Dear Prudence: The Nature and Normativity of Prudential Discourse 0198858264, 9780198858263

Philosophers have long theorized about what makes people's lives go well, and why, and the extent to which morality

154 35 1MB

English Pages 224 [223] Year 2021

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Cover
Dear Prudence: The Nature and Normativity of Prudential Discourse
Copyright
Dedication
Contents
Acknowledgements
Introduction
Part One: Prudence, Prudential Discourse, and Normativity
Part Two: The Nature of Prudential Discourse
Part Three: Prudential Discourse Is Normative. What Follows?
Part One: Prudence, Prudential Discourse, and Normativity
Chapter 1: Prudence as a System of Categorical Imperatives
1. Introduction
2. Grounds for Rejecting PVM (I): Prudential Error Theory
3. Grounds for Rejecting Prudential Value Matters (II): No Agent-Relative Reasons
4. Grounds for Rejecting PVM: (III) Humeanism about Reasons
5. Humeanism vs PVM
6. Conclusion
Chapter 2: Is Prudential Discourse Normative?
1. Introduction
2. The Parity Strategy
3. Markers of the Normative
3.1 Evaluative and Prescriptive
3.2 Authority
3.3 Affective Connection
3.4 Disagreement
3.5 Testimony
4. Why Think Prudential Judgements Are Not Normative Judgements?
5. Because Prudential Value Is Desire-Based?
6. Because Prudential Properties Reduce to Non-Normative Properties?
7. Because Prudential Judgements Are Invulnerable to Error Theory?
8. Conclusion
Part Two: The Nature of Prudential Discourse
Chapter 3: Prudential Language and Context (I): Good for and Needs
Part A—Evaluative Prudential Language: Good For
1. Introduction
2. Some Data about ‘Good for’ Talk: Unity and Diversity
3. First Proposal: ‘Good for’ and Attributive Goodness
4. A Better Account of ‘Good for’: End-Relationalism
5. Kraut’s Developmentalism versus End-Relationalism
6. Conclusion
Part B—Directive Prudential Claims
1. Introduction
2. ‘Needs’ Discourse
3. From ‘Must’ to ‘Needs’
4. ‘Must’ and ‘Needs’ in Prudential Discourse
5. Conclusion
Chapter 4: Prudential Language and Context (II): Contextualism and Pluralism
1. Introduction
2. Alexandrova’s Contextualism
3. Assessing Alexandrovan Contextualism
3.1 Threshold Dependence
3.2 Constitutive Dependence
4. Campbell’s Conceptual Pluralism
5. Assessing Campbell’s Case
5.1 Putative Conflict (I): Relation to the Subject’s Attitudes
5.2 Putative Conflict (II): Relation to Morality
5.3 Putative Conflict (III): Experience versus Desire Satisfaction
6. Conclusion
Chapter 5: Prudential Judgements and Motivation
1. Introduction
2. Prudential Judgements and Motivation
3. Objections to PJI
3.1 Objection 1: Pragmatic Explanations
3.2 Objection 2: Illicit Intuitions
4. A More Ambitious Thesis than PJI?
4.1 Possibility 1: Explanation by Conative State of Mind
4.2 Possibility 2: Explanation by Rational Requirement
5. A More Ambitious Thesis than PJI (Continued)?
5.1 A More Ambitious Thesis: Intra-prudentially
5.2 A More Ambitious Thesis: Inter-personally
6. Small Amendments to PJI
7. Implications of PJI 4: Cognitivism and the Alienation Constraint
7.1 Non-Cognitivismabout Prudential Judgements
7.2 The Anti-Alienation Constraint on Prudential Value
Part Three: Prudential Discourse is Normative. What Follows?
Chapter 6: Prudential Normativity, Moral Scepticisms, and Metaethics
1. Introduction
2. Why Be Moral and the Normativity of Prudence
3. A Prudential Companions-in-Guilt Argument
4. Why Be a Moralizer? Prudential Normativity and Revisionary Metaethical Views
5. Thick Concepts and Prudential Normativity
6. Conclusion
Chapter 7: Prudential Normativity: Realism and Anti-Realism
1. Introduction
2. First and Second-Order Questions about Morality
2.1 First-Order Question
2.2 Second-Order Question
3. First and Second-Order Questions about Prudential Normativity
4. Anti-Alienation and Anti-Realism
4.1 Sumner and Rosati’s Arguments for the Subject-Relativity of Prudential Value
5. Subjective Theories and Anti-Realism
6. Why It Would Be Better to Focus on Second-Order Prudential Questions
7. Conclusion
Chapter 8: Conclusion: Taking Stock and Looking Ahead
References
Index
Recommend Papers

Dear Prudence: The Nature and Normativity of Prudential Discourse
 0198858264, 9780198858263

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Dear Prudence

Dear Prudence The Nature and Normativity of Prudential Discourse GUY FLETCHER

1

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 © Guy Fletcher 2021 The moral rights of the author have been asserted First Edition published in 2021 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: 2021934209 ISBN 978–0–19–885826–3 DOI: 10.1093/oso/9780198858263.001.0001 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

For Debbie, Eli, and Gwen

Contents Acknowledgements

xi

Introduction1 PA RT O N E : P RU D E N C E , P RU D E N T IA L D I S C OU R SE , A N D N O R M AT I V I T Y 1. Prudence as a System of Categorical Imperatives

1. Introduction 2. Grounds for Rejecting PVM (I): Prudential Error Theory 3. Grounds for Rejecting Prudential Value Matters (II): No Agent-Relative Reasons 4. Grounds for Rejecting PVM: (III) Humeanism about Reasons 5. Humeanism vs PVM 6. Conclusion

2. Is Prudential Discourse Normative?

13

13 17 20 22 25 32

33

1. Introduction 33 2. The Parity Strategy 35 3. Markers of the Normative 37 4. Why Think Prudential Judgements Are Not Normative Judgements?44 5. Because Prudential Value Is Desire-Based? 45 6. Because Prudential Properties Reduce to Non-Normative Properties?48 7. Because Prudential Judgements Are Invulnerable to Error Theory? 52 8. Conclusion 54

PA RT T WO : T H E NAT U R E O F P RU D E N T IA L D I S C OU R SE 3. Prudential Language and Context (I): Good for and Needs Part A—Evaluative Prudential Language: Good For 1. Introduction

59

59

59

viii Contents 2.  Some Data about ‘Good For’ Talk: Unity and Diversity 3.  First Proposal: ‘Good for’ and Attributive Goodness 4.  A Better Account of ‘Good for’: End-Relationalism 5.  Kraut’s Developmentalism versus End-Relationalism 6. Conclusion

Part B—Directive Prudential Claims

1. Introduction 2.  ‘Needs’ Discourse 3.  From ‘Must’ to ‘Needs’ 4.  ‘Must’ and ‘Needs’ in Prudential Discourse 5. Conclusion

4. Prudential Language and Context (II): Contextualism and Pluralism

62 67 70 77 80

81

81 82 86 90 90

92

1. Introduction 2. Alexandrova’s Contextualism 3. Assessing Alexandrovan Contextualism 4. Campbell’s Conceptual Pluralism 5. Assessing Campbell’s Case 6. Conclusion

92 93 97 108 109 118

5. Prudential Judgements and Motivation

120

1. Introduction 120 2. Prudential Judgements and Motivation 121 3. Objections to PJI 124 4. A More Ambitious Thesis than PJI? 128 5. A More Ambitious Thesis than PJI (Continued)? 135 6. Small Amendments to PJI 138 7. Implications of PJI 4: Cognitivism and the Alienation Constraint139

PA RT T H R E E : P RU D E N T IA L D I S C O U R SE I S N O R M AT I V E . W HAT F O L L OWS ? 6. Prudential Normativity, Moral Scepticisms, and Metaethics 145 1. Introduction 2. Why Be Moral and the Normativity of Prudence 3. A Prudential Companions-in-Guilt Argument 4. Why Be a Moralizer? Prudential Normativity and Revisionary Metaethical Views 5. Thick Concepts and Prudential Normativity 6. Conclusion

145 146 150 159 164 166

Contents  ix

7. Prudential Normativity: Realism and Anti-Realism

167

8. Conclusion: Taking Stock and Looking Ahead

189

References Index

195 207

1. Introduction 167 2. First and Second-Order Questions about Morality168 3. First and Second-Order Questions about Prudential Normativity 170 4. Anti-Alienation and Anti-Realism 178 5. Subjective Theories and Anti-Realism 183 6. Why It Would Be Better to Focus on Second-Order Prudential Questions 186 7. Conclusion 188

Acknowledgements Chapters 1 & 2 partially overlap with ‘Taking Prudence Seriously’, in R. Shafer-­Landau (ed.), Oxford Studies in Metaethics: Volume 14 (Oxford University Press, 2019), 70–94. Chapter 3 partially overlaps with ‘Needing and Necessity’, in M. Timmons (ed.), Oxford Studies in Normative Ethics: Volume 8 (Oxford University Press, 2018), 170–92. Chapter  4 largely overlaps with ‘Against Contextualism about Prudential Discourse’, Philosophical Quarterly 69 (277) (2019): 699–720. This ma­ter­ial appears here by permission of Oxford University Press. Chapter 6 partially overlaps with ‘Pain for the Moral Error Theory? A New Companions-­in-­Guilt Argument’, Australasian Journal of Philosophy 96 (3) (2018): 474–82. This material appears here by permission of Taylor & Francis. I have incurred many debts, of many sorts, in writing this book. To keep this to a manageable length I will, with one exception, thank a person or organization only once, despite the fact that a great many people merit multiple appearances. I have been thinking about these issues since I was a PhD student from 2006 to 2009 in the Philosophy department at the University of  Reading. I must thank my supervisors—Brad Hooker and Philip Stratton-­Lake—especially as well as the other faculty and fellow graduate students for help and for making being a PhD student there such a wonderful experience. In more recent times, I have been fortunate enough to be at Edinburgh, where a number of my faculty and student colleagues— particularly James Brown, Michael Cholbi, Matthew Chrisman, Jonny Cottrell, Giles Howdle, Ellie Mason, Maggie O’Brien, Mike Ridge (who kindly read a full draft), Debbie Roberts, and Chad Stevenson—have been invaluable in reading, commenting upon, and discussing this work with me, sometimes more than they would have wished to, perhaps, over a sustained period. Other people who have been generous with their time, either in ­discussion, or in reading drafts, and just helpfully encouraging me in the

xii Acknowledgements idea that this project was worthwhile include: Anna Alexandrova, Krister Bykvist, Steve Campbell, David Copp, Chris Cowie, Stephen Darwall, Dale Dorsey, Jake Edwards, Steve Finlay, Alex Gregory, Daniel Groll, Louise Hanson, Dan Haybron, Max Hayward, Jessica Isserow, Guy Kahane, Eden Lin, Kathryn Lindeman, Brian McElwee, Euan MacDonald, Filipa Melo Lopes, Peter Momtchiloff, Alex Murphy, Jonas Olson, Bryan Pickel, David Plunkett, Brian Rabern, Jason Raibley, Connie Rosati, Richard Rowland, Alex Sarch, Justin Snedegar, Sarah Stroud, Bart Streumer, Nandi Theunissen, Teemu Toppinen, Joey Van Weelden, Pekka Väyrynen, Jonathan Way, Daniel Whiting, Eric Wiland, Daniel Wodak, Jack Woods, Alex Worsnip, and two referees for the publisher, one of whom remains anonymous to me (though I have my suspicions) and whose comments really went above and beyond. I must also thank Phil Dines for meticulous copyediting of the final manuscript. I must thank audiences from (at least): Montreal, Southampton, St Louis, Geneva (twice), Trinity College Dublin, Syracuse, Paris, California State Long Beach, the Groningen Metaethics Workshop, the Madison Metaethics Workshop, the Arizona Workshop in Normative Ethics, The Prudential Value Workshop in Edinburgh, the Ethics Seminar in Edinburgh, The Companions in Guilt Conference in Rome, for opportunities to present this work and for stimulating discussion of it. I must also thank the British Academy for the postdoctoral fellowship that supported the early stages of this project and the School of PPLS at Edinburgh, who provided invaluable sabbatical time. Various non-­academic events have delayed this book—multiple bouts of industrial action, unscheduled back surgery, months of home-­schooling, and probably more—but here, right in the middle of a global pandemic, I have finally finished it. Luckily for me, there are no such things as omens. Finally, my greatest debt is to Debbie Roberts, who has inspired, assisted, and tolerated this project to superhuman levels. She has saved me from errors, provided innumerable helpful suggestions, read mul­ tiple drafts of multiple chapters, and, over many years, indulged my fas­ cin­ation with these issues (and ability to turn discussions of many other topics into discussion of them) more than anyone could deserve. I dedicate this book, with love, to her.

Introduction For as long as I can remember I have held firm views about how people ought to live. Some were moral views. Others were about how people ought to live with respect to themselves and making their own lives go best. Luckily, I stumbled upon the part of philosophy where I could express and defend such views: the theory of well-­being or prudential value. Philosophers have long theorized about which things make people’s lives go well, and why, and the extent to which morality and self-­interest can be reconciled. By contrast, we have spent little time on m ­ eta-­prudential questions, questions about prudential discourse.1 This is surprising given that prudence is, prima facie, a normative form of discourse and, as such, cries out for further investigation of how exactly it functions and whether it has problematic commitments.2 It also marks a stark contrast from moral discourse, about which there has been extensive theorizing. By ‘prudential discourse’ I mean two things. Firstly, thought and talk about what’s good/better/best for or bad/worse/worst for, harms, benefits (etc.) someone. Secondly, thought and talk about what someone prudentially ought to do, has reason to do, must do, or needs to do, along with analogous claims about attitudes.3 Prudential discourse thus encompasses evaluative prudential discourse (‘good for’ etc.) and dir­ect­ ive prudential discourse (‘mustprudential’ etc.). Here are some paradigmatic instances:

1  Some exceptions to this are Railton (1989), Hooker (1991), Rosati (1995a; 1995b; 1996), and Darwall (2004). 2  Much of the history of moral philosophy presumes that prudential normativity is genuine and more easily justifiable than moral normativity. C.f. Darwall (2016). 3  A complication: ‘good for’ (etc.) talk extends beyond welfare subjects (beings with levels of well-­being). For example: ‘oil is good for engines’, ‘sand is bad for watches’. How to demarcate the boundary between these uses of ‘good for’ and those connected to well-­being is an issue that I bracket here but discuss later (Chapter  3, Part A). See Kraut (2007), Fletcher (2012), Finlay (2014), and Rosati (2020). Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0001

2  Dear Prudence ‘You’ll be better off if you have the surgery.’ ‘Her spouse’s death was terribly bad for her.’ ‘Sarah should take the job.’ ‘You need friends in order to live well.’

The lack of explicit attention to prudential discourse is surprising for at least three reasons. First, because such discourse is so important. We care deeply about how our lives go—beyond how they go morally speaking—and so prudential discourse is a central feature of human life. The inattention is also surprising because a better understanding of such discourse might help us to make progress (of at least some sort) in thinking about what makes people’s lives go best and why. Finally, prudential discourse is, prima facie, a normative form of discourse and these forms of discourse cry out for investigation of how they function and whether they have problematic commitments. Peter Railton succinctly presents some of the puzzles generated by prudential discourse thus: Statements about a person’s good slip into and out of our ordinary discourse about the world with nary a ripple4 [. . .] Statements about a person’s good present themselves as being about something with respect to which our beliefs can be true or false, warranted or unwarranted. Let us speak of these features as the descriptive side of discourse about a person’s good [. . .] Discourse about a person’s good also has another, prescriptive side, for to make or accept a judgment that something is good for someone is in some sense to recommend that thing to him and to those who care about him [. . .] Philosophers have been struck by the difficulty of understanding how evaluative language, such as discourse about a person’s good, could be at once essentially descriptive and essentially prescriptive.

4  Railton (1989: 155) is explicit here that he means well-­being or prudential value by ‘a person’s good’: ‘We are here occupied in this paper with one species of value, that which is realized in a person’s own good, sometimes called his well-­being or welfare.’

Introduction  3 Prudential discourse is thus philosophically interesting and practically important yet under-­explored. Having put prudential discourse into the foreground, here are some big-­picture meta-­prudential questions that we can ask about it: 1. Are there normative prudential reasons? 2. Are prudential judgements normative judgements? 3. Are these judgements motivationally special? 4. Is prudential discourse context-­sensitive in significant ways? 5. Does prudential discourse involve the ascription of normative properties (in either a deflationary sense or a robust sense) and, if robust, what is the nature of prudential properties? 6. Is prudential talk partly (or wholly) expressive? My aims in this book are threefold. I want, firstly, to make progress in answering some of these questions by defending particular views in response to them. Secondly, I want to motivate the project of investigating all of these questions (and more). That is, I want to show that they are worth further consideration. So even if you disagree with my specific answers to some of them, I will succeed, in part, if I convince you that they are worthy of investigation. A third aim of this book is to demonstrate two kinds of continuity between prudential discourse and other kinds of discourse. The first kind of continuity is that the same interesting questions and issues arise for prudential discourse. The second kind of continuity is that insights from and theories developed in other areas of philosophy can be usefully applied to prudential discourse and vice versa. As an example of the first continuity, I argue in Chapter  5 that the long-­standing debate about the connection between moral judgement and motivation arises also in the case of prudential judgements. Put another way, however plausible it is that moral judgements necessarily motivate, it is at least equally plausible that (at least some) prudential judgements necessarily motivate. One example of the second kind of continuity is the case I make in Chapter 3 for forms of contextualism about ‘good for’ talk and ‘needs’ talk. Therein I show how insights from the works of Stephen Finlay and

4  Dear Prudence Angelika Kratzer can be fruitfully applied to areas of prudential discourse, thereby improving our understanding of it. These meta-­prudential questions, and implications of the answers I defend, will occupy me for the rest of this book. I propose to answer them by giving a theory of various aspects of prudential discourse and by also arguing (in Chapter 2 and elsewhere) that the nature of prudential discourse means that it is a normative form of discourse exactly on a par with moral discourse. My way of setting up the issues is designed to isolate and separate two sources of disagreement. Thus, one is free to agree with at least some of my more specific claims about the nature of prudential discourse whilst rejecting the claim that such discourse is normative (and vice versa). One might suggest,5 sceptically, that this project is unnecessary on the grounds that prudential discourse has already been theorized about, given the recent tendency for people to talk in terms of the m ­ eta-normative, rather than the metaethical in particular. This deflationary suggestion embodies three mistakes. First, little attention has been paid to whether prudential discourse in particular is normative and thus to-­be-­included within a theory of the meta-­normative.6 At least in my experience of presenting ideas from this book, it is a substantive, controversial question whether prudential discourse is normative in the way that moral discourse is normative (and thus a natural bedfellow of moral discourse within the class of the meta-­normative). Moreover, whilst it is true that a great many philo­ sophers think it is obvious whether prudential discourse is normative, unfortunately they do not agree as to what that obvious truth is.7 On the positive side we have Steve Finlay: There is close isomorphism between morality and self-­interest. For one thing, each is a normative domain: there is a moral ‘ought’ and an ‘ought’ of self-­interest.8 5  Indeed, some have suggested this to me. 6  For analogous inquiries into other prima facie normative properties and domains, see Schroeder (2015) and Sylvan (2018). 7  I think it is fair to attribute to them the implicit view that the answer is obvious by the limited amount of space they spend arguing for their respective views. 8  Finlay (2007a: 138).

Introduction  5 And David Enoch who writes: Paradigmatic examples may however help clarify the kind of things I  refer to as normative[:] that I should go on a diet, that you have a reason to read Kant [. . .] that pain is pro-­tanto bad for the person whose pain it is[.]9

By contrast, others contend that prudential discourse is non-­normative. Bart Streumer comments: [T]hat a painting is beautiful can be a reason to look at it and can make it wrong to destroy it, and being a reason and being wrong are normative properties. But that does not mean that beauty is itself a normative property. [S]imilar claims apply to disgust, funniness, harm, benefit[.]10

And Jonas Olson writes: The second consideration that makes me hesitant to accept the idea that welfare should be seen as a normative concept is purely eco­nom­ ic­al. We have a plethora of normative concepts already[.] Why introduce yet another one if it seems that we can do without it?11

Thus, one of the questions that I take up, in Chapter 2 and elsewhere, is whether prudential discourse is itself a normative form of discourse. I answer in the affirmative; prudential discourse is a normative form of discourse, on a par with moral discourse. Returning to the suggestion that this project is unnecessary, on the grounds that prudential discourse has already been theorized about via theorizing about the meta-normative, a second problem is that little to no attention has been paid to potential similarities and differences between prudential discourse and moral discourse. Even if, as I shall argue, moral and prudential discourse are both normative, there may 9  Enoch (2011a: 1). The reading example could be read differently—as a moral reason to read—but the other examples are plausibly prudential. 10  Streumer (2017: 105) (my italics). 11  Olson (2006: 176). Granted, Olson’s claim is specifically about the concept welfare.

6  Dear Prudence still be enough differences between them to render implausible a ­one-­size-­fits-­all meta-­normative theory. At the very least, even if we were already convinced that prudential discourse is normative, we should examine it in detail to assess the plausibility of the view that there is one true account of all normative forms of discourse. By examining prudential discourse in specific detail, and in parallel with moral discourse, we can thus make progress in assessing the plausibility of this thesis: One-­Theory-­To-­Rule-­Them-­All:  There is one true meta-­theory for all forms of normative discourse. A third way in which this sceptical stance is unwarranted stems from the fact, mentioned above, that some of these issues are separable from the question of whether prudential discourse is normative. Some of these questions, and the answers I defend, are simply about what prudential discourse is like. These questions simply have not been taken up at any great length. Let me outline how I proceed, chapter by chapter, and the views defended therein.

Part One: Prudence, Prudential Discourse, and Normativity I begin, in Chapter 1, with the question of whether there are prudential reasons. After clarifying the nature of prudential reasons (what it would take for there to be such reasons) I examine, and reject, various positions that entail the denial of the claim that prudential value generates reasons. These include forms of nihilism about prudential value along with Humeanism about practical reasons. Thus I argue for this thesis: Prudential Value Matters (PVM):  Evaluative prudential facts generate directive prudential facts (including facts about prudential, normative, reasons for action and for attitudes). In Chapter 2, I argue for the following thesis:

Introduction  7 Prudential Judgements are Normative (PJN):  Prudential judgements are normative judgements. I begin by distinguishing various ways by which one could argue that prudential judgements are normative. After outlining the weaknesses of alternative strategies I outline my own ‘parity strategy’. Therein, I assume that moral judgements are normative and lay out a set of features that prudential judgements hold in common with moral judgements which thus provide reasons to think that prudential judgements are normative. I then argue that there is no disanalogy between moral and prudential judgements that suffices to make the latter non-­normative.

Part Two: The Nature of Prudential Discourse In this section, I examine prudential discourse both for its own sake and in order to pursue my case for its being a normative form of discourse. In Chapter  3, I examine ‘good for’ talk and ‘needs’ talk and argue for forms of contextualism about each of these. In the case of ‘good for’ talk I show how recent work by Steve Finlay (2014) on the context-­sensitivity of ‘good’ supplies a plausible theory of ‘good for’ talk. The value of the proposal lies in its being able to simultaneously accommodate both the uniformity and the diversity of ‘good for’ talk. I then move on to ‘needs’ talk, showing how work on deontic modals from, and inspired by, Angelika Kratzer (1977) can be applied in a way that illuminates how ‘needs’ talk functions in prudential contexts. In Chapter 4, I then examine whether, and how, well-­being talk in particular is context-­ sensitive. I examine arguments from Anna Alexandrova (2017) and Stephen Campbell (2016) that well-­being talk is fundamentally context-­sensitive. I argue that some parts of well-­being discourse—such as ‘doing well’ claims—are context-­sensitive but that well-­being talk is not fundamentally context-­sensitive or conceptually pluralistic. In Chapter  5, I take up the question of what connection there is, if any, between prudential judgements and motivation. I argue that a small

8  Dear Prudence set of prudential judgements are necessarily motivating in rational agents (such that an agent’s failing to be motivated in the presence of the judgement is, ipso facto, a rational failing).

Part Three: Prudential Discourse Is Normative. What Follows? In this part, I examine some of the implications of the thesis that prudential discourse is normative. These are of two sorts. The first sort, outlined in Chapter  6, are implications for ethics and metaethics. I argue that the normativity of prudential value can play an important role in the discussion of certain forms of moral and normative scepticism in a way that puts pressure upon these sceptical views as well as bearing on the debate about the nature of thick concepts. The second set of implications, explored in Chapter  7, are implications for theorizing about prudential value. Herein, I explore the fruitfulness of applying ideas from metaethics to questions about prudential value and prudential discourse and argue for an reinterpretation of some classic debates as meta-prudential debates about the nature of prudential value. I close, in Chapter 8, by trying to map out some of the issues that are raised but not pursued in the book, to provide a guide for future inquiry. The first question is: which meta-­prudential view should we adopt? And is it plausible that the true metaethical view also applies to prudential discourse? I lay out the options and explain ways in which the foregoing discussion is probative for these debates. Before moving on, let me make explicit one of the general background assumptions of this work and a related issue that I take no stand on. My discussion assumes what David Copp (ms) calls ‘normative pluralism’, the view that there are distinct domains of normativity including morality and prudence which issue distinct evaluations of states of affairs and actions (etc.) and distinct reasons, permissions, and obligations. I therefore reject, for instance, the attempt to treat all practical normativity as part of a (much more capacious) category of prudence.

Introduction  9 My discussion does not, officially, take a stand on how the requirements (etc.) of these respective domains are related or whether they are commensurable in a way that yields truths about what one ought to do all things considered or what one just plain ought to do. Whilst I tend to be sympathetic to such a view, and deploy the idea in making certain points, my main commitments are, I believe (and hope), fully com­pat­ ible with the view that practical reason is fundamentally fractured such that there can be facts about what you prudentially ought to do as well as facts about what you morally ought to do but no fact about what you ought to do all things considered.12 Let’s get started. Are there any prudential reasons?

12  For discussion of these issues see especially Baker (2017), Chang (2004), Copp (1997, ms), Dorsey (2013), Killoren (2019), Sidgwick (1981), Tiffany (2007).

PART ONE

PRUDE NC E , PRU DE NT IA L DISC OU R SE , A ND NOR MAT IVIT Y

1 Prudence as a System of Categorical Imperatives 1.  Introduction In this chapter1 I will defend this thesis: Prudential Value Matters (PVM):  Evaluative prudential facts generate directive prudential facts (including facts about prudential normative reasons for action and for attitudes).2 Let me say a little more up front about this claim, to make things precise. First, evaluative prudential facts are facts about what is good for, better for, best for someone (and the corresponding negatives) with respect to well-­being. Two examples: her spouse’s death was terribly bad for Jane; June will be better off is she has surgery. Directive prudential facts are facts about the prudential reasons someone has, or what they prudentially should, ought, must do, or need to do/have. Two examples: Sarah must take the job; Catherine should end her relationship. Second, to keep things simple, I will mostly focus on the specific claim that evaluative prudential facts generate prudential normative reasons for action (and for attitudes) but anyone sceptical of reasons in particular can run the arguments I give for other claims, such as ought claims. Third, by a ‘prudential reason’ I mean, to borrow an expression from Alex Worsnip, a reason that is ‘distinctively and fundamentally about the promotion of the agent’s own well-­being’.3 Such normative reasons 1  This title of this chapter is, of course, a reference to Foot (1972). 2  I grant that ‘matters’ might sound more axiological than directive. I use it in the absence of a better name. (‘Prudential value is normatively significant’ is too clunky, for example.) 3  Worsnip (2018: 236). Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0002

14  Dear Prudence are thus doubly distinctive in being both agent-­relative and provided by facts about the agent’s own well-­being.4 Here is an example: finishing this book would enhance my well-­being so that gives me a prudential reason to finish this book. Fourth, as referred to in PVM, prudential reasons are normative ­reasons. They are justifying reasons to do or to feel something as opposed to motivating reasons (or, reasons why someone did something) and they are genuinely authoritative. Thus, PVM is not merely noting that prudence recommends that you do what is good for you. It is trivial that a standard or outlook generates reasons relative-­to-­that-­standard.5 For example, the Goth standard gives you a Goth-­reason to wear black rather than yellow. But that does not settle the question of whether there is, really, a justifying reason to wear black rather than yellow. That is the question one asks oneself in wondering whether the Goth standard is genuinely authoritative.6 PVM, then, is the claim that the reason that arises when something is good for you is a genuinely normative reason. One final clarification. PVM contradicts what is known as the ­buck-­passers ‘negative thesis’, the claim that evaluative facts themselves do not generate reasons and only their grounds do so.7 PVM does not, however, deny that the grounds of evaluative prudential facts give prudential ­reasons. It simply holds that—whether or not the grounds do—the evaluative prudential facts themselves provide agents with prudential reasons.8,9 With these clarifications made, let me give some motivation for the thesis via three apparent instances where prudential reasons follow from evaluative prudential facts: 4  Hereafter I will often drop the ‘normative’ from ‘normative reason’ for readability. 5  On this, see, for example, Enoch (2011b). 6  On these issues see, for example, Baker (2017), McPherson (2018), and Wodak (2019). 7  For discussion see e.g., Scanlon (1998), Dancy (2000), Suikkanen (2005), Stratton-­Lake & Hooker (2006), Väyrynen (2006), Bedke (2011), Gregory (2014), and Rowland (2019). 8  Anyone who would prefer a slightly weaker form of the thesis, one fully agnostic with respect to buck-­passing, is welcome to substitute ‘necessitate’ for ‘generate’ to make: ‘Evaluative prudential facts necessitate directive prudential facts (including facts about prudential, normative, reasons for action and for attitudes).’ On this buck-­passing-­friendly view prudence would still be constitutively normative (just like morality could be on a buck-­passing view of deontic facts). For discussion, see Schroeder (2007: ch 3). 9  One of the issues here is that some might think that we must choose between allowing either the evaluative prudential facts or their grounds to bear the normative ‘weight’, so to speak. For critical discussion of this idea, see Johnson King (2019) and Schroeder (2009).

Prudence as A System of Categorical Imperatives  15 Ursula’s Upgrade:  Ursula is checking in for her flight. The person at the desk tells Ursula that she might get a free upgrade and that this will be decided between now and when she reaches the gate. Ursula will be better off if her ticket is upgraded. Thus, Ursula has prudential reason to hope that she is upgraded. Christine’s Career:  Christine is deciding between different careers. She could be a lawyer, a professor, or a poet. The jobs have very different upsides and downsides. Christine will be best off if she chooses to be a poet. She will be worst off if she chooses to be a lawyer. Thus, Christine has prudential reason to choose to be a poet. Helen’s Holiday:  Helen is deciding between three different vacations. One of them, a beach holiday, will be very relaxing. The second, a skiing holiday, will be very exciting. The third, a writer’s retreat, will be very intellectually stimulating. Helen will be equally well off if she chooses the beach holiday or the writer’s retreat, each of which is better for her than the skiing holiday. Thus, Helen has a prudential reason to choose either the beach holiday or the writer’s retreat over the skiing holiday. In each example I have assumed that there is at least a partial ranking of the options with respect to prudential value. In the case of Ursula, one option is superior to the other. In the case of Christine, there is a ranking of all three options with one unique winner. In the case of Helen, there is one option that is worse than the other two equal-­ranked better options. In each case, I claim, the prudential value of the different options gives prudential reasons for the relevant agent (for Ursula, a reason for attitude, and for the other two, a reason for action). These seem, I hope, like the kinds of judgements that we make on a day-­to-­day basis. The thesis defended in this chapter, PVM, is simply that this connection between something being (e.g.) best for an agent and their having prudential reasons to (e.g.) hope that it occur or to bring it about, is genuine. We now know what must be true for there to be prudential reasons. Are there any? It seems like there are. Across many decision-­making contexts, one assesses options at least partly with an eye to which is in one’s best interests. When deciding which career to take up, one

16  Dear Prudence con­siders among other things how the options will impact one’s well-­being. It is common for people to recommend an option to ­someone on the ground that ‘it will be good for you’ or to recommend something by gently prodding ‘wouldn’t you be better off if you . . . ’. Or think of the occasions where someone simply enjoins ‘do what’s best for you’.10 One might choose to be a lecturer rather than an aid worker because, although better impartially, the latter option will be worse for oneself. How to weigh these competing considerations is, of course, difficult. But it seems plausible that, in such decisions, one factor we weigh against and alongside the moral reasons we have is a set of prudential reasons, reasons to promote and secure our own good. The same seems true for smaller-­scale decisions (joining a gym, going on holiday, and the like). Suppose you are asked why you did something. Just as ‘it was the right thing to do’ seems like a justification for performing some action in most, if not all, contexts, so too ‘it was the best option for me’ serves as a justification in at least some circumstances.11 So evalu­ative prudential properties seem to generate prudential reasons for action. Why, then, would someone deny such prudential reasons? The rest of this chapter is targeted at those who, on various grounds, have denied, or would deny, PVM. There are two general routes to ­denying PVM that I will examine here. The first is by doubting that ­anything ever has prudential value and so, as a corollary, rejecting the claim that evaluative prudential properties generate reasons (I will ­consider this view in section 2). The second general ground for doubting PVM is to deny that there are prudential reasons, even if one allows that there is prudential value. There are two major views that are committed to such a denial. The first 10  I do not deny that these locutions can be used to make claims about what would be, generally, best, rather than best for the agent. For more on this, see Chapter 3 Prudential Language and Context (I): Good for and Needs. 11  Scanlon (1998: 126–7) is often cited as having argued that something’s being best for you (etc.) is not a reason to choose it. The passage is actually unclear because Scanlon—repeatedly—phrases the claim more weakly. For instance: ‘the idea of well-­being plays little role in explaining why we have reason to value these things’ (my italics). More substantively, as Roger Crisp (2013) points out ‘this latter claim [enjoyment adds to my well-­being] sounds peculiar only because we already know that enjoyment makes a person’s life better for them. And in some circumstances such a claim would anyway not be odd: consider an argument with someone who claims that aesthetic experience is worthless, or with an ascetic.’ With Crisp, I think that Scanlon misidentifies a purely pragmatic phenomenon, that of the redundancy of citing the evaluative prudential fact in many cases when asked to justify some action.

Prudence as A System of Categorical Imperatives  17 is the view that all reasons are agent-­neutral. That is, one might doubt that there are any reasons with respect to an agent’s own well-­being that are not shared by all agents.12 The second is the view that all reasons for action are fundamentally generated by something other than prudential value, specifically by desires.13 On this story, evaluative prudential properties do not themselves generate reasons, at least not in a fundamental way. Having briefly introduced these grounds for rejecting PVM, let me now try to reject them.

2.  Grounds for Rejecting PVM (I): Prudential Error Theory Someone might deny Prudential Value Matters because they are committed to the view that there is no prudential value. I take such a view— ‘prudential error theory’—to be pretty counterintuitive in the abstract. But notice that there are two ways that one could end up committed to it. One could be a global error theorist about the evaluative in general, thinking that there is no prudential value in the same way that there is no moral value, no aesthetic value and the like. Alternatively, one could be a local error theorist, by being an error theorist about prudential value specifically, without being an error theorist about other kinds of value. The difference between these views matters because of the difference in how counterintuitive they are and the extent to which we can use ­evidence from ordinary normative thinking to adjudicate between the error theorist and someone who holds PVM. In the remainder of this chapter, I will confine my attention to local prudential error theory (hereafter omitting the ‘local’), postponing discussion of the more global form of error theory until a later chapter.14 It is plausible that if there are any evaluative facts at all, two of them are (i) that at least some kinds of pleasure are good for their subjects and

12  A position explored at length in Nagel (1970). 13 A view traditionally associated with Hume (1739) and given its most sophisticated defence in Schroeder (2007). 14  See Chapter 6, Prudential Normativity, Moral Scepticisms, and Metaethics.

18  Dear Prudence (ii) that at least some kinds of pain are bad for their subjects. Given the plausibility of these claims, and the wide commitment to them in or­din­ ary evaluative discourse, I will assume that the burden of proof is on the prudential error theorist to give us a good reason to reject the idea that at least some things are good and bad for those who experience them. Let us then examine the arguments that have been offered, or might be offered, for prudential error theory.15 First, take Moore’s famous assessment of ‘good for’ talk: What, then, is meant by ‘my own good’? In what sense can a thing be good for me? [. . .] In short, when I talk of a thing as ‘my own good’ all that I can mean is that something which will be exclusively mine, as my own pleasure is mine (whatever be the various senses of this ­relation denoted by ‘possession’), is also good absolutely; or rather that my possession of it is good absolutely. The good of it can in no possible sense be ‘private’ or belong to me; any more than a thing can exist ­privately or for one person only. The only reason I can have for aiming at ‘my own good,’ is that it is good absolutely that what I so call should belong to me—good absolutely that I should have something, which, if I have it, others cannot have. But if it is good absolutely that I should have it, then everyone else has as much reason for aiming at my having it, as I have myself.16

This passage provides no good argument for prudential error theory.17 Moore argues that one cannot make prudential claims using ‘good for’ by, first, presupposing that all good for claims concern impersonal goodness and then explaining why impersonal goodness cannot belong to one person. But the relevant view here is that there is a kind of value, prudential value, that is distinct from impersonal goodness.18 This issue 15  It is difficult to find a completely unambiguous expression of (local) prudential error theory. I discuss Moore and Hurka by way of supposed, and possible, inspirations for the view rather than as clear adherents to it. 16  Moore (1903: 150). 17  There are other ways of reading the passage, such as its offering a reforming definition or a reductive analysis of ‘good for’. The error theoretic, incoherentist, reading is offered by Crisp (2013). 18  For discussion of these issues see Rosati (2008: 330–1).

Prudence as A System of Categorical Imperatives  19 is obscured because of Moore’s focus on the locution ‘good for’ and much of the passage only sounds remotely plausible by focusing on that particular expression. If we substitute ‘benefits’ into the passage it is much less rhetorically effective and the problem with the argument is clear. After all, we have no difficulty imagining that a benefit can be restricted to one person. Suppose I sell all my belongings and give you the proceeds. Who benefits? You do. This is perfectly clear, pace Moore.19 The second part of the passage makes clear why the wider dialectical context leads Moore to deny that things can benefit particular people or be good for them.20 Moore’s ultimate target is the form of egoism that holds that, fundamentally, everyone has reason only to aim at their own good. Moore’s aim in this passage is to cut such a view off at the knees by denying that there is even such thing to aim at. It would be much better though to be less ambitious and argue that whilst there is such a thing as my own good, things that benefit me, I have reasons in addition to the prudential reasons I have to promote these things. Most importantly for my purposes, it should be clear that accepting PVM carries no commitment to egoism nor, indeed, any view of the existence and relative weight of non-­prudential reasons. (PVM is compatible with orthodox impartial utilitarianism about reasons for action so long as the prudential reasons generated are only reasons for attitudes.) Some others who are prudential error theorists, or might fairly be read as expressing a prudential error theory, are contemporary Mooreans such as Tom Hurka. He writes: I think ‘good for’ is fundamentally confused, and should be banished from moral philosophy. The problem is not that the expression can be given no clear sense. It can, indeed, be given too many senses, none of which is best expressed by ‘good for’. As currently used, ‘good for’ is multiply ambiguous, and for none of its meanings is ‘good for’ the appropriate term.21 19  By the time he wrote Ethics, Moore (1912: 120) was happy to write: ‘I think, therefore, we must conclude that a maximum of true good, for ourselves, is by no means always secured by those actions which are necessary to secure a maximum of true good for the world as a whole.’ 20  Which looks like a Moorean fact to me. 21  Hurka (1987: 72).

20  Dear Prudence Hurka is absolutely correct that ‘good for’ is used in multiple ways.22 But this is inadequate as support for prudential error theory. One reason why it is insufficient is that prudential error theory requires rejecting a host of claims that are not expressed using ‘good for’. For example, prudential error theory is committed to the view that nothing ever harms anyone and that nothing ever benefits anyone, that nothing is ever better for someone than something else, that no one has a level of well-­being, that no one’s life deteriorates, or improves, and so on. Trying to figure out the things that do this—which things harm us, which things benefit us, and the like—is, of course, difficult. But that does not show that there is no sense to be made of something benefitting someone, or how their life is going. On the basis of the claims in Hurka’s passage, we do not have reason to accept prudential error theory. Remember that I have postponed the global kind of error theory—the one that rejects prudential value on the basis of rejecting all kinds of value. Here my attention was confined only to those who are happy to countenance other kinds of value but who think that there is some reason to reject prudential value in particular and would reject PVM on that basis. My response was, first, that this view is highly counterintuitive and thus we should only accept it on the basis of a good argument and, second, that we have been given no good argument for such a view. Having rejected this kind of ground for rejecting PVM, one based on denying that things have prudential value, I now move on to considering the second main way in which one might reject PVM: rejecting prudential reasons in particular.

3.  Grounds for Rejecting Prudential Value Matters (II): No Agent-­Relative Reasons One way of denying that evaluative prudential facts generate prudential normative reasons is to hold the view all reasons are agent-­neutral. This would be the view, for example, of someone who accepted a very broad form of utilitarianism, holding that well-­being is the sole (fundamental) 22  See Chapter 3, Prudential Language and Context (I): Good for and Needs.

Prudence as A System of Categorical Imperatives  21 generator of all reasons for action who denied that a person’s well-­being generates distinctive reasons for themselves, over and above the moral reasons that everyone’s well-­being generates for everyone.23 What is the difference between agent-­ relative and agent-­ neutral reasons?24 Whilst there is no consensus on how precisely to draw the distinction,25 we get the general idea via cases such as:   Option 1 Option 2

Vikram’s Well-­Being

Julia’s Well-­Being

500 600

1000 901

Suppose you must decide which option to bring about. Vikram and Julia are strangers (both to you and to each other). The reason that you have for choosing option 2 is plausibly fully general or agent-­neutral: the greater total well-­being. Now, suppose that Julia is making the decision (Vikram and Julia remain strangers to each other). If there are agent-­relative reasons then the situation is different from the first iteration in the following way.26 Whilst Julia has a reason to choose option 2 that is exactly like one you had, there is also a reason for Julia to choose option 1, namely that her own well-­being is higher under that scenario. Those who claim that all reasons are agent-­neutral must deny the existence of this reason for Julia. For them, there is no difference in the reasons that you or Julia would have in making these decisions. I think that such a view is implausible for reasons for action and that we can see that by considering the following kind of case:   Option 1 Option 2

Vikram’s Well-­Being

Julia’s Well-­Being

500 1

1 500

23  This would be a broad view because it would be utilitarianism about not only moral ­reasons but all practical reasons. 24  The terminology of agent-­neutral vs agent-­relative was, I believe, introduced by Parfit (1984). The modern debate about such reasons stems from Nagel (1970). 25  For detailed discussion of the extant proposals, see Ridge (2017). 26  I am here assuming that if there are agent-­relative reasons then there are agent-­relative reasons in this sort of case. Of course, one sympathetic to agent-­relative reasons might deny that there are any in this particular case.

