2012, Shpall 2013, and Broome 2013. 34  Schroeder 2004, p. 346, and Schroeder 2009, p. 227. 35  Kolodny 2005 develops asymmetry objections involving the direction of reasoning against the wide-scope interpretations of several rational requirements. See especially pp. 527–30. For a reply, see Way 2011, pp. 235–8. 36  The parenthetical is necessary if we take the relevant requirements of theoretical rationality to be wide-scope. If Aaron dropped his instrumental belief while also dropping his belief about the evidence being conclusive, then he wouldn’t violate a wide-scope requirement of theoretical rationality (but would violate a narrow-scope requirement). For the purposes of this argument, it doesn’t matter whether we take those requirements of theoretical rationality to be wide- or narrow-scope. Our aim here

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  77 Means–Ends Coherence goes, dropping his instrumental belief is as rational a  way of proceeding as intending the means, it doesn’t follow that, taking everything into account, especially norms of theoretical rationality, it’s as rational a way of proceeding as intending the means. (An analogy with Ross’s ethical theory might help: it may be that two ways of proceeding are on a par so far as the prima facie duty of beneficence goes without their being on a par, all things considered.37) Second, the wide-scoper can attempt to explain why one way of proceeding is more rational than another, taking everything into account.38 If we’re looking at Aaron’s particular case, this is fairly straightforward: since he thinks there’s conclusive evidence that he’ll pass only if he intends to study, his dropping his belief would be prohibited by a requirement of theoretical rationality, and so would be a less rational way of proceeding, all things considered, than intending to study or not intending to pass, both of which (we’ll assume) don’t involve Aaron in any further irrationality. We’re also in a position to explain our asymmetry intuitions—that is, why it seems that giving up an instrumental belief is a less rational way of complying than giving up an end or intending the means. We can do this by noting that Aaron’s epistemic situation is fairly typical: by and large, people hold instrumental beliefs they take to be strongly supported by the evidence. And so, by and large, proceeding by revising one’s instrumental belief would be a less rational way of proceeding than proceeding by intending the means or dropping the end. This explains why we have the asymmetry intuitions we have—specifically, why we think revising a belief would be less rational than, say, intending the means. The wide-scoper can also appeal to requirements of rationality besides Means–Ends Coherence to explain the asymmetries in reasoning mentioned above. For instance, some wide-scopers have endorsed a basing prohibition according to which Aaron is prohibited from revising his instrumental belief

is simply to defend the wide-scope interpretation of means–ends coherence, and either in­ter­pret­ation will do. However, I’ll assume that the those other theoretical requirements are narrow-scope, since doing so will allow me to more easily raise another skeptical worry about Means–Ends Coherence that I’ll address in Section 3.3.1. 37  Ross 1930. 38  There are several resources for a wide-scoper to draw upon in providing such explanations. The wide-scoper could appeal to other requirements of rationality (Brunero  2012), basing principles (Way  2011), basing prohibitions (Broome  2013), rational commitments (Shpall  2013), attitudinal pressures (Fogal  2016, pp. 158–62), among other things. I’m here illustrating a general strategy avail­able to the wide-scoper, the details of which could be developed in various ways. If the particular explanation given here is incorrect or incomplete, it won’t spell doom for the general strategy.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

78  Instrumental Rationality on the basis of his intending to pass and not intending to study.39 This basing prohibition will explain why it’s not acceptable for Aaron to reason from his not intending to study and intending to pass to the belief that intending to study isn’t really necessary for passing after all. And we can also specify the permissible bases on which to make revisions in attitudes, perhaps allowing Aaron to form an intention to study on the basis of his intention to pass and his belief that studying is necessary. Such prohibitions and permissions can be brought in to explain why certain ways of proceeding are rationally prohibited, while other ways of proceeding are rationally permissible.40 The im­port­ ant point is that we recognize that it’s not the job of instrumental rationality to condemn every such way of proceeding. Other requirements can be brought in to do that. In the remainder of this section, I will note in Section  3.3.1 how we can explain the asymmetry between dropping an end and intending the means along similar lines. However, this response may not suffice to counter every version of the asymmetry objection. I will consider in Section 3.3.2 one such version, which concerns the difference between escaping from and complying with a requirement, and I’ll offer a reply to it. I will conclude that wide-scopers can thus offer a good reply to what has been the most significant component in the case for narrow-scoping.41 39  See Broome 2013, pp. 140–1, and Way 2011, pp. 230–1. As Broome observes, the narrow-scoper will also need to appeal to basing prohibitions, since it is possible for agents to comply with narrowscope requirements on inappropriate bases. So, we shouldn’t think this puts the wide-scoper at any disadvantage. 40  Kiesewetter (2018) has developed a novel version of the asymmetry objection that won’t be met by appealing to basing permissions and prohibitions. He appeals to the Advanced Rational Process Constraint: “For every rational requirement R, R requires A to be in state S, only if [either A satisfies R or for every satisfaction state of R, S*, there is a rational process available to A which leads to S*” (2018, p. 145). A satisfaction-state of R is, roughly, a state such that the agent’s being in that state satisfies R. For instance, for the Wide Enkratic Requirement, not believing one ought to X and intending to X are both satisfaction states. Kiesewetter is aware that sometimes an agent would have a rational process available to her that would lead her not to believe she ought to X. (The case of Annie, who believes she lacks sufficient evidence for her belief that she ought to attend the lecture, might be spelled out to provide such a case.) But he goes on to observe that “the rational requirement claims we are considering are supposed to hold in all possible worlds; consequently the strong rational process constraint demands that A can give up his belief [that he ought to X] through a rational process in all worlds in which the wide-scope requirement applies to him and he does not satisfy it” (2018, p. 146). But now Kiesewetter’s strong constraint would be too strong, since it would rule out many narrow-scope requirements as well. Surely there are possible worlds in which an akratic agent would have no rational process available to her that would lead to an intention to X. Perhaps some irresistible impulse, or external interference, prevents her from forming an intention to X. But that’s no objection to the Narrow Enkratic Requirement. (The point here is that a rational process to a satisfaction state can fail to be available to you not only from your not having access to the relevant starting points, but also from your not having access to the relevant ways of proceeding from those starting points to a change in attitude.) 41  I won’t pretend to have addressed every argument made on behalf of narrow-scope in­ter­pret­ ations here. That would require too much space. But I’m focusing on those objections that have received the most attention in the literature and which strike me as the most pressing.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  79

3.3.1  The Means–Ends Asymmetry In outlining the two-part response above, I showed how the wide-scoper can account for the intuition that revising one’s instrumental belief is less rational than either intending the means or dropping the end. I there appealed to the fact that we most often take our beliefs to be strongly supported by the evidence. And that’s why revising an instrumental belief strikes us as the least rational way to proceed. We can employ a similar strategy to deal with the asymmetry between intending the means and dropping the end. More often than not, our ends are ones we think we ought not give up. And that’s why abandoning the end strikes us as a less rational way to proceed than intending the means. Of course, there are many exceptions to the rule: there are many cases where we think it’s permissible, or even required, for us to give up an end. But the usual case is one where we think we ought not give up our ends.42 (It’s not as common as cases in which we think we ought not give up our instrumental belief, which explains the asymmetry between giving up an instrumental belief and giving up the end.) Although dropping an end and intending the means are on a par as far as instrumental rationality goes, they are no longer on a par once we take into account these other beliefs about the reasons for our attitudes. And the usual presence of such beliefs explains our asymmetry intuitions—that is, it explains why dropping the end seems to us a less rational way of proceeding than intending the means.43 (A thought experiment: Imagine our ends were selected for us at random. Imagine that at the start of each day we wake up finding ourselves programmed with all sorts of bizarre intentions. We’d probably have the opposite thought: dropping an end is the more rational way of proceeding. This suggests that the fact that we typically

42  I’m not saying that the usual case is that we ought not to give up the end, just that the usual case is that we think that we ought not give up the end. Perhaps we often mistakenly think ourselves required to hold onto merely permissible ends. 43  Even if you don’t accept my claim that more often than not we believe we ought not give up our ends, we could still run an explanation of the means–ends asymmetry intuition along these lines, so long as you grant me that this belief is more common than the belief that one ought to give up one’s end. Note that there are three relevant kinds of belief one could have with regard to one’s ends: one could believe keeping one’s end is required, keeping one’s end is permissible but not required, and keeping one’s end is forbidden. In the second case, it’s rationally acceptable to give up the end—unless, as we noted in Chapter 2, doing so is part of a pathological pattern of indecisiveness (or involves some other form of irrationality)—and also rationally acceptable to intend the means instead. There’s no asymmetry here. The crucial point to note now is that beliefs of the third kind—beliefs that your end is forbidden—are far less common than beliefs of the first kind—that is, beliefs that your end is required. And that’s what, I would argue, explains the asymmetry intuition—that is, the intuition that giving up an end is “less rational” than coming to intend the means.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

80  Instrumental Rationality adopt ends that we think we ought not give up plays a role in explaining our asymmetry intuitions.) This explanation of our asymmetry intuitions holds that giving up an end and intending the means, while on a par as far as instrumental rationality goes, are no longer on a par once we also consider an agent’s other attitudes, particularly, an agent’s beliefs about whether she ought to have those ends. But this explanation gives rise to a worry: once we appeal to these other attitudes to give an account of how some particular agent is rationally required to escape a state of means–ends incoherence, it no longer seems necessary to have a requirement, like Means–Ends Coherence, which specifies merely that she is rationally required to escape such a state. For instance, suppose we say that when Aaron believes there’s very good evidence that he must intend to study if he’s to pass, and believes that he ought not give up his intention to pass, he’s rationally required to escape his state of means–ends incoherence by  coming to intend to study. Adding that there’s also a requirement, Means–Ends Coherence, which states that he’s rationally required to escape his means–ends incoherence seems redundant. And the same would go for cases where one believes one ought not intend the means to one’s ends. Suppose I intend to get to work on time, and come to realize, in light of good evidence, that I’ll get there on time only if I (intend to) drive at very unsafe speeds. I currently don’t intend to drive at very unsafe speeds, and I think I ought not come to intend to do so. Here, we might say, rationality requires me to escape my means–ends incoherence by not intending to get to work on time. Again, adding that there’s also a requirement, Means–Ends Coherence, which states merely that I’m rationally required to escape my state of means–ends incoherence seems redundant.44 (And we could make the same point with regard to a case where one thinks one should give up one’s instrumental belief.) This raises a question: Is Means–Ends Coherence unnecessary? 44  This line of thought is related to Niko Kolodny’s (2005) suggestion for how his narrow-scope “core requirements” of rationality—which are: C+: If one believes that one has conclusive reason to have A, then one is rationally required to have A; and, C-: If one believes that one lacks sufficient reason to have A, then one is rationally required not to have A (p. 524)—can explain the consistency requirement on belief. In the view suggested in Kolodny  2005, cases of inconsistent beliefs—say, believing P and believing ~P—aren’t governed by a wide-scope requirement, but are instead governed by a disjunction of narrow-scope requirements. Roughly, if you think the evidence for P is stronger, you should drop ~P, and if you think the evidence for ~P is stronger, you should drop P, and if you think the evidence is neither stronger nor weaker, you should drop both. So, we can say, of the person who has inconsistent beliefs, that either he is required to drop his belief that P, or he is required to drop his belief that ~P, or he is required to drop both. See Kolodny 2005, pp. 540–2. (This view differs from the “disjunctivist” view in Kolodny’s later papers, which concerns a disjunction of narrow-scope requirements of reason, not rationality. We’ll consider this view in the next chapter.) I think that versions of the two objections that follow in the text would also apply to Kolodny’s suggestion, though I won’t pursue that here.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  81 There are at least two reasons to think it is not. First, means–ends incoherent agents might not have the relevant beliefs about how well the reasons support their attitudes. But we still want some requirement in place stating that they must escape states of means–ends incoherence. Perhaps, if we assume that these agents are in a position to know about their means–ends incoherence, we could say that they should think about which way they should revise, and so ought to form the relevant beliefs.45 But, we still want some requirement in place that holds that, even if they fail to form those beliefs they ought to form, rationality requires them to escape from their states of means–ends incoherence. (One doesn’t get off the hook merely by failing to think about how well one’s attitudes are supported by reasons.) And Means–Ends Coherence does this. Related to this, there’s a worry that even if they form the relevant beliefs about the reasons for their attitudes, those beliefs might not determine exactly how they should escape states of means–ends incoherence. It’s true that such beliefs can sometimes determine exactly how one should proceed, as in the examples above of Aaron’s studying and my driving recklessly to work, but it doesn’t seem as though they always do. Suppose I think that my intending some end, and my intending some means, are both permissible but not required. Suppose, for instance, I believe that my updating the departmental webpage is supererogatory (and hence permissible but not required).46 But I’m currently incoherent with respect to this end: I believe I’ll update the webpage only if I put in an IT request by Friday, and I don’t currently intend, nor trust that I will intend, to do so. I also think that my putting in the IT request is supererogatory. In this case, my beliefs about the reasons for my attitudes don’t call for any specific revision. In dropping the end, I wouldn’t be doing anything I think I ought not do, nor anything I think I ought to do. Likewise, in coming to intend the means, I would not be doing anything I think I ought not to do, nor anything I think I ought to do. There is no specific revision, by my lights, that is required of me. Yet we still want to say that rationality requires me to escape my state of means–ends incoherence.47

45  Kolodny (2008a, p. 388) makes a suggestion along these lines. For a critique, see Way 2018, §4.4. 46  Way (2013; 2018) points to supererogatory actions as cases of “mere permissibility” that are troublesome for “disjunctivist” views. (More on this in Chapter 4.) 47  Perhaps one also has the belief that one ought to either not intend to update the webpage or intend to put in the IT request. (Here the belief isn’t about the reasons for or against having (or not having) a specific attitude, but about the reasons not to have a certain combination of attitudes: intending to update the website and not intending to put in the IT request.) But, again, we don’t want the irrationality of my having the combination to depend upon my believing I ought not have it. Means–Ends Coherence condemns my incoherence without requiring I have any such belief.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

82  Instrumental Rationality Second, and more importantly, we want to allow for the possibility that one could exhibit rationality or irrationality in the execution of one’s plans even when those plans are contrary to one’s assessments of what one should do. (This is the important phenomenon that R. Jay Wallace, following Aristotle, refers to as “cleverness” in the pursuit of ends one does not think worthwhile.48) Suppose that both Beth and Carrie believe they shouldn’t attend the lecture across campus, yet, against their better judgment, intend to do so. But only Carrie intends to do what’s believed to be necessary to get to the lecture, while Beth gets distracted along the way, failing to form the intentions she knows she must form to get there. Carrie exhibits a kind of rationality in the pursuit of her end, while Beth fails in this regard. But if all we had were principles that specify how Beth and Carrie are rationally required to proceed given their beliefs about what they ought to do—namely, that they should proceed by abandoning their ends—we’d be unable to find any rational difference between Beth and Carrie. But Means–Ends Coherence allows us to find such a difference: Carrie satisfies this requirement while Beth violates it. Carrie, but not Beth, is instrumentally rational with regard to her end. Let’s sum up the main ideas of this subsection. Our two-part strategy can explain the asymmetry between giving up an end and intending the means: So far as instrumental rationality goes, these two ways of proceeding are on a par. But our intuition that they are not on a par is explained by the fact that it’s more often than not on balance rational for one to proceed by intending the means. And this in turn is explained by looking to our normative beliefs about our ends, provided that we adopt the independently plausible assumption that we by and large have ends we think we ought not give up. Additionally, we’ve seen that allowing for these normative beliefs about attitudes to be relevant doesn’t give us grounds for thinking that Means–Ends Coherence is redundant or unnecessary.

3.3.2  Escape and Compliance However, there are versions of the asymmetry objection that remain, even after employing our two-part strategy. One might present the asymmetry objection in the following way: the wide-scope requirement, Means–Ends Coherence, issues the problematic prediction that, so far as instrumental rationality goes, abandoning the instrumental belief is a way of complying 48  Wallace 2001, pp. 1, 14–20.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  83 with a requirement of means–ends coherence. But that seems to be a way of escaping from a requirement of means–ends coherence, not a way of complying with it.49 And so we should reject the wide-scope formulation of the requirement. (Since this escape/comply asymmetry objection notes an al­leged­ly problematic prediction for the wide-scope interpretation “so far as instrumental rationality goes,” appealing to what rationality requires on balance won’t be helpful in countering this objection.) The medium-scoped and narrow-scoped formulations of means–ends coherence do better, since they both predict that abandoning an instrumental belief is a way of escaping from, not complying with, the requirement. So, we have a reason to prefer these formulations to the wide-scope formulation. Errol Lord presents the objection as follows: After all, it seems like, as far as instrumental rationality is concerned, you are doing something quite different when you give up your means–ends belief than when you take the means. When you give up your belief it seems like you are getting out of an obligation to take the means rather than doing the thing that the requirement of instrumental rationality requires you to do.50

He presents an example to illustrate the point. A graduate student intends to get a good job in philosophy, and, due to being dispirited, comes to believe that a necessary condition for doing so is having a famous advisor. However, after a good meal and helpful conversation with friends who provide several counterexamples, he comes to believe a famous advisor is not necessary. In this case, it seems as though when he comes to revise his belief after the good meal and conversation, he is escaping, and not complying with, a requirement of means–ends coherence. One strategy of reply would be to distinguish compliance with a requirement from conformity with a requirement.51 One conforms with a requirement when one does what is required. Compliance, however, involves intending to conform, and doing so. (And so Lord’s graduate student conforms, but does not comply, with Means–Ends Coherence, since he does what Means–Ends 49  For more on the distinction between escape and compliance and its relevance to the scope debate, see Schroeder 2004, pp. 352, 359; Lord 2014b, pp. 456–60; Schroeder 2014a, pp. 221–2; and Fogal 2016, pp. 68–80. The concern about compliance discussed in this section differs from the concern about compliance raised in our discussion of synchronic requirements in Section 3.1, which Revised Wide Synchronic MEC was designed to meet. 50  Lord 2014b, p. 460. For reasons mentioned in Chapter 1, I think it would be better to formulate it as a requirement to intend the means, rather than to take the means. But that doesn’t matter here. 51  Thanks to a referee for this suggestion.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

84  Instrumental Rationality Coherence requires, without intending to conform to that requirement.) This strategy of reply may go some ways toward meeting the objection, but is probably not enough. Note that in the passage quoted from Lord above, where he introduces the objection, he claims that giving up an instrumental belief seems to be more like getting out of an obligation than conforming (“doing the thing that the requirement of instrumental rationality requires you to do”) to an obligation. And so he would likely respond simply by reframing the escape/compliance asymmetry objection as an escape/conformity asymmetry objection. So, more needs to be said to meet it. (In what follows, I’ll put aside the compliance/conformity distinction and simply speak of “compliance,” since that’s the way the objection is put by those who make it. But keep in mind that in using “compliance” this way, I’m allowing for the possibility of “complying” with a requirement without intending to do so, perhaps even by pure accident.) I do agree that we can use our intuitions about whether one would escape or comply in proceeding a certain way to help us decide whether the requirement should be narrow-scoped or wide-scoped. This seems to help us in other domains. Consider conditional requirements of prudence.52 Here’s an example of a conditional prudential requirement from John Broome: prudence requires you to use a mosquito net, conditional on being in tropical Africa.53 Intuitively, leaving Africa would be a way of avoiding or escaping the requirement, not complying with it. In light of this, we should formulate the requirement so that “requires” has narrow-scope: if you are in tropical Africa, prudence requires you to use a mosquito net. Other conditional prudential requirements, however, might be more appropriately rendered as wide-scope requirements. Prudence requires you to use a mosquito net treated with deltamethrin, conditional on your not using one treated with permethrin. Here we think that “requires” has logical scope over the conditional; prudence requires that you either use a mosquito net treated with deltamethrin or you use one treated with permethrin.54 And our thinking that “requires” has wide scope here is supported by our intuitive judgment that both using a net treated with deltamethrin and using a net treated with permethrin are ways of 52  Legal requirements would provide good examples as well. Consider an example of a conditional legal requirement from Schroeder 2014a, p. 221: if you’re driving in New York, you are required not to turn right on red. We should not read the antecedent as being within the scope of “requires” since leaving (or not travelling to) New York seems to be a way of escaping (or avoiding) the requirement, not complying with it. 53  Broome 2013, p. 133. 54  This is my own example, not Broome’s, though I doubt he would object to it as an example of a wide-scope requirement.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  85 complying with what prudence requires of you. We’ve here used our intuitive judgments concerning escape and compliance to help us figure out whether these prudential requirements are narrow-scope or wide-scope requirements. And it seems that we can do the same thing when it comes to rational requirements as well. However, I confess that I don’t share the intuition that the graduate student escapes, rather than complies, with means–ends coherence when he gives up  his advisor belief. (Remember that we’re allowing for the possibility of non-intentional, accidental “compliance” with a requirement.) At this point, we might simply have a clash of intuitions—some of us will think the graduate student escapes while others, like me, will think he complies—and our in­tu­itions here are most likely informed by our prior thoughts about how the requirement of means–ends coherence should be formulated. However, rather than rest content with a clash of intuitions, I’ll offer an argument for thinking the graduate student complies with means–ends coherence. If the argument succeeds, it will show that, perhaps despite initial appearances, there’s no bullet biting here—there’s little or no theoretical cost in having to say that the graduate student complies with means–ends coherence when he gives up his advisor belief. Let’s turn to that argument. Instead of starting by thinking about means–ends coherence, where there is disagreement about whether there is an escape/compliance asymmetry, with some philosophers thinking that intending the means involves compliance while revising an instrumental belief involves escaping, and other philo­sophers (like me) thinking that both involve compliance, we can start by thinking of  related requirements where it is uncontroversial that there is no escape/ compliance asymmetry. Consider, for instance, the rational requirement against believing that P and believing that ~P, and consider someone who currently believes that P and believes that ~P. No one would seriously hold that dropping the belief that P escapes the consistency requirement while dropping the belief that ~P complies with it, or vice versa. (There’s simply no basis for any such claim. It’s not as though there’s something special about dropping a belief with a negation as part of its content.) Likewise, we could consider a consistency requirement on intentions, which prohibits one from intending to X and intending not to X. Again, there is no escape/compliance asymmetry here. No one would seriously maintain that dropping an intention not to X escapes the requirement while dropping the intention to X complies, or vice versa. Though the example of inconsistent beliefs above involves two beliefs (the belief that P and the belief that ~P), it’s uncontroversial that one could have inconsistency in a larger set of beliefs. Suppose, for instance, that I believe

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

86  Instrumental Rationality that P, believe that P→Q, and believe that ~Q. In this case, again, no one would seriously hold that there’s an escape/compliance asymmetry. No one thinks that by dropping my belief that P, I would comply the requirement to have consistent beliefs while by dropping my belief that P→Q, I would be escaping the requirement. (Again, there’s no basis for such a claim. It’s not as though there’s something special about dropping a belief whose content takes the form of a conditional.) So, for instance, let’s suppose that I have the following inconsistent set of beliefs: Anastasia will get a good job; Anastasia will get a good job only if she has a famous advisor; and, Anastasia does not have a famous advisor. When it comes to giving up these beliefs, dropping any one of them would comply with the consistency requirement on belief. There’s no escape/compliance asymmetry here either. Rationality requires not only that our intentions be consistent with one another, and that our beliefs be consistent with one another, but also that our intentions be consistent with our beliefs. As Michael Bratman puts it, there is “a rational demand that one’s intentions, taken together with one’s beliefs, fit together into a consistent model of one’s future.”55 Rationality prohibits me from, for instance, intending to X and believing I will not X. Like the consistency requirement on beliefs, this requirement extends to sets of attitudes larger than two. For instance, rationality prohibits me from intending to X, believing I will X only if I do not Y, and intending to Y. What is prohibited here is a certain combination of intentions and beliefs, specifically a com­bin­ ation that doesn’t, as Bratman puts it, fit together into a consistent model of one’s future. The consistency requirement here isn’t exclusively within the domain of practical rationality nor exclusively with the domain of theoretical rationality; rather, it concerns how our intentions and beliefs hang together. If one insists that this requirement exists exclusively within the domain of practical rationality, one might think that, when I intend to X and believe I will not X, revising the former attitude complies with the requirement while revising the latter escapes it. But that would involve working from a theoret­ ic­al assumption that isn’t well-motivated. As we noted earlier, the requirement covers both intentions and beliefs, requiring that their contents fit together into a consistent model of the future. There wouldn’t be any reason to assume that the requirement is “really” about intentions, and so compliance can occur only through revisions in intentions. Also, we should note that in the same way in which I can comply with the requirement of belief consistency by dropping a conditional belief, I can come to comply with the requirement of 55  Bratman 2009b, p. 29.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  87 intention-belief consistency by dropping a conditional belief, like a belief of the form “I will X only if I do not Y,” when I intend to X and intend to Y. So, it seems that we should think that someone complies with the requirement of intention-belief consistency when she gives up her belief, both in the twoattitude case, where she intends to X and believes she will not X, and in the three-attitude case, where she intends to X, intends to Y, and believes she will X only if she does not Y. Let’s now return to Lord’s example of the graduate student who revises his advisor belief after having a good meal and helpful conversation with friends. I want to consider two versions of this example. In the first version, we’ll assume that the graduate student, before he revises his belief, has the following prohibited combination of attitudes: an intention to get a good job, the advisor belief, and an intention not to have a famous advisor. In this version of the case, the requirement of intention-belief consistency is applicable, and dropping any one of the three attitudes would be a way of complying with the intention-belief consistency requirement. In the second version of the case, we’ll assume that he has the following prohibited combination of attitudes: he has an intention to get a good job, has the advisor belief, and doesn’t intend to have a famous advisor. (In the first case, he intends not to have a famous ad­visor; in the second case, he doesn’t intend to have a famous advisor.) My inclination is to treat this second case along the same lines as the first case, in that revising in any of three ways would be a way of complying with the requirement. The narrow-scoper who appeals to the escape/compliance asymmetry ­concerning means–ends coherence, however, treats the two cases differently. On this view, when the graduate student has the good meal and hears the counterexamples from his friends, and then abandons his belief that having a famous advisor is necessary for getting a good job, he in the first case complies with the relevant requirement, while in the second escapes it. But it’s hard to see why the two cases should be treated so differently, especially since the difference between them is ever so slight—intending not vs. not intending— and that difference isn’t at all relevant to how he proceeds in his revisions. In light of these considerations, it’s not unreasonable to think that the graduate student complies with means–ends coherence when, in light of the evidence, he gives up his advisor belief. Thinking this allow for us to treat cases of means–ends incoherence along the same lines as very similar cases of intention-belief inconsistency. So, I’m not inclined to think of it as a cost to the wide-scope view that it predicts that the graduate student complies rather than escapes. I think it’s an advantage.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

88  Instrumental Rationality Indeed, the narrow-scoper faces an explanatory burden. What would justify treating means–ends coherence so differently from intention-belief consistency, given the difference is ever so slight (not intending vs. intending not) and irrelevant? Or, if they are treated alike, what would justify treating intention-belief consistency so differently from a requirement prohibiting one from believing that P, believing that P→Q, and believing that ~Q (as in the example of my beliefs about Anastasia)? Or, if those are all treated alike— that is, if we treat dropping the conditional belief as a way of escaping—what would justify treating inconsistency in three beliefs (that P, that P→Q, that ~Q) so differently from inconsistency in two beliefs (that P, that ~P)? The widescoper avoids these troublesome questions.

3.4 Conclusion In this chapter, I have considered the question of how best to formulate rational requirements, especially the requirement of means–ends coherence. I  take rational requirements to be local requirements governing conflicts among our attitudes that can be formulated as synchronic or diachronic requirements. Means–Ends Coherence, for instance, governs the conflict between one’s intending an end, believing that intending some means is ne­ces­sary, and not intending those means. It can be understood as a synchronic requirement not to have that combination of attitudes at a time, and as diachronic requirements to avoid entering, and to escape from, that prohibited state, with the synchronic requirement plausibly explaining the diachronic ones. Using the cases of Annie and Aaron, I’ve developed arguments in favor of some wide-scope formulations, particularly Wide Enkratic and Mean-Ends Coherence. And similar arguments can be constructed for the wide-scope formulations of other rational requirements. I’ve also considered two im­port­ ant strategies that narrow-scopers have pursued in reply to these arguments: the introduction of a theory about how requirements are weighed up, and the introduction of cancelling conditions into the formulation of the requirements. However, I’ve given reasons to think each of these strategies involves the narrow-scoper in further difficulties. I have also considered what I take to be the strongest argument against the wide-scope formulation of rational requirements: the asymmetry objection. I’ve suggested a two-part strategy of reply, according to which we first note that requirements like Means–Ends Coherence tell us merely what’s required

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Scope  89 so far as instrumental rationality goes, and other requirements are brought in to explain our asymmetry intuitions. With this two-part strategy, we can explain why intending the means seems more rational than abandoning an end or abandoning an instrumental belief. Additionally, we have considered whether it’s a cost to say that in abandoning one’s ends or one’s instrumental belief one complies with, rather than escapes from, the requirement of means–ends coherence. I’ve suggested that if we think of means–ends coherence on the model of intention-belief consistency, there’s no cost in saying this.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

4 Normativity I Now that we have defended the wide-scope formulation of means–ends coherence, we can turn to the question of whether means–ends coherence is “normative” in the sense that we ought, or (more weakly) have a reason, to comply with it. This is a special sense of “normative,” which, while common, is by no means the standard usage among philosophers. For instance, Philippa Foot refers to the requirements of etiquette as “normative” while also thinking— and this was her point in bringing up such requirements—that there may be no reason for you to comply with them.1 So, we’ll stipulate that in this chapter and the next, we’ll be using “normative” in the “reason-implying sense” of the term.2 While there has been some recent discussion of whether rational requirements in general are normative, our question is narrower, since we are concerned with the normativity of means–ends coherence in particular.3 However, if we show that means–ends coherence isn’t normative, that would suffice to show that not all rational requirements are normative. We can distinguish two kinds of normativity which means–ends coherence might have: Strong Normativity:  You always ought to be means–ends coherent. Weak Normativity:  You always have a reason to be means–ends coherent. One could disagree with both of these claims and still hold that by and large one ought to be means–ends coherent, or by and large one has a reason to be means–ends coherent, since these views would fall short of saying we ought to, or have reason to, comply with means–ends coherence on every particular 1  Foot 1972, pp. 309–10. 2  This sense of “normative” is familiar from Parfit (2011, p. 144), though he uses it to refer to norms that imply reasons or apparent reasons. Raz (1999b, p. 67) also understands normativity in terms of reasons: “The normativity of all that is normative consists in the way it is, or provides, or is otherwise related to reasons.” Such approaches will not go on to distinguish between those reasons that matter and those that do not (e.g., Finlay 2006), since they take all reasons to matter. 3  Discussions of the normativity of rational requirements in general can be found in Kolodny 2005, Southwood 2008, Way 2009, Broome 2013, Ch. 11, and Ross 2012, among others. Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  91 occasion. For the same reason, one could disagree with both of these claims and still hold that our being disposed to be means–ends coherent best leads us, in the long run, to be as we ought to be.4 In asking whether requirements of rationality are normative in this sense, we are asking a question often asked of other sets of requirements. Philosophers of law have asked whether legal requirements are normative in the sense that we always ought, or have a reason, to obey them.5 And metaethicists have long been concerned with the question of whether moral requirements are normative in this sense. (Moral rationalists have defended the normativity of morality against those who are persuaded by Foot’s suggestion that moral requirements are on a par with the requirements of etiquette so far as their normativity goes.6) And, in theory, this “normativity question” could be posed for any set of requirements—requirements of prudence, morality, law, etiquette, Catholicism, chivalry, and so forth—as well as for any subset of those sets of requirements.7 We can ask, of any requirement, whether we ought, or have a reason, to comply with it.8 4 See Kolodny  2008b for skepticism about this claim, and Jennifer Morton  2011, especially pp.  575–7, for criticism of Kolodny’s skepticism. On Morton’s “ecological theory” (Morton  2011; Morton 2016; Morton and Paul forthcoming) we should approach the normativity of rationality by asking which deliberative norms (which Morton takes to be defeasible, stable, background habits and dis­posi­tions) are such that they best lead us to do or believe what we ought to do or believe in the long run, even if they occasionally lead us astray. The answer to this question, on her view, will depend on particular contextual facts about the agent and her environment, and is not answered by a priori the­ or­iz­ing. I’m sympathetic to this approach. However, it’s worth noting that it won’t help establish Strong (or Weak) Normativity, since it could be the case that one ought to have a disposition without it being the case that one ought (or even has a reason) to manifest the disposition on every particular occasion. 5  See, for instance, Smith 1973. 6  See, for instance, Shafer-Landau 2003, pp. 190–214. 7  Obviously, you “legally ought” to comply with legal requirements, “morally ought” to comply with moral requirements, and so forth. That’s true by definition. But, in asking whether a requirement is strongly normative, we’re interested in whether it’s also true that you ought to comply with it in that you have “most reason” to comply with it. We might call this “the ‘ought’ of practical reason.” (There’s a further question of whether we should understand this “ought” as being in some sense “objective,” or in some sense relative to an agent’s information.) 8  I’m here setting up the question of rationality’s normativity differently from how John Broome sets it up. Broome’s “normative question” about rationality asks whether the following claim is true: “Normativity of Rationality: Necessarily, if rationality requires N to F, that fact is a reason for N to F.” (2013, p. 192) Broome thinks the view that “necessarily, if rationality requires N to F, there is a reason for N to F” is “not a version of the claim that rationality is normative” since it merely “associates rationality with normativity” without saying that rationality is a source of normative requirements. (2013, p. 193) However, this saddles defenders of rationality’s normativity with an additional burden: they must, on Broome’s formulation of the question, defend the claim that the fact that rationality requires N to F is a reason for N to F. But would we suggest the same for those who defend the normativity of morality? Must they defend the view that the fact that morality requires N to F is a reason for N to F? If they denied this, but held that those features which go into making it the case that morality requires N to F (e.g., that N has promised to F, that F-ing relieves suffering) are reasons for N to F (or, more strongly, make it the case that N ought to F), would they thereby be denying the normativity of morality? It doesn’t seem so. For a related excellent criticism of Broome’s understanding of normativity, and defense of an alternative understanding, see Guindon 2016.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

92  Instrumental Rationality I’ve formulated Strong Normativity and Weak Normativity above so that they are applicable to either the wide-scope or the narrow-scope formulations of means–ends coherence. But the combination of the narrow-scope formulation and Strong Normativity is not promising. On the narrow-scope formulation, when Bob intends to kill his colleagues, and believes that to do so he must intend to poison them, rationality requires that he intend to poison them. If means–ends coherence is strongly normative, then it follows that Bob ought to intend to poison them. But that’s obviously false. What about the combination of the narrow-scope formulation and Weak Normativity? From that combination, it would follow that Bob has a reason to intend to poison them. This conclusion isn’t obviously false, so long as we emphasize that the reason is a lightweight one. But this view would still be vulnerable to the version of the bootstrapping objection developed in Chapter  2, and so should be rejected. As I mentioned in the previous chapter, the difficulty of combining ­narrow-scope formulations with the normativity of rationality is a common mo­tiv­ation for adopting the wide-scope formulations. Adopting a wide-scope formulation of means–ends coherence avoids these problematic predictions about the case of Bob. For instance, if we combine Means–Ends Coherence with Strong Normativity, we’d get the result that Bob ought to either not intend to kill his colleagues or not believe intending to poison them is necessary or intend to poison them. And if we combine Means–Ends Coherence with Weak Normativity, we’d get the result that Bob has a reason to either not intend to kill his colleagues or not believe intending to poison them is ­necessary or intend to poison them. These predictions concerning Bob don’t appear to be as troublesome. However, the approach taken in this book departs from this common motivation for the wide-scope formulations in at least two ways. First, as I argued in Chapter 3, there’s good reason to adopt the wide-scope formulations independently of how we answer the further questions concerning their normativity. Second, as we’ll see in Chapter  5, even using the wide-scope formulation, there are very good reasons to doubt the normativity—at least the strong normativity—of means–ends coherence. In this chapter and the following one, I’ll consider three different ways in which philosophers have appealed to normativity in order to explain, or explain away, the rational requirement to be means–ends coherent. First, in this chapter, we’ll start with Niko Kolodny’s forceful development of an idea found in the work of Joseph Raz: the idea that a proper understanding how normativity governs our beliefs and intentions will reveal the existence of formal requirements of coherence, like Means–Ends Coherence, to be a myth.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  93 More precisely, if we pay attention to the specific intentions and beliefs one ought to have—not “rationally ought,” but the “ought” of “most reason”—then we’ll see that in cases of means–ends incoherence, either one ought not intend the end, or one ought to intend the means, or one ought not believe one will achieve the end only if one intends the means. But, the myth theorist argues, we made an error in mistaking this disjunction of oughts for a disjunctive rational requirement, Means–Ends Coherence. Second, in Chapter 5, I will turn directly to the question of the normativity of Means–Ends Coherence, focusing in particular on Strong Normativity, since, as we’ll see, that’s what would be needed to explain the strictness of the rational requirement. Third, also in Chapter  5, I’ll consider the strategy, developed in different ways by Mark Schroeder and Jonathan Way, of accounting for means–ends coherence by an appeal to subjective reasons and the subjective ought. Each of these three strategies offers us a unique way of developing the nat­ural, intuitive idea that the rational requirement of means–ends coherence can be explained by looking at the attitudes, or combinations of attitudes, we ought to have (where this is the “ought” of “most reason” or, in the case of the subjective ought, an “ought” closely related to it). But I’ll conclude that we should be skeptical of all three of these strategies, and look elsewhere for an explanation of means–ends coherence.

4.1  Normative Disjunctivism One might start with the observation that in cases of means–ends incoherence, we can usually identify some specific attitude the agent ought to have but doesn’t, or ought not to have but does. In the case of Aaron, who intends to pass the test but doesn’t intend to study, which he knows he must do, it’s probably the case that he ought to form the intention to study. In the case of Bob, he clearly ought to give up his intention to kill his colleagues. And in other cases one ought to give up one’s instrumental belief. Suppose I believe, against the evidence, that I can eat healthily only if I intend to avoid foods with genetically modified ingredients, and I’m also means–ends incoherent: I intend to eat healthily and don’t intend to avoid food with genetically modified ingredients. Here, I should give up my belief that I can eat healthily only if I intend to avoid foods with genetically modified ingredients, since that belief isn’t sufficiently supported by the evidence. These cases might lead us to the conjecture that in general means–ends incoherent agents have some specific

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

94  Instrumental Rationality attitude they ought not have or lack some specific attitude they ought to have. We can call this view: Normative Disjunctivism:  If someone is means–ends incoherent, then either she ought not intend the end, or she ought not believe intending the means is necessary for the end, or she ought to intend the means.9 This view would allow us to hold that whenever one is fully responsive to reason—that is, whenever one holds the intentions and beliefs one ought to hold, and lacks the intentions and beliefs one ought to lack—one is means–ends coherent. We should distinguish Normative Disjunctivism from Strong Normativity. Both views hold that whenever you are means–ends incoherent, it’s guaranteed that you are not as you ought to be. But according to Strong Normativity, that’s because you ought to be means–ends coherent and are not. According to Normative Disjunctivism, that’s because there is some specific attitude you ought to have but don’t, or some specific attitude you ought not to have but do. The views are compatible; it could be that both Normative Disjunctivism and Strong Normativity are true. The theoretical interest in Normative Disjunctivism is largely due to the support it promises to provide for the theory that means–ends coherence is a myth. The myth theory, as Niko Kolodny puts it, attributing the idea to Joseph Raz, “denies, in particular, that there is a rational requirement of formal coherence that governs means and ends.”10 And so the myth theorist would deny means–ends coherence. Many philosophers believe there is some such requirement governing means and ends. So, we need to explain where they went wrong. And Normative Disjunctivism can figure into that error theory: our mistake was that we confused a disjunction of oughts with a rational requirement with scope over a disjunction (Means–Ends Coherence). We didn’t go wrong in thinking that when you’re means–ends incoherent, you violate some requirement. But we did go wrong in identifying what the

9 I’m here following Jonathan Way (2013, 2018) who refers to such views as versions of “Disjunctivism.” But I’m adding “Normative” to it to distinguish it from the Non-normative Disjunctivism that will be introduced in Chapter 7. 10  Kolodny 2008a, p. 366. See also Kolodny 2007b and 2008b, and Kiesewetter 2018, Chs. 9 and 10. Errol Lord (2014a, 2017) also defends a version of Normative Disjunctivism, but not in the service of a myth theory. (See especially Lord 2014a, p. 170.) Given the possibility of views that accept Normative Disjunctivism without endorsing the myth theory, it will be worth devoting attention to the question of whether Normative Disjunctivism is true, even if—as I’ll argue in Section 4.3—it doesn’t actually lend support to the myth theory.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  95 requirement is that you violate. It’s not that you violate a wide-scope requirement of rationality. Rather, you violate a requirement in that you have some specific attitude you ought not to have, or fail to have one you ought to have.11 So, we have two questions to consider in this chapter. First, is Normative Disjunctivism true? Second, if it is true, does it support the myth theory? I’ll now argue that there are several good reasons to think Normative Disjunctivism is false (Section 4.2) and doesn’t support the myth theory (Section 4.3).

4.2  Is Normative Disjunctivism True? Normative Disjunctivists think that whenever you are means–ends incoherent, either you ought to give up the end, or you ought to intend the means, or you ought to give up your instrumental belief. To reject this view, we need to find cases in which you are means–ends incoherent, but none of these claims about what you ought to intend or believe are true. I’ll present four such cases below, one in each of the first four subsections. In the fifth subsection, I’ll consider the prospects for a variation on Normative Disjuctivism which I’ll call “Normative Conjunctive Disjuctivism.” This view, I’ll argue, might avoid one of the four objections, but would be subject to the other three. Some of the cases I use in developing these four objections will also play a role in the arguments against Strong Normativity in Chapter 5.

4.2.1  “Ought” and Ability Many philosophers accept a version of ought-implies-can according to which if you are psychologically unable to do something, then it’s not the case that you ought to do it.12 (This view is compatible with the “ought” remaining in place when it’s merely very difficult for you to do so something, as it might be in many cases of compulsion and addiction.) On this view, if you are psychologically unable to revise an intention, then it’s not the case that you ought to 11  Kolodny also argues for such an error theory about the rational prohibitions against having inconsistent beliefs (Kolodny  2007b, pp. 235–6; 2008b, pp. 440–1) and inconsistent intentions (Kolodny 2008a). 12  See Vranas 2007 for a defense of a version of ought-implies-can along these lines, and for many additional references.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

96  Instrumental Rationality revise that intention. I’ll here assume the truth of a view along these lines, and consider its implications for Normative Disjunctivism. Let’s start with an observation made by Kieran Setiya in another, related context: it’s possible for someone to display instrumental irrationality (or rationality) with regard to some intention that cannot be revised. Setiya provides the example of a means–ends incoherent agent who intends to smoke a cigarette—it’s stipulated that he cannot revise this intention—and who also believes that to smoke a cigarette he must intend to buy a pack, but doesn’t intend to buy a pack.13 Since this is an example of means–ends incoherence, Normative Disjunctivism holds that either he ought not intend to smoke a cigarette, or he ought not have this instrumental belief, or he ought to intend to buy a pack. We can assume that there’s conclusive available evidence that he’ll smoke a cigarette only if he intends to buy a pack, so it’s not true that he ought not have this instrumental belief. And we can assume that, since smoking is very harmful and expensive, it’s not the case that he ought to intend to buy a pack. So, according to Normative Disjunctivism, he  ought not intend to smoke. But, as we saw, since he cannot revise his intention to smoke, it’s not the case that he ought to do so. So, Normative Disjunctivism is false. One way to resist the argument, besides challenging the relevant version of ought-implies-can, would be to deny the possibility of unrevisable intentions. In other words, one might argue that intentions are necessarily revisable. That might be true. But it doesn’t help us here, since, even if we’re not licensed to call our agent’s attitude—which has all the features of an intention except for revisability—an “intention,” and must instead call it something else, it’s still true that one could display instrumental rationality or irrationality with respect to this attitude, whatever it’s called. The person who has this attitude, but who gets distracted and doesn’t intend to buy a pack, while believing doing so is necessary for smoking, would clearly be displaying instrumental irrationality. Likewise, if a motivational failure keeps him from following through on his plan to smoke a cigarette—perhaps he’s too depressed to get out of bed and go buy a pack—he would be displaying instrumental irrationality. (Note also that the source of the unrevisability of the intention need not make a psychological difference in the agent. The unrevisability could, for instance, be due to a Frankfurtian counterfactual intervener, unknown to you, who would prevent you from revising your attitude were you to try.14 But suppose now you don’t try, but, through distraction or depression, fail to 13  See Setiya 2007, pp. 660–1.