22  Dear Prudence Suppose I am making the decision between these options and that there are no further relevant differences between these two outcomes and these two people are perfect strangers to me. In such a situation, I would need some way of resolving the tie between these outcomes, such as by flipping a coin. If all reasons are agent-­neutral then it would be true that if Julia were in the position of choosing, that she too would need to flip a coin in order to decide between these options. But that seems im­plaus­ ible. Presumably her own well-­being can at least act as a tiebreaker in this case and give her some reason to choose option 2. This suggests that there are agent-­relative prudential reasons even if they are only strong enough to break ties. Note, though, that the thesis I am defending is compatible with denying that there are agent-­relative reasons for action as long as there can be agent-­relative prudential reasons for attitudes. And this seems in­cred­ ibly plausible. Going back to the case above (where the two options have the same total well-­being) and where I am the person who must choose between the options, suppose Vikram and Julia know the situation. Don’t they each have at least a reason to prefer the outcome in which they do much better? Would it not be strange if they were each completely indifferent to which option I choose, given the huge difference in their well-­being on each scenario? I take this to be compelling evidence that there are agent-­relative, prudential, reasons for attitudes at the least. And that is sufficient to ward off this objection to Prudential Value Matters. I now move on to what I take to be the main, and likely most popular, ground for rejecting PVM: Humeanism about practical reasons.

4.  Grounds for Rejecting PVM: (III) Humeanism about Reasons On a Humean theory of reasons, all reasons for action fundamentally stem, or are explained by, an agent’s desires. Whenever someone has a reason to do something this is because the action is conducive to the fulfilment of a desire of theirs. Put more precisely:

Prudence as A System of Categorical Imperatives  23 Humeanism about Reasons:  X has a reason to Ф iff and because Ф-­ing promotes the fulfilment of a desire of X’s.27,28 One might wonder precisely why Humeanism leads one to deny PVM. The tension arises because PVM claims that evaluative prudential properties provide practical reasons for the relevant agent whereas, according to Humeanism, reasons are always provided by facts about the promotion of the fulfilment of desire. Humeanism and PVM will give identical verdicts about the presence of a reason in many cases—i.e. both will deliver the verdict that the rele­ vant agent has a reason to Ф. This will be true on any of the (presumably many) occasions where an agent has a desire that is fulfilled by something that is good for them. For example, suppose that: friendship is good for Agnes; Agnes desires friendship; and Agnes is in a position to form a friendship. According to Humeanism and PVM Agnes therefore has a reason to form a friendship. However, the views offer competing explanations of why Agnes has this reason. On the Humean view, the fundamental explainer of the reason is the fact that Agnes desires friendship. On the alternative view, the fundamental explainer of the reason is, simply, the fact that the option is good for the agent (and this reason remains in place even if the desire were extinguished). Thus, according to the Humean, the explanation of Agnes’ reason to form a friendship reaches its end once we note that Agnes can promote the fulfilment of her desire for friendship by forming a friendship. By contrast, according to Prudential Value Matters at least one reason that Agnes has to form a friendship is the fact that forming a friendship is good for her.

27  For canonical recent defences of the view see Schroeder (2007) and Markovits (2014). For the sake of space I focus on this, standard, Humean view, but many Humeans (and commentators) sophisticate the theory in different ways. Such ways include: allowing the reasons that an agent has to be explained by desires that they do not have but would have under some counterfactual condition (e.g., Williams 1981); allowing desires that the agent does not yet have but will have to be relevant (e.g., Lin 2015); allowing the desires of other agents to be relevant (e.g., Manne 2016). For dissent see, for example, Enoch (2005). 28  For discussion of the promotion relation see DiPaolo & Behrends (2015), Lin (2018), Sharadin (2015, 2016).

24  Dear Prudence This is a tricky point, so let me put it another way. The challenge to prudential reasons posed by the Humean theory of reasons is not that if Humeanism is true then there are no reasons to promote your own ­well-­being. Rather, the view is that there are no distinctively prudential ­reasons, namely reasons for action that stem ultimately from facts about well-­being.29 A Humean view can well allow that there are desire-­based reasons to promote one’s own well-­being (and the well-­being of others) whilst denying that prudential value gives rise to a distinct set of prudential reasons. A useful parallel is another corollary of the Humean view—namely that there are no distinctively aesthetic reasons. A Humean will hold that even though some things are aesthetically valuable and disvaluable, these properties do not themselves generate reasons. Rather, any reason to promote the aesthetically valuable is fundamentally grounded in desire. On such a view neither beauty, nor any other aesthetic property, generates reasons for action in its own right.30 Let me leave aside the point about the different explanations that Humeanism and PVM will offer in cases where both hold that the agent has a reason to do something that promotes their well-­being. More significantly, Humeanism and PVM will diverge in the verdicts they give in any case, real or conceivable, where Ф-­ing is good for the agent but the agent does not have a desire whose fulfilment will be promoted by Ф-­ing. Having explained the challenge to PVM from the Humean view of practical reasons, let us now assess the prospects for prudential reasons. Of course, the truth or otherwise of Humeanism about practical reasons is a major issue, so I do not therefore pretend to offer a comprehensive case against it. I aim only to show, firstly, that the view is counterintuitive with respect to reasons to promote well-­being in a number of ways. I also argue, in a more conciliatory way, that even the truth of Humeanism about reasons for action is compatible with the plausible claim that prudential properties generate reasons for attitudes. So one could still be a Humean about reasons for actions and accept a weakened form of PVM. 29  For helpful, detailed, discussion of these points see Worsnip (2018). 30 The issues here are structurally similar to those concerning Humeanism and moral ­reasons. Humeans hold that morality and its demands cannot fundamentally generate reasons, only desires can. On these issues and their connection see also Streumer (2017: §52).

Prudence as A System of Categorical Imperatives  25

5.  Humeanism vs PVM Here are three cases to provisionally demonstrate the ways in which Humeanism is counterintuitive with respect to reasons to promote well-­being: Cult Colin:  Colin is raised in a cult and lead to believe that his torment and suffering is necessary for the sake of humanity. He has many desires including the desire to suffer for the sake of humanity. None of his desires would be fulfilled by escaping the cult. However, were he to do so, he would lead a life with a much higher level of well-­being. Depressed Dave:  Dave is severely depressed. His condition means that he has no desire that would be fulfilled by his continued existence. However, if he were to get medical help, he would go on to lead a life with a much higher level of well-­being. Educating Emily:  Emily is a child who desires only to watch cartoons and eat ice cream. None of her desires would be fulfilled by going to school. However, were she to do so, she would lead a life with a much higher level of well-­being. In each case it is plausible that the relevant agent has a reason to perform some action (leave the cult, get help, go to school). And in each case that reason looks like it comes from facts about their well-­being and cannot be explained by the action promoting the fulfilment of their desires.31,32 These scenarios do not depend on the assumption of any particular first-­order theory of well-­being. Whilst it is easiest to see why it would be that (e.g.) Colin would have a much higher level of well-­being on the assumption of an objective list theory, or perfectionism, even the truth 31  For discussion of these, see Lin (2015). 32  I eschew discussion of two of the most famous cases from the literature introduced by Derek Parfit—the agony argument and Future Tuesday Indifference—on the grounds of their introducing distracting elements. Parfit’s agony argument (2011: 73–74) is complicated by the question of whether pleasure and pain themselves are desire-­based in nature. For discussion of the issues see, for example Sobel (2019) and Ventham (2019). Parfit’s (2011: 56) other famous example—Future Tuesday Indifference—also introduces complications about the nature of pleasure and pain but also many other issues about how to understand the psychology of such an indifferent agent. For forensic discussion and reconstruction of this example, and analogous ones, see Street (2009).

26  Dear Prudence of a desire theory of well-­being is compatible with the example.33 It would just need to be true that (e.g.) Colin would end up with a different, larger set of (fulfilled) desires, if he were to leave the cult than if he were to remain. Examples like Cult Colin, Depressed Dave, and Educating Emily exemplify one way in which Humeanism is revisionary about reasons and well-­being. For in these cases Humeanism entails (i) that the prudential value that would be gained by performing some action is not, itself, a reason to do it and (ii) that the absence of a desire that would be fulfilled by performing the action entails that there is in fact no reason to perform the action. But this is extremely counterintuitive, especially in the case of Dave. Our sense that Dave has good reason to get help is not undercut, I assume, by the discovery that his depression has become so bad that he has no desires served by not getting help. Rather, this is evidence of just how strong his reason to get help is. Cases like Colin, Dave, and Emily put pressure on Humeanism by showing how revisionary it is when it comes to well-­being (in a way that is, arguably, starker than its commitments with respect to morality).34 We can accentuate this by considering a different case:35 Balanced Belinda:  Belinda has a choice between two different career paths. Her desires are such that each option will promote the fulfilment of the same number of them (suppose for the sake of simplicity that her desires are all of the same intensity) so her desires leave the options weighted exactly equally. However, one option will give her a significantly higher level of well-­being than the other. Humeanism regards the two options as identical, in terms of reasons for action, given that each option promotes the fulfilment of the same 33  The simple theory I have in mind is [Desire Theory of Well-­Being]: something is good for someone if and only if and because it fulfils one of their actual desires. It is important to note this point because there is a tendency to suppose that if well-­being is itself desire-­based then reasons to promote well-­being must be desire-­based. 34  Starker, I think, because it is easy to think of internal reasons as being closely connected to one’s own well-­being (especially if one holds a desire-­based view of well-­being) and to fail to appreciate how often Humeanism will fail to deliver the reasons that we think agents have to promote their own well-­being. 35 Unlike the earlier cases, this case does require assuming the falsity of at least some, implausible, forms of the desire fulfilment theory of well-­being.

Prudence as A System of Categorical Imperatives  27 number of desires (of equal intensity). But that is to think that no amount of extra well-­being for one of the options can provide any normative weight, even as a tiebreaker. And that seems extremely counterintuitive. Couldn’t someone recommend that Belinda take one of the options on precisely this, prudential, ground? And couldn’t Belinda rationally choose one of the options on this ground? If so, then it looks like prudential properties can at least give reasons to break ties. Having laid out some cases designed to motivate PVM, let me turn now to Humean arguments against PVM.36 A standard Humean approach is to argue that if there were prudential reasons then we should expect to see them make certain differences in cases which, according to the Humean, they do not in fact make. One such argument is given by Alex Worsnip, thus: Anyone who thinks there are fundamentally prudential reasons must hold that your own well-­being carries some weight, in addition to whatever weight it has morally, that is not contingent on your conative states. But if this is so, then (at least normally) you are required, from the all-­things-­considered point of view, to assign your well-­being this extra weight in your deliberations. In my view, this is implausible. It is one thing to say that you are permitted to give it this weight if you want to. It is quite another to say that you make a mistake if you do not.37

And Max Hayward provides a memorable description of such a case thus: Nevertheless, after a week of monkish virtue at my desk, an evening of intoxicated indulgence issues a compelling call, consequences be damned. Pain and regret are simply the costs of the pleasure I seek, which preference is not, I think, ipso facto irrational. I have been

36  The most famous piece of rhetoric in favour of Humeanism over PVM is, of course, Hume’s infamous claims: ‘Tis not contrary to reason for me to chuse my total ruin, to prevent the least uneasiness of an Indian or person wholly unknown to me. Tis as little contrary to reason to prefer even my own acknowledg’d lesser good to my greater, and have a more ardent affection for the former than for the latter.’ (my italics) 37  Worsnip (2018: 244).

28  Dear Prudence known to make the exchange; I have no confidence that the pleasures always outweigh the pains in any sense; and I am not irrational.38

Taking the Hayward first, his charge is that pleasurably drinking to excess even when one knows that this will be outweighed, in the balance of pleasure and pain, is not irrational. To put it in the terms of the dialectic of this chapter, his position is that there is no—genuinely normative— reason to promote and protect your own good in a case like this.39 Worsnip’s argument, by contrast, is more abstract. He identifies an apparent consequence of prudential reasons and suggests that it is implausible. (What is common to both, though, is the idea that one’s own good does not, ipso facto, generate prudential, normative, reasons.) These are ­plausible claims, but the extent to which they tell against PVM, is overstated, I think, in a way that helps to clarify what is at stake in holding PVM. It is true that we do not expect someone to always pay close attention to what is good, or best, for themselves. And it is true that we do not condemn the person, even a friend, who is slightly imprudent. Going further, it is also true that we often condemn someone who is very concerned with what is best for themselves. These points, however, can be explained consistently with PVM and do not provide evidence against it. In fact, they are what we should expect if there are prudential reasons (among other normative reasons). Part of the PVM-­friendly explanation of these observations, which applies to what I will call Hayward’s tippler, is that it will often be difficult to tell whether some option was, or was not, in someone’s best interests, especially someone other than the person themselves (who cannot know, for instance, exactly how pleasurable a prospect the friend finds it to be). So, often, our failure to condemn what, in fact, is imprudent will be because it is not sufficiently clear whether it was imprudent.40 Another part of the explanation will be that, even if we stipulate that it is imprudent on this occasion (and known to be), it is obviously an unwise strategy to take a keen interest in tracking small imprudences. 38  Hayward (2019: 55–6). 39  This is a corollary of his general view that there are no prudential reasons or requirements. 40  Note that in the example this is just a regular Friday night, not the day before some important exam or similar.

Prudence as A System of Categorical Imperatives  29 A good friend does not meticulously track, and criticize, every instance of imprudence (though presumably they will keep track of large-­scale imprudence). And, even in the first-­person case, an individual will likely not promote their own well-­being as successfully if they focus too much on whether a particular decision—at least one of this magnitude—is prudentially best.41 This will be especially true in any case where we are deciding whether to promote our own good or someone else’s. It makes sense, both morally and prudentially, for us not to fastidiously track what is best for us.42 Thus, contra Worsnip, the existence of prudential reasons does not mean that ‘(at least normally) you are required, from the all-­things-­considered point of view, to assign your well-­being this extra weight in your deliberations.’ (my italics).43 In fact, they would precisely recommend your not doing so (or, at least, recommend against doing so all of the time). These points made in response to Worsnip and Hayward point to a constructive way forward. If the kinds of pragmatic facts cited above can explain why in normal circumstances it seems like prudential con­sid­er­ ations carry no normative weight,44 or why it can in fact be prudent not to weigh them in deliberation, that suggests that we should think about more extreme cases. One way to get there is to move from Hayward’s tippler to an example of someone who makes this choice day in, day out. Then it seems much less plausible that the person is not being irrational or failing to attend to a normative requirement to take care of themselves. Then take cases 41  These kinds of points are familiar, of course, from discussion of the paradox of hedonism. On this, see, for example, Railton (1984). 42  This point will be further supported if there are instances of prudential supererogation. It is easy, and common, to think of prudence as taking a straightforward maximizing form, such that one must prudentially always do what is best for oneself and everything else is criticizable. But it is unobvious why some prudential options are not good enough to render us free from criticism, even if some other option is better. For instance, there seem to be things that you prudentially should do but do not have to do. For discussion of prudential supererogation see McElwee (2017). 43  Let me note, in the interests of fairness, that Worsnip’s claim is explicitly hedged by an ‘at least normally’, so he does not commit himself to the conditional claim that if there are prudential reasons you must always assign them weight in your deliberations. But Worsnip does not outline the exceptions and the objection I make here is, I hope, responsive to the spirit of his objection. 44  It is (too) easy to infer the absence of reasons from what we have reason to say and do. See Schroeder (2007) and Snedegar (2013) for discussion.

30  Dear Prudence such as Diogenes syndrome or extreme self-­neglect. In such cases individuals fail to attend to their most basic needs and resist help, seemingly from a total lack of desire to help and take care of themselves. We treat this as a serious, and tragic, medical condition, and one that plausibly rests on a cognitive failing. If one accepts PVM one can say that the person who suffers from this condition ipso facto makes a mistake about whether they have reasons to take care of themselves and to want to do so. The Humean, by contrast, must say that such cases are problematic only if the person in the grip of such a condition thwarts a desire of theirs.45 But in the case where the person has no desires served by taking care of themselves, the Humean must say that these people are not overlooking genuine prudential reasons and, in fact, have no reasons to look after themselves.46 Given that a Humean account of reasons to promote our well-­being says that it is our desires that are fundamental to generating reasons for action, and not well-­being itself, it is part of the view that the general tendency that people have to be concerned for their own good is, simply, a matter of taste or just a feature of our psychology. Starting from that observation, let me give one final argument in favour of PVM over Humeanism. Take this case: Simple Stan:  Stan, a single man with no children, works in a w ­ ell-­paid job. He lives frugally, eating cheap food, forgoing any luxuries, and maintaining only his basic survival needs, in order to save as much money as possible. He is not saving for his future or for his retirement. He makes no such provision for that and knows he will be destitute in retirement. He saves only to send money, anonymously, to a person he met once in an airport departure lounge. He receives no feedback or information whatsoever on his donations. 45  Or, if they think that no such case is actual, that such a case would not be problematic because the agent would thwart no desires of theirs. 46  Worsnip (2018: 246) considers such a response: ‘[I]t might be conceded that you have some latitude over the normative weight of your own well-­being. But, it might be continued, isn’t there some minimum baseline below which you may not discount your own well-­being? I do not have to deny this. I simply have to insist that if there is such a minimum baseline, it is whatever weight your well-­being has morally.’ (My italics). But it is, I suggest, implausible that such self-­neglect would be completely unproblematic, if one were a hermit or alone on a desert island. Being intensely neglectful of oneself seems like a failing, even if one is not letting others down also.

Prudence as A System of Categorical Imperatives  31 What can a Humean say about Stan? They can note that he is unusual, of course. But, given that the Humean treats the reason to promote one’s own well-­being as being determined by one’s desires, they cannot say that he is making a mistake in doubting that he has reason to promote his own well-­being beyond the minimal level he needs in order to keep functioning and saving money. He is unusual but only in the way that people who do not like chocolate or puppies are unusual. But this does not seem strong enough as a criticism! Stan is not merely atypical or someone with different tastes from the rest of us. He fails to see the ­reasons that he has to make his life go better. Let me adopt a more conciliatory mode now. I have argued that Humeanism is counterintuitive in a range of cases precisely because it says that well-­being does not, itself, generate prudential reasons for action. But remember that the original issue we started off thinking about was the status of this claim: Prudential Value Matters (PVM):  Evaluative prudential facts generate directive prudential facts (including facts about prudential, normative reasons for action and for attitudes). Reasons for action are only one possible candidate here. Even if the foregoing arguments fail, and even if Humeanism were true of all reasons for action, there can still be other prudential reasons that evaluative prudential properties produce in the form of reasons for attitudes, such as preference or desire.47 We saw one example earlier which demonstrated the plausibility of prudential reasons for attitudes: Ursula’s Upgrade:  Ursula is checking in for her flight. The person at the desk tells Ursula that she might get a free upgrade and that this will be 47  Is it peculiar to examine the possibility of Humeanism about reasons for action but not for attitudes? I do not deny that some will be Humeans about both. But it is hard to know what the orthodox Humean position is on this issue. In A Treatise of Human Nature (1738: 2.3.3.6), Hume famously declared there to be no reasons for preference: ‘Tis as little contrary to reason to prefer even my own acknowledg’d lesser good to my greater, and have a more ardent affection for the former than for the latter’ (my italics). By contrast, contemporary Humeans typically leave unclear the extent to which their views extend beyond reasons for action. Two exceptions: Markovits (2014)—who is explicit in rejecting Humeanism for reasons for belief— and Cowie (2014a; 2016a) who explicitly extends Humeanism to reasons for belief.

32  Dear Prudence decided between now and when she reaches the gate. Ursula will be better off if her ticket is upgraded. In this case, Ursula does not have reason to do anything to get an upgrade, given that she cannot affect her chances. Presumably, though, she has reason to hope that she is upgraded, stemming from the fact that she would be better off. Further, this does not seem dependent upon Ursula having some relevant desire which the hope would promote the fulfilment of. Thus, even if Humeanism is true for reasons for action, PVM (or, more precisely, a slightly weaker version of PVM) could still be true because of prudential reasons for attitudes.48,49 Whilst I hope to have made a case for Prudential Value Matters on the basis of prudential reasons for action that are, ultimately, explained by evaluative prudential properties, it would suffice for my official goal of establishing Prudential Value Matters if there are prudentially grounded reasons for attitudes.

6.  Conclusion In this chapter, I have sought to establish Prudential Value Matters, the claim that evaluative well-­being properties generate prudential reasons. I took the prima facie plausibility of the claim as a starting point and focused my efforts on trying to undermine objections. These were, respectively: scepticism about prudential value, the view that all reasons are agent-­neutral, and Humeanism about practical reasons. If Prudential Value Matters is true, then evaluative well-­being properties give us distinctively prudential reasons. That is one meta-­prudential thesis. In Chapter 2 I move on to look at prudential thought and talk, asking whether such thought and talk is normative. 48  Slightly weaker because one would need to replace ‘reasons for action and for attitudes’ with ‘reasons for action or for attitudes’. 49  One could fruitfully argue from reasons for attitudes to reasons for action on the grounds that it would be weird if prudential facts could give us reasons for attitudes but could not give us reasons for action. But the Humeanism-­compatible-­compromise is sufficient for my purposes here. For discussion of reasons for action and attitudes in another normative domain— the aesthetic—see (e.g.) Whiting (forthcoming).

2 Is Prudential Discourse Normative? 1.  Introduction In Chapter 1, I argued, in answer to the first of the ‘big picture’ questions distinguished in the introduction, that there are prudential reasons. In this chapter, I shift my focus to the second of the big picture ­meta-­prudential questions—Is prudential discourse normative? As I mentioned in the introduction, many philosophers think that there is a clear answer to the question of whether prudential discourse is normative.1 Unfortunately, though, they disagree on what that answer is and their discussion tends to be in passing, rather than sustained. For example, on the positive side we have Daniel Wodak, who writes: [T]he expressivist offers an explanation of the meaning of claims about what we ‘ought to do,’ or what we ‘morally ought to do.’ But this ignores the varieties of normativity. Antigone’s and Creon’s claims about what they ‘ought to do’ are relativized to moral, legal, prudential, and rational standards.2

Similarly, Alex Gregory writes: To that extent, other normative domains that are less social in nature seem to lack any necessary connection with our reactive attitudes. Think, for example, about prudential judgments.3

And Ishtiyaque Haji begins his discussion thus: 1  I think it is fair to attribute to them the (implicit) view that the answer is clear by the limit­ed amount of space they spend arguing for their respective views. 2  Wodak (2017: 217). 3  Gregory (2017a: 40). Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0003

34  Dear Prudence Various normative judgments are conceptually tied to freedom. For instance, we are at home with the view that moral responsibility requires control [. . .] Still further, and not surprisingly, we find this same sort of conceptual association between freedom and judgments of prudential obligation.4

We also find Joshua Gert who writes that the normative: includes—amongst many other things—the beautiful, the disgusting, the funny, the rational, the moral, the good and the bad in various forms, harm and benefit, health and disease, function and purpose, and truth and falsity.5

By contrast, we find Bart Streumer, who, replying to Gert, argues: [T]hat a painting is beautiful can be a reason to look at it and can make it wrong to destroy it, and being a reason and being wrong are normative properties. But that does not mean that beauty is itself a normative property. I think that similar claims apply to disgust, funniness, harm, benefit[.]6

We also find Jonas Olson: The second consideration that makes me hesitant to accept the idea that welfare should be seen as a normative concept is purely eco­nom­ ic­al. We have a plethora of normative concepts already[.] Why introduce yet another one if it seems that we can do without it?7

I take this disagreement, and the scant attention paid to the question of whether prudential discourse is a form of normative discourse, to justify investigation of the issue. Having introduced the question, let me now outline how I will try to answer it (and how I will not). 4  Haji (2012: 3). 5  Gert (2012: 1). 6  Streumer (2017: 105). 7 Olson (2006: 176). On Olson’s reference to the ‘second’ consideration: as I read him, Olson’s first objection is not to the normativity of prudential judgements per se but, spe­cif­ic­ al­ly, to the combination of that view and Darwall’s account of the nature of such judgements.

Is Prudential Discourse Normative?  35

2.  The Parity Strategy I will assume that some region of discourse is normative if the judgements essential to it are normative judgements. Thus, I will argue that prudential discourse is normative by arguing for the following more specific thesis: Prudential Judgements are Normative (PJN):  Prudential judgements are normative judgements. A natural retort at this point: what makes a judgement a normative judgement? Before outlining my strategy for establishing PJN, let me introduce and set aside three alternate proposals whose weaknesses are instructive. First, I do not answer the question of what makes a judgement a normative judgement via some meta-­normative theory. For instance, if one characterizes normative judgements as essentially involving ascription of non-­natural, irreducibly normative properties then one’s char­ac­ter­ iza­tion has immediately assumed a form of non-­naturalism or robust realism (or error theory).8 By contrast, if one characterizes normative judgements as essentially desire-­like states then one has immediately assumed a form of non-­cognitivism (and the same goes for other parochial proposals). It is not my ambition here to argue for one specific meta-­prudential theory, only the more minimal thesis that prudential discourse is a normative form of discourse. I thus need a less parochial way of characterizing normative judgements.9 Second, I do not want to treat the normativity of prudential judgements as settled by the fact that such judgements are made using 8  For discussion of various options for what makes a property normative, see Streumer (2017: §45). 9  Let me say a little about why the equivalent task in metaethics is typically not undertaken, to make what I am doing maximally clear. In metaethics, the subject matter is moral judgements and it is normally part of the background to that inquiry that moral judgements are normative judgements and, relatedly, have a number of interesting features. The main task is then to work out the best theory of this set of normative judgements. Here, I am trying to establish the equivalent claim of normativity but for prudential judgements. The next step (not undertaken here) would be to argue for a specific meta-­prudential theory. Interestingly, things are different in the case of meta-aesthetics. As made clear by Hanson (2018), philosophers tend towards a specific set of meta-­aesthetic views—anti-­realist ones—but without denying that the aesthetic is a normative domain that belongs alongside the moral domain.

36  Dear Prudence evaluative and directive terms. The trouble with this test, considered alone, is that there are many kinds of judgements that deploy evaluative and directive vocabulary that seem not to belong in the same theoretical category as, e.g., moral judgements, the paradigmatic instance of normative judgement.10 Familiar examples here are judgements about attributive goodness—good kettles and good toasters etc.—the rules of clubs and games, legal judgments, and etiquette judgements. Though these judgements are normative in some sense of ‘normative’ they are not normative in the way moral judgements are. Third, one might suggest that prudential judgements are normative if they ascribe properties that entail normative reasons. Unfortunately, this proposal collapses the distinction between normative properties and normatively-­relevant properties (and between normative judgements and normatively-­ relevant judgements).11 Many judgements ascribe properties that entail normative reasons but not all of these are plausibly normative judgements themselves. For example: ‘a phone is ringing in the cinema’, ‘the fridge is empty’, ‘that’s a snake’ are judgements that ascribe properties that entail normative reasons without themselves being normative judgements. I thus cannot characterize normative judgements in terms of reasons-­entailment.12 The alternate strategy that I will pursue I call the ‘parity strategy’. This is to argue for PJN by analogy with moral discourse. It is widely held that morality is a normative domain. Thus, if we can find relevant similarities between moral judgements and prudential judgements, and no significant disanalogy, that gives us grounds for holding that prudential judgements are normative. I will pursue such a strategy, arguing, first that the features of moral judgements that metaethicists appeal to, or would appeal to, when articulating, explaining, and justifying the claim that moral judgements are normative are also possessed by prudential judgements. I then consider various objections to the parallel.

10  I use ‘normative’ in the broad sense that encompasses the evaluative rather than contrasting with it. 11  This distinction (in terms of facts) comes from Parfit (1997). 12  What about strengthening the claim to necessary reason-­entailment? I think not. Pleasure and pain facts are plausible candidates for being necessarily reason-­entailing without, themselves, being normative facts.

Is Prudential Discourse Normative?  37 Let us then look at some of the grounds that are offered for the claim that moral judgements are normative (which I will call ‘markers’ of the normative) to see how well they carry over to the prudential case.13 My claim is not that any of these alone is a sufficient condition of some judgement type being normative. It is easy to find false positives— namely, forms of non-­normative judgements that have at least one of these features. However, it seems plausible that there are no forms of non-­normative discourse that share all of these features. Finally, let me emphasize that these markers serve as evidence that some type of judgement is normative. I do not take them to constitute the nature of normative judgement. Normative types of judgements have these features (or at least some subset of them) because they are normative, whatever precisely that comes to, rather than being normative because they have these features.14

3.  Markers of the Normative 3.1  Evaluative and Prescriptive One feature of moral judgements commonly cited in connection with their normativity is that they are evaluative, critical, or prescriptive. Moral judgements do not merely represent how the world, agents, and their actions are. Rather they are about how the world or its constituents ought to be, how we have reason to make it, how it would be good, bad, better, best, or worst for it to be. This feature of moral judgements is reflected in the vocabulary in which they are made and expressed. We use evaluative terms (‘good’, ‘bad’, ‘value’, ‘best’) and directive terms (‘rights’, ‘wrong’, ‘permitted’, ‘ought’, ‘should’, ‘must’).

13  Just to emphasize, my main claim here is that any plausible story of why moral judgements are normative will carry over to prudential judgements. So one could reject any (or all) of my claims about the marks of the normativity of moral judgements. 14  To expand: I take the points I am about to develop to be general articulations of the data points that we use when formulating metaethical views, data that must be explained, or explained away, by such views. For a clear example of this methodology see Ridge (2014).

38  Dear Prudence Here we find a perfect fit. Prudential judgements are equally evalu­ative, critical, or prescriptive. They do not merely represent how the world, agents, and their actions are but instead how the world or its constituents ought to be, how we have reason to make it, how it would be good, bad, better, best, or worst for it to be. We also find the same evalu­ative (‘good for’, ‘better for’) and directive (‘reason’, ‘ought’, ‘should’, ‘must’, ‘needs’) vocabulary used to make prudential judgements.15

3.2  Authority Moral judgements are typically held up as having a kind of authority which in turn gives them a special place in deliberation. It is often taken to be a feature of moral discourse that morality makes a difference to the reasons an agent has such that if an action is morally wrong then this entails reasons for the agent not to perform that action and reasons for the agent to feel shame or guilt and to be blamed by others.16 This authoritative nature of morality is also reflected in the kind of centrality with which it is treated in deliberation where deciding what to do allthings-­considered is closely connected with deciding what one morally ought to do.17 It is an everyday feature of conversation and deliberation that accepting that it would be morally wrong to perform some action is typically treated as settling whether one ought to do it all-­thingsconsidered. Furthermore, even where people allow that morality does not settle the question of what an agent ought to do all-­things-­considered, it is typically common ground that morality makes a significant difference to what an agent ought to do, by entailing reasons. Here we have a close fit between moral and prudential judgements. As argued in the previous chapter, prudential value plausibly makes a difference to the reasons an agent has; if an option is best for someone then this means that they have at least some, prudential, reason to prefer 15  On the use of the modal ‘needs’ in prudential discourse, see Chapter 3. 16  That is not to deny the existence of scepticism on this point (e.g., the Why Be Moral? challenge). 17  I here assume the coherence of ought all-­things-­considered. But see Tiffany (2007), Baker (2017), and Copp (ms).

Is Prudential Discourse Normative?  39 that option. And, often, deciding what would be best for oneself settles what one ought to do all-­things-­considered. More commonly, in discussion we either presuppose the significance of prudential considerations or accept them as authoritative by default, when introduced, in contrast to other sets of considerations where we are, or can be, called upon to explain or justify their relevance. One might be sceptical as to whether prudence has quite such a central role in deliberation as that which we find with morality. I agree that there is a less close connection between what one ought to do all-­thingsconsidered and what one prudentially ought to do. But this, I suggest, is specifically because moral demands plausibly trump prudential demands across many (though not all) contexts. If we focus on contexts where moral demands are screened off the normative authority of prudence is clearer to see. In these contexts there is a close connection between what one ought to do and what one prudentially ought to do; that some option would be (sufficiently) bad for you is treated as settling whether you ought to bring about that option (and vice versa for what would be best for you). Furthermore, in contexts where moral demands are in play, we treat prudential considerations as to-­be-­weighed against moral requirements and one of the most plausible sets of cases where morality fails to settle what one ought to do all-­things-­considered is the set in which prudence specifically, rather than any other prima facie normative domain, trumps morality because morality’s demands are too costly. This suggests that prudence enjoys the same kind of authority as morality.18

3.3  Affective Connection For at least a central subset of moral judgements, there is a close connection between them and certain affective responses (including reactive attitudes).19 For example, judging that morally I ought not to perform 18 Though, again, this depends on one’s view about cases where prudence can override morality in determining what one ought to do all-­things-­considered. 19  ‘Close connection’ is deliberately vague. The most hardcore of externalists about moral motivation can accept this thesis as long as they think it just notes a correlation.

40  Dear Prudence some action is closely connected to having some sort of negative attitude towards it. Typically, someone who makes the judgement that some action is morally wrong will blame and feel anger towards others who perform the action and guilt towards themselves if they perform such an action. And if an agent judges that, morally, they ought to perform some action we expect them (even if defeasibly) to be motivated to perform that action. We make predictions about affective responses and mo­tiv­ ation in light of moral judgements and we make these predictions on the basis that an agent who falsified them would be criticizable in some way.20 Here, again, we find a perfect parallel with prudential judgement. In the case of prudential judgements, or at least the analogous central subset of them, we find a close connection between those judgements and affective responses (including reactive attitudes). Judging that an action is what one prudentially ought to do is closely connected to being at least somewhat motivated to perform that action and, in cases where one has no agency to bring it about, judging that some option would be best for oneself is closely connected to hoping that that option comes about. Making one’s own situation much worse, even if there is no cost to others, is closely connected to self-­pity and self-­recrimination. Thus, we similarly predict that agents have responses that line up with their prudential judgements and we make these predictions on the grounds that to lack such responses would be a failing.21

3.4  Disagreement Moral discourse is striking in the way that it permits of radical dis­agree­ ment without manifestation of conceptual incompetence or loss of common subject matter.22 People of radically divergent general worldviews and across large differences in social and cultural milieu seem equally to make moral judgements and to be in the condition of agreeing and dis­ 20  For representative discussion of these issues see (e.g.) Smith (1994), Svavarsdóttir (1999). 21  I return to these issues in Chapter 5. 22 See Hare (1952), Horgan & Timmons (1991). For scepticism about these cases see Dowell (2016).

Is Prudential Discourse Normative?  41 agree­ing morally, despite widespread differences in their wider set of beliefs. If there are limits to what one can think is morally wrong without thereby showing oneself to be incompetent with moral concepts, these limits seem lax, permitting a large range of fundamental moral disagreement.23 Here too we get a perfect fit. Prudential discourse also permits extremely radical disagreement, without manifestation of conceptual incompetence or loss of a common subject matter. Stephen Darwall provides an argument on precisely these lines as follows: [I]t seems possible for two people who care about someone, S, to coherently disagree about whether something, X, is good for S, even though they agree completely about all the non-­normative facts concerning X and S.  Suppose, for example, that X is a pleasant illusory belief of S’s, say, that S’s novel has sold 10,000 copies (when in fact it has sold only 12). It would seem that two people could be agreed about everything else, but simply disagree about whether this pleasant illusory belief is good for S or makes some contribution to his welfare, other things being equal . . . If this is right, then welfare . . . must be an explicitly normative concept.24

We can broaden out from single illusory pleasant beliefs to note that people of radically divergent general worldviews and across large differences in social and cultural milieu seem equally to make prudential judgements and to agree and disagree prudentially, despite widespread 23  Foot (1958: 512) gives the example of thinking it is morally wrong to look at hedgehogs in the light of the moon or run clockwise around trees. Perhaps this is a limit on what can count as a moral judgement. For the attempt to motivate a wide range of conceptual restrictions on moral judgements see Cuneo & Shafer-­Landau (2014). I think that the same applies to prudence. Perhaps there are such fixed points when it comes to prudence (e.g., that one cannot coherently hold that the only thing with prudential value is looking at hedgehogs in the light of the moon.) Morality and prudence thus seem on a par here. 24  Darwall (2004: 10) gives this as a direct argument to argue for the view that prudential judgements are normative (along with his particular account of their content) as we see from the conditional in the last sentence of the quotation. This argument is, rightly, criticized by Olson (2006) on the grounds that this disagreement phenomenon is found in other, non-­ normative, domains, such as genre judgements. Olson’s objection succeeds, I think, against the argument that this feature is sufficient for a type of judgement to be normative. For relevant discussion of legal disagreement see (e.g.) Plunkett & Sundell (2013).

42  Dear Prudence differences in their wider set of beliefs. Close to home, we have the example of Nozick’s experience machine. Some philosophers hold views of well-­being where there is nothing prudentially deficient about life in the experience machine whatsoever (assuming a large enough supply of hedons).25 By contrast, many philosophers think that such a life would hold little to no prudential value, because what matters prudentially is the exercise and development of our human capacities. Some philo­ sophers think that getting what you want is what matters fundamentally, others think that some things have prudential value independently of whether we desire those things. Moreover, it is easy to imagine radically different theories of prudential value across different communities (constituting prudential ‘twin earth’ scenarios).26 We can, for instance, imagine a pair of communities where one has a set of practices that suggests a hedonistic conception of prudential value whilst the practices in the other community suggests a perfectionist conception of prudential value. For example, in the first community, people advise their loved ones to develop their capacities just in case by doing so they thereby maximize enjoyment, whereas in the second community they advise their loved ones to only seek pleasures where this is connected to the maximal exercise and development of their capacities. Similarly, in the first community, those that are envied are the ones who experience most pleasure, whilst in the second community, it is those who develop and exercise their capacities most, even if their lives are known to be less pleasurable. From cases like these, it seems safe to say that prudential discourse, like moral discourse, permits a wide range of fundamental disagreement, without talking past each other or manifesting conceptual incompetence.

3.5  Testimony One interesting feature of moral judgement is that, unlike with common-­ or-­garden empirical judgements, it seems weird and problematic to

25  Nozick (1974: 42).

26  For more discussion of this issue see Chapter 6.

Is Prudential Discourse Normative?  43 defer to another entirely when making moral judgements. To give an example, it seems completely fine for you to believe, simply because I told you, that the stadium is on 21st Street. By contrast, it seems weird and problematic for you to believe, simply because I told you, that boxing is morally wrong.27 The ‘simply because I told you’ qualifiers in the example are doing a lot of work, so let me spell out the idea more precisely. The relevant set of cases, where moral deference is problematic, are those that involve direct reliance upon purely moral content offered in testimony. So, the relevant cases are those where a thinker forms the moral judgement that P solely on the basis of Y’s testimony that P (where P is some moral proposition without specification of a ground).28 Now, what exactly is problematic about direct reliance upon purely moral content is not important for my purposes and many plausible candidates have been identified.29 What I am interested in showing here is the claim that moral judgements have this feature. Moreover, they plausibly share this feature with some other forms of practical normativity. This explains why we find the same problematicity when it comes to, for example, aesthetic judgements (‘This painting is beautiful. Debbie told me.’).30 Turning now to prudential judgements, we find, I suggest, that they behave like moral judgements when it comes to (the relevant class of) cases of forming judgements on the basis of the testimony of others. Whilst there is nothing amiss in my deferring to a doctor about how to get healthy, it is problematic for me to simply defer to the doctor, or 27  One aspect of the weirdness is that to profess (e.g.) that boxing is wrong on the basis of testimony, comes off as a joke or as insincere. Compare, ‘the stadium is this way, Gwen told me’ with ‘boxing is morally wrong, Elijah told me.’ 28  The implicit contrasts are thus with: Impure Moral Testimony: Testimony whose content is: (i) some morally relevant descriptive fact OR (ii) some morally relevant descriptive fact indicated as grounds for some moral fact. AND Indirect Reliance upon Moral Testimony: X forms or sustains the moral judgement that P partly on the basis of Y’s testimony that P. (These formulations are mistakenly factive in mul­ tiple ways that do not affect the main point). 29  For discussion of the problematicity (or not) of moral defence see: Lord (2018), Hopkins (2007), Hills (2009, 2019), Howell (2014), Jones (1999), McGrath (2009, 2011), Nickel (2001), Sliwa (2012), Fletcher (2017), Driver (2006), Enoch (2014), Crisp (2014), Lewis (2020), Callahan (2018), Groll & Decker (2014), Mogensen (2017), and Wodak (forthcoming). 30  On this see, for example, Hopkins (2011), Lord (2018), and Whiting (2015).

44  Dear Prudence anyone else, about whether I would be better off sacrificing athletic achievement for some increased degree of health (or vice versa) or whether to become a parent. When it comes to these kinds of fundamental prudential questions, simply taking someone’s word for it is strange and seemingly problematic, just like in the moral case.31 To sum up this section, I have introduced five markers of the normativity of some set of judgements—five (sets of) ideas that are typically appealed to when articulating the normativity (and theoretical interest and importance) of moral judgements. I then argued that we find sufficiently close, often perfect, parallels in the case of prudential judgements. This gives us grounds for holding that prudential judgements are normative, on a par with moral judgements. As I noted earlier, to show that prudential judgements are normative by analogy with moral judgements I need to establish relevant parallels between them and show that there is no significant disanalogy. I will now consider objections to the effect that there is a significant disanalogy between moral and prudential judgements and so PJN is false.