14  I discuss such cases in Brunero 2010b.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  97 intend the means. I don’t see how the unrevisability of your intention makes any difference to your rationality. You seem just as irrational as you would be were you to proceed the same way in a world with no such intervener.) In short, it’s natural to think that the agent in this example violates some requirement of means–ends coherence. So, if the myth theorist is going to appeal to Normative Disjunctivism in support of skepticism about any such requirement, then this must be a case in which the predictions of Normative Disjuctivism hold true—that is, it must be a case in which either one ought not have the intention (or whatever it is) for the end, or ought not have the instrumental belief, or ought to intend the means. But, I’ve argued, those predictions do not hold true. In the reply above, I noted that one can display instrumental irrationality of the usual sort, perhaps due to distraction or a failure in motivation, with respect to an intention to smoke which cannot be revised. But another possibility is that the agent, while unable to revise his intention to smoke, takes care to limit the influence of this intention on his other attitudes and actions, insofar as he is able to do so. In such a case, his not intending to buy a pack of cigarettes may not be a rational failure at all, but a kind of rational success: the agent succeeds in restricting the influence of this unendorsed but unrevisable intention. These are interesting cases.15 But all I need for my argument here is that there are some possible cases in which one displays means–ends incoherence of the usual sort with respect to the unrevisable intention. In such cases, the predictions of Normative Disjunctivism do not hold true.

4.2.2  Advantageous Incoherence It’s often noted that in certain contexts various forms of irrationality can prove advantageous. Thomas Schelling famously wrote that it’s “perfectly clear that it is not a universal advantage in situations of conflict to be inalienably and manifestly rational in decision and motivation.”16 For instance, he notes, there 15  I won’t pursue here the question of whether we need to revise Means–Ends Coherence to accommodate such cases. It may be that we don’t. If we see Means–Ends Coherence as specifying a necessary condition for being (fully) rational—namely, that you must not have the combination of intending an end, believing intending some means is necessary, and not intending the means—then such cases might not be counterexamples to Means–Ends Coherence, since the agent here still seems, in virtue of being unable to modify his intention to smoke, not fully rational. (Even if the agent is responding rationally to that underlying irrationality by seeking to limit the influence of that intention to smoke, he is nonetheless not fully rational, though perhaps through no fault of his own.) For more on this way of understanding rational requirements, see Broome 2013, pp. 110–16. 16  Schelling 1960, p. 18.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

98  Instrumental Rationality could be a situation in which one could evade extortion by taking a drug that renders one irrationally non-responsive to threats. This example suffices to show that it “is not true . . . that in the face of a threat it is invariably an advantage to be rational, particularly if the fact of being rational or irrational cannot be concealed.”17 We’re here concerned not with rationality in general, but with means–ends coherence in particular. But it seems that the point could be made in this context as well: it’s not universally advantageous to be means–ends coherent. Here’s a fictitious example. (Any resemblance to real persons living or dead is purely coincidental.) The Philosophy Department Chair regularly distributes various service responsibilities and projects to faculty members in the Philosophy Department. Professor Henri Liable, who genuinely intends to do his fair share of service, has a long track record of being unreliable when it comes to following through with the projects he takes on. Henri displays only those vices usually tolerated in academic circles—absent-mindedness, ­procrastination, etc.—without doing anything to suggest any intellectual incompetence. The Chair, who can read people quite well, sees that Henri genuinely intends to help out, but cannot be depended upon to follow through. As a result, the Chair shifts various service responsibilities away from Henri and toward the other, more practically competent, members of the Department, thereby freeing Henri up to focus more on his research. Since Henri genuinely intends to help out, and volunteers for various projects, he never loses the good will of the Chair, who sees that he’s not faking incompetence in order to avoid service. But Henri’s instrumental irrationality ensures that the more burdensome service responsibilities go to other members of the Department, which is to Henri’s advantage. In this case, there’s a strategic advantage to being means–ends incoherent. The feature mentioned by Schelling above obtains in this case: Henri’s instrumental rationality or irrationality is not concealed. If he’s instrumentally rational, more projects will get thrown his way, and he’ll have much less time for research. If he’s instrumentally irrational, fewer projects will get thrown his way, and he’ll have much more time for research. In this situation, it seems natural to think that Henri ought to be instrumentally irrational.18 In doing so, he would uphold his reputation for not following through, and

17  Schelling 1960, p. 18. See also Parfit 1984, pp. 12–13. 18  We’ll set aside the relevant moral reasons for Henri to contribute his fair share, or assume they’re outweighed in this case.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  99 gain all of the advantages that come with that, without losing the good will of the Chair. In this case, the advantage of having more time for research comes from Henri’s general reputation for not following through on his projects. But since particular instances of means–ends incoherence will help cultivate and maintain that reputation, we can say that there are advantages that come from particular instances of means–ends incoherence. Suppose Henri has some particular incoherent combination of attitudes: he intends to get his new course certified for general education credit, believes that to do so he must intend to get in touch with the relevant administrators, but doesn’t intend, nor trust that he will later intend, to get in touch with the relevant administrators. We’ll assume there’s conclusive evidence for his instrumental belief, and so that belief is one he ought to have. According to Normative Disjunctivism, then, either he ought to intend to talk with the administrators or he ought to give up his intention to get the course certified. But it’s not true that he ought to intend to talk with the administrators, since talking with them will demonstrate his effectiveness, just increasing the likelihood of more service being thrown his way. But it’s also not true that he ought to give up his intention to get the course certified, since in doing so, he would lose the good will of the Chair. Rather, it seems that he should continue to be means–ends incoherent, getting off the hook of having more burdensome responsibilities without angering the Chair. But if that’s right, the prediction issued by Normative Disjunctivism is false.19 One way for the Normative Disjunctivist to avoid this result would be to adopt a skepticism about the existence of the “wrong kind” of reasons for intentions. One might first propose some way of distinguishing the “right” and “wrong” kind of reasons for intentions—no easy task—and then argue that Henri’s reason to intend to get the course certified (that he’ll keep the good will of the Chair) is of the “wrong” kind, and hence not really a reason for him to intend to get the course certified. And since what one ought to do 19  I’ve argued above that Henri ought to be means–ends incoherent, since doing so is advantageous. But doesn’t that, one might object, also make it rational for him to be means–ends incoherent? This thought might be supported by some conception of rational behavior as behavior directed toward what is, or what is expected to be, to one’s advantage. In reply, it’s worth noting that we wouldn’t accept a similar objection concerning epistemic rationality: showing that having some belief or combination of beliefs would be to your advantage won’t suffice for showing that the belief or combination of beliefs is rational. And that’s because we have some independent standards of theoretical rationality that aren’t entirely driven by thoughts about to what’s to your advantage. (See Kelly 2003.) In the same way, I don’t think that showing that some pattern of attitudes is to your advantage suffices to show that the pattern is rational. Again, we have some independent standards—particularly, standards prohibiting you from being “at odds with yourself,” to borrow Jonathan Dancy’s (2009, p. 96) apt phrase—that aren’t entirely driven by thoughts about what’s to your advantage.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

100  Instrumental Rationality is determined by the balance of reasons—this reason now being excluded from that balance—we can say that he ought not intend to get the course certified.20 And so this case is no counterexample to Normative Disjunctivism. I’m not optimistic about this strategy. On the leading account of how to distinguish reasons of the right kind from those of the wrong kind, the right kind of reasons are “object-given” while the wrong kind are “state-given.”21 On this view, the right kind of reasons to intend to X will be considerations favoring the object of your intention (your X-ing) while the wrong kind of reasons will be considerations favoring your being in the state of intending to X without also favoring the object of your intention. (Kavka’s (1983) toxin puzzle is a standard example: the billionaire’s incentive is a state-given reason for you to intend to drink the toxin, since it favors your intending to drink the toxin without also favoring your drinking the toxin.22) But skepticism about the existence of state-given reasons to intend is implausible. Suppose you are considering two equally good summer vacation trips, one to Provincetown and one to Block Island, and you know that intending to go to Block Island will fill you with intense feelings of joy while intending to go to Provincetown will leave you cold. This is a state-given reason to intend to go to Block Island, since it favors your intending to go to Block Island without favoring going to Block Island. (It’s part of the example that you get the additional good feelings from your intending to go, not from your going. And there’s nothing unusual about this: there are some things the planning of which is more enjoyable than others. We can assume here that you’ll get those planning pleasures if you intend to go to Block Island, even if you don’t end up going. And if you go there without intending to go—perhaps you reluctantly agree to have someone else surprise you with vacation plans at the last minute—you won’t get them.) Although going to Provincetown will be just as good as going to Block Island, if we add in the great joy you’ll experience in intending to go to Block Island, then you ought to intend to go there. But if we were skeptics about the existence of state-given reasons, we’d have to say that there’s no 20  See Way 2012b for a defense of wrong kind of reasons skepticism. 21  For an important criticism of this view, see Schroeder 2012. On the distinction between stategiven and object-given reasons, see Parfit 2001, p. 21. 22  An interesting feature of Kavka’s case is that it raises the question of whether one can intend to drink the toxin tomorrow when one knows that by then the money will already be (or not be) in one’s bank account, and that one’s future self, insofar as he is rational, will not drink the toxin. In Kavka’s view, one can intend to drink the toxin tomorrow only with “extreme difficulty” (1983, p. 35). But the difficulty arises not from any feature of the reason (the $1M incentive), but from the agent’s beliefs about what his future self will do. We can see this by thinking of a case with the same reason poses no difficulty: suppose the billionaire offers to put $1M in your bank account if you intend to go to Block Island. It’s easy for you to form that intention.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  101 more reason for you to intend to go to Block Island than to intend to go to Provincetown.23 And that isn’t plausible.24

4.2.3  Practical and Theoretical Reason Many moral philosophers are attracted to views about well-being, reasons, or value which appeal not to an agent’s actual desires, but to her “idealized” desires—roughly, the desires she would hold were she fully aware of, or fully imaginatively acquainted with, all the relevant facts—particularly, with facts about what it would be like for her to realize her desires.25 (Such views differ in several respects, particularly with regard to how they spell out the relevant idealization.) For instance, on an informed desire view of what’s good for a person, we aren’t forced to say that the satisfaction of a person’s actual desires contributes to a person’s good when those desires are based on false beliefs. Consider this example from Peter Railton.26 Lonnie, a traveler who feels uneasy, desires to settle his stomach and thinks that do so he must have a drink of milk. However, the milk will in fact increase his dehydration and make him feel worse. What Lonnie really needs is a drink of water. On Railton’s view of a person’s good, it’s not Lonnie’s actual desires, but his idealized desires, which are relevant. Here is Railton’s explanation of the relevant idealization: Give to an actual individual A unqualified cognitive and imaginative ­powers, and full factual and nomological information about his physical and 23  Those who deny that such considerations are reasons to intend to go to Block Island will often offer a consolation prize: such considerations are reasons to want to intend to go to Block Island, or ­reasons to try to bring it about that one intends to go to Block Island. But such consolation prizes aren’t sufficient to avoid the objection. To break the tie, and say that I ought to intend to go to Block Island, we need to allow that such considerations are reasons to intend to go to Block Island. 24  Jacob Ross 2012, p. 156, makes a similar point with his Buridan-Kavka case. One possible reply would be to reject the theory of the “right kind” of reasons as “object-given” reasons (and the “wrong kind” of reasons as “state-given” reasons) and replace it with another theory. But that would require that we have some other theory. And, moreover, the right kind/wrong kind distinction is sometimes itself presented—independently of the theory of what the distinction amounts to—in a way that would rule out this reason to intend to go to Block Island. (For instance, Jonathan Way (2012a, p. 497), in explaining the right kind/wrong kind distinction writes: “In the case of intention, the right kind of reasons to intend derive from reasons to perform the action intended. The wrong kind of reasons derive from other benefits of intending….”) However, there might be room for optimism here, since the reason to intend to go to Block Island doesn’t share all the typical features of the wrong kind of reasons (difficulty of one’s being motivated by them, intuitive irrelevance to rationality, etc.) See Schroeder’s (2012, pp. 458–61) discussion of the “earmarks” of the wrong kind of reasons. 25  For examples of such views, see Griffin 1986, Williams 1981, and Lewis 1989. 26  Railton 1986, pp. 174–5.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

102  Instrumental Rationality psychological constitution, capacities, circumstances, history, and so on. A will have become A+, who has complete and vivid knowledge of himself and his environment. We now ask A+ to tell us not what he currently wants, but what he would want his non-idealized self A to want – or, more generally, to seek – were he to find himself in the actual conditions and circumstances of A.27

Lonnie+ would want that Lonnie seek out water, not milk. If we understand what’s good for Lonnie in terms of such idealized desires, we can then say that drinking the milk isn’t good for Lonnie, while drinking the water is. Along similar lines, Bernard Williams develops a view of practical reasons according to which an agent may have no reason to satisfy some desire when that desire is based on a false belief.28 Suppose you want to drink the contents of the glass in front of you, which you think is a gin and tonic, your favorite drink. However, unbeknownst to you, the bartender has swapped out the gin for gasoline. In Williams’s view, you have no reason to drink the contents of the glass in front of you, although you may believe that you do. The move from actual desires to idealized desires might be motivated by a concern for extensional adequacy. We can avoid saying, in these particular cases, that it would be good for Lonnie to drink the milk, or that there’s a reason for Williams’s agent to drink the gasoline and tonic. And that’s an advantage. But there’s an independent rationale for idealizing as well. Intuitively, when Williams’s agent orders the gin and tonic, and gets the gas­oline and tonic and spits it out, he would complain that he didn’t “really want” to drink the contents of the glass. What he really wanted was a gin and tonic, and he wanted to drink the contents of the glass only because he mistakenly thought the glass contained gin and tonic. Likewise, when Lonnie realizes that the milk has further upset his stomach, he’s likely to say that he didn’t “really want” to drink the milk. What he really wanted was something to settle his stomach, and he only wanted the milk because he mistakenly thought it would settle his stomach. In giving priority to our idealized desires over our actual desires, we do better at having our theory (whether of well-being or reasons) reflect what we really want. David Sobel provides a more precise statement of this rationale for appealing to idealized, rather than actual desires:

27 Railton  1986, pp. 173–4. See also Smith  1995, pp. 109–12, on the “advice model” and the “ex­ample model.” 28  Williams 1981, p. 102.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  103 The rationale I find most persuasive is right on the surface, and it is what I have always assumed motivated and guided various idealization proposals. The rationale for granting the idealized agent information and experience is to provide her with a more accurate understanding of what the option she is considering would really be like. This explains the shape that the idealization proposals take and provides a clear rationale for idealization.29

As Sobel notes, when we say that an agent didn’t “really” want some option, what we mean is that she wanted it only because she was ignorant of what it would be like to have that option. As he puts it: “Our informed desires are, in a sense, more genuinely for their object. And such desires are what we or­­ dinar­ily would have referred to as what we ‘really want’. ”30 Railton offers a theory of what’s good for a person, and Williams offers a theory of what there’s reason for a person to do. But it’s natural to think that such theories will inform our theory of what an agent ought to do. For instance, if there’s nothing good that will come from Lonnie’s drinking the milk, and there’s no reason—or, at least, no strong reason—for him to drink the milk, then it seems that we shouldn’t think that he ought to intend to drink it. But now we are in a position to see a problem for Normative Disjunctivism. Suppose Lonnie has the following attitudes: (i) he intends to settle his stomach, and (ii) he believes he’ll settle his stomach only if he intends to drink milk. According to Normative Disjunctivism, either Lonnie ought not intend to settle his stomach, ought not believe he’ll settle his stomach only if he intends to drink milk, or ought to intend to drink milk. But it’s not true that Lonnie ought not intend to settle his stomach. That’s a permissible end for him to have. And, as we noted above, given that Lonnie desires to drink milk only because of his false belief that it’ll settle his stomach, it’s not true that he ought to intend to drink milk. That means that Normative Disjunctivism is committed to the view that Lonnie ought not believe: (S)  He’ll settle his stomach only if he intends to drink milk. But what’s troublesome about this is that we’ve arrived at a claim about what Lonnie ought not believe without saying anything about the evidence for or 29  Sobel 2009, p. 343.

30  Sobel 2009, p. 347.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

104  Instrumental Rationality against (S). Moreover, this claim about what Lonnie ought not believe will hold even when the evidence overwhelmingly supports (S). Suppose that medical science has long held, based on numerous peer-reviewed studies, that milk is the only effective means for settling one’s stomach. Moreover, suppose that on every past occasion in which Lonnie has felt uneasy, a glass of milk has done the trick, while nothing else has made him feel any better. We could even suppose that, as he’s drinking, God tinkers with the properties of water and milk, changing which one is an effective means to settling Lonnie’s stomach. Until this moment of divine intervention, it’s been the case that milk would settle Lonnie’s stomach, but now, only water will do the trick. It seems odd to say that Lonnie, unaware of God’s recent intervention, ought not believe milk is the only effective means to settling his stomach. In short, even though (S) is false, it seems that there could be situations in which one permissibly believes it. (Similarly, there could be situations in which someone could permissibly not believe some true proposition.31) One might object: if we’re using idealization to determine what’s good for a person, or what a person has reason to do, why shouldn’t we also use idealization to determine what a person ought to believe? So, just as we say that a fully informed version of Lonnie wouldn’t want actual Lonnie to want to drink the milk, we can say that a fully informed version of Lonnie wouldn’t want actual Lonnie to believe that he’ll settle his stomach only if he drinks milk. If we’re going to look to the wishes of informed advisors in practical matters, why not do so in theoretical matters as well? The answer is easy: the rationale for looking to idealized versions of ­ourselves in the practical case doesn’t carry over to the theoretical case. While it’s plausible and intuitive to say that the desires I have only because I’ve misunderstood what the option would be like aren’t “genuine” desires or what I “really” want, the analogous claims for beliefs aren’t plausible at all. The beliefs I have only because I am not well informed are in no sense less genuine or less real. Consider an example. Suppose I want to watch the Cardinals play and falsely believe they’re playing the Cubs at Wrigley on May 12, and I have an instrumental desire to see the May 12 game at Wrigley. In fact, on May 12, the Cubs are playing the Mets—a game I have no interest in watching. You’d understand me perfectly well when, upon finding this out, I say, “Well, I didn’t really want to see the May 12th game. What I wanted was to watch the

31  There’s a debate in epistemology on the question of whether the truth of some proposition, regardless of the evidence available, provides sufficient reason to believe it. For arguments that it doesn’t, and for further references, see McHugh and Way 2017.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  105 Cardinals, and I thought they were playing on the 12th.” But you won’t be able to make heads or tails of my saying, “Well, I didn’t really believe the Cardinals were playing on May 12th.” It’s not clear what this could mean. The problem here isn’t with the relevant counterfactuals. Just as it’s true that were I fully informed, I wouldn’t desire to go to the May 12 game, it’s true that were I fully informed, I wouldn’t believe the Cardinals are playing on May 12. The problem, rather, is with the claims that such counterfactuals license about our attitudes: while it makes sense for me to say I didn’t really desire to see the May 12 game, it doesn’t make sense for me to say I didn’t really believe that the Cardinals were playing on May 12. So, the rationale for idealization, clearly articulated by Sobel above, doesn’t carry over to the theoretical domain. Yet, if Normative Disjunctivism is true, idealizers would be forced to accept a rather strong claim about what we ought to believe: that regardless of the evidence for P, if P is false, you ought  not believe it. Not everyone will find this objection to Normative Disjunctivism compelling. You won’t find it compelling if (i) you don’t find this to be an unattractive view about what one ought to believe, or (ii) you don’t find much attractive about idealizing when it comes to matters relevant to what one ought to do.32 However, it’s clear that Normative Disjunctivism does come with some substantive commitments about practical and theoretical reason. It rules out a combination of views that I suspect many philosophers find attractive.

4.2.4  Mere Permissibility Normative Disjunctivism faces a well-known difficulty when it comes to ends and means that are merely permissible—that is, permissible but not required.33 For examples, we might look to the cases of ties and incommensurability discussed in Chapter 2. Suppose there are two equally good cans of tomatoes on the shelf in front of you—one on the right side and one on the left side—and you need just one of them. And suppose that a means to getting the can on the right is walking to the right and a means to getting the can on the left is walking to the left. You are permitted but not required to get the can on the right. (And you are permitted but not required to walk to the right.) And you 32  Those philosophers attracted to Normative Disjunctivism have tended to adopt “perspectival” views where what you ought to do is determined by the reasons you possess, or relative to your evidence, or something similar. See, for instance, Lord 2017, and Kiesewetter 2018, Ch. 8. 33  See Way 2012a, p. 489; 2013, pp. 598–606; and 2018, §4.1. See also Kolodny 2007b, pp. 251–2, Kolodny 2008b, pp. 452–3, and Kiesewetter 2018, pp. 263–8, for a discussion of this problem.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

106  Instrumental Rationality are permitted but not required to get the can on the left. (And you are permitted but not required to walk to the left.) But here’s the problem: you could be incoherent with respect to these ends and means. You could, for instance, intend to get the can on the right, believe that to do so you must intend to walk to the right, but not intend to walk to the right. And you could do so while having a perfectly respectable instrumental belief. If Normative Disjunctivism is true, then either (i) you ought not intend to get the can on the right, or (ii) you ought to intend to walk to the right. But neither of these claims are true: (i) is false because it’s permissible for you to intend to get the can on the right, and (ii) is false because it’s merely permissible (i.e., not required) for you to intend to walk to the right. Or suppose that you’re choosing between a career in philosophy and a career in the circus, and you’d be just as happy, well-off, etc., in either one. Here again, ends and means are merely permissible. You are permitted but not required to intend to pursue the career in philosophy. (And you are permitted but not required to intend to research graduate schools.) And you are permitted but not required to intend to pursue the career in the circus. (And you are permitted but not required to research training for the circus.) But, again, here’s the problem: you could be incoherent with respect to these ends and means. You could, for instance, intend to pursue a career in philosophy, believe that to do so you must intend to research graduate schools, but not intend to research graduate schools. And you could have a perfectly re­spect­ able instrumental belief. If Normative Disjunctivism is true, then either (i) you ought not intend to pursue the career in philosophy, or (ii) you ought to intend to research graduate schools. But the first claim is false since intending to pursue that career is permissible, and the second claim is false since intending to research graduate schools is not required.34 However, we could get around this worry if Intentions Provide Reasons or Intentions Provide Tie-Breaking Reasons were true. That would allow us to say that your intending to take the can on the right provides a reason that tips the scales, making it the case that you ought to intend to walk to the right. And your intending to pursue the career in philosophy provides a reason to research graduate schools that tips the scales, making it the case that you ought to intend to research graduate schools.35 (The career-choice case would 34  As Jonathan Way (2013, p. 600) observes, we could also use cases of supererogation to illustrate the problem of mere permissibility. 35  See Schroeder  2009, pp. 237–8; Lord  2014a, pp. 167–70; and the “modest bootstrapping” of Cheng-Guajardo 2014, pp. 496–7. For an excellent discussion of various problems with this strategy, see Kiesewetter 2018, pp. 268–75.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  107 require that the reason be substantially weighty, since, as is often observed, the addition of lightweight reasons in such cases doesn’t suffice to tip the scales. If you offer me a dollar to take the can of tomatoes on the left, then that’s what I ought to do. But if you offer me a dollar to become a philosopher, it’s still neither the case that I ought to become a philosopher nor the case I ought to become a circus performer.) And that would get around the mere permissibility objection to Normative Disjunctivism: in such cases, one ought to intend the means. However, in Chapter 2 I argued that there’s good reason to resist the view that intentions provide reasons, even in tie-breaking situations. I argued that these views lead to unacceptable bootstrapping. We can now see one payoff of those arguments: we can reject this strategy for defending Normative Disjunctivism. Moreover, we can also see one payoff from the particular conception of bootstrapping I defended in Chapter 2, that of bootstrapping as incoherent advice. Errol Lord, in defense of one version of Normative Disjunctivism,36 has suggested that while it would be objectionable bootstrapping to say that intentions provide reasons, it isn’t objectionable ­bootstrapping, or otherwise problematic, to allow that intentions can “intensify” the weight of already existing reasons to take the means, thereby making it the case that, in our examples, one ought to intend to walk to the right, and one ought to intend to research graduate schools.37 However, even allowing intentions this role of intensification would lead to the kind of unacceptable bootstrapping I was worried about. Recall the Southwest/United case: I intend to fly United but I’m permitted to change my mind and intend to fly Southwest instead. But if we allow that my intention makes it the case, whether by providing an additional reason, or by intensifying existing reasons, that I ought to click the blue square, which would book the flight with United, we would get incoherent advice: “You should click the blue square, but it’s permissible to change your mind and fly Southwest instead.” Such advice would likely be met with suspicion: “But if I should click the blue square, which would make it such that I fly United, how can it be that I’m permitted to intend to fly Southwest instead?”38 36 Lord 2014a; 2017. 37  Lord 2014a, pp. 167–70. Lord writes: “According to Intentions Intensify, one’s intentions don’t affect the reasons in favor of pursuing the end. They affect the reasons to take the means. Moreover, they do this not because they add a lightweight reason, but by intensifying” (2014a, p. 169). For general discussion of the concept of intensification (and attenuation) of reasons, see Dancy 2004, pp. 41–2. 38  As Jonathan Way (2013, pp. 603–6) observes, the Promotional Significance View of Chapter 2 will be of no help to the Normative Disjunctivist here. As we noted earlier, it’s an important feature of the Promotional Significance View that one’s intending an end does not provide any reason to intend or

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

108  Instrumental Rationality We would get similarly incoherent advice in the examples we’re currently considering if we allowed for intentions to provide or intensify reasons to take means. For example, suppose that, in the case in which I intend to get the can of tomatoes on the left, the intensification makes it the case that I ought to walk left. How can it also be permissible for me to change my mind and intend to grab the can of tomatoes on the right? It seems again that our normative theory is issuing incoherent advice.39 In summary, Normative Disjunctivism faces a significant difficulty when it comes to cases of mere permissibility—that is, cases in which one is incoherent with respect to ends and means which are permitted but not required. We could save Normative Disjunctivism by adopting a version of the ­intentions-provide-reasons view. But, as we saw in Chapter  2, such views license un­accept­able bootstrapping. We’re better off giving up on Normative Disjunctivism.

4.2.5  Normative Conjunctive Disjunctivism There might be a version of Normative Disjunctivism that can avoid the problem of mere permissibility. Niko Kolodny has suggested that instead of having a disjunctivist account which holds that whenever you’re means–ends incoherent there’s some specific attitude you ought to have (or not have), we can have an account which holds that whenever you’re means–ends incoherent, either you have a belief you ought not have or there’s some other pattern of intentions you ought to have but don’t have.40 Note that the following four patterns of intentions are possible: take some sufficient means to that end. But we could imagine a tie-breaker case, or a case of incommensurability, where the means, the intending of which the agent believes to be necessary for the end, is a sufficient means. For instance, we could suppose that I intend to grab the can of tomatoes on the left, and believe that to do so I must intend to move my arm forward and close my grip around it. But moving my arm forward and closing my grip suffices for grabbing the can on the left. Since this is a case of “mere permissibility,” prior to my forming any intention, it’s not the case that I ought to intend to move my hand forward and close my grip around the can on the left. And the Normative Disjunctivist needs it to be the case that my intention makes it the case that I ought to intend to move my hand forward and close my grip around the can on the left. But the Promotional Significance View is no help here, since, as we noted, it doesn’t generate reasons to intend or take sufficient means. (See also Kiesewetter 2018, pp. 272–4, for further discussion of these and related difficulties.) 39  As we noted in Chapter 2, we could put the objection in terms of the transmission of oughts to necessary means. Suppose I ought to walk left. Plausibly, my persisting in my intention is a necessary means to my walking left. (I won’t walk left if I change my mind.) By a principle governing the transmission of oughts to necessary means, it follows that I ought to persist in my intentions. So, it’s not also permissible for me to change my mind. 40  For the justification for considering patterns of intentions, as opposed to individual intentions, see Kolodny 2008a, pp. 369–70.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  109 A B C D

Intend the end Intend the end Not intend the end Not intend the end

&   &   &   &  

Intend the means Not intend the means Intend the means Not intend the means

According to this view, in cases of means–ends incoherence, either you have some belief you ought not have, or there is some other pattern of intentions besides B that you ought to have—that is, either (i) you ought to have A, (ii) you ought to have C, (iii) you ought to have D, (iv) you ought to have either A or C, (v) you ought to have either A or D, (vi) you ought to have either C or D, or (vii) you ought to have either A, C, or D.41 Since the patterns of intentions here all involve conjunctions, we could call this view Normative Conjunctive Disjunctivism.42 One immediate question is whether we should accept that ought distributes over conjunction—that is, whether it follows from O(P & Q) that O(P). If it does, then it would follow from “you ought to have A [intend the end and intend the means]” that “you ought to intend the means.” Similarly, it would follow from “you ought to have C [not intend the end and intend the means]” that “you ought not intend the end,” and from “you ought to have D [not intend the end and not intend the means] that “you ought not intend the end.”43 And such results are incompatible with cases of mere permissibility, which are defined as cases in which it’s not the case that you ought not intend the end, since it’s permissible, and not the case that you ought to intend the means, since that’s merely permissible. However, perhaps there are reasons to deny that ought distributes over conjunction.44 In Jackson and Pargetter’s (1986, pp. 235, 247) famous example of Professor Procrastinate, it seems right to say that 1)  he ought to accept the invitation and write the review,

41 This mirrors the non-cognitivist error theoretic account of intention consistency given in Kolodny 2008a. See especially p. 382, including fn. 50. This is also the “Ought Pattern 2” in Kolodny and Brunero 2013, §3.1. The view here comes close to Jonathan Way’s “subjective reasons” account of means–ends coherence, according to which if one’s instrumental belief is true, one ought not have pattern B. We’ll consider this account in Chapter 5. 42 Jonathan Way (2018, fn. 9) refers to Normative Disjunctivism as “full disjunctivism,” and Normative Conjunctive Disjunctivism as “partial disjunctivism.” 43  It would also follow from (iv) “you ought to have either A or C [(intend the end & intend the means) or (not intend the end & intend the means)] that you ought to intend the means, and from (vi) ”you ought to have either C or D [(not intend the end & intend the means) or (not intend the end & not intend the means) that you ought not intend the end. But (v) and (vii) are in the clear. 44  Kolodny rejects the distribution of oughts over conjunction. See fn. 21 of 2008a.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

110  Instrumental Rationality but also, since he won’t write it, that 2)  he ought not to accept the invitation. If ought distributes over conjunction, then it follows from (1) above that he ought to accept the invitation, and so it comes out both that he ought to accept the invitation and that he ought not to accept the invitation. Yet that conjunction seems false. We could avoid this by denying that ought distributes over conjunction. However, even Jackson and Pargetter accept a weaker principle: (P*)  If S ought to do A and B, and S does A and B, then S ought to do A. (It’s assumed that were Procrastinate to accept the invitation he would not write the review, and so that case is no counterexample to P* since the “S does A and B” clause won’t be met.45) But even on this weaker principle, we’d be forced to conclude that, when you both intend the end and intend the means, it follows from “you ought to have A [intend the end and intend the means]” that “you ought to intend the means,” the truth of which is incompatible with the case being one of mere permissibility. But the Normative Conjunctive Disjunctivist could sidestep these issues by saying that in cases of means–ends incoherence towards merely permissible ends, it could be that (vii) you ought to have either A, C, or D. And it doesn’t follow from this that you ought not intend the end, nor that you ought to intend the means. That seems sufficient for getting around the problem of mere permissibility. Understanding Normative Conjunctive Disjunctivism in this way brings it closer to some other approaches to the normativity of means–ends coherence that we’ll consider in the next chapter, such as the straightforward conjunction of Means-End Coherence and Strong Normativity, and Jonathan Way’s ­version of the “subjective ought” approach, neither of which have any trouble with cases of mere permissibility. But, as we’ll see, some of the concerns about those other views would also apply to Normative Conjunctive Disjunctivism. (In particular, we’ll see that it’s difficult to explain why it’s not the case that one ought to have the combination of intending an end and not intending the means in every case in which one’s instrumental belief is true.) For now, we

45  Jackson and Pargetter 1986, p. 248.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  111 can conclude that Normative Conjunctive Disjunctivism might avoid the problem of mere permissibility. However, it remains subject to the three objections to Normative Disjunctivism outlined above. First, consider Setiya’s case in which the ­cig­ar­ette smoker cannot alter his intention to smoke. Since he is unable to not intend to smoke, he’s also unable to have these combinations: [not intending to smoke and intending to buy a pack] and [not intending to smoke and not intending to buy a pack]. And so it’s not the case that he ought to have either of these combinations. And among the combinations he can have—[intending to smoke and intending to buy a pack] and [intending to smoke and not intending to buy a pack]—it seems that the latter is preferable. With the latter combination, he’ll end up not getting the pack of cigarettes he intends to get, but with the former combination, he’ll increase the probability of dying a horrible early death from cancer. That’s far worse. So, it doesn’t seem to be true that he ought to have some combination other than [intending to smoke and not intending to buy a pack]. Second, the problem of advantageous incoherence remains in place. Professor Henri Liable has most reason to have the instrumentally incoherent combination of attitudes. So, it’s not the case that he ought to have a combination other than B.  Third, we would still have the problem that the view prevents us from idealizing when it comes to practical reason while not extending that idealization beyond its rationale to the the­or­et­ic­al domain. We are prohibited from saying, for instance, that Lonnie ought to intend to settle his stomach and not intend to drink milk, while also saying that it’s permissible for him to believe that milk is the only effective means to settling his stomach. So, Normative Conjunctive Disjunctivism wouldn’t be a tremendous improvement over Normative Disjunctivism.

4.3  Does It Support the Myth Theory? If Normative Disjunctivism is true, then for every violation of Means–Ends Coherence, one has some particular attitude one ought not to have, or lacks some particular attitude one ought to have. This leads to the following conjecture: the idea that there is a rational requirement of Means–Ends Coherence is a myth. In fact, there are just these requirements to have, or not have, specific attitudes. Compliance with these requirements guarantees that one will be coherent, and so incoherence is a sign that something else has gone wrong—specifically, that one has some particular attitude one ought not to have or lacks some particular attitude one ought to have. And that explains

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

112  Instrumental Rationality why we were drawn in by the myth. But, really, there is no requirement of Means–Ends Coherence. I’ll argue that Normative Disjunctivism doesn’t lend support to the view that Means–Ends Coherence is a myth. This point can be brought out by considering some variations on our case of Bob, who intends to kill his colleagues, believes that to do so he must intend to poison them, but doesn’t intend to poison them. I’ll develop this point in three, related ways. First, it’s intuitive that there are two ways in which Bob goes wrong. First, he has an intention to kill his colleagues, which he ought not have. Second, he fails to intend those means believed to be necessary for killing them. As Jonathan Way puts it, there’s a “further problem” besides his having an intention he ought not to have.46 However, the worry is that, so far as Normative Disjunctivism goes, there is only one way in which Bob goes wrong: he has an intention he ought not to have. We need some requirement in place to predict the further way in which Bob goes wrong. But if Means–Ends Coherence is a piece of mythology, we don’t have the resources to do this. Second, let’s imagine that Bob, while initially means–ends incoherent, comes to form the intention to poison his colleagues. Suppose at the start he’s unaware of his incoherence, but he then becomes aware of it and reflects on possible ways of escaping it, ultimately deciding to poison his colleagues. Intuitively, he has come to satisfy some requirement by resolving his ­incoherence.47 If Means–Ends Coherence is a genuine requirement, we can explain this intuition: Bob satisfies some requirement because he transitions from violating to satisfying Means–Ends Coherence. But if Means–Ends Coherence is a myth, we are unable to explain this. So far as Normative Disjunctivism goes, there’s no requirement that Bob satisfies.48 Had Bob 46  See Way 2012a, p. 489, and Way, 2018, §4.2. One might find this point question-begging, since a proponent of the myth theory could simply deny that there’s any further problem here. See ChengGuajardo 2014, p. 493. However, I think it’s a cost to be forced to say that there’s only one way Bob goes wrong. Other things equal, we should have a view that allows for Bob to go wrong in two ways. 47  See Kolodny 2007b, p. 237 and Kolodny and Brunero 2013, §3.1. 48  One might argue: if Bob believed that he ought to poison his colleagues, then when he came to intend to do so, he thereby satisfied the enkratic requirement. And so he satisfied some requirement. There are two problems with this line of reasoning. First, there is no reason to assume Bob had this belief. Perhaps he hadn’t thought about whether this is something he ought to do. Or perhaps he was akratic: he believed he ought not poison them, but came to intend to do so anyway. (See Kolodny 2008a, pp. 387–90 for further discussion of this concern.) Second, the enkratic requirement is a requirement of rationality, not a requirement of reason. But why wouldn’t our myth theorist concerning Means– Ends Coherence provide a parallel account for the enkratic requirement as well? On the myth-the­or­ et­ic­al Normative Disjunctivist account of the enkratic requirement, someone who believes she ought to X and does not intend to X either ought not to believe she ought to X or ought to intend to X. And, in Bob’s case, he ought not believe that he ought to poison his colleagues, since that belief is false. But, intuitively, Bob nonetheless satisfies some requirement in resolving his incoherence by intending to poison his colleagues. We still haven’t explained how he satisfies some requirement in proceeding as he does.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  113 resolved his incoherence by abandoning his end of killing his colleagues, then the Normative Disjunctivist could say that Bob satisfies some requirement: the requirement not to intend to kill his colleagues. But the Normative Disjunctivist cannot explain how Bob satisfies some requirement in proceeding as he does.49 Third, the myth theorist is unable to account for Kant’s insight, discussed in Chapter  1, that both the physician and the poisoner are subject to similar “imperatives of skill” regardless of whether the ends they adopt are ones they ought to adopt. As Kant observes, “whether the end is rational and good is not at all the question here.”50 Despite the physician having a good end and the poisoner having a bad end, both would go wrong in the same way if they failed to intend the means they know to be necessary for their ends. Now, if Means–Ends Coherence is a genuine requirement of rationality, we can explain how both the physician and the poisoner go wrong in the same way: both of them violate Means–Ends Coherence. But if it’s not a genuine requirement of rationality, as the myth theorist alleges, then it’s hard to see how this can be explained. So far as Normative Disjunctivism goes, the physician and the poisoner go wrong in two very different ways: the physician fails to intend means he should intend and the poisoner has an end he shouldn’t have. In summary, even if Normative Disjunctivism is true, it doesn’t lend support to the view that Means–Ends Coherence is a piece of mythology. As we’ve seen, Normative Disjunctivism is unable to explain (i) how Bob encounters a further problem in virtue of being incoherent, (ii) how Bob would satisfy some requirement when he escapes incoherence by coming to intend the means, and (iii) how the way in which Bob goes wrong in being incoherent is the same way one could go wrong with respect to good ends. Means–Ends Coherence, in contrast, is able to explain all three. First, Bob encounters a further problem because he violates both the prohibition against intending to 49  One might argue that in the same way that it’s alleged to follow—by “deontic inheritance” within Standard Deontic Logic, according which if P→Q, then O(P)→O(Q)—from your being required to mail the letter that you are required to mail it or burn it (Ross  1941), it follows from Bob’s being required not to intend to kill his colleagues that he is required not to intend to kill his colleagues or intend to poison them. And Bob satisfies this requirement when he intends to poison them. This accounts for the thought that Bob satisfies some requirement when he intends to poison them. (See Lord 2014a, pp. 171–2.) But, besides involving a highly controversial inference, this makes requirements too cheap to come by. It follows by the same reasoning that Bob is required not to intend to kill his colleagues or burn their letters, and that he is required not to intend to kill his colleagues or dance wildly, and even that he is required not to intend to kill his colleagues or not intend to poison them, and so forth. By doing anything, Bob would satisfy some requirement. But what we want to account for is the intuitive thought that by intending the means, Bob would achieve some rational success that wouldn’t be achieved by doing anything whatsoever, including not intending the means. (For further discussion, see Kolodny 2007b, p. 239, fn. 18, and Kolodny and Brunero 2013, §3.1) 50  Kant 1785/1997, p. 26. [4:415].

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

114  Instrumental Rationality kill his colleagues and Means–Ends Coherence. Second, Bob would satisfy some requirement by coming to intend to poison his colleagues since he would satisfy Means–Ends Coherence. Third, Means–Ends Coherence applies regardless of the moral and rational status of the agent’s ends, and hence can explain why the incoherent poisoner and incoherent physician go wrong in the same way. Benjamin Kiesewetter (2018) has argued that the myth theorist has resources to meet some of these challenges.51 He has argued, following a suggestion from Kolodny, that one alternative explanation of the intuition that Bob comes to satisfy some requirement when he comes to intend the means is that Bob’s incoherence provides him with “second-order” evidence that there is some specific attitude he ought not have but does or doesn’t have but ought to. This generates a requirement to review and reassess the reasons he takes himself to have. According to Kiesewetter, someone “who is lead astray by her own reassessment, and therefore escapes incoherence in the wrong direction, might nevertheless count as satisfying a second-order requirement to review and reassess a particular group of attitudes.”52 This might explain the in­tu­ ition that Bob satisfies some requirement when he intends the means. Kiesewetter also argues that this requirement to review and reassesses can help with the “further problem” problem: in addition to having the problem of intending to do something he ought not do, Bob has the further problem of needing to reconsider his reasons (or, if he has already done so and held onto his intention, he has the problem of having done so wrongly).53 In my view, it is doubtful that such a requirement to review and reassess one’s reasons can explain the intuition that Bob satisfies a requirement when he comes to intend the means. Bob’s reassessment of his reasons may occur prior to any revisions in attitudes Bob makes. He may, for instance, reflect on his reasons, and then, at a later time, come, or not come, to intend the means. (To use Harman’s terminology, we can distinguish the phase of reflection from the phase of revision in Bob’s reasoning, and suppose that Bob has already undergone all the relevant reflection and all that remains is for him to actually revise his attitudes.54) If we take care to specify that Bob has already undergone a re-assessment of his reasons, we would still have the intuition that by continuing on to then intend the means (in the phase of revision), Bob would 51  Kiesewetter 2018, pp. 242–6. The strategies outlined there are general ones for defending myththeoretical approaches to rational requirements against such objections. I’ll here apply them to means–ends coherence in particular. (See pp. 292–3 for Kiesewetter’s own application of some of these strategies to instrumental rationality in particular.) 52  Kiesewetter 2018, p. 243. 53  Kiesewetter 2018, p. 246. 54  See Harman 1986, p. 2.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  115 satisfy some requirement. So, in that case, the intuition that Bob would satisfy some requirement by intending the means wouldn’t be explained by pointing to a requirement to reassess. He’s already done that.55 Additionally, even if we set up the example so that Bob has already conducted the relevant reassessment, and, unfortunately, has remained firm in his intention to kill his colleagues—and so he’s got two distinct failures (his original failure in intending to kill his colleagues, and a second failure to correct this original failure)—it still seems as though there would be yet another way for Bob to go wrong, not in the phase of reflection, but in the phase of revision, when he doesn’t go on to intend the means. Kiesewetter offers a second possible explanation of the satisfaction in­tu­ ition: Bob satisfies some requirement because he engages in correct reasoning when he proceeds to base his intention for the means on his intention for the end and his instrumental belief.56 But this doesn’t seem to explain the satisfaction intuition. The intuition we have is that Bob satisfies some requirement when he intends the means. But correct reasoning, as John Broome has persuasively argued, is a matter of permissibly basing some attitudes on others, and need not involve the satisfaction of any requirement.57 And basing permissions do not seem well-suited to explaining the intuition that there is a requirement Bob satisfies by intending the means. Moreover, even if we allow that correct reasoning is best understood in terms of basing requirements, that won’t help us with the “further problem” problem.58 When we list out basing requirements and permissions, we specify the required and permissible bases for an attitude. (For instance, we might say that it’s permissible, and perhaps even required, that Bob’s intention to poison be based on his intention to kill and his instrumental belief, and we might say that it’s impermissible for his intention to poison to be based on his belief that he is tall.) But those basing requirements and 55  A similar problem would arise for the suggestion that the means–ends incoherent agent who revises in the wrong direction nonetheless satisfies a requirement to make a decision about whether or not to intend the means or give up the end. (See Kiesewetter 2018, p. 285.) On this suggestion, we could say that when Aaron, who we’ll suppose should intend to study, instead gives up his intention to pass, he nonetheless satisfies a requirement to decide between intending the means and giving up the end. But note there’s a gap here as well. Aaron could decide to give up his intention to pass, without actually doing so. But we still have the intuition that Aaron would go on to satisfy some requirement by actually revising his attitudes—by giving up his intention to pass. 56  Kiesewetter 2018, pp. 243–4. 57  As Broome puts it, “correctness for reasoning is a matter of which ways it is rationally permissible to reason” (2013, p. 247). See also pp. 256–9. 58  In discussing a version of this problem (which he labels the “distinctive violation problem”), Kiesewetter (2018, p. 246) cautiously claims that some, but perhaps not all, cases of incoherence will involve incorrect reasoning. So, he might not disagree with the point that follows in the text.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

116  Instrumental Rationality permissions would not be requirements to go on to form or revise attitudes in the first place. They tell us the permissible and impermissible bases for Bob’s intention to poison, were he to have that intention. But that’s different from a requirement to form or revise attitudes to begin with. And we could very well suppose Bob goes wrong by not starting to reason his way to a revision in his attitudes.59 Kiesewetter suggests a third possible explanation of the satisfaction in­tu­ ition: Bob’s intending the means manifests a rational disposition that’s valuable in many other contexts.60 For instance, in circumstances in which Bob intends to do something he ought to do, the possession of such a disposition would likely ensure that he ends up doing what he ought to do. And so such a ­dis­pos­ition would be valuable. On this suggestion, when we think Bob has satisfied some requirement by intending the means, we are instead praising Bob for having, and manifesting, this valuable disposition. However, I’m not optimistic that such an explanation will work. Note that I’ve told you very little about Bob. I’ve told you nothing about his character, and nothing about his rational dispositions. It’s perfectly compatible with the description given so far that Bob has no such rational disposition. In that case, his intending to poison his colleagues would therefore manifest no such dis­ pos­ition. But it nonetheless seems as though doing so would satisfy some requirement. So, it’s not clear how an appeal to such rational dispositions, or other features of Bob’s character, would play the relevant explanatory role. Kiesewetter also appeals to rational dispositions to deal with the “further problem” problem. The idea here is that when Bob fails to intend the means, that is evidence that “he lacks this disposition or that his disposition is poorly trained.”61 Kiesewetter then claims that our intuitions that Bob goes wrong in some further way—that is, in addition to his not intending as he should— “can be accounted for in evaluative rather than deontic terms—they can be understood as character evaluations.”62 In short, we’re inclined to think that Bob goes wrong in some way because we evaluate his character negatively. However, this proposal also has its limitations. It’s possible for people to act out of character. It’s possible for me do something generous without being a generous person, to do something cowardly without being a cowardly ­person, and so forth. In these cases, assuming we already know a bit about 59  It’s also possible that Bob may transition to a change in attitudes through some automatic process that doesn’t count as reasoning (or at least “active” reasoning) but would nonetheless involve a permissible basing. See Broome 2013, Chs. 13 and 14, for further discussion. 60  Kiesewetter 2018, p. 244. 61  Kiesewetter 2018, p. 245. 62  Kiesewetter 2018, p. 246.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity I  117 the person’s character, we won’t change our evaluation of the person’s character based on a single generous act, or a single cowardly act, and so forth. So, let’s set up our example of Bob so that we know a bit about his character. Suppose his instrumental irrationality is highly uncharacteristic of him. In that case, we won’t arrive at any negative character evaluations of him. But it still seems that there is some way in which he goes wrong in not intending the means. So, the intuition that he goes wrong in a further way isn’t explained by negative character evaluations. In summary, I remained unconvinced by Kiesewetter’s proposals. Means–Ends Coherence, however, offers a straightforward and compelling ex­plan­ation of why Bob faces a further problem in not intending the means, and of why Bob would satisfy a requirement in coming to intend to poison his colleagues, and of why, when he fails to intend the means, Bob goes wrong in the same way as Kant’s physician would, were he to fail to intend to procure the necessary means to healing his patient.

4.4 Conclusion In this chapter, I have explored the first of three attempts to appeal to normativity to explain, or explain away, means–ends coherence. I’ve presented four arguments against Normative Disjunctivism, a view which holds that whenever one is means–ends incoherent, there is some specific attitude one should have but doesn’t, or some specific attitude one does have but shouldn’t. (We’ll return to some of the same examples employed in those arguments in Chapter  5, since they will pose related difficulties for the combination of Strong Normativity and Means–Ends Coherence.) Despite our rejection of Normative Disjunctivism, the view will provide a useful model for a related view that I’ll defend in Chapter 7, Non-normative Disjunctivism, which has a structure similar to Normative Disjunctivism, but can easily sidestep all four of the arguments I’ve given here against it. I have also argued in this chapter that even if Normative Disjunctivism were true, it wouldn’t lend support to the myth theory.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

5 Normativity II In the previous chapter, we considered the first of three ways in which philo­ ­ sophers have appealed to normativity to explain, or explain away, means–ends coherence. I argued there that Normative Disjunctivism isn’t true, and wouldn’t support a myth theory even if it were true. In this chapter, we’ll consider the two additional views I mentioned at the start of Chapter 4: first, the view which straightforwardly combines Strong Normativity and Means–Ends Coherence, and second, the view which holds that we can explain means–ends coherence by an appeal to subjective reasons and the subjective ought. I’m going to argue that we should be skeptical of these two views, and look elsewhere for an explanation of means–ends coherence. In Section 5.1, I’ll begin with a discussion of a central source of skepticism about both Strong and Weak Normativity: the difficulty in specifying the reason(s) to be means–ends coherent. In response to such skepticism, some have argued, drawing inspiration from Prichard’s critique of “moral philoso­ phy,” that rationality provides its own reasons, and so the demand to dis­ cover reasons outside of these “reasons of rationality” is misplaced. I’ll argue (Section 5.1.1) that there are significant difficulties with this analogy to Prichard’s critique, and that the demand to produce reasons to be means–ends coherent is a legitimate demand, and (Section 5.1.2) is not easily met. In Section 5.2, I’ll turn to arguing directly against Strong Normativity, pre­ senting three separate arguments against it, building on the work of Kieran Setiya, Joseph Raz, and others. These arguments have all received some crit­ ic­al discussion in the literature. I’ll show how each of these arguments can withstand, or be modified to accommodate, the various objections that have been put to them. And, in some cases, the arguments can be developed on weaker assumptions than commonly thought. The result is, I think, a very strong case against Strong Normativity. In Section  5.3, I’ll consider the strategy, developed in different ways by Mark Schroeder and Jonathan Way, of accounting for means–ends coherence by an appeal to subjective reasons and the subjective ought. I’ll argue that Schroeder’s proposal requires that we accept a highly controversial thesis Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  119 about the nature of intentions, while Way’s approach is unable to avoid many of the objections facing the other approaches we’ve considered.

5.1  Reasons to be Rational Whereas Strong Normativity holds that you always ought to be means–ends coherent, Weak Normativity holds merely that you always have a reason to be  means–ends coherent. In other words, according to Weak Normativity, you always have a reason not to have the combination of intending an end, believing intending some means is necessary for the end, and not intending the means. While Weak Normativity might be true, it’s not a promising avenue for explaining the rational requirement of means–ends coherence. That’s because rational requirements, like means–ends coherence, are, as John Broome puts it, “strict”—they are requirements; they demand something of us—but this “strictness” or “stringency” would be unexplained by an appeal to pro tanto reasons.1 After all, you could have a reason to do something and vio­ late no requirements whatsoever in not doing it, as when you have greater rea­ son to do something else instead. So, saying that you have a reason to not have a certain combination of attitudes leaves it unexplained why you are required not to have this combination. However, “oughts” are strict. So, if Strong Normativity is true, then the combination of intending an end, believing intending some means is necessary for the end, but not intending the means, is a combination you ought not have. And this would explain the “strictness” of means–ends coherence. Given that we’re here interested in how normativity might explain means–ends coherence, we’ll focus primarily on Strong Normativity. However, the central reason for skepticism about Strong Normativity threatens Weak Normativity as well. Both views are committed to there being a reason to be means–ends coherent. And if there is a reason to be means–ends ends coherent, we should be able to say what that reason is. But, as Niko Kolodny argues in “Why be Rational?,” it’s difficult to do so. (Kolodny is concerned with reasons to be rational in general, but we’re here focusing on one rational requirement.) Such reasons could be either instrumental or non-instrumental. But it’s hard to see what the instrumental reasons could be, since it’s not clear what good would be promoted by one’s becoming 1 Broome  2000, p. 81. See also Schroeder (2009) on accounting for “stringency” of rational requirements.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

120  Instrumental Rationality means–ends coherent. Suppose Bob comes to intend to poison his colleagues, or suppose he gives up his instrumental belief. These are both ways of becoming coherent that don’t seem to promote anything of value. And it’s also hard to see what the non-instrumental reasons could be. As Kolodny puts it, “We have intrinsic reasons to care about persons, relationships, ­justice, art, science, the natural environment, and so on, for their own sake. All  of that is familiar enough. But is being subjectively rational another substantive value that we actually weigh against these others?”2 One’s having coherent attitudes doesn’t seem to be an intrinsic good to be included along­ side the others on that list. But if we can identify neither an instrumental nor a non-instrumental reason to be means–ends coherent, we should be skeptical that there is any such reason, and hence skeptical of both Strong and Weak Normativity. However, one might suspect that Kolodny’s request for reasons to be rational rests on a mistake—in particular, the same kind of mistake that Harold Prichard accused moral philosophers of making when they tried to provide non-moral reasons to be moral. In Prichard’s view, much of moral philosophy—indeed, the “whole subject” in his view—was devoted to answering the “Why be moral?” challenge—a challenge which traces back at least as far as Glaucon’s skepticism at the start of Book II of Plato’s Republic.3 In Prichard’s view, moral philosophy attempts to provide an answer to such skeptics: Yet, like Glaucon, feeling that somehow he ought after all act in these [morally required] ways, he [the skeptic] asks for a proof that this feeling is justified. In other words, he asks ‘Why should I do these things?’, and his and other people’s moral philosophizing is an attempt to supply the answer, i.e. to supply by a process of reflection a proof of the truth of  what he and they have prior to reflection believed immediately and without proof.4

Prichard goes on to argue that none of the existing answers to the skeptic’s challenge, such as answers which try to show how morality is actually in one’s self-interest, are successful. But he thinks the project of trying to answer this question (“moral philosophy”) rests on a mistake, specifically, 2  Kolodny 2005, p. 545. As he puts it elsewhere, it’s “outlandish” to put the kind of “psychic tidi­ ness” required by coherence requirements, like means–ends coherence, “alongside such final ends as pleasure, friendship, and knowledge” (2008a, p. 386). 3  Prichard 1949, p. 1. 4  Prichard 1949, pp. 1–2.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  121 “the mistake of supposing the possibility of proving what can only be apprehended directly by an act of moral thinking.”5 We could put Prichard’s central suggestion as follows: rather than taking on the project of searching for non-moral reasons to engage in some piece of moral behavior—say, paying one’s debts—moral philosophers should simply recognize that there are moral reasons to do so, reasons which are apprehended through an act of moral thinking. Some philosophers have thought that a move parallel to Prichard’s response to the “Why be moral?” challenge could be made in response to Kolodny’s “Why be rational?” challenge: rather than search for non-rational reasons to comply with requirements of rationality, we should recognize that there are reasons of rationality to do so. Both Nicholas Southwood and Nadeem Hussain have independently put forth this suggestion, noting the parallel to Prichard’s view.6 Here’s Southwood: For the normativity of rationality is a matter of reasons that are internal to rationality, not reasons that are external to it. It is a matter, if you like, of reasons of rationality, not independent reasons to obey rationality. Rather than giving us any reason to conclude that rationality isn’t normative, the failure to locate an independent justification for obeying rational requirements should do nothing whatsoever to undermine our belief in the normativity of rationality.7

If this is right, both the “Why be rational?” and the “Why be moral?” chal­ lenges are misguided in that they seek a justification in terms of reasons out­ side of their respective domains, when one is readily available within those domains. If this strategy—I’ll call it “the Prichard Analogy”—is successful for rational requirements in general, we’ll have an easy answer to Kolodny’s challenge. And we could then say there are reasons of rationality to be means–ends coherent. I’ll now argue (Section 5.1.1) that the Prichard Analogy isn’t successful, and that we need to take seriously the demand for reasons to be means–ends coherent. I’ll then consider (Section 5.1.2), skeptically, some attempts to meet this demand.

5  Prichard 1949, p. 16. 6  See Southwood  2008, esp. pp. 17–19, and Hussain, unpublished, esp. §5. John Broome (2013, p. 204), who takes rationality to be non-instrumentally normative, also notes this parallel to Prichard and morality. 7  Southwood 2008, p. 18.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

122  Instrumental Rationality

5.1.1  Evaluating the Prichard Analogy I won’t here raise any questions about whether Prichard’s response to the moral skeptic succeeds. I’ll assume that it does. (If you don’t think that it does, you probably won’t think that the Southwood/Hussain view succeeds either, in which case you’re on my side.) Rather, I’ll argue that the reasons to think Prichard’s response to the moral skeptic succeeds do not carry over into the rational domain, and so are inadequate as a response to the rational skeptic. In short, I’ll argue that the Prichard Analogy is not a good analogy. First, we should note that analogies along these lines would likely not prove convincing in other debates about the normativity of various requirements. Consider the debate over the normativity of legal requirements. Suppose after surveying and finding unsatisfactory the various proposed reasons to obey the law—including the best versions of the gratitude and social contract argu­ ments in the Crito, the best versions of the “fair play” arguments found in Hart and Rawls, and the best arguments of act- and rule-consequentialists, etc.—one then says, channeling Southwood, “The normativity of law is a mat­ ter of reasons that are internal to law, not reasons that are external to it. It is a matter of the reasons of law, not independent reasons to obey the law.” We’ve made a mistake, according to this view, in looking for prudential and moral reasons to obey the law; we can instead simply point out that the law provides its own reasons.8 Such a line of reasoning seems obviously problematic as a contribution to a debate over the normativity of law. (If not there, you might also think of similar reasoning done in defense of the normativity of etiquette, chivalry, Catholicism, and so forth.) If we don’t find such arguments persua­ sive in these others domains, why should we find them persuasive in the rational domain? Prichard’s allegation that moral philosophy has committed a “mistake” in seeking a justification from outside its domain might seem compelling, if it does, because the moral skeptic is seeking a justification for what she “prior to

8  It should be noted that there is a perfectly innocent sense in which we can speak of the “reasons of law,” “reasons of etiquette,” and so forth. Within systems of norms, there may be considerations that play a pro tanto, or contributory, favoring role. (Suppose someone legally ought to pay state tax on an honorarium. It’s compatible with this that there are legal reasons favoring his paying tax on it, and legal reasons favoring his not paying tax on it. Lawyers will sometimes point to such reasons. And judges will sometimes have to weigh competing legal reasons to determine what one is legally required to do.) But there still remains a question of normativity, in our sense of that term, to be raised about the system of norms. For instance, we can ask, “If I’m legally required to X, is it true that I ought, or have a reason, to X?” and in asking that question, we clearly don’t have in mind “reasons of law” in this innocent sense.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  123 reflection believed immediately or without proof.”9 (It’s worth remembering that the case of the moral skeptic is analogous, for Prichard, to the epistemo­ logical skeptic, who, seeing the possibility of his being mistaken, wonders whether he really knows that 7 × 4 = 28. In both cases, prior to skeptical doubt, the initial stage is one of “unquestioning confidence.”10) Moreover, Prichard provides an epistemological account that would, if correct, justify one’s maintaining that initial confidence, thus removing the need to search for non-moral reasons to be moral. In Prichard’s view, what we ought to do is “self-evident” and would be apparent to any sufficiently “thoughtful” and “developed” person, provided that all the relevant consequences of the rele­ vant action, and all the relevant relations, are specified.11 Since these moral claims are self-evident, there’s no need to seek any non-moral justification for them; they need no evidence other than that which they themselves provide. Whatever you think of Prichard’s moral epistemology, it’s doubtful that plausible analogues of these claims would hold for rational requirements, like means–ends coherence. While I am highly confident that I ought not cause you unnecessary suffering for fun, I am not, nor have I ever been, similarly confident that I ought not have a combination of attitudes fitting the follow­ ing schema: intending to X, believing I’ll X only if I intend to Y, not intending to Y. Nor is it correct to dismiss those who believe means–ends coherence is a myth, such as Joseph Raz and Niko Kolodny, as insufficiently thoughtful or insufficiently rationally developed. Like the question of the normativity of law, the question of rationality’s normativity is a contested one, and won’t be settled by such a quick and easy appeal to self-evidence. Thus, insofar as the allure of Prichard’s argument relies upon his moral epistemology, particularly his appeal to self-evidence in the moral (and mathematical) cases, that argu­ ment won’t extend to support to the normativity of rational requirements. So far I’ve argued that the basic argumentative strategy we’re considering here would be obviously problematic when employed in debates about the normativity of, say, law or etiquette. And so we need some special reason to think that it would work within the domain of rationality. Additionally, I’ve argued that Prichard’s moral epistemology—particularly his claims about the moral skeptic’s initial high degree of confidence in the self-evident truths of morality—won’t support analogous views when it comes to the normativity of rationality. But one might also have a more general concern: the reasons a philosopher would likely adduce for the existence of reasons of morality— that they figure in our best first-order normative theory, survive reflective 9  Prichard 1949, pp. 1–2.

10  Prichard 1949, p. 14.

11  Prichard 1949, pp. 8–9.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

124  Instrumental Rationality equilibrium, and so forth—cannot be assumed also to support the existence of reasons of rationality. Perhaps they do. But that’s a substantive claim that needs to be argued for, and the Prichard Analogy, by itself, doesn’t constitute any such argument.

5.1.2  The Costs of Incoherence Since the request for reasons to be rational—and, in particular, to be ­means–ends coherent—is a legitimate request, we now need to consider whether such reasons can be identified. A natural strategy would be to appeal to the costs, broadly understood, of being instrumentally incoherent. There are two ways one might do this. First, in support of Strong (or Weak) Normativity, one might argue that there are costs to intending an end, believing intending the means is necessary, but not intending the means. Second, in support of Normative Conjunctive Disjunctivism (or the related version of the subjective ought approach we’ll consider in Section  5.3), one might argue that when one’s belief that intending the means is necessary is true, there are costs to having the combination of intending the end but not intending the means. Let’s start with the latter approach, looking at the costs of incoherence assum­ ing that your instrumental belief is true. It seems that when you intend some end that you will not in fact achieve (since you don’t intend the means that are in fact necessary), you’ll be wasting resources. As Niko Kolodny observes, intending some end increases the likelihood that you will take costly means to that end, which will all be for nothing if you don’t intend the necessary means.12 And, moreover, having that intention could interfere with the successful formation and execution of other intentions. Suppose you’re instrumentally incoherent with respect to your intention to get the can of tomatoes on the left. Plausibly, it will be more difficult for you to form and execute an intention to get the one on the right. This decreases the likelihood of your getting any tomatoes at all, or at least makes the process of getting them clumsier than it need be. And, as Jonathan Way argues, the mere intending of an end in such cases— regardless of whether it’s accompanied by the further spending of resources, or whether it frustrates your other goals—is a waste of mental effort. For one thing, intending to do something involves your monitoring your progress 12  Kolodny 2008a, p. 369. Kiesewetter 2018, pp. 282–3, also points to how means–ends incoherence risks wasting valuable resources.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  125 toward achieving the end, and, for another, it disposes you toward intending further means, both of which involve some mental effort.13 As Way argues, “intending is a way of tying up psychological resources—resources which are limited and for which there is always another use.”14 So, intending in such cases is itself a waste of mental resources. I agree with Kolodny and Way that intending very often involves such costs, but the question we need to consider is whether such costs are always present, and whether they always have weight sufficient to make it such that one ought not have the combination of intending the end and not intending the means. And with respect to these questions, I have my doubts. Suppose that, following in the footsteps of someone I admire, I now intend to spend my sixtieth birthday doing volunteer work in a foreign country. Since I’ve got a long way to go before my sixtieth birthday, there’s no immediate op­por­tun­ ity to waste resources on costly means, such as flights or accommodations. Nor is it clear how having this intention interferes with the formation and execution of other intentions. Unlike the example above, in which my intend­ ing to get the can of tomatoes on the left increases the difficulty of my getting any tomatoes at all, my intention regarding how to spend my sixtieth birthday exists in deliberative isolation in that it doesn’t present any psychological obstacle to my intending and doing other valuable things. Nor is it clear how there would be a wasteful expenditure of psychological resources. Perhaps my mental life isn’t so cluttered as to make my formation of this intention come at the expense of other valuable projects. Also, perhaps devoting further atten­ tion to my progress toward this end, even if I don’t follow through on it, wouldn’t be a waste of time—perhaps doing so is valuable as means to selfdiscovery, or to increasing my knowledge of the needs of distant strangers. You might think of it this way: suppose that I didn’t intend to spend my sixtieth birthday doing such volunteer work, but merely hoped or desired to do so. And suppose that I then devoted some mental energy to thinking about mat­ ters concerning the distant needy and what could be done to help them. You’d be unlikely to think that I’m wasting my time and mental energy on this. So, why should you think of these same activities as a waste of time and mental energy when they’re prompted by an intention to spend my sixtieth birthday doing volunteer work? In short, the problem is that it’s not clear that we should construe all such expenditures of attention and mental energy as a cost.

13  Way 2012a, p. 504.

14  Way 2012a, p. 504.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

126  Instrumental Rationality The overall worry here is analogous to the familiar concerns about ­consequentialist approaches to the normativity of law.15 Although we can think of many cases in which breaking the law yields bad results, it isn’t always the case. Running the stop sign in my empty office parking lot at 5:00 a.m., when it’s obvious no one else is around, risks no bad results whatsoever and is convenient. And keeping up with the speed of traffic, even when everyone else is going above the speed limit, is usually the safer bet. And so it’s hard to  see how considerations concerning the consequences of obedience and disobedience will yield the result that we always ought not disobey the law. In the same way, it’s hard to see how an appeal to the consequences of intending an end without intending the means will make it such that one always ought not have this combination. Indeed, my example above seems to be a case of “no-cost incoherence,” or at least a case in which none of the costs identified by Kolodny and Way are present. Another approach to understanding the costs, broadly understood, of incoherence is suggested by Michael Bratman, in defense of the normativity of Means–Ends Coherence.16 In Bratman’s view, there’s a reason of selfgovernance to comply with rational requirements of consistency and coher­ ence, and this accounts for the “non-instrumental normative significance” of Means–Ends Coherence.17 Whenever someone is means–ends incoherent— whenever they intend an end, believe (truly or falsely) that intending some means is necessary for it, but don’t intend the means—they fail to govern themselves. And Bratman, drawing on work by Harry Frankfurt, offers us an explanation of why this is so: in such cases, there is no place where the agent stands with respect to the end. As Bratman puts it: And now the point to note is that if I – a planning agent – intend E, but also intend what I believe to be incompatible with E, or do not intend what I  believe to be necessary means to E, then, barring special circumstances, there is no clear answer to the question of where I stand with respect to E. In intending E, I seem to include E within my practical standpoint; but in intending something I believe incompatible with E, or in failing to intend 15  See Schauer and Sinnott-Armstrong 1996, pp. 222–7. The examples that follow are taken from their discussion. 16  Bratman doesn’t defend Strong Normativity, since “it is too strong to say that there is always an overriding reason to avoid such incoherence.” (2009c, p. 420) We will see in Section  5.2.2, some ­reasons why Bratman thinks this is so. I’ve argued above that Weak Normativity isn’t enough to explain the “strictness” of Means–Ends Coherence, or, as Bratman puts it, the fact that Means–Ends Coherence is a “demand of practical rationality” (2009c, p. 421, my emphasis). 17  Bratman 2009d, p. 230.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  127 believed necessary means to E, I seem to exclude E. In this local area of my psychology, there is a structure of planning attitudes whose functioning in the guidance of my thought and action does not unequivocally either favor or reject E. So with respect to E, there is no clear place where I stand.18

Since being means–ends incoherent on any particular occasion will constitute a failure to have a place where one stands with respect to an end, which itself constitutes a failure of self-governance, it follows that we have a reason against being means–ends incoherent on any particular occasion (since, as noted above, we have a reason against our failing to govern ourselves). However, it’s not clear that Bratman is correct in thinking that all cases of means–ends incoherence involve someone’s lacking a clear place where he stands.19 On an ordinary understanding of this phrase, it seems that there are cases in which it is perfectly clear where a means–ends incoherent agent stands. Suppose that Bob, who fails to intend to poison his colleagues, strongly and unwaveringly believes that he ought to kill them, and is critical of himself for failing to intend to poison them. It seems natural to say that Bob’s stance with regarding to the end of killing his colleagues is this: he is for killing them. That’s where he stands. However, it’s worth noting that Bratman draws this particular phrase from Frankfurt’s work on identification, and it functions as a technical term in his argument. Bratman explains the idea as follows: “In the background is Harry Frankfurt’s insight that we need an account of what it is for an agent to iden­ tify with a certain thought or attitude—of what it is for a thought or attitude to speak for the agent, to be part of where the agent stands, to have agential authority.”20 (Attitudes with agential authority, Bratman explains, are those which are such that “when they guide, the agent governs.”21) Now the ques­ tion of how to understand agential authority—whether in terms of one’s 18  Bratman 2009d, p. 236. See also 2009c, p. 431: “If I intend E but I do not now intend known necessary means intending which now I know to be necessary, there is no clear answer to the ques­ tion, ‘Where do I stand?’ with respect to E. With respect to this end, there is as yet no relevant fact of the matter about where I stand.” 19  I’ll here grant for the sake of argument Bratman’s claim that we have a reason to govern our­ selves, as well as his claim that a failure to have a place where one stands with respect to an end consti­ tutes a failure of self-governance. I’ll challenge the premise that means–ends incoherence necessarily involves lacking a place where one stands with respect to an end. 20 2009c, p. 430. The phrase appears in Frankfurt  1988, p. 166, in a passage discussing Gary Watson’s (1975) criticism of Frankfurt’s hierarchical model. Showing that there’s a conflict between an agent’s higher-order and lower-order desires “is hardly adequate to determine – with respect to that conflict – where (if anywhere) the person himself stands.” 21  2009c, p. 430. The concept of agential authority is further explored in several of the essays in Bratman 2007.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

128  Instrumental Rationality judgments about the good, one’s values, one’s second-order desires, one’s ­self-governing policies, etc.—is a large question which I don’t need to weigh in on here. For on any of the leading theories, it seems perfectly possible for the relevant attitudes to line up in favor of one’s carrying out the end, and against one’s not intending the means believed to be necessary. In my vari­ation on the case of Bob, I assumed that his beliefs about what he ought to do, along with a disposition to be self-critical, line up in this way. But we could make the point on any leading theory of self-governance. We could set up the example so that, say, Bob’s values, or second-order desires, or whatever, are so aligned. Of course, I do not deny here that there’s a conflict among an agent’s ­attitudes (understood broadly to include the attitudes the agent has and lacks) in cases of means–ends incoherence. I do think there are such conflicts, and they are governed by a rational requirement, Means–Ends Coherence. What I’m denying, rather, is that such conflicts must always involve a person being conflicted such that there’s no place where he stands with respect to the end. In Bob’s case, there is a conflict among his attitudes, but we shouldn’t say that Bob is conflicted such that there’s no place where he stands with respect to killing his colleagues. Once we factor in his norma­ tive beliefs, dispositions to be self-critical, etc., it’s clear that he’s not con­ flicted at all: he wholeheartedly favors killing his colleagues. (There might be some cases where one doesn’t intend the means because one is conflicted, having second thoughts, etc. But there’s no reason to assume all cases are like this.) Frankfurt himself draws a related distinction, noting how, when there is a conflict among an agent’s desires, the agent’s identifying with one of them changes the nature of the conflict, not by removing the conflict among desires, but by eliminating the conflict within the person regarding where he stands: Suppose that a person with two conflicting desires identifies with one rather than the other. . . . Quite possibly, the conflict between the two desires will remain as virulent as before. What the person’s commitment to the one eliminates is not the conflict between it and the other. It eliminates the con­ flict within the person as to which of these desires he prefers to be his motive. The conflict between the desires is in this way transformed into a conflict between one of them and the person who has identified himself with its rival. That person is no longer uncertain which side he is on, in the conflict between the two desires, and the persistence of this conflict need not subvert

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  129 or diminish the wholeheartedness of his commitment to the desire with which he identifies.22

Likewise, I do not deny that there remains a conflict between Bob’s intending the end and his not intending the means he believes to be necessary for it. But I do deny that when Bob identifies with the end—however that’s understood— there’s still uncertainty about which side he is on. In that case, it seems there’s a clear place where he stands: he’s for killing his colleagues. Let’s return to the passage from Bratman quoted above. I think we’re now in a position to see exactly where it goes wrong. Bratman says something about means–ends incoherence that seems true: “In this local area of my psych­ology, there is a structure of planning attitudes whose functioning in the guidance of my thought and action does not unequivocally either favor or reject E.” But the crucial question for Bratman is why we should limit our attention to this local area of the agent’s psychology. Couldn’t an agent’s other attitudes make it clear whether he favors E, rejects E, or—adding a third option—is undecided with respect to E? After all, in the case of Bob, his nor­ mative belief and dispositions of self-criticism make it perfectly clear where he stands: he favors killing his colleagues. Without an answer to this crucial question, the conclusion in Bratman’s next sentence (“So with respect to E, there is no clear place where I stand.”) does not follow. I should clarify that the point against Bratman doesn’t hinge upon whether we’re always willing to say that a means–ends incoherent agent “lacks a place where he stands,” where this is understood as a ordinary English phrase. Rather, the point concerns the more technical use of this phrase taken from Frankfurt. (We could of course grant that the phrase provides a useful way of characterizing irrational incoherence—much like Dancy’s gloss that the ir­ration­al agent is “at odds with himself ”—but then Bratman’s claim that we always have a reason to have a place where we stand would amount to just another way of saying that we always have a reason to be rational, which is precisely the point that stands in need of support.) Bratman’s idea is that when one is means–ends incoherent, one “lacks a place where one stands” in the sense that one lacks attitudes with respect to E-ing that have “agential author­ ity” and “speak for the agent.” And he then argues we always have a reason to have such attitudes—a view I’ve conceded for the sake of argument.23 But I’ve 22  Frankfurt 1988, p. 172. 23  Kiesewetter’s (2018, p. 105) critique of Bratman challenges this view, but doesn’t take issue with the crucial claim from Bratman that I’m disputing here. Kiesewetter (2018, p. 106) also observes that

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

130  Instrumental Rationality argued that on any leading theory of agential authority, we could imagine Bob’s other attitudes line up in such a way that his means–ends incoherence doesn’t threaten the existence of agentially authoritative attitudes with respect to E-ing. So, Bratman’s proposal does not succeed. In summary, Bratman’s proposal encounters a difficulty similar to those encountered by other attempts to point to the costs of being means–ends incoherent: while it’s true that some cases of incoherence will involve the specified costs, it’s not clear that all would. In particular, it’s not clear that all cases of means–ends incoherence would make it such that there’s no place where the agent stands with respect to the relevant end. Let’s take stock of where we are. In the previous subsection, I argued that there’s a legitimate demand to specify reasons to be means–ends coherent, one that can’t be met simply by appealing to “reasons of ration­ ality” in the same way Prichard appeals to moral reasons to counter the moral skeptic. We’ve now seen that it’s not easy say what those reasons are. But remember that a defense of Strong Normativity requires not just that we be able to spe­cify such reasons, but that we take the reasons be decisive or conclusive. And this thought, I’ll now go on to argue, is problematic in at least three ways.

5.2  Against Strong Normativity I’ll present three independent arguments against Strong Normativity, one in each of the following subsections. Each of these arguments will give us suffi­ cient reason to reject Strong Normativity. (The first two will involve cases familiar from the critique of Normative Disjunctivism in Chapter  4.) I’ll assume throughout, for reasons mentioned earlier, that we’re working with the wide-scope requirement, Means–Ends Coherence.

Bratman’s proposal threatens to prove too much. Wouldn’t we also have reasons of self-governance to avoid lacking a place where we stand in contexts that involve no irrationality, like when we are reasonably withholding belief? I suppose that Bratman could accept this, but argue that the relevant evidential considerations provide reasons to withhold belief that outweigh the reasons of self-governance to make up one’s mind (while also observing that such countervailing reasons wouldn’t be present for ordinary cases of irrationality, like akrasia or means–ends incoherence). However, exploring this option would take us too far afield. And, as I’ve argued, we can reject Bratman’s argument on other, more fundamental, grounds: the crucial claim that means–ends incoherence always leaves you with­ out a place where you stand is false.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  131

5.2.1  Advantageous Incoherence Let’s start by returning to the observation made earlier (Chapter  4, Section 4.2.2), due to Thomas Schelling, that irrationality can sometimes be to one’s advantage. In our example of advantageous means–ends incoherence, Henri intends to get the course certified, believes that to do so he must talk to the relevant administrators, but doesn’t intend to do so. His intending to get it certified secures the good will of the Chair, who reliably detects such intentions, while his not intending to talk with administrators decreases the probability of future burdensome service assignments. In this case, like in Kavka’s toxin puzzle, there are benefits to having an intention that don’t derive from what would come about were the intention successfully executed. I argued that Henri’s case poses a problem for Normative Disjunctivism, since there is no specific attitude that Henri ought to have but doesn’t have, nor that he ought not to have but does have. But the case is equally well an objection to those who hold both Means–Ends Coherence and Strong Normativity, since those two views together predict that Henri ought not have the combination of atti­ tudes he has, and that prediction strikes us as false. Indeed, it seems he ought to have the incoherent combination of attitudes. We considered one possible response on behalf of the Normative Disjunctivist already: Henri’s state-given reasons to intend to get the course certified are “the wrong kind of reasons” to intend to get the course certified, and hence, according to wrong-kind-of-reasons skeptics, not really reasons at all. If we excluded such reasons, the Normative Disjunctivist could argue that Henri ought not intend to get the course certified. That same strategy could also help the defender of Strong Normativity, who could argue that with such reasons off the table, Henri ought not to have the combination of attitudes he has. However, I rejected this strategy, since skepticism about state-given ­reasons to intend is implausible: such skepticism would predict that I have no more reason to intend to go to Block Island than to intend to go to Provincetown, even though, the destinations being equally good, intending to go to Block Island would fill me with intense joy.24 So, Strong Normativity is still threatened by cases of advantageous incoherence.25 24  I won’t discuss here how state-given and object-given reasons are to be weighed against one another, although I assume that they sometimes can be. For a discussion of issues related to such weighing, with respect to evidential and pragmatic reasons for belief in particular, see Reisner 2008. 25  Even though we must give up Strong Normativity, there might be a consolation prize: we might be able to say that excluding state-given reasons one always ought to be Means–Ends Coherent. Or, as Jonathan Way (2010, p. 229) puts it, we could say that there’s always “conclusive object-given reason” to be means–ends coherent. However, this view would require that we see the reasons to satisfy

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

132  Instrumental Rationality But a defender of Strong Normativity could make another reply here. Jacob Ross (2012) aims to defend the strong normativity of rational requirements against objections involving cases of advantageous incoherence (what he calls “the pragmatic reasons problem”), without endorsing any general skepticism about the bearing of state-given reasons on what we ought to intend.26 If we can apply Ross’s strategy to Means–Ends Coherence in particular, we’d be able to say that Henri ought not have the combination of attitudes he has, without having to buy into any general skepticism about the existence or relevance of state-given reasons, which is good since such skepticism yielded the counter­ intuitive result in the Block Island/Provincetown case. However, although Ross applies his strategy to several requirements of rationality, including the rational prohibition against having inconsistent intentions, he notes that the application of his strategy to other requirements, including Means–Ends Coherence, is beyond the scope of his paper.27 But given how promising his strategy proves for those other rational requirements, it’s worth exploring the prospects for extending it to Means–Ends Coherence. In the remainder of this subsection, I’ll explain the two principles that form the heart of Ross’s response to “the pragmatic reasons problem,” and then show how he applies them in defense of the strong normativity of the rational prohibition on inconsistent intentions. But I’ll then argue that there would be significant difficulties in extending Ross’s strategy to Means–Ends Coherence. Ross begins by observing that one of the functions of intentions is the ­settling of deliberation: “once one has formed the intention to X in C, one no longer needs to deliberate about whether to X in C.”28 As he puts it, the inten­ tion to X in C could serve as a “surrogate for deliberation” in that one can rely on the intention when it comes time to X, rather than again weighing the Means–Ends Coherence as object-given reasons, and it’s not clear to me that this is so. The standard candidate reasons, discussed in Section 5.1.2, seem to be state-given: the costs of having an intention that won’t succeed, and the self-governance that could be achieved by having a place where one stands. (It was by intending to take the means, not by taking the means, that the means–ends incoherent agent would achieve the kind of self-governance Bratman has in mind. So, the relevant reasons here aren’t object-given.) Moreover, there seems to be a clear difference between prohibitions on combinations of attitudes (such as that on intending to drink and intending to drive) that are explained by the reasons against the objects of the intentions (the reasons against drinking and driving), and prohibitions on combinations of attitudes (such as that on intending to drink and intending not to drink) that aren’t explained by the reasons against the objects of the intentions. (Since drinking and not drinking is not even possible, there are no pros and cons of doing it.) I’m inclined to say that the former case involves object-given reasons against a combination of attitudes, while the latter involves state-given reasons against a combination of attitudes. (But for an alternative take on the distinction, see Way  2010, pp. 230–2.) So, I doubt that the reasons to have coherent attitudes would be object-given. 26  Ross 2012, pp. 140–1, 155–9. 27  Ross 2012, pp. 179–80. 28  Ross 2012, p. 166. I’ve changed his “ϕ”s to “X”s.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  133 reasons for and against X-ing in C.29 And he then argues that intentions could not play this role were it possible for one to have sufficient reason to intend to X in C without also having sufficient reason to X in C. For were this possible, once someone formed an intention to X in C, there would remain a further, unsettled question about whether to X in C.30 A generalization of this point leads him to adopt the Commitment Transmission Principle:  if M [a mental state] commits one to X-ing in C, then one has sufficient evidence-relative reason to have M only if one has sufficient evidence-relative reason to X in C.31 If this Principle is correct, then, since intentions commit one to X-ing in C, it follows that if you lack sufficient evidence-relative reason to X in C, then you lack sufficient evidence-relative reason to intend to X in C. Ross then goes on to defend an agglomeration principle for attitudes: Commitment Agglomeration Principle:  If a mental state M consists in having some set of attitudes, then M commits one to acting in the conjunction of the ways in which one is committed to acting by having the mental states in this set.32 Together with the Commitment Transmission Principle, this Principle allows Ross to conclude that if a combination of attitudes commits one to X-ing and Y-ing in C, then, if you lack sufficient evidence-relative reason to X and Y in C, then you lack sufficient evidence-relative reason to have that combination of attitudes.33 Let’s consider now how these principles would apply to a case in which an agent intends to X in C and, at the same time, intends not to X in C, and in which an eccentric billionaire will put a million dollars in the agent’s bank account if he continues to have these intentions. The agent in this case has a set of attitudes which, by the Commitment Agglomeration Principle, commit him to [X-ing and not X-ing] in C. But, since [X-ing and not X-ing] in C is impossible, the agent lacks sufficient reason to do so. And by the Commitment Transmission Principle, it follows that he lacks sufficient reason to have the combination of attitudes he has. Since he lacks sufficient reason to have

29  Ross 2012, p. 166. 30  Ross 2012, p. 166–7. I won’t assess this argument here. 31  Ross 2012, p. 167. 32  Ross 2012, p. 168. I’ll accept this principle for the sake of argument. 33  In what follows, I’ll omit “evidence-relative” to avoid distracting clutter.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

134  Instrumental Rationality this combination of attitudes, he ought not have them.34 Thus, we arrive at the strong normativity of the rational prohibition on inconsistent intentions. The beauty of Ross’s account, which he recognizes, is that it doesn’t require any general skepticism about the existence or relevance of state-given reasons. And so he could perfectly well allow that I ought to intend to go to Block Island rather than Provincetown.35 However, one might worry: what if we make the Provincetown vacation slightly more appealing than the Block Island vacation, so that I don’t have sufficient reason to go to Block Island. By the Commitment Transfer Principle, I now lack sufficient reason to intend to go to Block Island. But given the significant benefits of intending to go—if we need to add more, we could introduce an eccentric intention-detecting billionaire who’ll reward my intending to go to Block Island—this seems to be a counterintuitive result. Ross is aware of this worry.36 And he offers a response: Not only does intending to X commit one to X-ing, but it also commits one to closing de­lib­er­ation on whether to X. So, in the example in which I intend to go to Provincetown, I have thereby committed myself to closing deliberation on the question of whether to go to Provincetown. But I lack sufficient reason for closing deliberation on this question. After all, closing off deliberation makes it highly unlikely I’ll intend to go to Block Island and gain the rewards that come with doing so. (As Ross notes, “if he reopens deliberation, he may change his mind, but if he does not reopen deliberation, he will not change his mind.”37) And by the Commitment Transmission Principle, since I lack sufficient reason for closing deliberation, I lack sufficient reason for intending to go to Provincetown. But no parallel line of reasoning would threaten my intention to go to Block Island were I to form such an intention, and so Ross can deliver the intuitive verdict that I ought to intend to go to Block Island. It appears that Ross has offered us a plausible way to defend the strong nor­ mativity of intention consistency without being forced to deny that pragmatic benefits can be relevant to decision-making in cases like our Block Island/ Provincetown case. Let’s now consider whether this strategy could be employed to defend the strong normativity of Means–Ends Coherence against the problem of advantageous incoherence. One difficulty, off the start, is that it’s hard to see how the means–ends incoherent agent has a mental state with a set of attitudes that would commit 34  Ross takes the fundamental normative “ought” to be evidence-relative. See especially pp. 162–4. 35  This case has the same structure as his “Buridan-Kavka” case. See Ross 2012, p. 156. 36  This case has the same structure as his “Thinking Outside the Box” case. See Ross 2012, p. 158. 37  Ross 2012, p. 174.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  135 him to doing anything he lacks sufficient reason to do. The problem with the means–ends incoherent agent is that he has two attitudes and lacks another: he intends an end, believes intending a means is necessary for that end, and doesn’t intend the means. And the two attitudes he has are also possessed by the person who is means–ends coherent, and so it would be problematic if they committed him to doing something he lacks sufficient reason to do, since the means–ends coherent agent would also be so com­ mitted. (One way around this would be to say that he has a third attitude— a belief that he doesn’t intend the means—and these three attitudes together commit him to doing something he lacks sufficient reason to do. But, as we’ll see in the next chapter, our failures to intend the means aren’t always transparent to us.) But perhaps we can work with a broader understanding of “set of attitudes” so that it includes both the attitudes one has and the attitudes one lacks. One could then argue that the means–ends incoherent agent has a set of attitudes that commit him to doing something he lacks sufficient reason to do. But the challenge is to say what it is that the means–ends incoherent agent is committed to doing. Suppose I intend to travel to Charleston by train tomorrow and believe that to do so I must intend to buy a ticket, but, incoher­ ently, don’t have any such intention. Perhaps we could say, on the model of Ross’s treatment of the norm of intention consistency, the following: I have a mental state with a set of attitudes (broadly understood to encom­ pass both the having and lacking of attitudes) that commit me to traveling to Charleston by train tomorrow without buying a train ticket. Since this is impossible, I do not have sufficient reason to travel to Charleston by train tomorrow without buying a train ticket, and, by the Commitment Transmission and Agglomeration Principles, I lack sufficient reason to have that set of attitudes.