4.  Why Think Prudential Judgements Are Not Normative Judgements? One non-­starter is the idea that prudential judgements are non-­normative because the prudential, as a domain, does not contain obligations, prohibitions, and permissions. We might put this thus: (1)  For a judgement to be normative it must be a judgement about a prohibition, obligation, or permission. (2)  Prudential judgements are not judgements about prohibitions, obligations, and permissions. Therefore, (3)  Prudential judgements are not normative. 31  This point is obscured, I think, by the fact that most prudential advice is impure because in the kinds of context in which we give prudential advice we typically presuppose some substantive view of the prudential good of the person and attempt to identify productive and constitutive means to it.

Is Prudential Discourse Normative?  45 Where are the weak points in this argument? I am unsure about (2). It seems plausible that there are prudential prohibitions. If one simply neglects oneself for no further purpose then one is criticizable (and not merely morally criticizable).32 However, whilst (2) is plausibly deniable, the argument fails because (1) is false. As pointed out above, judgements about the moral goodness and badness of states of affairs are normative. Similarly, judgements about the moral goodness or badness of actions and characters are another counterexample to (1). That doing the right thing for the wrong reason is morally bad or that cruelty makes one a morally bad person are paradigmatic instances of normative judgements even if they are not judgements about permissions etc. This should not be a controversial claim. In outlining his error theory about the normative, J.L. Mackie writes: There are no objective values . . . [This] is meant to include not only moral goodness[.]33

Given that evaluative judgements, such as moral goodness judgements, are part of the set of normative judgements, we can reject (1) whatever we think about prudential judgements. Finally, at least part of the set of judgements at issue here are prudential reasons judgements. If there are such things, then these are clearly paradigmatically normative in precisely the way that the argument supposes prudential judgements not to be. Thus this argument is a non-­starter. Let us turn to a more promising argument.

5.  Because Prudential Value Is Desire-­Based? One ground for rejecting PJN that might be offered is the claim that prudential value is desire-­based and so prudential judgements are non-­normative. This is an inference that I think is widespread but implicit

32  Yes, even on a desert island. Or so I am inclined to think. 33  Mackie (1990: 15), my italics.

46  Dear Prudence in people’s thinking about prudential value. It is also one that is, I think, endorsed by Mackie in an overlooked passage where he considers the scope of his error theory. He writes: But if we think rather of a counsel of prudence as being related to the agent’s future welfare, to the satisfaction of desires that he does not yet have—not even to a present desire that his future desires should be satisfied—then a counsel of prudence is a categorical imperative, different indeed from a moral one, but analogous to it.34

As I read this, Mackie holds that whether moral and prudential judgements are analogous (with respect to normativity and the truth of error theory) stands or falls with the first-­order desire theory of well-­being.35 Transposed to the dialectical context of this chapter, the relevant argument is that a desire-­based theory of well-­being is true, that this entails that prudential properties are non-­normative properties, which means that prudential judgements are non-­normative. (Because this objection to PJN assumes the truth of a particular first-­order theory, I will talk in terms of prudential properties and facts but this should be interpreted in a sufficiently minimal way as to beg no questions between e.g., cognitivism vs non-­cognitivism about prudential discourse.) To fairly examine this objection, let us assume the following theory: Desire Theory of Well-­Being:  Something is good for someone if and only if and because it fulfils one of their actual desires. Would the truth of this theory entail that prudential properties and judgements are non-­normative? No it would not. The argument assumes that the fact that something fulfils a person’s desire is a non-­normative fact, which we can grant for the sake of argument. Would that suffice to show that the left-­hand side, the prudential fact, is a non-­normative fact? No, that’s too quick. Just because a prudential fact is explained by a non-­normative fact, it doesn’t follow that it is a non-­normative fact. 34  Mackie (1990: 29). Obviously a lot hangs on the ‘related to’ in interpreting this passage. 35  For creative reconstruction of what Mackie is doing here see Ridge (2020).

Is Prudential Discourse Normative?  47 To see why, note the failure of the analogous argument for the non-­normativity of morality, on the assumption of the dependence of the moral on the non-­moral. Suppose this moral theory is true: Moral Theory:  It is morally wrong for X to Φ if and only if and because X’s Φ-ing does not maximize pleasure. On this kind of view, a fact about total pleasure that would be generated by an action—a paradigmatic non-­normative fact—explains a moral fact. Would the truth of this explanatory claim, coupled with the right hand side being non-­normative, mean that the moral fact is also a non-­normative fact? Plainly not. We do not simply infer from the nonnormative nature of the facts that make actions right and wrong that facts about rightness and wrongness are, themselves, non-­normative. We find many cases where non-­normative facts ground normative facts. For example, facts about pain and about temperature ground normative facts about what people have reason to do. We see this from examples such as: temperature fact:  the fact that it is minus 15 degrees outside gives you a reason to take a coat. pain fact:  the fact that it would be painful gives you a reason not to let the dog bite you. Thus, it would be a mistake to conclude that prudential facts are non-­ ­ normative simply because the facts that ground them are non-­normative. Just to be clear about the dialectic here, my claim is that prudential value being desire-­based would not entail that prudential properties and judgements are non-­normative. For even if the facts that ground prudential facts are non-­normative, this does not, itself, settle the nature of prudential facts.36 And we can see that this inference is dubious by looking at other domains.

36  On this see Hampton (1998) and Bedke (2010).

48  Dear Prudence The way in which this argument fails extends to other theories of well-­being. Just as the truth of a desire theory of well-­being is insufficient to show that prudential properties and judgements are nonnormative, the truth of hedonism (or any other theory) would be similarly insufficient. Thus, we cannot point to the non-­normativity of the grounds of prudential facts to show that prudential properties and judgements are non-­normative.

6.  Because Prudential Properties Reduce to Non-­Normative Properties? Let us now consider a sophisticated successor to the objection that we just rejected. This objection holds that prudential properties are identical to desire-­fulfilment properties and so that is why prudential discourse is non-­normative. To make maximally clear how this differs from the previous objection, on this new proposal it is not merely that desirefulfilments have prudential value. Rather, on this view, what it is to be good for someone just is to fulfil their desire. This view thus suggests a reductive account of the evaluative prudential property of being good for someone.37 It combines this with the claim that such a reductive account entails that prudential judgements are non-­normative. This objection need not be run with the specific claim that it is desirefulfilment that the property of being good for someone is identical to. Another natural candidate would be that evaluative prudential properties are identical to hedonic properties—such that to be good for someone just is to be pleasurable for them (and to be bad for someone just is to be painful for them). What, then, to make of this objection? There are two weaknesses to it, in the form of heavy argumentative burdens which the objection incurs (burdens which make it difficult to provide a direct reply to the objection, without it being worked out in a lot more detail). First, for the objection to work, the objector must supply the reductive account of the property of being good for someone. For reasons 37  For discussion of these general issues see Schroeder (2005, 2007).

Is Prudential Discourse Normative?  49 familiar from metaethics, this is no mean feat. As argued above, one of the normative markers of prudential discourse which it shares with moral discourse is the possibility of radical disagreement. Such dis­ agree­ment provides one obstacle to identifying the property of being good for someone with (e.g.) the property of fulfilling the person’s desire (or, similarly, some hedonic property).38 The second argumentative burden that this strategy takes on is that of showing that the reductive account of the property of being good for someone entails that this property and judgements ascribing it are non-­normative. Put another way, suppose that the property of being good for someone just is some hedonic property (and so on for the other evalu­ative prudential properties). The question now is: why does that mean that prudential properties and judgements are non-­normative? This strategy must therefore show that a form of reductive naturalism about prudential properties is incompatible with prudential judgements being normative judgements.39 An analogous objection, this time focused on directive prudential properties, is the view that prudential reasons are not normative reasons (and so on for other directive prudential properties) and that this means that prudential judgements are not normative judgements. That is, someone might grant that there are prudential reasons, and oughts, etc. but hold that these are somehow not normative or, at least, not normative in the way that moral reasons are. One model that might be appealed to in support of this is the idea that prudential reasons are like reasons (or ‘reasons’) of etiquette or that what it is for something to be prudentially required is similar to what it is for something to be required by law, or by a company code of conduct, or by etiquette (a position we examined in Chapter 1).40 Another model would be that prudential reasons are hypothetical, rather than cat­egor­ ic­al, reasons (another position that we saw Chapter 1). This is no phantasm. We find real instances of this position in the debate about what to do with moral discourse if error theory is true. In 38  See Heathwood (2009) for discussion of this issue. 39  On these issues see, for example, Schroeder (2005). 40  For discussion of these issues see Dorsey (2016: ch. 1), Olson (2011a, 2011b), and Stroud (1998: 172).

50  Dear Prudence that debate, it is frequently treated as utterly unproblematic to invoke prudential reasons even within a dialectical context that supposes that there are no moral reasons (and perhaps no normative reasons whatsoever). For example, Richard Joyce writes: But it is important to remind ourselves that even the eliminativist error theorist will still have plenty of good and strong reasons—many of them self-­interested reasons—for being nice to her fellows[.]41

This passage, like many others, embodies the view of prudential reasons as non-­normative (or, at least, as normative in some different way or to a lesser extent, than moral reasons). It is important to distinguish between the claim that there are no prudential reasons and this latest position, namely that there are prudential reasons but that they are not bona fide normative reasons of the sort that could make prudential judgements normative judgements on a par with moral judgements. What could the difference between prudential and moral reasons consist in, such that prudential reasons are non-­normative? As we saw earlier, it cannot be merely that prudential reasons are grounded in desires because a desire-­fulfilment theory of well-­being is true. That is compatible with their being normative. Rather, the property of being a prudential reason must be radically different from the property of being a moral reason, such that the prudential one is non-­normative. We saw in Chapter 1 that there is a genuine clash between Prudential Reasons and Humeanism about reasons. This means we cannot treat prudential reasons as distinct from moral reasons by treating them as hypothetical, rather than categorical, reasons. This is because, spelled out, this is the view that there are no prudential reasons. If we have ­reasons to promote well-­being only because that promotes the fulfilment of our desires, then that is to think that evaluative prudential facts themselves do not, fundamentally, generate reasons (which is, ipso facto, to deny that there are prudential reasons). There is thus no way of

41  Joyce (2007: 75). I discuss this issue again in Chapter 6.

Is Prudential Discourse Normative?  51 holding that there are prudential reasons and that they are non-­normative by being hypothetical. As with the objection we considered in section 5, the best version of the idea is that being a prudential reason just is some relation such as that of promoting fulfilment of the agent’s desire, or generating pleasure for the agent. And, again, as with the previous objection, the person making this objection has two jobs to do. They must do the hard work of supplying a plausible reduction of the relation of being a prudential reason and show that this relation (and judgements ascribing it) are thereby nonnormative. Furthermore, they must also supply a plausible account of why prudential discourse (like moral discourse) possesses the markers of normativity, in stark contrast to the kinds of domains that might be appealed to as models for this reduction of prudential discourse (namely law, codes of conduct, and etiquette).42 To put it rhetorically: if prudential reasons are so different from moral reasons, why do they seem so similar? Before moving on, let me note that whilst I considered the objections in this section individually neither alone can be a fully general objection to PJN for the following reason—PJN encompasses evaluative and directive judgements and each of these objections targets only one of these sets. This means that these objections, individually, could only establish that one part of prudential discourse is non-­ normative. Someone who wanted to argue that prudential discourse is wholly non-­normative must either (a) combine these objections (or similar ones) to show that neither evaluative nor directive prudential judgements are normative or (b) collapse these tasks into one by treating all prudential normative properties as reducing to one and showing how that property (and judgements ascribing it) reduces to something non-­normative.43 These tasks are significant and it behoves those who wish to treat prudential normativity as second-­class normativity to undertake them.

42  These domains have some of the markers but do not possess them all. In each case, the connection to affect and motivation is implausible. It is also difficult (especially for codes of conduct and etiquette) to hold that there could be the same kind of radical disagreement. 43  For example, a buck-­passing view of all prudential properties would only need to provide one reduction—of the property of being a prudential reason.

52  Dear Prudence

7.  Because Prudential Judgements Are Invulnerable to Error Theory? A different reason one might be sceptical of the normativity of prudential value is that one might think that prudential judgements are invulnerable to a certain kind of error theory. The relevant kind of error theory is a global error theory about normativity. On such a view, normative properties would have a certain nature if they were genuine properties, but no normative properties are instantiated.44 The person wielding this argument starts by noting that an obvious consequence of some domain of facts being normative is that they are within the set of facts targeted by a global normative error theory, such that the following argument would be valid: (1)  (Assumption) Prudential facts are normative. (2)  If global error theory about the normative is true then there are no normative facts. Therefore, (3)  If global error theory about the normative is true then there are no prudential facts. The person wielding this argument then uses the implausibility of (3) to cast doubt upon (1). Here is how they might put it: ‘Global normative error theory would rule out a lot (including morality). But even the truth of global normative error theory would not entail that there are no prudential facts. Even if global normative error theory were true, pain would still be bad for us!’45 We can render this as the following argument:

44  For moral error theory the locus classicus is Mackie (1990). For discussions of more generalized kinds of normative error theories see, for example, Cuneo (2007), Lillehammer (2007), Bedke (2010), Olson (2011), Rowland (2013,  2016), Cowie (2014b,  2016a), and Streumer (2017). 45  I find this style of argument tempting in the case of attributive goodness. Global normative error theory would not entail that there are no good toasters or good kettles.

Is Prudential Discourse Normative?  53 (I)  Global normative error theory could be true.46 (II)  Nihilism about prudential value (such that e.g., pain is never bad for us) could not be true. Therefore, (III)  Prudential facts aren’t targeted by global normative error theory. Therefore, (IV)  Prudential facts aren’t normative. A few pointers about the argument. Premise (II) contains the ‘never’ qualifier because it should be agreed on both sides that it might be true that some pains are not bad for us. Depending on one’s first-­order view, these might be pains that are deserved, pains that are very mild, or pains that are either not minded or which are relished (readers may supply their own example).47 The argument does not require accepting global normative error theory. The argument is independent of one’s meta-­normative views. It only requires that one accept that if prudential value facts are normative then they would be targeted by global normative error theory. It is the un­accept­abil­ity of the conditional claim that the argument relies upon. The proponent of the argument’s main claim is that it just cannot turn out to be true that (e.g.) pain is never bad for people. They might point to a paradigm case: the excruciating pain felt by an entirely innocent person and note the implausibility of saying that no pain like that is ever bad for the person who experiences it. This kind of argument might be powerful in some cases. For example, take a form of natural law theory that simply identifies legal properties with moral properties. Against such a view, one might point out the conditional conclusion that if global normative error theory is true then there are no laws and argue that the implausibility of that conditional 46 This is consistent with Streumer’s (2013,  2017) argument that we cannot believe the global normative error theory. All that it requires is that we can accept that the global normative error theory might be true. 47  See Bradford (2020). One might of course hold an unusual first order view of pain and its (lack of) prudential disvalue. If so, one should substitute some other prudential claim that one finds to have the requisite plausibility.

54  Dear Prudence casts doubt on the idea that legal properties are moral properties (or any other robustly normative properties). In this dialectical context, however, the argument is not very powerful. Once we see the parallels between moral discourse and prudential discourse that we saw above, then we have a plausible case for thinking that prudential discourse is normative. With that in place, we cannot treat our credence in the claim that at least some pains are bad for people as independent of the truth of the claim that global normative error theory might be true. Rather, unless we have some other ground for thinking that prudential judgements are non-­normative, premise (II) is evidence against premise (I). Put another way, the implausibility of prudential nihilism is evidence against global normative error theory, given the observation that prudential discourse shares the markers of normative judgements.48

8.  Conclusion In this chapter I have argued that prudential discourse is normative. I did so by focusing on the questions of whether prudential judgements are normative and by examining the similarity between moral judgements and prudential judgements. To that end, I laid out some of the markers of the normativity of moral judgements before arguing that we find these same features with respect to prudential judgements. There is thus a provisional case for treating prudential judgements as normative judgements, on the assumption that moral judgements are normative. Subsequently, I examined and sought to undermine some arguments against the normativity of prudential judgements, some of which were premised on the idea that prudential value is somehow desire-­based. I argued that none of these arguments succeeds in showing that prudential value, reasons, and judgements are non-­normative. We thus have a provisional case for the view that prudential discourse is a normative form of discourse, on all fours with moral discourse. Let me close this chapter by providing a reminder of the modesty of my ambitions both here in this chapter and in this book. I aim to defend 48  I return to error theory and prudential normativity in Chapter 6.

Is Prudential Discourse Normative?  55 some particular claims about prudential discourse and to show that prudential discourse is normative. I have not sought to defend a specific meta-­prudential theory.49 In the next part (Part  2), I defend a collection of views about the nature of prudential thought and talk. These chapters can be read in two ways. They can be read as elaborations upon the way in which prudential discourse has the markers of the normative distinguished above. Thus, the next part of the book can be read as providing additional support for the claim that prudential discourse is normative, as evidenced by its having these features. The alternative way of reading the next part of the book is as bracketing the question of whether prudential discourse is normative and as simply an inquiry into the nature of prudential discourse by examining various aspects of it in detail. Readers may choose their own adventure.

49  Indeed, as mentioned at the outset, one of my aims is to provoke inquiry into such meta-­ prudential questions.

PART T WO

THE NAT U R E OF PRU DE NT IA L DISC OU R SE

3 Prudential Language and Context (I) Good for and Needs

In the next two chapters, I examine prudential language. The set of prudential vocabulary encompasses at least the following terms, which we can sort into the following categories: Evaluative:  ‘good for’, ‘bad for’, ‘best for’, ‘worst for’, ‘better for’, ‘worse for’, ‘benefit’, ‘harm’ Directive:  ‘reason’, ‘ought’, ‘must’, ‘needs’, ‘may’.1 I will use this division to structure this chapter. In Part A, I explore evaluative prudential language, focusing mostly on ‘good for’. In Part B, I move to directive prudential language, focusing on ‘needs’ and ‘must’ claims. In each case I show how work in philosophy of language can be applied to prudential talk in a way that illuminates how it works. The general lesson is that prudential language is an instance of evaluative and modal thought and talk more generally. Thus, the best theories of these apply just as well to the use of evaluative and directive terms in prudential contexts.

Part A—Evaluative Prudential Language: Good For 1.  Introduction It is commonplace for philosophers writing on prudential value, or ­well-­being, to note that there is some kind of plurality to the way that we 1  As will become clear through this chapter, many (perhaps all of these) can be used to make prudential claims or non-­prudential claims. Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0004

60  Dear Prudence use the expression ‘good for’.2 For example, take the following, typical, set of sentences: (1)  ‘Light colours are good for small spaces.’ (2)  ‘Oil is good for an engine.’ (3)  ‘It would be good for Jane to be head of department.’ (4)  ‘The repeal of Obamacare is good for the Tea Party.’ (5)  ‘Water is good for plants.’ Examples such as these have led many to hold that there are many senses of ‘good for’ and that when we engage in prudential discourse we use some specific, prudential sense of the expression. Connie Rosati gives an especially clear expression of this view as follows: We commonly talk about what is good for someone or something. But the expression ‘good for’ seems to refer to differing relations depending on the context of use. Sometimes when we say that X is good for Y, we seem to refer to the instrumental relation X stands in to Y, as when we say that hammers are good for pounding nails. Sometimes we seem to refer to a maintenance relation between X and Y, whereby X will put or keep Y in the condition standard users would want Y to be in, as when we say that changing the oil regularly is good for a car. Sometimes we seem to refer to the perspectival relation that holds when X is good from the point of view of Y, as when we say that advancing the ‘tea party’ agenda is good for tea party adherents but not for their political opponents. And sometimes, when we say that X is good for Y, we seem to refer to the welfarist relation X stands in to Y, when X contributes to what is variously called Y’s welfare, well-­being, self-­interest, flourishing, or (at least where Y is a person) personal good. My interest herein will lie with the welfarist good-­for relation, and my aim will be to defend a particular view about the structure and nature of this good-­for value.3

2 For example: Kraut (2007), Rosati (2009, 2020), Behrends (2011), Fletcher (2012), Korsgaard (2013), Finlay (2014), and Theunissen (2018). 3  Rosati (2020: 241).

Prudential Language and Context (I)  61 In a similar fashion, I previously argued: [W]e use the locution ‘good for’ [. . .] to express different relations. The good for relation that I am interested in here is that connected to ­well-­being . . . But there are many other ways in which we use the phrase ‘good for’ that clearly do not express the same relation and it is helpful to distinguish these[.]

Jeff Behrends expresses a similar view when, after posing the following questions, he goes on to argue that there is no unified analysis of ‘good for’ claims: Human beings, though, are not the only subjects of good-­for claims. For example, we should all agree that a healthy diet is good for a pet dog, that water and sunlight are good for roses, and that regular oil changes are good for cars. What is it, though, for something to be good for a plant? Are oil changes good for cars in the same way that pleasure is good for a person? What we need to know is whether there is a single relation that is appropriately used in each of these contexts, or whether there are in fact multiple relations that are sometimes expressed when we claim that something is good for some subject.4

For my purposes, the main ideas expressed in these passages (ideas that are commonplace in philosophical discussion of prudential value) are: (A)  There are different senses or meanings of ‘good for’. (B) Within this plurality is a specific sense employed within prudential thought and talk. This combination of views is understandable, given the apparent dissimilarity between (1)–(5) and the examples in the passage from Rosati. However, I do not think that it is correct, at least on a straightforward interpretation (more on this caveat shortly). 4  Behrends (2011: 121). Behrends’ argument is, roughly, that ‘good for’ talk is sometimes, but not always, to be understood in terms of function and thus there is no unified analysis.

62  Dear Prudence A related question, one that crops up in discussion more than in print, is the question of which things are welfare subjects. Sentences like the above make clear that we talk about what is good for a wide range of things (engines, ants, economies, and a great many others). Is our doing so sufficient to make all of these things welfare subjects? Do ants and engines have levels of well-­being? If not, how do we distinguish between ‘good for’ talk that is connected to welfare and that which is not? In this part, I show how an end-­relational account of ‘good’ of the sort proposed by Paul Ziff (1960), and recently developed in detail by Steve Finlay (2014), provides a plausible account of ‘good for’ talk and il­lu­min­ ates these issues. I argue for the proposal by showcasing the intrinsic merits of the view and by demonstrating its superiority to rival pro­posals, including Richard Kraut’s recent defence of what he calls ‘developmentalism’ and a view that treats all ‘good for’ talk as concerning attributive goodness. The part will proceed as follows. I first (section 2) take a closer look at ‘good for’ talk and use that to motivate some data points for, and constraints on, an account of ‘good for’. I then (section 3) provide and assess a first proposal for how to understand ‘good for’ talk. Whilst this proposal ultimately fails, the way in which it fails is instructive for formulating and assessing the alternate, end-­relational, proposal. I present this in section 4 and outline its merits. Finally, in section 5, I outline the main competitor proposal, Richard Kraut’s developmentalism, and argue that it is inferior to the view defended herein. Finally, before moving to the main discussion, let me note at the outset that the proposal for understanding ‘good for’ talk that I will defend could, instead, be read as an interpretation and precisification of (A) and (B). I have no quarrel with that way of reading the view. Either way, end-­relationalism represents an improvement in our theoretical understanding of ‘good for’ talk.

2.  Some Data about ‘Good for’ Talk: Unity and Diversity Remember our initial set of ‘good for’ sentences: (1)   ‘Light colours are good for small spaces.’ (2)   ‘Oil is good for an engine.’

Prudential Language and Context (I)  63 (3)   ‘It would be good for Jane to be head of department.’ (4)   ‘The repeal of Obamacare is good for the Tea Party.’ (5)   ‘Water is good for plants.’ and the claims that Rosati, earlier-­Fletcher, Behrends etc. hold on the basis of sets of examples such as these: (A)  There are different senses or meanings of ‘good for’. (B) Within this plurality is a specific sense employed within prudential thought and talk. One challenge for claims (A) and (B) is that if ‘good for’ in (1) really has a different meaning from ‘good for’ in (2) and (3) then we should not expect these ‘good for’s to have a common translation from English into other languages. Yet when we examine other languages, we find that the expression ‘good for’ as it appears in (1) is given a common translation with ‘good for’ as it appears in (2) and (3). For example: French (1*)  ‘Les couleurs claires sont bonnes pour les petits espaces.’ (2*)  ‘L’huile est bonne pour un moteur.’ (3*)  ‘Ce serait bon pour Jane d'être chef de département.’ (4*)  L’abrogation d’Obamacare est bonne pour le Tea Party. (5*)  ‘L’eau est bonne pour les plantes.’ Spanish (1**)  ‘Los colores claros son buenos para espacios pequeños.’ (2**)  ‘El petróleo es bueno para un motor.’ (3**)  ‘Sería bueno que Jane fuera la cabeza del departamento.’ (4**)  ‘La derogación de Obamacare es buena para el Tea Party.’ (5**)  ‘El agua es buena para las plantas’. The same goes for other languages.5 This is reason to prefer an analysis of ‘good for’ claims in which we do not have different meanings between sentences (1)–(5) and the like. 5  Specifically: German, Dutch, and Polish. One limitation of this evidence is that it does not come from different families of languages.

64  Dear Prudence Another reason to prefer an analysis that does not generate different meanings is the observation that ‘good for’ fails a standard test for ambiguity: conjunction reduction.6 If we take two sentences that each use an ambiguous term and conjoin them into one sentence, the resulting sentence is zeugmatic. For example, if we combine ‘Jane fired the gun’ and ‘Jane fired the plumber’ we get the zeugmatic ‘Jane fired the plumber and the gun’, which gives us evidence that ‘fired’ is ambiguous. If, then, ‘good for’ were ambiguous, we should expect to find the same effect when we concatenate ‘good for’ claims across sets of diverse subjects as follows: (6)  ‘Water is good for plants, engines, dogs, and humans.’7 Sentence (6) is slightly unusual. But it is not problematic or zeugmatic. In that respect, it differs from: #(7)  ‘Bill Gates and chocolate cake are both rich.’ Sentence (7) sounds problematic and functions as a (weak) joke because ‘rich’ is being used to ascribe two wholly distinct properties that merely happen to be ascribed with homonymous terms. We now have our first data point. We should theorize about ‘good for’ talk in a way that can explain why it is not merely an accident of English that we use ‘good for’ across (1)–(5) etc. and why we can concatenate ‘good for’ claims.8 The second data point tends in the opposite direction, emphasizing the diversity of ‘good for’ claims. Looking again at our initial set of ‘good for’ sentences (1)–(5): (1)  ‘Light colours are good for small spaces.’ (2)  ‘Oil is good for an engine.’ 6  For more on this and other tests for ambiguity see Sennet (2016). 7  To emphasize, many writers introduce claims such as ‘oil is good for engines’ and ‘water is good for people’ to introduce the relevant question—whether there is a single sense of ‘good for’. As Behrends (2011: 3) helpfully put it: ‘Are oil changes good for cars in the same way that pleasure is good for a person?’ My move here is to claim that our ability to concatenate ‘Water is good for engines and people’ is evidence against any view that says there are diverse meanings of ‘good for’ between the original pair of sentences. 8  Whilst it could be coincidence, that hypothesis is presumably our last resort and I will ignore it herein.

Prudential Language and Context (I)  65 (3)  ‘It would be good for Jane to be head of department.’ (4)  ‘The repeal of Obamacare is good for the Tea Party.’ (5)  ‘Water is good for plants.’ Clearly, there is more than one way to interpret each of these. This is because, as Rosati (2020) suggested, ‘good for’ can, in context, be used to make different claims. Sentences (3)–(4) provide clear cases of such variety. Take sentence (4) first. This sentence can be used to make at least three different claims. It can be used to make a claim either about (a) what the Tea Party believes to be good or (b) what would benefit the Tea Party as a movement or (c) what would benefit its members (readings (b) and (c) arise due to ambiguity in ‘Tea Party’) We can see this by disambiguating as follows: (4a) ‘The repeal of Obamacare is good for the Tea Party. They think it’s a great thing.’ (4b) ‘The repeal of Obamacare is good for the Tea Party. Their influence will grow with this significant achievement of their goals.’ (4c) ‘The repeal of Obamacare is good for the Tea Party. Their members will have much lower tax bills.’ In context, a speaker may intend to convey more than one of (4a)–(4c). (4a) is the most important for my present point. (4a) is an instance of a general tendency to use ‘[adjective] for [noun]’ constructions as a way of ascribing attitudes (including, but not restricted to, beliefs). For ex­ample, we might say that ‘turquoise is green for him and blue for me’, that ‘the film was scary for me’, that ‘the food was spicy for me’, and so on. Finally, though the grammatical form is different, we see the same pattern in ‘For Trump, Climate Change is a hoax perpetrated by China.’ Henceforth, I ignore ‘good for’ claims where they are used as a means of ascribing attitudes, as in (4a).9

9  So all subsequent claims about ‘good for claims’ should be read as ignoring such attitude ascription claims.

66  Dear Prudence Turning to (3), this sentence can be used to make a claim about what is in Jane’s best interests or, alternatively, about what would be, generally, good. Someone might disambiguate these thus:10 (3a)  ‘It would be good for Jane to be head of department. She’d make everything much better.’ (3b)  ‘It would be good for Jane to be head of department. She’ll become more confident as a result of taking it on.’ The difference between these is made clear if the person were to go on to claim: (3c)  ‘It would be good for Jane to be head of department. She’d make everything much better. It’d be terrible for her, admittedly.’ In sentences (1)–(5) we therefore see variety of some sort. This gives us our second data point; an account of ‘good for’ claims must explain the diversity that ‘good for’ claims exhibit. Putting this together with the discussion above we thus have two data points to guide inquiry into ‘good for’, which I call ‘diversity’ and ‘unity’: Diversity:  ‘Good for’ claims are different in some way that we must explain. Unity:  Whatever that difference is, we should explain it in a way that makes sense of the fact that it is not a contingent artefact of English that we use the same expression for these claims. The aim is to provide a proposal for how to understand ‘good for’ claims that properly accommodates these two data points. In the next section, I discuss a proposal for understanding ‘good for’ claims as claims about attributive goodness. This proposal, I argue, ultimately fails but is worth discussing for two reasons. First, because it is one which I think many 10  Fletcher (2012: 2) uses a similar example to make this distinction. Finlay (2014: §2.4) treats these cases as involving ellipsis, such that (3a) is elliptical for ‘It would be good for Jane, for Jane to be head of department. She’d make everything much better.’ That seems plaus­ ible to me.

Prudential Language and Context (I)  67 would find attractive. Second, because seeing why it fails serves to ­highlight the superiority of the Ziff-­Finlay, end-­relational view defended herein (which I present in the subsequent section).11

3.  First Proposal: ‘Good for’ and Attributive Goodness Suppose we interpreted ‘good for’ across these sentences as follows: Good For: Attributive (GFA):  to be good for an X is to be a (­productive or constituent) means to its becoming, or remaining, an attributively good X.12

An ‘attributively good X’ is a good X or something that is good as an X. For example, an attributively good toaster is a good toaster. GFA has some virtues. It gets the right answers in the case of these sentences: (2)  ‘Oil is good for an engine.’ (8)  ‘Windows are good for houses.’ (9)  ‘Servicing is good for watches.’ This makes sense. After all, what could it be for something to be good for an engine except to be something that conduces to its being a good engine? And how could windows be good for a house except by making it a good (or better) house? And how could servicing be good for a watch without being conducive to its being a good watch. Given the fit

11  In describing the set of sentences above, and distinguishing the claims that they can be used to convey, I twice disambiguated them by clarifying that they were prudential claims. This was the case for sentences (4c) and (3b) where these sentences were used to make claims about what would benefit people or be in their interests. Is this enough to establish the claims above, that there’s a plurality of senses and a specific, well-­defined sense of ‘good for’ employed within prudential discourse? No, because even though some of the ‘good for’ claims concern prudential value that does not establish claim (B), which holds that there is a specific sense or meaning of ‘good for’ which relates to prudential value. 12  This is inspired by, but different from, Geach’s (1956) discussion of attributive goodness. A similar view, though not intended to apply to all ‘good for’ claims is discussed by Fletcher (2012).

68  Dear Prudence between these sentences and GFA, we should thus see whether GFA is plausible as applied across the whole set of ‘good for’ claims. Does GFA give plausible results across the whole range of good for claims? I think not. For whilst it seems exactly right when it comes to claims about engines, houses, and watches, GFA is mistaken when it comes to paradigmatic cases of welfare subjects (things with levels of well-­being). Consider: (10)  ‘Human company is good for dogs.’ (11)  ‘A varied environment is good for gorillas.’ (12)  ‘Pleasure is good for humans.’ GFA tells us that what it is for something to be good for an F is, necessarily, to be something that conduces to its being a good F. But in the case of (10)–(12), this does not seem plausible. Outside of contexts that make other standards salient—such as assessing dogs in a dog show—a good dog is, presumably, a well-­behaved dog. But it doesn’t seem plaus­ ible that there is a necessary connection between something’s being conducive to being a well-­behaved dog and it being something that enhances a dog’s well-­being. The same is true in the case of gorillas. Talk of what is good for a gorilla seems connected to what would make the gorilla happy and what would give the gorilla a higher level of well-­being rather than what would make it a good gorilla (whether that is to be understood as a well­behaved gorilla, a good instance of a gorilla etc.). The problem is most pronounced in the case of claims about what is good for humans. It just does not seem to be part of the meaning of ‘good for’ as it appears in (12) that what is good for humans makes them better humans. On the assumption that being a better human is to be a better-­behaved human,13 this position faces challenges from the fact that at least some pleasures and pains seem good and bad for humans irrespective of their connection to behaviour. We need not think about

13 One might want to supply some other candidate interpretation of being better qua human, such as better instance-­of-­a-­human. The challenge then is to make this alternative coherent and plausible.

Prudential Language and Context (I)  69 the difficult cases, such as the flourishing of the wicked. Mundane cases suffice to illustrate the point. The pain I experience at the dentist is bad for me, but it is not at all clear that it makes me vicious or inclines me to act wrongly.14 There is thus a disconnect between claims like (2), (8), and (9), for which GFA seems undeniable, and claims like (10)–(12), for which it is, well, eminently deniable. The difference between living things and artefacts makes a difference to how to understand ‘good for’ claims made about them. In the case of artefacts, the view that good for F means ‘is a means to being a good F’ seems plausible. In the case of living things, or, at least those examples we have just considered, it does not seem plaus­ ible that being good for an F is necessarily a matter of being conducive to their being a good F. From the failure of GFA, we might be tempted to reject the view that ‘good for’ claims are univocal. That is, we might think that GFA is the correct analysis of some of these claims (those concerning artefacts) but false for others (those concerning paradigmatic welfare subjects) and so there is a distinct sense of ‘good for’ for artefacts, for which GFA is correct, and another sense, one appropriate to welfare subjects, for which some other analysis is correct. That is plausibly what drives the picture that I presented at the start of this part, as outlined by Rosati, earlier­Fletcher, Behrends, etc. This would be premature. In the next section we shall see a schema for understanding all ‘good for’ claims, one that nicely explains exactly why these sentences are univocal, such that they receive common translations and allow of concatenation, whilst also explaining the differences between uses of ‘good for’. This view can explain why GFA is plausible for the range of cases that it is a good fit as well as enabling us to explain the difference between ‘good for’ claims.

14  One might accept a philosophical view that postulates a necessary connection between what generates virtue (or merely right behaviour) in humans and what generates well-­being but, that is a strong, substantive claim and not plausibly part of the meaning of ‘good for’.

70  Dear Prudence

4.  A Better Account of ‘Good for’: End-­Relationalism The best way to understand ‘good for’ talk is in fact the kind of ­end-­relational view of ‘good’ defended by Steve Finlay, following Ziff. The two main claims here (upon which I expand shortly) are as follows. First, that ‘good for’ is used simply to make claims about what promotes some relevant outcome.15 Second, the relevant outcome is sometimes the relevant subject’s well-­being.16 To understand this account properly we need to drill down into ‘promotes’, ‘relevant’ and ‘outcome’ but first let me issue a clarification. Ziff introduced a predecessor of this type of view thus: ‘good’ has associated with it the condition of answering to certain interests, which interests are in question being indicated either by the element modifying or the element modified by ‘good’ or by certain features of the context of utterance.17

Ziff ’s discussion is thus focused on ‘interests’, and he uses a broad conception of ‘interests’.18 By contrast, Finlay presents the view in terms of ‘ends’ but is explicit that an end need not be any person’s actual end. This is a matter of taste, but I prefer to forestall possible confusion here by framing things in terms of outcomes.19 Nevertheless I retain the end­relational name. Claims made using ‘good for’ are claims about what promotes some relevant outcome.20 Something promotes an outcome by being either a productive means to it or by being a constitutive means to it (or both). 15  There is no restriction on what the relevant outcome can be (and no requirement that it be valuable in any way). Many thanks to Nandi Theunissen for discussion here. 16  Finlay (2014: 32). See also Ziff (1960). 17  Ziff (1960: 218). 18  For rejection of Ziff ’s view as too narrow, even on his broad conception of ‘interests’ see Finlay (2014: 31). My reading of Ziff is that his conception of ‘interest’ might be broad enough to overlap with the end-­relational view—because, plausibly, we can say ‘in the interest of [x]’ as a way of referring to any outcome—but that this unclarity stems partly from his use of ‘interests’ and is reason to put the view in terms of outcomes. 19  Finlay himself also sometimes uses ‘outcomes’ talk, so this is not a major divergence from his discussion. 20  I am here sidestepping issues about the nature of the promotion relation. For discussion see: Finlay (2006), Schroeder (2007), and subsequent discussion such as DiPaolo & Behrends (2015) and Sharadin (2015).

Prudential Language and Context (I)  71 For example, a cement mixer is good for building houses. Why? By being a productive means of building good houses. Similarly, money is good for running for President because it is a productive means of running a campaign for President. By contrast, mutual respect is good for a relationship. Why? By being partly constitutive of a good relationship. Cool nerves are good for an assassin. Why? By being partly constitutive of being a good assassin.21 Thus, good for claims are claims about what (constitutively or productively) promotes some relevant outcome. Turn now to ‘relevant outcome’. The relevant outcome is sometimes explicitly stated in a good for claim, as in: (13)  (14)  (15)  (16) 

‘Oil is good for maintaining an engine in good condition.’ ‘Windows are good for making a house good to live in.’ ‘Light colours are good for making small spaces look attractive.’ ‘Servicing is good for maintaining a working watch.’

Very often, though, the relevant outcome is left implicit. This is presumably for reasons of conversational efficiency because the relevant outcome is easily recoverable from the context. This is clear in these cases: (1)  ‘Light colours are good for small spaces.’ (2)  ‘Oil is good for an engine.’ Unless context defeats these interpretations, (1) is naturally interpreted as concerning the aesthetic value of small rooms and (2) is interpreted in terms of the outcome of a well-­functioning engine. In the case of entities with a function, such as engines, this acts as the defeasible default interpretation of the sentence.22 So what is good for an F, where F has a function or end, is something that promotes F’s fulfilment of its function or realization of its end.23 For example:

21  It is likely a productive means also, of course. It is hard to find instances of constitutive means which are not plausibly productive means also. 22  As Finlay (2014: 37) puts it: ‘Since having a function is simply to serve some end, any functional noun-­phrase ‘K’ makes salient an end e.’ 23  Behrends (2011: 126) makes the same claim thus: ‘Being good for an artifact is a matter of promoting the function of an artifact.’

72  Dear Prudence (17)  ‘Couples’ counselling is good for X & Y’s marriage.’ (18)  ‘Sharpening is good for scissors.’ (19)  ‘Immigration is good for the economy.’ In each case, the audience understands the sentence in terms of the function of the relevant subject. Despite some differences between them, marriages, scissors, and economies are all entities with a function of promoting certain kinds of outcome and claims (17)–(19) are understood in terms of promotion of those. Here, then, is the general, end-­relational, recipe for ‘good for’ claims: End-­Relationalism (ER):  X is good for Y just in case X promotes relevant outcome O. ‘Good for’ claims vary with respect to what relevant outcome O is and how it is determined. O is sometimes explicit, as in sentences (13)–(16). At other times it is implicit. Where it is unspecified, and Y is something with a function, promotion of Y’s ability to perform that function is the default interpretation for a ‘good for’ claim. This fact explains why GFA, which we considered above, yielded correct answers across many cases. In the case of an object with a function, that determines the set of ­outcomes O that is relevant to a ‘good for’ claim and anything that ­promotes the ability of a functional F to fulfil its function will also be conducive to its being a good F. End-­Relationalism also nicely explains our immediate understanding of the relevant outcomes in sentences (17)–(19). With this understanding of how ‘good for’ works in general, and how it works in the case of entities with a function, let us go back to some difficult cases from earlier: (10)  ‘Human company is good for dogs.’ (11)  ‘A varied environment is good for gorillas.’ (12)  ‘Pleasure is good for humans.’ (5)  ‘Water is good for plants.’

Prudential Language and Context (I)  73 In each of these cases the relevant subject is a living thing.24 In the case of plants, I think it most plausible that talk of what is good for a plant is interpreted simply in terms of promotion of the plant’s health or, more broadly, its survival and reproduction. By contrast, things are less clear when it comes to claims about gorillas, dogs and humans. That is because (10)–(12) do not always trigger the interpretation in terms of survival and reproduction.25 Rather, I suggest, in the case of these sentences it seems much more plausible that the relevant outcome is promotion of their well-­being. For whilst humans, dogs, and gorillas are living things, they (unlike plants) are of sufficient complexity to live well or badly, over and above surviving and reproducing. They have lives that can go well (or badly) for them. Thus, claims about what is good for them are not, by default, understood as claims about what promotes their survival and reproduction. One piece of evidence for this is the contrast between these sentences and: (20)  ‘Moisture is good for bacteria.’ Bacteria survive and reproduce, but, I assume, they are not sufficiently complex to live well or badly. They are not welfare subjects.26 For that reason, we interpret (20) in terms of moisture being conducive to the survival and reproduction of bacteria, but, absent special contexts, we do not do the same in the case of humans, dogs, gorillas, and anything else that is a welfare subject. With welfare subjects, promotion of their well-­being acts as the default value of O for a good for claim about them.

24  Some hold that plants have a function—of surviving and reproducing. On that assumption, (5) will be understood by default as about what would promote the plant’s survival and reproduction. I take no stand on whether plants have functions. 25  For detailed discussion, and rejection, of the proposal that ‘good for’ claims made about living things be construed in terms of their function, see Behrends (2011: §§5–7). 26  What precisely makes something a welfare subject is an issue I will return to below.