But this line of reasoning is unpromising, since the central component of it— the claim that traveling to Charleston by train tomorrow without buying a train ticket is impossible—may well be false. Suppose the University Travel Agency has already booked my train, and so buying a ticket isn’t necessary for travelling to Charleston. And suppose that there’s strong evidence available to me that this is so, but that evidence hasn’t registered with me. Since traveling to Charleston tomorrow without buying a train ticket is perfectly possible, it doesn’t follow that I lack sufficient reason (whether evidence-relative or not) to do this. Indeed, it seems to be precisely what I ought to do. And so we

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

136  Instrumental Rationality cannot apply the Commitment Transmission Principle in the same way Ross did for the case of inconsistent intentions. In short, unlike the case of someone who intends to X and intends not to X, it’s hard to see how means–ends incoherent agents are always committed to doing something they lack sufficient reason to do. And so it’s hard to see how we can use the Commitment Agglomeration Principle and Commitment Transmission Principle to explain the strong normativity of Means–Ends Coherence.38 Let’s now turn to the other main component of Ross’s treatment of the problem of advantageous incoherence. Ross was worried that the Commitment Transmission Principle would issue counterintuitive results in cases like our modified Block Island/Provincetown case, in which the Provincetown vacation is slightly better, but there are very significant advantages that come from intending to go to Block Island. It seemed that the Commitment Transfer Principle would yield the bad result that I ought not intend to go to Block Island. Ross claims that it need not yield this result, since the other alternative—intending to go to Provincetown— would commit me to not reopening deliberation, which would decrease the probability of my getting the benefits that would come from intending to go to Block Island instead. However, the problem with Ross’s response is that it would seem equally applicable to cases of advantageous means–ends incoher­ ence. Suppose that Prof. Henri Liable resolved his incoherence and intended to talk with the administrators. This intention, like any other, commits him to not reopening deliberation on this topic. But if he doesn’t reopen deliberation, he won’t get the benefits that come along with continued practical incompetence (namely, the reduced service load). So, by the Commitment Transmission Principle, since he lacks sufficient reason to not reopen deliberation, he lacks sufficient reason to intend to talk with the administrators. The only difference is that in the modified Block Island/Provincetown case, the cost of not reopening deliberation is that one loses out on the benefits that would come from intending to go to Block Island, while in Henri’s case, he loses out on the benefits that would come from not intending the means. But since both cases involve a costly commitment to not reopening de­lib­er­ation, the Commitment Transmission Principle would apply in both cases. And that means that just as our vacationer ought not intend to go to Provincetown, 38  The same point could be made with respect to the normativity of intention–belief consistency (the rational prohibition on my, say, intending to X, intending to Y, and believing I will X only if I will not Y). If my belief is false, then [X-ing and Y-ing] could be perfectly possible and reasonable. And so Ross’s account of the normativity of simple intention consistency won’t extend to intention–belief consistency.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  137 because doing so commits him to not deliberating further, thereby forgoing the benefits that come from a change of mind, Henri ought not intend to talk to the administrators, since doing so commits him to not deliberating further, thereby forgoing the benefits that come with continued incoherence. In short, if the line of reasoning Ross proposes to avoid counterintuitive results in cases like our modified Block Island/Provincetown case is acceptable, it would also apply equally well to cases of advantageous means–ends coherence, and so is no help to the defender of the strong normativity of means–ends coherence. In summary, I’ve here considered Ross’s strategy for solving “the pragmatic reasons problem” for the strong normativity of rationality. Ross’s strategy requires that we show how irrational combinations of attitudes commit one to doing something one lacks sufficient reason to do. But, while it’s plausible that inconsistent intentions commit one to doing something impossible and hence insufficiently supported by reasons, it’s not clear how we could say something similar for means–ends incoherence. Moreover, Ross’s strategy for allowing the pragmatic benefits of intentions to be relevant in some cases, as in our modified Block Island/Provincetown case, seems too permissive, since that strategy would also allow for the pragmatic benefits in Henri’s case to be rele­ vant, thereby threatening Strong Normativity.

5.2.2  Transmission to Necessary Means A second argument against Strong Normativity, due to Kieran Setiya, con­ cerns cases in which an agent is means–ends incoherent, but unable to modify his end or his instrumental belief.39 So, the only way he can comply with Means–Ends Coherence is by intending the means. According to Strong Normativity, this person ought to be means–ends coherent. But if the only way to do so is by intending the means, it would follow, by a plausible trans­ mission principle, that he ought to intend the means. (Here is Setiya’s formu­ lation of that transmission principle: “Transmission: If you should do E, all things considered, and doing M is a necessary means to doing E, you should do M, all things considered, too.”40 We’re here, roughly, plugging in “satisfy Means–Ends Coherence” for E, and “intending the means” for M.) But the 39  See Setiya 2007, pp. 660–1, who draws upon Greenspan 1975, p. 265. The argument can also be found in Schroeder 2009, p. 277, and Bedke 2009, pp. 686–7. 40  Setiya 2007, p. 656. See also Schroeder 2009, p. 234; Bratman 2009c, p. 424; and Kiesewetter 2015, p. 922, for similar versions of this principle.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

138  Instrumental Rationality problem is that it might not be the case that he ought to intend the means, as illustrated by Setiya’s example of someone who intends to smoke, believes that to do so he must intend to buy a pack, but doesn’t intend to buy a pack. In this case, if the end and instrumental belief are unalterable, it would follow from Strong Normativity, Means–Ends Coherence, and Transmission that he ought to intend to buy a pack.41 But that’s clearly not the case. As we noted earlier when considering a related objection to Normative Disjunctivism (Chapter 4, Section 4.2.1), it won’t help to argue for the impos­ sibility of unalterable beliefs or intentions, since even if we’re not licensed to call these attitudes “intentions” and “beliefs,” it’s still clear that one can display instrumental rationality or irrationality with respect to them, and so the same problems will arise. But there are some more promising strategies for resisting the argument that we should consider. In particular, we should consider chal­ lenges that have been raised to the crucial premise, Transmission.42 One way to challenge Transmission is to return once more to the case of  Professor Procrastinate, introduced by Jackson and Pargetter (1986). Procrastinate ought to accept the invitation and write the review. A neces­ sary means to doing so is accepting the invitation. It thus follows from Transmission that he ought to accept the invitation. But, given that he won’t write the review, it’s not the case that he ought to accept the invitation. And so Transmission is false.43 John Broome has challenged Transmission with a similar example: Suppose that you ought to see a doctor and to do so you must take a day off work. But you won’t see a doctor and will instead just sit around and waste the day away. Here, it seems that you shouldn’t take the day off work, even though it’s a necessary means to doing something you ought to do.44 There are other ways to challenge Transmission. Jonathan Way has observed that while Transmission is plausible when we plug in some action for “E,” questions arise when we plug in attitudes, like “intending to X” or “believing that P.”45 Suppose that you ought to intend to X (say, because X-ing will bring about some very good result), but the only way you’ll intend to X is if you also intend to Y, where there is no reason for you to Y. In this case, your reason to 41  The derived “ought” would be the same kind of “ought” that figures in Strong Normativity. 42  We should distinguish Transmission from deontic inheritance, according to which if P→Q, then O(P)→O(Q). This is the inference rule that gives rise to Alf Ross’s (1941) paradox: that we can infer from “you ought to mail the letter” that “you ought to mail the letter or burn it.” Ross’s paradox doesn’t arise for Transmission since “mailing the letter or burning it” is not plausibly thought to be a neces­ sary means to mailing it. (It doesn’t facilitate your mailing the letter, nor make it more probable that you mail the letter, nor help bring it about that you mail the letter, and so forth.) 43  See Kolodny, 2018, §4. 44  Broome 2013, p. 126. 45  See Way 2010, p. 226.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  139 intend to X would be object-given (since it’s explained by your ­reasons to X) but your reasons to intend to Y would be state-given (since, we’re assuming, there’s no reason for you to Y). This shows that the following transmission principle is false: if there’s conclusive object-given reason to E, and doing M is necessary for doing E, there’s conclusive object-given reason to M. (Just plug in “intend to X” for E and “intend to Y” for M to see that it’s false.)46 Now, if one were to adopt a skepticism about the existence of state-given reasons, then it would also follow that Transmission is false, since there would be no reason for you to intend to Y, despite your intending to Y being a necessary means to your intending to X and it being the case that you ought to intend to X.47 That gives us another argument against Transmission. We could respond by challenging these two objections directly, for instance by resisting the argument against Transmission involving Professor Procrastinate,48 or by defending the existence of state-given reasons to intend. But it might be easier to see if we could run Setiya’s argument on weaker premises. We can get around the Professor Procrastinate objection by work­ ing with a weaker transmission principle applicable to necessary and sufficient means: Weaker Transmission:  If you should do E, all things considered, and doing M is a necessary and sufficient means to doing E, you should do M, all things considered, too. In the case of Professor Procrastinate, his accepting the invitation is necessary but not sufficient for his doing what he ought to do. Likewise, in Broome’s example, your taking the day off is necessary but not sufficient for your seeing the doctor. So, such cases pose no threat to Weaker Transmission.49 But Weaker Transmission is enough to get Setiya’s argument off the ground, since in Setiya’s example (in which the agent intends to smoke, believes buying a pack is necessary, but cannot alter these two attitudes) intending the means is 46  See Way 2010, pp. 226–8. See also Way 2012a, p. 499. 47  In the 2010 paper, Way doesn’t opt for skepticism about state-given reasons for intention (see 2010, p. 229), but in his 2012b, he defends skepticism about the “wrong kind” of reasons, which would license skepticism about state-given reasons to intend, on the assumption that these are reasons of the wrong kind. It’s not clear to me how Way’s observations here about the transmission of object-given reasons would amount to a counterexample to Setiya’s Transmission without such skepticism. After all, the defender of Transmission could happily concede that the reasons to intend to Y are state-given. So long as it’s the case that one ought to intend to Y, Transmission is unthreatened. 48  See Kiesewetter  2015 for a defense of Transmission against such objections. Kolodny  2018 is critical of Transmission, but proposes his own principle, General Transmission, and argues that it can be used instead to get Setiya’s argument running. (See especially §7.) 49  Way (2010, p. 225, fn. 32) also makes a point along these lines.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

140  Instrumental Rationality both necessary and sufficient for satisfying Means–Ends Coherence. So, if  we combine Strong Normativity, Means–Ends Coherence, and Weaker Transmission, we’d still get the undesirable conclusion that the smoker ought to intend to buy a pack. We can thus sidestep the objection involving these alleged counterexamples to Transmission. What about the second objection to Transmission, which concerns the transmission of object-given reasons? I think we could instead work with a weaker and highly plausible principle concerning the transmission of objectgiven reasons, which I’ll now explain. Way is correct that object-given reasons to intend will not transmit along the lines of Setiya’s Transmission: it doesn’t follow from there being conclusive object-given reason to intend to X, and your intending to Y being necessary for your intending to X, that there’s conclusive object-given reason to intend to Y. But the following principle about the transmission of object-given reasons seems plausible: Object-Given Reasons Transmission:  If there’s conclusive object-given reason to A or B, and it is not possible for you to A, then there’s conclusive object-given reason to B. Suppose that there’s conclusive object-given reason to either not intend to drink or not intend to drive. (It’s because drinking and driving presents a sig­ nificant risk to others and yourself that you ought not both intend to drink and intend to drive.) Now, suppose you must intend to drink. It would follow from Object-Given Reasons Transmission that there’s conclusive object-given reason not to intend to drive. And that seems to be precisely the right result. After all, your reason—that drinking and driving poses a significant risk to others and yourself—is no less applicable. Now, assuming for the sake of argument that there’s conclusive objectgiven reason to be means–ends coherent50—that is, to either not intend the end or not hold the instrumental belief or intend the means—we could use Object-Given Reasons Transmission to support the conclusion that Setiya’s smoker, who cannot modify his end or his instrumental belief, has conclusive object-given reason to intend to buy a pack. We still generate the undesirable conclusion from plausible premises and Strong Normativity. To avoid the undesirable conclusion, we should reject Strong Normativity.

50  I registered some skepticism about this idea earlier (note 25, this chapter).

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  141 In summary, Setiya’s argument requires much weaker assumptions than is often thought. Instead of using Transmission, we could work with Weaker Transmission and Object-Given Reasons Transmission to show how Strong Normativity and Mean-Ends Coherence together yield the counterintuitive prediction that the smoker ought to intend to buy a pack (and also has conclusive object-given reason to do so, provided that he has conclusive object-given reason to be means–ends coherent). Thus, we should reject Strong Normativity.51

5.2.3  Transmission to Sufficient Means While it seems plausible that oughts transmit to necessary means, or perhaps more narrowly to means which are both necessary and sufficient, no one thinks that they transmit to merely sufficient means. In other words, no one thinks that if one ought to E, and M is a sufficient means to E-ing, then one ought to M. Suppose I ought to get a baseball glove for my son for his eighth birthday. One sufficient means of getting a glove is buying one from the local sporting goods store. But another sufficient means is stealing a glove cherished by my neighbor’s kid. If oughts transmitted to sufficient means, then it would follow that I ought to buy one from the store, and that I ought to steal one from my neighbor’s kid. But it’s clearly not true that I ought to steal the glove from my neighbor’s kid. But it doesn’t seem as implausible to say in this case that there’s a reason to steal the glove from my neighbor’s kid, so long as we take care to note that this reason is heavily outweighed by the reasons not to steal the glove. After all, it’s

51  Michael Bratman (2009c) also offers an important reply to Setiya’s argument, which I consider in detail elsewhere (Brunero  2010b). Bratman’s reply is somewhat concessive to the critic of Strong Normativity in that he concedes Setiya’s example is one in which Means–Ends Coherence applies, but in which there’s no reason to be means–ends coherent. But Bratman offers a promising explanation of why there is a reason to be means–ends coherent in other cases, but not in this particular case. As we saw earlier, on his view, the reason to be means–ends coherent is a reason of self-governance. He argues that when we fail to be means–ends coherent, we are such that “there is no clear answer to the question ‘Where do I stand?’ with respect to E [the intended end]” and so we fail to govern our­ selves (2009c, p. 434). We have a reason to avoid such failures of self-governance. But this reason of self-governance is present only when certain enabling conditions obtain—specifically, only when we are capable of the relevant kind of self-governance. And he argues that when one cannot modify one’s intentions and instrumental beliefs—as in the case of Setiya’s smoker—the relevant kind of selfgovernance is not possible. However, I am skeptical of his claim that whenever one cannot modify one’s intentions and instrumental beliefs, one is incapable of self-governance. The impossibility of modifying one’s intentions and instrumental beliefs might be due to some cause—perhaps a Frankfurtian coun­ terfactual intervener—that poses no threat to the possibility of relevant self-governance, across all leading theories of self-governance. See Brunero 2010b for further discussion.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

142  Instrumental Rationality pretty easy to identify what the reason is: by stealing the glove I would thereby be getting a baseball glove for my son. Additionally, as we noted in Chapter 2, we can appeal to Gricean conversational pragmatics, as Mark Schroeder has done, to explain why we should be cautious about our intuitions about the non-existence of reasons in such cases.52 Ordinarily, when someone says that there’s a reason for someone to do something, we assume they are speaking about a relatively weighty reason, since lightweight reasons would not be worth mentioning or thinking about. For the same reason, when someone says that there’s no reason to steal the glove, they’re likely denying the existence of a weighty reason to steal it. But the non-existence of such weighty reasons is perfectly compatible with the existence of lightweight reasons to steal the glove. (As Schroeder observes, our intuitive resistance to the claim that there’s a reason to steal the glove diminishes once the other party to the conversation makes it explicit that the reason is a lightweight one, and once again when that person specifies what the reason is—in this case, that doing so provides a glove for my son on his birthday.) So, it might be that we could endorse a weaker principle concerning transmission to sufficient means: Transmission to Sufficient Means:  If you ought to E, and M is a sufficient means to E-ing, then there’s a reason to M.53 If that’s right, since I ought to get a baseball glove for my son, there’s a reason for me to steal the glove from my neighbor’s kid. But, as Joseph Raz has observed, there is now a problem for the com­bin­ ation of Strong Normativity and Means–Ends Coherence. If Bob ought to be means–ends coherent, and a sufficient means to being means–ends coherent 52  Schroeder 2007, pp. 92–7. 53  A proper formulation of this principle would have to take into account cases of superfluous means. (For further discussion, see Raz  2005, pp. 5–6, and Kolodny  2018, §4). Suppose I ought to make coffee at 7:00, and I’ve set the entirely reliable automatic timer on the coffee pot for then. But another sufficient means for making coffee at 7:00 is pressing the “on” button at 7:00, which cancels the timer function and turns on the pot. Intuitively, although pushing the button is a sufficient means for making coffee at 7:00, I have no reason to take it since it’s entirely superfluous. (Things might be different if the timer were somewhat unreliable.) Consider also some cases in which E is disjunctive. Suppose I ought not to write philosophy while high on cocaine; in other words, I ought to either not write philosophy or not be high on cocaine. One sufficient means to this is not being high on cocaine. Another sufficient means to this is not writing philosophy. But it doesn’t seem like this provides me with any reason not to write philosophy right now. (I’m not high on cocaine.) Not writing philosophy, although sufficient for my either not writing philosophy or not being high on cocaine, is entirely superfluous, since I’m not high on cocaine. Kolodny (2018, §7) notes that his principle, General Transmission, could replace Transmission to Sufficient Means, avoiding such counterexamples, while still allowing Raz’s argument to go through.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  143 is his intending to poison his colleagues, then there’s a reason for Bob to intend to poison his colleagues.54 What is Bob’s reason? As Raz explains: A roundabout way to identify the reason is to say that it is a reason to avoid being in a situation in which one would be in breach of that reason. And one would be in breach if one both intends E and fails to do M. There are two ways to avoid being in that situation. One is to abandon the intention to do E. The other is to do M.  So one has both a reason to do M and a reason to abandon one’s intention to do E (though no reason to do both, because once one does one of them the reason to do the other lapses).55

So, we must accept that there’s a reason for Bob to intend to poison his colleagues.56 However, it’s not clear that this is a bitter pill to swallow. After all, the same strategies we employed in defense of Transmission to Sufficient Means would work equally well here. We could simply accept the conclusion that there’s a reason for Bob to intend to poison his colleagues, taking care to specify that this reason is heavily outweighed by the reasons not to intend to poison them. And we can appeal to Gricean conversational pragmatics to explain away our intuitions that there’s no such reason.57 In short, if you think that the strategy employed in defense of Transmission to Sufficient Means against the counter­ examples above works, you should also think it works here in defense of the combination of Means–Ends Coherence and Strong Normativity. However, there’s a better way to develop Raz’s argument, so that the chal­ lenge it presents isn’t so easily met. Remember that Means–Ends Coherence is a wide-scope requirement with which you could comply in three ways: by intending the means, dropping the end, or dropping the instrumental belief. Each of these three ways of proceeding is sufficient for compliance with Means–Ends Coherence. And so by Transmission to Sufficient Means and 54  Raz also thinks that it would follow from the weaker claim that Bob has a reason to be means– ends coherent that he has a reason to intend to poison his colleagues. (See Raz 2005, p. 12.) But that need not concern us here, since we’re limiting our focus to Strong Normativity. For discussion of Raz’s argument, see Rippon 2011. 55  Raz 2005, p. 12. 56  It might be even worse: you have a reason to poison your colleagues, too. After all, Means–Ends Coherence applies to all agents, and so all agents ought to avoid the prohibited combination [intend­ ing to kill one’s colleagues, believing one must intend to poison them to do so, not intending to poison them], a sufficient means to which is intending to poison them. (See Rippon 2011, p. 12.) One could avoid this counterintuitive result by limiting the applicability of means–ends coherence to agents who already have the relevant intentions and beliefs, perhaps along the lines of the wide-scope requirement Revised Wide Synchronic MEC, discussed in Chapter 3, §1. 57  As Schroeder 2005 notes, in response to Raz.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

144  Instrumental Rationality Strong Normativity, it would follow that there’s a reason for the means–ends incoherent agent to drop his instrumental belief. In our example, there’s a reason for Bob not to believe that he must poison his colleagues if he’s to kill them.58 But this seems false.59 We haven’t been given any evidence against that belief, and so it’s not clear how evidential considerations would provide any reason against his so believing. And pragmatic reasons for belief, if there are such reasons, involve something of value being promoted by one’s believing or not believing some proposition (e.g., believing you’re handsome improves your confidence; believing in heaven increases your chances of admission). But what good comes from Bob avoiding incoherence in this way? It won’t make him happier, more popular, or bring him any rewards.60 So, it seems false to say that there’s a reason for Bob not to believe that poisoning is neces­ sary for killing. And Gricean conversational pragmatics will be of little help here. That strategy might be able to help explain how, despite initial intuitions to the contrary, there could be a reason to do something deeply immoral, like poi­ son one’s colleagues. But it won’t help explain how there could be a reason for Bob not to believe intending to poison is necessary for killing when there are no evidential considerations supporting his dropping his belief, and there are no pragmatic benefits that come from his doing so. Note that once we cancel any relevant pragmatic implicatures—say, by making it explicit that we’re listing out all the pros and cons, regardless of weight—it still seems to us that Bob has no such reason. It might be tempting to think the problem I’ve posed for Strong Normativity here is relatively insignificant. Specifically, one might worry that reasons are 58  Transmission principles are standardly thought to extend to constitutive means. (For instance, see Raz 2005, p. 5.) For example, if you ought to do something exciting on your birthday, and going skydiving constitutes an exciting activity, then, by Transmission to Sufficient Means, you have a reason to go skydiving on your birthday. In our example, each of these three ways of proceeding constitutes compliance with Means–Ends Coherence. And so, if Strong Normativity is true, there would, by Transmission to Sufficient Means, be a reason to proceed in each of these three ways. 59  One reader has correctly observed that we could avoid the counterintuitive result if we formu­ lated Transmission to Sufficient Means so that the relevant reasons never transmit to reasons to believe (or not believe). This suggestion avoids the difficulty, but strikes me as ad hoc. One way to see why it’s ad hoc is to observe that no such restriction would be supported by the rationale for transmission articulated in the passage from Raz quoted above. Raz notes in that passage that we have reasons to avoid being in breach of the reasons we have. But Bob could avoid being in breach of the (alleged) reason to be means–ends coherent in three ways: by not intending the end, by intending the means, and by not having his instrumental belief. Any of these three ways of proceed­ ing avoid the relevant breach. If that’s the rationale for the transmission principle, it doesn’t support the proposed restriction. 60  In the philosophical literature on value, standard candidates for what has final value include happiness, autonomy, knowledge, and the like. As far as I know, no one has seriously proposed adding the avoidance of means–ends incoherence to that list. So, it won’t help to suggest that Bob’s not being means-end incoherent itself has final value.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  145 cheap—there are already a whole bunch of reasons for a whole bunch of ­attitudes—and so it’s not so bad to be forced to accept just one more reason for Bob not to have a particular belief. But this misunderstands the problem. The problem isn’t just that we have to add to our stock of reasons, but that we’re also forced to acknowledge a whole new category of theoretical reason. We’re familiar with evidential reasons for believing, or not believing, P. We’re also familiar with the (more controversial) pragmatic reasons for believing, or not believing, P. Such reasons, it is alleged, are provided by the benefits, or costs avoided, in having, or not having, a particular belief. (Such reasons are more controversial in that evidentialists will deny that such pragmatic bene­ fits are really reasons for belief.61) But we’re now forced to acknowledge theoretical reasons that have neither a basis in evidence (nor in any epistemic goals or aims) nor a basis in the pragmatic benefits, or harms avoided, in having the belief—a whole new, mysterious kind of theoretical reason. That strikes me as a significant cost. In summary, if we accept Strong Normativity and Means–Ends Coherence and Transmission to Sufficient Means, we’d have to accept that there’s a reason for Bob not to have his instrumental belief, even in cases where there are no epistemic reasons for him not to have it, and no good comes from his not having it. But that seems false. And so we have yet a third reason to reject Strong Normativity.

5.3  Subjective Oughts There’s another approach to explaining means–ends coherence that we should consider. It’s an approach which makes use of the concept of a subjective ought. Whereas what one objectively ought to do is determined by the rele­ vant facts, what one subjectively ought to do is determined by one’s beliefs, or one’s rational beliefs, or something similar. Suppose, to return to Williams’s example, the bartender has switched out the gin for gasoline, but you have no grounds for thinking he’s done so, and reasonably believe the glass contains the gin and tonic you want. You subjectively ought to drink the contents of the glass, but objectively ought not to do so.62 The distinction is a promising one to bring into discussions of rationality, since some authors take this distinction between objective and subjective oughts to be the same as the distinction between what one has most reason to 61  See, for instance, Shah 2006.

62  Williams 1981, p. 102.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

146  Instrumental Rationality do and what it would be most rational to do. Here, for instance, is Derek Parfit’s presentation of the latter distinction: As rational beings, we can ask: What do we have most reason to want, and do? What is it most rational for us to want, and do? These questions differ in only one way. While reasons are provided by the facts, the rationality of our desires and acts depends instead on what we believe, or – given the evidence, ought rationally to believe. When we believe the relevant facts, these questions have the same answers. In other cases, it can be rational to want, or do, what we have no reason to want, or do. Thus, if I believe falsely that my hotel is on fire, it may be rational for me to jump into the canal; but I may have no reason to jump.63

Even though I have no reason to jump, I have a belief which is such that were it true, I would have a reason to jump. (As Parfit puts it, it’s a belief whose truth would give me a reason to jump.64) Since rationality and the subjective ought might be linked up in this way, we should consider whether the subjective ought could help us explain the particular rational requirement to be means–ends coherent.65 This strategy has been developed in different ways by Mark Schroeder (2009) and Jonathan Way (2012a). On Schroeder’s view, means–ends incoherent agents subjectively ought to intend the means. And that’s because whenever an agent intends to X, he believes that he objectively ought to X.66 And in the same way that my belief that the hotel is on fire is such that its truth ensures that I objectively ought to jump, the incoherent agent’s belief that he objectively ought to X is such that its truth ensures, trivially, that he objectively ought to X. And so we can say that he subjectively ought to X.67 And then suitable 63  Parfit 1997, p. 99. See also Parfit 2011, pp. 33–6. 64  Parfit 2011, p. 34. 65  Way  2009 argues that this strategy (of understanding rationality in terms of the reasons we would have were our beliefs true) would be a way of vindicating the normativity of rationality. As he sees it, if a concept (like goodness, or ought, or—on this strategy—rationality) is explained in terms of reasons, that’s enough for it to count as normative. (After all, as he points out, the error theorist would be committed to denying the truth of claims about what’s good, what we ought to do, and—on this strategy—what’s rational.) This is a broad conception of normativity. On this conception of normativ­ ity, even a set of requirements that call for us to do the opposite of what we have most reason to do— call them “Costanza-requirements”—would count as normative. (Since Costanza-requirements are explained in terms of reasons, their truth would also be denied by the error theorist.) This might lead us to suspect that Way’s conception of normativity is overly broad. But I won’t pursue this question here. (For further discussion, see Kiesewetter 2012.) 66  Schroeder 2009, p. 236. 67  Schroeder 2009, p. 230. One might reasonably worry that allowing the truth of normative beliefs, like a belief that one objectively ought to X, to generate subjective oughts or reasons would allow for

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  147 transmission principles will allow us to derive the conclusion that he subjectively ought to intend the means he believes necessary for X-ing.68 This is a quick description of Schroeder’s account. But we’re already in a position to see that its central component—the claim that intending to X involves believing one objectively ought to X—is, as Schroeder recognizes, highly controversial. One challenge to this claim concerns whether it would hold for the tie-breaker and incommensurability cases discussed in Chapter 2. Schroeder considers tie-breaker cases—what he calls cases of “picking”—and argues that in such cases people would be aware that intentions provide ­reasons.69 For instance, a person would understand that in choosing the can of tomatoes on the left, that choice makes a normative difference by providing a “very trivial reason” that tips the scales, and so she would believe, after hav­ ing made the choice, that she ought to get that can. (Schroeder doesn’t con­ sider cases of incommensurability, where, as we’ve noted, the addition of very trivial reasons won’t “tip the scales.” To deal with such cases, the reason pro­ vided by one’s intention would have to be weightier.) Although we’ve already argued against the intentions-provide-reasons view in Chapter  2, what’s important for Schroeder’s argument is that the agent believes that intentions make a normative difference in these cases, whether or not this is actually so. However, I have my doubts. In choosing one of two incommensurable options—say the philosophy career over the circus career—I might (perfectly reasonably) think that this choice hasn’t made it such that I now ought to pursue the philosophy career. I could think the normative situation is one of incommensurability, both before and after my choice. Likewise, when someone comes to intend to do something she thinks is supererogatory, she likely won’t think that her so intending makes the option now required, so that now she ought to do it. In these cases, one intends to X while believing it’s not the case that one ought to X. Yet the intention to X puts in place a requirement to be means–ends coherent that would remain unexplained on the present strategy. Not only would Schroeder have to show that in these cases, contrary to appearances, the person who intends to X also believes she ought to X, but he would also have to show that she doesn’t believe it’s not the case she ought to X. Suppose that Schroeder only establishes that in intending to X she believes she ought to X. In that case, the people in these examples would have us to come by them too easily (especially if we take the subjective ought to be linked to what we rationally ought to do). For this worry, see Parfit 2011, p. 119. 68  For those principles, see Schroeder 2009, pp. 234, 245.

69  Schroeder 2009, pp. 237–8.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

148  Instrumental Rationality inconsistent beliefs: believing both that it’s not the case that she ought to X and that she ought to X.70 Now, if we consider just the latter belief, we can say that she subjectively ought to X, since the truth of the contents of that belief would make it the case that she objectively ought to X. But what if we take into account the former belief as well? The truth of its contents would make it not the case that she objectively ought to X, and so we wouldn’t then be able to generate the conclusion that she subjectively ought to X. (And, of course, in general, we should consider all of the agent’s beliefs in determining what she subjectively ought to do.71 After all, if the person in Parfit’s example also believed the canal was frozen and thirty stories below, then it would no longer be the case that he subjectively ought to jump into the canal. Considering just his belief that the hotel is on fire would lead us astray.) And without it being the case that she subjectively ought to X, we can’t use the relevant transmis­ sion principles to get the desired conclusion that she subjectively ought to intend the means believed necessary for X-ing. Schroeder has more recently distanced himself from the specific proposals in his paper and endorsed Jonathan Way’s particular development of this ­general strategy of appealing to subjective oughts to explain means–ends coherence.72 Way’s account works with the medium-scoped formulation of means–ends coherence: Medium Means–Ends Coherence:  If one believes one will X only if one intends to Y, then rationality requires that [if one intends to X, then one intends to Y].73 70  The problem identified here would also emerge for Schroeder’s suggestion (p. 239) of a weaker thesis: that intending to X involves believing one has adequate reason to X. I could believe I don’t have adequate reason to stay up late but intend to do so anyway. Since the contents of my beliefs here would be inconsistent, the same problem would emerge. Schroeder also floats the possibility of weakening the thesis another way: intending to X involves taking it that one ought to X, and then holding that “your subjective reasons depend on what you either believe or take to be true” (p. 239). But we’d still have the same problem, since the contents of one’s belief (“It’s not the case that I ought to X”) and one’s taking (“I ought to X”) would be inconsistent. The problem also isn’t avoided with Schroeder’s final proposed weakening (p. 245), according to which intending to X involves believing there’s a reason to X, since it still seems possible for me to think I have no reason to stay up late but intend to do so anyway. (Additionally, this proposal runs into further trouble, since we need to have it come out that the means–ends incoherent agent subjectively ought to intend the means, in order to explain the “strict­ ness” or “stringency” of means–ends coherence. And the truth of this belief would only ensure that there’s a subjective reason to X that would transmit to a subjective reason to intend the means to X-ing. And that would leave strictness unexplained. A similar problem would apply to the first weaker thesis above—the thesis that intending to X involves believing one has adequate reason to X—since one could have adequate reason to X, fail to X, and yet be perfectly as one ought to be, as when one also has adequate reason to do something else instead. So, this would also leave strictness unexplained.) 71  As Schroeder himself acknowledges on p. 243. 72  See Schroeder 2014c, p. 11. See Way 2010, 2012a. 73  See Way 2010, p. 223, and Way 2012a, p. 494. (I’ve taken the liberty of changing Way’s formula­ tion to include “intends to” in the content of the instrumental belief. He acknowledges this addition is necessary and omits it only for the sake of brevity (2012a, pp. 489,494).)

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  149 Think back to Parfit’s example in which you believe your hotel is on fire and are rationally required to jump into the canal. We can explain this rational requirement by noting that you have a belief whose truth would give you con­ clusive reason to jump into the canal. More generally, we could hold that, as Way puts it, “you are rationally required to A if you believe some things whose truth would give you conclusive reason to A.”74 If we adopt this framework, we could then explain Medium Means–Ends Coherence if the following were true: Conclusive Reasons Claim:  If one will X only if one intends to Y, then there’s conclusive reason to [if one intends to X, then one intends to Y ]. For this would ensure that the means–ends incoherent agent has a belief whose truth would give him conclusive reason to do what he is rationality required to do according to Medium Means–Ends Coherence. Unlike Schroeder’s account, Way’s account works with a non-normative belief (that one will X only if one intends to Y). It also differs from Schroeder’s account in that what rationality requires is disjunctive (to either not intend to X or intend to Y), whereas on Schroeder’s account what rationality requires is non-disjunctive (to intend to Y). It has an advantage over Schroeder’s account in that it doesn’t require us to adopt any controversial view about how inten­ tions necessarily involve beliefs about what one ought to do. However, the account would be subject to four objections we’ve already considered, which make me skeptical about the prospects for its success. I’ll here briefly review those four objections. First, in Chapter  3, Section  3.2.1, I argued against Medium Means–Ends Coherence, noting that it is subject to counterexamples. I argued that some­ times one would display no irrationality in continuing to have the com­bin­ ation of intending to X and not intending to Y and revising one’s instrumental belief instead. Since Way’s account requires that we work with Medium Means–Ends Coherence, it would be subject to that objection. The remaining three objections concern the Conclusive Reasons Claim. Second, as we saw in our discussion of the alleged costs of incoherence (Section 5.1.2), it’s possible for an agent to have a merely permissible end—e.g., my doing volunteer work on my sixtieth birthday—where there are no external or psychological costs involved in being incoherent with respect to that end, in which case it’s not 74 Way  2012a, p. 495. This formulation gets revised later, in ways that won’t matter for our discussion.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

150  Instrumental Rationality clear that one ought not have the combination of intending to X and not intending to Y. Third, there might even be cases where one ought to have that combination, as when there are strong pragmatic reasons to do so (as discussed in Section  5.2.1). Fourth, in cases in which one cannot alter one’s end, the applicable transmission principles would generate the false prediction that one ought to intend the means (as discussed in Section 5.2.2).75 Since these four objections apply to Way’s proposal, I’m skeptical about the prospects for an explanation of means–ends coherence along these lines.

5.4 Conclusion In this chapter and the previous one, I’ve considered three important strat­egies which appeal to normativity—where “normativity” is understood to involve claims about what one ought, or has reason, to do—to explain, or explain away, means–ends coherence. First, the Normative Disjunctivist holds that means–ends incoherent agents will either intend an end they ought not to intend, have a belief they ought not to have, or fail to intend a means they ought to intend. If this is correct, it raises the prospect of dismissing the rational requirement of means–ends coherence as a myth. I’ve argued that Normative Disjunctivism faces four sig­ nificant objections, and even if it were true, it doesn’t provide us with good grounds for dismissing means–ends coherence as a myth. Second, Strong Normativity holds that one ought to be means–ends coher­ ent. But even on the wide-scope formulation of means–ends coherence, this strategy encounters three significant problems: the “pragmatic reasons prob­ lem,” a problem concerning the transmission of oughts to necessary means in cases involving unalterable attitudes, and a problem concerning the transmis­ sion of reasons to sufficient means. I’ve also considered and rejected the approach of those who, inspired by Prichard’s critique of “moral philosophy,” argue that the search for “external” reasons to be rational is unnecessary since rationality provides its own reasons. I think that the demand to produce “external” reasons to be rational is a legitimate demand, and it’s not easy to see how to meet it. A third, more indirect, strategy holds that means–ends incoherent agents will have some belief whose truth would ensure that some relevant normative claim holds, thereby allowing us to appeal to the idea of a subjective ought to 75  See Setiya 2007, p. 662. We’ve already addressed Way’s response to this objection in Section 5.2.2.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Normativity II  151 explain means–ends coherence. But Schroeder’s way of developing this proposal requires that we commit ourselves to a controversial claim about the nature of intentions, while Way’s version of the proposal encounters several of the objections which confront the other strategies we’ve considered. It might have seemed natural to think that we can explain why rationality requires us to means–ends coherent by looking at what attitudes or com­bin­ ations of attitudes we ought to have (where this is the “ought” of “most rea­ son” or, in the case of the subjective ought, one closely related to it). But it turns out that each of these different strategies for developing this idea faces significant difficulties. So, we might need to look elsewhere. In the next chap­ ter, we’ll consider a different approach, one which aims to explain rational requirements governing intentions by looking to rational requirements governing beliefs.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

6 Belief Cognitivism about practical rationality is the view that some rational requirements governing intentions can be explained by the rational requirements governing beliefs. Cognitivists have tended to focus on two requirements of practical rationality in particular: the consistency requirement on intentions, and Means–Ends Coherence. Cognitivism about these practical requirements might seem promising. After all, we’re very comfortable speaking of consistency and coherence requirements governing beliefs. And many have thought that intentions involve beliefs in some way. And so it’s natural to think that the rational requirements governing those involved beliefs might explain the rational requirements governing those intentions. To get a sense of how such an explanation might go, consider how a cognitivist might explain a simplified version of a consistency requirement governing intentions, Intention Consistency:  Rationality requires that [if one intends to X, then one does not intend not to X]. The cognitivist could first point to a similar consistency requirement governing beliefs: Belief Consistency:  Rationality requires that [if one believes that P, then one does not believe that ~P]. And she could then defend a thesis about the way intentions involve beliefs: Strong Belief Thesis:  Intending to X involves believing that one will X.1

1  The Strong Belief Thesis is endorsed by Hampshire and Hart 1958, Grice 1971, Harman 1976, 1986, Velleman 1989, 2007, and Davis 1984, and opposed by Audi 1973, Davidson 1980a, Ludwig 1991, and Mele  1992. Bratman  1987 also expresses skepticism about the thesis, and develops an account of intention that doesn’t rely on it. For a very helpful overview of various possible views of the intention–belief connection, see Adams 2007, esp. pp. 143–7. Adams also opposes the Strong Belief Thesis. Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  153 If the Strong Belief Thesis is true, whenever one intends to X and intends not to X, one has inconsistent beliefs about what one will do: a belief that one will X and a belief that one will not X. This suggests that Intention Consistency can be explained by Belief Consistency and the Strong Belief Thesis: it’s ir­ration­al to have inconsistent intentions because, in doing so, you’ll have inconsistent beliefs about what you will do, which is irrational. This is a simplified version of a consistency requirement on intentions since, as we noted earlier, rationality requires not just that our intentions be consistent with one another, but also that they be consistent with our beliefs. For example, I would be irrational if I intended to travel to New York this afternoon and intended to travel to San Francisco this afternoon, while believing that if I travel to New York this afternoon, I won’t travel to San Francisco this afternoon. But the cognitivist employing the Strong Belief Thesis will have no trouble here since the associated beliefs—that I’ll travel to New York this afternoon, that I’ll travel to San Francisco this afternoon, and that if I travel to New York this afternoon, I won’t travel to San Francisco this afternoon—are jointly inconsistent. Cognitivist approaches to practical rationality have both their defenders (including Gilbert Harman, David Velleman, R.  Jay Wallace, and Kieran Setiya) and their critics (most notably, Michael Bratman).2 In this chapter, I’ll be concerned with the particular question of whether a cognitivist can ­successfully explain Means–Ends Coherence. I’ll start (Section 6.1) by considering the prospects for a cognitivist account of Means–Ends Coherence that appeals to the Strong Belief Thesis. I’ll argue we should reject such accounts since the Strong Belief Thesis is either false or it’s unable to do the explanatory work the cognitivist needs it to do. I’ll then (Section 6.2) consider the prospects for a cognitivist account that doesn’t appeal to the Strong Belief Thesis, but instead to some weaker thesis about the connection between intention and belief. I’ll start with Wallace’s cognitivist account, present two objections to it, and then present an account that’s similar but avoids these objections. I think this is the best available cognitivist account of Means–Ends Coherence. However, the best isn’t good enough. I’ll argue (Section  6.3) that even if cognitivists are able to show that every instance of means–ends incoherence involves incoherence in belief, they will still need to show that the theoretical requirements explain the practical ones. I’ll argue that there are two reasons 2 See Harman  1976,  1986, Velleman  1989,  2007, Wallace  2006, and Setiya  2007. See Bratman 1987, 2009a, 2009b.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

154  Instrumental Rationality to doubt cognitivism’s explanatory thesis. First, we should expect a theory that explains Means–Ends Coherence to also be able to explain closely related practical requirements, but cognitivism seems unable to do this. Second, if we consider all the relevant theoretical requirements that apply, and not simply select out some and ignore others, cognitivism issues false predictions about the rationality of ways of escaping from a state of means–ends incoherence.

6.1  Cognitivism with the Strong Belief Thesis Some philosophers have developed cognitivist explanations of Mean-Ends Coherence using the Strong Belief Thesis.3 Let’s consider how such an ex­plan­ ation might go. Suppose I’m instrumentally incoherent. Suppose that, after an enjoyable stay in Charleston, I intend to travel back to New York this afternoon, believe I’ll get there only if I intend to buy an airplane ticket, but don’t intend to do this. According to the Strong Belief Thesis, I believe: (1)  I will travel to New York this afternoon. If we pair this belief with my instrumental belief: (2)  I will travel to New York this afternoon only if I intend to buy an ­airplane ticket, and apply the theoretical rational requirement: Closure:  Rationality requires that [if one believes that P, and believes that if P then Q, then one believes that Q]4 then rationality requires that our instrumentally incoherent agent either give up his belief (1), which, by the Strong Belief Thesis, would involve his not 3  See Setiya 2007, pp. 663–71. Setiya’s account differs slightly from the sketch given below in ways that won’t matter for the following discussion. 4  As stated, this requirement might demand too much of us. One strategy of reply is to claim that Closure is a principle of “ideal rationality” which does not imply that any blame or criticism is appropriate for failures to live up to it. (See Setiya 2007, pp. 665–6). Another strategy would be to modify the principle in some way. John Broome (2013, pp. 157–8), for instance, adds a clause requiring one to care about whether Q is true: “Rationality requires of N that, if N believes at t that p, and N believes at t that if p then q, and if N cares at t whether q, then N believes at t that q” (p. 157). Other modifications are possible. I’ll assume the cognitivist will be working with some acceptable formulation of this principle. (For a discussion of the relationship between logic and reasoning relevant to the worry about demandingness, see Harman 1986, pp. 11–20.)

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  155 intending to travel to New York this afternoon, and so would involve his escaping instrumental incoherence, or give up his belief (2), which would also constitute an escape from instrumental incoherence, or come to believe: (3)  I intend to buy an airplane ticket. This gets us quite close to a cognitivist explanation of Means–Ends Coherence, but not all the way there, since it seems possible for one to believe one intends to do something without actually intending to do it. But if one can close this gap by showing how it is independently theoretically irrational to have false beliefs about one’s intentions in this context, then one would have a cognitivist explanation of Means–Ends Coherence in terms of Closure and the the­or­ et­ic­al rational requirement not to have false beliefs about one’s intentions in this context. But can this gap be closed in this way? Ordinarily, the presence of a false belief need not involve one in irrationality, especially if we understand rationality as a matter of coherence among one’s attitudes. It is of course true that a false belief can be criticized for its falsity, but it’s hard to see how a false belief necessarily involves attitudinal incoherence, whether the belief is about the “external” world, or about one’s own psychology (one’s desires, other beliefs, hopes, fears, intentions, etc.). So, there are questions about whether cognitivists can close this gap.5 (I’m personally skeptical about the prospects of doing so.) But for our purposes here, I’m going to assume that the cognitivist can find some way to close the gap and show how in falsely believing, say, (3) above, one would be theoretically irrational. Even then, however, the cognitivist account here is only going to be as plausible as the Strong Belief Thesis. According to the Strong Belief Thesis, intending to X involves believing one will X. But this need not preclude one from also believing one’s belief could be false. Just as I could believe that my friend Dana will be in New York this afternoon, but also believe that I could be wrong about this, when I intend to travel to New York this afternoon, and so believe that I will, I could also believe that I could be wrong about this. The Strong Belief Thesis is com­pat­ ible with one’s being a fallibilist about the involved beliefs.6 However, one might still worry that the Strong Belief Thesis is vulnerable to counterexamples. Critics of the thesis usually present cases involving an agent intending to do something difficult, where, it is alleged, the agent intends to X, but doesn’t 5  For an attempt to close this gap, see Setiya 2007, pp. 670–1. For arguments that the gap remains open, see Bratman 2009a, §4, and Brunero 2009b, §2. 6  See Harman 1986, p. 92.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

156  Instrumental Rationality believe that she will X. Some examples include a golfer intending to sink a difficult putt, an amateur basketball player intending to make a half-court shot, someone intending to leap across a wide gap, and someone intending to lift a heavy log. Defenders of the Strong Belief Thesis aren’t persuaded by these alleged counterexamples, and they usually reply by denying that these really are cases in which one intends to sink the putt, make the shot, leap across the gap, and so forth.7 Additionally, they have presented arguments for why we should accept the Strong Belief Thesis. We’ll now turn to this debate. I’ll first consider (Section 6.1.1) some im­port­ ant arguments for the Strong Belief Thesis, and I’ll argue that none of them give us sufficient reason to accept the Strong Belief Thesis. I’ll then consider (Section  6.1.2) the case against the Strong Belief Thesis, particularly these alleged counterexamples involving actions thought to be difficult. I’ll argue that the argumentative maneuvers (such as appeals to “intentions to try to X”) needed to save the Strong Belief Thesis from these counterexamples, if successful, undermine the cognitivist’s broader explanatory project. So, the cognitivist employing the Strong Belief Thesis faces a dilemma: she must either concede that the Strong Belief Thesis is false or maintain that it’s true but unable to play the explanatory role that the cognitivist needs it to play. Either way, cognitivist explanations employing the Strong Belief Thesis fail.

6.1.1  The Case for the Strong Belief Thesis In recent work, David Velleman argues that we need to accept the Strong Belief Thesis if we are to explain certain characteristic functions of intentions, and explain why intentions are governed by rational requirements like Means–Ends Coherence and Intention Consistency. He presents four arguments. First, Velleman reminds us of Elizabeth Anscombe’s view that the natural expression of an intention to X is “I am going to X.” It seems that if one is sincerely asserting this, then one believes that one will X. As Velleman 7  One way to proceed would be to consider linguistic evidence about agential expressions of an intention (or lack of an intention) in such cases. Suppose the golfer makes the difficult putt. On the one hand, if she were to say, “I intended to make that,” she comes across as bragging. (Harman 1986, p. 91) However, this might merely be a feature of Gricean conversational pragmatics; notice that the implicature of overconfidence can be cancelled by, for instance, her saying, “I intended to make that, but didn’t really think I would.” And there’s also some linguistic evidence pointing in the other direction: it would sound odd for the golfer to say, “I had no intention of making it.” (Adams 2007, pp. 151–2). We’ll return to these cases in Section 6.1.2. I think we can avoid putting much weight on such arguments. Our question is more manageable: can the Strong Belief Thesis be true in a way that lends support to the cognitivist explanation of Means–Ends Coherence? I think we can answer this question in the negative without having to rely on linguistic evidence of this sort.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  157 observes, the assertion, “I am going to X, but I don’t believe that I will” appears to be an instance of Moore’s paradox.8 One might think this is some evidence for the Strong Belief Thesis. But I don’t think this is right, since Anscombe’s remark could be evidence for the Strong Belief Thesis only if it is understood as a universal generalization—that is, as something like, “For all intentions, the natural expression of the intention is ‘I am going to X’ ”—and that generalization seems to be shown to be false by the very same examples that challenge the Strong Belief Thesis. For instance, for the person intending to make the difficult putt, “I am going to make it” doesn’t seem to be the natural expression of his intention. Something expressing less confidence—like “I am going to try to make it”—would better reflect his appreciation of the difficulty involved in making the putt, and hence be the more natural thing to say in this context. And no Moore-paradoxical results emerge if this less confident assertion is paired with one’s asserting, “But I don’t believe that I will.”9 Let’s now turn to a second argument from Velleman—one central to our concerns here—which holds that the Strong Belief Thesis is needed to explain means–ends coherence. Velleman writes: Why, for example, should an agent be rationally obliged to arrange means of carrying out an intention, if he is agnostic about whether he will in fact carry it out? Suppose that I form an intention to fly to Chicago next Tuesday, well knowing that I often forget to take trips that I have planned. (I am even more forgetful than Bratman.) Buying a ticket for my flight to Chicago will turn out to have been a waste of money if I forget to take the trip. . . . But why should I be categorically required to invest in means whose benefits I am not yet prepared to believe in? If I am still entertaining the possibility that a ticket will go to waste, why shouldn’t I weigh its expected benefits against those of alternative investments?10

The idea here seems to be that if we think one could intend to fly to Chicago without believing one will do so, then we have no explanation for why there would be a rational requirement to intend the means of buying a ticket, for

8  Velleman 2007, pp. 206–7. 9  As an anonymous referee correctly observed, the defender of the Strong Belief Thesis could deny that one intends to X in such cases, and insist that one only intends to try to X—a strategy which, I’ll argue in the next subsection, isn’t helpful to the cognitivist—and then insist that it’s still the case that the natural expression of an intention to X is “I am going to X.” However, this shows that the argument from Anscombe’s remark will be convincing only to those who already accept a controversial response to the alleged counterexamples. 10  Velleman 2007, p. 205.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

158  Instrumental Rationality doing so should be viewed by the agent as a potential waste of money given his agnosticism about whether he’ll go. Velleman might be right that there would be no categorical requirement for the agnostic traveler to intend to buy the ticket. But it doesn’t follow from this that Means–Ends Coherence doesn’t apply to the agnostic traveler. Means–Ends Coherence is a wide-scope requirement; what it requires is that one either intend the means, or abandon the end, or give up one’s instrumental belief. It doesn’t follow from one’s not being required to intend the means that this disjunctive requirement isn’t in place. In Chapter 3, we discussed one motivation for the wide-scope formulation of means–ends coherence: it often happens that we recognize an end of ours as imprudent, immoral, or otherwise unreasonable. In such cases, it seems perfectly rational for an agent to give up his end. But if we understood means–ends coherence as a narrow-scope requirement—a categorical requirement to intend the means, as Velleman puts it—we would have to concede that someone who abandons an end instead of intending the means violates this rational requirement.11 But that seems to be the wrong result; abandoning the end instead could involve no irrationality whatsoever. And Velleman’s example seems to be precisely the kind of case that motivates the wide-scope formulation in this way. In being agnostic about one’s success in getting to Chicago, and seeing the significant expense involved in buying the ticket, the agent would be perfectly rational in giving up the end instead of intending the means. There are two points to make about Velleman’s argument. First, even in cases where one isn’t agnostic, but instead believes one will carry out what one intends, it could be rational for one to abandon one’s end instead. For instance, one could realize that the expense involved in intending the means is so great that even if one achieves one’s end, it isn’t worth it. In such cases, it would be false to say that there’s a categorical requirement to intend the means. So, insisting on the Strong Belief Thesis won’t be enough to block this allegedly bad result. Secondly, and more importantly, this allegedly bad result isn’t actually a bad result. There’s no cost in our having to say that, in Velleman’s example, there’s no narrow-scope requirement to intend the means. There would be a cost in saying that there’s no applicable requirement of means–ends coherence. But it doesn’t follow from there being no narrow-scope requirement to

11  See Chapter 3, Section 3.2.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  159 intend the means that the wide-scope requirement, Means–Ends Coherence, is not in place.12 Velleman offers a third argument for the Strong Belief Thesis: if the Strong Belief Thesis were not true, then we would be unable to explain how intentions function so as to coordinate behavior. Velleman writes: When an intention coordinates behavior, the agent and his associates proceed on the assumption of its being executed – which would be an odd way to proceed if the agent himself were agnostic on the question. If I am agnostic as to whether I will be in Chicago, why should anyone plan or act on the assumption of my being there. And why should anyone hesitate to plan or act in ways inconsistent with that eventuality.13

In short, Velleman argues that since we already believe that intentions play a role in coordinating behavior, and the Strong Belief Thesis is necessary to explain how intentions play that role, we should believe in the Strong Belief Thesis as well. Velleman’s argument identifies two specific coordinating roles for intentions: intentions coordinate the behavior of both the agent herself and the behavior of her associates. However, it’s not clear how this latter coordinating role provides support for the Strong Belief Thesis, since what matters for such coordination is what the associates believe, not what the intending agent herself believes, about whether the intention will be successfully executed. But the Strong Belief Thesis is a thesis about what the intending agent herself believes. But, putting aside this worry for a moment, it’s not clear that when an intention coordinates behavior, one’s associates always “proceed on the assumption of its being executed,” though they may sometimes do. Suppose we are teammates on a basketball team and you intend to make a shot, and, aware of your intention, I position myself for a rebound. I’m clearly not 12  Bratman 2009b, §8, also observes Velleman’s mistake of understanding Means–Ends Coherence as a narrow-scope requirement to intend the means. In replying to Velleman’s argument, Bratman goes on to develop the idea that intentions have a distinctive aim, in much the same way that belief is sometimes thought, by Velleman and others, to have a distinctive aim. In particular, Bratman suggests that “intentions aim at the coordinated control of action that achieves what is intended” and this explains why Means–Ends Coherence is a rational requirement. We’ll return to this in Chapter 7. But we don’t need any such thesis to show that Velleman’s argument doesn’t succeed. Rather, we can simply note that a crucial premise of Velleman’s argument—namely, that if there’s no requirement to intend the means, then means–ends coherence does not apply—is false and so this argument for the Strong Belief Thesis is unsound. 13  Velleman 2007, p. 206.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

160  Instrumental Rationality proceeding on the assumption that you’ll execute your intention. Had I proceeded on that assumption, I would have hurried back down the court to set up on defense. Or suppose I’m in the stands watching the game, and someone offers me a bet on whether you’ll make the shot. I’d be irrational to proceed on the assumption you’ll successfully execute your intention. I should instead consider the probability of your doing so and calculate the expected utility of accepting the bet. And much the same goes for the agent herself. When one coordinates behavior with one’s own intentions, one need not proceed on the assumption that one’s intention will be successfully executed. The shooter might position herself for a rebound. And she might not bet the farm on making the shot. So, it’s a mistake to characterize the coordinating role of intentions merely in terms of how agents (the actor herself or her associates) plan on the assumption that the intention will be successfully executed. Sometimes we do this. But sometimes we don’t, and we instead consider the probability that the intention will be successfully executed. So, with this in mind, we should return to our question: is the Strong Belief Thesis needed to explain the co­ord­in­at­ing role of intentions? Now it seems as though some weaker thesis about the intention–belief connection might actually do a better job of explaining the coordinating role. Consider, for instance, Robert Audi’s view of intention. Audi doesn’t accept the Strong Belief Thesis, but thinks that intending to X involves believing that X-ing is more likely than not.14 (According to Audi’s view, both Shaquille O’Neal and Wilt Chamberlain intended to make free throws—barely— whereas Ben Wallace hoped, but did not intend, to make them—assuming all were aware of their appalling free throw percentages.15) On this view, sometimes one will intend to X and believe one will X, whereas other times one will intend to X and merely believe the probability of one’s X-ing is greater than (.5). Audi’s view—which I’m not endorsing here—seems to do a better job explaining the coordinating role of intentions than the Strong Belief Thesis does. Since it allows that sometimes one intends to X and believes one will X, it can account for all those cases in which we plan on the assumption that one 14 Audi 1973. 15  Wallace shot 42% from the line for his career, whereas Wilt shot 51% and Shaq 53%. Mele 1992 uses the example of a free throw shooter to argue against Audi’s view. See especially pp. 131–2; 136–7. If we look at certain commonly accepted functions of intentions—specifically, the functions of initiating and guiding an agent’s actions, coordinating behavior, terminating practical reasoning, etc.—then we won’t notice a difference between, say, O’Neal and Wallace shooting a free throw. So, it seems odd to say that one intends and the other doesn’t.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  161 will X. But since it also allows that one can intend to X without believing one will X, it can also account for some cases where one plans in light of the probability that one will X. So, it turns out that Velleman is wrong, and the Strong Belief Thesis isn’t needed to explain the coordinating role of intentions.16 Once we consider that not all coordination involves proceeding on the assumption of successful execution, this argument for the Strong Belief Thesis fails. Of course, Velleman might deny that these really are intentions. He might say that in cases where the behavior of oneself or others is coordinated based on some assessment of probabilities, rather than on an assumption of successful execution, we’re not dealing with an agent’s intentions—maybe only intentions to try, or intentions in the “goal-state” sense of the term.17 One odd thing about this possible response, which we noted earlier, is that whether something counts as an intention to X shouldn’t depend upon whether or not someone else plans on the assumption that I will X. What does their planning have to do with my beliefs? Additionally, such a reply would employ strategies that, I’ll argue in the next subsection, are unhelpful to the cognitivist. Also, such a reply would appear to rig the argument from the start: it would be using the Strong Belief Thesis to determine what counts as an intention, thereby narrowing what counts as the relevant coordinating phenomena to be explained, and then saying that we should believe in the Strong Belief Thesis because it’s necessary to explain the relevant coordinating phenomena. It wouldn’t give us an independent reason for believing in the Strong Belief Thesis to begin with. Velleman offers a fourth argument for the Strong Belief Thesis, arguing that we can’t explain a consistency requirement on intentions unless the Strong Belief Thesis is true: But why should my intentions be subject to a requirement of consistency if I can remain cognitively uncommitted to their truth? If I am agnostic as to whether I’ll be in Chicago on Tuesday evening, why should my plans for Tuesday evening have to be consistent with my being there?18

16  Holton 2008 makes the related point that an appeal to partial, as opposed to all-out, belief would allow for a response to Velleman’s coordination argument for the Strong Belief Thesis. (Audi’s view is that intention involves an all-out belief that success is more likely than not, not a partial belief in success.) Holton argues that intentions, coupled with partial beliefs in success, can play a role in co­ord­in­ at­ing the behavior of self and others. (As Holton correctly observes, one’s informing others of one’s uncertainty of success facilitates that coordination.) 17  We’ll explore these options in more detail in Section 6.1.2. 18  Velleman 2007, p. 206.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

162  Instrumental Rationality Velleman specifically asks why he shouldn’t also make dinner plans somewhere local for Tuesday evening, knowing he won’t be able to both be in Chicago and keep the dinner reservation. The first thing to note is that we should distinguish cases of inconsistency in intentions, which is irrational, from cases of contingency planning, which need not be irrational. If I doubt I’ll succeed in going to Chicago, it may be perfectly rational to put in place a backup plan that involves dinner reservations at a local restaurant.19 We aren’t tempted to describe such cases of contingency planning as involving an intention to eat at the restaurant; rather, we would say that one intends to eat there if one stays in town. And this contingency plan isn’t inconsistent with one’s planning to be in Chicago instead. Additionally, even if such contingency planning involved inconsistency, this wouldn’t give us grounds for accepting the Strong Belief Thesis, since we engage in contingency planning even when we believe we’ll succeed in our intentions. For instance, when I intend to clear the small pond on the easy par 3 and believe I will, I may still carry an extra ball just in case I’m wrong. So, adopting the Strong Belief Thesis won’t help one avoid this supposed inconsistency in intentions. So, let’s assume we’re not dealing with a case of contingency planning, but rather a case of someone who intends to be in Chicago and intends to eat at a local restaurant, knowing he can’t do both. Velleman asks “why should my intentions be subject to a requirement of consistency if I can remain cognitively uncommitted to their truth?” There are two ways to understand this question. On one reading, Velleman is suggesting that if intentions did not involve cognitive commitment, then there would be no consistency requirement governing intentions. But since there is a consistency requirement governing intentions, it follows that intentions do involve cognitive commitment. On a second reading, Velleman is asking what explanation can be given of a consistency requirement on intentions if it’s not the cognitivist one. Start with the first reading. I don’t think Velleman is right that there wouldn’t be a consistency requirement on intentions if the Strong Belief Thesis were false. Let’s assume it’s false. Let’s assume again that Audi’s weaker view of intention is correct, so that intending to X involves believing X-ing is more likely than not. It doesn’t follow that we’re now rationally permitted to have inconsistent intentions. Consider an example. Suppose I know that Farmer Joe grows five fruits (apples, oranges, plums, pears, and peaches) and 19  Of course, one wouldn’t want to invite others to dinner without telling them this is just a contingency plan—that would risk rudeness.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  163 every day selects at random only four of those fruits to bring to the Farmer’s Market. If Audi’s view is correct, in order to intend to buy apples at the Farmer’s Market, I must believe it’s more likely than not I’ll buy apples. But that doesn’t present a barrier to my intending to buy apples; I can here intend to buy apples and believe it’s more likely than not that I will. Likewise, I can intend to buy oranges, since I believe it’s more likely than not that I’ll buy oranges. The same goes for the other fruits. But now suppose that I tell you, knowing that Joe brings only four fruits to the market, that I intend to buy apples, intend to buy oranges, intend to buy plums, intend to buy pears, and intend to buy peaches. You would rightly accuse me of irrationality in having these five intentions, while believing Joe will bring only four fruits. This example suffices to show that Velleman is wrong: even if the Strong Belief Thesis is false, one could still have intentions that are rationally criticizable for their inconsistency.20 On the second reading, Velleman is challenging his opponent to provide some other explanation of the consistency requirement on intentions if it’s not going to be the cognitivist one. This is a hard question. We’d have to provide an account of intentions that shows why having intentions whose contents are inconsistent (or inconsistent with the contents of one’s beliefs) is irrational— and, of course, do so without appealing to the involvement of beliefs. Moreover, we’d have to explain how intentions differ from other attitudes, such as desires, where having that attitude towards inconsistent contents need not involve irrationality. There is, after all, nothing irrational about both wanting to go for a smoke (because it’s pleasant) and wanting not to go for a smoke (because it’s healthy), but there is something irrational about both intending to go for a smoke and intending not to go for a smoke. We’ll take up this project in Chapter  7, since the proposed explanation of Means–Ends Coherence presented there extends to cover the consistency requirements on intentions as well. But our task here is to evaluate the cognitivist’s explanation 20  One might try to circumvent this objection by insisting that the intentions here are conditional ones: I intend to buy oranges if I can, and I intend to buy apples if I can, and so forth. It’s clear that we often do have conditional intentions. I may intend to go home early if I finish my work, or intend to go buy some more milk if there’s none in the fridge. But, as Davidson (1980a, p. 94) observed, it seems appropriate to attribute a conditional intention to an agent only if that agent has explicitly contemplated that contingency in advance. (This seems to explain our reluctance to recast all of our intentions as conditional intentions. We don’t say, for instance, that I intend to work this afternoon on the condition that my building isn’t on fire—even though it’s true I wouldn’t work this afternoon were it on fire—since I’ve never contemplated that contingency.) With this in mind, we could return to our example and make it explicit that our shopper hasn’t contemplated the contingencies in advance, and so it wouldn’t be appropriate to recast the intentions as conditional ones. It’s poor planning on the shopper’s part not to have given this some thought, but it seems possible to be a poor planner in this way. Thanks to Emma Borg for helpful discussion on this point.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

164  Instrumental Rationality of the requirements of practical rationality. And, regardless of whether you accept the non-cognitivist account I’ll give in Chapter  7, the lack of an alternative theory of the consistency requirement on intentions wouldn’t constitute a good reason for accepting the Strong Belief Thesis.

6.1.2  The Case against the Strong Belief Thesis The main argument against the Strong Belief Thesis consists of counterexamples. Here’s one: Aaron intends to pass the test while remaining agnostic on the question of whether he will, since he knows it’s a challenging test. As we’ll see, defenders of the Strong Belief Thesis will insist that the case is not correctly described: if Aaron doesn’t have the belief that he’ll pass the test, then he doesn’t intend to pass the test—perhaps he only intends to try to pass the test, or perhaps he only has the goal of passing the test. And the same could be said in reply to the common counterexamples mentioned earlier (intending to make the difficult putt, sink the half-court shot, leap across the wide gap, lift the heavy log, etc.).21 The problem with these lines of reply, I’ll argue, is that while they might save the Strong Belief Thesis, they are of no help to the cognitivist attempting to explain means–ends coherence. In other words, the Strong Belief Thesis is either false, or it’s unable to do the explanatory work the cognitivist needs it to do. Let’s start with Gilbert Harman’s suggestion that the alleged counterexamples don’t involve intentions to do the act believed to be difficult, but only intentions to try to do the act.22 For instance, on Harman’s suggestion, Aaron intends to try to pass the test, and the golfer intends to try to sink the putt. We need to be careful about what’s being claimed here. It might be plausible to think that when a speaker says, “I intend to try to X,” she’s expressing the thought that she intends to X, and the “try to” is there to express her doubts about success. But this view, of course, would be of no help to the Strong 21  Another often-discussed counterexample is Michael Bratman’s (1987, p. 32) famous case of an absent-minded cyclist who intends to stop by the bookstore on the way home, but, aware of his tendency to go on “autopilot” once on his bike, doesn’t believe that he will. He’s agnostic about whether he’ll succeed in doing what he intends to do. While the other examples I mention here all involve one having doubts about what will come of one’s actions, Bratman’s example involves one having doubts about whether one will perform the requisite actions to begin with. 22  Harman 1986, pp. 90–4. It’s less clear that this strategy helps with Bratman’s bicyclist example, mentioned in the previous footnote. Suppose we say that the bicyclist intends to try to stop by the bookstore. According to the Strong Belief Thesis, he believes he will try to stop by the bookstore. But it’s not clear that this is so. He might be agnostic about whether he will actually make an attempt, given his forgetfulness.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  165 Belief Thesis, since this view concedes that the Strong Belief Thesis is false. For Harman’s suggestion to save the Strong Belief Thesis, it must be the case that in the alleged counterexamples what one intends is to try to X—and, moreover, one has this intention to try to X without also intending to X. Some have argued against the possibility of one’s intending to try to X without also intending to X.23 But I think Harman is right to see this as a possibility. After all, it seems possible for one to try to X without intending to X. Consider this variation on an example from Kirk Ludwig. Suppose I’m sure the battery on my car is dead, but my wife is skeptical of my claim that the car won’t start. I might go outside with her and try to start the car in order to show her that it won’t start. In this case, I try to start the car, but I don’t intend to start the car.24 (Instead, I intend to demonstrate to her that it won’t start.) So, it’s possible to try to X without intending to X. And it’s then a small step to the possibility of intending to try to X without intending to X. It seems natural to think that after talking with my wife and before getting into the car, I intend to try to start the car, without intending to start the car. So, it’s possible to intend to try to X without intending to X. At least, the case for this possibility is plausible enough to grant the possibility to Harman. I think Harman’s move saves the Strong Belief Thesis, but it’s of little help to the cognitivist attempting to explain Means–Ends Coherence. Let’s suppose we follow Harman in thinking that Aaron only intends to try to pass the test. Intuitively, if Aaron were to believe that he’ll pass the test only if he intends to study, but have no intention to study, he would be criticizable as means–ends incoherent. But Aaron might think that his intending to study, while necessary for passing the test, isn’t necessary for trying to pass the test. (After all, he might recall the last time he took a test and failed it because he didn’t study. He tried and failed to pass the test, but didn’t fail to try to pass the test. So, as he sees it, intending to study isn’t necessary for trying to pass the test.) So, if Aaron’s intention is, as Harman suggests, merely an intention to try to pass the test, he is no longer criticizable as means–ends incoherent in failing to 23  See McCann 1991, p. 29. 24  Ludwig 1991, p. 262. Ludwig’s example has the car start up, against all expectations. He uses the example to establish a different point: that even when one believes X-ing is impossible, one can intentionally X. David Chan (2016, p. 166), however, uses Ludwig’s example to establish the possibility of trying to X without intending to X. Jennifer Hornsby (1995, pp. 526–7) considers a similar example, crediting it to James Thomson: Edie wants to persuade someone that a stone is too heavy to lift, and so tries to lift it. Hornsby uses the example to show that one can have a reason to try to X that’s independent of her reason for X-ing. (Edie has a reason to try to lift, but not a reason to lift.) But Hornsby also (p. 528) takes the example to illustrate the possibility of intending to try to X without intending to X: Edie, who thinks it’s impossible to lift the stone, intends to try to lift the stone (to demonstrate that impossibility) without intending to lift the stone.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

166  Instrumental Rationality intend to study. He’s no longer failing to intend means believed necessary for achieving his end. But, intuitively, this is the wrong thing to say; surely he is criticizable for being means–ends incoherent here. One might reject the coherence of my description of Aaron’s psychology— specifically my supposing that he believes his intending to study is necessary for passing but not for trying to pass. On this objection, Aaron must think that intending to study is necessary for both passing and trying to pass. But I can’t see any grounds for insisting that his beliefs must be so. Our description of the case seems to be a perfectly plausible description of someone’s psych­ology. He’s got good evidence, from past occasions, for his belief that intending to study isn’t necessary for trying to pass the test. (Maybe he’s even tried to pass tests just to show us that he can’t do so without studying, in the same way I tried to start the car just to show my wife it won’t start.) Moreover, his beliefs could be true: it could be that Aaron can’t pass without intending to study, but he can try to pass without intending to study. So, I don’t see any good grounds for rejecting the coherence of my description of Aaron’s ­psychology. And so the objection remains.25 In summary, the problem here is that there is, intuitively, a requirement of means–ends coherence that applies to Aaron. If the Strong Belief Thesis is false, the cognitivist can’t account for this requirement in the standard way (as in the sample explanation of Means–Ends Coherence given at the start of Section 6.1). But if we save the Strong Belief Thesis by saying that these cases involve only intentions to try, we still are unable to explain this requirement, since the agent, while believing that forming a certain intention (the intention to study) is necessary for achieving an end, might not believe that forming that intention is necessary for trying to achieve that end. And so what should count as a case of criticizable means–ends incoherence isn’t so counted by the cognitivist. Another way to save the Strong Belief Thesis from these counterexamples, originally suggested by Harman but developed by Velleman in Practical Reflection, is to argue that “intend” is ambiguous, and the Strong Belief Thesis holds for only one sense of “intend.” Velleman writes: 25  It might be objected that “trying to X” is ambiguous between “making an attempt at X-ing” and “doing the best one can to X” and that my objection works only on the first interpretation of “trying.” (For an important related discussion of trying in connection with the debate over cognitivism, see Archer 2018.) It’s true that I’m understanding “trying” along the lines of the first interpretation. That’s because it’s only on that understanding of “trying” that I can make sense of the possibility, brought out by the example from Ludwig, of intending to try to X without intending to X. More importantly, the second understanding of “trying” wouldn’t save the Strong Belief Thesis from counterexamples. I can intend to do the best I can to X while being agnostic about whether I will in fact do the best I can to X. (Suppose X-ing is some complex and difficult task.)

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  167 The words ‘intention’ and ‘intend’ are thus ambiguous. They are used to denote, on the one hand, the agent’s attitude toward outcomes that are settled, from his perspective, at the close of deliberation and, on the other hand, his attitude toward outcomes whose pursuit is the topic of his deliberation but whose attainment is not thereby settled. In other words, they are used to denote both plan-states and goal-states of the agent.26

On this view, Aaron would have the goal of, but not plan on, passing the test, since the outcome isn’t “settled from his perspective.” (And the same would go for our golfer, basketball player, leaper, and log-lifter.) But since the Strong Belief Thesis is restricted to the sense of “intend” denoting “plan-states,” such examples are not counterexamples to the thesis. I’m not sure whether “intend” is ambiguous in this way, nor whether, if it is, these are the only two senses. But this strategy, while saving the Strong Belief Thesis, will be of no help to the cognitivist. Even if we concede that Aaron, due to doubts about success, doesn’t intend (in the “plan-state” sense) to pass, this doesn’t let him off the hook so far as means–ends coherence goes. He would still be irrational if he didn’t intend to study while believing studying to be necessary. And the cognitivist would have to explain that irrationality. In short, you don’t get off the hook when it comes to means–ends irrationality simply because you have doubts about whether you’ll succeed in some difficult attempt. Worrying about obstacles to your success doesn’t give you license to undermine yourself by failing to intend means believed necessary. In summary, although the Strong Belief Thesis might be a philosophically respectable but controversial thesis, I think we should be much less confident that it could both be true and successfully employed in a cognitivist ex­plan­ ation of means–ends coherence, since those maneuvers needed to save it from counterexamples (saying one merely intends to try, or that “intend” is am­bigu­ous) make it unsuitable to play the necessary role in the cognitivist explanation.

6.2  Cognitivism without the Strong Belief Thesis Perhaps the cognitivist could account for Means–Ends Coherence without relying on the Strong Belief Thesis. I’ll start with a view proposed by R. Jay Wallace, and consider some difficulties it faces. I’ll then introduce another, 26  Velleman 1989, p. 112. See also Harman 1986, pp. 93–4.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

168  Instrumental Rationality related view that can avoid these difficulties. This latter view is, I think, the best available cognitivist account of Means–Ends Coherence.

6.2.1  Towards a Cognitivist Account without Strong Belief Wallace’s account relies on a weaker thesis about the connection between intention and belief: intending to X involves believing X is possible. On this thesis, the person who intends to travel to New York will believe: (4)  It is possible for me to travel to New York. Wallace assumes that his instrumental belief takes the following form: (5)  It is possible for me to travel to New York only if I intend to buy an airplane ticket. Now suppose that the means–ends incoherent agent, who doesn’t intend to buy the ticket, is “minimally self-aware” and so believes: (6)  It is not the case that I intend to buy an airplane ticket. In that case, the means–ends incoherent agent, Wallace argues, would have inconsistent beliefs. He argues that if you are means–ends incoherent, you will be left in effect with the following incoherent set of beliefs (assuming you are minimally self-aware): the belief that it is possible that you do x, the belief that it is possible that you do x only if you also intend to do y, and the belief that you do not intend to do y. The incoherence of these beliefs is a straightforward function of the logical relationship among their contents, suggesting that the normative force of the instrumental principle can be traced to independent rational constraints on your beliefs – in particular, to constraints on certain combinations of beliefs . . .27

This cognitivist explanation of Means–Ends Coherence doesn’t require the Strong Belief Thesis. 27  Wallace 2006, p. 106.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  169 However, one might have worries about this weaker thesis as well. Couldn’t one intend to X while being agnostic about whether one can X? Suppose I intend to go shopping this evening, but I’m not sure whether the only shopping mall in town is open, and so I’m agnostic about whether I can go shopping. It’s not that I believe I can’t go shopping; I just don’t believe I can. It might be poor planning on my part to form that intention before finding out whether the mall is indeed open, but it seems possible to do so.28 Additionally, one might have some doubts about the logical form of the instrumental belief in (5), which may not be obvious. Perhaps what one believes is that the combination of traveling to New York and not intending to buy a ticket is not possible, so that the logical form would be: ~◇(N & ~T), where “N” is “I travel to New York” and “T” is “I intend to buy a ticket.” On Wallace’s view, the inconsistency of the beliefs is “a straightforward function of the logical relationship among their contents” which he takes to be: (4’)  ◇N (5’)  ◇N → T (6’)  ~T.

But if the logical relationship among their contents were instead (4’)  ◇N (5’’)  ~◇ (N & ~T) (6’)  ~T

then the contents of the involved beliefs wouldn’t be inconsistent. (To see that there’s no inconsistency here, consider another example: one might consistently believe Hank isn’t going to Nashville, isn’t going to Tennessee, can go to Nashville, can go to Tennessee, but can’t go to Nashville and not go to Tennessee.)29 28  One possible way around this objection would be to again insist that the intention here is a conditional one: I intend to go shopping this evening if I can. (On Wallace’s thesis, it would follow that I believe that I can go shopping if I can—which, it seems, would be a belief in a trivial truth.) As I noted in note 20 (this chapter), it seems appropriate to attribute conditional intentions only when the agent has contemplated the contingency in advance, and we can simply assume that this hasn’t happened in this example. 29  Note that it won’t help Wallace to claim that intending to travel to New York this afternoon involves believing (4’’) ◇ (N & ~T). That would indeed generate an inconsistency in belief with (5’’), ~◇ (N & ~T), but that would make one’s belief (6’) irrelevant to the inconsistency in belief. One’s not

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

170  Instrumental Rationality Here’s a reason for thinking Wallace has misidentified the logical form of the instrumental belief. Suppose he is right that the logical form of the instrumental belief is ◇N → T, the contrapositive of which is ~T → ~◇ N. Possibility and necessity are defined in terms of one another, such that ~◇N → □~N, and so it follows from these two claims that ~T → □~N. But this seems wrong. This reading licenses us to detach the claim that □~N via modus ponens from ~T. But what I should conclude is that I’m not travelling to New York this afternoon, not that it’s necessary that I’m not travelling to New York this afternoon. I don’t take my not travelling to New York this afternoon to be necessary; rather, I only take it to be necessary that it’s not the case that I both travel to New York this afternoon and not buy a ticket. So, it seems better to understand the logical form of one’s instrumental belief as □ (~T → ~N), instead of ~T → □~N.30 Perhaps there’s a way to defend some specific conception of possibility that allows us to avoid these two objections to Wallace’s account.31 But it might be easier to develop a version of cognitivism that preserves the central insights of Wallace’s approach, but avoids the difficulties that come with talk of possibility. I’ll here outline such an account.32 First, let’s follow Wallace in assuming that the instrumentally incoherent agent who is minimally self-aware will believe: intending the means, and so believing one doesn’t intend the means, wouldn’t matter—you’ve got an inconsistency regardless. But, if we’re trying to explain Means–Ends Coherence, whether one intends the means should matter. 30  See Hughes and Cresswell 1996, pp. 14–16. The sense of “necessary” in my instrumental belief here shouldn’t be understood as involving logical, metaphysical, or natural necessity. After all, I know it’s perfectly possible that some Air Force pilot could offer me a free ride to New York at the last mi­nute. I just can’t rely on such outlandish possibilities in my practical reasoning. 31  We might understand the possibility involved here as epistemic possibility, so that our subject believes, for instance, in (4), that for all he knows, it’s possible that he travels to New York. (Thanks to Mike Titelbaum for this suggestion.) Since it seems mere belief, not knowledge, is relevant to the alleged inconsistency in (4)–(6), the relevant sense of possibility would be doxastic possibility, which David Chalmers (2011, pp. 62–3) helpfully defines as follows: “A scenario [a ‘maximally specific way the world might be’] is doxastically possible for a subject if and only if it is not doxastically ruled out by any of the subject’s beliefs.” This might circumvent our two objections. First, it’s plausible that our agent believes that going shopping isn’t doxastically ruled out by any of his beliefs. Second, the detachment of □~N isn’t as worrisome, since this should be understood as stating that N is ruled out by some of the subject’s beliefs. However, one might worry that once we establish this as the relevant sense of “possibility” at work, we should worry about Wallace’s intention–belief thesis. Is it true that one can intend to X only if one believes X is doxastically possible? Perhaps not. Indeed, cases in which one is aware of one’s own instrumental irrationality—specifically, when one is aware that one doesn’t, but must, intend the means, if one is to X—might be cases where a subject doesn’t believe X-ing is doxastically possible. Wallace’s thesis now appears false. Moreover, if Wallace’s thesis were true, we’d have to say that the subject here doesn’t really intend to X, and so there is no instrumental irrationality for him to be aware of. That’s implausible. In short, the thesis that intending to X involves believing X is possible seems plausible to us, if it does, when we have in mind some other notion of possibility—perhaps logical, metaphysical, or natural possibility—but if we specify that what we mean by the thesis is that “intending to X involves believing X isn’t ruled out by one’s other beliefs” then the thesis becomes less plausible, and if true, threatens to make awareness of instrumental irrationality impossible. 32 This account is a slight variation on the account that’s presented in §3.4.3 of Kolodny and Brunero 2013.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  171 (7)  I do not intend to M. Second, let’s assume that the agent’s instrumental belief takes the following form: (8)  If I do not intend to M, I certainly will not E. If we apply Closure, then rationality requires that he believe: (9)  I certainly will not E, or revise his other beliefs. Third, let’s assume a thesis about the connection between intention and belief that’s far weaker than any we’ve discussed so far: Very Weak Belief Thesis:  Intending to X involves not believing that one certainly will not X. According to this thesis, someone who believes (9) does not intend to E. On this account, whenever one is means–ends incoherent, one has beliefs that violate Closure. Moreover, if a means–ends incoherent agent were to come to comply with Closure, he would make revisions that would also remove him from his state of means–ends incoherence. We’re assuming that the means–ends incoherent agent believes both (7) and (8). So, he should, according to Closure, either come to believe (9), which, by the Very Weak Belief Thesis, will involve his not intending to E (thereby removing him from a state of means–ends incoherence), or give up his instrumental belief (8) (thereby removing him from a state of means–ends incoherence) or give up his belief (7), which, we’ll assume for now, will involve him forming the intention to M (thereby removing him from a state of means–ends incoherence). This account differs from Wallace’s account in at least two ways. First, instead of holding that intentions involve having a certain belief, it holds that intentions involve not having a certain belief. Second, like the account we considered in Section  6.1, it employs Closure, rather than a consistency requirement on beliefs. I think this account is preferable since it better avoids the two objections to Wallace’s account. (First, while it seems the shopper can intend to go shopping while being agnostic about whether he can, it’s much less plausible that he can intend to go shopping while believing he certainly won’t. Second, the detachment of (9) via modus ponens isn’t as implausible as the detachment of □~N, since (9) is compatible with the possibility of E-ing.)

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

172  Instrumental Rationality But of course this is not to say that there won’t be reasonable objections to the components of this account. For one thing, it’s not clear that even the Very Weak Belief Thesis is true.33 For another, both this account and Wallace’s account face an obvious objection: what if someone were to falsely believe she intends the means? Such a person, it appears, wouldn’t violate any theoretical requirements of rationality, but would be instrumentally incoherent in not intending the means. Let’s turn to that objection.

6.2.2  Unknown Failures to Intend Suppose I never notice that I don’t intend to buy a ticket to New York. Or suppose I falsely believe instead that I do intend to buy a ticket. I violate Means–Ends Coherence, but yet I don’t have the relevant belief that would allow for either of these cognitivist explanations of Means–Ends Coherence to get off the ground. It won’t help to deny the possibility of such ignorance and false beliefs. Nor could we easily claim such a person would be irrational, if we understand rationality as a matter of coherence among one’s attitudes, since an agent’s being unaware of, or having false beliefs about, the absence of an intention need not involve any incoherence. Perhaps the cognitivist’s best bet is simply to concede that one isn’t instrumentally irrational in such cases. In other words, someone can get off the hook, so far as the charge of instrumental irrationality goes, just by being ignorant of, or having false beliefs about, her not intending the means. This response might not be as far-fetched as it sounds. After all, we can also get off the hook by being ignorant of, or having false beliefs about, the necessary means to our ends. Additionally, precise formulations of the requirement will aim to take into account rational self-trust. Recall Setiya’s match-striker example from Chapter 1.34 I intend to start a fire and know that to do so I must strike a match, but I don’t now intend to do so, since I trust that I will form that intention later, once the wood and kindling are in place. It doesn’t seem right to accuse me of irrationality. To get off the hook, so far as the charge of instrumental irrationality goes, I can form some (perhaps false) beliefs about my future intentions. But, if we allow that false beliefs about one’s future intentions can get one off the hook, why not also allow that false beliefs about one’s current intentions can get one off the hook as well? 33  Anscombe (1957, p. 94) argues against it by giving several purported counterexamples. See also Holton (2008, p. 54), who is convinced by Anscombe’s counterexamples. Nonetheless, it is the least controversial cognitivist thesis concerning the intention–belief connection we’ve considered here. 34  Setiya 2007, p. 668.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  173 Perhaps this response to the objection isn’t convincing. If it isn’t, I don’t think it poses any threat to the view being defended here—that is, the view that this cognitivist account of Means–Ends Coherence is better than any other available cognitivist account, especially accounts that employ the Strong Belief Thesis—since those accounts will also face a version of this same objection. Recall the account from Section 6.1. Couldn’t I come to believe (3)—that I intend to M—without intending to M? This objection is just as much an objection to the account of Means–Ends Coherence based off the Strong Belief Thesis. And so it doesn’t challenge my contention that the account sketched in Section 6.2.1 is the best available cognitivist account of Means–Ends Coherence.