74  Dear Prudence It is consistent with, and a virtue of, end-­relationalism that we can always force the interpretation in terms of survival and reproduction (or any other outcome) by explicitly mentioning it. For example: (21)  ‘Human company is good for dogs’ health/survival and reproduction.’ (22)  ‘A varied environment is good for gorillas’ health/survival and reproduction.’ (23)  ‘Pleasure is good for humans’ health/survival and reproduction.’ But with living creatures complex enough to have a level of well-­being, the default interpretation of good for claims made about them takes the relevant outcome O to be their well-­being. Let me recap. I argued that this is the most plausible account of ‘good for’ claims: End-­Relationalism:  X is good for Y if and only if X promotes relevant outcome O. Relevant outcome O is either (a) provided explicitly within the sentence or (b) generated by context. I have also discussed three different values for O and how they are generated by context if not explicitly stated: (I) In the case of non-­living things with a function, promotion of the fulfilment of that function provides the default value of O. (II)  In the case of living things, O is, by default, either: survival and reproduction (in the case of basic livings things). OR prudential value (in the case of more complex living things, things ­capable of a level of well-­being).27 27  As Finlay (2014: 34) puts it: ‘Our hypothesis is therefore that “good” sentences are fundamentally end-­relational, and that patient-­and interest-­relativity are merely special cases of this.’ I read Korsgaard (2013) as fundamentally rejecting this way of construing the relation between ‘good for’ and ‘good’.

Prudential Language and Context (I)  75 The variety between these explains the variation in ‘good for’ claims that we observe between: (2)     ‘Oil is good for an engine.’ {default outcome: fulfilment of function} (5)     ‘Water is good for plants.’ {default outcome: survival and reproduction} (20) ‘Moisture is good for bacteria.’{default outcome: survival and reproduction} (22) ‘A varied environment is good for gorillas.’ {default outcome: well-­being} (23)  ‘Pleasure is good for humans.’ {default outcome: well-­being} However, ER explains why we can concatenate ‘good for’ claims across diverse entities such as: ‘water is good for engines, bacteria, dogs, and gorillas’. In such a sentence we use ‘good for’ univocally; they are all claims about what promotes some relevant outcome O. It is just that O is determined differently for dogs and humans than for bacteria and engines. End-­ Relationalism yields plausible interpretations of ‘good for’ claims. It nicely handles all of the cases introduced above and it makes clear both the unity across the cases and the diversity between them. It brings out the way in which sentences like the set we started with are different (because they are concerned with different outcomes) whilst also bringing out the commonality of such claims (that they are all claims about what promotes some relevant outcome). Another major virtue of end-­relationalism is that it is extremely flex­ ible. It explains why we can make meaningful, even if false, claims about an enormous range of entities. Here is one: ‘Orange juice is good for the Catholic Church.’ End-­Relationalism explains our ability to understand this bizarre sentence by understanding the relevant outcome for the subject of the sentence. End-­Relationalism also enables us to explain why some ‘good for’ claims are so hard to interpret. Even though we can talk about what is good for a wide set of things, as Finlay points out: ‘without special context it’s hard to make sense of talk about what is good for a rock, a piece

76  Dear Prudence of lint, a shadow, or Wednesdays.’28 Why, absent special context, does talk of what’s good for a rock make no sense? Presumably it is because a rock is a non-­living thing which also has no function and so we lack a candidate for the relevant outcome. End-­Relationalism is also combinable with the full range of substantive views about prudential value, including some that I have been rejecting hitherto in this part. For instance, I have claimed that ‘good for bacteria’ is different from ‘good for humans’—because in the former case (alone) the relevant outcome is by default, survival and reproduction. But this is an additional commitment over and above end­relationalism. One could just as easily combine it with the claim that, for living things, the relevant outcome for a ‘good for’ claim is always survival and reproduction. Similarly, those who embrace a thoroughgoing Aristotelianism are free to combine end-­relationalism with the view that the default relevant outcome is always the fulfilment of the function of the subject of the claim (whether that is survival and reproduction or something else).29 End-­Relationalism is thus virtuously non-­committal. Finally, end-­relationalism makes completely clear that the question which often gets posed in discussion of ‘good for’ talk—of which things are welfare subjects—is a wholly separate, substantive, issue. ‘Good for’ talk is sometimes about promotion of well-­being but the fact that we can meaningfully talk of what is good for (e.g.) an engine does not commit us to thinking that engines are welfare subjects. End-­Relationalism, to its credit, takes no stand on which things are welfare subjects. So far we have examined ‘good for’ talk and I have argued that end­relationalism is the correct understanding of such talk. Let us briefly revisit the claims with which I began the part: (A)  There are different senses or meanings of ‘good for’. (B)  Within this plurality is a specific sense employed within prudential thought and talk. Let us also remember the two data points from earlier:

28  Finlay (2014: 31).

29 Aristotle, Nicomachean Ethics, 1.7; Foot (2000: 94).

Prudential Language and Context (I)  77 Diversity:  ‘Good for’ claims are different in some way that we must explain. Unity:  Whatever that difference is, we should explain it in a way that makes sense of the fact that it is not a contingent artefact of English that we use the same expression for these claims. End-­Relationalism explains these two data points. Unity is explained by ‘good for’ claims being claims about what promotes some relevant ­outcome. Diversity is, in turn, explained by the huge range of relevant outcomes. Given this, I think that (A) and (B) are correct on one interpretation, as long as we interpret ‘senses’ in terms of the variety of relevant outcomes O. But we should not think of these as different senses or differences in literal meaning across uses of ‘good for’ or that there is a special prudential sense of ‘good for’.30 Having outlined the merits of end-­relationalism as an account of ‘good for’ talk, let me compare the proposal with the most detailed competitor proposal in the extant literature: Richard Kraut’s developmentalism. I argue that the most significant problem for Kraut’s proposal is avoided by end-­relationalism, further adding to its plausibility.

5.  Kraut’s Developmentalism versus End-­Relationalism In his What is Good and Why, Richard Kraut develops a view about ‘good for’ talk which marks a contrast with end-­relationalism and the views that we saw at the beginning of the part.31 Kraut’s view is dis­tinct­ ive in holding that ‘good for’ talk is unitary in meaning, being concerned always with the promotion of flourishing.32 Here are some key parts of Kraut’s view (I provide labels for ease of reference):

30  Due to constraints of space I omit discussion of ‘better for’ and ‘best for’. For discussion see Finlay (2014). 31  Kraut (2007). 32  Actually, for reasons that will become clear below, it might be accurate to think of Kraut’s view as a version of Ziff ’s view but one which constrains ‘interests’ more narrowly than Ziff does, such that the relevant kind of ‘interest’ is always flourishing.

78  Dear Prudence [A] For an immense variety of living things—trees and shrubs, no less than every member of the animal kingdom—certain things are good and others bad.33 [B] S does not have to be a living creature for it to be the case that G is good for S. Dry air is bad for pianos. Sugar is bad for gas tanks. Sand is bad for watches. Whatever enhances the performance of an artifact or its ability to play its role is good for it; whatever damages it or detracts from its suitability to achieve its purpose is bad for it.34 [C] We also speak of what is good for activities: fair weather is good for flying, knives for cutting, wet snow for making snowballs. These will not call for separate treatment.35 [D] But the word ‘good’ does not vary in its meaning as it is applied to these many diverse subjects. We say, ‘That kind of barley is good for horses but not for human beings’. But we do not mean that it is good­in-­a-­horsey sense for horses and bad-­in-­a-­human-­sense for humans.36 [E] Living things are not the only things for which some things are good and other things bad. Some things are good for a car, other things bad. But truths about what is good for a car depend on truths about the ways in which they serve the good of human beings. They are designed to improve conditions for living beings. So we can say: everything that is good for S, whether S is living or not, either promotes or is part of flourishing.37

There is much that I agree with here. Claim [D] seems exactly right, and I have already endorsed something close to claims [A] and [B]. Trouble arises from the core of Kraut’s developmentalism, which is claim [E]. We see the core problem with Kraut’s proposal by seeing what it rules out as an acceptable use of ‘good for’ but which end-­relationalism permits. Take some arbitrary outcome: landing a watermelon on Mars. Various things are no help in promoting that outcome: bow-­ties, toddlers, and paper planes. But some things—computers, renewable fuel 33  Kraut (2007: 3). 36  Kraut (2007: 3).

34  Kraut (2007: 3). 37  Kraut (2007: 132).

35  Kraut (2007: 3, n2).

Prudential Language and Context (I)  79 sources, refrigerators, and the like—are highly conducive to this ­outcome. So, computers, unlike toddlers, are useful for promoting the outcome of landing a watermelon on Mars. Can we then say, from this, that computers are good for landing a watermelon on Mars? On end-relationalism, the answer is clearly yes. This seems like the right answer; computers are a productive means of generating this outcome. In stark contrast, on Kraut’s view, we must know much more about the situation. In particular, we must know that landing a watermelon on Mars promotes or is part of flourishing. But that seems too restrictive. Furthermore, there are true ‘good for’ claims even where we stipulate no positive connection to flourishing. Behrends gives the example of claims about what is good for a terror machine, a machine whose function is to maximize suffering and Finlay uses the example of a doomsday device which is designed to destroy the universe.38 In each case it seems clear that there are true ‘good for’ claims to be made about these things—such as that ‘oil is good for the terror/doomsday machine’— even though, by hypothesis, they do not promote flourishing.39 End­Relationalism correctly predicts that anything that effectively promotes some relevant outcome can be truly said to be good for producing that outcome (whatever its value). How then do end-­relationalism and Kraut’s proposal relate? Kraut’s proposal is that ‘good for’ is univocal and that the outcome relevant to a ‘good for’ claim is always flourishing (either of the agent or someone else). In terms of the data points that we noted earlier: Diversity: ‘Good for’ claims are different in some way that we must explain. Unity: Whatever that difference is, we should explain it in a way that makes sense of the fact that it is not a contingent artefact of English that we use the same expression for these claims.

38  Behrends (2011: 125), Finlay (2014: 31) 39  In case you are sceptical, think about betterness claims. Suppose the machine will break down if it gets above a certain temperature and it is approaching that temperature now. It seems clear that cool water is better for the terror machine than hot water.

80  Dear Prudence Kraut’s view scores highly with respect to Unity. If ‘good for’ really were as Kraut suggests, it would be clear why there is a common translation of ‘good for’s into other languages. In every case, the use of ‘good for’ would be to talk about the very same relation. Kraut’s proposal fails, however, with respect to diversity. In insisting that the outcome relevant to a ‘good for’ claim is always flourishing, Kraut’s view is unduly restrictive. End-­Relationalism, by contrast, treats claims about promoting outcomes connected to prudential value as subclass of ‘good for’ claims rather than definitive of the whole. We can see the advantage of this when we consider claims like ‘computers are good for landing a watermelon on Mars’ and ‘oil is good for the terror machine.’40 Pace Kraut, and exactly as predicted by end-­relationalism, we can understand and interpret this claim without having to take a stand on whether, for example, the terror machine promotes anyone’s well-­being. Overall, whilst Kraut’s proposal successfully captures one of the data points for ‘good for talk’—the unity of ‘good for’ talk—it does so in a way that is too restrictive. By treating all ‘good for’ claims as concerning promotion of prudential value it loses out heavily with respect to diversity. End-­Relationalism does well, and does better than developmentalism, by holding that the outcome relevant to a ‘good for’ claim is sometimes prudential value but that this is not part of the meaning of ‘good for’. Rather it is just that something’s well-­being is sometimes the relevant outcome. As Finlay (2014: 34) puts it: ‘Our hypothesis is therefore that “good” sentences are fundamentally end-­relational, and that patient- and interest-­relativity are merely special cases of this.’

6. Conclusion In this part I have argued that Finlay’s end-­relationalism provides an illuminating account of ‘good for’ claims in general, one on which ‘good for’ claims are claims about what promotes some relevant outcome. The distinctively prudential ‘good for’ claims are not ones in which ‘good for’ 40  Of course a terror machine would run on oil!

Prudential Language and Context (I)  81 has some specific, prudential, meaning. Rather, they are simply those where the relevant outcome is, itself, the subject’s well-­being. My inquiry was guided by the following data points: Diversity: ‘Good for’ claims are different in some way that we must explain. Unity: Whatever that difference is, we should explain it in a way that makes sense of the fact that it is not a contingent artefact of English that we use the same expression for these claims. As I noted at the beginning, many philosophers respond to the range of ‘good for’ claims by holding: (A)  There are different senses or meanings of ‘good for’. (B)  Within this plurality is a specific sense employed within prudential thought and talk. End-­Relationalism is either a rejection of (A) and (B) or a precisification of them, depending on how strictly we interpret ‘senses’ and ‘­meanings’.41 I take no stand on which of these is the correct interpretation. Either way, I hope to have shown that end-­relationalism is a promising way to accommodate both the diversity and the unity of ‘good for’ talk.

Part B—Directive Prudential Claims 1.  Introduction The terms we have examined so far are used to make prudential claims about what would be good for some welfare subject. These are evaluative 41  There are many related questions that this chapter has not addressed. For example, my account of ‘good for’ talk assumes that such talk is often used to make claims about promotion of prudential value. Of course, nothing in this discussion so far has established that such talk is in good order, that we should not be error theorists about prudential value. For critical discussion see Moore (1903), Hurka (1987), Rosati (2009). I take up these questions in Chapters 1 and 6.

82  Dear Prudence prudential claims. At the beginning of the chapter I distinguished these from directive prudential terms such as ‘reason’, ‘ought’, ‘must’, ‘needs’. We see these terms deployed in the following sentences: (25) ‘John ought to take more exercise.’ (26) ‘Jane needs a transplant.’ (27) ‘Gloria must have water.’ (28) ‘Hillary should take that job.’ However, each of these sentences can be used with respect to different kinds of modality. For example, they can be used to make moral claims. My interest, however, is in the use of modal terms (those highlighted in the sentences above) to make prudential claims. My plan is to leave aside ‘ought’ and ‘should’ and restrict my attention to ‘needs’ and ‘must’. This is because the question of how ‘should’ and ‘ought’ relate to ‘must’ is a vexed one, and getting into that debate would take us too far afield. My assumption is that, given the plausibility of the claim that the best theory of ‘needs’ talk applies to its appearance within prudential discourse, whatever story works for distinguishing ‘should’ and ‘ought’ from ‘must’ in general will carry over to the use of these terms in prudential discourse.

2.  ‘Needs’ Discourse Claims about needs are a ubiquitous feature of everyday practical discourse. It is therefore unsurprising that needs have long been a topic of interest in moral philosophy, applied ethics, and political philosophy. Philosophers have devoted much time and energy to developing t­ heories of the nature of human needs and the like.42 Philosophical discussion of needs typically begins with a pair such as: (29)  ‘Hillary needs water.’ (30)  ‘Hillary needs 2,383 delegates to win the nomination.’ 42  Needs are also a central issue in other disciplines, including development studies, psych­ ology, economics (particularly issues of healthcare resource allocation).

Prudential Language and Context (I)  83 Such a pair is used to support the idea that there is an important ­difference between (29) and (30) and between claims of their respective types more generally. Very roughly, the standard story is that claims like (29) are normative, or important, whilst claims like (30) are in some way secondary or less significant. Different terms are used to label this purported distinction between claims like (29) and (30). These include: ‘Absolute vs Instrumental needs’; ‘Fundamental vs Instrumental needs’; ‘Non-­Contingent vs Contingent needs’.43 Philosophers working on needs are typically committed to the following claims: (1)  There are different kinds of needs. (2)  Within the different kinds of needs is a privileged class of needs that is especially normatively significant. When it comes to the question of what appears on each side of the distinction (i.e., what we are distinguishing between), different philo­ sophers provide different answers. In some places, the claim is that there are multiple types of needs: Contingent needs are requirements for contingent ends, which the needing being might or might not have [. . .] Non-­contingent needs, by contrast, are necessary conditions for non-­contingent aims that the needing being could not but have (like life).44 A definition of ‘need’ requires a distinction [. . .] between fundamental and instrumental needs[.]45

Sometimes the claim is that there are multiple concepts of need. A conceptual distinction between instrumental needs and absolute needs can be drawn[.]46 43  Absolute vs Instrumental [McLeod (2014), Wiggins & Dermen (1987)], Fundamental vs Instrumental [Thomson (2005)], Non-­Contingent vs Contingent [Reader & Brock (2004), Schuppert (2013)]. These categorizations are contentious in many ways. They are also far from clearly equivalent. However, that is not important for my purposes. 44  Reader & Brock (2004: 252). 45  Thomson (2005: 175). 46  McLeod (2014: 293).

84  Dear Prudence Sometimes the claim is that there are multiple senses of ‘need’. [A]lthough there is an instrumental sense of ‘need’ where we can ask for some purpose to be specified in a non-­elliptical version of the ‘needs’ claim and there are no limits on what this purpose is (except the limits of what can be of any conceivable concern to anyone), there is another sense of ‘need’ by which the purpose is already fixed, and fixed in virtue of the meaning of the word . . . We have then to assign at least two senses to ‘need’ if we are to assign the right significance to the sorts of thing people use the word to say and to understand the special argumentative force of needs claims.47

With these ideas on the table, philosophers of need then try to find the grounds of that distinction, or precisely what absolute/fundamental/ non-­contingent/basic needs have in common that makes them special, in contrast with the merely instrumental needs. This way of thinking about ‘needs’ claims mirrors the standard view of ‘good for’ talk I examined and rejected above. There is the suggestion of there being multiple senses (etc) and a particular significant sense. As I argued above, it is better not to postulate ambiguity or multiple concepts to account for diverse uses of the same word (at least other things equal). Such a move seems under-­motivated by the relevant data here, namely claims like (29) and (30). The difference, such as there is, between (29) and (30) does not look like a conceptual difference, there seems too much commonality between the two claims. It equally seems implausible that ‘needs’ is used in two different senses in (29) and (30). The similarity between (29) and (30) seems utterly different from the ambiguity exhibited by, for example, ‘That’s Rahul’s date[romantic] for the party’ and ‘That’s Rahul’s date[temporal] for moving in.’48 47  Wiggins (1987: 9). 48  I had originally used an example of the supposedly ambiguous bank [financial] vs bank [river] but this isn’t clearly ambiguity, as opposed to polysemy. For discussion see Viebahn & Vetter (2016), citing Elbourne (2011), who write: ‘the meanings of “bank” (financial institution/riverbank) have a common etymological origin in a Proto-­Germanic expression for a shelf or a bench.’

Prudential Language and Context (I)  85 This seeming lack of a conceptual difference is reinforced by the claims made by needs theorists in trying to distinguish the privileged class of needs from the merely instrumental ones. The common idea is that instrumental needs claims are claims about what is necessary for the obtaining of some goal whereas the privileged class of needs claims are claims about what is necessary for (e.g.) the avoidance of harm. But this does not look like a deep difference and certainly not a conceptual difference. The only difference is in whether the relevant state of affairs is (e.g.) the avoidance of harm or something else. But it is not at all clear that this yields a difference in how ‘needs’ functions in (29) and (30). A further worry about (1)—the claim that there are different kinds of needs—and the general context in which it is motivated is that theorists working on needs have tended to focus on a restricted range of claims that are made by using ‘needs’. When we look at the range of sentences that deploy ‘needs’ we find many claims about need do not obviously fit into the categorizations that are offered by needs theorists. Take these claims: (31)  ‘Triangles need three sides.’ (32)  ‘America needs campaign finance reform.’ (33)  ‘Blatter needs to be punished.’ These do not seem obviously to fit into the categorizations offered above. Yet despite their difference from claims like (29) and (30) it seems implausible to think that there is a semantic difference in ‘needs’ across the examples. What we need is some way of recognizing that claims like (29) and (30), and other claims involving ‘needs’, can be different but without treating this as a matter of multiple senses of ‘needs’ or multiple concepts. We need to recognize the diversity within needs-­discourse whilst preserving what is unified across it. Thankfully, there is an alternative, more unified, way of understanding needs claims like (29) and (30). We reach this by taking some ideas from recent work on ‘must’ and then applying them to needs claims.

86  Dear Prudence

3. From ‘Must’ to ‘Needs’ Take the following set of sentences: (34)  ‘Morally speaking, John must tell the truth.’ (35)  ‘John must take more exercise.’ (36)  ‘John thinks that he is Queen Elizabeth so he must think that he lives in Buckingham Palace.’ (37)  ‘John must be here by 3pm given the quiet roads.’ (38)  ‘Drinking water must be clean.’ For a long time, the standard response to sets of sentences like these was to postulate multiple senses of ‘ought’ and ‘must’, or multiple concepts, to account for the diversity exhibited by (34)–(38).49 The work of Angelika Kratzer provides a way to resist that kind of view.50 Her pi­on­ eer­ing work on modals shows how we can preserve a uniform semantics for modals like ‘must’ (and ‘ought’) that is compatible with the differences in how they function in sentences (34)–(38). (From here, I leave aside ‘ought’ and focus on ‘must’.51) ‘Must’ is used to make strong necessity claims. Simplifying a little, the Kratzerian analysis of such strong necessity claims is that they are claims about what is true in all of the possibilities consistent with the relevant law or standard (where this might be explicitly mentioned within the sentence itself, as in sentence (34), or contextually salient).52 To see how

49  This issue has been discussed much more with respect to ‘ought’ than ‘must’. For discussion of ‘ought’ see Ewing (1953), Harman (1975), and Parfit (2011). For prescient discussion see especially the exchange between Xenakis (1957) and Glassen (1960). One issue I leave aside here is the debate between those who think that there is ambiguity in ‘ought’ between ‘agential’ and ‘evaluative’ ought claims. Discussing that would take me too far afield here. See e.g., Wedgwood (2006), Schroeder (2011), and Chrisman (2012). 50  Kratzer (1977). Also Wedgwood (2006), Dowell (2011), Finlay (2014), and Chrisman (2015). 51  I focus on ‘must’ because ‘needs’ is like ‘must’ in being a strong necessity modal and I want to sidestep (i) whether ‘ought’ is a strong necessity modal and (ii) how to understand weak necessity modals within the Kratzerian framework. 52  One detail I omit, purely for simplicity, is the restriction on the set of possibilities (the ‘modal base’). A more accurate statement would thus be that, on Kratzer’s view, strong necessity claims are claims about what is true in all of the relevant possibilities consistent with the relevant standard. Another detail I omit, again for simplicity, is that Kratzer’s analysis involves the idea of partial ordering.

Prudential Language and Context (I)  87 this works, let us start with two example sentences. On the Kratzerian view, ‘Morally speaking, John must tell the truth’ expresses the prop­os­ ition that in all possible worlds in which the moral standards are adhered to, John tells the truth.53 By contrast, ‘What goes up must come down’ expresses the proposition that in all worlds consistent with the laws of nature, objects that go up come down. On this Kratzerian view, ‘must’ has the same meaning in (34)–(38). In each case it functions to generate a proposition about what is required by some law or standard (hereafter I will simply use ‘standard’, for brevity). The key difference between the sentences is a difference in the relevant standard. Returning to the set of sentences above, sentence (34) is a claim about what is required by the moral standard. Conversely (35) is a claim about what is required by the prudential standard. By contrast (36) uses the standard of rationality whilst (37) uses an evidential stand­ ard and (38) uses a general evaluative standard. A nice feature of this view is the way that it combines unity with flexi­ bil­ity. Must claims have a unified logical structure; we do not need to postulate different senses of ‘must’ to account for (34)–(38). At the same time, it is flexible; one can plug in any arbitrary standard whatsoever and generate meaningful must claims and without endorsing the relevant standard. For example, ‘To uphold Bullingdon club tradition, he must [something horrible].’ My suggestion is that we take the Kratzerian story for ‘must’ and apply it to ‘needs’ claims.54 This is justified by the fact that ‘needs’, like ‘must’, is a strong necessity modal.55 We see this from the similarity between these needs claims and the equivalent must claims: • ‘You should give it back. In fact, you need to do so.’ • ‘You should give it back. In fact, you must do so.’ • ‘He should give it back but you need to do so.’ 53  I follow Kratzer’s presentation in construing possibilities in terms of possible worlds. But the general approach does not depend on a commitment to possible worlds, only possibilities. 54  Philosophers working on modal language have seemed to neglect ‘needs’ discourse. One exception is Finlay (2016: 196) who briefly mentions ‘needs to’ as a normative verb. 55  I am here talking of ‘needs’ as a verb, rather than as a noun. It is interesting that in English we have the noun ‘needs’ but lack one for ‘must’, ‘ought’, ‘should’ (i.e., #‘his musts/oughts/ shoulds are as follows. . .’).

88  Dear Prudence • ‘He should give it back but you must do so.’ • ‘I need to leave now.’ • ‘I must leave now.’ There is also the general similarity between the needs claims we started with and the equivalent must claims: (29)    ‘Hillary needs water.’ (30)    ‘Hillary needs 2,383 delegates to win the nomination.’ (29*)  ‘Hillary must have water.’ (30*)  ‘Hillary must have 2,383 delegates to win the nomination.’ Finally, there is the problematic nature of: # ‘Hillary must have water but she doesn’t need it.’ # ‘Hillary must have 2,383 delegates to win the nomination but she doesn’t need them.’56 From this I conclude that ‘needs’, like ‘must’, is used to make strong necessity claims, claims about what is required by a body of laws or a standard.57 Crucially, like ‘must’, ‘needs’ is used to make claims about different kinds of necessity. Take the following set of needs claims: (29) (30) (31) (32) (33) (34)

‘Hillary needs water.’ ‘Hillary needs 2,383 delegates to win the nomination.’ ‘Triangles need three sides.’ ‘America needs campaign finance reform.’ ‘Blatter needs to be punished.’ ‘Blatter needs a corkscrew to open that wine.’

56 One can get a reading of sentences like these which are not problematic. But that is only if the sentence makes claims about two different standards or use different flavours of modality. For example, ‘Given that she has a well, Hillary must have water but she doesn’t need it.’ 57  There is also the fact that ‘need’ is used for the negation of ‘must’ claims where ‘mustn’t’ would be incorrect. For example: ‘Must I return the book tonight?’; ‘No you needn’t.’ (c.f. ‘No you must not return the book tonight.’).

Prudential Language and Context (I)  89 Claim (30) is a claim about what is necessary in order to win the ­nom­in­ation within the election rules. By contrast, (31) is a claim about what is logically necessary. Claim (32) is a claim about what is necessary according to an evaluative standard. Claim (33) is a claim about what is morally necessary.58 Claim (34) is a claim about what is nomologically ­necessary. (I bracket for the moment the question of how to interpret (29).) Given this parallel between ‘must’ and ‘needs’, we should take the orthodox treatment of ‘must’ claims and apply it to ‘needs’ claims.59 Remember that the Kratzerian analysis of strong necessity claims is that they are claims about what is true in all of the possible worlds consistent with some relevant standard. Applying this to needs claims we get the following picture: ‘Hillary needs 2,383 delegates to win the nomination’ conveys the proposition that in all worlds where the election rules are adhered to and in which Hillary wins the nomination, she has 2,383 dele­gates. By contrast, ‘Blatter needs to be punished’ conveys the prop­ os­ition that Blatter is punished in all of the worlds consistent with the moral standard. Let me briefly restate the package of claims I have argued for in this section, which I will refer to as ‘needs claims as modal claims’ Needs claims as modal claims (NCAMC) - Needs claims are strong necessity claims, claims about what is true in all possibilities where some standard is met or where some stand­ard is met and some state of affairs obtains.60 - Nothing in the meaning of ‘needs’ places a restriction on either the relevant standard or the relevant state of affairs. 58  Or legally necessary, depending on the context. 59 One difference between ‘must’ and ‘needs’ is that ‘must’ is always an auxiliary verb whereas ‘needs’ can either be a regular verb or an auxiliary verb. This difference does not matter though. Even when ‘needs’ is a regular verb, as in (34) ‘Blatter needs a corkscrew to open that wine’, it is still used to make claims about what is necessary (it is just that it is used to make claims about what is necessary, relative to some kind of necessity, for some state of affairs to obtain). Thus, ‘needs’ is like ‘must’ in being used to make strong necessity claims. 60  This feature of the view has the consequence that there are lots of needs claims which are true but weird to assert. These will be the set of needs claims about anything that necessarily correlates with a true needs claim. The same applies to needs claims about necessary truths. For example, ‘Hillary needs 2+2=4’ will come out as true.

90  Dear Prudence - The relevant standard can either be explicitly mentioned in the needs claim or recoverable from context. Let me now connect NCAMC to prudential discourse.

4. ‘Must’ and ‘Needs’ in Prudential Discourse From NCAMC we can see how prudential directive claims are made using ‘need’ and ‘must’. We make such claims when the standard relevant to the ‘need’ or ‘must’ claim is a prudential standard. For example, if we are talking about Hillary’s welfare and I say: (29)  ‘Hillary needs water.’ I convey the proposition that in all worlds that meet the prudential standard, Hillary has water. Similarly, to say that people need friends is to claim that in all of the possibilities that meet the prudential standard, people have friends. There are many questions that this view rightly leaves open. One such question is whether the relevant standard for a strong prudential necessity claim can vary between contexts or if there is always exactly one standard that is relevant to prudential necessity claims. For example, are prudential ‘needs’ claims always about maintaining some specific min­ imal level of well-­being? Or are they about remaining above some context-­sensitive level of well-­being? I will, in Chapter  4, examine whether prudential language is more pluralistic and context-­sensitive, beyond what we have seen so far.

5.  Conclusion In this chapter, I examined instances of two types of prudential normative terms—evaluative terms and directive terms. In particular, I focused on ‘good for’ (and ‘better for’, ‘best for’) and ‘needs’ (and ‘must’). In each case, we saw how a plausible general account of such terms carried over to their use in prudential discourse.

Prudential Language and Context (I)  91 For ‘good for’, I defended end-­relationalism about ‘good for’ claims in general, one where such claims are claims about what promotes some relevant outcome. The distinctively prudential ‘good for’ claims were those where the relevant outcome is, itself, the subject’s well-­being. For ‘needs’ and ‘must’, I applied the Krazterian contextualist account where these are strong necessity modals, claims about what is true in all of a set of possibilities that meet a particular standard. The distinctively prudential ‘needs’ and ‘must’ claims are those where the relevant stand­ ard is the prudential standard. The proposals defended in the chapter had a similar set of virtues. Generalizing our data points from earlier: Variety:  ‘Good for’ claims and ‘needs’ claims are different in some way that we must explain. Unity:  Whatever that difference is, we should explain it in a way that makes sense of the fact that it is not a contingent artefact of English that we use the same expression for these sets of claims. we can see how end-­relationalism and NCAMC capture these data points. In each case, we have a general recipe coupled with an ex­plan­ ation of the diversity between ‘good for’ claims, in terms of the relevant outcome, and ‘needs’ claims, in terms of the relevant standard. We get the specifically prudential forms of discourse when we assume that the outcome relevant to a ‘good for’ claims is the agent’s well-­being and when we assume that the standard relevant to assessing a ‘needs’ or ‘must’ claim is a prudential standard. There are many questions that this chapter has not addressed. For example, my accounts of ‘good for’ and ‘needs’ assume that there is such a thing as prudential value, and nothing I have argued here establishes that. Some other questions that I have not addressed can be more precisely stated using the account of evaluative and prudential language offered here. For example, at least some have wondered whether there is even greater contextual flexibility to prudential discourse than I have allowed here. They wonder whether there is a single concept (or set of concepts) for prudential value or if there is in fact a plurality of some sort. I address these issues in the next chapter.

4 Prudential Language and Context (II) Contextualism and Pluralism

1. Introduction In the previous chapter, I provided accounts of two parts of prudential language: evaluative claims (‘good for’, ‘better for’, ‘best for’) and directive claims (‘needs’, ‘must’). These accounts relied on the idea that claims made using ‘good for’ and ‘needs’ are, respectively, claims about what promotes a subject’s well-­being or enables them to meet some prudential standard. Therein I allowed that prudential ‘good for’ and ‘needs’ talk are context-­sensitive.1 In this chapter, I consider whether prudential discourse is contextsensitive in a further, more radical, way. I do this by considering two recent cases for strong forms of context-­sensitivity and pluralism, cases made by Anna Alexandrova and Steve Campbell. Alexandrova gives a general argument for the view that the semantic content of terms like ‘well-­being’ and ‘doing well’ varies across different contexts. In a similar vein, Campbell proposes that there are plural prudential concepts at play in prudential discourse (and in philosophical reflection upon such discourse) despite the surface appearance, and general assumption, of uniformity. Here is how I proceed. I will first examine Alexandrova’s con­text­ual­ ism. I will argue that Alexandrova draws our attention to a further way in which prudential discourse is sensitive to context (when it comes to 1  This was for three reasons: (1) Some (many) ‘good for’ and ‘needs’ claims aren’t prudential at all: ‘acid is good for dissolving metals’, ‘a triangle needs three sides’. (2) Even ‘good for’ talk which is broadly prudential is different in the case of welfare subjects and e.g. plants re­spect­ ive­ly. (3) Even ‘needs’ claims which are prudential are context-­sensitive because the relevant threshold for someone’s needing something is plausibly context-­sensitive. Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0005

Prudential Language and Context (Ii)  93 judgements and claims about someone ‘doing well’). However, I will reject the deeper, more radical, kind of contextualism that she posits, showing how all of the relevant data can be accommodated by a noncontextualist view. I then consider and reject Campbell’s conceptual pluralism.

2.  Alexandrova’s Contextualism Anna Alexandrova has recently outlined a strong form of contextualism about well-­being talk. Let me first describe the data that she introduces before outlining and assessing the range of options that she considers in response to it. First, Alexandrova points out that people make apparently prudential claims, and ask (e.g.) ‘how are you doing?’, in a rich variety of contexts. She gives three cases where someone asks someone, Masha, how she is doing: (1) a Good Samaritan when Masha has fallen on ice; (2) a good friend in confidence; (3) a social worker checking on Masha as a new parent. In each case, it is stipulated that the interlocutor asks Masha how she is doing. Alexandrova contends that: [W]e witness three judgements that are ostensibly about Masha’s well-­being. Nothing changes in her life, and yet in each case a different standard of well-­being is used. The Good Samaritan has in mind neither flourishing nor positive mental states but rather the physical comfort of a heavily pregnant woman walking on ice. So long as Masha is not terribly in pain and can get home all right, the Good Samaritan is justified in judging her to be well. In the second case, the caring friend’s concern is a richer notion of well-­being—probably closest to what philosophers call ‘well-­being’. When she asks Masha how she is doing, she has in mind whether Masha is fulfilling her hopes and whether she is depressed. So the friend justifiably concludes that Masha is not doing well. Finally, the social worker is employing yet another notion of well-­being, most akin to quality of life. To use Thomas Piketty’s term, Masha is a member of the patrimonial middle class, which makes it likely that she will not fall through the cracks

94  Dear Prudence when she hits a vulnerable stage in life. She also has lots of people and resources to count on. Those two things are enough for well-­being as far as the social worker is concerned. [T]he threshold that separates well-­being from ill-­being—that is, how much of a given good Masha must have in order to qualify as doing well—and the factors that count for well-­being appear to vary across our three cases. I call these threshold and constitutive dependence, respectively. How should philosophy accommodate them?2

It is important to distinguish the data that Alexandrova identifies from the best explanation of the data. In order not to prejudge the latter question, I will characterize the data points in rough terms. The first one is: Data point 1:  Across different contexts, the uses of (i) ‘doing well’ and (ii) ‘are you OK?’ seem importantly different. A second datum that Alexandrova introduces is the plurality of ways in which ‘well-­being’ is used across different academic disciplines: This problem is magnified when we turn our attention to the scientific, rather than the everyday, context. Here, we do not even need to assume that ‘How are you doing?’ is a question about well-­being. Researchers all across the social and medical sciences use the term ‘well-­being’ freely and abundantly . . . It is a term used to refer to a minimal quality of life in development economics; to a health-­related quality of life in medicine; to a child’s access to decent schooling, healthcare, and parental love in disciplines that study children; to mental health in psychiatry and clinical psychology; and so on and so forth.3

Let me recast this as follows: Data point 2:  The uses of ‘well-­being’ across different disciplines seem importantly different.

2  Alexandrova (2017: 8).

3  Alexandrova (2017: 9).

Prudential Language and Context (Ii)  95 To make things maximally clear up front, my strategy in the subsequent discussion is to agree with Alexandrova about the data points (as I have characterized them) but (mostly) disagree about the best explanation of them. Alexandrova describes three possible views that one could take in  response to these observations: Circumscriptionism, Differential Realization, and—her preferred view—Contextualism: Circumscriptionism The first possibility is just to deny the significance of the diversity in question. One could claim that this diversity, to the extent that it exists, is a mistake or an instance of linguistic carelessness on the part of those who use ‘well-­being’ outside its proper context. Well-­being proper is that general, all-­things-­considered evaluation that philosophers have been concerned about.4 Differential Realisation Another possibility is to accommodate the diversity that the Circumscriptionist rejects. Perhaps there is a stable content for well-­being expressions (an assumption shared with the Circumscriptionist), but different states realise well-­being in different circumstances [. . .] For instance, well-­being might consist purely in one’s emotional balance when ascribed to a depression sufferer, or in one’s access to basic medicine and education in an environment of deep poverty, or in realisation of one’s dreams and ideals when we evaluate someone’s life as a whole. On this view, the semantic content of the term ‘well-­being’ does not change with each change in the environment. Only the truth-­makers of the state ‘is doing well’ (or ‘is not doing well’) change with context.5 Contextualism [C]ontextualism about well-­being would maintain that the semantic content of sentences in which ‘well-­ being’ and its cognates occur depends, at least in part, on the context in which it is uttered. A developmental economist might just mean something different by ‘well-­being’ than does a clinical psychologist. On this view it is impossible to speak of well-­being simpliciter. Rather, the content of well-­being assertions 4  Alexandrova (2017: 5).

5  Alexandrova (2017: 5).

96  Dear Prudence needs to be indexed to specific circumstances (doctor’s visit, poverty relief on country-­wide scale, heart-­to-­heart conversation with a friend, etc.). Since these circumstances will inevitably differ from situation to situation, so will the semantic content of ‘well-­being’. In one situation it will connote a concern of a doctor for their patient, in another of a social worker for his clients, or of a therapist for her depressed patient, and so on and so forth. The context of an all-­things-­considered evalu­ ation privileged by philosophers is just that: one of the many contexts in which well-­being is in question.6 Alexandrova contends that contextualism is the best explanation of the relevant data. I think that this is a mistake. Before moving on to explain why, let me first present the three alternatives in a slightly different way, to make the differences between them maximally clear: Circumscriptionism:  ‘well-­being’ (like ‘doing well’) only ever refers to exactly one state type, regardless of context. Differential Realization:  (1) ‘well-­being’ (like ‘doing well’) only ever refers to exactly one state type, regardless of context. (2) different properties ground that property across different contexts. The following schematic example helps to demonstrate differential realization: In context A, ‘S is doing well’ ascribes property P, whose grounds are x, y, z. In context B, ‘S is doing well’ ascribes property P, whose grounds are a, b, c. The third alternative that Alexandrova introduces is contextualism: Contextualism:  (1) ‘well-­being’ (like ‘doing well’) refers to different state types across different contexts (2) different properties ground the relevant property across different contexts. 6  Alexandrova (2017: 5).

Prudential Language and Context (Ii)  97 The following schematic example helps to demonstrate contextualism: In context A, ‘S is doing well’ ascribes property P,  whose grounds are x, y, z. In context B, ‘S is doing well’ ascribes property Q, whose grounds are a, b, c.

3.  Assessing Alexandrovan Contextualism I intend to argue that Alexandrova’s contextualism, as outlined above, is not the best explanation of the data points that she presents. Let me state my specific theses up front to aid the reader. My first main claim is that the use of ‘doing well’ is context-­sensitive. Different levels of well-­being count as doing well in different contexts. My second main claim is that, pace Alexandrova, the use of ‘well-­being’ etc. in different academic contexts is best explained by an innocent form of circumscriptionism. Finally, pace Alexandrova, I hold that in everyday contexts we make judgements about different aspects of well-­being and how well someone is doing with respect to those aspects. I now move on to assessing Alexandrova’s contextualism and arguing that my view is superior. Let us start by considering data point 2. It is undeniable that people from diverse subjects label that which they study as ‘well-­being’. The view that this is best explained by (something like) circumscriptionism might seem to manifest dogmatism. However, we see that this is not the case once we realize the close connection between the distinct subjects of these studies. We can distinguish between at least the following. First, we have a person’s level of well-­being, either at a time or over a period of time (including a whole life). Secondly, we have the positive and negative constituents of well-­being (that which has non-­instrumental prudential value and disvalue). Thirdly, there are instrumental means to improving a person’s well-­being. Fourthly, there are necessary means or pre-­conditions to a person either having a level of well-­being or their level of well-­being

98  Dear Prudence improving. There are also reliable indicators of well-­being. Finally, there are different levels of well-­being that we might be interested in. One of these is the level of a person’s well-­being that qualifies as a good life or doing well. Another one is the level of a person’s well-­being that qualifies as a minimally good life. (There are, of course, further distinctions we could make.) These distinctions are subtle and, crucially, unimportant for many subjects beyond philosophy because the practical difference between them is minimal or non-­existent. For example, the difference between constituents of well-­being, means to improving well-­being, and necessary preconditions of having or improving a level of well-­being would (presumably) not be especially important to many outside of philosophy. After all, the general point holds across these categories that, in general, the more someone has of the relevant thing, the better off they are. I think circumscriptionism, minus the claims of carelessness, is true for the use of ‘well-­being’ across different subjects, such as the social and medical sciences. When people in different subjects use these terms they refer to distinct things (and usually things distinct from well-­being itself). These might be something from one, or more than one, of the categories identified above or from categories that require further similar distinctions. They may also simply function as stipulated technical terms, ones not intended to play the same role as the everyday term ‘well-­being’. Contrary to Alexandrova’s formulation of circumscriptionism, it is not that these other subjects manifest carelessness. It is rather that the theoretical and practical purposes served by these subjects make it appropriate for them to label different—but closely related—things as ‘well-­being’. Some subjects are interested in particular constituents of well-­being, or in particular ways of promoting well-­being, some are interested in particular levels of well-­being. Other subjects are interested in these questions but restricted to sub-­groups of the population, such as children. Thus, they are talking about a cluster of closely related things, all of which are the subject of study because they are connected to well-­being or aspects of it. It is thus no accident, and no grounds for

Prudential Language and Context (Ii)  99 criticism, that different subjects label these different things ‘well-­being’.7 This is a perfectly innocent kind of loose talk. Having discussed the second data point, let me now move on to data point 1. Alexandrova provides this useful distinction between two different kinds of contextual variation (where each is essential to her contextualism): [T]he threshold that separates well-­being from ill-­being—that is, how much of a given good Masha must have in order to qualify as doing well—and the factors that count for well-­being appear to vary across our three cases. I call these threshold and constitutive dependence[.]