6.3  Problems for Cognitivism’s Explanatory Claim But I don’t think the best available cognitivist account is good enough. It might be true that whenever one is means–ends incoherent, one also has beliefs that violate requirements of theoretical rationality. But cognitivism makes a further claim: those theoretical requirements explain the practical ones. However, one might doubt this explanatory claim.35 One might think that even if every violation of Means–Ends Coherence involved a violation of requirements of theoretical rationality, the latter requirements don’t explain the former. I’ll give two reasons for doubting the explanatory claim. First, note that once we reject the Strong Belief Thesis, cognitivism doesn’t seem to be a plausible strategy for accounting for the closely related consistency requirement on intentions. Weaker views about the intention–belief connection seem unable to do the trick. For instance, as Bratman observes, if we work with Wallace’s view that intending to X involves believing it is pos­ sible to X, we can’t explain what’s wrong with intending to X and intending not to X since the associated beliefs (it is possible to X; it is possible not to X) are perfectly consistent with one another.36 If we took Audi’s view of 35  The explanatory claim tends to receive much less attention than the cognitivist project of showing that means–ends incoherence guarantees theoretical irrationality. One notable exception to this is Lord, 2018. 36  See Bratman 2009b, §3 and Ross 2009, p. 245. There might be a way around this worry. Suppose that intentions agglomerate in the sense that if I intend to X and intend to not X, I also intend to X and not X. By Wallace’s view of the intention–belief connection, it follows that I believe it’s possible that I X and not X. But that’s inconsistent with my belief that I can’t both X and not X. And so I’ve got inconsistent beliefs. The agglomeration thesis here, however, is very controversial—much more controversial than Bratman’s claim that such agglomeration is rationally required. (As Bratman  2009b, §3, observes, appealing to any such rational requirement of intention agglomeration here would be of little help to the cognitivist since that rational requirement of intention agglomeration isn’t itself given a

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

174  Instrumental Rationality intentions, the associated beliefs would be inconsistent (X-ing is more likely than not X-ing; not X-ing is more likely than X-ing) but such a view doesn’t help once we expand our stock of intentions. It’s irrational to intend to A, intend to B, and intend to C, while believing that if one As and Bs, one will not C. But the associated beliefs (A-ing is more likely than not A-ing; B-ing is more likely than not B-ing; C-ing is more likely than not C-ing) need not be inconsistent. (If the odds of A-ing and the odds of B-ing are each only slightly better than a coin toss, then it wouldn’t be more likely than not that one both As and Bs, and so it could still be that C-ing is more likely than not C-ing.) One might wonder what other explanation can be given of the consistency requirement, and whether that same explanation could then be given for Means–Ends Coherence. If we can find another theory that can explain both of these requirements of practical rationality, we would have reason to prefer that theory to cognitivism.37 But there is a second, more serious, worry about the cognitivist’s explanatory claim. The worry is that the cognitivist’s explanation appears to work only because she considers some applicable theoretical requirements while ignoring others. Once we consider all the applicable theoretical requirements, it no longer appears plausible to think that Means–Ends Coherence is explained by the requirements of theoretical rationality. Consider someone who violates Closure by believing P, believing P→Q, but not believing Q. When one notices one holds this irrational combination of attitudes, one should look to one’s reasons for belief in order to determine the appropriate direction of response. Let’s assume that P→Q is a fixed background belief. If one thinks there are really strong reasons for believing P, and relatively weak reasons for not believing Q, then one should escape this irrational state by coming to believe Q. However, if one thinks there are strong reasons not to believe Q, and relatively weak reasons for believing P, one should revise one’s cognitivist explanation. We would no longer have what Bratman calls “pure cognitivism.”) What the thesis here says is that, as a matter of psychological fact, the objects of separate intentions also form the compound object of a single intention. One might have worries about this. If I intend to write a book, and I intend to order soup for lunch, do I also now intend to write a book and order soup for lunch? Adding this intention to my psychological set doesn’t help explain anything that couldn’t be just as well explained by the two individual intentions alone. Also, at the extreme, this view would require that we take the objects of all of my intentions and insist that I also have a single intention to pursue a very large compound object! I don’t see what would be gained by complicating the description of my psychology in this way. Thanks to Severin Schroeder for helpful discussion on this point. 37  I don’t think there are sufficient grounds for assuming every requirement of rationality is going to be explained in the same way. But I do think it would be odd if our explanation of the consistency and means–ends coherence requirements on intention turned out to be radically different. I’ll provide, in Chapter 7, an explanation that covers both.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Belief  175 beliefs in the modus tollens direction instead, and cease to believe P. In short, one’s assessment of the relevant reasons for belief determines the appropriate direction of revision.38 But now consider our means–ends incoherent agent who, on our cognitivist account, believes (7) he doesn’t intend to buy an airplane ticket, and believes (8) if he doesn’t intend to buy an airplane ticket, he certainly will not travel to  New York, but does not believe (9) that he certainly will not travel to New York. Let’s suppose that he considers the relevant reasons for belief. Presumably, the evidence that he doesn’t intend to buy an airplane ticket will be rather strong. Introspection isn’t foolproof, but it will normally provide very strong evidence that he doesn’t intend to buy an airplane ticket—and there doesn’t seem to be any evidence on the other side. Taking into account the requirements of theoretical rationality governing the direction of revision, it seems that in all but the most unusual cases, revision should proceed in only one direction: coming to believe he certainly will not travel to New York (thereby ceasing to intend to travel to New York). Cognitivism thus predicts an asymmetry in the direction of response.39 But, as Means–Ends Coherence is usually understood, there is no such asymmetry. (Indeed, if there’s any temptation toward thinking there’s an asymmetry, it would be toward proceeding in the other direction: coming to intend the means.40) Rather, one could remove oneself from a state of means–ends incoherence by either abandoning the end, or by coming to intend the means. This symmetrical feature of Means–Ends Coherence is unexplained by the cognitivist account, and so the cognitivist account is inadequate. In summary, when we consider the requirements of theoretical rationality in an appropriately holistic way—that is, we don’t select one requirement and ignore how it interacts with other relevant rational requirements—then we see that the requirements of theoretical rationality offer a poor explanation of Means–Ends Coherence. Theoretical rationality requires that we respond to 38  I’ll here avoid the complicated task of specifying the principles of rationality governing such revision. But I do think such principles are principles of rationality, not principles of reason. One’s assessment of the reasons could be mistaken. For instance, it could be that the person who thinks there are strong reasons to believe P is mistaken, and there are really strong reasons to believe ~Q instead. Here, rationality requires him to revise his attitudes in such a way that he comes to have a belief not well-supported by reasons. (A similar point could be made for the practical case.) 39  As we noted in Chapter 3, Section 3.3, such asymmetries have played an important role in arguments against the wide-scope formulation of rational requirements. For instance, some have argued that such asymmetries put pressure on us to construe the enkratic requirement as a narrow-scope requirement to intend to do what we believe we ought to do. If cognitivism is true, we should expect to feel the same pressure toward construing means–ends coherence as a narrow-scope requirement (to abandon the end). That we don’t feel such pressure is a reason to think cognitivism is false. 40  See Chapter 3, Section 3.3.1.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

176  Instrumental Rationality states of incoherence by revising in light of our assessment of the relevant reasons for belief, which we take to be provided by the evidence, while prac­tical rationality requires that we respond to states of incoherence by revising in light of our assessment of the relevant reasons for action. These rational requirements may not push in the same direction. Specifically, as I’ve argued, the theoretical requirements issue the false prediction that we should almost always respond to instrumental incoherence by giving up on our ends. So, even if the cognitivist can establish that violations of Means–Ends Coherence involve violations of theoretical rationality, she hasn’t yet established that the requirements of theoretical rationality explain Means–Ends Coherence.

6.4 Conclusion In this chapter, I’ve considered the merits of some possible ways of developing a cognitivist account of Means–Ends Coherence. I’ve argued that we should reject the account which relies on the Strong Belief Thesis, since that thesis is either false or unable to do the explanatory work the cognitivist needs it to do, and the arguments for that thesis aren’t convincing. I’ve argued that the cognitivist would do better to work with an account modeled on Wallace’s account, but which employs a weaker belief thesis, avoids talk of possibility and necessity, and employs Closure instead of a consistency requirement. But it’s worth remembering that both accounts struggle when it comes to establishing that means–ends incoherence guarantees theoretical irrationality. On the account I sketched which employs the Strong Belief Thesis, it could be that one falsely but rationally believes (3), in which case you’d have means–ends incoherence without theoretical irrationality. And on the account I sketched which employs the Very Weak Belief Thesis, it could be that one doesn’t intend the means, but doesn’t believe (7), in which case you’d again have means–ends incoherence without theoretical irrationality. But even if we resolved these worries, and means–ends incoherence guaranteed theoretical irrationality, we still would need to establish that the theoretical requirements explain the practical ones. And I’ve argued that we should be worried about the cognitivist’s explanatory thesis. The best available cognitivist account of Means–Ends Coherence leaves the consistency requirement on intentions unexplained, and the explanation it gives of Means–Ends Coherence appears to work only because it considers some theoretical requirements in abstraction from others. In light of these difficulties, I don’t think we should look to the requirements of theoretical rationality to explain Means–Ends Coherence.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

7 Intention In the previous chapter, we considered attempts to explain some of the requirements of consistency and coherence governing intentions by appealing to the requirements of consistency and coherence governing beliefs. We’ll now consider a different approach. We’ll consider whether we can give an explanation of these practical requirements that parallels the explanation we give of the theoretical requirements. (That will require, among other things, that we take up the project of considering how the theoretical requirements themselves are explained.) The resulting picture will be a non-cognitivist one in that the practical requirements are not explained by the theoretical ones. I’ll start by outlining the basic approach, the central components of which will be explained in greater detail in the remainder of the chapter. Here’s the main idea: A belief is an attitude with a constitutive aim: truth. It is part of the “job description” of belief to correctly represent how things are. We can use this constitutive aim to explain rational requirements governing beliefs. For instance, if you have inconsistent beliefs—suppose you believe that P and believe that ~P—then the logical relations among the contents of your attitudes ensures that one of these beliefs fails to satisfy belief ’s constitutive aim: either your belief that P is false or your belief that ~P is false. We can give a parallel explanation for the consistency requirement governing intentions. An intention is an attitude with a constitutive aim, one that I’ll call the aim of “controlled action.” I’ll explain what this means later, but we can note here that one way to fail to achieve controlled action is to fail to act at all. If you have inconsistent intentions—suppose you intend to X and intend to ~X—then the logical relations among the contents of your attitudes ensures that one of your intentions will fail to achieve its constitutive aim: either you won’t X or you won’t ~X.1 By appealing to the constitutive aims of both belief and intention, we can explain rational requirements, like intention–belief consistency and means–ends coherence, which govern combinations of beliefs and intentions. Suppose you 1  I’m here working with an expansive understanding of “action” such that both X-ing and ~X-ing would count as actions. Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

178  Instrumental Rationality violate intention–belief consistency by intending to travel to San Francisco this afternoon, intending to travel to New York this afternoon, and believing that if you travel to New York this afternoon, you won’t travel to San Francisco this afternoon. You have a combination of attitudes the logical structure of which guarantees a constitutive aim will not be met: either your belief is false, or you have an intention that won’t issue in controlled action. Likewise, suppose you violate means–ends coherence by intending to take the train to Charleston tomorrow, believing that to do so you must intend to buy a ticket, but not intending to buy a ticket. In this case, either your instrumental belief is false, or you won’t achieve your aim of taking the train to Charleston. The view here resembles the Normative Disjunctivist approach to means–ends coherence from Chapter  4 in that, roughly speaking, on both views, when you violate means–ends coherence, some specific attitude is out of line. But whereas the Normative Disjunctivist holds that it’s assured you’ll have some belief or intention you ought not have (or don’t have an intention you ought to have), this view makes no claims about the attitudes one ought, or ought not, have. So, I’ll call this view “Non-normative Disjunctivism.” On this view, it’s assured you’ll either have some intention that won’t achieve its constitutive aim or have some belief that won’t achieve its constitutive aim. An appeal to constitutive aims offers us a promising way to explain the “normativity” (in the broader sense) of these rational requirements of consistency and coherence governing beliefs and intentions. Very roughly, the normativity of the requirements governing beliefs and intentions is explained by these attitudes themselves—in particular, by their constitutive aims. That’s a rough sketch of the basic idea. The remainder of this chapter will be devoted to filling out the two main components of it. First, in Section 7.1, I’ll consider in what sense there is an “aim of intention.” The suggestion that there is an aim of intention analogous to the aim of belief, and that this aim of intention can help explain requirements of consistency and coherence governing intentions, originates in the work of Michael Bratman, but is only given a very brief treatment in two related papers primarily devoted to criticizing the cognitivist approaches to intention consistency and means–ends coherence.2 In those two papers, Bratman suggests that just as beliefs aim at truth, intentions aim at coordinated, effective control of action. Since Bratman’s presentation of the idea is very brief, and it hasn’t received any critical discussion, I’ll spend some time considering how it might be developed. I’ll focus in particular on 2  Bratman 2009a, pp. 23–7, and Bratman 2009b, pp. 48–54.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  179 two issues. First, not all of the ideas collected under the umbrella of the slogan “belief aims at truth” would be promising avenues for developing Bratman’s suggestion that intentions analogously aim at coordinated, effective control of action. So, I will try (Section 7.1.1) to isolate the way in which there can be an “aim of intention” analogous to the aim of belief. Second, I’ll focus (Section 7.1.2) on, and criticize, Bratman’s particular conception of the aim of intention, and also his particular account of how that aim explains the rational requirements of consistency and means–ends coherence governing intentions. I’ll then say more (Section 7.1.3) about how we should understand the aim of intention. (Obviously, many philosophers besides Bratman have written about the nature and aims of intention and intentional action, and so the focus on Bratman’s particular suggestion in this section might appear to be unduly narrow.3 There are two justifications for the narrow focus. First, exploring the advantages and disadvantages of the wide variety of accounts of intention available in the action theory literature would be difficult, if not impossible, in a chapter of this length. That might require a book in itself. Second, Bratman’s approach, despite my disagreement with its details below, strikes me as the most promising one for developing an account of means–ends coherence in terms of constitutive aims, such that I think it’s worth spending time seeing how the approach might be developed in some detail, even if that comes at the expense of considering rival approaches.) Second, in Section  7.2, I’ll look more carefully at Non-normative Disjunctivism and show (Section  7.2.1) how it has several advantages over rival accounts, and can avoid all of the defects of Normative Disjunctivism. I’ll then discuss (Section 7.2.2) another advantage of the view: this approach to means–ends coherence promises (as suggested above) to explain several other closely related rational requirements, such as intention consistency and intention–belief consistency. There are, however, limitations on how far the approach extends. It won’t cover, for instance, the rational prohibitions on intransitive preferences and akrasia. It would be wonderful if we had a grand theory that would explain all of these requirements in the same way. But in the absence of such a theory, we can rest content with a view that explains means–ends coherence and several closely related requirements in the same way. Yet another advantage of this general approach—the approach of looking 3  Other well-known accounts of the aim of intention and action include Velleman’s account (2000a, p. 22), according to which the constitutive aim of action is knowledge of what one is doing, and Wedgwood’s account, according to which practical reasoning aims at choiceworthiness in action (2007, pp. 101, 104).

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

180  Instrumental Rationality to constitutive aims to explain rational requirements—is that even if it turns out that rational requirements are not (weakly) normative, we need not slide into a myth theory about them. I’ll explain (Section 7.2.4) why there would be  a temptation in the direction of a myth theory, and how appealing to constitutive aims allows an avenue of resistance.

7.1  The Aim of Intention Bratman notes that there are several aspects of the idea that “belief aims at truth.” First, there is a descriptive thesis that beliefs tend to track the truth—a feature of belief that helps us distinguish it from other attitudes. Second, there is a normative thesis that beliefs are criticizable if they are false. (As Bratman notes, David Velleman once held that this second, normative thesis is de­riv­ able from the first, descriptive thesis, but in more recent work Velleman has walked back that claim.4) Third, there is the idea, found in Bernard Williams’s classic discussion of the aim of belief in “Deciding to Believe,” that, as Bratman puts it, “part of the explanation of how, for agents like us, beliefs track the truth will appeal to an internal norm that assesses beliefs in terms of their truth.”5 Bratman includes all three of these aspects under the idea that “beliefs aim at truth”—an idea which, he notes, could also be expressed by saying that truth is a “constitutive aim” of belief. Bratman tells us that cognitivists about practical rationality will likely pursue a two-part strategy along the following lines: first, they will explain prac­tical requirements of rationality, like means–ends coherence and intention consistency, in terms of requirements of theoretical rationality—this is what we considered in Chapter  6—and, second, they will explain the requirements of theoretical rationality in terms of belief ’s constitutive aim (truth). But Bratman notes the availability of another, non-cognitivist strategy: “if we are attracted to this idea that norms on belief are tied to its constitutive aim of truth, then we can argue, in a parallel fashion, that norms on intention are tied to the (or anyway, a) constitutive aim of intention.”6 Just as the cognitivist, in the second part of the two-part strategy, uses the constitutive aim of belief (truth) to explain theoretical rational requirements, the non-cognitivist, who rejects the first part of the cognitivist’s two-part

4  See Velleman 2000b, 2000c, and, for his more recent views, Shah and Velleman 2005. 5  Bratman 2009a, p. 23. 6  Bratman 2009a, p. 25.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  181 strategy, could use the constitutive aim of intention to explain practical rational requirements. Bratman goes on to fill in the details of this strategy. First, there is the idea that just as truth is the aim of belief, “coordinated, effective control of action” is the aim of intention. Second, there is the idea that this constitutive aim explains the requirements of consistency and means–ends coherence governing intentions: Each intention aims at its realization in coordination with one’s overall system of intentions. Coordination involves consistency among one’s intentions, given one’s beliefs; effective control requires that one intend means intending which one knows to be needed to achieve intended ends. So it is plausible that we can see the norms of consistency and means–ends rationality – norms characteristic of intention – as related to the (or, an) aim of intention in a way that parallels the relation between analogous norms on belief and the truth-aim of belief.7

Although I agree with Bratman’s general strategy of appealing to the constitutive aim of intention as part of an explanation of intention–belief consistency and means–ends coherence, I don’t agree with the particular way in which Bratman carries out this strategy. In Section 7.1.2 below, I’ll explain my reservations about this strategy with respect to both intention–belief consistency and means–ends coherence. But before doing so, we first need to consider (Section  7.1.1) in more detail the idea that there is an aim of intention that parallels the aim of belief. Bratman correctly notes that the slogan “belief aims at truth” is an umbrella for several ideas. But these ideas can be interpreted in various ways, not all of which would be promising avenues for developing his beliefintention ana­logy. Since I wish to take on board the strategy of appealing to an aim of intention to explain the requirements of consistency and coherence governing intention—although, again, I disagree with the particular way Bratman carries out this strategy—I should spend some time discussing in what sense there is an aim of intention analogous to the aim of belief. So, let’s start by looking at some of the ideas under the umbrella of the slogan “beliefs aim at truth.”

7  Bratman 2009a, p. 25.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

182  Instrumental Rationality

7.1.1  The Aim of Belief The metaphor of belief ’s aim is sometimes used to refer to the direction of fit of belief.8 Beliefs aim to fit the world. And this marks a contrast with desires (and intentions), which do not aim to fit the world, but to have the world fit them. (As it is often put, desires have a “world-to-mind” direction of fit, while beliefs have a “mind-to-world” direction of fit.) If you believe P and P is not the case, your belief is mistaken, while there may be nothing wrong whatsoever with your desiring that P when P is not the case. Likewise, while you are disposed to abandon your belief that P upon becoming aware of conclusive evidence that ~P, you are not disposed to abandon your desire that P upon becoming aware of conclusive evidence that ~P.9 Anscombe famously illustrates these different directions of fit with the example of a shopper buying items on his shopping list while being followed by a detective recording the items the shopper purchases. If there’s a discrepancy between the shopper’s list and what the shopper buys, then “the mistake is not in the list, but in the man’s performance” whereas if there’s a discrepancy between the detective’s record and what the shopper buys, then “the mistake is in the record.”10 But the metaphor of belief ’s aim is also sometimes understood as a normative thesis about the nature of belief. For instance, Ralph Wedgwood writes: It is often claimed that beliefs aim at the truth. Indeed, this claim has often been thought to express an essential or constitutive feature of belief. But this claim is obviously not literally true. Beliefs are not little archers armed with little bows and arrows: they do not literally “aim” at anything. The claim must be interpreted as a metaphor. I propose to interpret this claim as a normative claim – roughly, as the claim that a belief is correct if and only if the proposition believed is true.11

Wedgwood tells us that he isn’t using “correct” as a synonym for “true.” (That would almost render the thesis entirely trivial.) Rather, to say that an

8  Here, for instance, is a passage from Mark Platts 1979, pp. 256–7, which is quoted and endorsed by Michael Smith 1994, p. 112: “Beliefs aim at the true, and their being true is their fitting the world; falsity is a decisive failing in a belief, and false beliefs should be discarded; beliefs should be changed to fit the world, not vice versa. Desires aim at realization, and their realization is the world fitting with them; the fact that the indicative content of a desire is not realized in the world is not yet a failing in the desire, and not yet any reason to discard the desire; the world, crudely, should be changed to fit with our desires, not vice versa.” 9  Smith 1994, pp. 113–15. 10  Anscombe 1957, p. 56. 11  Wedgwood 2002, p. 267.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  183 attitude is correct is to say that one has “got things ‘right’; one’s mental state is ‘appropriate’,” whereas to say that an attitude is incorrect is to say that one has “got things ‘wrong’ or made a ‘mistake’. ”12 On this view, to say that a belief aims at truth is to say that a belief is correct, or appropriate, if and only if the prop­os­ition believed is true. Now, it may well be that both of these interpretations of the metaphor of belief ’s aim are correct—that is, it may well be that belief has a mind-to-world direction of fit, and that a belief is correct, or appropriate, if and only if the proposition believed is true. But I think it’s clear that we cannot take both of these ideas on board to understand Bratman’s suggestion that there’s a parallel aim of intention. Suppose, paralleling Wedgwood’s normativist interpretation of truth as the aim of belief, we said that just as truth is the correctness condition for belief, coordinated, effectively controlled action is the correctness condition for intention. The idea seems flawed for two reasons. First, there’s the point about the different directions of fit of belief and intention. It need not be a mark against an intention that it hasn’t yet resulted in the intended action. For instance, there’s nothing “defective” or “inappropriate” or “mistaken” about the items on Anscombe’s shopper’s list prior to his going shopping. So, we don’t want to make coordinated, effectively controlled action the correctness condition for intention. While there’s something incorrect about the mental state of belief when the believed proposition is not true, there need be nothing incorrect about the mental state of intention when the intended action has not yet come about. Second, even if an intention results in coordinated, effectively controlled action, that’s not sufficient for counting the intention as “correct” or “appropriate.” Suppose Kant’s poisoner effectively carries out his intention to kill someone by poisoning him. That’s hardly a good ground for calling his intention “correct” or “appropriate.” If we’re looking to understand whether the poisoner’s intentions (and actions) are correct, or appropriate, we’d instead consider, perhaps among other things, whether poisoning a person is something one ought to do.13 And consideration of this question would lead us to conclude that his intention to kill his colleagues by poisoning them is in­appro­pri­ate or incorrect.

12  Wedgwood 2002, p. 267. In the introduction to a volume on the aim of belief, Timothy Chan (2013, p. 3) classifies Wedgwood’s view as a “normativist” interpretation of the aim of belief, which is “the view that belief, by its very nature or essence, possesses the normative property of having truth as its correctness condition.” 13  A more plausible view would have the correctness condition for intention be choiceworthiness, or something similar. See Wedgwood 2007, pp. 101, 104.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

184  Instrumental Rationality For belief, it seems that direction of fit and correctness conditions are aligned: it’s plausible to say both that a belief aims to fit the world, and that a belief is correct, or appropriate, just when it fits the world. But not so for intention, at least working with something like Bratman’s suggestion that coordinated, effectively controlled action is a constitutive aim of intention. Here, direction of fit and correctness conditions are not aligned. It isn’t plaus­ ible to say both that an intention aims to have the world fit it, and that an intention is correct, or appropriate, just when the world fits it. This latter claim is false for the two reasons mentioned above. So far, we’ve discussed two ways of understanding the thesis that belief aims at truth. Some have understood the thesis as a way of characterizing belief ’s direction of fit, while others have understood the thesis as specifying the correctness conditions for belief. A third way of understanding the thesis is found in Bernard Williams’s classic discussion of the aim of belief in “Deciding to Believe.” Williams’s discussion of the thesis is brief, since it is introduced at the beginning of an extended argument against doxastic voluntarism, which is the main topic of that paper.14 Williams classifies three related ideas under the single heading “belief aims at truth.” He first notes that truth is a “dimension of assessment” of belief in that we can assess beliefs, but not many other attitudes, as true or false. Second, he notes that to believe that P is to believe that P is true. A third point, related to the second, is that “to say ‘I believe that P’ itself carries, in general, a claim that P is true.”15 Williams notes that this last point is connected to Moore’s paradox. G.E.  Moore had observed that it is paradoxical for someone to say “I believe that it’s raining, but it isn’t.” There’s nothing paradoxical when I say about my past self “I  believed yesterday that it was raining, but it wasn’t” or when I say of someone else, “She believes that it is raining, but it isn’t.” It’s only for first-­ personal assertions in the present tense that the paradox arises. I can’t coherently say, “I believe that it’s raining, but it isn’t.” If these Moore-paradoxical assertions indicate one way in which belief aims at truth, we might consider whether there are analogous paradoxical assertions regarding intention and action. Perhaps parallel to “I believe that P, but ~P” we might have “I intend to X, but will not X.” However, the latter assertion doesn’t seem paradoxical. As we noted in Chapter 6, in cases of difficult action—say, my attempting to make a difficult putt—I might say

14  Williams 1973. For an argument for the compatibility of doxastic voluntarism with the aim of belief, see Reisner 2013. 15  Williams 1973, p. 137.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  185 something like, “I intend to make it, but I won’t.” (I’m not thinking of a typical difficult putt in which I aim to put the ball close to the hole so as to reduce the likelihood of my three-putting—or, if we’re being honest, four-putting—but a case where my aim is to make the putt. Suppose I must make it to win, and there are no prizes for second place or ties.) Assertions like “I intend to make the putt, but I won’t make it” don’t come across as paradoxical in the way that assertions like “I believe I’ll make the putt, but I won’t make it” do.16 The lesson here is that we should be cautious: we shouldn’t think that in speaking of an “aim of intention” we must bring on board all of the ideas associated with philosophical discussions of the aim of belief. In particular, I think we should reject the idea that when we say that coordinated, effectively controlled action is the aim of intention we are specifying the correctness conditions for intentions. And, if we follow Williams in thinking that Moore’s paradox illustrates a way in which belief aims at truth, we shouldn’t expect to find analogous Moore-paradoxical assertions involving intentions. But it does seem that, despite being cautious, we can identify a sense in which there is an aim of intention analogous to the aim of belief. As Peter Railton notes in discussing the aim of belief, it’s the “job description” of belief to correctly represent how things are. As Railton puts it, “getting things right is what a belief presents itself as doing.”17 That’s the role of belief within one’s psychological apparatus. And we can assess whether a particular belief succeeds or fails with respect to this role: a successful belief will be one that is true, while an unsuccessful belief will be one that is false. Call this the “success condition” for beliefs. Parallel to this, Bratman could argue that the “job description” of intention is to yield coordinated, effective control of action. That’s the role of an intention within one’s psychological apparatus. And we can assess whether a particular intention succeeds or fails with respect to this role: a successful intention will issue in coordinated, effectively controlled action, while an intention would be unsuccessful, if, say, it doesn’t issue in any action at all. Call this the “success condition” for intentions. In a slogan, a successful belief gets things right and a successful intention gets things done. What support can we give to the idea that the “job description” of belief is to correctly represent how things are? David Velleman has argued that the aim of getting things right helps distinguish belief from other cognitive attitudes. On a first pass, we might say that what’s unique about belief is that believing

16  The point here is just about ordinary assertions. It may be that philosophical reflection on the nature of intention leads us to deny that one really intends to make the putt. 17  Railton 1994, p. 74.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

186  Instrumental Rationality that P involves regarding P as true. That certainly distinguishes belief from non-cognitive attitudes, like desire. One can desire that P without regarding P as true. But it doesn’t distinguish belief from other cognitive attitudes, such as assuming, or imagining. When we assume that P, we regard P as true for the purposes of advancing some argument or line of inquiry. And when we imagine that P, we regard P as true for the purpose of “stimulating or vicariously satisfying our desires.”18 So, believing that P, assuming that P, and imaging that P all involve regarding P as true. What makes belief distinctive, argues Velleman, is that in believing P, one regards P as true, not for some heuristic purpose (as with assuming) nor for some motivational purpose (as with imagining), but for the purpose of getting things right: “we believe a proposition true when we regard it as true for the sake of thereby getting the truth right with respect to that proposition: to believe something is to accept it with the aim of doing so only if it really is true.”19 However, we need not think of this aim of getting the truth right with respect to P as an agential intention (though of course an agent may have such intentions): Truth must be the aim of belief, but it need not be an aim on the part of the believer; it may instead be an aim implicit in some parts of his cognitive architecture. When his beliefs change in the face of evidence or argument, he might be described as trying to arrive at the truth, as if he were motivated by a desire. But this description might be a personification of aims that are in fact sub-personal.20

Beliefs can be described as aiming or “trying to arrive at the truth” given the ways in which we systematically form, revise, and abandon beliefs in light of indicators of truth (evidence or argument).21 This receptivity of belief to indicators of truth allows us to say that belief has a certain role to play within our psychological apparatus—the role of getting things right—a role which isn’t shared by other attitudes, even those which involve regarding a proposition as true.

18  Velleman 2000b, p. 183. 19  Velleman 2000b, pp. 183–4. Velleman elaborates on this in “The Aim of Belief,” but the point there is the same: to believe a proposition is to regard it as true with the “aim of getting its truth-value right, by regarding it as true only if it really is” (Velleman 2000c, p. 247). 20  Velleman 2000a, p. 19. See also, 2000b, p. 184 and 2000c, p. 253. 21  As Velleman notes, this is compatible with the possibility of factors other than truth influencing belief: “But my thesis is not that belief is completely shielded from mechanisms that tend to make it  false; my thesis is that belief is necessarily subject to mechanisms designed to make it true.” Velleman 2000c, p. 254.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  187 What support can we give to the idea that the “job description” of intention is coordinated, effective control of action? Bratman notes that “intentions are embedded in a planning system that tracks coordination and effective control and systematically adjusts, when need be, in their direction.”22 This systematic adjustment allows us to say that an intention is “aiming” or “trying” to bring about coordinated, effective control of action. (I’ll raise some concerns about this, especially regarding coordination, in the next subsection.) More generally, Bratman suggests that his planning theory of intention, articulated in Intentions, Plans, and Practical Reason, can provide support for attributing this constitutive aim to intention.23 In that work, Bratman had emphasized the role of intentions in, among other things, disposing one to take steps which bring one closer to action. For instance, a person who intends some end, as opposed to merely desiring that it be brought about, will be disposed to make relevant sub-plans and take facilitating steps at the appropriate times. This feature of intentions provides further support for the claim that intentions constitutively aim at coordinated, effective control of action. Talk of an “aim” of intention might seem paradoxical. But it’s important that in thinking there is a constitutive aim of intention, we need not think intentions are necessarily accompanied by some further agential intention. After all, this would threaten a regress: if the constitutive aim of my intention to do something itself involved a further agential intention, then, since that further agential intention would also have a constitutive aim, it would itself involve a further agential intention, which would also have a constitutive aim, and so on ad infinitum. But we can avoid the regress simply by rejecting the idea that the constitutive aim of intention involves a further agential intention. This would also preserve the parallel with Velleman’s account of the constitutive aim of belief, which, as we saw in the passage quoted above, need not involve an agential intention to arrive at the truth. So far, I’ve argued that it’s plausible to think that there’s an aim of intention analogous to the aim of belief provided that we develop the point in terms of the success conditions of these attitudes, which are given by the “job descriptions” of the attitudes—that is, specifications of their roles within our psy­cho­logic­al apparatus. But we can now also see an important difference between intentions and beliefs: for intentions, but not for beliefs, success conditions and correctness conditions come apart. Kant’s poisoner could successfully execute his intention to poison his victim, without that intention being a correct

22  Bratman 2009a, p. 25.

23  Bratman 2009a, p. 26. See also Bratman 1987.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

188  Instrumental Rationality intention.24 But for belief, correctness and success go together. When a belief is true, that belief is both correct, or “appropriate” (as Wedgwood puts it), and successful—that is, it’s doing its job of getting things right. (And when the belief is false, it is both incorrect, or inappropriate, and failing to do its job.) Success and correctness here go hand in hand. Of course, I’m not denying that it’s possible to have some basis for calling the intention of Kant’s poisoner “unsuccessful.” There are many standards and measures to which one could appeal in assessing some particular intention. Perhaps, for instance, by the standards of morality, it doesn’t count as a success. That’s fine. The point here is just that it is successful qua intention—that is, it succeeds by the standard presented in intention’s “job description” since it yields coordinated, effectively controlled action. So, we can admit that a particular intention can be successful qua intention, but unsuccessful by some other measure. (Likewise, in a case where someone believes something false because it is pleasing, we might say that the belief is unsuccessful qua belief—it fails by the standard presented in belief ’s “job description” since it is false—but is successful insofar as it increases happiness.25) We need not, and should not, say that what we’ve called “success conditions” captures every way in which we use the terms “success” and “failure” with regard to these attitudes. Let’s sum up the main ideas of this subsection. I’ve been looking at the broad strategy of argument according to which there’s a constitutive aim of intention analogous to the constitutive aim of belief. I’ve argued that we should be cautious, since not every idea under the umbrella of the slogan “belief aims at truth” will apply straightforwardly to the idea of an aim of intention, but also that we should be optimistic, since it seems that a parallel could be drawn with respect to the success conditions of these two attitudes.

24  Shah  2008, p. 13, makes a similar point about how correctness and success conditions come apart in this way, but he is concerned with successful action rather than successful intention. 25  The belief might also be unsuccessful qua belief but a successful means to some epistemic goal. Consider the following example from Peter Railton (1994, p. 73). Suppose a mathematician comes to believe, falsely, that a certain theorem T has been proven, and this false belief gives her confidence to keep working on developing a procedure for producing new proofs—she would ­otherwise have given up on the project—and this results in her having many new true, relevant mathematical beliefs. Her believing that T has been proven is an instrumentally valuable belief, and, moreover, it is instrumentally valuable with respect to an epistemic goal: the goal of having true, relevant beliefs. But that belief doesn’t count as a successful one, as we’re understanding it, since the belief is false: the particular belief that T has been proven presents itself as representing reality accurately but fails to do so.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  189

7.1.2  Bratman’s Account Let’s now turn to Bratman’s specific account, according to which “coordinated, effective control of action” is a constitutive aim of intention. As we saw in the passage quoted at the start of this section, Bratman uses this constitutive aim to explain both the consistency requirement on intentions and means–ends coherence. Let’s start with his explanation of means–ends coherence. The idea that intention’s constitutive aim includes effective control of action is crucial to Bratman’s explanation of means–ends coherence: “effective control requires that one intend means intending which one knows to be needed to achieve intended ends.”26 That’s what allows Bratman to say that means–ends coherence is explained by intention’s constitutive aim. However, there’s an immediate problem with the word “knows” in the above quotation. Means–ends coherence applies to those means we believe we must intend to achieve our ends, even if we don’t know we must intend them. If you falsely believe you must intend some means to achieve an end, there is a requirement in place prohibiting you from also intending that end without intending those means. So, even if Bratman is right that effective control requires that one intend known necessary means, that won’t allow us to explain means–ends coherence in its full generality, since that requirement extends to believed necessary means, even if they are not known. It might be tempting to think that this is an insignificant slip on Bratman’s part, and that we could easily fix it by replacing “known to be needed” with “believed to be needed.” However, it doesn’t seem to be a slip. In elaborating on this idea in another paper, Bratman talks throughout of needed means, rather than means believed to be needed: it is plausible to suppose that intentions aim at coordinated control of action that achieves what is intended. . . . If our intentions in favor of ends are to be effective, they normally need to function by way of our intending needed means. . . . So, if one is to achieve the characteristic aim of one’s intention – an aim in favour of coordinated control of action that is effective in the pursuit of what is intended – one needs, in such cases, to intend the means.27

Moreover, it’s easy to see why Bratman presents the point this way: while it’s true that a failure to intend means which are actually necessary would prevent

26  2009a, p. 25. See also 2009b, p. 54.

27  Bratman 2009b, p. 54, emphasis in original.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

190  Instrumental Rationality one from acting effectively, it’s hard to see how a failure to intend means which are falsely believed to be necessary would undermine the possibility of the relevant effective action. Suppose, for instance, that I intend to turn on my television and falsely believe that to do so I must first replace the batteries in my remote control. (Unbeknownst to me, my wife has already replaced the batteries. And so there’s no need for me to do so.) However, I’m means–ends incoherent in that I don’t intend, nor trust that I will later intend, to replace the batteries in my remote. Perhaps I’m absent-minded, or distracted, and so I  never form the intention to replace the batteries. But here my failure to intend means falsely believed to be necessary is no obstacle to effective action: I could pick up the clicker, point it at the television, and turn it on. So, Bratman faces a dilemma. If his account merely makes use of the claim that effective action requires that one intend known necessary means, then, although this might be true, it wouldn’t be the right kind of basis for an ex­plan­ation of means–ends coherence, which concerns means believed to be necessary, regardless of whether that belief is true or false. On the other hand, if his account makes use of the claim that effective action requires that one intend believed necessary means, it would be the right kind of basis for an explanation of means–ends coherence, but that explanation would fail since the claim is false: when one falsely believes some means is necessary, a failure to intend the means need not threaten the possibility of effective action. I can effectively turn on the television, in the example above. In fairness to Bratman, I should note that he doesn’t take himself to have given a full defense of how this constitutive aim of intention explains the rational requirement of means–ends coherence. However, it’s hard to see how this problem could be avoided. If Bratman is to use the constitutive aim of intention (particularly, effective control of action) to explain means–ends coherence in its full generality, he would have to endorse the claim that ef­fect­ive control of action is prevented when one doesn’t intend the means believed to be necessary, even when that belief is false. And that’s precisely what seems hard to believe. Looking ahead, the account given by the Non-normative Disjunctivist isn’t subject to this problem, since that account uses the constitutive aims of both belief and intention to explain means–ends coherence. On that account, the means–ends incoherent agent either has a false instrumental belief, which is a failure to achieve belief ’s constitutive aim, or won’t do what he intends to do, which is a failure to achieve intention’s constitutive aim. For instance, when I intend to turn on my television and believe I won’t do so unless I put batteries in the remote, but, incoherently, don’t intend to put batteries in the remote, either I’m wrong about the need for batteries, or I won’t

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  191 turn on my television. This view doesn’t commit us to thinking that when my instrumental belief is false, my incoherence prevents me from effectively ­controlling action. It only prevents me from effectively controlling action when my belief about the necessary means is true. (This account thus has a significant advantage over Bratman’s account.) Bratman also uses the constitutive aim of intention to account for the consistency constraint on intentions.28 Here, the crucial component of intention’s constitutive aim is the idea that intention aims at coordinated, effective control of action. As we saw in the passage quoted at the start of this section, Bratman holds that coordination requires consistency among one’s intentions given one’s beliefs. A noteworthy feature of Bratman’s view here is that each individual intention aims for coordination with one’s other intentions and beliefs.29 So, in a case in which one intends to X and intends to ~X, thereby violating a consistency constraint on intentions, both one’s intention to X and one’s intention to ~X fail to achieve their constitutive aims of coordinated, effective control of action, simply because both fail to be “coordinated.” As Bratman acknowledges, on this conception of the aim of intention, there’s a significant difference between the aim of belief and the aim of intention.30 When we say that belief aims at truth, we hold that a particular belief that P aims at getting things right with respect to P. There’s no mention, in the specification of the belief ’s aim, of one’s other beliefs. So, in looking to see whether some particular belief achieves its constitutive aim, we don’t need to consider any belief but that particular one. But matters are different when it comes to intentions. A particular intention to X aims for effective control of X-ing in coordination with one’s other intentions. Unlike belief, the aim of an intention to X makes reference to one’s other intentions. So, in looking to see whether some particular intention achieves its constitutive aim, we do need to consider the agent’s other intentions. Thus, on Bratman’s view, there’s an important disanalogy between the aim of belief and the aim of intention. Should we accept Bratman’s account? The step from intention’s constitutive aim (coordinated, effective control of action) to an explanation of consistency requirements is fairly immediate: “coordinated” control is simply understood to require compliance with consistency requirements. But then all the weight of the argument is on the claim that “coordination,” so understood, is a constitutive aim of intention. How should we understand this claim? Again, it’s worth noting that talk of constitutive aims is slippery. As we saw in the 28  Bratman 2009a, p. 25; 2009b, pp. 49–52. 30  Bratman 2009b, p. 52.

29  Bratman 2009b, p. 52.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

192  Instrumental Rationality previous subsection, there are several ideas under the umbrella of the slogan “belief aims at truth,” and it’s unclear how many of these ideas we are taking on board in talking about an analogous aim of intention. But if we follow my suggestion of understanding constitutive aims in terms of success conditions, then Bratman’s claim would entail that when an intention fails to achieve “coordination,” that intention is unsuccessful qua intention—that is, it fails to succeed with regard to its specific role within our psychological apparatus. But that claim seems implausible to me. It’s standardly remarked that in cases of irrational incoherence, there is something amiss with the combination of attitudes. For instance, when one intends to X and intends to ~X, there is something amiss with this com­bin­ ation of intentions. But Bratman’s account would commit us to the further claim that there’s also something amiss with each particular attitude in this combination. Specifically, we would have to say that both the intention to X and the intention to ~X are unsuccessful qua intentions. Moreover, we would be committed to thinking this even when one successfully executes one of these two intentions. If I succeed in X-ing under the guidance of my intention to X, that intention still fails to achieve its constitutive aim because of the mere presence another intention, my intention to ~X. (And this would be so even if the intentions were “compartmentalized,” or part of a “fragmented” psychology, as they often are in cases of inconsistency, and the intention to ~X didn’t present any psychological obstacle to my executing my intention to X.)31 It’s hard for me to see why we shouldn’t say that the ­particular intention to X here has done its job (while of course endorsing the claim, which nearly everyone accepts, that something is amiss in the combination of attitudes). Bratman does provide a brief rationale for saying that in cases of inconsistency, each particular intention fails to achieve its constitutive aim of co­ord­in­ ation. He writes: Why say this? Because our intentions and planning attitudes are elements in a coordinating system. The central role of this system is to guide practical thought and action by way of a coordinated representation of our practical future in the world as we find it. A system of attitudes that were not responsive to pressures for coordination within that system would not be a planning system.32 31  On the idea of “compartmentalization” or “fragmentation” see Lewis 1982, p. 436. 32  Bratman 2009b, p. 52.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  193 What Bratman says here might be right. But I don’t think it establishes the controversial view above. That’s because Bratman’s claim here is a claim about the role of a system in guiding practical thought and action, not the role of specific intentions within that system. It might be true that a planning system has the role of guiding practical thought and action by way of a coordinated representation of our practical future—and so the system would aim at “co­ord­in­ation”—without it being true that specific intentions within this system have this aim. To conclude that particular intentions have this aim from the fact that the system of which they are a part has this aim, would be to commit the fallacy of division. A bundle of boards may have the property of being strong. But it doesn’t follow that an individual board in the bundle has this property. Likewise, the complex structure of intentions, planning attitudes, and beliefs that make up what Bratman calls our “planning system” may aim for “coordination” without that being the aim of any individual intention. So, absent a further explanation of why particular intentions would inherit the aims of the system of which they are a part, Bratman’s rationale is unconvincing.33 Looking ahead, we can use Non-normative Disjunctivism to account for the consistency requirement on intentions without having to rely on Bratman’s controversial thesis that intentions aim at coordination with other intentions. Suppose that instead of working with Bratman’s account, according to which a constitutive aim of intention is “coordinated, effective control of action” we held the weaker view that a constitutive aim of intention is “effective control of action.” (This has the advantage of preserving the parallel with belief, since, as we noted above, the constitutive aim of belief doesn’t make reference to one’s other beliefs.) According to the Non-normative Disjunctivist, whenever you violate a requirement of intention consistency, the logical relations among the contents of your intentions ensure one of them won’t be realized. And whenever you violate a requirement of intention–belief consistency, the lo­gic­al relations among the contents of your attitudes ensure that either your belief is false, or one of your intentions won’t be realized. And this means that either you have a belief that won’t realize its constitutive aim, or you have an intention that won’t realize its constitutive aim. In a case in which some intention won’t realize its constitutive aim, that’s simply because it won’t issue 33  I don’t think it makes a difference that we’re dealing with functional systems. The circulatory system aims to supply cells with nutrients and oxygen-rich blood. But this need not be the aim of a particular part of that system, such as the large vein in my left arm, or the mitral valve in my heart. In the same way, the aim of the planning system as a whole need not be an aim of a particular part of that system, like my intention to turn on the television.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

194  Instrumental Rationality in action. The account thus doesn’t require commitment to the controversial thesis that an intention can fail with respect to its constitutive aim because of its failure to coordinate with other intentions.34 Let’s take stock. We’ve been working with Michael Bratman’s proposals for using the constitutive aim of intention—coordinated, effective control of action, in his view—to explain the norms of coherence and consistency governing intentions. And I’ve raised some problems for Bratman’s proposals. With regard to means-end coherence, it’s not clear how every instance of means–ends incoherence would undermine effective control of action, and so it’s not clear how this constitutive aim could explain means–ends coherence. An analogous problem doesn’t hold for consistency, since “coordinated” is understood so that, by definition, any failures of intention consistency, or intention–belief consistency, are failures of coordination. But here the problem is that it’s not clear why we should accept this account of intention’s aim to begin with. I’ve also suggested that Non-normative Disjunctivism provides a better way to explain these rational requirements. The explanation it gives of means–ends coherence extends to cover every case in which the requirement applies. And with regard to consistency, we can work with a weaker account of intention’s constitutive aim (by jettisoning “coordination”), thereby allowing for an account of intention’s constitutive aim that parallels belief ’s constitutive aim (in that neither makes reference to an agent’s other attitudes). Before turning directly to Non-normative Disjunctivism, I want to briefly consider which parts of Bratman’s account, in light of the arguments I’ve presented so far, we should continue to take on board with us.

34 There are some aspects to Bratman’s discussion (2009b, pp. 50–2) I’m passing over here. Bratman is primarily responding to Velleman, particularly to some of the arguments Velleman gives for the Strong Belief Thesis. (We discussed these arguments in the first section of Chapter  6.) Velleman is willing to concede that it’s irrational to intend to do something while believing it’s impossible. Bratman sees an opening: if we can argue for a requirement of intention agglomeration— whereby intending to A and intending to B rationally requires intending to A&B (and explain this intention agglomeration requirement in a non-cognitivist way by appealing to intention’s constitutive aim) then cases of intention–belief inconsistency will be cases where one either violates the agglomeration requirement (by intending A, intending B, and not intending A&B) or intends to do something while believing it’s impossible (intending to A&B while believing it’s impossible to A&B). We thereby give an account of intention–belief consistency without relying on Velleman’s cognitivism. I won’t here explore this particular suggestion. It still requires that we adopt the controversial thesis that individual intentions constitutively aim at coordination. (That’s needed for the non-cognitivist account of intention ag­glom­er­ation.) Furthermore, I’m skeptical about the alleged rational requirement of intention ag­glom­er­ation. (Exploring this would take us too far afield. See Zhu  2010 for an overview of several objections. Zhu defends a revised version of the intention agglomeration requirement, though not a version which would support the purposes to which it is put here.) And it seems that the Non-normative Disjunctivist has a more straightforward account of intention–belief consistency that doesn’t need to go by way of intention agglomeration. So, we’ll pass over these aspects of the Bratman-Velleman exchange.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  195

7.1.3  Controlled Action If we make the above modification and drop “coordinated,” we are left with “effective control of action” as intention’s constitutive aim. For convenience, I’ll shorten this to “controlled action.” For the purposes of defending Nonnormative Disjunctivism, all we need is an account of intention’s aim such that in failing to act, one fails to achieve the constitutive aim of intention. However, there’s good reason to think, as Bratman does, that controlled action, not mere action, is a constitutive aim of intention. After all, if you intended to turn on the television, but did so by accidentally sitting on the remote control, you will have in a sense done what you intended to do (you turned on the television), but we wouldn’t think your intention succeeded with respect to its role. An initial thought might be that what’s needed for an intention to be successful is that you act because of your intention. After all, in this example, the agent didn’t turn on his television because of his intention to turn it on, but because he sat on the remote control. But this won’t provide the needed fix, since when intentions cause the intended action “in the wrong way” they aren’t fulfilling their role. Suppose, borrowing and slightly modifying an example from Roderick Chisholm, you intend to kill someone by running him over, but, as you are backing out of your driveway on your way to find him, you absentmindedly run over someone who turns out to be your intended target.35 Your intention caused you to run him over, but we wouldn’t say that you’ve successfully executed your intention in this case. Likewise, we could appeal to a slight variation on Donald Davidson’s famous mountain climber case: suppose the climber intends to let go of the rope, which would cause another climber to fall, and the thought of this so unnerves him that he lets go of the rope unintentionally.36 Again, although the intention causes, or at least plays a causal role in, the climber’s dropping the rope, we wouldn’t say that he successfully executes his intention. In these cases, although the agent’s intention is causally relevant, it doesn’t guide or control the relevant action (running him over, dropping the rope). Of course, it’s a notoriously difficult task to say precisely what such control or guidance amounts to—to say how the case of Davidson’s climber differs from ordinary cases where one successfully executes an intention to let go of a rope—but that’s not a question we need to take up here. We could think of “controlled action” as a placeholder for whatever that correct account turns 35  See Chisholm 1964, p. 617.

36  See Davidson 1980b, p. 79.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

196  Instrumental Rationality out to be.37 But I think reflection on these examples gives us some prima facie justification for siding with Bratman in thinking that controlled action, not mere action, is the constitutive aim of intention. However, I won’t pursue the question further, since, as I said, all that Non-normative Disjunctivism needs is the weaker claim that a failure to act is a failure to achieve intention’s constitutive aim. In discussing the aim of belief, Velleman noted that the aim of getting things right could help distinguish belief from other cognitive attitudes, such as assuming or imagining. Parallel to this, the aim of controlled action could help distinguish intentions from other conative attitudes, such desiring or wishing. Suppose I have a desire to hike the Appalachian Trial (AT). I find my thoughts often directed to matters related to hiking the AT, and find the prospect of doing so pleasing.38 But, reasonably, I never form an intention to do so, since my familial and professional obligations are far more important. But that’s no obstacle to my continuing to desire to hike the AT. I could even continue to have this desire were hiking to become physically impossible for me, perhaps because of a severe injury. In these cases, I have a desire (to hike the AT) that doesn’t issue in controlled action. But we don’t think that my desire here failed to do something it was supposed to do. In other words, we don’t think that issuing in controlled action is part of the “job description” of desires. The above point is compatible with the endorsement of a theory of desire that posits a counterfactual connection between desires and action.39 One could hold that if I desire to hike the AT, then I would hike the AT were I not bound by competing familial or professional obligations, and were there no competing desires, and were it physically possible for me to do so, and so on. In holding this, one isn’t committed to thinking that in both the actual and counterfactual circumstances, the role of desire within one’s psychological apparatus is to issue in controlled action. Moreover, it’s worth noting that the connection between desire and action even in the counterfactual circumstances would be mediated by intention, at least for normal cases involving creatures like us. It’s not that in such circumstances my desires would simply sweep me up into acting. The point, rather, is that in the relevant counterfactual circumstances, I would decide—that is, form an intention—to hike the 37  An intention could issue in controlled action without every component of the action being under agential control. I may intend to drink to the point of passing out, and guide myself to doing so (buying liquor, pouring drinks, and so forth), without the particular way in which I pass out being under my control. Thanks to a reader for this point and example. 38  The idea of directed attention is important for Scanlon’s account of one sense of “desire.” See Scanlon 1998, p. 39. 39  Smith (1994, p. 113), for instance, endorses a counterfactual account of desires along these lines.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  197 AT. So, even if we expect desire to issue in action in the counterfactual circumstances, it would be via the formation of an intention. And the role of that intention would be to issue in my hiking the AT. Likewise, wishes don’t aim at controlled action. Suppose I wish that I hadn’t been so angry with you last year. Obviously, I can’t change the past. If wishes shared intention’s constitutive aim, we’d have to say that all wishes for things to have gone differently are unsuccessful qua wishes. Such wishes, we’d be committed to thinking, aim to change the past and fail. That isn’t plausible. The role of wishes in our psychology, if they have one, isn’t to issue in controlled action, or action of any other kind. Let’s sum up the main conclusions of this section. I’ve considered Bratman’s strategy of arguing that there’s an aim of intention, analogous to the aim of belief, which can explain the rational requirements of consistency and coherence governing intentions. Although not all ideas under the “belief aims at truth” slogan carry over well to Bratman’s suggestion that there’s an analogous aim of intention, there is a way to make sense of the idea in terms of the success conditions of belief and intention. But I’ve also disagreed with Bratman’s particular conception of the aim of intention, particularly its appeal to “co­ord­ in­ation,” and also with the way Bratman uses that conception to explain the rational requirements of consistency and coherence on intention. Let’s now see if there’s another, better way to do that.

7.2  Non-Normative Disjunctivism Let’s suppose that you’re means–ends incoherent. You intend to take the train from New York to Charleston, believe that to do so you must intend to buy a ticket, but you don’t intend to buy a ticket. The Non-normative Disjunctivist looks to the logical relations among the contents of your attitudes, and the constitutive aims of those attitudes, to explain why something is amiss in this case. Here’s one way of putting the explanation: if your belief that you must intend to buy a ticket in order to take the train to Charleston achieves its constitutive aim (truth), then you’ll take the train to Charleston only if you intend to buy a ticket. But since you don’t intend to buy a ticket, you won’t take the train to Charleston. If this is so, your intention won’t succeed with respect to its constitutive aim. So, given your combination of attitudes, either you’re wrong about the need to intend to buy the ticket, and so your belief fails to achieve its constitutive aim, or you won’t take the train to Charleston, and so your intention will fail to achieve

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

198  Instrumental Rationality its constitutive aim. In other words, your failing to intend to buy a ticket has ensured a constitutive aim failure: either your belief is false or your intention will not succeed. And the logical relations among the contents of your attitudes—specifically, how the content of your intention hooks up with the content of your belief via modus ponens—is a crucial part of the above explanation of why this is so. (You might also ensure a constitutive aim failure simply by intending to do something that cannot be done. But we don’t think that there need be any irrationality here. You might simply be unaware of the impossibility of doing what is intended.) There are two crucial components at work in this explanation: the logical relations among the contents of the attitudes, and the constitutive aims of the attitudes. To see how each component is crucial, we could note how slight variations with respect to each would suffice to remove the irrationality. Consider first the constitutive aims of the attitudes involved. Suppose we swapped out your intention to take the train to Charleston with a wish to take the train to Charleston. (We’ve here made no alteration to the contents themselves, just to the attitudes you have toward those contents.) As I argued earl­ ier, wishes do not share intention’s constitutive aim, and so there need not be a constitutive aim failure when you wish to X and fail to X. But assume your other attitudes remain the same: you still believe that you’ll take a train to Charleston only if you intend to buy a ticket and don’t intend to buy a ticket. Since the logical relations among the contents remain the same as before, we can again say that your not intending the means guarantees that either your belief is false or your wish won’t come true. But there’s no guarantee of constitutive aim failure here, since an unfulfilled wish is not a constitutive aim failure, and there’s also no irrationality. Now suppose that we instead vary our original case in another way: instead of believing that you’ll take the train to Charleston only if you intend to buy a ticket, you believe you’ll intend to buy a ticket only if you’ll take the train to Charleston. You no longer think your intending to buy a ticket is necessary, but do think it is sufficient, for taking the train to Charleston. (This would of course be a highly unusual belief to hold.) Again, as before, you intend to take the train to Charleston and don’t intend to buy a ticket. Note that your not intending to buy a ticket does guarantee a constitutive aim failure here: you won’t succeed in your intention to take the train to Charleston. But the logical relations among the contents of your attitudes is not relevant to the ex­plan­ ation of why that constitutive aim failure is guaranteed. Rather, the ex­plan­ation is given simply by the fact that you don’t intend some means the intending of which is in fact necessary to take the train to Charleston. And there’s no irrationality here.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  199 We could make the same points with regard to our requirement of intention– belief consistency. (We’ll discuss consistency requirements more generally in Section  7.2.2.) Suppose that, irrationally, you intend to take the train to Charleston, believe that if you take the train to Charleston, you’ll travel south of the Mason-Dixon line, and intend not to travel south of the Mason-Dixon line. You’ve guaranteed a constitutive aim failure—if your belief is correct, then one of your intentions won’t succeed; if both of your intentions succeed, then your belief is false—which is explained by the logical relations among the contents of your attitudes. But if we swapped out your intentions for desires, keeping the contents the same, so that now you desire to take the train to Charleston and desire not to travel south of the Mason-Dixon line, then there would be no guarantee of a constitutive aim failure, and no ir­ration­al­ity. Likewise, if we varied the case by supposing that you don’t believe that if you travel to Charleston, you will travel south of the Mason-Dixon line—we’ll suppose that you’re not great at geography and history—you will still have guaranteed a constitutive aim failure, since either your intention to travel to Charleston won’t succeed or your intention not to travel south of the MasonDixon line won’t succeed. But this assured failure won’t be explained by the logical relations among the contents of your attitudes, but instead by the geographical fact that Charleston is south of the Mason-Dixon line. And there’s no irrationality here. You are bad at geography and history, not irrational. In the remainder of this section, I’ll explore several advantages of the Non-normative Disjunctivist approach to means–ends coherence. I’ll begin (Section 7.2.1) by noting how the approach is able to avoid several objections I’ve made against rival views in previous chapters. I’ll then argue (Section 7.2.2) that the approach can also offer promising explanations of a class of rational requirements (including belief consistency and closure, intention consistency, and intention–belief consistency) that are intuitively closely related to means–ends coherence. That’s another advantage. I’ll also show (Section 7.2.3) how the approach has the resources to explain the pressure we feel to comply with means–ends coherence when we’re made aware of failures to comply. Yet another advantage (Section  7.2.4) is that the view promises to remove one source of attraction to the myth theory, by explaining how there could be a non-conventional, non-normative source of requirements.

7.2.1 Advantages One advantage of the view is that it meets the Kantian desideratum of providing an account of means–ends coherence that applies regardless of the normative

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

200  Instrumental Rationality status of the agent’s ends. With Kant, we can say that “whether the end is rational and good is not at all the question here.”40 For both Kant’s phys­ician and Kant’s poisoner, when there’s means–ends incoherence, the logical relations among the agent’s attitudes ensures that either the relevant instrumental belief is false or the relevant intention won’t be achieved—that is, it ensures that either the belief fails to hit its aim (truth) or intention fails to hit its aim (controlled action). Of course, the poisoner has a problem the phys­ician doesn’t have: he has an intention he ought not have. But we’ve here identified a problem they both share. We’ve identified the way in which they both have set themselves up to fail. The view also avoids the “bootstrapping” objection from Chapter 2. That’s because the view doesn’t hold that intentions generate reasons. Intentions do  put in place a standard of success. But it’s perfectly possible, on this view,  that there’s no reason in favor of the successful execution of some intention. Similarly, Non-normative Disjunctivism isn’t committed to Strong Normativity—the view that we always ought to be means–ends coherent, and so can avoid the three problems we raised for that view in Chapter 5, such as  the problems concerning the transmission of oughts to necessary and ­sufficient means. Although the view is a close cousin of Normative Disjunctivism, it doesn’t inherit any of the four objections we raised to that view. First, in the case of Setiya’s smoker, the Normative Disjunctivist ran into difficulty in that the psy­cho­logic­al impossibility of the smoker’s revising his intention to smoke would, if a suitable version of ought-implies-can is true, block the claim that he ought to revise his intention to smoke. But there’s no analogous “constitutiveaim-implies-can” principle that we need to worry about. The psy­cho­logic­al impossibility of his revising his intention is no barrier to the claim that his not intending to buy a pack, if he’s right about that being necessary, will render his intention to smoke unsuccessful. Second, the Normative Disjunctivist was committed to thinking that even in cases where instrumental incoherence was advantageous, there was some specific attitude that ought to be revised or acquired. The Non-normative Disjunctivist can say one ought to remain incoherent in such cases. It would still be true that the logical relations among the contents of the agent’s attitudes ensure a constitutive aim failure. But it’s no part of the view that one ought to avoid this. Third, the Normative Disjunctivist who is attracted to idealization when it comes to specifying what you ought to do would be forced to adopt the view 40  Kant 1785/1997, p. 26 [4:415].

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  201 that you ought to believe what’s true, regardless of the evidence available to you. The Non-normative Disjunctivist can remain neutral on how we approach the questions of what you ought to do and what you ought to believe. When you’re means–ends incoherent, either your instrumental belief is false or your intention will be unsuccessful. There are further questions about how this fact bears, if it does at all, on what you ought to do or believe. Non-normative Disjunctivism doesn’t come with substantive commitments regarding how these further questions are to be answered. Fourth, Normative Disjunctivism struggled with cases of merely per­mis­ sible ends—that is, ends that we are permitted but not required to intend. But such cases pose no difficulty for the Non-normative Disjunctivist. Regardless of whether one’s intentions are required, merely permitted, or forbidden, the same explanation is given in terms of the constitutive aims and contents of the relevant attitudes. Additionally, since the view is a non-cognitivist account, it avoids all the various problems for the cognitivist views discussed in Chapter 6. It doesn’t commit us to a controversial thesis about how intention involves belief, allows that you can be instrumentally irrational even when you are unaware of your failure to intend the means, doesn’t generate counterintuitive predictions about the appropriate direction of revision in cases of instrumental incoherence, and so forth. Lastly, as we’ve already noted, the view provides a way of carrying out Bratman’s strategy of using the aim of intention to explain means–ends coherence while avoiding the main problem with Bratman’s approach. On Bratman’s approach, explaining means–ends coherence in its full generality requires that we endorse the claim that means–ends incoherence, even in cases where one has a false instrumental belief, prevents an intention from achieving its constitutive aim. For Non-normative Disjunctivism, we need only endorse the claim that means–ends incoherence prevents an intention from achieving its constitutive aim when one’s instrumental belief is true. So, the view has a lot going for it. I should say more about the origins of this approach. As we’ve noted already, we’re borrowing the idea from Velleman (and others) that belief constitutively aims at truth. And we’re borrowing (and modifying) the idea from Bratman that intention analogously aims at controlled action. And the “Disjunctivism” part of the approach is borrowed from Kolodny. In the course of defending a myth theory about intention–belief consistency, Kolodny noted in passing the following observation: Certain Failure: If one intends to X, intends to Y, and believes that if one X’s, then one does not Y, then either one’s intention to X does not succeed, or

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

202  Instrumental Rationality one’s intention to Y does not succeed, or one’s belief that if one X’s, then one does not Y is false.41

However, Kolodny immediately goes on to note that this observation won’t yield his desired conclusion that when one intends to X, intends to Y, and believes that if one X’s then one does not Y, it’s assured that one violates a “requirement of reason” in having one of these three attitudes. In other words, this observation won’t support the Normative Disjunctivism that Kolodny wants to defend in order to advance his myth theory about intention–belief consistency. In guaranteeing that either one has an intention that won’t succeed or one has a belief that is false, one does not thereby guarantee that either one has an intention one ought not have or has a belief one ought not have. The problem, he notes, is that having a false belief need not violate a requirement of reason. (All the available evidence might support P, in which case you ought to believe it, even if P is false.) And being unsuccessful in executing an intention need not involve violating a requirement of reason. (Perhaps the intention was one you had no reason to form in the first place.) As Kolodny puts it: “The difficulty is that one need not violate a requirement of reason by having a false belief, or by not succeeding in an intention.”42 So, Certain Failure won’t get Kolodny where he wants to go. That’s why it is dismissed so quickly. But we’re not looking to go in Kolodny’s direction. (Indeed, if the arguments in Chapter 4 are correct, we should accept neither Normative Disjunctivism, nor a myth theory about practical requirements.) And, for our purposes, Certain Failure might do quite well. If we follow Velleman in thinking that beliefs constitutively aim at truth, and follow Bratman (to some extent) in thinking that intentions constitutively aim at controlled action, then, if Certain Failure is true, whenever one intends to X, intends to Y, and believes that if one X’s one does not Y, it’s assured that there’s a constitutive aim failure— that is, either the belief is false or one of the intentions won’t issue in controlled action.43 The resulting mixture—one part Velleman, one part Bratman, with a twist of Kolodny—offers a promising way to understand the rational requirement of intention–belief consistency. (We’ll return to this again in a

41  Kolodny 2008a, p. 368. 42  Kolodny 2008a, p. 368. 43  Non-normative Disjunctivism goes beyond Kolodny’s statement of Certain Failure in two ways. First, it supplements that thesis with an account of the constitutive aims of intention and belief. Second, it notes that what explains the assurance of a constitutive aim failure here is the logical structure of the contents of your attitudes.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  203 moment.) And all we’ve done above is extend this approach to account for means–ends coherence.

7.2.2  Extension to Other Rational Requirements I’ll now consider how well the strategy of Non-normative Disjunctivism ­generalizes. Specifically, does the approach offer a way to explain other rational requirements besides means–ends coherence? I’ll argue that there are grounds for optimism regarding the extension of the approach to closely related requirements of consistency and coherence governing beliefs and intentions. That’s an advantage of the approach. (Other things equal, it would be odd to explain means–ends coherence one way, but explain intention consistency and intention–belief consistency in other ways.) But I’ll also argue that there are grounds for pessimism regarding the extension of the approach to other rational requirements, such as the prohibitions on intransitive preferences and akrasia. Let’s start with consistency norms. We’re in a position to provide an ex­plan­ation that unifies the three commonly discussed consistency requirements governing intentions and beliefs: intention consistency, belief consistency, and intention–belief consistency. For all of these requirements, the logical relations among the contents of the attitudes governed by the requirement ensure a constitutive aim failure. If you intend to X and intend to ~X, then one of these intentions will be unsuccessful qua intention. If you believe that P and believe that ~P, then one of these beliefs will be unsuccessful qua belief. And, as we noted with Certain Failure above, if you intend to X, intend to Y, and believe that if you X you will not Y, then either one of your intentions will be unsuccessful qua intention or your belief is unsuccessful qua belief.44 The account assumes we already have an understanding of what’s involved in the contents of attitudes being inconsistent. We can say that a set of sentences is inconsistent just when their logical form ensures that not all of them are true. (There are more precise model-theoretic and proof-theoretic def­in­ itions of inconsistency available. We could say that a set of sentences is semantically inconsistent if and only if it has no model, and that a set of sentences is syntactically inconsistent if and only if it can be used to prove (P&~P). But the rougher gloss will do for our purposes.) But that’s not yet to 44 The account of consistency here avoids Bratman’s controversial claim, discussed above, that “coordination” is a component of intention’s constitutive aim.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

204  Instrumental Rationality say what makes the attitudes inconsistent in cases governed by these three consistency requirements. (As we noted earlier, not all combinations of attitudes toward inconsistent contents are irrational. If we substituted desires in for intentions, there would be no irrationality.) In other words, it’s not yet to say anything about what makes you inconsistent in having intentions and beliefs with inconsistent contents.45 What we need to add, on this view, is that the logical form ensures a constitutive aim failure. That’s what makes your attitudes, and not just their contents, inconsistent. In short, you’re inconsistent because your attitudes are arranged so that logic dictates that not all of them can achieve their constitutive aims.46 That’s the way in which you’re “at odds with yourself ” when you violate consistency requirements. Non-normative Disjunctivism can also explain the requirement for you to believe the logical consequences of your beliefs. Suppose you believe that P, believe that P→Q, but don’t believe that Q.47 The logical relations among the contents of your attitudes ensure that either your belief that P is false, your belief that P→Q is false, or Q is true and you don’t believe it. But does this last possibility involve a constitutive aim failure? Not believing Q is the absence of an attitude. How can we say that there’s a constitutive aim failure in the absence of an attitude? We should, however, remember our initial description of how belief aims at truth: belief is an attitude that tends to come into existence in response to indicators of truth, and go out of existence in response to indicators of falsity. As Velleman puts it, beliefs are characterized as aiming or “trying to arrive at the truth” given the ways in which we systematically form, revise, and abandon beliefs in light of indicators of truth (evidence or

45  For a related helpful discussion of inconsistency in attitudes, see Schroeder  2008, pp. 43–4. There’s a difference between the explanation of these consistency constraints, on the one hand, and the explanation of means–ends coherence and the requirement to believe the logical consequences of one’s beliefs (discussed below), on the other. The conflict governed by consistency constraints is a conflict between the having of some attitudes and the having of others, and that allows us to identify inconsistency entirely in the contents of the attitudes, whereas with means–ends coherence and the requirement to believe the logical consequences of one’s beliefs, the conflict is between the having of some attitudes and the lacking of others, and so we can’t identify incoherence entirely in the contents, since the lack of an attitude has no content. Nonetheless, the logical relations among the contents of the attitudes you do have are essential components in the explanations of why you’ve ensured constitutive aim failures. 46  I have avoided the difficult task of formulating the rational requirements of consistency governing belief. One possible response to the preface and lottery paradoxes would be to allow for the rational permissibility of inconsistency in sufficiently large sets of beliefs. If that response is on the right track, then we would need to supplement our constitutivist account by noting that a constitutive aim failure guaranteed by the logical relationship among the contents of one’s beliefs doesn’t suffice for irrationality. We would need to add some additional condition, perhaps that the set of beliefs is sufficiently small. 47  Suppose also that you care about whether Q, or meet whatever conditions there are for the requirement to be applicable.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  205 argument). So, it would be incorrect to think that constitutive aim successes or failures can occur only with respect to those beliefs we already have. Rather, they can occur with respect to the formation of beliefs as well. Perhaps it would be better to say that one’s belief-forming mechanisms, rather than any particular belief, have the constitutive aim of arriving at the truth with respect to some particular proposition. That would be a helpful way of understanding constitutive aims so that they are applicable both to conflicts between the having of beliefs and the having of other beliefs, and conflicts between the having of beliefs and the absence of other beliefs. Putting the view this way also has the advantage of allowing us to explain the limited applicability of the rational requirement to believe the logical consequences of one’s beliefs. We have to make some decisions about the domains in which we employ those belief-forming mechanisms. I could, after all, spend a lifetime exploring the logical consequences of my mathematical beliefs. But this would be a waste of time and energy, and come at the expense of other valuable forms of inquiry. And that’s why there’s no general requirement to believe the logical consequences of one’s beliefs. Instead, the requirement is restricted in some way—perhaps, to use Broome’s preferred formulation, it is restricted to contexts in which you care about the truth of propositions that logically follow from the contents of your current stock of beliefs. So far, I’ve argued that our approach to explaining means–ends coherence— that is, the approach of looking to how the logical relations among the contents of attitudes help explain the guarantee of constitutive aim failure—can be extended to explain other rational requirements as well, such as intention consistency, belief consistency, intention–belief consistency, and the requirement to believe the logical consequences of your beliefs. In all of these cases, the logical relations among the contents of attitudes help explain the guarantee of constitutive aim failure. However, there are clear limits to how far the approach will go. It’s of no help with regard to the prohibition on intransitive preferences. Suppose I irrationally prefer apples to bananas, bananas to cherries, and cherries to apples. Does this combination of preferences ensure that the constitutive aim of one of these preferences won’t be met? For starters, it’s not clear what the constitutive aim of a preference is, if there is one. Some philosophers, such as Jacob Ross, have argued that one’s preferring A to B plays the psychological role of settling the matter of what to choose when given the choice between A and B. (Forming such preferences saves deliberative resources, in that one won’t have to think over what to choose when presented with the choice between A and B. It’s a “surrogate for deliberation,” as

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

206  Instrumental Rationality Ross puts it.48) On Ross’s view, given this psychological role, one’s preferring A over B commits one to choosing A over B when given the choice between these alternatives. Ross doesn’t put the point in terms of constitutive aims, but one might hold that a preference of A over B constitutively aims at a choice of A over B when the options present themselves. But, even assuming this is correct, the problem is that in preferring apples to bananas, bananas to cherries, and cherries to apples, there’s no assurance of a constitutive aim failure. For one thing, I might never be given a choice of fruits. For another, I might be given only pairwise choices, and succeed in choosing apples over bananas when given that choice, succeed in choosing bananas over cherries when given that choice, and succeed in choosing cherries over apples when given that choice. Unlike having inconsistent beliefs, or inconsistent intentions, my having intransitive preferences doesn’t ensure a constitutive aim failure. There might be other ways of explaining what’s irrational about intransitive preferences—perhaps my vulnerability to being a money pump49— but our approach won’t cover it. And our approach is also no help with regard to the rational prohibition on akrasia. I can believe, correctly, that I ought not eat any more apples and yet successfully intend to eat more of them. There’s no constitutive aim failure here; my belief and intention are both successful.50 I’m skeptical that there is some single unifying theory that will explain all the requirements of rationality governing our attitudes in the same way. I think I’ve shown what some requirements (means–ends coherence, intention consistency, etc.) have in common. But the Non-normative Disjunctivist approach, I’ve argued, would be unpromising as an explanation of all rational requirements, since it won’t cover prohibitions on intransitive preferences and akrasia. Of course, other things being equal, it would be preferable to

48  Ross 2012, §6.2. 49  See Davidson, McKinsey, and Suppes 1955. 50  Nor would the approach explain all the requirements that might come under the umbrella of “instrumental rationality.” For instance, perhaps, as John Broome (2013, 169–70) suggests, there’s also a rational requirement for us to choose the best means to our ends. (Broome, however, puts such a requirement aside, given the difficulty of formulating the requirement, which would require us to specify what makes one means better than another.) It’s not clear to me how Non-normative Disjunctivism could help explain a requirement to choose the best means. It is worth noting that among the possible rational requirements under the umbrella of “instrumental rationality,” means– ends coherence has attracted the most attention from philosophers. I suspect that this is so for at least two reasons. First, unlike, say, a requirement to choose the best means, it seems relatively easy to arrive at a precise formulation. Second, it bears an affinity with other commonly discussed rational requirements, like the requirements of consistency governing beliefs and intentions. Non-normative Disjunctivism explains that affinity by pointing out features common to those requirements.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  207 have a view that explains all of the requirements on our attitudes in the same way. But I don’t think such a view is available.51

7.2.3  Explaining Pressure When I point out to someone that she is means–ends incoherent, or irrational in some other way, it’s unlikely that she would respond by saying “So what?” or something similarly dismissive. Instead, it’s likely that she would feel pressure to revise her attitudes so as to escape irrationality. But why would this be so? What explains the felt pressure to revise in the direction of compliance? Other philosophers skeptical of the normativity (in the sense captured by Strong Normativity or Weak Normativity) of rationality have thought they’ve had to explain this felt pressure. For instance, Niko Kolodny52 proposes his Transparency Account as a way of doing so. If we allow that all rational requirements fit the model of Kolodny’s “core requirements” in that they are, roughly, requirements to have the attitudes one believes one ought to have, and lack the attitudes one believes one ought not have—or we allow that all requirements can be derived from these core requirements—we would have a way of explaining rational pressure: we feel compelled to comply because as it 51  One very promising (but ultimately unsatisfactory, in my view) unifying theory has been offered by Alex Worsnip. Here is his view: A set of attitudinal mental states is jointly incoherent iff it is (partially) constitutive of the mental states in question that, for any agent that holds these attitudes, the agent is disposed, when conditions of full transparency are met, to give up at least one of the attitudes. That is, human agents are disposed such that they are (at least normally) not able to (or at least find it difficult to) sustain such com­bin­ations of attitudes under conditions of full transparency. A putative coherence requirement is genuine iff every combination of states that it forbids is jointly incoherent.  (2018, p.188) For instance, on Worsnip’s view, the requirement of Belief Consistency is a genuine requirement, since  in conditions of full transparency—that is, in “conditions under which the agent knows, and explicitly and consciously believes, that she has the states in question, without self-deception, mental fragmentation, or any failure of self-knowledge (pertaining to those attitudes)” (p. 188)—the agent will be disposed to give up one of the beliefs. However, I worry that this view generates both too many and too few requirements. As for the worry about too many: suppose I have a mistaken higher-order belief. Suppose, for instance, that I believe that P and I believe that I don’t believe that P. I admit there is something odd about this com­ bin­ation of attitudes, but I don’t think that we would say that I’m irrational in this case. (I’ve just made a mistake about my own psychology. But a false belief, even if about one’s own mind, does not make for irrationality.) Yet I would be disposed to give up one of these beliefs under conditions of full transparency. As for the worry about too few: the theory has difficulty allowing for the irrationality of clear-eyed akrasia. Worsnip offers some defense of his theory against this latter worry, consideration of which would take us too far afield. 52  Kolodny 2005, pp. 513, 557–60.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

208  Instrumental Rationality seems us we ought to comply with them.53 In other words, the requirements are apparently normative. Kolodny recognizes that there’s a significant challenge in explaining how coherence requirements like belief consistency, intention consistency, and means–ends coherence can be derived from the core requirements, which involve beliefs about reasons. In Chapter 3, Section 3.3.1, I noted briefly some considerations that would give us reason to be skeptical of the idea that means–ends coherence can be derived from requirements governing beliefs about reasons. The basic worry was that means–ends incoherent agents might not have the relevant beliefs about the reasons for their attitudes, or might have beliefs that wouldn’t necessitate any particular revision, such as the belief that an end is merely permissible (neither required nor forbidden). I won’t review those points here. But it does mean that I won’t be in a position to rely on Kolodny’s Transparency Account to explain the felt pressure towards being means–ends coherent. I need another explanation—one that doesn’t look toward our beliefs about reasons. However, I think that such an explanation is readily available to the Nonnormative Disjunctivist. Let’s start by recalling what licenses us to say that beliefs aim at truth and intentions aim at controlled action. We’re licensed to say that beliefs aim at truth given the ways in which we systematically form and revise beliefs in response to evidence and argument—the indicators of truth. And we’re licensed to say that intentions aim at controlled action given the ways in which intentions involve various dispositions geared toward the production of controlled action (e.g., dispositions to engage in further planning, dispositions to intend means believed necessary for the end, dispositions to remove obstacles to one’s acting as one intends, and, plausibly, second-order dispositions to be self-critical when such first-order dispositions fail to manifest at the appropriate times). Such an understanding of the roles of beliefs and intentions in our psychologies gives us enough material to explain felt pressure towards compliance.54 We can hold that an intention, unlike a wish 53  As Kolodny (2005, pp. 513) puts it, “The normative pressure . . . [he] feels to comply with the rational requirement, by forming the attitude, derives from how things seem to him – from the reason that, as it appears to him, he has.” 54  Finlay  2009 goes further and claims that intentions guarantee compliance with means–ends coherence, since whenever one intends to E and believes M-ing is necessary for E-ing, one thereby intends to M. And so he’s skeptical that means–ends coherence is a genuine requirement of rationality on the grounds that it’s impossible to violate it. On this view, the felt pressure to comply with means– ends coherence wouldn’t be something to be explained, but something to be explained away. I lack space for a treatment of Finlay’s view here. But it’s worth noting that the Non-normative Disjunctivist could adopt the first part of Finlay’s view about what’s involved in intending—the claim that in intending to E, one intends to take the necessary means to E (where this is read de dicto, not de re)—while

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  209 or a desire, just is the kind of mental state that generates pressure in the direction of being means–ends coherent. Part of what it is for you to intend an end, as opposed to merely desire it or hope that it obtains, is for you to feel pressure to intend believed necessary means when it’s pointed out to you that you fail to do so. (I’m here assuming that in pointing this out to you, I’m not also providing you with any evidence or argument that would give you good reason to revise your instrumental belief. Were I to do so, you would also feel pressure in the direction of compliance with Means–Ends Coherence, but as a by-product of your beliefs being appropriately responsive to that evidence or  argument.) We thus have a simple explanation of felt pressure towards means–ends coherence. We don’t need to rely upon an agent’s beliefs about normative reasons to explain pressure toward compliance. It can instead be explained by looking at the very attitudes governed by means–ends coherence, intentions and beliefs, and the roles they play within our mental lives.

7.2.4  Social Practices, Normativity, and Myths There’s yet another advantage of the general strategy of looking to the constitutive aims of attitudes to explain rational requirements governing those attitudes: this general strategy promises to resolve a puzzle about the status of those rational requirements—a puzzle which, when left unresolved, might provide us with grounds for accepting a myth theory about those requirements. I’ll start by explaining the puzzle, and then say how I think this general strategy can solve it.55 The puzzle emerges once we cast doubt on the normativity of rational requirements. So far, I’ve focused primarily on one rational requirement, means–ends coherence, and I’ve focused on one way in which it could be normative: Strong Normativity. That’s because I was interested in whether we could explain the rational requirement of means–ends coherence in terms of a requirement of reason to be means–ends coherent. In Chapter 5, I argued that Strong Normativity is false. But one might think that worries about the normativity of means–ends coherence cut deeper in two ways. First, one might think that Weak Normativity is also false. (After all, it’s not clear what denying the second—the claim that when one intends to take the necessary means to E (where this is read de dicto, not de re), and one believes M-ing is necessary, one also intends to M. 55  Versions of this puzzle can be found in Kiesewetter 2018, p. 43; Worsnip 2018, pp. 186–8; and Way, 2018, pp. 490–1.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

210  Instrumental Rationality the reason is to be means–ends coherent in every instance. It doesn’t seem to be good in itself—recall Kolodny’s worries about an obsessive concern for “psychic tidiness”—nor does it seem that we can point to some good that would be promoted by your compliance with this requirement in each and every case.) And, second, one might think that these worries about the normativity of means–ends coherence are also worries about the normativity of rational requirements in general. After all, it seems that we can raise the same challenges to the normativity of any rational requirement whatsoever. Those who think the worries cut deeper in these two ways would have to admit that the requirements of rationality differ from the requirements of prudence and morality. For prudence, we can say that if prudence requires you to X, then there’s a reason for you to X. And for morality, we can say that if morality requires you to X, then there’s a reason for you to X. (Perhaps we could also say, more strongly, that if morality requires you to X, then you ought to X, where this is the “ought” of practical reason, not the moral “ought.”) But for rationality, we can’t say that if rationality requires you to X, there’s a reason for you to X. In short, the requirements of prudence and morality are normative, whereas the requirements of rationality are not. Now, this is not yet puzzling. After all, there are lots of requirements that aren’t normative. Think of the requirements of chess, law, grammar, etiquette, and so forth. The rules of chess prohibit me from moving my rook diagonally. State and city laws require me to bring my vehicle to a complete stop at a stop sign. The requirements of contemporary English grammar prohibit me from saying “most unkindest” (though it was apparently acceptable in Shakespeare’s time). Etiquette, Philippa Foot tells us, requires that an invitation in the thirdperson be answered in the third-person. But is there always a reason to comply with these requirements? The question of whether we always have reason to comply with the requirements of chess, law, grammar, and etiquette is up for debate. But the relevant point here is that we don’t think that if that debate yielded a negative answer, we’d have to give up on the idea that these requirements exist. The rules of chess, for instance, will still require that you not move your rook diagonally, even if in some particular circumstance, there’s no reason for you to comply with them. And, as Foot notes concerning her example of a rule of etiquette, “the rule does not fail to apply to someone who has his own good reasons for ignoring this piece of nonsense, or who simply does not care about what, from the point of view of etiquette, he should do.”56 These thoughts might, at first glance, provide some comfort to the person 56  Foot 1972, p. 308.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  211 who is skeptical about the normativity of rationality: just as the requirements of chess, law, grammar, and etiquette would continue to be applicable even if they aren’t normative, the requirements of rationality would continue to be applicable even if they aren’t normative. However, this comfort is short-lived. For all these other non-normative requirements (chess, law, etiquette, grammar), we can provide some ex­plan­ ation of why the requirements would continue to be applicable even if they are not normative: the requirements have their source in some contingent social practice or rule. The requirement not to move rooks diagonally can be traced back to the rules of the game set by FIDE and other governing bodies. The prohibition on rolling stops at stop signs can be traced back to city and state laws which have been passed by the appropriately authorized legislative bodies, not repealed by courts, and so forth. The prohibition on saying “most unkindest” is a linguistic convention, and the norms governing responses to invitations are social conventions. No doubt the full story of how social practices ground each of these requirements is complex. (It’s no easy task to spe­ cify, for instance, the complete grounds for some rule being a valid legal requirement while another rule is not.) But we think that some explanation along these lines can be given—that is, we think there’s a way to trace these requirements back to social practices or conventions, the existence of which will explain why the requirements would continue to apply even when they aren’t normative.57 But the problem is that such explanations seem unavailable in the case of rational requirements. What social rule or convention would explain the consistency constraint on intentions, or means–ends coherence? It’s not clear how such an explanation could be given. Moreover, looking toward contingent social rules or conventions seems to be looking in the wrong direction, since the requirements of rationality are supposed to govern all possible agents. Jonathan Way puts it well:

57  Although the typical examples of requirements I give above (etiquette, games, grammar, law) are explained in terms of social practices, I don’t need to commit myself to the view that the relevant practices that explain genuine requirements are necessarily social. Suppose, alone on an island, I follow a rule to exercise in the morning. Suppose that it isn’t merely that I “make it a rule” of exercising every morning in the sense that I do so habitually, but I criticize myself for lapses, am guided by the rule, and so forth. (See Dworkin  1967, 30.) If we think of these exercise requirements as genuine requirements, they will be explained in terms of contingent practices, though those practices wouldn’t be social. However, this possibility won’t help avoid the problem discussed in the main text, since the requirements of rationality govern all possible intentions and beliefs, and so won’t be explained by contingent practices of any sort, social or individual. I’ll continue to speak of social practices, since those would be involved in the explanations of the commonly cited examples of requirements of law, games, etiquette, and grammar.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

212  Instrumental Rationality Requirements of this sort [law, etiquette, etc.] are naturally understood in terms of contingent social practices. However, rational requirements, if true at all, seem to be necessary – after all, in motivating the idea that there are rational requirements we did not need to make any assumptions about the actual world; all possible agents who have inconsistent beliefs or intentions, or who are means-end incoherent, seem thereby to be irrational. But if rational requirements are necessary, they cannot be understood in terms of contingent social practices. In this respect, rational requirements seem more like the basic requirements of morality or prudence, which are also widely held to be necessary.58

So, the explanation we gave for how the requirements of chess, law, etiquette, and grammar could persist even in the absence of reasons to comply with them—an explanation that appeals to social practices or conventions—seems unavailable in the case of rational requirements. Now here’s the puzzle: If rational requirements are unlike the requirements of prudence and morality, since they aren’t normative, and are unlike the requirements of chess, law, etiquette, and grammar, since they can’t be traced to social rules or conventions, then why should we think they are genuine requirements at all? It might help to put the puzzle in the form of an argument for a myth theory about rational requirements: 1)  All genuine requirements are either normative (like prudential and moral requirements) or grounded in social practices and conventions (like the requirements of chess, law, etiquette, and grammar). 2)  Rational requirements are not normative. 3)  Rational requirements are not grounded in social practices and conventions. C)  Therefore, rational requirements are not genuine requirements. If this argument is sound, then it’s a very quick route from denying the normativity of rational requirements to thinking they’re a piece of mythology.59 There’s been a lot of discussion of the second premise of this argument, which is highly controversial. The third premise seems unassailable to me. I  want to focus on the first premise. If we can find fault with that premise, we’ll have a way of resisting the slide into a myth theory, even if rationality 58  Way 2018, 491. 59  Although this version targets rational requirements in general, we could formulate a narrower version that targets only means–ends coherence.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  213 turns out not to be normative. The key to rejecting the first premise, I’ll argue, is recognizing that there’s a source for genuine, non-conventional, non-normative rational requirements: constitutive aims. In this chapter, I’ve been considering the approach of looking at the ­constitutive aim of intention to explain the requirements of consistency and coherence governing intentions. I’ve considered Michael Bratman’s particular way of developing this idea, and presented some criticisms of it. And I’ve also suggested another way to use the constitutive aim of intention to explain these rational requirements: Non-normative Disjunctivism. Besides Bratman’s account and Non-normative Disjunctivism, there may yet be other ways to use the constitutive aim of intention to explain the rational requirements governing intention. I here want to abstract away from any particular account, and discuss the advantages of the general strategy of using the constitutive aims of attitudes to explain requirements governing those attitudes. Note that if we explain rational requirements in terms of the constitutive aims of the attitudes governed by the requirements, we wouldn’t be explaining them in terms of some contingent social rule or convention. If some aim is constitutive of some attitude, then part of what it is to have that attitude is to have that aim. And so, for example, every possible belief would be aimed at getting things right, and every possible intention would be aimed at controlled action. These constitutive aims are thus a natural place to look for an explanation of rational requirements, which, as Way notes in the passage quoted above, apply to every possible agent, regardless of her social circumstances. In tracing rational requirements back to the very nature of the attitudes they govern, we would thereby avoid tracing them to some contingent social source. But, on the other hand, these constitutive aims aren’t normative (in the reason-providing sense of “normative”). As we noted in Section  7.1, Kant’s poisoner could successfully execute his intention without his doing so being correct or appropriate, and without having any reason to do so. In specifying the constitutive aims of attitudes, we need not commit ourselves to thinking that there’s any reason to achieve those aims. Our appeal to constitutive aims in this chapter differs from much of the work done by constitutivists in moral philosophy, whose project is to show that morality itself can be traced back to the constitutive aims of action, attitudes, or agency.60 For instance, David Velleman has argued that just as belief constitutively aims at truth, action constitutively aims at self-knowledge, and this 60  See Velleman 2000b, Korsgaard 2009, and Katsafanas 2013. For criticism of this approach, see Enoch 2006.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

214  Instrumental Rationality aim can be used to explain how we have external reasons—in the sense of “external” familiar from Bernard Williams’s work—to act, including moral reasons.61 And Christine Korsgaard has argued that both moral and instrumental requirements are constitutive of action.62 These are highly ambitious and controversial projects. But the modest constitutivism outlined in this chapter doesn’t share these aims. It looks to the constitutive aim of attitudes merely to explain rational requirements governing those attitudes.63 We aren’t using constitutive aims to derive normativity (in the reasons-providing sense), morality, or anything of the sort. In short, if we can successfully trace a rational requirement back to the ­constitutive aims of the attitudes governed by the requirement—that is, if something along the lines of the strategies Bratman and I have pursued can succeed—then we’ll have a source for non-conventional, non-normative requirements: the constitutive aims of the attitudes themselves. And that would allow us to reject the first premise in the argument above.