According to Alexandrova, ‘doing well’ judgements manifest threshold dependence whereas ‘well-­ being’ judgements manifest constitutive dependence. It will be useful to examine these separately, partly because I want to grant threshold dependence for ‘doing well’ judgements (and the like) as a genuine phenomenon, but deny constitutive dependence, the more radical element of Alexandrova’s contextualism.

3.1  Threshold Dependence Alexandrova is right that whether someone counts as doing well depends upon context. This is analogous to the way that whether someone counts as tall depends on the context.8 In different contexts, there are different minimum heights that one must be to count as tall. In a similar fashion, different levels of well-­being will count as doing well in different contexts because the relevant qualifying level of well-­being will be different. For example, take the question of whether our friend Jules is doing well. In one context it is true to say that Jules is not 7  We find the same pattern with the use of ‘happiness’. Similar patterns are observable for ‘utility’, ‘welfare’, and the like. See, though, Wierzbicka (2004) on ‘happiness’ talk in particular and the difficulties it poses for cross-­cultural investigations. 8  On the question of whether ‘tall’ really is context-­sensitive, see e.g. Stanley (2002) and Borg (2012)

100  Dear Prudence doing well. Perhaps they are, by far, the worst off of our friends. But in another context it will be true to say that Jules is doing well because, of the people alive presently, they have one of the highest levels of well-­being (it is just that all of their friends are doing even better). There are thus two different judgements that can be made of Jules. They are doing well for someone alive right now but they are not doing well for someone in their friendship group. Here is another example to demonstrate the truth of threshold dependence. You are in an accident and severely injured. After a few days your injuries have begun treatment and you are recovering in bed. Your friend comes to see you a few days apart and, upon returning, remarks ‘wow, you’re doing really well.’ In this context clearly the threshold for doing well has been shifted down from the default. You are not doing well on the everyday standard for doing well. But you are doing well for someone who was recently in a bad accident. There will be countless other ways in which we can set a minimum level of well-­being such that someone counts as doing well or not doing well. This is achieved by implicit or explicit indexing of these judgements as in ‘doing well for an X’ where ‘X’ can be related to sex, age, occupation, the period of history the person lives, or any other way of categorizing. For example we might say that ‘Jules is doing well for an old man.’ The addition of ‘for an old man’ adjusts the level of well-­being that the subject must have in order to count as doing well. How does it do this? Presumably by provoking us to make our judgements sensitive to our antecedent expectations of someone in that group’s level of well-­being. For example, we judge that old men typically have lower levels of well-­being than the general population of humans and then assess Jules’ level of well-­being against this particular standard. Thus, I agree with Alexandrova that our judgements about whether someone is doing well, doing OK, or doing badly manifest context-­dependence. That is because in addition to a fully general assessment of someone’s level of well-­being we can, and do, commonly assess that level against a contextually determined threshold for doing well. Threshold dependence is thus a real phenomenon. But notice that it does not support Alexandrova’s con­text­ ual­ism, given that that view encompasses two claims:

Prudential Language and Context (Ii)  101 Contextualism:  (1) ‘well-­being’ (like ‘doing well’) refers to different state types across different contexts (2) different properties ground the relevant property across different contexts. We have seen that (1) is true, applied to ‘doing well’, by seeing that ‘doing well’ refers to different levels of well-­being across different contexts. But nothing so far has shown that (2) is true. The explanation of threshold dependence—that different levels of well-­being count as doing well across different contexts and that we can often index someone’s level of well-­being (against people of the same age or gender)—was entirely compatible with well-­being itself being context-­insensitive (like height). We thus need to examine (2) by examining constitutive dependence.

3.2  Constitutive Dependence Alexandrova argues that different goods are relevant to different assessments of well-­being in different contexts. Let us remember the three cases that Alexandrova gives to motivate this claim: The Good Samaritan [. . .] So long as Masha is not terribly in pain and can get home all right, the Good Samaritan is justified in judging her to be well. In the second case, the caring friend’s concern is a richer notion of well-­being—probably closest to what philosophers call ‘well-­being’. When she asks Masha how she is doing, she has in mind whether Masha is fulfilling her hopes and whether she is depressed. So the friend justifiably concludes that Masha is not doing well. Finally, the social worker is employing yet another notion of well-­being, most akin to quality of life. To use Thomas Piketty’s term, Masha is a member of the patrimonial middle class, which makes it likely that she will not fall through the cracks when she hits a vulnerable stage in life. She also has lots of people and resources to count on. Those two things are enough for well-­being as far as the social worker is concerned.9 9  Alexandrova (2017: 7).

102  Dear Prudence No single response adequately explains all of these data points. The last case, the social worker, seems one for which circumscriptionism is plaus­ible. As we saw above, a cluster of closely related things are called ‘well-­being’ across different subject matters, and it is plausible that the social worker’s job is to assess the extent to which one has certain resources that are (at least) instrumental means to well-­being. She is there to assess the extent to which people have these resources, on the presumption that they are reliable means to (or preconditions of) sufficient prudential value for parents and infants. Here is evidence for this interpretation. Suppose the social worker reports ‘Masha is doing well’ and a passing philosopher interjects and presses the social worker on whether she really knows that Masha is doing well. Presumably, if pressed like this, the social worker would not reply (e.g.) that these goods that (she has verified that) Masha has access to are, really, the only constituents of well-­being and so yes Masha is, ipso facto, doing well. After all, it is clear that one could have money and people close to you but not live well. Rather, it seems likely that the social worker would reply by saying that her job is to see that Masha has these things (on the assumption that they are relevant to, because re­li­ able sources of, well-­being) and that everything beyond this is outside her remit. Her ‘doing well’ report was just loose talk about reliable means to well-­being. The different contexts where well-­being reports are made will sometimes exhibit (perfectly reasonable) loose talk for which circumscriptionism is plausible. That explains what goes on in the social worker case. The other two cases, however, manifest a different kind of vari­ ation, one for which circumscription (alone) as a response is implausible but which, I will argue, do not tell in favour of contextualism (or differential realization). Instead, these cases show how we focus on different aspects of well-­being in different contexts, depending on what is salient. The Good Samaritan asks fallen Masha if she is OK. How should we interpret this?10 Plausibly, the Samaritan is asking something about

10 There’s some plausibility to circumscriptionism given that Masha’s inquiry might be about something related to, but distinct from well-­ being, such as whether Masha has broken something.

Prudential Language and Context (Ii)  103 Masha’s present level of well-­being, in light of the highly salient fact that Masha has just fallen over on ice. I suggest that the Good Samaritan is asking about one aspect of Masha’s well-­being, her hedonic state (balance of pleasure over pain).11 When the Samaritan judges, or reports, that Masha is doing OK they are not judging her overall level of well-­being. They are judging her to be doing sufficiently well in one respect, or along one dimension, of well-­being. We see evidence for this by imagining different ways that the conversation could go. Version 1 Samaritan:  ‘Are you OK?’ Masha: ‘I’m fine, thanks.’ Samaritan:  ‘I’m glad you’re OK.’ Samaritan (later, to third party): ‘I checked on her and she was OK.’ Third party:  ‘No she isn’t. I’ve seen her around looking sad, and she doesn’t seem to have any friends around here, and her partner just lost their job.’ Samaritan:  ‘Right, fine. But physically she is doing OK.’ Version 2 Samaritan:  ‘Are you OK?’ Masha: ‘I’m fine, thanks. But my life is going terribly. I don’t have any friends or money and my partner just lost their job.’ In version 1, Good Samaritan makes a judgement about an aspect of well-­being—Masha’s hedonic condition (and perhaps also her physical condition). Whilst, for brevity, they will express this by saying that she’s ‘doing OK’, their judgement is that she is doing OK in this respect. Evidence for this comes from what they will say when pressed by a third party—namely that, in spite of the surface appearance, they are not making a judgement about her overall well-­being, only one aspect of it.

11  Perhaps coupled with her physical state (on the assumption that this is a productive means to prudential value and disvalue or perhaps even a constituent).

104  Dear Prudence In version 2, Masha’s first sentence allows Samaritan to form a judgement about one aspect of her well-­being. Masha’s second sentence, however, changes the focus of the conversation by making relevant her well-­being as a whole forcing that to become the common ground of the conversation.12 Once Masha does this, Samaritan cannot, as in version 1, simply address one aspect of her well-­being and say (e.g.) ‘I’m glad you’re OK.’ Such an utterance would be bizarre and seem uncaring. Similarly, Samaritan will have to report the incident to third party differently from in version 1, something like ‘Physically she was OK, but she was doing very badly overall.’ My proposal is that Masha’s falling over, and this being known to both parties, makes salient one aspect or dimension of her well-­being—the hedonic aspect. Samaritan’s inquiries into how Masha is doing are then treated as inquiries into how she is doing in that respect. But this is not fixed. Either party can shift the conversation to a more general, or fully general, assessment of well-­being. Here is an analogy to support the point. Suppose an electrician is employed by the government to provide free wiring checks for those soon to be having children living with them. The electrician checks an apartment and reports to their supervisor ‘the apartment is safe.’ Clearly, though, there are sources of danger for a child beyond electrical faults, so the electrician has not determined that it is safe overall. Suppose that a passer-­by points this out. Presumably the electrician will concede that their talk was loose—they haven’t verified that the apartment is safe overall—but clarify that their job is to assess one aspect of overall safety and that everything else is beyond their remit and was excluded from what they were reporting. I have offered an account of what goes on in the cases of the social worker and the Good Samaritan. In each case, the context is one where they are focused on either one aspect of well-­being or some particular set of productive or constitutive means to well-­being. In each case, this is not a fixed feature of the conversation, it is easy for it to shift to a more and less general assessment of the relevant person’s well-­being. 12 On conversational common grounds, see especially Grice (1989), Lewis (1979), and Stalnaker (2002).

Prudential Language and Context (Ii)  105 How does the analysis I have given here compare with Alexandrova’s? Importantly, my analysis contradicts Alexandrova’s constitutive dependence claim and thus her contextualism generally: Contextualism:  (1) ‘well-­being’ refers to different state types across different contexts (2) different properties ground the relevant property across different contexts. According to Alexandrova, across these different contexts, ‘­well-­being’ refers to different states which are, in turn, grounded by different properties. My alternative proposal is that, across these different contexts, ­ people refer to different aspects of one common thing: ­well-­being. There is not a plurality of states ascribed by ‘well-­being’ with a plurality of grounds. Rather, there is contextually determined focus on different aspects of well-­being combined with a certain degree of loose talk. How can we tell the difference between these two views? Note, first, that on Alexandrova’s view there will be a lot of talking past each other. For example, suppose that the friend in the Masha story meets the Samaritan. In response to the question: ‘how is Masha doing?’ the friend says: ‘Masha is not doing OK’, whereas the Good Samaritan says, ‘Masha is doing OK.’ On Alexandrova’s view, there is no tension whatsoever between these two judgements or their expression in this fashion by the two interlocutors. After all, the Good Samaritan attributes property P, whose grounds are x, y, z whereas the friend denies the attribution of property Q, whose grounds are a, b, c. Given that P and Q are, according to Alexandrova, distinct properties,13 there is no disagreement here and nothing to clarify or reconcile. On my alternative, aspectualist, analysis, Friend and Samaritan are not simply talking past each other, as on Alexandrova’s view. Rather, they have one common subject matter—well-­being—and (at least one of 13  One question for a radical form of contextualism like Alexandrova’s is what makes these properties related, such that they are both well-­being judgements. Let me report a hunch: it will be difficult to answer this question in a plausible way without contextualism turning into aspectualism. The most promising route would be to defend a polysemy view of how we use ‘well-­being’. On polysemy in this kind of context see Fogal (2016) and more generally Viebahn & Vetter (2016).

106  Dear Prudence them is) engaged in acceptable loose talk in expressing their view. So rather than simply proceeding as if their claims are not in tension with each other—‘Yes. And Masha is not doing OK.’—they instead recognize that at least one of them is engaged in loose talk and then make more explicit their non-­ conflicting propositions (Samaritan asserts that Masha is doing OK physically speaking, whereas Friend asserts that Masha is not doing OK overall). Remember the first of the data points that Alexandrova introduced: Data point 1:  Across different contexts, the uses of (i) ‘doing well’ and (ii) ‘are you OK?’ seem importantly different. It seems plausible that there is something different between the judgements that Samaritan and friend make in the Masha cases. But Alexandrova’s view explains this such that there is too great a difference. It is too great a difference because, on her contextualist view, there is no disagreement at all between: Samaritan:  ‘Masha is doing OK.’ Friend: ‘Masha is not doing OK.’ On the alternative view that I have developed here, we preserve the sense that Friend cannot reply to Samaritan like this because we have at least an initial appearance of disagreement, even if it turns out that this is explained as loose talk (and that the judgements that the interlocutors hold do not strictly conflict). They do not conflict because they are about different aspects of one common thing. The aspectualist view I have described here thus does a better job of explaining data point 1—that judgements of whether someone is doing OK exhibit variation across different contexts. It points out that we make judgements about different, particular aspects of well-­being but that, for reasons of conversational efficiency, we often express these as if they were judgements of the whole. Unlike Alexandrova’s contextualism, this view explains both of the following points. First, why there is a need for Samaritan and Friend to clarify their utterances if made together.

Prudential Language and Context (Ii)  107 Secondly, the fact that there is an important difference between their judgements. By examining Alexandrova’s case for contextualism about well-­being, we have seen some of the ways in which prudential discourse is contextsensitive. These were as follows: (1) Threshold dependence:  In different contexts, different levels of well-­being count as doing well. (2) Aspectualism:  In different contexts (academic and everyday) we are interested in different aspects of well-­being or in different close relations to well-­being. Let me recapitulate what we have seen so far in this chapter. I have examined Anna Alexandrova’s case for radical contextualism about well-­being discourse. Alexandrova drew our attention to apparent differences in the way that terms such as ‘is doing well’, ‘well-­being’, and ‘doing OK’ are used across different contexts. She argued that this data supports a radical form of contextualism about prudential discourse. I offered an alternative account which consisted of four claims. First, there is (perfectly innocent) looseness in use of ‘well-­being’ across different academic subject matters. Secondly, Alexandrova is correct that there is threshold dependence with respect to expressions such as ‘doing well’, so different levels of well-­being count as doing well in different contexts. Thirdly, we sometimes assess someone’s level of well-­being rela­tive to groups to which they belong. Fourthly, in different contexts we pick out different aspects of well-­being and engage in loose talk when making claims about them. My alternative is thus a mild form of context-­sensitivity, one that retains the idea that well-­being is fundamentally context-­insensitive, even if it is a matter of context which aspects of well-­being we talk about and which levels of well-­being count as doing well. In the rest of this chapter I address another threat to the view that prudential value talk is conceptually unitary and context-­insensitive. This comes in the form of Steve Campbell’s claim that there are plural prudential value concepts.

108  Dear Prudence

4.  Campbell’s Conceptual Pluralism Campbell starts off by pointing out that work on well-­being depends on the assumption of a single subject matter: Hedonists, desire-­fulfillment theorists, perfectionists, and objective-­list theorists generally take themselves to be in genuine disagreement with each other over a common subject matter that is both coherent and significant. Likewise, analyses of well-­being are often presented as casting new light on the concept or property of well-­being.14

This is undeniable as a claim about the literature. Campbell, however, thinks that we should reject the assumption. I will first present and explain Campbell’s view before moving on to examine the justification he gives for it. Campbell’s view is as follows: Conceptual pluralism:  Despite the appearance of a univocal subject matter, prudential discourse involves distinct (sets of) concepts, meaning that at least some apparent instances of fundamental substantive prudential disagreements are, in fact, instances of talking past each other. There is a clear affinity between this and Alexandrova’s contextualism, but Campbell’s view is explicitly framed as a claim about distinct con­ cepts. Here is a reconstruction of the first part of Campbell’s argument for conceptual pluralism: (1)   Prudential discourse manifests mutually conflicting commitments.15 (2)  If prudential discourse manifests mutually conflicting commitments, there is no univocal subject matter. 14  Campbell (2016: 408). 15  Campbell’s discussion is officially about prudential theorizing, so officially he has in mind the context of philosophers discussing prudential value. I take the discussion in the first half of the chapter to provide a case against conceptual pluralism among non-­philosophers who engage in prudential discourse. For brevity, I will discuss Campbell’s view as if it is about prudential discourse in general, but my arguments focus on the more theoretical context that he is officially focused upon. Many thanks to Steve Campbell for discussion here.

Prudential Language and Context (Ii)  109 (I will momentarily give some examples of the mutually conflicting commitments referred to in (1).) Suppose for the moment that we accept (1) and (2). What could we conclude on their basis? There are at least two options. We might hold an incoherentist error theory about prudential discourse.16 On such a view, although within prudential discourse, we presume that there is a single, coherent, subject matter for w ­ ell-­being, we actually use the concept in fundamentally inconsistent ways, resulting in the discourse being incoherent. Alternatively, we could conclude that despite the appearance of a single set of concepts, and so one subject matter, there is a set of distinct concepts, so conceptual plur­al­ism is true. Campbell leaves aside the first, incoherentist error-­theoretic, possibility. For the sake of tractability, I will also assume that we have reason to reject incoherentism, such that anyone who accepts (1) & (2) could infer conceptual pluralism.17 Thus, Campbell’s argument can be interpreted as: (1) Prudential discourse manifests mutually conflicting commitments. (2) If prudential discourse manifests mutually conflicting commitments, then there is no univocal subject matter. (3) If (1) and (2), then either an incoherentist error theory is true or conceptual pluralism is true. (4) Incoherentist error theory is false. Therefore, (5) Conceptual pluralism is true. I now move on to assessing Campbell’s argument.

5.  Assessing Campbell’s Case Campbell is clearly correct that prudential thought and talk, both by  philosophers and non-­ philosophers, manifests inconsistent 16  I will discuss other error theories in Chapters 1 & 6. 17  The case I give against Campbell’s conceptual pluralism also tells against incoherentist prudential error theory.

110  Dear Prudence commitments. On one interpretation of premise (1), it is made true by the fact that we find first-­order prudential disagreements, such as on whether knowledge is beneficial for its own sake or only instrumentally. Given that we are not, for example, tempted by contextualism about ‘economic growth’ claims on the basis of disagreement within economic discourse, Campbell’s argument needs more than simple first-­order disagreement. Thus, for certain forms of disagreement, the truth of premise (1) is insufficient to justify premise (2). For Campbell’s argument to work we must therefore identify deep, fundamental, disagreement within prudential discourse. This is clear from the way he seeks to justify premise (1). He aims to identify highly general commitments within prudential discourse and to show that these are incompatible. Before outlining these putative conflicting commitments, let me outline the strategy for responding to them. I will argue that we do not find disagreements within prudential discourse of the sort that can justify premise (2) of Campbell’s argument. This will be for different reasons for different putative commitments; thus, my response is (of necessity) piecemeal. In some cases, a kind of defusing strategy is available. This is because the commitments turn out, on closer inspection, to be genuine but non-­conflicting. For other putative conflicts, I argue that the commitments are non-­fundamental and best explained in a way that does not favour conceptual pluralism. I will now examine the conflicting commitments within prudential discourse that Campbell puts forward in support of premise (1).

5.1  Putative Conflict (I): Relation to the Subject’s Attitudes Campbell claims that some of the ways in which we think about prudential value suggest that well-­being is closely connected to the subject’s attitudes whilst other ways suggest that well-­being is not closely related to the subject’s attitudes. As examples of the former commitment Campbell points out our tendency to think that self-­sacrifice must be felt and that reward and punishment are necessarily connected to feelings:

Prudential Language and Context (Ii)  111 A concept of well-­being that does not bear any essential connection to the attitudes of the person will be ill suited to characterize the sort of ‘benefits’ and ‘costs’ that we seek to bestow on others when we reward, give gifts, do favors, punish, and take revenge.18

As examples of the corresponding conflicting commitment, one that downplays the connection between well-­being and subjects’ attitudes, Campbell points to our willingness to think of certain states or properties of a person as deserving of pity, irrespective of the person’s own attitude towards them. He gives the following example: [J]udgements of pitiability, enviability, and luckiness need not depend upon the subject’s own attitudes toward these things. I can intelligibly think someone who is descended from one of the world’s great novelists is enviable and lucky in that regard even if she herself is completely unmoved by this fact about her genealogy. I can believe that my neighbor is pitiable for lying to others even if she has no reservations or regrets about it, and even takes great pride in her skills of deception. This suggests that the concept of well-­being must be quite broad and must allow for the possibility that something’s being good or bad for a person bears no essential connection to his or her favorable or un­favor­able attitudes.19

My reply is that, properly understood, the plausible ideas here are not in conflict. I agree with Campbell that it is intelligible that someone is pitiable for lying to others or that it is enviable to be descended from one of the world’s great novelists. My reservations therefore concern the other putative requirement, that well-­being be essentially connected to the attitudes of the person as well as the claim that these two commitments are in conflict. Campbell’s description of the requirement suggests that it is a fundamental commitment of prudential discourse that for someone to be punished, or rewarded, or the subject of revenge, etc. that must be partly 18  Campbell (2016: 411).

19  Campbell (2016: 410).

112  Dear Prudence explained by their having the relevant kinds of attitude (presumably a pro-­attitude for a reward, a con-­attitude for a punishment). In reply, note, first, that it is not obvious that it is a fundamental commitment of prudential discourse that for someone to be punished, or rewarded, or the subject of revenge, etc. that must be partly explained by their having the relevant kinds of attitude. Suppose that Y takes their revenge upon X by simply preventing good things from happening to X, unbeknownst to them. For example, they prevent some superior job offer which would have improved X’s life from accruing. Similarly, suppose that a kindly benefactor rewards you by preventing something bad from happening to you. For example, your car gets an erroneous ticket but your benefactor, realizing that it will be expensive for you to prove this, simply pays it for you (thus you never learn, or suspect, that you were to experience this setback). These are plausible cases of revenge and reward. If so, then it is not necessary for something to be revenge or reward that it brings about actual pro or con attitudes. Furthermore, we can explain away such an essential connection by explaining why it might seem like there is such a connection even where there is not. To do so, we need only to note that it will often be the case that punishments, favours, revenge, and reward will trigger and be partly constituted by pro and con attitudes. And that holds for two ­reasons. First, that many of the good and bad prudential states are themselves attitudinal at least in part (pain and pleasure being plausible examples). Second, that in typical cases we know that we are being punished etc. and feel bad about that. But this is to stop far short of postulating the kind of essential connection Campbell suggests is a commitment of prudential discourse. A natural reply here, on Campbell’s behalf, is to appeal to either dispositional pro-­attitudes and con-­attitudes or hypothetical attitudes. That is, Campbell might argue that my cases work against only a crude understanding of this essential connection, namely one between punishment (etc.) and actual and occurrent attitudes but that they do not show that there is no connection between punishment (etc.) and dispositional attitudes. This leads to my second reply. Suppose, for the sake of argument, that there is an essential connection between punishment (etc.) and dispositional (or hypothetical)

Prudential Language and Context (Ii)  113 con and pro attitudes. How much does this show with respect to the concept of prudential value? Plausibly, not much. After all, there is no incompatibility between holding that some, but not all, aspects of prudential value are essentially connected to the agent’s pro or con attitudes. And one might argue that even if revenge has such a connection, this does little to establish that prudential value generally relates to the agent’s attitudes in some deep way. One way of defending this ­manoeuvre is by arguing that we have a connection to the agent’s pro and con attitudes in the case of punishment and reward because of the nature of punishment and reward themselves (namely that it is in the essential nature of these things that they are related to the agent’s attitudes). On such a view, there is no deep conflict between (i) the idea that punishment and revenge have an essential relation to the agent’s attitudes (albeit hypothetical or dispositional ones) and (ii) the idea that prudential value does not have such a connection in general. Campbell’s claim that we are deeply committed to an essential connection between attitudes and well-­being, and that this is in tension with another commitment about well-­being, can thus be resisted in two ways. First, we can deny that there is such a connection. Second, we can dispute that the connection demonstrates a fundamental feature of the concept of prudential value. If either of these is correct, then Campbell has not identified a genuinely conflicting pair of commitments within prudential discourse.

5.2  Putative Conflict (II): Relation to Morality Campbell puts forward a second candidate inconsistency in prudential discourse—that although we sometimes have a morality-­including conception of well-­being (such as in the passage above, where it seems intelligible that someone merits pity on account of lying), we also sometimes have a morality-­excluding conception of well-­being. As he puts it: Still other aspects of the standard picture suggest a more restrictive concept of well-­being—in particular, one that screens off the possibility that acting morally, in and of itself, is good for us. The egoist is

114  Dear Prudence standardly defined as one whose sole ultimate aim is the promotion of his or her own well-­being. Yet, many philosophers have seen the egoist as a natural critic of morality and someone who must be convinced that being moral can, by some indirect route, serve his or her own best interests. They do not seriously entertain the possibility that the egoist might view being moral as an important component of the good life. What may underlie this tendency is an assumption that well-­being is morality-­excluding in the following sense: it is either impossible or deeply implausible that being moral is intrinsically good for us.20

From these observations, Campbell concludes: All of this appears to indicate that the standard picture of well-­being is a conflation of two or more concepts. A single concept could not possibly satisfy all of these demands. It cannot be the concept of something that is both independent from and dependent upon the attitudes of the subject. It cannot both include and exclude being moral as a possible component of well-­being.21

Pace Campbell, I think that there is a single concept of well-­being which both includes and excludes being moral as a possible component of well-­being, at least once we disambiguate. Let us distinguish something’s being a non-­instrumental prudential value—its being non-­instrumentally good for you—from its being in your best interests. With this distinction in hand, one could hold that acting morally (or, more strongly, exemplifying virtue) is good for the agent whilst also accepting that it is deeply implausible that being moral is always in our best interests. Invoking this distinction between what is good for someone and what is in their best interests is not ad hoc. The distinction between something’s being good for someone to some extent and its being best for them is one that ordinary prudential thought and historical work on well-­being has been slow to recognize. It is perfectly coherent to hold that being moral is not always in one’s best interests even if one thinks

20  Campbell (2016: 411).

21  Campbell (2016: 411).

Prudential Language and Context (Ii)  115 that some vices or some immoral behaviour are bad for the person who performs them and thus worthy of pity.22 Finally, note that the idea that it is not plausibly in one’s best interests to be moral can be resisted. Aristotelian theories in ethics aim to show that virtue and the good life are deeply compatible (if not identical). But we can imagine any number of other theories that would reconcile these commitments. Suppose one held that friendship is a major source of prudential value, if not the only one, but that friendship (or only the most beneficial kinds of friendship) is only possible between people who act morally. On such a view there would be a close connection between acting well and living well. My point here is not to defend this conception. My point is only to show how there are resources from within ordinary prudential thought to render these commitments consistent. Thus, pace Campbell, a single concept could satisfy all of these demands.23 So far I have shown how some of the putatively conflicting commitments offered by Campbell can be made consistent. I now move on to assessing a final, stronger candidate for such conflicting commitments that could be offered, before showing how it too can be undermined.

5.3  Putative Conflict (III): Experience versus Desire Satisfaction Two everyday prudential platitudes are that what you do not know cannot hurt you and that it is always good for you to get what you want. These claims exemplify two tendencies in prudential thought. The first is the tendency to think that prudential value is necessarily experiential, in the sense that everything that is non-­instrumentally good or bad for you is experienced (or, at least, affects your experience). This is what motivate some to defend an ‘experience requirement’ on prudential value. Roughly put, this is the requirement that for something to be a 22  Note the way that the distinction is elided at the beginning of Plato’s Republic. 23  There is also the plausibility of treating our pity for someone’s immoral behaviour as being directed towards the consequences of being known to be immoral: punishment, ostracization, and the like.

116  Dear Prudence non-­instrumental prudential value it must either be experienced by the relevant agent or make some difference to their experience. If there is such a requirement, then it is true that what you do not know (more precisely, what you do not experience) cannot hurt you (non-­instrumentally). The second tendency is to think that prudential value is necessarily a matter of getting what you want. It is clear that everyday prudential thought displays some commitment to the idea that getting what you want is good for you. This has led some philosophers to defend an antialienation constraint on prudential value (holding that having a pro attitude towards P is necessary for P being good for you) or, going further, to treat desire-­satisfaction as necessary and sufficient for prudential value.24 These two constraints, however, cannot both be satisfied. Here are cases that illustrate the clash between them:25 Deluded Delphine:  Delphine believes that she is married to a wonderful person and that they have a secure, loving relationship, all she ever wanted. However, and unbeknownst to her, a tragic case of mistaken identity means that she is being monitored by the police and is married to an undercover police officer. This person has deceived Delphine into loving them and does not care at all about her. All of her friends are also undercover officers who pretend to like her in order to monitor her. Her entire romantic and personal life is a sham.26 Hard to reach Hillary:  Hillary is a research scientist who desires three things. She wants Leicester City to win the premier league, she wants her research to be recognized by her peers, and she wants her selfpublished fiction to be well received. Luckily for her, all of these desires are satisfied! Sadly, though, Hillary knows none of this. She is on a solo long-­term research expedition to a remote location and cannot receive 24  On anti-­alienation constraints, see for example Dorsey (2017), Hall & Tiberius (2016), Hawkins (2014), Fletcher (2013), and Railton (2003). One of the most sophisticated proposed constraints—one which does not take prudential value to be constrained by the actual, present, attitudes of the person—comes from Rosati (1996). For discussion of her proposal, see especially Sarch (2011). 25  Originally used in: http://blogs.lse.ac.uk/theforum/knowing-­whats-­good-­for-­you/ 26  Case based on the UK Spycops case.

Prudential Language and Context (Ii)  117 updates from the outside world. She is confident that her desires are not satisfied—they were long shots, after all—and when she left for the base she had excellent evidence that her desires would never be fulfilled. She is very unhappy. In the first case, to the extent that one endorses the view that prudential value is necessarily experiential (and more specifically hedonic), one will hold that Delphine’s life is going well. After all, it is a pleasant sham and she never finds out (we can stipulate). By contrast, to the extent that one thinks that prudential value is a matter of getting what you want, one will judge that this life is going badly. In the other example, to the extent that well-­being is experiential, Hillary’s life is going badly. By contrast, to the extent that one thinks that prudential value is a matter of desire fulfilment, it is going well. This is, then, another place where the professed commitments of prudential discourse manifest an apparent inconsistency. Does this tell in favour of conceptual pluralism? Again, I think not. It is highly dubitable that prudential discourse manifests a deep commitment to the idea that getting what you want is, ipso facto, good for you. I thus want to grant that these ideas conflict but stand my ground against the idea that both commitments are really manifested within prudential discourse. Cases like Hillary function in the literature as a reductio of the view that desire-­satisfaction is, itself, of prudential value. For that reason, there are few philosophers who hold the view that desire-­satisfaction per se has prudential value.27 And the reasons for this lack of attraction to the view are obvious. It yields odd results in cases like Hillary, where the desire is fulfilled, but the agent has absolutely no idea that it is. Furthermore, we can explain why people have typically closely associated desire with well-­being, through the simple observations (analogous to those offered above) that people typically desire that their lives go well and they typically desire the things that are plausible for candidates for prudential value (pleasure, friendship etc.). Desire satisfaction is a useful proxy for what is good for someone, even if not a perfect 27  The view has a peculiar status in the philosophical literature on well-­being. It is one of the mainstays of the discussion but has few historical or contemporary defenders in its pure form.

118  Dear Prudence indicator (because of cases like Hillary). In this way we can explain away the idea that prudential discourse is deeply committed to the idea that desire satisfaction per se has prudential value. By doing this we elim­in­ate another candidate for a pair of mutually conflicting deep commitments within prudential discourse. Campbell’s argument concluded that conceptual pluralism is the best explanation of the disagreement we find when we examining prudential discourse. I have argued against the data he offered as evidence of conceptual pluralism. But I want to offer an independent argument for why conceptual pluralism is not the best explanation of prudential disagreement. It is easy to overstate both the extent of this and the extent to which it is disagreement that motivates pluralism. The sort of disagreement that would motivate pluralism would be one where there was systematic non-­overlap in the judgements that people made about which things were good and bad for people or which lives were going well or badly. But it seems clear that these kind of judgements are subject to wide agreement. Where we find disagreement is in the explanation of prudential value, on the question of why things are good or bad for people, or why someone’s life is going well or badly. But when it comes to these questions, a plausible explanation is, simply, that philosophical questions are difficult despite the fact that they can seem easy. Our disagreement about the explanatory questions of why fundamentally someone’s life is going well or going badly reflects the difficulty of philosophical questions, and the subtlety of the difference between different substantive views of well-­being, not the truth of conceptual pluralism about prudential value.

6.  Conclusion In this chapter, I have examined two views of prudential discourse as radically context-­sensitive. I argued that we do not have a reason to favour Alexandrova’s radical contextualism over an alternative aspectualist view, coupled with the concession that prudential discourse is context-­sensitive in benign ways. I then argued, contrary to Campbell,

Prudential Language and Context (Ii)  119 that the inconsistent commitments and disagreement that we see in prudential discourse do not provide good evidence for conceptual plur­al­ ism about prudential discourse. I do not pretend to have provided a master argument against prudential discourse contextualism. Rather, I hope to have shown that we do not yet have a good case for con­text­ual­ ism about prudential discourse, as well as providing both a licence for pessimism in this regard and an alternative to contextualism in the form of aspectualism.

5 Prudential Judgements and Motivation 1. Introduction One of my main theses in this work is that prudential discourse is normative, on a par with moral discourse. In an earlier chapter, I argued, programmatically, that the features of moral judgements which are typ­ ic­al­ly pointed to as explanations of the normativity of moral discourse are also to be found within prudential judgements. One such feature that I briefly discussed there was the motivational profile of prudential judgements. In this chapter, I explore in detail how prudential judge­ ments are related to motivation. I proceed by examining a number of possible theses concerning their interrelation, and grounds of these the­ ses. I argue for the following claim:1 Prudential Judgement Internalism (PJI):  At least one type of prudential judgement (judgements about what one prudentially ought to do, among current options) is necessarily connected to motivation in rational agents.2 Here is the plan for the discussion. I begin by arguing for PJI in section 2 before considering an objection to it in section 3. In sections 4 and 5 I consider the prospects for more ambitious, general forms of prudential judgement internalism than PJI. I do this by examining possible explanations of the truth of internalist theses including PJI. I argue that the two main ways of extending PJI are implausible and so we cannot sustain anything more ambitious than PJI. In section 6 I make small amendments to PJI and give its final statement before (section 7) discussing the relation 1  More precisely, I argue for a slight refinement of this thesis (whose extra details would be a distraction here). 2  This form of internalism was introduced by Smith (1994). For discussion of it see (among others): Dreier (1990), Mason (2008), Svavarsdóttir (1999). Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0006

Prudential Judgements and Motivation  121 between PJI and questions concerning the nature of prudential judgements and the purported anti-­alienation constraint on prudential value.

2.  Prudential Judgements and Motivation Take the following case: Deliberating Denise:  Denise has reached the end of her undergraduate studies and is deciding what to do with her future. She spends time thinking about the various options open to her and concludes that one of them—law school—is what she prudentially ought to do. Suppose we know this all to be true—Denise sincerely judges that, pru­ dentially, she ought to go to law school (and that this is an option open to her)—what do we expect? Putting this question in terms of what we ‘expect’ masks two related questions, so let me distinguish them. First, do we predict that she will take the option? Not necessarily, I take it. We are accustomed to the fact that people sometimes act against their own best interests even as, correctly, judged by themselves. Someone might eschew an option that (they think that) prudentially they ought to take in favour of the option they judge to be best for par­ ticular other people or for others more generally. We do not assume that Denise will necessarily take the option that she judges that prudentially she ought to take. Similarly, we do not expect her to take the option in the sense of thinking that she has to, in light of her judgement. We do not think that Denise is necessarily criticizable for failing to take the option of going to law school, even if she judges that, prudentially, she ought to take it. She might, for instance, be moved more by what would be best for her family or for humanity in general. Those, then, are two things that we do not expect, in light of Denise’s judgement. However, we do expect Denise to be at least somewhat mo­tiv­ated to take the option of law school. Given her judgement that going to law school is what she ought to do (prudentially speaking) she cannot be left completely motivationally indifferent. Thus we predict that she will be—at least somewhat—motivated to take the option of law

122  Dear Prudence school.3 And we make this prediction on the grounds that failing to be so motivated would be to manifest a failing.4 That is, there would be something problematic in instantiating the following combination: (i)   Judging that some accessible option is prudentially required. (ii)  Failing to be motivated at all to pursue that option. In support, note that if someone reports being in this combination of states we would not accept that report at face value; we would look for some alternate explanation. We might wonder whether they were being insincere or if they were self-­deceived as a way of showing that in fact they do not instantiate the combination (because they do not really think that the relevant option is prudentially required). However, even if we concluded that they were sincere, so we could not explain away their report of being in this combination, we would treat this as a situation in which something had gone awry, as something to be remedied.5 Someone reporting being in this combination of states would engender sympathy from those who care for them. Being wholly indifferent in the face of your own judgements of what you prudentially ought to do is to manifest a deficiency of some sort. So far, I have argued that there is a failing in not being motivated to pursue what you judge to be required, prudentially, to do. But what kind of failing is it? Instantiating (i) and (ii) is not plausibly a moral failing or not merely a moral failing. Someone left cold by what they judge that they prudentially ought to do does not seem to manifest only a moral deficiency (if any moral deficiency at all).6 The most plausible candidate for the deficiency that this combination exemplifies is that of a rational failing, that instantiating the combination 3  For the sake of readability I sometimes remove the ‘given Denise’s judgement’ qualifica­ tion, which should be read as in play throughout the discussion unless otherwise indicated. 4  Reminder: I will discuss possible grounds of these claims in section 4. 5  Note that one can think this even if one is glad that the person is not motivated by their judgement (perhaps it is disastrously mistaken), for it still impugns their capacities in ways which are problematic. And a concerned other need not seek to remedy the situation by (sim­ ply) getting them to change their motivational state. (Compare: you believe that the moon is made of cheese and believe that cheese is edible whilst doubting that the moon is edible). 6  For arguments that it would be a moral failing (though agnostic on whether it would be another kind of failing), see Ross (2002: 25–6).

Prudential Judgements and Motivation  123 is a form of irrationality.7 There is something deficient, rationally speaking, in an agent who makes judgements about what they prudentially ought to do, among their available options, without being at all motivated to take those options. We can support this claim by comparing the combination we have been examining so far with these paradigmatic instances of irrationality: (iii) Judging that some proposition is best supported by your evidence. (iv)  Failing to be motivated at all to believe that proposition.8 (v) Judging that some action is what you ought to do all-­things-­ considered. (vi)  Failing to be motivated at all to perform that action. It is uncontroversial, I assume, that an agent who manifests these combinations is deficient, rationally speaking.9 If someone reported being in one of these combinations of states we would be reluctant to take this at face value. And if we eventually accepted that they indeed manifested this combination we would treat this as a rational failing. It is useful to draw a contrast between the case of Denise’s com­bin­ ation of states and some partly analogous combinations that are not plausible instances of irrationality. For example: ( vii) Judging that some action is required by some legal code. (viii) Failing to be motivated at all to perform that action. (ix) Judging that some action will increase the number of emails that have ever been sent. (x) Failing to be motivated at all to perform that action. In these latter two cases, it is highly implausible that there is, ipso facto, a deficiency in an agent simultaneously in both of these states. One does not necessarily manifest irrationality simply by being indiffer­ ent to the fact that some action is required by some legal code or would 7  More precisely, a failure to be fully rational. 8  Assumption: you have time to consider the matter etc. 9  On this see, Aristotle’s Nicomachean Ethics, Davidson (2004), Holton (2009), Stroud (2003).

124  Dear Prudence increase the number of emails ever sent.10 If someone reported being in one of these combinations of states we would take this report at face value and not think that there was anything to remedy. My claims are, first, that the combination we started with—judging that something is prudentially required and being indifferent—is sig­ nifi­cant­ly different from these kinds of cases, and belongs with the com­ bin­ations described above and, second, that it does so because this combination is a rational failing.11 Thus reflection on cases like Denise’s lends support to the following claim: Prudential Judgement Internalism (PJI):  At least one type of prudential judgement (judgements about what one prudentially ought to do, among current options) is necessarily connected to motivation in rational agents. From consideration of cases like Denise’s, we thus have a case for PJI. I now want to proceed by, first, addressing grounds for rejecting PJI. With these safely overcome, I move on to the question of whether we can establish a more ambitious thesis, by examining which other prudential judgement types are candidates for being necessarily mo­tiv­ at­ing in some set of agents.

3.  Objections to PJI Remember the case of Denise, from above: Deliberating Denise:  Denise has reached the end of her undergraduate studies and is deciding what to do with her future. She spends time thinking about the various options open to her and concludes that one of them—law school—is what she prudentially ought to do. Given the description of the case, if PJI is true then either Denise is at least somewhat motivated to pursue law school or she is irrational in 10  For related discussion of the connection between legal judgments and motivation see Rosati (2019). 11  Reminder, talk of something being a rational failing should be understood as a failure to be fully rational.

Prudential Judgements and Motivation  125 virtue of not being so motivated. How might someone reject PJI? More precisely, how might someone argue that Denise would not manifest irrationality if she were not motivated, at all, to act in accordance with her judgement?

3.1  Objection 1: Pragmatic Explanations The first type of response to consider is one that claims that our judge­ ment that Denise would have to be motivated by her judgement on pain of irrationality is, in fact, an illicit product of the way that the case is set up. More precisely, the objector would claim that our expectation that Denise is motivated in the presence of her judgement of what she pru­ dentially ought to do stems from features of the case other than her making the relevant judgement. For this response to work, we must identify a plausible confounding factor, one that is illicitly generating our intuitions in cases like Denise. The most plausible candidate for such a confounding factor here is the fact that the case is described as one in which Denise deliberates about her future, that she is deciding what to do. Thus, the objector might sug­ gest, this is a deliberative context and that is why we expect her judge­ ment to be accompanied by motivation. But, so the objection continues, if the context were different, we would not expect her, on pain of ir­ration­al­ity, to be motivated when she makes the relevant judgement. Such a response is weak, however. Compare the original version of the case with this variant: Deliberating Denise:  Denise has reached the end of her undergradu­ ate studies and is deciding what to do with her future. She spends time thinking about the various options open to her and concludes that one of them—law school—is what she prudentially ought to do. Deliberating Denise (alternate):  Denise has reached the end of her undergraduate studies and is deciding what to do. She spends time thinking about the various options open to her and concludes that one of them—law school—entails moving to a different city.

126  Dear Prudence I presume it is uncontroversial that, in the alternate case, Denise could (rationally) make the judgement and not be motivated accordingly. One could rationally judge that some option entails moving to a different city without being motivated to do so. Thus someone’s instantiating this combination: (xi)  Judging that some accessible option entails moving to a ­different city. (xii)  Failing to be motivated at all to pursue that option. seems not to impugn their rationality whatsoever. The point of examining this alternate case is to point to another prac­ tically relevant judgement, one that Denise might plausibly make in the same context, where failing to be motivated to act is not plausibly a rational failing. If we agree that this is true in the alternate case then this undermines the objector’s hypothesis that our judgement in the original Denise case—that either Denise is motivated or she is irrational—is il­legit­im­ate­ly triggered by the judgement having been made within a deliberative context. To reinforce this reply, consider the corresponding contrast case, one where someone forms a judgement about what they prudentially ought to do outside a deliberative context. Suppose, for example, that Denise is not deliberating about her future, she is just watching TV. A friend, see­ ing that there is a serious, unexpected, threat to Denise, sends her a message: ‘you need to get out of your house and drive to the next town!’ Denise trusts this person and so forms the judgement that she needs to get out of her house. What do we expect? I take it that we predict that she will be motivated, at least somewhat, to get out of her house and drive to the next town. Further, I assume that we make this prediction on the basis of the fact that Denise would be criticizable if she were not mo­tiv­ ated, given her belief. Suppose the imagined objector were to reply: ‘no, no, once she has formed the belief, she is, ipso facto, in a deliberative context.’ That might be true. But this is not a dialectical move that the person making this objection can make, for fear of undermining their objection. Their objection relies upon the possibility of prudential ought judgements outside of deliberative contexts.

Prudential Judgements and Motivation  127 The objection we have been considering has two parts. First, that Denise is not irrational simply in virtue of instantiating this combination: (i)   Judging that some accessible option is prudentially required. (ii)  Failing to be motivated at all to pursue that option. and, second, that the intuition that she is so irrational is illegitimately triggered by the rest of the example as I described it, namely that she is in a deliberative context. In reply I have offered two cases, one of a judgement about what pru­ dential requirement that is formed outside of a deliberative context (where lack of motivation does constitute irrationality), and one of a judgement formed within a deliberative context (where lack of mo­tiv­ ation does not constitute irrationality). This serves to undermine the objection that our judgements about which combinations of belief and (lack of) motivation are irrational depends upon whether the context is a deliberative context.

3.2  Objection 2: Illicit Intuitions Someone might seek to resist the claim that it is irrational to instantiate combination (i) and (ii) by arguing that my case for such a rational requirement illicitly trades on only using plausible instances of the rele­ vant belief when I presented the idea. For example, in the cases above, I used the example of believing that, prudentially speaking, one ought to go to law school. Someone might worry that the general plausibility of this belief gives the general rational requirement not to co-­instantiate (i) and (ii) an inflated degree of plausibility. Though I think that this objection fails, it does draw attention to a consequence of PJI that I should make fully explicit, namely that it imposes no restrictions on the relevant belief. Here then is an example to both demonstrate this and to reply to the objection under consideration: Alexis and the Aliens:  Alexis believes that, prudentially, she ought to be picked up by aliens and taken to their planet, where she would rule as

128  Dear Prudence Queen for the rest of time. Alexis believes that this will only happen if she stands at the top of the tallest hill in her neighbourhood. Clearly, and sadly, cases of such unusual beliefs occur. Suppose that the case is as stated and thus Alexis sincerely believes that prudentially she ought to stand at the top of the tallest hill in her neighbourhood. Taking the arguments from earlier and applying them here, I am com­ mitted to the view that, given Alexis’s belief, she would be less than fully rational if not motivated to stand at the top of the tallest hill in her neighbourhood. Contra this imagined objector, this seems like exactly the right thing to say. It is uncontroversial that something has gone awry in Alexis’s psychology in her coming to form this belief. But given that she sincerely holds it, it would be better, rationally speaking, for her to be motivated to stand at the top of the hill. If she formed this belief and was not motivated to act accordingly then her psychology would embody two significant flaws. By contrast, if she has the rele­ vant motivation then, even if her belief formation is flawed, there is something in her psychology that is functioning properly. Given the failure of these objections, I hope to have shown that it is irrational, by itself, to instantiate combination (i) and (ii). If so, then PJI is true.

4.  A More Ambitious Thesis than PJI? Suppose, as I have argued, that PJI is true. Given that PJI is quite a nar­ row thesis, applying only to a small subset of prudential judgements, we might wonder whether a more ambitious thesis—one applying to a wider range of prudential judgements—could be defended. I will now examine the prospects for this. There are two major ways in which a thesis could apply more widely than PJI. It could be wider inter-personally or intra-prudentially: Inter-­personally:  An analogue of PJI that is true of a wider range of subjects of prudential judgements, specifically judgements about what ­others prudentially ought to do.

Prudential Judgements and Motivation  129 Intra-­prudentially:  An analogue of PJI that is true of a wider range of prudential judgements, such as good for, better for, prudential reason, judgements. (These could be combined. A maximally ambitious analogue of PJI would be wider in both respects simultaneously by applying to every prudential judgement about any agent.) Before addressing the question of whether either way of extending PJI is plausible, it will be helpful to pause and briefly lay out some possible grounds of the truth of theses like PJI. (By ‘grounds’ I mean possible explanations of the truth of PJI, and similar theses.) Seeing these pos­sible explanations of PJI, and the like, will help us to see which other commit­ ments follow from these theses and help us to assess their plausibility.

4.1  Possibility 1: Explanation by Conative State of Mind One way of explaining a thesis like PJI—one that postulates a neces­ sary condition between prudential ought judgements about oneself and motivation—is by holding that it is true because of what these judgements are: conative states. On this kind of proposal it is the nature of the conative state of mind that one is in when one forms a judgement about what is best for oneself that guarantees the connec­ tion to motivation. To make this a little less abstract, let us introduce the following assumption: Humeanism about Motivation (‘Humeanism’):  Belief and desire are necessarily distinct existences and motivation requires the presence of a desire. Let me also introduce a thesis that is stronger than PJI, PJI*: Prudential Judgement Internalism* (PJI*):  At least one type of prudential judgement (judgements about what one prudentially ought to do, among current options) is necessarily connected to motivation in all agents.

130  Dear Prudence According to PJI*, judgements about what one prudentially ought to do are necessarily connected to motivation in all agents. If PJI* is true then, on the assumption of Humeanism, these judgements are desire-­like states. That would be entailed by the fact that belief and desire are dis­ tinct existences and that prudential ought judgements are necessarily connected to motivation. Thus if PJI* were true then this would be because prudential ought judgements are desires. Thus, this thesis about prudential judgements and their connection to motivation, if true, would be best explained via a conative state of mind. The key difference between PJI* and PJI is that PJI* holds that the connection between these judgements and motivation is necessary in all agents. PJI, by contrast, holds only that the connection is necessary in all rational agents. This difference enables us to see a different way of  explaining connections between prudential judgements and motivation.

4.2  Possibility 2: Explanation by Rational Requirement Suppose one thought that PJI* was too strong because judgements about what one prudentially ought to do do not necessarily motivate, even if their failing to marks the agent out as, ipso facto, irrational. If, instead, one subscribed only to PJI: Prudential Judgement Internalism (PJI):  At least one type of pruden­ tial judgement (judgements about what one prudentially ought to do, among current options) is necessarily connected to motivation in rational agents then we might ask how that thesis could be grounded. We cannot deploy the explanation-­via-­state-­of-­mind strategy from earlier. That is because if judgements about what one prudentially ought to do are desires then there will be no gap between forming the judgement and motivation (because holding the judgement constitutes being motivated) precluding the possibility of forming the judgement but, irrationally, failing to

Prudential Judgements and Motivation  131 be motivated.12 PJI, therefore, requires a different kind of explanation, one that is not based upon the nature of prudential judgements themselves. PJI could instead be grounded in the following way. Suppose that one is rationally required not to be in certain combinations of states.13 For example, one is rationally required not to (outright, knowingly) both believe P and believe not P. Instantiating that combination is a form of irrationality. Suppose, also, that one of the combinations that is ration­ ally prohibited is the combination of {judging that prudentially one ought to do something} and {not being motivated accordingly}. On this proposal, PJI is true not because the relevant prudential judgements are desires—this view perfectly well allows that they are beliefs—but on the normative ground that there is a rational requirement to be motivated in accordance with your prudential ought judgements.14 We have introduced possible explanations of two theses that postulate a necessary connection between prudential judgements and (some set of) agents. These were the following combinations of thesis-­and-­possible­ground: PJI* & Explanation via conative state of mind. PJI & Explanation via rational requirement. With these combinations in mind, let us now return to the question of whether there is some plausible thesis that is wider than PJI. Take this combination first:

12  This is to assume that to desire is, ipso facto, to be motivated. Some distinguish these. For discussion, see (e.g.) Gregory (2017b: 207), Swartzer (2015), Arpaly & Schroeder (2013). 13  I am assuming that these requirements are ‘wide scope’ requirements. For discussion of rational requirements and wide scoping see, for example, Broome (2007), Brunero (2010), Way (2011), Lord (2014). 14  One might wonder how genuinely explanatory this account is. Does it just reassert the phenomenon to be explained—that one is irrational if not motivated in accordance with one’s judgement about what is best for oneself? That is only a problem if the idea seems implausible in itself. But I take it that it does seem plausible that there is a rational requirement not to be indifferent in the face of judgements about what is best for yourself and, moreover, that this is a plausible candidate for a fundamental truth about rationality.

132  Dear Prudence Prudential Judgement Internalism* (PJI*):  At least one type of prudential judgement (judgements about what one prudentially ought to do, among current options) is necessarily connected to motivation in all agents. PLUS Explanation of PJI* via conative state of mind. There is a certain instability in this combination. This arises from the fact that PJI* applies only to a subset of prudential judgement types (judgements about what that agent prudentially ought to do, among their present options) but it is purportedly explained by the nature of conative states of mind. This prompts the following question: if pruden­ tial ought judgements about one’s current options are desires, what about the other prudential judgements (judgements about what is best for oneself, what is better for oneself, what others ought to do, and the like)? There are two possibilities. Either: 1.  All prudential judgements are desires. OR 2. Only some prudential judgements are desires (judgements about what one prudentially ought to do, among one’s current options). Option (2) involves postulating a disconnect between one subset of pru­ dential judgements and the rest. But it seems implausible that a subset of prudential judgements are a wholly different state of mind (type) from all of the others, that whilst most prudential judgement (types) are beliefs, one subset of prudential judgements are desires. More conces­ sively, someone who takes option (2) takes on the burden of giving a plausible case for a radical division between prudential judgements.15

15  Readers familiar with the metaethical debate over moral/motivational judgement inter­ nalism will remember that people typically argued for non-­cognitivism as a general thesis about moral judgements on the basis of judgement internalism. A striking feature of that debate was the focus on a narrow set of moral judgements (typically moral ought judgements about one’s current options) when trying to motivate the general thesis.

Prudential Judgements and Motivation  133 On the assumption that few, if any, who are sympathetic to PJI* will take option (2) I will assume that most will deem option (1) to be superior. Notice, though, that once someone takes option (1) then they are committed to a much more general thesis than PJI*. In fact, they are committed to a maximally general analogue of PJI*. For if all prudential judgements are desires then this will be true: PJI**:  Every prudential judgement is necessarily connected to mo­tiv­ ation in all agents. Thus if one seeks to explain PJI* via conative state of mind then one will likely end up committed to the view that all prudential judgements are necessarily connected to motivation in all agents.16 That means, for one thing, that judgements about what is slightly prudentially good for the agent themselves will necessarily motivate that agent. More prob­lem­at­ ic­al­ly, judgements about what would be slightly prudentially good for (historically or geographically) distant others will necessarily motivate all agents. Thus, on this view, if I judge that it would be good for some distant stranger if they received £10 in the post then I am necessarily motivated (to at least some extent) to send them £10.17 For now, I want to bracket the question of how plausible this commit­ ment is. I am interested for the moment in simply pointing out that there is a short step from PJI* to PJI**.18 Let us go back to the second pairing of thesis-­plus-­possible-­grounds: PJI  PLUS Explanation via rational requirement. Suppose someone holds PJI and that PJI is true because of a rational requirement. Is there then an equivalent pressure to embrace a more general thesis than PJI, analogous to that we saw above (for those who 16  I.e., each agent will be motivated in accordance with every one of their prudential judgements. 17  If that does not seem problematic enough, notice that the view will be committed to a necessary connection to motivation (or something affective at least) in the case of judgements about what would be slightly good for historically remote agents. 18  On the assumption that it is undesirable to treat one kind of prudential judgement as a different state of mind type.

134  Dear Prudence accept PJI* and explain it via conative state of mind)? Remember that PJI is the following claim: Prudential Judgement Internalism (PJI):  At least one type of prudential judgement (judgements about what one prudentially ought to do, among current options) is necessarily connected to motivation in rational agents. If we thought that PJI was true in virtue of a rational requirement, is one driven to accepting a more general analogue of PJI? The answer is no. For notice that there is a significant difference in plausibility between thinking that rationality prohibits both: (i) Judging that some accessible option is what you prudentially ought to do. (ii) Failing to be motivated at all to pursue that option. And thinking that rationality prohibits both: (xiii) Judging that some accessible option (e.g. sending £10) would be good for some distant stranger. (xiv)  Failing to be motivated at all to pursue that option. In support of this, imagine that someone reported instantiating com­ bin­ation (xiii) and (xiv). Would one be surprised? I assume not. Would one doubt the sincerity of someone who reported being in this com­bin­ ation? Again, I assume not. It seems rationally permissible to be indiffer­ ent in the face of a judgement about what would be slightly good for some distant stranger. This yields two points. First, that accepting PJI, and accepting it on the basis of a story about what rationality requires, does not compel one to accept a maximally general analogue of PJI. Second, that accepting PJI on this kind of basis provides a way of rejecting some more general theses than PJI, on the grounds that it does not seem plausible that rationality requires, e.g., being motivated in accordance with one’s judgements of what would slightly benefit a distant stranger. But, usefully, it also leaves open if

Prudential Judgements and Motivation  135 there are any more general theses than PJI which are true. It does not settle the question of whether there are any prudential judgements, beyond prudential ought judgements about one’s current options, that are necessarily motivating in rational agents. There would thus be a live issue: which combinations of prudential judgement and indifference does rationality permit? By contrast, accepting PJI, or anything more ambitious than PJI, on the basis of a conative state of mind story imme­ diately forces one to adopt PJI**, the maximally general analogue of PJI.

5.  A More Ambitious Thesis than PJI (Continued)? In the previous section, I described two strategies for explaining theses like PJI. I pointed out that explaining theses like PJI in terms of states of mind generates a pressure to accept a maximally general claim about prudential judgements and motivation whereas explaining them in terms of a rational requirement does not. I want now to directly con­ sider the prospects for a more general thesis than PJI. If we accept PJI and we think that PJI is explained by a rational requirement, can we justify a more general thesis, one applying to a larger class of prudential judgements?

5.1  A More Ambitious Thesis: Intra-­prudentially Consider this thesis: Prudential Judgement Internalism (PJI 2): Every self-­directed pruden­ tial judgement is necessarily connected to motivation in rational agents. By ‘self-­directed’ I mean that the judgement is a prudential judgement about one’s own options (characterized first-­personally), so about what is good for oneself, or what one has a prudential reason to do etc. If PJI 2 is true then either every prudential judgement I make about my options motivates me, at least somewhat, or I am irrational.

136  Dear Prudence Let us consider the plausibility of PJI 2 by considering a variant on an earlier example: Deliberating Dugald:  Dugald has reached the end of his undergraduate studies and is deciding what to do. He spends time thinking about the various options open to him and, as he does so, judges that one of them—law school—would be good for him to at least some extent. If we know that this description is correct, what do we expect? I will assume that that we do not expect this judgement to be necessarily accompanied by action. It would be problematic, and bizarre, if judge­ ments about what is good for oneself to some degree necessarily prompted action. But do we expect Dugald to be motivated at least a lit­ tle, given his judgement? I am not completely sure here, but it does seem to me both possible, and rationally permitted, for Dugald not to be motivated to go to law school, despite his judgement that the option is good for him to at least some extent. In deliberation we typically are aware of at least some of the large number of options that will be good for us to at least some extent, and it seems implausible that for each such judgement we must be motivated, on pain of irrationality. Suppose, for example, that one knows that one option is what one prudentially ought to do and someone correctly points out that some alternative has some prudential value. I do not think that we would doubt the sincerity of someone who reports being indifferent towards that option, despite acknowledging that there is something to be said for it, prudentially speaking. It thus seems to me that we should reject the view that every self-­directed prudential judgement will be accompanied by motivation, unless the agent is (ipso facto) irrational. If this is correct, there is a division within self-­directed prudential judgements between (at least) (i) judgements about what one pruden­ tially ought to do and (ii) judgements about what would be good for oneself to some extent. One is rationally required to be motivated in the presence of the former judgements but not the latter.19 19  One issue which my discussion so far has ignored is the precise set of prudential judge­ ments that are plausible candidates for the rational requirement to be motivated in conjunction with them. So far I have discussed prudential ought claims. One might wonder, though, about

Prudential Judgements and Motivation  137 What about the other way of making PJI more ambitious, by extending it inter-­personally?

5.2  A More Ambitious Thesis: Inter-­personally The other way of moving to a more ambitious thesis than PJI is to an analogous claim about what any agent prudentially ought to do. We can formulate this explicitly as the following claim: Prudential Judgement Internalism (PJI 3):  Prudential judgements about what anyone prudentially ought to do are necessarily connected to motivation in rational agents. Is it plausible that there is a rational requirement not to be indifferent in the face of any judgement about what would someone prudentially ought to do? Again, it seems to me that this is not a plausible candidate for a rational requirement. There are an enormous number of actions that someone else prudentially ought to take where it does not seem plausible that every such judgement we might make about such options entails a rational requirement on motivation (or the like). To make things more concrete, take the following truth. I look out the window and see one of my neighbours in their garden, one whom I have never met. My neighbour ought, prudentially, to be eating healthily, tak­ ing sufficient exercise, saving for their retirement, and so on. I judge all of this but such judgements are not accompanied by any motivation on my part. And this does not seem like a rational failing. As evidence for this note, again, that one does not doubt at all the sincerity of someone who professes to be indifferent about the prudential requirements of

these prudential judgements—[prudentially] I must PHI, [prudentially] I should PHI, it would be [prudentially] best for me to PHI. It is not obvious how prudential ought and must judge­ ments relate to the judgement that something is best for oneself, nor how ‘must’, and ‘should’ judgements relate to ‘ought’ judgements. However, I take all of the arguments given so far con­ cerning ought judgements to carry over to ‘must’ judgements. I address this issue more in section 6.

138  Dear Prudence some stranger. One does not seem to manifest a rational failing merely in virtue of being in this combination of states.20 In stark contrast with judgements about what oneself prudentially ought to do, it seems rationally unimpeachable to be indifferent in the face of a judgement about the prudential requirements of some distant other. Perhaps I manifest virtue if I hope that my neighbour does what she prudentially ought to do, but I do not seem to be irrational if I do not desire so. Furthermore, take a case where a stranger’s doing what they pruden­ tially ought to do is highly detrimental to my own interests. Suppose that if Mandeep accepts the job offer then this will be bad for me (e.g., her superior ability will result in my being made redundant). In such a case it seems implausible that I am irrational if I fail to desire or hope that Mandeep do what she prudentially ought to do. It is not merely that I am allowed to prefer overall that she fail to act in her own best inter­ ests. Rather, it is not plausibly irrational if I do not desire to some extent that Mandeep do what she prudentially ought to do. And this is despite the fact that my judgements about what Mandeep prudentially ought to do, or would be best for her, are prudential judgements.

6.  Small Amendments to PJI In the previous section we started with PJI and sought to discover if there is a plausible thesis that is wider intra-­prudentially or interpersonally. I first examined a counterpart of PJI—PJI2—which applies to a wider range of first-­personal prudential judgements, such as judge­ ments about what would be prudentially good to some extent, for the agent. I rejected the claim that such judgements are necessarily mo­tiv­at­ ing on pain of irrationality. I then moved on to another counterpart of PJI—PJI 3—which applies to a wider range of prudential ought judge­ ments, by applying to judgements about what other agents prudentially 20  To put the point the other way, one might worry that the amoralist manifests no rational failing (even if they manifest abundant other failings) and that to claim that they do is, at bot­ tom, a kind of bluff. But there seems no bluffing involved in thinking that one is irrational if indifferent in the face of one’s judgements about what would be best for oneself.

Prudential Judgements and Motivation  139 ought to do. Again, I rejected the claim that such judgements are neces­ sarily motivating on pain of irrationality. I have argued for PJI and have argued against each of two wider claims than PJI, on the grounds that they postulate implausible rational constraints. However, we should broaden PJI to encompass a slightly broader class of cases than those considered so far, by making two amendments. I have explicitly focused on prudential judgements of what one prudentially ought to do. But all of the arguments for the con­ nection between these judgements and motivation (and affect) in rational agents seem equally well to apply to the following judgement types: (a) judgements of what one prudentially must do and (b) judge­ ments about one’s own needs. If one judges that one prudentially must do something or that one prudentially needs something, then either one has the corresponding motivational and affective states or one is, ipso facto, irrational. These two amendments take us from PJI to this broader thesis, PJI 4, the official statement of the view defended in this chapter: Prudential Judgement Internalism 4 (PJI 4):  Judgements of what one prudentially ought or must do, and one’s own needs, are necessarily connected to motivation and affect in rational agents. Having argued for PJI 4 in this chapter, I want to finish by addressing possible implications (one real, one illusory) of PJI 4 for two further debates.

7.  Implications of PJI 4: Cognitivism and the Alienation Constraint 7.1  Non-­Cognitivism about Prudential Judgements In metaethics there is a long history of using the debate about the truth of moral judgement internalism (often ‘motivational judgement internalism’)—roughly, the view that moral judgements are necessarily mo­tiv­at­ing—as a proxy war for the debate between cognitivists, who

140  Dear Prudence hold that moral judgements are purely belief-­ like states, and noncognitivists, who hold that moral judgements are desire-­like states. In brief, the idea was that if moral judgement internalism was true then, assuming Humeanism, this would entail non-­cognitivism about the nature of moral judgements.21 The dialectical strategy was to leverage the plausibility of the claim that all moral judgements are connected to motivation, and thus must be desire-­like states, by noting the oddness of judging that something is morally wrong whilst being left unmoved in light of that judgement. I want to make explicit something which was left somewhat implicit earlier, namely that the arguments in this chapter have tended the other way with respect to prudential judgements. I have argued that the strat­ egy of explaining PJI and the like in terms of conative state of mind is implausible precisely because it is implausible that prudential judge­ ments are necessarily motivating as a class, even in rational agents. Whilst there is a (narrow) class of prudential judgements for which it is plausible that there is a necessary connection between the judgement and motivation (or affect) in rational agents, for many prudential judge­ ments there is no barrier, either psychological or rational, to making the judgement but being left indifferent by it. If so, then there is not an argument from the affective profile of prudential judgements to noncognitivism about prudential value judgements.22 On the contrary, the affective profile of prudential judgements is some evidence for cognitiv­ ism about prudential judgements.

7.2  The Anti-­Alienation Constraint on Prudential Value Finally, let me forestall a possible misapplication of PJI 4.

21  There is a large literature on these issues. For representative moves and countermoves see: Cuneo (1999), Hare (1952), Gibbard (2003), McDowell (1979), Ridge (2015), Smith (1994), Svavarsdóttir (1999), Wedgwood (2004). 22  At least, not a pure form of non-­cognitivism. Things are more complicated in the case of hybrid theories and at least one hybrid non-­cognitivist, Ridge (2015) defends a restricted form of internalism of the sort proposed here. For a sample of hybrid theories see Fletcher & Ridge (2014).

Prudential Judgements and Motivation  141 Many have argued that it is a constraint of some sort on prudential value theorizing that what is good for an agent must in some way be constrained by their endorsement (their desire or some other kind of attitude). Some have argued that prudential value is in fact subject to an anti-­alienation constraint whilst others have argued against it, and the debate over this putative constraint is one of the major fault lines in con­ temporary work on well-­being.23 A classic articulation of the motivation behind the constraint come from Railton: It does seem to me to capture an important feature of the concept of intrinsic value to say that what is intrinsically valuable for a person must have a connection with what he would find in some degree com­ pelling or attractive, as least if he were rational and aware. It would be an intolerably alienated conception of someone’s good to imagine that it might fail in any such way to engage him.24

It might be tempting to think that PJI 4 is probative for the question of whether there is an anti-­alienation constraint. That is, someone might think that the truth of PJI 4 entails the anti-­alienation constraint (or at least provides strong evidence for it). Herein I do not take up the ques­ tion of whether there is such an anti-­alienation constraint on prudential value. Rather, adopting an argument from Alex Sarch,25 I aim to show that the truth of PJI 4 does not entail such a constraint. The easiest way to see that the truth of PJI 4 does not have such impli­ cations is to note that, as formulated, PJI 4 says nothing about the truth of the relevant prudential judgement. Remember our starting example: Deliberating Denise:  Denise has reached the end of her undergraduate studies and is deciding what to do with her future. She spends time

23  I return to this idea in Chapter 7. See also: Dorsey (2017), Fletcher (2016a: ch. 3 appen­ dix), Hall & Tiberius (2016), Hawkins (2014), Fletcher (2013). 24  Railton (2003). A sophisticated proposed constraint comes from Rosati (1996). Rosati argues for her constraint on the basis of something relevantly like, but more general than, PJI. For discussion of her argument see Sarch (2011). 25  Sarch (2011: §2.1). See also Dorsey (2017). For discussion of similar issues in the case of moral judgements and motivation see Wedgwood (2004).

142  Dear Prudence thinking about the various options open to her and concludes that one of them—law school—is what she prudentially ought to do. This case and the subsequent discussion took no stand on whether going to law school really is the option that Denise prudentially ought to take. Furthermore, we can easily vary the case by adding the stipulation that some other option—engineering school—is, in fact, better for her, irre­ spective of her judgement to the contrary, or that in fact she is very much mistaken and that going to law school would, in fact, be very bad for her. Neither of these amendments undermines our expectation that Denise will be motivated at least somewhat to go to law school, given that she sincerely judges that that option is best for her. Thus the plausibility of PJI 4 does not depend upon the assumption that the judgement in question is true. We expect people to be motivated by their judgements about what is best for them both when such judgements are true and when they are false.26 In light of this, we cannot derive an anti-­alienation constraint on prudential value—that something only actually has prudential value when it is, or would be, endorsed by the agent—from the truth of PJI 4.

26  Remember, too, the case of Alexis and the Aliens.

PART THREE

PRUDEN T IA L DISC OU R SE I S NORMAT IV E . WHAT F OLLOWS ?

6 Prudential Normativity, Moral Scepticisms, and Metaethics 1.  Introduction In earlier chapters, I argued for two theses. First, that prudential value gives agents normative, prudential reasons. Second, that prudential judgements are normative judgements (so prudential discourse is on a par with moral discourse). I now spell out some ramifications of these theses. I proceed by examining, in a programmatic fashion, four differ­ ent areas of inquiry about morality and moral discourse, showing how the theses hitherto defended in this book affect them. My ambitions in this chapter are fairly modest. I am concerned to point out how the theses that prudential value and prudential discourse are normative affect certain questions in ethics and metaethics. In doing so I sometimes distinguish some different argumentative choices that one might take in response and I make some provisional claims about their viability. However, I do not pursue the various options in detail, leaving that as work for another time. Here is how I proceed. In section 2, I look at one form of moral scep­ ticism, that found within the ‘why be moral?’ debate and point out the parallel with the question ‘why be prudent?’. Section  3 examines her­ men­eut­ic moral error theory and proposes a companions-­in-­guilt argu­ ment based on the normativity of prudential discourse. In section  4, I examine arguments given within the literature on revisionary metaethical views, pointing out and questioning their commitment to prudential justifications. In section  5, I show how the normativity of prudential properties applies to a central debate about thick concepts, that between reductionists and non-­reductionists about such concepts.

Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0007

146  Dear Prudence

2.  Why Be Moral and the Normativity of Prudence A classic question in moral philosophy is why one should be moral. Morality gives us requirements and obligations, many of which can be onerous to adhere to, so why should I do as morality requires? What reasons do I have to go along with it? One type of answer to this question rejects the presupposition of a need for some further reasons to be moral (put another way, it rejects the presupposition that the demands of morality are not normatively authoritative).1 A second, historically influential, type of answer, the one I focus upon here, is that we should be moral because it is good for us to be so. More precisely, the argument is that it is, or is highly likely to be, in an agent’s best interests to be moral where ‘being moral’ includes both acting as morality requires (etc.) and attempting to cultivate specific moral virtues. We find a classic expression of this answer to the ‘why be moral?’ question in the early books of Plato’s Republic. It occurs both in the ini­ tial framing of the problem: We must now examine, as we proposed before, whether just people also live better and are happier than unjust ones.2 [. . .] Therefore, praise justice as a good of that kind, explaining how—because of its very self—it benefits its possessors and how injustice harms them.3

as well as Socrates’ description of what he sets out to do in the rest of the dialogue: Glaucon and the others begged me not to abandon the argument but to help in every way to track down what justice and injustice are and what the truth about their benefits is.4

and, finally, in the description of what has been shown: 1  For example, see Prichard (1912). 2  Book I, 352c. 3  Book II, 367d. This framing of the problem is only one way of understanding the why be moral challenge. As I mention below, my discussion here is not supposed to apply to all forms of the why be moral challenge. 4  Book II, 368c.

Prudential Normativity and Metaethics  147 [I]f we are persuaded by me . . . we’ll always hold to the upward path, practicing justice with reason in every way . . . Hence, both in this life and on the thousand-­year journey we’ve described, we’ll do well and be happy.5

We find the same kind of argument throughout the history of moral phil­oso­phy.6 For example, in his discussion of whether moral phil­oso­ phy rests on a mistake, Prichard describes this answer thus: [T]hey state that we ought to do so and so, because, as we see when we fully apprehend the facts, doing so will be for our good, i.e. really, as I would rather say, for our advantage, or, better still, for our happiness[.]7

Let us take a step back and ask a very basic question. How would this kind of connection between being moral and one’s own interests answer the ‘why be moral?’ question? The most plausible explanation, I take it, is the background assumption of this familiar thesis: Prudential Value Matters (PVM):  Evaluative prudential facts generate directive prudential facts (including facts about prudential, normative, reasons for action and for attitudes).8 So, spelt out in full, this type of answer to the ‘why be moral?’ challenge assumes that evaluative well-­being facts generate normative reasons for the agent and it is that which underwrites the guarantee that if we show that being moral is best for the relevant agent,9 then we will have shown that they have reason to be moral.10 5  Book X, 621c. 6  As Foot (1958–9: 100) puts it: ‘In the Republic it is assumed that if justice is not a good to the just man, moralists who recommend it as a virtue are perpetrating a fraud.’ 7  Prichard (1912). It is an interesting question why Prichard thinks that ‘happiness’ would be the best way to formulate this view. 8 There is abundant evidence that we make this background assumption. As Worsnip (2018: 237) puts it: ‘prudential reasons (i.e., reasons to do with the acting agent’s self-­interest)— which frequently appear on lists of different kinds of reasons—seem often to receive a free pass.’ 9 Or most likely best for, a complication I ignore here. 10  I do not mean to deny that there are alternate readings of the why be moral challenge and how prudence is relevant to it. For example, one might think that the response simply takes it to be a brute fact about human psychology that people desire their own well-­being or, more specifically, that they desire their own happiness.

148  Dear Prudence This answer to the ‘why be moral?’ question clearly assumes that prudential value is normatively significant. After all, it is not only true that, without this assumption, the question would recur but that this is so obviously the case. Recognizing the way in which this answer to the why be moral question relies upon the normativity of prudence opens up a new, modest way of responding to it.11 We can respond to the person who poses the question by asking why we should treat morality and prudence so differently. Why assume that the demands of morality are not, themselves, normatively authoritative (they must be backed up by prudential reasons) whilst treating prudence as normatively authoritative? Why make the assumption that showing that something is in someone’s interests settles whether they have reason to do it (especially when, according to the challenge, showing that the action would be wrong does not)? The ‘why be moral?’ questioner may respond to this challenge in dif­ ferent ways. The major division between responses stems from whether they accept that their implicit asymmetric treatment of morality and prudence is unjustified. Suppose, first, that they accept that their asymmetric treatment of morality and prudence is unjustified. Then they can respond either by (i) treating morality as intrinsically authoritative (and thus cease to ask ‘why be moral?’) or by (ii) doubting that prudence is intrinsically authoritative (and thus start to ask ‘why be prudent?’). Alternatively, they might take the second route, that of arguing that treating morality and prudence differently is justifiable. Doing so places a burden upon them; that of providing such a justification for treating morality and prudence differently, such that prudence enjoys the status of needing no further justification whilst morality does not. I guess some will take route one. Within that set of people, I assume that most will take sub route (ii), by coming to have doubts about pru­ dence, rather than being reassured about morality. Either way, I assume

11  This is a form of companions-­in-­guilt argument, of course. I leave this labelling out for fear of generating confusion between this part of the chapter and the next part.

Prudential Normativity and Metaethics  149 that the first general route will prove less popular than the second route, so I will leave it aside here.12 Suppose the ‘why be moral?’ questioner takes route two instead. How might they proceed? They might argue that the distinction between one­ self and other agents suffices to distinguish prudence from morality. That is, they might argue that morality’s demands are not normatively authoritative by themselves because they are other-regarding whereas prudential demands are self-regarding. It is unclear that this claim is true, given that it is non-­obvious that there are no self-­regarding moral requirements.13 But even if we suppose that there are no such require­ ments, it is unclear why the truth of this self/other distinction justifies the asymmetry that treats the normative authority of prudence as secure whilst harbouring doubts about it for morality. On one reading it begs the question, by simply pointing out that morality is morality and pru­ dence is prudence. That cannot be the answer. Or, perhaps it can. Perhaps a fairer reading of the response is that it is claiming that it is plausibly a basic normative truth that we have reasons to be prudent and that this is much less plausible in the case of morality, and so prudence alone legitimately enjoys the default presumption of normative authority. Of course this view is coherent and it might be true but I take it that this kind of brute fact explanation is an answer of last resort. If the ‘why be moral?’ questioner does not take this option, what other story have they to tell about why prudence is intrinsically authori­ tative whilst morality is not? Determining whether any other argument of this form can be made to work is a challenge for another day.14 In this section, I have shown that the ‘why be moral?’ challenge or, more precisely, one version of it,15 rests on the assumption that prudential 12  Darwall (2016) has recently argued in favour of an opposite asymmetry, holding that morality is intrinsically authoritative whereas prudence is not. His discussion seeks not to interpret but, rather, to overturn the standard dialectical set-­up by arguing that it is easier to show that morality is authoritative than it is to show that prudence is. 13 It is tempting to see self-­regarding moral requirements as misidentified prudential requirements. 14 One possibility is that they will offer a version of the argument we encountered in Chapter 2, that prudential value is desire-­based. 15  There are at least two other forms of why be moral challenge which are not affected by the discussion here. One is the form deployed by Thrasymachus at the beginning of the Republic. Another is that deployed by Hume equally against both morality and prudence.

150  Dear Prudence value is normative. I have also shown that once we make this assumption explicit, the burden shifts to the person who is sceptical of the normative authority of morality. It is they who must provide a story about how morality and prudence differ such that either the ‘why be prudent?’ ques­ tion does not arise, or arises but is answerable, because of a difference between morality and prudence. I do not pretend to have shown that they cannot meet this burden, just that it is their burden. In this section, I have shown how making the normativity of pruden­ tial discourse explicit can help to shift the burden back onto one form of moral sceptic. This kind of sceptic is someone who grants that there are moral requirements but doubts that they are genuinely authoritative over them. Let me now move on to applying the normativity of pruden­ tial discourse to a different form of moral sceptic, the error theorist who thinks that moral discourse is fundamentally flawed because there are no moral requirements.

3.  A Prudential Companions-­in-­Guilt Argument Hermeneutic moral error theorists (hereafter ‘error theorists’) hold that our existing moral discourse—thought and talk about what we morally ought to do, what we have moral reason to do and the like—is funda­ mentally flawed because it is committed to the existence of normative properties (or facts or relations) that do not exist. The property com­ monly focused upon in discussion of error theory is the property of being a categorical reason for action.16 However, the error theorist is not best formulated solely in terms of categorical moral reasons.17 The scope of Mackie’s error theory is clear when he writes:

16  Reminder: a categorical reason for action is one that is relevantly independent of the agent’s desires. For example, if I promise to give you my car, but doing so would not fulfil a desire of mine, there is still a moral reason for me to do so, a reason generated by my promise. It is widely but not universally accepted that moral discourse is committed to categorical reasons. 17 For discussion see Bedke (2010), Hampton (1998), Olson (2014: ch. 6), Streumer (2017: ch. 8).

Prudential Normativity and Metaethics  151 There are no objective values . . . The claim that values are not objective, are not part of the fabric of the world, is meant to include not only moral goodness . . . but also other things that could be more loosely called moral values or disvalues—rightness and wrongness, duty, obligation[.]18

According to error theory, moral discourse is committed to an array of normative properties that are sui generis, irreducibly normative, features of actions (etc.).19 In holding that the properties ascribed within moral discourse are sui generis irreducibly normative features of actions, the error theorist contends that they cannot be identified with any onto­ logic­al­ly innocent, natural property. The error theorist combines this view about the commitments of moral discourse with the denial that such properties exist. The upshot is that nothing is morally permissible (or impermissible) or morally good (or bad), and so on. Naturally, many baulk at the idea of moral discourse being wholly in error. A popular strategy for replying to the view by, or on behalf of, moral realists, has been to find so-­called ‘companions in guilt’. These are forms of non-­moral discourse that seem in good standing—an error theory is implausible for them—but which share the putatively prob­ lematic features of moral discourse. Wielders of these arguments then use the good standing of the second kind of discourse, coupled with its sharing the putatively problematic features of moral discourse, to argue that moral discourse must be equally in good standing.20 This general strategy has been most thoroughly explored in connec­ tion with epistemic discourse (thought and talk about e.g. what we are justified in believing).21 The merits of the strategy are clear. Nonetheless, doubts have been raised concerning the extent to which epistemic dis­ course shares the problematic features of moral discourse. If epistemic

18  Mackie (1990: 15). 19  This is not to deny that one could combine error theory with the claim that all normative properties are identical, or reduce, to categorical reasons for action, only to point out that this is a further commitment. 20  See Lillehammer (2007) for discussion of a selection of such arguments. 21  Especially Cuneo (2007).

152  Dear Prudence discourse does not share these feature then, even if it is in good standing, it makes no trouble for the moral error theorist. In this section, I show how the claim that prudential discourse is nor­ mative opens up the possibility of a new companions-­in-­guilt argument, one more powerful than the argument from epistemic discourse. I start with a more precise characterization of the epistemic companions-­inguilt argument. I then explain a reductionist reply to the epistemic argu­ ment that seems available to the moral error theorist. Finally I present the prudential companions-­in-­guilt argument and argue for its su­per­ ior­ity over the epistemic argument. A common reply to the moral error theorist has been to argue that epistemic discourse is equally committed to irreducibly normative properties. If such parity is real, then moral and epistemic discourse stand or fall together. Proponents of the argument have assumed that few will embrace epistemic error theory and that we will, instead, reject moral error theory on the grounds that epistemic discourse successfully ascribes irreducibly normative properties. Cowie nicely spells out the epistemic companions-­in-­guilt argument (‘Epistemic Argument’) thus: Epistemic Argument (1)  Parity premise:  If the arguments for the moral error theory (what­ ever they may be) are sufficient to establish that the moral error theory is true, then those arguments (or appropriate analogues of them) are sufficient to establish that the epistemic error theory is true. (2)  Epistemic existence premise:   The epistemic error theory is false. So: (3)  The arguments for the moral error theory are not sufficient to estab­ lish that the moral error theory is true.22

22  Cowie (2014b: 408). For discussion of this and related arguments see especially Cuneo (2007), Rowland (2013, 2016), Das (2017).