7.3 Conclusion This chapter has explored Bratman’s suggestion that just as there’s an aim of belief (truth) that can explain the rational requirements of consistency and 61  See Velleman 2000b. See also 2000a, pp. 26–9. On internal and external reasons, see Williams 1981. 62  Korsgaard writes: “A good action is one that constitutes its agent as the autonomous and efficacious cause of her own movements. These properties correspond, respectively, to Kant’s two imperatives of practical reason. Conformity to the categorical imperative renders us autonomous, and conformity to the hypothetical imperative renders us efficacious. These imperatives are therefore constitutive principles of action, principles to which we necessarily are trying to conform insofar as we are acting at all” (2009, xii). Our modest constitutivist project differs in that it’s concerned with the constitutive aim of intention, not action, and aspires to explain the rational requirements related to hypothethical imperatives, without aiming in any way to explain Kant’s categorical imperative. 63  Although the ambitious constitutivist program in moral philosophy isn’t a good model for the kind of view outlined in this chapter, better models are found elsewhere. Within legal philosophy, Lon Fuller has argued that there is an “internal morality of law”—that is, a set of standards (e.g., that law should be promulgated, should not be contradictory, should be clear, should not be retroactive, should not require the impossible) that can be derived from the very nature of law, which, for Fuller, is “the enterprise of subjecting human conduct to the governance of rules” (1969, p. 106). We might see Fuller’s project as seeing which normative (in the broader sense) standards can be derived from the constitutive aim of law. But Fuller didn’t think that all normative standards governing law are derived in this way: there is also an “external morality of law”—standards by which we could, for example, condemn some particular law as substantively unjust or harmful. So, Fuller’s view is a better model for the kind of view outlined here, though there are of course differences. (Additionally, if we held that there isn’t always a reason to comply with the “internal morality of law”—perhaps think of the harm of promulgating substantively unjust laws—then Fuller would provide us with an additional way of rejecting the first premise in the above argument: the requirements of law’s “internal morality” would then not be normative (in the reasons-providing sense), but nor would these requirements be based on some contingent social rule or convention; rather, the requirements would be derived from the very nature of law itself.)

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Intention  215 coherence governing beliefs, there’s an aim of intention (coordinated, ef­fect­ive control of action) that can explain the rational requirements of consistency and coherence governing intentions. Although I’ve taken issue with some aspects of Bratmans’s approach—particularly with his inability to explain the applicability of means–ends coherence to cases involving false instrumental beliefs, and his appeal to “coordination” as part of intention’s constitutive aim—I think Bratman’s suggestion provides a promising way to explain means–ends coherence. We do need to be cautious, however, since some of the ideas involved in the slogan “belief aims at truth” won’t easily transfer over to intentions, while others will. (Specifically, I’ve argued that we should focus on the success conditions, rather than the correctness conditions, of these attitudes.) By tracing rational requirements back to the constitutive aims of the attitudes they govern, we can allow for the possibility of rational requirements that are neither normative nor conventional. The advantage of this is that if it turns out that Weak Normativity is false, we won’t be forced into accepting a myth theory. I’ve suggested one particular way of using the constitutive aim of intentions and beliefs to explain means–ends coherence, which I’ve labeled “Non-normative Disjunctivism.” This view avoids the objections to Normative Disjunctivism, Strong Normativity, and cognitivist approaches to means–ends coherence. Also, it promises to explain several closely related rational requirements governing intentions and beliefs. It won’t explain every requirement of rationality, but that’s a lot to ask, and it doesn’t seem as though we have another view on the table that succeeds in this regard.

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

8 Conclusion In these very brief concluding remarks, I’ll sum up the central claims defended in this book, and return to the two sources of philosophical interest in means–ends coherence I had mentioned in the introductory chapter. Let’s start with the central claims. In Chapter 2, I agreed with those who think that intentions don’t “bootstrap” reasons into existence, but I argued that it’s difficult to say precisely what’s so objectionable about bootstrapping. Several versions of the bootstrapping objection in the literature seem too strong, since they overextend to rule out promises as a source of reasons. I’ve proposed recasting the objection as a worry about the incoherent advice that would be issued by our normative theory if intentions provided reasons. However, even though we should reject the idea that intentions provide ­reasons, we shouldn’t reject the idea that intentions put in place a rational requirement of means–ends coherence. (In Chapter  4, I argued against a “myth theory” about means–ends coherence, and in Chapter 7, I presented an account that would allow us to resist a temptation toward a myth theory if it turned out that there isn’t always a reason to be means–ends coherent.) But there’s a debate about how we should formulate this rational requirement of means–ends coherence between those who favor a “wide-scope” formulation and those who favor a “narrow-scope” formulation. In Chapter 3, I set forth what I hope is a strong case for the wide-scope formulation, while responding to the most significant objections raised by narrow-scopers. What’s the normative status of this wide-scope requirement? In Chapter 5, I rejected as unhelpful the Prichardian strategy of saying that the requirement provides “its own” reasons, and noted how it’s difficult to specify the reason to be means–ends coherent in general. I also considered, in both Chapter 4 and Chapter  5, three ways in which one could start with claims about the attitudes, or combinations of attitudes, one ought, or ought not, have, and then proceed to explain, or explain away, means–ends coherence. None of these strategies proved fruitful in the end. Normative Disjunctivism encounters several forceful objections, including the problem of mere permissibility. (That problem is made more trenchant by the results of Chapter 2, since we cannot Instrumental Rationality: The Normativity of Means–Ends Coherence. John Brunero, Oxford University Press (2020). © John Brunero. DOI: 10.1093/oso/9780198746935.001.0001

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

Conclusion  217 solve the problem by allowing intentions to bootstrap reasons into existence.) Strong Normativity faces the pragmatic reasons problem, and also runs into difficulty when combined with plausible transmission principles governing necessary and sufficient means. And, lastly, views which appeal to subjective reasons seem to either involve controversial theses about the nature of intention, or run into many problems similar to those facing the first two views. In Chapter  6, I considered and rejected cognitivist attempts to explain means–ends coherence by appealing to the rational requirements governing beliefs. But, in Chapter 7, I’ve suggested a new approach. Instead of explaining means–ends coherence by appealing to the attitudes or combinations of attitudes we ought, or ought not, have, or by appealing to rational norms governing beliefs, we would do better to follow up on Bratman’s neglected suggestion to look toward the constitutive aim of intention to help explain some of the rational requirements governing intentions. I’ve suggested a particular way of understanding constitutive aims (in terms of success conditions), and proposed a particular way of using constitutive aims (Non-normative Disjunctivism) to explain means–ends coherence and closely related requirements in a way that avoids some pitfalls of Bratman’s own approach. Those are the central claims of the book. Let’s now return to those two separate sources of philosophical interest in means–ends coherence that I identified in Chapter 1, and consider how these central claims relate to them. First, there was the older debate between those, like Mackie, who thought, roughly, that hypothetical imperatives were unproblematic while categorical imperatives were not, and those, like Hampton and Korsgaard, who thought that skepticism about the reality of moral requirements would translate into skepticism about the reality of instrumental rationality as well. Second, there was the more recent debate about the normativity of rationality—that is, the debate about whether we always ought (or, more weakly, have a reason) to comply with rational requirements. As a standard example of a rational requirement, means–ends coherence has figured prominently in this debate. How do the central claims of this book relate to these two debates? With regard to the second, more recent debate, we considered the relationship between rationality and normativity in Chapter  5, where I rejected Strong Normativity and raised some doubts about several attempts to specify the reason to be means–ends coherent in general. However, I’ve left it open whether Weak Normativity is true. But we can conclude that rationality isn’t strongly normative; in other words, it’s not the case that one always ought to do what one is rationally required to do.

OUP CORRECTED PROOF – FINAL, 24/02/20, SPi

218  Instrumental Rationality What about the first, older debate? Here my remarks are much more speculative. It seems that if the account given in Chapter 7 is correct—that is, if we’re able to explain means–ends coherence in terms of the constitutive aims of intention and belief—then we shouldn’t think that skepticism about the reality of moral requirements would translate into skepticism about the reality of means–ends coherence any more than it would translate into skepticism about the reality of requirements of chess, etiquette, law, or grammar. Although means–ends coherence can’t be traced to some social rule or convention, it can be explained in terms of the constitutive aims of the very attitudes the requirement governs. And I don’t think these constitutive aims—given how we’re understanding them in terms of the success conditions of the relevant attitudes—would raise the same metaphysical, epistemological, and motivational worries that led antirealists like Mackie to doubt that moral requirements are part of the fabric of the universe. Matters would be different if we understood means–ends coherence such that we ought not be means–ends incoherent in the same sense in which we ought not intentionally cause human suffering for fun—in that both would be contrary to the “ought” of practical reason. Then it would be unclear why skepticism about the latter claim about what we ought not do wouldn’t also involve us in ­skepticism about the former. But we’re rejecting Strong Normativity, and we don’t have to understand means–ends coherence in this way. As I said above, these remarks are speculative, and it would be inappropriate to get into the details of this meta-normative debate here. However, I hope that even if you’re not convinced by any of the central claims of this book, you are convinced that the project of understanding means–ends coherence—what it requires of us and why—is a philosophically interesting topic in its own right, independently of its bearing on these two debates. And I also hope to have made some small contribution to that project.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bibliography Adams, F., 2007, “Trying with the Hope” in M.  Timmons, J.  Greco, and A.  Mele (eds.), Rationality and the Good, Oxford: Oxford University Press. Anscombe, E., 1957, Intention, Oxford: Basil Blackwell. Archer, A., 2018, “Trying Cognitivism: A Defence of the Strong Belief Thesis” Theoria 84, 140–56. Audi, R., 1973, “Intending” Journal of Philosophy 70(13): 387–403. Audi, R., 1990, “Weakness of Will and Rational Action” Australasian Journal of Philosophy 68(3): 270–81. Bedke, M., 2009, “The Iffiest Oughts: A Guise of Reasons Account of End-Given Conditionals” Ethics 119(4): 672–98. Behrends, J., 2015, “Problems and Solutions for a Hybrid Approach to Grounding Practical Normativity” Canadian Journal of Philosophy 45(2): 159–78. Bratman, M., 1987, Intentions, Plans, and Practical Reason, Cambridge: Harvard University Press. Bratman, M., 2007, Structures of Agency: Essays, Oxford: Oxford University Press. Bratman, M., 2009a, “Intention, Belief, and Instrumental Rationality” in D.  Sobel and S. Wall (eds.), Reasons for Action, Cambridge: Cambridge University Press, 13–36. Bratman, M., 2009b, “Intention, Belief, Practical, Theoretical” in S.  Robertson (ed.), Spheres of Reason, Oxford: Oxford University Press, 29–62. Bratman, M., 2009c, “Intention, Practical Rationality, and Self-Governance” Ethics 119(3): 411–43. Bratman, M., 2009d, “Intention Rationality” Philosophical Explorations 12(3): 227–42. Broome, J., 2000, “Normative Requirements” in J. Dancy (ed.) Normativity, Oxford: Blackwell Publishers, 78–99. Broome, J., 2001, “Are intentions reasons? And how should we cope with incommensurable values?” in C. Morris and A. Ripstein (eds.), Practical Rationality and Preference: Essays for David Gauthier, Cambridge: Cambridge University Press, pp. 98–120. Broome, J., 2003, “Practical Reasoning” in J. Bermùdez and A. Millar (eds.), Reason and Nature: Essays in the Theory of Rationality, Oxford: Oxford University Press, pp. 85–112. Broome, J., 2005, “Does Rationality Give Us Reasons?” Philosophical Issues 15: 321–37. Broome, J., 2008a, “Is Rationality Normative?” Disputatio 11: 153–71. Broome, J., 2008b, “Reply to Southwood, Kearns and Star, and Cullity” Ethics 119: 96–108. Broome, J., 2013, Rationality Through Reasoning, Oxford: Wiley Blackwell. Brunero, J., 2007, “Are Intentions Reasons?” Pacific Philosophical Quarterly 88(4): 424–44. Brunero, J., 2009a, “Reasons and Evidence One Ought” Ethics 119(3): 538–45. Brunero, J., 2009b, “Against Cognitivism about Practical Rationality” Philosophical Studies 146(3): 311–25. Brunero, J., 2010a, “The Scope of Rational Requirements” Philosophical Quarterly 60(238): 28–49. Brunero, J., 2010b, “Self-Governance, Means-Ends Coherence, and Unalterable Ends” Ethics 120 (3): 579–91.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

220 Bibliography Brunero, J., 2012, “Instrumental Rationality, Symmetry, and Scope” Philosophical Studies 157(1): 125–40. Brunero, J., 2013, “Rational Akrasia” Organon F 20(4): 546–66. Brunero, J., 2015, “Review of Mark Schroeder, Explaining the Reasons We Share” Ethics 126(1): 238–44. Carroll, L., 1895, “What the Tortoise Said to Achilles” Mind 4(14): 278–80. Chalmers, D. J., 2011, “The Nature of Epistemic Space” in A. Egan and B. Weatherson, eds. Epistemic Modality, Oxford: Oxford University Press. Chan, D., 2016, Action Reconceptualized: Human Agency and its Sources, London: Lexington Books. Chan, T., 2013, “Introduction” in T. Chan, ed., The Aim of Belief, Oxford: Oxford University Press. Chang, R., 2009, “Voluntarist Reasons and the Sources of Normativity” in Reasons for Action, D. Sobel and S. Wall (eds.), New York: Cambridge University Press. Chang, R., 2013a, “Commitments, Reasons, and the Will” in R. Shafer-Landau (ed.) Oxford Studies in Metaethics, Vol. 8, Oxford: Oxford University Press, 74–113. Chang, R., 2013b, “Grounding Practical Normativity: Going Hybrid” Philosophical Studies 164: 163–87. Cheng-Guajardo, L., 2014, “The Normative Requirement of Means-Ends Rationality and Modest Bootstrapping” Ethical Theory and Moral Practice 17(3): 487–503. Chisholm, R., 1964, “The Descriptive Element in the Concept of Action” Journal of Philosophy 61(20): 613–25. Cullity, G., 2008, “Decisions, Reasons, and Rationality” Ethics, 119(1): 57–95. Dancy, J., 2000, Practical Reality, Oxford: Oxford University Press. Dancy, J., 2004, Ethics without Principles, Oxford: Oxford University Press. Dancy, J., 2009, “Reasons and Rationality” in S. Robertson (ed.) Spheres of Reason, Oxford: Oxford University Press, 93–112. Davidson, D., 1980a, “Intending” in Essays on Actions and Events, Oxford: Oxford University Press. Davidson, D., 1980b, “Freedom to Act” in Essays on Actions and Events, Oxford: Oxford University Press. Davidson, D., J. C. C. McKinsey and P. Suppes, 1955, “Outline of a Formal Theory of Value, I” Philosophy of Science 22(2): 140–60. Davis, W., 1984, “A Causal Theory of Intending” American Philosophical Quarterly 21(1): 43–54. Dreier, J., 2001, “Humean Doubts about Categorical Imperatives” in E.  Millgram (ed.) Varieties of Practical Reasoning, Cambridge, MA: MIT Press. Dworkin, R., 1967, “The Model of Rules” The University of Chicago Law Review 35: 14–46. Enoch, D., 2006, “Agency, Shmagency: Why Normativity Won’t Come from What Is Constitutive of Action” Philosophical Review 115(2): 169–98. Estlund, D., 2011, “What Good Is It? Unrealistic Political Theory and the Value of Intellectual Work” Analyse & Kritik 33(2): 395–416. Finlay, S., 2006, “The Reasons that Matter” Australasian Journal of Philosophy 84:1–20. Finlay, S., 2009, “Against All Reason? Scepticism about the Instrumental Norm” in  C.  Pigden (ed.) Hume on Motivation and Virtue, Basingstoke/New York: Palgrave-Macmillan. Finlay, S., 2010, “What Ought Probably Means, and Why You Can’t Detach It” Synthese 177(1): 67–89. Fogal, D., 2016, Bad Attitudes: Rationality and Its Discontents, Ph.D.  Dissertation, New York University, Department of Philosophy.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bibliography  221 Foot, P., 1972, “Morality as a System of Hypothetical Imperatives” Philosophical Review 81(3): 305–16. Frankfurt, H., 1988, “Identification and Wholeheartedness” in The Importance of What We Care About: Philosophical Essays, Cambridge, UK: Cambridge University Press, 159–76. Fuller, L., 1969, The Morality of Law: Revised Edition, New Haven: Yale University Press. Greenspan, P., 1975, “Conditional Oughts and Hypothetical Imperatives” The Journal of Philosophy 72(10): 259–76. Grice, P., 1971, “Intention and Uncertainty” Proceedings of the British Academy, 57: 263–79. Griffin, J., 1986, Well-Being, Oxford: Oxford University Press. Guindon, B., 2016, “Sources, Reasons, and Requirements” Philosophical Studies 173(5): 1253–68. Hampshire, S. and H. L. A. Hart, 1958, “Decision, Intention, and Uncertainty” Mind 67: 1–12. Hampton, J., 1998, The Authority of Reason, Cambridge, UK: Cambridge University Press. Harman, G., 1976, “Practical Reasoning” Review of Metaphysics, 29(3): 431–63. Harman, G., 1986, Change in View: Principles of Reasoning, Cambridge: MIT Press. Harman, G., 2001, “General Foundations versus Rational Insight,” Philosophy and Phenomenological Research 63(3): 657–63. Hedden, B., 2015, “Time-Slice Rationality” Mind 124(494): 449–91. Hill, T., 1973, “The Hypothetical Imperative” Philosophical Review 82(4): 429–50. Holton, R., 2004, “Rational Resolve” Philosophical Review 113(4): 507–35. Holton, R., 2008, “Partial Belief, Partial Intention” Mind 117(1): 27–58. Hornsby, J., 1995, “Reasons for Trying” Journal of Philosophical Research 20: 525–39. Hsieh, N., 2008, “Incommensurable Values,” in The Stanford Encyclopedia of Philosophy (Fall 2008 Edition), Edward  N.  Zalta (ed.), URL = http://plato.stanford.edu/archives/ fall2008/entries/value-incommensurable/ Hughes, G.  E. and M.  J.  Cresswell, 1996, A New Introduction to Modal Logic, London: Routledge. Hussain, N., unpublished manuscript, “The Requirements of Rationality”. Jackson, F. and R. Pargetter, 1986, “Ought, Options, and Actualism” Philosophical Review, 95(2): 233–55. Jollimore, T., 2005, “Why is Instrumental Rationality Rational?” Canadian Journal of Philosophy 35(2): 289–307. Kamm, F., 2000, “The Doctrine of Triple Effect and Why a Rational Agent Need Not Intend the Means to His End” Proceedings of the Aristotelian Society 74: 21–39. Kant, I., 1785/1997, Groundwork of the Metaphysics of Morals, (ed. Mary Gregor), Cambridge, UK: Cambridge University Press. Katsafanas, P., 2013, Agency and the Foundations of Ethics: Nietzschean Constitutivism, Oxford: Oxford University Press. Kavka, G., 1983, “The Toxin Puzzle” Analysis 43(1): 33–6. Kearns, S. and D. Star, 2008, “Reasons: Explanations or Evidence?” Ethics 119: 31–56. Kearns, S. and D.  Star, 2009, “Reasons as Evidence” in R.  Shafer-Landau (ed.), Oxford Studies in Metaethics, Vol. 4, Oxford: Oxford University Press, 215–42. Kelly, T., 2003, “Epistemic Rationality as Instrumental Rationality: A Critique” Philosophy and Phenomenological Research 66(3): 612–40. Kiesewetter, B., 2012, “A Dilemma for Parfit’s Conception of Normativity” Analysis 72(3): 466–74. Kiesewetter, B., 2015, “Instrumental Normativity: In Defense of the Transmission Principle” Ethics 125(4): 921–46. Kiesewetter, B., 2018, The Normativity of Rationality, Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

222 Bibliography Kolodny, N., 2005, “Why be Rational?” Mind 114(455): 509–63. Kolodny, N., 2007a, “State or Process Requirements?” Mind 116(462): 371–85. Kolodny, N., 2007b, “How Does Coherence Matter?” Proceedings of the Aristotelian Society 107(3): 229–63. Kolodny, N., 2008a, “The Myth of Practical Consistency” European Journal of Philosophy 16(3): 366–402. Kolodny, N., 2008b, “Why be Disposed to be Coherent” Ethics 118(3): 437–63. Kolodny, N., 2011, “Aims as Reasons” in R. J. Wallace, R. Kumar, and S. Freeman (eds.) Reasons and Recognition: Essays on the Philosophy of T.M. Scanlon, New York: Oxford University Press. Kolodny, N., 2018, “Instrumental Reasons” in D.  Star (ed.) The Oxford Handbook of Reasons and Normativity, Oxford: Oxford University Press. Kolodny, N. and J.  Brunero, 2013, “Instrumental Rationality” in E.  N.  Zalta (ed.), The Stanford Encyclopedia of Philosophy (Spring 2013 Edition). Korsgaard, C., 2008, “The Normativity of Instrumental Reason” in The Constitution of Agency: Essays on Practical Reason and Moral Psychology, Oxford: Oxford University Press. Korsgaard, C., 2009, Self-Constitution: Agency, Identity, and Integrity, Oxford: Oxford University Press. Lavin, D., 2004, “Practical Reason and the Possibility of Error” Ethics 114(3) 424–57. Lee, W., 2018, “Willing the End Means Willing the Means: An Overlooked Reading of Kant” Ergo 5(16): 419–44. Lewis, D., 1982, “Logic for Equivocators” Noûs 16(3): 431–41. Lewis, D., 1989, “Dispositional Theories of Value” Proceedings of the Aristotelian Society, Supplementary Volumes 63: 113–37. Liberman, A. and M. Schroeder, 2016, “Commitment: Worth the Weight” in E. Lord and B. Maguire (eds.) Weighing Reasons, Oxford: Oxford University Press. Lipsey, R.G. and K.  Lancaster, 1956–1957, “The General Theory of Second Best” The Review of Economic Studies 24(1): 11–32. Lord, E., 2014a, “The Coherent and the Rational” Analytic Philosophy 55(2): 151–75. Lord, E., 2014b, “The Real Symmetry Problem(s) for Wide-Scope Accounts of Rationality” Philosophical Studies 170(3): 443–64. Lord, E., 2017, “What You’re Rationally Required to Do and What you Ought to Do (Are The Same Thing!)” Mind 126(504): 1109–54. Lord, E., 2018, “The Explanatory Problem for Cognitivism about Practical Reason” in C. McHugh, J. Way, and D. Whiting (eds.) Normativity: Epistemic and Practical, Oxford: Oxford University Press. Ludwig, K., 1991, “Impossible Doings” Philosophical Studies 65(3): 257–81. Mackie, J., 1977, Ethics: Inventing Right and Wrong, Harmondworth: Penguin. McCann, H., 1991, “Settled Objectives and Rational Constraints” American Philosophical Quarterly 28(1): 25–36. McHugh, C. and J. Way, 2017, “Objectivism and Perspectivism about the Epistemic Ought” Ergo 4(5):121–45. Mele, A., 1992, “Intention, Belief, and Intentional Action” in Springs of Action, Oxford: Oxford University Press. Morton, J., 2011, “Towards an Ecological Theory of the Norms of Practical Deliberation” European Journal of Philosophy 19(4): 561–84. Morton, J., 2016, “Reasoning under Scarcity” Australasian Journal of Philosophy 95(3): 543–59. Morton, J. and S.  Paul, forthcoming, “Norms of Practical Reasoning” in R.  Chang and K. Sylvan (eds.) Routledge Handbook of Practical Reason, London: Routledge.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bibliography  223 Nagel, T., 1970, The Possibility of Altruism, Princeton: Princeton University Press. Parfit, D., 1984, Reasons and Persons, Oxford: Clarendon Press. Parfit, D., 1997, “Reason and Motivation” Proceedings of the Aristotelian Society, Supplementary Volume, 71(1): 99–130. Parfit, D., 2001, “Reasons and Rationality” in D. Egonsson, B. Petersson, J. Joselfsson, and T.  Ronnow-Rasmussen (eds.) Exploring Practical Philosophy: From Action to Values, Aldershot: Ashgate. Parfit, D., 2011, On What Matters, Vol. 1, Oxford: Oxford University Press. Paul, S. K., 2011, “Deviant Formal Causation” Journal of Ethics and Social Philosophy 5(3): 1–23. Peirce, C. S., 1878, “How to Make Our Ideas Clear” Popular Science Monthly 12: 286–302. Piller, C., 2013, “The Bootstrapping Objection” Organon F 20(4): 612–31. Platts, M., 1979, Ways of Meaning, London: Routledge and Kegan Paul. Podgorski, A., 2016, “A Reply to the Synchronist” Mind 125(499): 859–71. Podgorski, A., 2017, “Rational Delay” Philosophers’ Imprint 17(5): 1–19. Pollock, J. and J. Cruz, 1999, Contemporary Theories of Knowledge, 2nd Edition, Lanham, MD: Rowman and Littlefield Publishers. Prichard, H. A., 1949, “Does Moral Philosophy Rest on a Mistake?” in Moral Obligation: Essays and Lectures, Oxford: Clarendon Press, 1–17. Railton, P., 1986, “Moral Realism” The Philosophical Review 95(2): 163–207. Railton, P., 1994, “Truth, Reason, and the Regulation of Belief ” Philosophical Issues 5: 71–93. Raz, J., 1975, “Reasons for Actions, Decisions, and Norms” Mind 84 (336): 481–99. Raz, J., 1999a, “Incommensurability and Agency” in Engaging Reason, Oxford: Oxford University Press. Raz, J., 1999b, “Explaining Normativity: On Rationality and the Justification of Reason” in Engaging Reason, Oxford: Oxford University Press. Raz, J., 2005, “The Myth of Instrumental Rationality” Journal of Ethics and Social Philosophy, 1(1): 1–28. Reisner, A., 2008, “Weighing Pragmatic and Evidential Reasons for Belief ” Philosophical Studies 138(1): 17–27. Reisner, A., 2009, “Unifying the Requirements of Rationality” Philosophical Explorations 12(3): 243–60. Reisner, A., 2013, “Leaps of Knowledge” in T. Chan, ed., The Aim of Belief, Oxford: Oxford University Press. Rippon, S., 2011, “In Defense of the Wide-Scope Instrumental Principle” Journal of Ethics and Social Philosophy 5(2): 1–21. Rippon, S., 2014, “Were Kant’s Hypothetical Imperatives Wide-Scope Oughts?” Australasian Journal of Philosophy 92(4): 783–8. Rosati, C., 2011, “The Importance of Self-Promises” in H.  Sheinman (ed.) Promises and Agreements: Philosophical Essays, Oxford: Oxford University Press, 124–55. Ross, A., 1941, “Imperatives and Logic” Theoria 7(1): 53–71. Ross, J., 2009, “How to be a Cognitivist about Practical Reason” in R. Shafer-Landau (ed.) Oxford Studies in Metaethics, Vol. 4, Oxford: Oxford University Press, 243–82. Ross, J., 2012, “Rationality, Normativity, and Commitment” in R.  Shafer-Landau (ed.) Oxford Studies in Metaethics, Vol. 7, Oxford: Oxford University Press, 138–81. Ross, W.  D., 1930, The Right and the Good, P.  Stratton-Lake (ed.), Oxford: Oxford University Press. Scanlon, T., 1998, What We Owe to Each Other, Cambridge, MA: The Belknap Press of Harvard University Press.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

224 Bibliography Scanlon, T., 2004, “Reasons: A Puzzling Duality?” in Reasons and Value: Themes from the Moral Philosophy of Joseph Raz, R. J. Wallace, P. Pettit, S. Scheffler, and M. Smith (eds.), Oxford: Oxford University Press. Scanlon, T., 2008, Moral Dimensions: Permissibility, Meaning, Blame, Cambridge, MA: The Belknap Press of Harvard University Press. Schauer, F. and W.  Sinnott-Armstrong, 1996, The Philosophy of Law: Classic and Contemporary Readings with Commentary, Fort Worth, TX: Harcourt Brace. Schelling, T., 1960, The Strategy of Conflict, Cambridge, MA: Harvard University Press. Schroeder, M., 2004, “The Scope of Instrumental Reason” Philosophical Perspectives, 18: 337–64. Schroeder, M., 2005, “Instrumental Mythology” Journal of Ethics and Social Philosophy, Symposium 1: 1–12. Schroeder, M., 2005b, “The Hypothetical Imperative?” Australasian Journal of Philosophy 83(3): 357–72. Schroeder, M., 2007, Slaves of the Passions, Oxford: Oxford University Press. Schroeder, M., 2008, Being For, Oxford: Oxford University Press. Schroeder, M., 2009, “Means-End Coherence, Stringency, and Subjective Reasons” Philosophical Studies 143(2): 223–48. Schroeder, M., 2012, “The Ubiquity of State-Given Reasons” Ethics 122(3): 457–88. Schroeder, M., 2014a, “Hypothetical Imperatives, Scope, and Jurisdiction” in Explaining the Reasons We Share: Explanation and Expression in Ethics, Vol. 1. Oxford: Oxford University Press, 216–26. Schroeder, M., 2014b, “Scope for Rational Autonomy” in Explaining the Reasons We Share: Explanation and Expression in Ethics, Vol. 1. Oxford: Oxford University Press, 227–39. Schroeder, M., 2014c, “Introduction” in Explaining the Reasons We Share: Explanation and Expression in Ethics, Vol. 1. Oxford: Oxford University Press, 1–18. Schwitzgebel, E., 2015, “Belief ” in E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy (Summer 2015 Edition). Setiya, K., 2007, “Cognitivism about Instrumental Reason” Ethics, 117(4): 649–73. Setiya, K., 2014, “Intention,” in E. N. Zalta (ed.) The Stanford Encyclopedia of Philosophy (Spring 2014 Edition). Shafer-Landau, R., 2003, Moral Realism, Oxford: Oxford University Press. Shafer-Landau, R., 2006, “Ethics as Philosophy: A Defense of Ethical Nonnaturalism” in T. Horgan and M. Timmons (eds.) Metaethics After Moore, Oxford: Oxford University Press. Shah, N., 2006, “A New Argument for Evidentialism” Philosophical Quarterly 56(225): 481–98. Shah, N., 2008, “How Action Governs Intention” Philosophers’ Imprint 8(5): 1–19. Shah, N. and D.  Velleman, 2005, “Doxastic Deliberation” Philosophical Review 114(4): 497–534. Shpall, S., 2013, “Wide and Narrow Scope” Philosophical Studies 163(3): 717–36. Shpall, S., 2014, “Moral and Rational Commitment” Philosophy and Phenomenological Research 88(1): 146–72. Smith, H., 1997, “A Paradox of Promising” Philosophical Review 106(2): 153–96. Smith, M., 1994, The Moral Problem, Oxford: Blackwell Publishers. Smith, M., 1995, “Internal Reasons” Philosophy and Phenomenological Research 55(1): 109–31. Smith, M. B. E., 1973, “Is There a Prima Facie Obligation to Obey the Law?” The Yale Law Journal 82(5): 950–67.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

Bibliography  225 Smith, M.  N., 2016, “One Dogma of Philosophy of Action” Philosophical Studies 173: 2249–66. Sobel, D., 2009, “Subjectivism and Idealization” Ethics 119(2): 336–52. Southwood, N., 2008, “Vindicating the Normativity of Rationality” Ethics 119(1): 9–30. Titelbaum, M., 2015, “How to Derive a Narrow-scope Requirement from Wide-scope Requirements” Philosophical Studies 172(2): 535–42. Velleman, J. D., 1989, Practical Reflection, Princeton: Princeton University Press. Velleman, J. D., 2000a, “Introduction” in The Possibility of Practical Reason, Oxford: Oxford University Press, 1–31. Velleman, J. D., 2000b, “The Possibility of Practical Reason” in The Possibility of Practical Reason, Oxford: Oxford University Press, 170–99. Velleman, J. D., 2000c, “The Aim of Belief ” in The Possibility of Practical Reason, Oxford: Oxford University Press, 244–81. Velleman, J. D., 2007, “What Good is a Will?” in A. Leist & H. Baumann (eds.), Action in Context, Berlin: de Gruyter, 193–215. Verbeek, B., 2014, “On the Normativity of Intentions” Topoi: An International Review of Philosophy 33(1): 87–101. Vranas, P., 2007, “I Ought, Therefore I Can” Philosophical Studies 136(2): 167–216. Wallace, R. J., 2001, “Normativity, Commitment, and Instrumental Reason” Philosophers’ Imprint 1(4): 1–26. Wallace, R. J., 2006, Normativity and the Will, Oxford: Clarendon Press. Watson, G., 1975, “Free Agency” Journal of Philosophy 72 (8): 205–20. Way, J., 2009, “Two Accounts of the Normativity of Rationality” Journal of Ethics and Social Philosophy, Discussion Note. Way, J., 2010, “Defending the Wide-Scope Approach to Instrumental Reason” Philosophical Studies 147(2): 213–33. Way, J., 2011, “The Symmetry of Rational Requirements” Philosophical Studies 155(2): 227–39. Way, J., 2012a, “Explaining the Instrumental Principle” Australasian Journal of Philosophy 90(3): 487–506. Way, J., 2012b, “Transmission and the Wrong Kind of Reason” Ethics 122(3): 489–515. Way, J., 2013, “Intentions, Akrasia, and Mere Permissibility” Organon F 20(4): 588–611. Way, J., 2018, “Reasons and Rationality” in D. Star (ed.) The Oxford Handbook of Reasons and Normativity, Oxford: Oxford University Press. Wedgwood, R., 2002, “The Aim of Belief ” Philosophical Perspectives 36(s16): 267–97. Wedgwood, R., 2007, The Nature of Normativity, Oxford: Oxford University Press. Wedgwood, R., 2011, “Instrumental Rationality” in R. Shafer-Landau (ed.) Oxford Studies in Metaethics, Vol. 6, Oxford: Oxford University Press, 280–309. Wiland, E., 2012, Reasons, London: Continuum Books. Williams, B., 1973, “Deciding to Believe” in Problems of the Self, Cambridge: Cambridge University Press. Williams, B., 1981,“Internal and External Reasons” in Moral Luck, Cambridge: Cambridge University Press, 101–13. Worsnip, A., 2018, “What is (in)coherence?” in R. Shafer-Landau (ed.) Oxford Studies in Metaethics, Vol. 13, Oxford: Oxford University Press. Zhu, J., 2010, “On the Principle of Intention Agglomeration” Synthese 175(1): 89–99.

OUP CORRECTED PROOF – FINAL, 18/03/20, SPi

OUP CORRECTED PROOF – FINAL, 25/02/20, SPi

Index Anscombe, E.  156–7, 182–3 asymmetry  76–88, 175 escape vs. compliance  82–8 Audi, R.  160–3 belief aim of, see constitutive aims closure requirement on  154, 171, 205 consistency of, see consistency Strong Belief Thesis  152, 154–67 Very Weak Belief Thesis  171 bootstrapping  33–41, 92, 107, 141n.51, 200 Bratman, M.  5, 26–7, 33, 35–6, 86, 126–30, 153, 178–81, 187, 189–94, 201–3 Broome, J.  26–7, 33–4, 36–7, 84–5, 91n.8, 115–16, 119, 138–40 Carroll, L.  16 Certain Failure  201–2 Chang, R.  26, 46–51 Chisholm, R.  195 constitutive aims  177, 212–13 of belief  177, 180–8 of intention  177–9, 187–97 see also constitutivism see also correctness conditions see also success conditions constitutivism 213–14 controlled action  195–7 correctness conditions  182–4 Cullity, G.  35 cognitivism 152–4 see also non-cognitivism coordination 159–61 consistency of belief  3, 59, 85–6, 177, 203 of intention  2, 85–6, 133–4, 152, 161–4, 173–4, 177, 191–4, 203

intention-belief  86–9, 136n.38, 177–8, 199, 203 contingency planning  162 Dancy, J.  1 Davidson, D.  195 Deontic Inheritance  113n.49, 138n.42 direction of fit  182 Dreier, J.  16 enkratic requirement  60, 69–70, 205–6 fallacy of division  193 Finlay, S.  2n.4, 208n.54 Foot, P.  10–11, 90–1, 210–11 foreseen side-effects  20–2 Frankfurt, H.  126–30 Fuller, L.  214n.63 Hampton, J.  16, 18, 217 Harman, G.  37n.32, 114–15, 153, 164–7 Hussain, N.  121–4 hypothetical imperatives  8–15 vs. means-ends coherence  13–15 idealization 101–5 identification 126–30 incoherent advice  39–41, 107–8 intentions normative significance of  25–7 as “tie-breakers”  26, 31, 106–7 promotional significance of, see Promotional Significance View providing reasons  30–1, 106–7, 147 features of  31–2, 156–64, 187 self-justification of  36–8, 43–6 persistence requirement  45 unalterable  96–7, 111, 137–41, 200

OUP CORRECTED PROOF – FINAL, 25/02/20, SPi

228 Index intentions (cont.) to try  164–6 ambiguity of  166–7 ignorance of  172–3 agglomeration of  173n.36 conditional 163n.20 aim of, see constitutive aims intransitive preferences  205–6 instrumental belief  19–22, 82–8, 189 irrationality costs of  124–6 advantages of  132–7 Jackson, F. and R. Pargetter  55, 109, 138–40 Kant, I.  8–15, 113, 117 Kearns, S. and D. Star  29 Kiesewetter, B.  78n.40, 114–17 Kolodny, N.  17–18, 60–1, 65, 67, 72, 92–3, 108, 119–20, 124, 201 Korsgaard, C.  15–16, 25–6, 32–3, 36–7, 47, 213–14, 217 Liberman, A.  71–3 Ludwig, K.  165 Lord, E.  73–6, 82–8, 107 Mackie, J.  15, 217 mere permissibility  105–8 Morton, J.  91n.4 Moore’s paradox  156–7, 184–5 myth theory  111–17, 179–80, 209–14 Nagel, T.  11 non-cognitivism  177, 180–1, 201 see also cognitivism Non-normative Disjunctivism  117, 178, 190–1, 193–4, 197–214 vs. Normative Disjunctivism  178, 200–2 normativity senses of  4, 90 transmission of, see transmission of rationality  17–19, 90–3, 209–11, 217 strong vs. weak  90–1, 119 Normative Disjunctivism  25, 93–111 Normative Conjunctive Disjunctivism 109–11 see also Non-normative Disjunctivism

Parfit, D.  28, 52, 54, 145–6 Prichard, H.  7, 118, 120–4 pressure 207–9 Professor Procrastinate  55, 109–10, 138–40 promises  33–6, 38 Promotional Significance View  27, 55–8 Railton, P.  101, 185 rationality as attitudinal coherence  1–4, 155, 172 requirements of, see rational requirements scope of  15–17, 217–18 normativity of, see normativity disadvantageous  97–101, 130–7 dispositions toward  90–1, 116–17 reasons of  121–4 rational requirements  2–3 synchronic vs. diachronic  62–7 applicability of  63–5 local 67 wide-scope formulations  68–71, 143–4, 158 and weighing  71–3 and cancelling conditions  73–6 and symmetry, see asymmetry satisfaction of  112–13 strictness of  119 Raz, J.  48–9, 92–3, 118, 142–4 Ross, J.  132–7, 205–6 reasoning reflection vs. revision  114–15 and satisfaction of requirements  115–16 and asymmetry  76–8; see also asymmetry reasons normative vs. motivating  28 distinguished from rational requirements 29–30 as primitive  28–9 as evidence  29n.14 transmission of, see Transmission wrong kind and right kind  99–100, 131 object-given and state-given  100–1, 131, 138–40 to be rational  119–30 subjective, see subjective oughts Scanlon, T.  26, 28, 51–4 Schelling, T.  98–9, 131

OUP CORRECTED PROOF – FINAL, 25/02/20, SPi

Index  229 Schroeder, M.  25–6, 32–3, 36–7, 65, 71–3, 92–3, 118–19, 141–2, 146–8 self-governance 126–30 self-trust  22–4, 172 Setiya, K.  22–4, 96, 111, 118, 137–41, 153, 200 Sobel, D.  102–3, 105 Southwood, N.  121–4 subjective oughts  145–50 success conditions  185–8 Toxin Puzzle  39, 100–1, 131 transmission of reasons  11–12 of oughts to necessary means  40, 137–41

of reasons to sufficient means  141–5 Transparency Account  207–8 underdetermination  47–51, 106–7, 147 see also mere permissibility Velleman, D.  153, 156–64, 180, 185–7, 196, 201–5, 213–14 voluntarism 47–51 Wallace, J.  153, 167–72 Way, J.  92–3, 110–11, 118–19, 124–5, 138–40, 146n.65, 148–50, 211 Wedgwood, R.  182, 187–8 Williams, B.  102, 145, 180, 184–5, 213–14 Worsnip, A.  207n.51