Prudential Normativity and Metaethics  153 Having introduced the moral error theory and the epistemic argument let me explain a common error-­theory-­friendly reply to the argument before presenting my prudential alternative. In response to the epistemic argument outlined above, one might accept (1) and seek to reject (2), by arguing for error theory about epi­ stem­ic discourse. There are various ways that an error theorist might seek to do this.23 A more common reply—the focus of the literature on the epistemic argument—has been to undermine premise (1) by casting doubt on the claim that epistemic discourse is committed to the same problematic properties as moral discourse is. One way to put the point is to say that epistemic error theory is false because epistemic discourse does not, in fact, ascribe ontologically problematic properties.24 The trouble for the epistemic argument comes from the fact that, for epistemic discourse, there is a plausible case for a reduction of the prop­ erties ascribed within the discourse to something ontologically unprob­ lematic. On such a view, whilst epistemic discourse uses the same normative terms as moral discourse (good belief forming mechanisms, reasons to believe, what we ought to believe, etc.) within epistemic dis­ course the terms are used to ascribe different, ontologically respectable, properties.25 What might such a reduction be like? The suggestion developed and examined at most length is that epistemic discourse ascribes alethic properties, such as obtaining or preserving true beliefs. So, for example, the property of being an epistemic reason just is the property of being something that raises the probability of believing the truth (and so on for other properties ascribed by epistemic discourse). There are of 23  For example, by arguing that moral and epistemic discourse are actually entangled such that there is no division between the two (so epistemic discourse inherits the problematic com­ mitments of moral discourse.) For discussion of this see Cuneo (2007: ch. 2). Another possibil­ ity would involve there being pragmatic encroachment on the epistemic, because epistemic properties are appropriately related to moral reasons for action. On this, see Fantl & McGrath (2007) and Hawthorne & Stanley (2008). 24  Fletcher (2009), Heathwood (2009), Olson (2011b), Cowie (2014a, 2014b, 2016a). 25  There are two versions of this response. The first is that epistemic discourse ascribes nor­ mative properties but that the discourse is in good standing because it ascribes reductively normative properties. The alternative is to hold that epistemic discourse is in good standing because the properties that it ascribes are non-­normative. The difference between these two ways of replying does not matter for my purposes here, given that either will falsify the parity premise.

154  Dear Prudence course tricky questions to ask about the precise relationship(s) between epistemic claims and alethic properties (and generally whether this reduction succeeds). One licence for optimism for the strategy highlights an apparent dis­ analogy between the epistemic and the moral. This is the fact that there is an ontologically unproblematic property—true belief—that plausibly plays a central role in epistemic discourse, such that something cannot qualify as epistemic discourse without being connected to that property. By contrast, there seems to be no such candidate in the case of moral discourse. There does not seem to be an ontologically innocent prop­ erty, or set of properties, such that something does not count as moral discourse unless this property plays a central role in such discourse.26 As Heathwood puts it: The sentence ‘this is likely, given my evidence, but it’s not reasonable for me to believe it’ does have an air of incoherence about it in a way that axiological statements—even such patently false ones like ‘suffer­ ing is intrinsically good’—never do.27

This feature of the epistemic gives some plausibility to the claim that epistemic discourse can be reduced—that it uses the same vocabulary as moral discourse but in an innocent fashion—to talk about truth and the promotion of true beliefs, rather than to anything ontologically problematic.28 This yields an asymmetry between the moral and the epi­stem­ic which gives grounds for doubting the parity premise of the epistemic argument. This response has three merits. First, it is plausible that anything unconnected to truth and the promotion of true beliefs is not a form of epistemic discourse. Second, the property of promoting true belief is, plausibly, ontologically unproblematic, and so there is some reason to think that the epistemic could be reduced to the ontologically unprob­ lematic. Third, no analogue of these points plausibly holds for moral discourse (a claim that is common ground with the moral error theorist). 26  Moore (1903: ch1). 27  Heathwood (2009: 90) 28  For discussion see Cowie (2016a: §4).

Prudential Normativity and Metaethics  155 Let me clarify the dialectic at this point. A common reply to the epi­stem­ic argument has been that, irrespective of the status of (2), there is reason to doubt (1)—that epistemic discourse is relevantly on a par with moral discourse. This reason to doubt (1) comes from the plausibility of a reductive account of epistemic normativity. I do not claim that this reduction is uncontroversial or that it has been achieved.29 Rather, my claim is only that this is a plausible ground for doubting premise (1) and that, other things being equal, it would be better to find a companions-­inguilt argument without this possible weakness. This would all be moot if there were no other candidates for companions-­ in-­ guilt arguments. Indeed, some proponents of this reductionist response to the epistemic error theory have presented it as a master argument against all forms of the companions-­in-­guilt strat­egy.30 But this is to overlook the possibility of mounting such an argument based on prudential discourse, once we recognize that prudential discourse is normative. Let me present the argument (adapting Cowie’s formulation): The Prudential Argument (1) Prudential Parity premise:  If the arguments for the moral error the­ ory are sufficient to establish that the moral error theory is true, then those arguments (or appropriate analogues of them) are sufficient to establish that the prudential error theory is true. (2) Prudential existence premise: The prudential error theory is false. So: (3) The arguments for the moral error theory are not sufficient to estab­ lish that the moral error theory is true. Support for premise (1) comes from the arguments in previous chapters that show that prudential discourse is normative like moral discourse.

29  Someone attracted to (e.g.) a deontological conception of epistemic justification might deny that truth plays the kind of role that is built upon by the reductionist view that I have been considering. 30  See for example Cowie (2016a).

156  Dear Prudence In support of (2), it seems undeniable that some things are good or bad for people, that some people lead lives that go well for them (or vice versa), and that some outcomes are better (or worse) than others for someone. For example, aren’t at least some pains that are experienced by at least some subjects bad for those subjects? Could it really be the case that no pain could ever be bad for anyone?31 Note that premise (2) does not assume any particular substantive the­ sis about prudential value. It relies on the claim that something is good (or bad) for people. Thus the error theorist who seeks to reject premise (2) must defend the following: nothing is either good or bad, beneficial or harmful, better or worse, for anything. That would commit them to the view that there are no welfare subjects, that nothing has a level of well-­being. Similarly, they must reject the thesis of Chapter 1—Prudential Value Matters—and hold that no agent ever has prudentially grounded reason for anything. I do not claim that premise (2) is indubitable. An error theorist might be able to come up with an argument for rejecting that premise (and any error theorist who accepts (1) must do so). But, crucially for my pur­ poses here, premise (2) seems at least as plausible as the equivalent premise of the epistemic argument does. That is, denying premise (2) of the prudential argument, and holding, for example, that nothing is bet­ ter or worse for anyone at all is at least as implausible as holding, for example, that no belief is more justified than any other. The most likely response by the moral error theorist, I will assume, is to try to reject premise (1).32 How might they seek to do so? Presumably by taking the model that we saw above, for the reduction of epistemic discourse, and applying it to prudential discourse. So is there a plausible reduction of prudential discourse to the ascription of a set of onto­logic­al­ly unproblematic properties?33 I think not. Let me explain why. 31  For example, read the horrifying opening of ‘The body of the condemned’ in Foucault’s Discipline and Punish. 32  I take the following to be evidence of this likelihood claim: to my knowledge, no one has ever defended a form of prudential, normative, error theory. There are narrowly moral error theorists such as Joyce (2007: 75) who are happy to talk about the prudential reasons that we have. Conversely, we get more general normative error theorists, like Olson (2006), Streumer (2017), and Joyce (2001: 50) who think that prudential judgements are true but non-­normative. 33  There are plausible candidates for normative properties that prudential discourse might be reducible to. For example, Darwall’s (2004) claim that prudential value is what it would be rational to want in so far as you care about someone.

Prudential Normativity and Metaethics  157 The most likely proposal for such a reduction, I assume, is that pru­ dential properties reduce to hedonic properties: for example, that being good (or bad) for a welfare subject just is being pleasurable (or painful) for them, and so on for the other prudential properties (better than, worse than, etc.). The difficulty for this proposed reduction is that, whilst hedonic properties, plausibly, are ontologically innocent, they do not seem like plausible candidates for the properties ascribed by pru­ dential discourse (nor does it seem like there are other plausible candidates). It would be evidence against the proposal that prudential properties are identical to hedonic properties if it was plausible that there can be communities of agents who make prudential judgements without ascribing hedonic properties. And, more generally, it counts against any proposed reduction of the prudential to the ontologically unproblematic if we can construct plausible scenarios where there is prudential dis­ course that does not ascribe those properties.34 It seems plausible that we can construct such scenarios, the pruden­ tial analogue of Horgan & Timmons’ (1991) Moral Twin Earth thought experiment.35 On prudential twin earth 1, agents have a form of dis­ course connected to the exercise and development of unique human capacities. They hold that theoretical contemplation is the only uniquely human capacity, and it is plausible that they hold that only this activity has prudential value (and that its value in no way depends upon it being pleasurable). Evidence for this commitment comes from the advice that they give each other (‘only have fun if it helps your contemplation’, ‘only exercise if it helps your contemplation’), and from how they structure their lives both as individuals and as a society. For example, people only non-­instrumentally desire contemplation for themselves, they reliably seek it out for themselves and for their loved ones, they envy those with more contemplation time than themselves, and so on.36

34  These points are not decisive against the proposals. Rather, it is some evidence against the reduction, and provides the task for a reductionist account of explaining why it seems like there can be (e.g.) prudential discourse that does not ascribe the relevant properties. 35  See also Hare’s (1952) missionary and the cannibals. 36 There is more to say about how precisely we determine that these are prudential judgements.

158  Dear Prudence By contrast, on prudential twin earth 2, agents have a similar form of discourse that is connected to pleasure and pain. It is plausible that they hold that only these hedonic phenomena have prudential value (or dis­ value). Evidence for this comes from the advice that they give (‘have fun!’, ‘only exercise if it feels good’) and from how they structure their lives as individuals and society. This pair of scenarios exhibits the following features. It seems plaus­ ible that agents in both worlds make prudential judgements and that they manifest a disagreement in conceptions of prudential value.37 They disagree, for example, about the prudential value of having fun and doing exercise despite the lack of a plausible candidate for an onto­logic­ al­ly unproblematic property that their judgements commonly ascribe. This suggests that we cannot reduce prudential properties to hedonic properties in particular, and it seems like the argument can be run with equal plausibility for other ontologically innocent properties (prolong­ ing life, life-­satisfaction, desire-­satisfaction, etc.). In contrast with epistemic discourse, where there is some plausibility to the idea of a reduction of the epistemic to the alethic, there seems to be little to no plausibility in the idea that prudential discourse reduces to ontologically unproblematic properties. Given this difference in plausi­ bility, the parity premise seems secure for the prudential argument (and more secure than for the epistemic argument). The moral error theorist who wants to reject premise (1) thus needs to give an account of how prudential discourse can be reduced to the ontologically unproblematic.38 Furthermore, they must do so within the constraint that the account must not carry over to moral discourse, on pain of ceasing to be moral error theorists. I do not pretend to have shown that that is impossible, but the argument above constitutes a licence for pessimism. There is no obvious candidate reduction available for prudential discourse. The alternative option for the error theorist is to accept premise (1), bite the bullet, and try to argue that prudential error theory is plausible. To

37  We saw this point also in Chapter 2, in discussion of the markers of normativity. 38  They could put this either as the claim that prudential discourse is normative, but not sufficiently like moral discourse, or that it is non-­normative.

Prudential Normativity and Metaethics  159 do this, they must make plausible not only that we lack moral obligations to alleviate the pain of people whom we could easily help, but, further, that such pain is never bad for those who experience it. Once we see the plausibility of the claim that prudential discourse is normative we get a new strategy for resisting the error theory (or, at least, a new variant on a familiar strategy) that is more plausible than the most-­explored existing proposal. If the prudential is normative then the dialectical costs of moral error theory are much higher than previously acknowledged. The moral error theorist must hold that pains are not only something that we have no moral reason to alleviate but that, in fact, they are never bad for those who experience them. Thus, no one has reason to alleviate or prevent pains to others or even to themselves. Having shown how the normativity of prudential discourse affords a new reply to the moral error theorist, let me now move on to showing how it also throws a spanner in the works of a project downstream from moral error theory, that of assessing what we should do with moral dis­ course if moral error theory is true.39

4.  Why Be a Moralizer? Prudential Normativity and Revisionary Metaethical Views In the previous section, I focused upon hermeneutic moral error theory and the consequences for its proponents if prudential discourse is simi­ larly normative. Moral error theorists have sometimes combined their hermeneutic view of existing moral discourse with a revisionary metae­ thical view, a theory of whether and, if so, how, to revise moral discourse, given the truth of moral error theory.40 Discussions of such revisionary proposals are typically premised on the assumption of hermeneutic error theory (either as a real commitment or in order to draw condi­ tional conclusions about what we should do if moral error theory were

39  Caveat: not everyone who writes in this debate assumes the truth (even for the sake of argument) of moral error theory, a complication I deal with shortly. 40  See for example: Joyce (2001), Nolan, Restall, & West (2004), Garner (2007), Kohler & Ridge (2013), Olson (2014: Part 3), Ingram (2015), Lutz (2014), Cuneo & Christy (2011).

160  Dear Prudence true) and those will be my focus here.41 However, before carrying on let me note a terminological awkwardness. Within this debate, some are self-­styled ‘conservationists’ about moral discourse, arguing that we should not do anything about moral dis­ course in the face of the truth of moral error theory. In the absence of a perspicuous term, and for the sake of brevity, I will refer to all views on the ‘what now?’ question as ‘revisionary’ views and the question of what to do with moral discourse, on the assumption of error theory, as a revi­ sionary issue, despite this slightly misleading nomenclature. It is commonly observed that claims about what we should do, given the truth of the error theory, must be handled with care. If moral error theory is true then there is nothing that we morally ought to do about moral discourse (or anything). If the ‘ought’ were a moral ‘ought’, any proposal of this form would be straightforwardly self-­ refuting or contradictory. I will first demonstrate that discussions of revisionary metaethical views frequently make prudential claims within arguments for and against revisionary views before then explaining why such arguments are undermined by the normativity of prudential discourse. If this is correct, many arguments given in favour of revisionary views are equally self-­defeating as moral arguments for and against revisionary views. Finally, I consider possible responses to and limitations of my argument. When arguing for and against revisionary metaethical views, philo­ sophers frequently invoke the prudential. They give prudential justifica­ tions for certain views, make explicit or implicit reference to prudential considerations, describe the revisionary views of others as being con­ cerned with prudential considerations, and generally treat the pruden­ tial as uniquely relevant to adjudicating between the claims of different revisionary theories. To show that this is not an idiosyncrasy of a par­ ticular contributor to the debate, or a mere slip, here is a sample of passages: 41  One proposal—Ingram (2015)—is meant to be compatible with the truth of moral real­ ism. This interesting discussion lies beyond the scope of my argument, which applies only to discussions based on hermeneutic moral error theory. Another interesting outlier is Streumer (2017: §74, esp. 188). For Streumer, the revisionary question is moot because ‘our inability to believe the error theory undermines these revisionary alternatives to the theory.’

Prudential Normativity and Metaethics  161 [Moral discourse has] side-­effects some of which benefit some people at the expense of others, while others do more harm than good to almost everyone[.]42 Mackie did not draw quite the consequences one might have expected from this position [hermeneutic moral error theory]. If a vocabulary embodies an error, then it would be better if it were replaced by one which avoids the error. We could better describe this by saying that our old, infected moral concepts or ways of thought should be replaced by ones that serve our legitimate needs but avoid the mistake.43 The major argument used by Mackie and by Nolan, Restall, and West, is that we should keep the moral overlay because we are better off with it than we are without it.44 First . . . moral thought and discourse—are central for solving collective decision problems in mutually beneficial ways.45 But it is important to remind ourselves that even the eliminativist error theorist will still have plenty of good and strong reasons—many of them self-­interested reasons—for being nice to her fellows.46

It is striking that it is regarded as so utterly uncontroversial within this debate to appeal to prudential considerations and, to my knowledge, no one has argued that these prudential considerations are inadmissible once hermeneutic moral error theory is assumed. There is thus a wide­ spread assumption that prudential discourse and prudential value are non-­normative.47 If, however, prudential discourse is normative, like moral discourse, then appealing to prudential considerations in the dialectical context of hermeneutic error theory is illegitimate. To do so renders revisionism just as self-­undermining as if we treat the ‘should’ in the question ‘what should we do with moral discourse?’ as a moral ‘should’. For if 42  Mackie (1980: 154). 43  Blackburn (1993: 149). Blackburn’s point, and the wider discussion of which it is part, muddies the hermeneutic vs revisionary distinction. But that does not affect my point in using this passage. 44  Garner (2007: 504). 45  Kohler & Ridge (2013: 433). 46  Joyce (2007: 75). 47  Or, more precisely, that they are either non-­normative or normative in some way moral discourse isn’t.

162  Dear Prudence prudential discourse is normative in just the same way as moral dis­ course, and if moral error theory is true, no one has prudential reason to do or to feel anything, there is nothing that we prudentially should do, nothing that is prudentially good or bad for anyone, and so on and so forth. The argument of this section complements the previous section in the following way. If the argument in the previous section was correct then the hermeneutic error theorist has a bigger bullet to bite than previously acknowledged, given that the cost of moral error theory is prudential error theory. The argument in this section is that the hermeneutic error theorist who has a revisionary view also inherits a further problem. This further problem is that they cannot give prudential grounds for revi­ sionary responses to the truth of hermeneutic error theory, given the normativity of prudential discourse. Plausibly, this leaves the revision­ ary project hamstrung, on the assumption that, without prudential grounds to appeal to, there will be insufficient plausible justifications for any revisionary view. Those who engage in debates about revisionary views have a choice.48 They might argue that prudential discourse is not normative in the same way as moral discourse, and is thus able to survive the truth of her­men­ eut­ic moral error theory, making appeals to prudential considerations within this debate wholly legitimate.49 The preceding chapters tried to close off this second option and I have already discussed that option within this chapter, so I will leave it to one side. Alternatively, error theorists (and others engaged in this debate) might accept that prudence is normative in the same way as moral dis­ course, and thus equally vulnerable to the arguments that establish the moral error theory, but simply swear off appealing to prudential consid­ erations when motivating and undermining revisionary views. The cost

48  Reminder: my discussion applies to the large subset of the literature which assumes error theory (even if they only do so for the sake of argument). 49  In later work, Joyce (2007: 52–3) is explicit on this: ‘The moral error theorist usually allows that we can still deliberate about how to act, she thinks that we can still make sense of actions harming or advancing our welfare (and others’ welfare), and thus she thinks that we can continue to make sense of prudential “oughts”.’

Prudential Normativity and Metaethics  163 of this is the unlikelihood of a compelling case for a revisionary form of moral discourse, once prudential grounds are ruled out. There is, technically, a third option. They could argue that their argu­ ments were not prudential to start with. But this response is utterly implausible, given, among other things, the frequency with which explicitly prudential claims are made in this debate (I provided only a small sample above).50 I will thus ignore this possibility. The first option lands the moral error theorist with the task of show­ ing that prudential discourse is not in fact normative in the way that moral discourse is. Thus, if the error theorist has such a reply, then they can use it within that debate and here as well. If, however, they opt to take the other tack—arguing that the revision­ ary project is viable even without appeal to prudential considerations— can this be made to work? It seems difficult to imagine such a story having much force. Even if they offer some hypothesis as to how moral discourse tends to lead, for example, to the fulfilment (or the frustra­ tion) of our desires, or to our being happy (or unhappy), and the like, these factors would seem to lose their significance within a dialectical context in which it is assumed that nothing holds prudential value. Moreover, appeal to prudential considerations in general (‘do this, it’s for your own good’) is rhetorically more powerful than the alternative (‘do this, it will increase your happiness/pleasure/. . .’) precisely because it does not identify specific things which the individual to whom it is addressed may or may not care about.51 These considerations are, pre­ sumably, precisely why all of the arguments in this debate focus exclu­ sively on giving prudential grounds for revisionary views because, once morality is screened off, these are the only powerful justifications remaining. In this section, I showed that the debate about revisionary views in metaethics—attempts to answer the question of what to do with moral discourse if error theory is true—presuppose that prudential discourse can survive the truth of hermeneutic moral error theory (by being nonnormative). Those engaged in this debate must therefore either provide 50  Better to admit you had your hand in the cookie jar than to deny that you have hands. 51  Many thanks to Jessica Isserow for discussion here.

164  Dear Prudence an argument that prudential discourse is not normative like moral discourse or give up their dependence on prudential arguments, instead providing non-­prudential arguments for revisionary views.

5.  Thick Concepts and Prudential Normativity In this, final section, I spell out a final implication of the normativity of prudential discourse. In metaethics, there is a debate between two dif­ ferent views of so-­called ‘thick concepts’.52 Thick concepts are those, like ‘courageous’, ‘cruel’, and the like, which in some way combine evaluation and description (as contrasted with ‘thin’ concepts such as ‘good’ and ‘bad’). For example, to judge that someone is cruel is to evaluate them negatively, but it is more specific than simply holding that they are bad. It tells you particular details about the way that they are bad, or something like that. A long-­running battle in metaethics has been between reductive and non-­reductive views of thick concepts. The full details of that debate are interesting but spelling them out in full will take us too far afield. To put it roughly, proponents of reductive views of thick concepts think that thick concepts are a combination of distinct evaluative (normative) con­ tent and non-­ evaluative (non-­ normative) content. Non-­ reductivists deny this claim of separability. One way in which the dialectic proceeds is by the proponents of reductive views offering candidate proposed reductions of thick concepts into normative and non-­normative compo­ nents, as proof of concept (excuse the pun). How does the normativity of prudential judgements relate to this? Notably, when offering such reductions of thick concepts, defenders of reductive views frequently include prudential content within the pur­ portedly non-­normative part of the concept. Here is a representative example; Daniel Elstein & Tom Hurka, in the process of defending the reductive programme in general, offer the following analysis of ‘courage’:

52  For an overview of the issues see Roberts (2017a). For recent discussions, see Kirchin (2017), Väyrynen (2013), Eklund (2017).

Prudential Normativity and Metaethics  165 ‘[A]ct x is courageous’ can be analyzed as something like ‘x is good, and x involves an agent’s accepting harm or the risk of harm for him­ self for the sake of goods greater than the evil of that harm, where this property makes any act that has it good’, and where, again, the second ‘good’ is an embedded evaluation.53

We thus find prudential content—‘harm or the risk of harm’—within the officially non-­normative part of the analysis. Thus, if prudential dis­ course is normative, this reduction has not analysed ‘courage’ into sep­ ar­ate normative and non-­normative components, as they claim. Rather, it has only separated out different types of normative component, one of which is prudential.54 This is, of course, just one putative analysis. But the problem posed for the reductive view of thick concepts by the normativity of prudential discourse is more general. If one wants to offer reductive views of thick concepts one must either (i) provide a plausible case that prudential properties are non-­normative, so legitimately includable, OR (ii) eschew reference to prudential content such as ‘harm’, ‘well-­being’ etc. in one’s reductive analysis. The first option has been considered and rejected multiple times in the preceding discussion so I won’t rehash those points here. The diffi­ culty with the second option is that any non-­normative counterpart to (e.g.) harm, such as ‘reduced happiness’, which the reductivist might try to use instead will make for implausible analyses by being implausibly specific. Unless one is using ‘happiness’ in a very thin way, such that it is equivalent to well-­being, if we substitute ‘reduced happiness’ for ‘harm’, the analysis of courage offered above is implausible. After all, it does not seem to be plausibly a necessary truth, let alone a conceptual one, that courage involves risking lowering one’s level of happiness in particular. The same will apply, I suggest, for other attempted reductions of thick

53  Elstein & Hurka (2009: 527). I take the ‘something like’ qualifier to be spelled out later (p. 531) thus: ‘We are not committed to all the details of the above analyses, which are only sketches intended to illustrate the general resources the reductive view has.’ It seems plausible to me that the point I am making here—that the prudential is being straightforwardly included within the non-­descriptive—does not fall foul of this qualifier. 54  For broader discussion of this point see Roberts (2017b: 209), (2018: 14).

166  Dear Prudence concepts. For good reason, such candidate reductions make abundant appeal to prudential content because thin prudential terms like ‘harm’, ‘well-­being’, etc. are necessary to make the analyses plausible. But in doing so they undermine their professed goal to be offering reductions of thick concepts into normative and non-­normative parts. In short, the normativity of prudential properties makes harder the task of those defending a reductive view of thick concepts.

6.  Conclusion Considered together, these issues reveal the strangely ambivalent man­ ner in which philosophers treat prudential value and prudential dis­ course. In some modes, such as when thinking about the why be moral question, it is treated without argument as intrinsically authoritative. By stark contrast, in the context of thinking about moral error theory and thick concepts, it is treated as wholly, and obviously, non-­normative. If prudential discourse is normative then the second of these is, of course, mistaken. But even if it is not, we have still found an inconsist­ ency in how we think about prudence, between treating it as in­trin­sic­ al­ly authoritative and treating it as immune to normative error theory.55

55  For what it is worth, I do not think that the applications of the normativity of prudence considered here exhaust its meta-­normative significance.

7 Prudential Normativity Realism and Anti-­Realism

1.  Introduction In earlier chapters, I argued that prudential discourse is normative, like moral discourse. If that is true, then we should expect to be able to take resources from metaethics and fruitfully apply them to meta-­prudential questions and vice versa. This chapter tries to do precisely that. In particular, it aims to show how various long-­standing debates about prudential value can be enhanced by importing ideas from metaethics and how bringing these two lines of inquiry together generates a number of significant benefits both inside and outside of theorizing about prudential normativity. I begin (section 2) by introducing one of the major divisions within moral philosophy—roughly (for the moment) that between (i) the normative grounds of moral facts and (ii) the nature of moral facts— before (section 3) showing how this general distinction applies to the prudential domain. I then turn to two debates about prudential value and show how these can be either reinterpreted as, or replaced by, meta-­prudential debates about the nature of prudential value. The first of these (section 4) is the debate over the anti-­alienation constraint on prudential value. The second (section 5) is the attempt to distinguish between subjective and objective theories of well-­ being. Finally, (­section 6), I explain the philosophical pay-­offs from reinterpreting these debates as meta-­prudential questions about the nature of prudential value.

Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0008

168  Dear Prudence

2.  First and Second-­Order Questions about Morality Here is a moral fact: . The history of moral philosophy has helpfully distinguished two kinds of question that we can ask about this fact. The first is a question about the grounds of the fact. The second question is about the nature of this fact. Let me make these more precise.

2.1  First-­Order Question What makes it true that it is wrong for Donald to refuse to rent out his property to black people? Why is that action wrong? Moral philosophy has distinguished various candidates with the main contrast being between consequentialist answers—the action is wrong because it fails to maximize value—and non-­consequentialist answers, which reject the claim that facts about value alone explain facts about rightness, substituting some other explanation (such as that the behaviour is wrong because of the interplay of prima facie duties).1 I will call these different accounts of the grounds of wrongness, of what makes some act (etc.) wrong.

2.2  Second-­Order Question It is wrong for Donald to refuse to rent out his property to black people. But what is the nature of this fact, or wrongness facts more generally? In property talk, what sort of property is wrongness? Here we find competing accounts, worked out in great detail within contemporary metaethics.2 One division is between: views which treat 1 Paradigmatic consequentialist answers are those given Mill (1861), Moore (1903). Paradigmatic non-­consequentialist answers are given by: Kant (1785), Ross (1930), Scanlon (1998). 2  One complication here: people of different metaethical stripes will characterize this question differently. At least some metaethicists would not be happy with taking this use of ‘properties’ and ‘facts’ at face value. It is compatible with what I say to interpret these claims in a deflationary way. Further, some metaethical views might be understood as trying to collapse, or as committed to collapsing, the distinction I am introducing here. That is OK for my purposes and would be akin to kicking away the ladder after one has used it to climb.

Prudential Normativity: Realism and Anti-Realism  169 moral facts as natural facts and views which treat moral facts as nonnatural facts.3 Another dividing line, within answers to the second-­order question, runs between realist views and anti-­realist views. According to realist views is a fact that is independent, in the relevant way, from the attitudes of actual (or possible) agents.4 The general idea is that moral facts (or properties) are not identical to, constituted by, nor dependent upon, what agents think or feel. On the realist view, moral facts are, in that respect, like facts such as and . By contrast, according to anti-­realist views, is a fact that is identical to, constituted by, or relevantly dependent upon, the attitudes of some actual or possible agent. In this respect, on the anti-­realist view, moral facts and properties are like facts about what is fashionable or cool (or the properties of being fashionable or being cool) in that there isn’t a set of facts about what is cool that is out there waiting for us to discover, independent of human responses. Rather, for something to be cool just is a matter of how humans, or some relevant subset, feel about that thing. The simplest versions of moral anti-­realism would be the view that moral facts are identical to, constituted by, or relevantly dependent upon, the judger’s own attitudes. Put in property talk, on an anti-­realist view, wrongness is an attitude-dependent property.5 Applied to our 3  Paradigms of the former are Boyd (1988), Brink (1986), Sturgeon (1985), Railton (2003), and Foot (2000). Paradigms of the latter are Moore (1903), Ross (1930), Cuneo, (2007) Enoch (2011a). 4 For general discussion of realism and the attitude-­independence constraint see Joyce (2016) and Finlay (2007b). Realism is sometimes characterized in a way that includes the existence claim that such facts/entities exist and are independent whereas other times it is characterized using only the attitude-­independence constraint (if such facts exist (existed) they are (would be) independent), where, e.g., a moral error theorist counts as a realist. I am using the former, more common way, of understanding realism to enable me to talk in terms of moral facts. But what I say is translatable into other preferred metaethical idiolects (see footnote 2). 5  Trying to spell out the attitude-­independence constraint is hard because moral facts seem to be sensitive to agents’ attitudes in some ways that no realist should ever want to deny. For example, if you find yourself with certain kinds of desires, you’re morally required to seek help (example inspired by one of Dancy’s (2002: 125)). The customary realist way to handle such cases is to treat, e.g., problematic attitudes as merely capable of triggering moral requirements which are, themselves, attitude-­independent. For discussion of these issues in connection to prudential reasons, see Sobel (2019) and for helpful discussion of the same issues in the aesthetic domain see Hanson (2018: §3).

170  Dear Prudence example moral fact, our basic form of anti-­realism would hold that the fact is identical to, constituted by, or relevantly dependent upon Donald’s (or some relevant counterpart’s) aversion to his not renting out his property to black people. This division between realism and anti-­realism is quite simple, and there are a great many differences between representatives of each camp.6 But even this division is sufficient for my broader purposes here. Let me combine this with the earlier distinction, to map out the relevant space of views when it comes to second-­order questions in the case of morality: Realism: moral facts (/properties) are attitudeindependent.

Anti-Realism: moral facts (/properties) are attitudedependent.

Naturalist Realism: moral facts (/properties) are attitudeindependent and natural.

Naturalist Anti-Realism: moral facts (/properties) are attitudedependent and natural.

Non-Naturalist Realism: moral facts (/properties) are attitudeindependent and non-natural.

Having explained the distinction between first and second-­ order questions with respect to morality, I want to now apply these ideas to prudential normativity to show how analogous questions arise there.

3.  First and Second-­Order Questions about Prudential Normativity Take this case: it is beneficial for Michelle to read a book. We now have an evaluative prudential fact about which we can ask two questions.7 First question: why is reading a book good for Michelle (what makes it good for her)? Second question: 6  Along with disputes about how to characterize each camp. 7  Henceforth, I mostly omit ‘evaluative’ and just talk in terms of prudential facts/properties but, unless otherwise specified, I am always referring to evaluative prudential facts/properties in particular.

Prudential Normativity: Realism and Anti-Realism  171 what is the nature of this fact? Is it an attitude-­dependent fact or an attitude-­ independent one? For example, is it, simply, a fact about Michelle’s own attitudes? In property talk, is the property of being good for someone identical to some attitudinal property? To the extent that these questions are distinguished, contemporary work on prudential value has been mostly organized around the first of these issues. Familiarly, there are hedonistic theories, desire theories, as well as perfectionist theories and objective list theories, which offer competing explanations of prudential value. These theories offer different explanations of why Michelle’s reading a book has the property of being good for her. Hedonism says that it is good for her by being an instance of pleasure, perfectionism says that it is good for her by being the exercise and fulfilment of her human capacities, and so on. By stark contrast, the distinction between first and second-­order questions is hardly ever made, and is perhaps conflated, and there has been little official attention given to the second-­order question of the nature of evalu­ ative prudential facts and properties (or what kind of fact it is, that reading the book is good for her).8 Nevertheless, we can see how the distinction drawn above between realism and anti-­realism applies here also. One could accept realism about prudential facts, holding that the fact that Michelle’s experience is good for her is relevantly attitude-­independent. For example, you could hold a perfectionist first-­order theory of well-­being and hold that what makes something good for someone is its being the exercise and fulfilment of their essential human capacities. You would then combine that with the view that the property of being good for someone is some further non-­attitudinal property.9 8  Some exceptions are Hooker (1991), Rosati (1996,  2020), Darwall (2004). Rosati and Darwall both explicitly take up the question of the nature of prudential facts, or what it is for something to be good for someone. For example, Rosati (2020) defends the view that the relation of being good for someone is (p. 242) ‘a reason-­giving relation of fit between a welfare object and a welfare subject, where a welfare subject is a valuable being’. As I read this, it is slightly orthogonal to the issues I am examining in this chapter because different accounts of what ‘fit’ consists in will yield realist, anti-­realist, naturalist, non-­naturalist versions of this proposal. 9  Another example in case it would be useful at this point: you could hold a hedonistic theory of well-­being, holding that only pleasure is good for people, so what makes something non-­instrumentally good for someone is, simply, its being an instance of pleasure. And you

172  Dear Prudence Alternatively, one could hold that prudential facts should be construed in a anti-­realist way, holding that the fact that Michelle’s experience is good for her is a fact that is relevantly dependent upon the attitudes of some agent (Michelle’s or someone else’s).10 We see that there is a real difference between these questions and that it is equally possible to distinguish them in the prudential realm as in the moral realm. However, I expect that at least some will be sceptical of the value of doing so, perhaps asking: if it is worthwhile to distinguish these questions, why have they not been clearly distinguished, especially in comparison to work on morality?11 Partly to combat this kind of scepticism but also to make the issues clearer, let me offer an explanation for why I think that these issues have been less clearly distinguished within debates about prudential value than in the case of morality. My hypothesis is that this partly stems from the fact that, in the prudential case, second-­order views are closely related to first-­order views, in a way that prevents us easily seeing their separability in principle. More precisely, some second-­order views either entail first-­order views or at least form natural combinations (by which I mean, whether or not one view really does entail the other, it is at least plausible that it does). Thus, to a greater extent than in the case of morality, answers to these questions plausibly come, or might reasonably be thought to come, as a package deal. Here is an example of such a plausible package deal. Suppose that, when pressed on the second-order question of the nature of prudential facts, one held that prudential facts just were facts about pleasurable experience (a hedonic property identity view). Call this: Hedonism (2nd order):  Evaluative prudential facts/properties are identical to hedonic facts/properties. (E.g. = < reading a book is pleasurable for Michelle>.)

could combine that with the view that the property of being good for someone is some further, non-­attitudinal, property. 10  Arneson (1999) is one instance of someone explicitly discussing prudential realism and presents objective theories as realist. Many thanks to Dale Dorsey for a reminder of this. 11  For example, in the case of morality, we have different subdisciplines: normative ethics and metaethics.

Prudential Normativity: Realism and Anti-Realism  173 If one held this second-­order view, which first-­order views would be on the table? It seems plausible that only one would remain: Hedonism. This is because regular, first-­ order, Hedonism would plausibly be thought to follow from Hedonism (2nd order).12 A hedonistic theory on the meta-prudential question of the nature of prudential facts plausibly entails a hedonistic view on the first-­order question of the grounds of prudential facts. Similarly, suppose that one held the view that the property of being good for someone just was the property of fulfilling their desire (a desire-­fulfilment property identity view):13 Desire-­Fulfilment (2nd order):  Evaluative prudential facts/properties are identical to desire-­fulfilment facts/properties. (E.g. = .) What first-­order views would be on the table, on the assumption of this view? It is plausible to think that only one would remain: regular ­first-­order desire-­fulfilment theory. If evaluative prudential facts are identical to facts about desire-­fulfilment then it seems unavoidable that things hold prudential value just in case they fulfil the agent’s desires. Thus, if one held a desire-­fulfilment view on the meta-prudential question of the nature of prudential facts, one would likely feel committed to a ­desire-­fulfilment view on the first-­order question of the grounds of prudential facts. So far I have given two examples in which second-­order views either entail, or would reasonably be thought to entail, the equivalent ­first-­order views, to demonstrate the idea that these things are liable to be implicitly treated as package deals (which I think explains, in part, why these questions have gone underexplored). Here, now, is another case where we get (at least) a plausible connection but this time going from first-­order view to second-­order view. 12  Again, to emphasize, I am not asserting that the one view really does entail the other, only that this claim is plausible. 13  Darwall (2004: ch. 2), one of the few to discuss second-­order views of prudential value, discusses exactly this possibility, of a desire-­fulfilment second-­order view.

174  Dear Prudence Suppose that on the first-­ order question, one held the simple desire-­ ­ fulfilment theory such that things have prudential value for ­people because they fulfil their desires. That would be to hold: DFT (1st order):  Something is good for a welfare subject if and only if and because it fulfils a desire of theirs. (E.g. reading a book is good for Michelle if and only if and because reading a book fulfils one of Michelle’s desires.) Suppose one were then prompted on the second-­order question—the nature of the fact that reading a book is good for Michelle. It would be natural, presumably, to combine the first-­order desire view with a simple anti-­realist view, holding that the fact that reading a book is good for Michelle is a fact about Michelle’s actual attitudes.14 In the case of desire views (and hedonistic views) we can see how holding one of these views on one of these questions at least makes it plausible to hold the equivalent view on the other question. This demonstrates the way that these things are liable to be treated as a package deal. And it is that which, in part, explains why these things are less often distinguished in the case of prudential facts than for moral facts. It is important to note though that first-­order views and second-­order views are variously combinable. For example, suppose one held a particular pluralist objective list theory of prudential value, such as this one (just for illustration). There are three basic good types—pleasure, knowledge, and achievement—and everything has non-­instrumental prudential value simply in virtue of being a token of at least one of these types.15 Applied to the case of Michelle, such a view would give this answer to the first question:

14  This might only hold on one reading of the DFT view—the one on which it is the object of the desire that is good for the welfare subject, rather than the combination [desire + its satisfaction]. For discussion of these two forms of desire view see e.g. Van Weelden (2019). I always here understand the DFT view as the former, object-­of-­desire, view. Things get even more complicated when we consider more complex forms of desire theory. 15  I borrow this helpful way of putting the view from Lin (2016b).

Prudential Normativity: Realism and Anti-Realism  175 OLT (1st order):  Something is good for someone if and only if and because it is a token of at least one of the basic goods: pleasure, knowledge, achievement. (E.g. reading a book is good for Michelle if and only if and because it is a token of pleasure, knowledge, achievement.16) This first-­order commitment could be combined with either realist or anti-­realist theories of the nature of prudential facts. In schematic form, these proposals would be: OLT + Realism:  if and only if and because it is a token of one of the three basic goods. OLT + Anti-­ Realism:  if and only if and because it is a token of one of the three basic goods. Different realist accounts and anti-­realist accounts will supply different candidates for R and A. One anti-­realist view would be that property A just is the property of being what an idealized agent would desire under some relevant set of conditions, such as the condition of full information.17 One realist view would be that property R is some biological property. Finally, an alternative realist view is that R is some sui generis normative property.18 In a similar vein, it is clear that if one started off with some answers to the second-­order question this would not commit one to any particular answer on the first-­order question. One could start off by embracing some form of Platonistic realism and go on to defend various first-­order theories (hedonism, perfectionism, or the like). Conversely, if one started off with the anti-­realist view that prudential facts just are facts 16  For the sake of readability, I’m leaving out complications such as being instrumentally vs non-­instrumentally good for the subject. 17  As Rosati (1996: 298) points out in connection with the alienation constraint, ‘full information’ here must exclude information about prudential value, if the view is not to collapse into a form of realism. 18  A further alternative is that it is partly constituted by some normative property such as the property of being generally valuable. One instance of such a view is my (2012) defence of a Locative Analysis of prudential value.

176  Dear Prudence about the desires of some idealized agent, then this can presumably be combined with various first-­order views.19 If this is right then we have uncovered two things. First, that, at the very least, one kind of first-­order view—objective list theory—does not commit one at all on the second-­ order, meta-­ prudential question. Second, there are second-­order theories which do not commit us to some particular first-­order theory.20 This, hopefully, dispels scepticism as to whether these questions are really distinguishable. Of course, it is one thing to point out a possibility, quite another to show that it is significant. But it is interesting and important that there has been so little official inquiry into the nature of prudential facts, as contrasted with the facts that ground them or generate prudential value. And it is worthwhile to investigate the nature of the facts ascribed within prudential discourse. Should we adopt realism about prudential value or anti-­realism?21 Furthermore, my hunch is that whilst there has been little official attention given to second-­order questions about prudential value, these questions have in fact been lurking in the background of various debates. Here is one example. A common objection to objective theories of ­well-­being is that they are problematically arbitrary or explanatorily unsatisfying in some way. For example, Bradley writes: [P]luralism seems objectionably arbitrary. Whatever the composition of the list, we can always ask: why should these things be on the list? What do they have in common? What is the rational principle that yields the results that these things, and no others, are the things that are good?22

By itself, this can seem like an unfair objection to objectivist pluralist theories and I argued previously:

19  Reminder: the idealization cannot include access to pre-­existing prudential facts, without ceasing to be realist. 20 To emphasize, I think there are multiple coherent combinations of first-­order and second-­order views. 21  Or neither, by rejecting descriptivism? 22  Bradley (2009: 16).

Prudential Normativity: Realism and Anti-Realism  177 [T]he objective list theory is no more burdened by these challenges than any other theory of well-­being. We can ask: ‘why is pleasure (or knowledge or . . .) alone of prudential value?’ or ‘what is the rational principle that determines that pleasure (or knowledge or . . .) contributes to well-­ being?’ The same goes for desire-­ fulfillment theory. Desire-­ fulfillment theorists spend little or no time providing an ex­plan­ation of why desire fulfillment contributes to well-­being.23

and was happy to think that this objection to objective list theories was simply an example of philosophers being better at finding faults (even the very same fault!) in theories other than their own. Things look different, though, once we notice the close connection between hedonistic and desire-­ based first-­ order theories of well-­ being and their ­second-­order counterparts. Perhaps this challenge is much less frequently (i.e. hardly ever) made against hedonism and desire-­fulfilment theories for good reason—namely that it is so plausible that these the­or­ ies would be combined with the analogous second-­order views. That is, if you were a hedonist, you would be likely to accept a hedonic identity theory of being good for someone whereby this property just is the property of being pleasurable. For such a hedonist the explanatory demand that Bradley is posing to the objective list theory would be either moot or very easily satisfied. On such a possibility, hedonism would be the true first-­order theory of well-­being because of the nature of prudential properties themselves.24 I do not offer this as a claim about what is going on ‘inside the heads’ of defenders of hedonism and desire-­fulfilment theory but rather as an interpretation of the general debate about first-­ order theories of ­well-­being, one that makes good sense of why this explanatory objection is frequently levelled against objective list theories of well-­being but almost never against hedonism and desire-­fulfilment theories. Let me now move on to another example where second-­order questions about prudential value have been lurking in the background of a

23  Fletcher (2016b: 154). 24  This specific possibility is noted in Heathwood (2013: 7) and Lin (2020: 39) and briefly discussed in Lin (2016a).

178  Dear Prudence debate: the ‘alienation constraint’ on prudential value. To make my claim more precise, I think that as prudential theorizing shifted away from the straightforwardly first-­ order debate between hedonism vs desire-­fulfilment theory vs objective list theory (etc.) and into thinking about the alienation constraint on prudential value it did so because this debate serves as a proxy war for the second-­order debate between realism and anti-­realism about prudential value. I will provide some evidence for this interpretation in the next section. I will also, subsequently, explain why it would actually be better if we focused fully explicitly on the debate between realism and anti-­realism about prudential value and decoupled it from the debate about the alienation constraint. Doing so enables the real choice points to better be seen and the real costs and benefits of each option, and set of options, to be properly assessed, by importing the useful advances from these debates in metaethics. Focusing explicitly on the choice between realism and anti-­realism about prudential value would also be advantageous in helping us to assess the thesis that there is one true meta-normative theory (and in seeing which philosophers subscribe to it).

4.  Anti-­Alienation and Anti-­Realism Theorizing about prudential value has increasingly shifted in recent decades away from straightforward first-­order debates between hedonism and its opponents and onto the question of whether there is an ­anti-­alienation constraint on prudential value.25 Many have argued that it is a necessary condition of prudential value that what is good for an agent must in some way be subject to their endorsement (a desire or some other kind of attitude), at least under some conditions. The classic articulation of the motivation behind the constraint comes from Peter Railton thus: It does seem to me to capture an important feature of the concept of intrinsic value to say that what is intrinsically valuable for a person 25  Rosati (1996), Sarch (2011). We also saw this issue briefly in Chapter 5.

Prudential Normativity: Realism and Anti-Realism  179 must have a connection with what he would find in some degree ­compelling or attractive, as least if he were rational and aware. It would be an intolerably alienated conception of someone’s good to imagine that it might fail in any such way to engage him.26

The debate over the putative constraint is one of the major fault lines in contemporary work on well-­being. It is important to note that the alien­ ation constraint is only held to be a necessary condition for prudential value. This allows it to function as part of a (non-­question-­begging) argument in favour of some first-­ order theories of well-­ being and against others. Thus the constraint is used as an argument for the broad class of subjective theories and against objective theories of well-­being (which specifically reject this necessary condition).27 Most discussion of the alienation constraint has gone into seeking either to defend it, to reject it, or to show how it (or the intuitions behind it) can be accommodated by objective theories. By contrast, and as even those sympathetic to the constraint have noted, little attention has gone into answering the question of why there would be such a constraint on prudential value.28 As Rosati puts it: Given the widespread tendency to defend accounts of [prudential] value in a way that presupposes internalism, it is surprising how little direct defense has been given for the thesis that a plausible account of a person’s good must satisfy internalism. Those defenses that have been offered have rarely received adequate development.29

As I said above, my hunch is that philosophers working on prudential value started discussing the anti-­ alienation constraint in large part because it functions as a proxy war for the debate between realism and anti-­realism about prudential value. Given that philosophers have not often supplied arguments for or explanations of the truth of this 26  Railton (2003: 47). For precise discussion of the constraint see e.g. Rosati (1996), Fletcher (2016a: appendix), Dorsey (2017). 27  For discussion and justification for this way of formulating objective theories, see Fletcher (2016b). 28  But see Dorsey (2017) for an exception to this. 29  Rosati (1996: 299).

180  Dear Prudence alienation constraint, it is hard to conclusively demonstrate this. I will make my case using the two most sophisticated defences of the alien­ ation constraint, from Wayne Sumner and Connie Rosati. In each case there is an explicit commitment to anti-­realism about prudential value behind their endorsement of the alienation constraint.

4.1  Sumner and Rosati’s Arguments for the Subject-­Relativity of Prudential Value Here are some claims made by Wayne Sumner in his discussion of the ‘subject-­relativity’ or ‘subjectivity’ of prudential value (as will become clear, any theory that has an alienation constraint is subjective, as Sumner carves things up). He distinguishes between subjective and objective theories in terms of whether they hold that prudential value depends (somehow) upon attitudes: [A] theory treats welfare as subjective if it makes it depend, at least in part, on some (actual or hypothetical) attitude on the part of the welfare subject. [. . .] Subjective theories make our well-­being logically dependent on our attitudes of favour and disfavour. Objective theories deny this dependency.30

He then argues that subjectivity is a necessary condition for an adequate theory of well-­being, because of the perspectival character of prudential value: [S]ubjectivity turns out to be a necessary condition of success in a theory of welfare. If I am right then objective theories can be ruled out of consideration as a category, all of them inadequate precisely because they are objective.31

30  Sumner (1996: 38).

31  Sumner (1996: 27).

Prudential Normativity: Realism and Anti-Realism  181 What subjectivity and the perspectival nature of prudential value amount to is made fully explicit thus: To claim that welfare is subjective is to claim that it is mind-­dependent— that and nothing more. [. . .] Prudential value is therefore perspectival because it literally takes the point of view of the subject. Welfare is subject-­relative because it is subjective.32

Thus we can translate Sumner’s line of thought into the idea that anti-­realism is true of prudential value and it is this which delivers the anti-­alienation constraint. Connie Rosati is similarly explicit in arguing for the alienation constraint on the basis of meta-­prudential anti-­realism. We see this in her deployment of the following argument: A second argument for internalism about a person’s good begins with an argument for subjectivism, according to which value exists only in virtue of subjectivity . . . It is a natural step from this thought to the thought that value itself must be a complex motivational property . . . If value can exist only if there are creatures who can be affected by and react to their world, then value, and more specifically, goodness for a person, must be a motivational property . . . The only alternative might seem to be that the property of being good for a person is a Moorean, nonnatural property, but this alternative introduces special meta­phys­ ic­al and epistemological problems . . . Considerations about the metaphysics of value show that a plausible account of a person’s good must satisfy existence internalism.33

Here Rosati clearly expresses support for anti-­realism about prudential value and explicit rejection of realism about prudential value (given that Moorean non-­naturalism is a paradigm of realism). To forestall misunderstanding let me note two points. First, nothing that I have argued in this section constitutes any objection whatsoever

32  Sumner (1996: 43).

33  Rosati (1996: 313).

182  Dear Prudence to the alienation constraint or to anti-­realism (or the arguments from the alienation constraint to anti-­realism). Rather, I have simply been trying to make maximally clear that these writers are explicitly arguing from an anti-­realist second-­order view about prudential value to the anti-­alienation constraint. Second, Sumner and Rosati are crystal clear in arguing that the alien­ ation constraint follows from the truth of anti-­realism about prudential value. I do not claim that there is anything unclear in their discussions and I hope they would accept my characterization of what they are doing. But what is striking is that these meta-­prudential arguments for the anti-­alienation constraint, and general expressions of support for anti-­realism about prudential value, have not been presented in these terms and have not engendered more official focus upon the ­meta-­prudential issues. Many subsequent discussions of the alienation constraint make no mention of its plausibly being the corollary of ­anti-­realism on the second-­order question of the nature of prudential value. Instead it is often presented as a standalone constraint.34 Thus, it has continued to be true that, as Rosati herself claimed: ‘[T]hose who accept this thesis [the alienation constraint] have tended at best only implicitly to recognize the kind of connection to motivation that the central intuition behind internalism requires.’35 I hope to have provided some evidence that the debate over the ­anti-­alienation constraint on prudential value can be interpreted as serving as a proxy war for the debate over anti-­realism and realism about prudential value.36 I now turn to a distinct, but closely related, debate in work on prudential value where I think we can offer a similar

34  For example Hawkins (2014: section 5). Dorsey (2017: 704), by contrast, argues for the constraint, and he does so with an argument that he specifically claims is meta-­prudentially neutral: ‘The relationship to value argument [Dorsey’s argument for the internalism constraint] does not rely on any particular metaethical claim, nor need it imply the non-­existence of attitude-­independent facts about value.’ 35  Rosati (1996: 299). 36  Confession: One piece of data which doesn’t obviously fit well with this general story is the class of hybrid theories which tend to supplement an otherwise objective theory with an anti-­alienation-­satisfying extra element. Still, I take the minority status of these views (both in terms of amount of exploration of them but also how few philosophers have explicitly defended them) to fit with my overall story of alienation serving as a choice point between those with realist leanings and those of an anti-­realist bent.

Prudential Normativity: Realism and Anti-Realism  183 diagnosis: the attempt to distinguish between, and categorize, subjective and objective theories of well-­being.

5.  Subjective Theories and Anti-­Realism Much energy has gone into the attempt to divide first-­order theories of well-­being into those which are subjective and those which are ob­ject­ ive. Without endorsing it,37 Bradley gives this representative example of such a distinction: Subjectivism about well-­being (version 1): An individual’s well-­being is determined entirely by that individual’s own subjective psy­cho­ logic­al states. Objectivism about well-­being (version 1): An individual’s well-­being depends at least in part on something other than that individual’s own subjective psychological states.

Chris Heathwood provides a similar distinction thus: Subjectivists maintain that something can benefit a person only if he wants it, likes it, cares about it, or it otherwise connects up in some important way with some positive attitude of his. Objectivists deny this, holding that at least some of the things that make our lives better do so independently of our particular interests, likes, and cares.38

People have made various suggestions for how to distinguish between subjective and objective theories. I think, however, that we should give up trying to divide up theories of well-­being in this way. The trouble comes from the fact that these (types of) distinctions deploy claims 37  Bradley (2014: 221). 38  Heathwood (2014: 202). Eagle-­eyed readers might note that Bradley and Heathwood’s criteria for subjectivism might come apart in the case of classifying hedonism. In the chapter that this derives from, Heathwood discusses whether hedonism counts as a subjective view, settling on the view that ‘whether hedonism qualifies as an objective or a subjective theory depends on which general approach to the nature of pleasure is correct[.]’

184  Dear Prudence about dependence (and related ideas) rather than something more precise. Typically, they will say that, according to subjective theories of well-­being, well-­being depends upon subjective states. Yet, as others have observed,39 and as we have already seen above, there are different ways of understanding the claim that prudential value depends upon sub­ject­ ive states. One such way is to hold that the bearers of prudential value are subjective states (e.g. states of consciousness are the bearers of prudential value). Another is to think that something’s status as a bearer of prudential value only holds on condition of the presence of subjective states in some relevant agent (e.g. something is valuable on condition that an agent desires that thing or takes pleasure in it). And there may be others. My positive suggestion is that we take seriously the idea of ditching this distinction between theories and instead ask the questions that we distinguished above, analogues of the questions that have been asked about morality. For example, for some putative instance of prudential value—Michelle’s reading a book—we can ask: 1. Why Michelle’s reading a book has prudential value (where the answer will be: because it is pleasurable or because it fulfils her desire or [. . .]). 2. The nature of the fact that Michelle’s reading a book has prudential value (where the answer will be: Michelle’s reading a book has realist property R or anti-­realist property A). This would constitute an improvement over the subjective vs objective distinction by focusing attention on the important issue it was, plaus­ ibly, attempting to track. Let me explain the line of thought here. Those who have been invested in trying to differentiate subjective theories and objective theories must have thought that the distinction is important. After all, we can obviously divide up theories in lots of trivial ways (alphabetically, chronologically etc.). The subjective/objective distinction was either a distinction among the prudential value makers (answers to question 1) or between different theories of the nature of 39  Bradley (2014: 221).

Prudential Normativity: Realism and Anti-Realism  185 prudential value (answer to question 2). But question 2 is the more important, fundamental, issue here.40 Thus, it is charitable to interpret the attempt to divide up theories into subjective and objective as being concerned with the nature of prudential value. Some support for this interpretation comes from Bradley, who writes: It might be that philosophers—subjectivists, in particular—are led to believe that there is an important distinction between objective and subjective theories of well-­being for reasons having little to do with judgments about what sorts of lives are valuable. They might rather be motivated by metaphysical and epistemological concerns. A likely motivation for being a subjectivist about well-­being is a commitment to a naturalistic metaphysical worldview that does not allow for an irreducible, non-­natural property of goodness for an individual (this may be in part because of epistemological worries about such properties). If you are suspicious of such properties, but you believe that some things are good for you and some things are bad for you, it is natural to think that all this amounts to is that you want or like some things and do not want or like others.41

I think Bradley’s diagnosis is exactly right. The attempt to divide up theories into subjective and objective is better understood as attempting to draw second-­order distinctions between realist vs anti-­realist (and Naturalist vs Non-­Naturalist) theories of the nature of prudential facts, not a distinction between prudential value-­makers.42

40  Notice that people have not sought to divide first-­order moral theories into subjective/ objective in a way analogous to dividing up theories of well-­being. In the moral case, talk of the subjectivity/objectivity of morality has long been interpreted, correctly, as about the nature of moral properties and judgements. 41  Bradley (2014: 236). Similar ideas are expressed by Lauinger (2012: 129). 42  Lin (2017: 355) is a commendably clear exception, writing: ‘I understand subjectivism to be a view in normative ethics, not metaethics. It concerns which things have the property of basic goodness for you (and why), not what this property is. Thus, it is neutral on whether this property is irreducibly normative, and on whether it is natural. Two non-­naturalists could disagree about whether something has the non-­natural property of basic goodness for you if and only if you have A toward the thing in C.’ However, Lin writes this in the process of rejecting subjectivism, so the worry that we have a constraint with no explanation does not apply.

186  Dear Prudence I have argued, first, that we should think of the attempts to distinguish between subjective and objective theories of well-­being as tracking the distinction between realism and anti-­realism about prudential value and, second, that we can ditch the traditional subjective/objective distinction and replace it with the questions and theories that I distinguished above. Still, one might reply that the point of doing so is still unclear. That is, one might grant that we could do this but wonder why we should. Let me explain, then, why I think it would be beneficial to focus more explicitly on second-­order prudential questions.

6.  Why It Would Be Better to Focus on Second-­Order Prudential Questions In the preceding sections, I argued that there are two long-­standing debates about prudential value where it is plausible that the underlying issue was, really, the distinction between realism and anti-­realism about prudential value, such that we could refocus those debates more ex­pli­ cit­ly on that issue. I also suggested that it would be better if we focused explicitly towards that issue, rather than these proxy issues. I want to now outline some of the advantages of doing so. A first advantage of explicitly focusing on the choice between realism and anti-­realism about prudential value is that we could immediately import a lot of ideas from metaethics, to help us understand the various options and the costs and benefits of each. The debates about the alien­ ation constraint and the subjective/objective distinction are both worked out in less detail than the equivalent positions and debates in metaethics. We would thus gain a more sophisticated understanding of the relevant issues by adopting this common framework for thinking about the important issues. By theorizing about realism and anti-­realism we avoid the reduplication of effort and error and will more easily be able to see how these debates about prudential value feed into, and are affected by, broader debates within and beyond the meta-­normative. Second, focusing on the choice between realism and anti-­realism will enable us to also simultaneously pursue the question of the nature of prudential discourse and prudential value facts alongside the question

Prudential Normativity: Realism and Anti-Realism  187 of whether there is one true meta-­normative theory (or at least one theory that is true of moral discourse and prudential discourse). Let me call this thesis: One-­Theory-­To-­Rule-­Them-­All:  There is one true meta-­theory for all forms of normative discourse. For example, suppose we find some knockdown argument against prudential realism. This entails that either moral realism is false or that One-­Theory-­To-­Rule-­Them-­All is false. Conversely, we might find that One-­Theory-­To-­Rule-­Them-­All is so plausible that it serves to make prudential realism seem more plausible (at least to those sympathetic to moral realism). All of this is much more straightforward if people explicitly defend anti-­realism about prudential value rather than, for example, the claim that there is an anti-­alienation constraint on prudential value. A third reason for focusing on meta-­prudential debates like the choice between realism and anti-­realism, rather than these more parochial debates (or terms of debate), is that it immediately raises certain questions that otherwise go ignored. One such question, which is almost never asked in the case of prudential discourse, is whether to accept descriptivism about prudential discourse as opposed to some form of anti-­descriptivism such as expressivism or inferentialism. That question about prudential discourse has yet to be taken up at any length because there has been so little explicit inquiry into meta-­prudential questions.43 For at least these three reasons it would be better if we focused more explicitly on second-­order debates about prudential discourse, debates about the nature of prudential value, alongside debates about the grounds of prudential value, as opposed to categorize theories as subjective or objective or trying to determine if there is an anti-­alienation constraint. 43 Rosati (1996: 310) mentions non-­cognitivism in one of her arguments for the anti-­ alienation constraint. That, along with Railton (1989) are, to my knowledge, the only mentions of the view in the context of prudential value and prudential discourse and there has never been a card-­carrying non-­cognitivist about prudential value (though Smart (1973: 22) defends a form of hybrid non-­cognitivism about happiness claims). In fact, many non-­cognitivists about moral discourse implicitly, sometimes explicitly, accept the view that prudential discourse is non-­normative. For examples see Chapter 6.

188  Dear Prudence

7. Conclusion In this chapter I have sought to demonstrate the benefits that arise from importing resources from metaethics and applying them to debates about prudential value (on the assumption that prudential discourse is normative). These pay-­offs came from taking familiar debates about prudential value—such as the attempt to divide theories into subjective or objective, and the attempt to establish or reject an anti-­alienation constraint—and recasting them as debates about the truth of realism or anti-­realism about prudential value. I argued that this is possible and that it is desirable for three reasons. First, that it allows us to use insights and distinctions from metaethics to illuminate the issues. Second, that it allows us to pursue the question of whether there is a true second-­order theory for both moral and prudential discourse. Finally, that it prompts us to ask certain neglected questions about prudential discourse, such as whether to accept (the hitherto implicitly assumed) descriptivism.

8 Conclusion Taking Stock and Looking Ahead

In the introduction to this book I distinguished the following questions: 1. Are there normative prudential reasons? 2. Are prudential judgements normative judgements? 3. Are these judgements motivationally special? 4. Is prudential discourse context-­sensitive in significant ways? 5. Does prudential discourse involve the ascription of normative properties (in either a deflationary sense or a robust sense) and, if robust, what is the nature of prudential properties? 6. Is prudential talk partly (or wholly) expressive? In the previous chapters, I have made a case for explicit answers to some of these questions (namely 1, 2, 3 and 4).1 In this chapter, I will first briefly recapitulate my case for those answers. I will also point out some of the questions that arise from the views I defend herein, which must be attended to in order to answer the questions that I did not explicitly address (questions 5 and 6). I began (Chapter 1) by arguing that there is a set of genuinely normative prudential reasons. I then (Chapter  2) shifted to the question of whether prudential discourse is a normative form of discourse. Having identified some features as ‘markers’ of the normative, I then set myself the task of showing that we find these features within prudential discourse. I did that explicitly in Chapter 2, but we also saw the same features arise as part of my giving a general account of prudential discourse (Chapters 3, 4, 5). 1  Namely: yes, yes, yes, and not as much as some people suggest. Dear Prudence: The Nature and Normativity of Prudential Discourse. Guy Fletcher, Oxford University Press (2021). © Guy Fletcher. DOI: 10.1093/oso/9780198858263.003.0009

190  Dear Prudence I first (Chapter 3) showed how to narrow down from the general set of ‘good for’ claims to claims about prudential value. Rather than identifying some specific sense of ‘good for’ that is prudential, I argued that Finlay’s end-­relationalist view is the most plausible account of ‘good for’ talk. I then showed how a similarly contextualist view, this time from the work of Angelika Kratzer, can be fruitfully applied to prudential uses of modal vocabulary such as ‘ought’, ‘must’, and ‘needs’ claims. In this discussion we saw clearly that one of the markers of normative discourse—its being critical and evaluative—was satisfied by prudential discourse. Prudential judgements are couched in directive (‘reasons’) and evaluative (‘good for’, ‘best for’, ‘bad for’) vocabulary and clearly constitute evaluations relative to some standard. We also saw that one of the normative flavours of ‘needs’, ‘must’, and ‘ought’ claims is prudential, meaning that we find another commonality in the deployment of modal vocabulary in the prudential domain. I also showed how prudential standards’ being context-­sensitive—such as in the case of ‘doing well’ claims—did not lend support for deeper, conceptually pluralistic, forms of contextualism about prudential discourse and that such views, as defended by Alexandrova and Campbell, are under-­motivated by the relevant data. Another marker of the normative that prudential discourse satisfies is that of playing a central role in deliberation. As we saw in Chapter  1, prudential value generates reasons for action and for attitudes across a vast array of practical contexts, the prudential status of one’s options is deemed of central practical relevance and is very often taken to determine what one ought to do all-­things-­considered.2 Furthermore, prudence is one of the standards that seems rationally non-­optional, in a way that marks it out as different from standards like etiquette and fashion. We also saw (Chapter 5) that there is a tight connection between the prudential judgements that one forms and motivation. At least some prudential judgements are necessarily connected to motivation in fully rational agents, such that a form of judgement internalism is true of (at least) these judgements.

2  On the assumption that there is such a thing.

Conclusion: Taking Stock and Looking Ahead  191 As we have also seen, such as in Chapter 4, prudential value concepts are extremely permissive when it comes to disagreement. Fundamental prudential disagreement does not incline us to diagnose agents as talking past one another in a way that would be reasonable if we found fundamental intractable disagreement in other domains. This fits with prudential discourse being normative. In these respects we have found clear parallels between moral discourse and prudential discourse. We have also found no significant disanalogy between moral and prudential discourse to disqualify the latter as a form of normative discourse. For example, I considered and rejected the idea that somehow prudential discourse is non-­normative because prudential value is desire-­based. In the absence of some further disanalogy between moral and prudential discourse I thus conclude that prudential discourse is normative, like moral discourse. But, as I noted in the introduction, one can reject this overall thesis but still separately accept the specific claims I made about the nature of prudential discourse. In Chapters 6 and 7 I showed how thinking of prudential discourse as normative can be fruitful for both metaethical and meta-­prudential questions. In Chapter 6 I showed how the normativity of prudential discourse bears on various forms of moral scepticism and projects connected to such views, along with debates about the nature of thick concepts. In Chapter 7 I proceeded in the opposite direction, showing how conceptual resources from metaethics can be fruitfully imported into theorizing about prudential value. Suppose that I am right that prudential discourse is normative, like moral discourse. Then we must ask certain questions (including but not limited to questions 5 and 6 from above). Are prudential judgements belief-­like, desire-­like, or some combination of the two? I noted above that there have been few, if any, non-­cognitivists about prudential judgements. Was that due to a failure to think of prudential judgements as normative (a tendency we also saw with respect to discussion of revisionary metaethical views) or, instead, due to the implicit judgement that non-­cognitivism is an implausible theory of prudential judgements? If we eschew non-­ cognitivism about prudential judgements, and instead embrace cognitivism, there is then the question of whether

192  Dear Prudence realism is true for prudential discourse. A striking feature of philo­soph­ic­al work on prudential value is its implicit treatment of prudential judgements as beliefs which ascribe properties. But beyond this, there has been very little discussion of the nature of such properties. As noted above, it is plausible that there has been some discussion of realism and anti-­realism about prudential discourse—though not in those terms and only as part of the background to the discussion of the purported anti-­alienation constraint on prudential value. For that reason, the full range of issues relevant to the realism and anti-­realism debate have not yet been fruitfully applied to the case of prudential judgements. It seems likely that many will favour forms of constructivist anti-­realism about prudential value, identifying prudential value properties with the property of what would be the object of the agent’s pro-­attitudes under some set of specified conditions.3 Orthogonal to the realism vs anti-­realism debate is the debate about naturalism. If we are cognitivists about prudential judgements, and treat such judgements as ascribing properties, should we also be naturalists? Do prudential judgements ascribe some naturalistic property (whether that can be specified in non-­prudential terms or not)? Alternatively, should we accept non-­naturalism about prudential value? Finally, should we accept error theory about prudential value discourse? Is prudential discourse in the business of ascribing properties that are never instantiated (be they natural or non-­natural) in a way analogous to that put forward by moral error theory?4 One question that these observations prompt is whether the same meta-­normative view is true for moral discourse as for prudential discourse. This would be the case if the following is true: One-­Theory-­To-­Rule-­Them-­All:  There is one true meta-­theory for all forms of normative discourse.5 3  As we saw earlier (Chapter 7) in the case of Rosati (1996) and Sumner (1996). 4  I hope to explore prudential error theory in future work. 5  There are, of course, a lot of details to unpack as to what counts as ‘the same’ meta-­theory. One could adopt descriptivism across the normative but offer different more specific accounts for, e.g., moral discourse than for prudential discourse. What I have in mind here is the person inclined to offer radically different treatments of moral and prudential discourse. For example, a kind of reductive naturalism (such as an hedonic identity view) about prudential discourse combined with non-­cognitivism about moral discourse.

Conclusion: Taking Stock and Looking Ahead  193 One way of rejecting this thesis (‘one theory. . .’) would be to treat all meta views as in agreement with respect to the major questions (states of mind, nature of the relevant form of language, nature of the properties the discourse purports to ascribe) and as disagreeing only as to whether such property ascriptions are successful. That is, one might hold the same set of views about moral and prudential discourse except that one is a realist about prudential judgement but an error theorist about moral judgement (or aesthetic judgement or some other region of discourse). In fact, I argued above, it is less plausible that prudential discourse is in error than that moral discourse is in error. Some authors have presented their views as meta-­normative views, or explicitly made clear that their metaethical view is supposed to apply beyond the moral in particular.6 In their case, accepting that prudential discourse is normative commits them to showing that there are no significant differences between moral and prudential discourse, such that the same meta-­normative view applies to (at least) these two domains. We saw above that there is at least a prima facie case for some more fundamental and interesting differences between moral and prudential discourse. Here is one example of this: a very general form of internalism which applies to all moral judgements seems prima facie plausible (which is evidence of non-­cognitivism about moral judgements). By contrast, such an extensive form of internalism seems implausible for prudential judgements (which is evidence for cognitivism about prudential judgements). A second place where we find grounds for treating moral and prudential discourse differently is that the case for constructivism seems much more plausible in the prudential case. People frequently invoke the subjectivity of prudential value in arguing for claims that are plaus­ ibly construed either as forms of constructivism about prudential value (such as the claims made by Rosati and Sumner) or in arguing for first-­order views which fit naturally with constructivism about prudential value (such as the desire-­fulfilment theory of prudential value). For many, it is obvious that prudential value admits of some form of constructivist treatment (in a way that does not extend to the moral case). 6  One particularly explicit instance of this is Enoch (2011a: ch. 1).

194  Dear Prudence Whatever one thinks of the strengths of constructivism in either domain, it is at least prima facie more plausible about prudential value than morality.7 I hope to have persuaded you of some particular theses about prudential discourse and the overarching claim that prudential discourse is normative, like moral discourse. If it is, then a new exciting area of meta-­normative theorizing opens up before us.

7  Cards on the table: I reject constructivism for both prudential and moral judgements. But I see the appeal in the prudential case.

References Alexandrova, A. (2017) A Philosophy for the Science of Well-Being (Oxford: Oxford University Press). Aristotle (340 BCE), Nicomachean Ethics. Arneson, R. (1999) ‘Human Flourishing Versus Desire Satisfaction’, Social Philosophy and Policy 16 (1): 113–142. Arpaly, N. & Schroeder, T. (2013) In Praise of Desire. Oxford University Press. Baker, D. (2017) ‘Skepticism about “Ought” Simpliciter’. In R. Shafer-Landau (ed.), Oxford Studies in Meta-Ethics: Volume 12 (Oxford: Oxford University Press). Bedke, M. (2010) ‘Might all Normativity be Queer?’, Australasian Journal of Philosophy 88 (1): 41–58. Bedke, M. (2011) ‘Passing the Deontic Buck’. In R.  Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 6 (Oxford: Oxford University Press). Behrends, J. (2011) ‘A New Argument for the Multiplicity of the Good-for Relation’, Journal of Value Inquiry 45(2): 121–33. Blackburn, S. (1993) ‘Errors and the Phenomenology of Value’. In Essays in QuasiRealism (New York: Oxford University Press), 149–65. Borg, E. (2012) Pursuing Meaning. (Oxford: Oxford University Press). Boyd, R. (1988) ‘How to Be a Moral Realist’. In G.  Sayre-McCord (ed.), Essays on Moral Realism (Cornell University Press), 181–228. Bradford, G. (2020) ‘The badness of pain’, Utilitas 32(2): 236–52. Bradley, B. (2009) Well-being and Death (Oxford: Oxford University Press). Bradley, B. (2014) ‘Objective Theories of Well-Being in Eggleston’. In B. Eggleston & D.  E.  Miller (eds.), The Cambridge Companion to Utilitarianism (Cambridge University Press). Brink, D.  O. (1986) ‘Externalist moral realism’, Southern Journal of Philosophy 24 (S1): 23–41. Broome, J. (2007) ‘Wide or narrow scope?’, Mind 116 (462): 359–70. Brunero, J. (2010) ‘The scope of rational requirements’, Philosophical Quarterly 60 (238): 28–49. Callahan, L.  F. (2018) ‘Moral Testimony: A Re-Conceived Understanding Explanation’, Philosophical Quarterly 68 (272): 437–59. Campbell, S. (2016) ‘The Concept of Well-Being’. In G. Fletcher (ed.), The Routledge Handbook of Philosophy of Well-Being (London: Routledge). Chang, R. (2004) ‘All Things Considered’, Philosophical Perspectives 18 (1): 1–22. Chrisman, M. (2012) ‘“Ought” and Control”’, Australasian Journal of Philosophy 90 (3): 433–51. Chrisman, M. (2015) The Meaning of ‘Ought’: Beyond Descriptivism and Expressivism in Metaethics (Oxford: Oxford University Press).

196 References Copp, D. (1997) ‘The Ring of Gyges: Overridingness and the Unity of Reason’, Social Philosophy and Policy 14 (1): 86–101. Copp, D. (ms.) ‘Normative Pluralism and Skepticism about Ought Simpliciter’. Cowie, C. (2014a) ‘In Defence of Instrumentalism about Epistemic Normativity’, Synthese 191 (16): 4003–17. Cowie, C. (2014b) ‘Why Companions in Guilt Arguments Won’t Work’, The Philosophical Quarterly 64 (256): 407–22. Cowie, C. (2016a) ‘Good news for Moral Error Theorists: A Master Argument Against the Companions in Guilt Strategy’, Australasian Journal of Philosophy 94 (1): 115–30. Cowie, C. (2016b) ‘Revisionist Responses to the Amoralism Objection: A Reply to Julia Markovits’, Ethical Theory and Moral Practice 19: 711–723. Crisp, R. (2013) ‘Well-being’. In Edward N. Zalta (ed.), The Stanford Encyclopedia of Philosophy. Crisp, R. (2014) ‘II—Moral Testimony Pessimism: A Defence’. Aristotelian Society Supplementary Volume 88 (1): 129–43. Cuneo, T. (1999) ‘An Externalist Solution to the “Moral Problem” ’, Philosophy and Phenomenological Research 59: 359–80. Cuneo, T. (2007) The Normative Web (Oxford: Oxford University Press). Cuneo, T. & Christy, S. (2011) ‘The Myth of Moral Fictionalism’. In M. Brady (ed.), New Waves in Metaethics (Basingstoke: Palgrave MacMillan), 85–102. Cuneo, T. & Shafer-Landau, R. (2014) ‘The Moral Fixed Points: New Directions for Moral Nonnaturalism’, Philosophical Studies 171 (3): 399–443. Dancy, J. (2000) ‘Should we pass the buck?’, Royal Institute of Philosophy Supplement 47: 159–73. Dancy, J. (2002) Practical Reality (Oxford: Oxford University Press). Darwall, S. (2004) Welfare and Rational Care (Princeton: Princeton University Press). Darwall, S. (2013) ‘Moral Obligation: Form and Substance’. In Morality, Authority and Law: Essays in Second Personal Ethics 1 (Oxford: Oxford University Press). Darwall, S. (2016) ‘Making the “Hard” Problem of Moral Normativity Easier’. In E. Lord & B. Maguire (eds.), Weighing Reasons (Oxford: Oxford University Press). Das, R. (2017) ‘Bad News for Moral Error Theorists: There Is No Master Argument Against Companions in Guilt Strategies’, Australasian Journal of Philosophy 95 (1): 58–69. Davidson, D. (2004) ‘How is Weakness of the Will Possible?’. In his Essays on Actions and Events (Oxford: Oxford University Press). DiPaolo, J. & Behrends, J. (2015) ‘Reason to Promotion Inferences’, Journal of Ethics and Social Philosophy 9: 1–9. Dorsey, D. (2013) ‘Two Dualisms of Practical Reason’. In R.  Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 8 (Oxford: Oxford University Press), 114–39. Dorsey, D. (2016) The Limits of Moral Authority (Oxford: Oxford University Press). Dorsey, D. (2017) ‘Why should Welfare “Fit”?’, Philosophical Quarterly 67 (269): 685–24. Dowell, J. (2011) ‘A Flexibly Contextualist Account of Epistemic Modals’, Philosopher’s Imprint 11 (14): 1–25.

References  197 Dowell, J. (2016) ‘The Metaethical Insignificance of Moral Twin Earth’. In R. ShaferLandau (ed.), Oxford Studies in Metaethics: Volume 11 (Oxford: Oxford University Press). Dreier, J. (1990) ‘Internalism and speaker relativism’, Ethics 101 (1): 6–26. Driver, J. (2006) ‘Autonomy and the Asymmetry Problem for Moral Expertise’. Philosophical Studies 128 (3): 619–44. Eklund, M. (2017) Choosing Normative Concepts (Oxford: Oxford University Press). Elbourne, P. (2011) Meaning: A Slim Guide to Semantics. (Oxford: Oxford University Press). Elstein, D. Y. & Hurka, T. (2009) ‘From Thick to Thin: Two Moral Reduction Plans’, Canadian Journal of Philosophy 39 (4): 515–35. doi:10.1353/cjp.0.0063 Enoch, D. (2005) ‘Why Idealize?’ Ethics, 115: 759–787. Enoch, D. (2011a) Taking Morality Seriously (Oxford: Oxford University Press). Enoch, D. (2011b) ‘Reason-giving and the law’. In Leslie Green & Brian Leiter (eds.), Oxford Studies in Philosophy of Law (Oxford University Press). Enoch, D. (2014) ‘A Defense of Moral Deference’. Journal of Philosophy 111 (5): 229–58. Ewing, A. C. (1953) Ethics (London: English Universities Press). Fantl, J. & McGrath, M. (2007) ‘On Pragmatic Encroachment in Epistemology’, Philosophy & Phenomenological Research 75 (3): 558–89. Finlay, S. (2006) ‘The Reasons That Matter,’ Australasian Journal of Philosophy, 84(1): 1–20. Finlay, S. (2007a) ‘Too Much Morality’. In Paul Bloomfield (ed.), Morality and Self-Interest (Oxford: Oxford University Press). Finlay, S. (2007b) ‘Four Faces of Moral Realism’, Philosophy Compass 2 (6): 820–49. Finlay, S. (2014) Confusion of Tongues (Oxford: Oxford University Press). Finlay, S. (2016) ‘ “Ought”: Out of Order’. In M.  Chrisman & N.  Charlow (eds.), Deontic Modality (Oxford: Oxford University Press). Fletcher, G. (2012) ‘The Locative Analysis of Good for Formulated and Defended’, Journal of Ethics and Social Philosophy 6 (1): 1–26. Fletcher, G. (2013) ‘A Fresh Start for the Objective List Theory of Well-Being’, Utilitas 25 (2): 206–20. Fletcher, G. (2016a) The Philosophy of Well-Being: An Introduction (Routledge). Fletcher, G. (2016b) ‘Objective List Theories’. In G.  Fletcher (ed.), Routledge Handbook of Philosophy of Well-Being (Routledge). Fletcher, G. (2017) ‘Moral Testimony: Once more with feeling’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 12 (Oxford: Oxford University Press). Fletcher, G. (2018) ‘Pain for the Moral Error Theory? A New Companions-in-Guilt Argument’, Australasian Journal of Philosophy. Fletcher, G. (2018) ‘Needing and Necessity’. In M. Timmons (ed.), Oxford Studies in Normative Ethics: Volume 8 (Oxford University Press). Fletcher, G. (2019) ‘Taking Prudence Seriously’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 14 (Oxford: Oxford University Press). Fletcher, G. & Ridge, M. (eds) (2014) Having It Both Ways: Hybrid Theories and Modern Metaethics (Oxford: Oxford University Press).

198 References Fogal, D. (2016) ‘Confusion of Tongues: A Theory of Normative Language, by Stephen Finlay’, Ethics 127 (1): 281–8. Foot, P. (1958) ‘Moral Arguments’, Mind 67 (268): 502–13. Foot, P. (1958–59) ‘Moral Beliefs’, Proceedings of the Aristotelian Society, New Series 59: 83–104. Foot, P. (1972) ‘Morality as a System of Hypothetical Imperatives’, Philosophical Review 81 (3): 305–16. Foot, P. (2000) Natural Goodness (Oxford: Oxford University Press). Foucault, M. (1977) Discipline and Punish (Vintage Books). Garner, R. (2007) ‘Abolishing Morality’, Ethical Theory and Moral Practice 10 (5): 499–513. Geach, P. (1956) ‘Good and Evil’, Analysis 17 (2): 33–42. Gert, J. (2012) Normative Bedrock: Response-Dependence, Rationality, and Reasons (Oxford: Oxford University Press). Gibbard, A. (2003) Thinking How To Live (Cambridge, MA: Harvard University Press). Glassen, P. (1960) ‘The Senses of “Ought” ’, Philosophical Studies 11: 10–16. Gregory, A. (2014) ‘A very good reason to reject the buck-passing account’, Australasian Journal of Philosophy 92 (2): 287–303. Gregory, A. (2017a) ‘Are All Normative Judgements Desire-Like?’, Journal of Ethics and Social Philosophy 12 (1): 29–55. Gregory, A. (2017b) ‘Might Desires Be Beliefs About Normative Reasons?’ In J. Deonna & F. Lauria (eds.), The Nature of Desire (Oxford University Press). Grice, H.  P. (1989) ‘Indicative conditionals’. In Studies in the Way of Words (Cambridge, MA: Harvard University Press). Groll, D. & Decker, J. (2014) ‘Moral Testimony: One of These Things is Just Like the Others’, Analytic Philosophy 55 (1): 54–74. Haji, I. (2012) Reason’s Debt to Freedom: Normative Appraisals, Reasons, and Free Will (New York: Oxford University Press). Hall, A. & Tiberius, V. (2016) ‘Well-Being and Subject Dependence’. In G. Fletcher (ed.), Routledge Handbook of Philosophy of Well-Being (Routledge). Hampton, J. (1998) The Authority of Reason (Cambridge: Cambridge University Press). Hanson, L. (2018) ‘Moral Realism, Aesthetic Realism, and the Asymmetry Claim’, Ethics 129 (1): 39–69. Hare, R. M. (1952) The Language of Morals (Oxford: Clarendon). Harman, G. (1975) ‘Moral relativism defended’, Philosophical Review 84 (1): 3–22. Hawkins, J. (2014) ‘Well-Being, Time, and Dementia’, Ethics 124 (3): 507–542. Hawthorne, J. & Stanley, J. (2008) ‘Knowledge and Action’, The Journal of Philosophy 105 (10): 571–90. Hayward, M. (2019) ‘Practical Reason, Sympathy and Reactive Attitudes’, Noûs 53 (1): 51–75. Heathwood, C. (2009) ‘Moral and Epistemic Open Question Arguments’, Philosophical Books 50: 83–98. Heathwood, C. (2013) ‘Reductionism in Ethics’. In H.  LaFollette (ed.), The International Encyclopedia of Ethics (Chichester: Wiley-Blackwell).

References  199 Heathwood, C. (2014) ‘Subjective Theories of Well-Being’. In Ben Eggleston & Dale Miller (eds.), The Cambridge Companion to Utilitarianism (Cambridge University Press), 199–219. Hills, A. (2009) ‘Moral Testimony and Moral Epistemology’, Ethics 120: 94–127. Hills, A. (2019) ‘Moral Testimony: Transmission Versus Propagatiom’, Philosophy and Phenomenological Research 99 (3): 694–713. Hobbes, T. Leviathan (1651). Holton R. (2009) Willing, Wanting, Waiting (Oxford: Clarendon Press). Hooker, B. (1991) ‘Theories of welfare, theories of good reasons for action, and ontological naturalism’, Philosophical Papers 20 (1): 25–36. Hooker, B. (2005) ‘Review of Welfare and Rational Care by Stephen Darwall’, Mind 114 (454): 409–13. Hopkins, R. (2007) ‘What is Wrong with Moral Testimony’, Philosophy and Phenomenological Research 74: 611–34. Hopkins, R. (2011) ‘How to be a Pessimist about Aesthetic Testimony’, Journal of Philosophy 108 (3): 138–57. Horgan, T. & Timmons, M. (1991) ‘New Wave Moral Realism Meets Moral Twin Earth’, Journal of Philosophical Research 16: 447–65. Howell, R. (2014) ‘Google Morals, Virtue, and the Asymmetry of Deference’, Noûs 48: 389–415. Hume, D. (1739) A Treatise of Human Nature. Hurka, T. (1987) ‘ “Good” for and “good for” ’, Mind 96 (381): 71–3. Ingram, S. (2015) ‘After Moral Error Theory, After Moral Realism’, Southern Journal of Philosophy 53 (2): 227–48. Johnson King, Z. (2019) ‘We Can Have Our Buck and Pass It, Too’. In R.  ShaferLandau (ed.), Oxford Studies in Metaethics: Volume 14 (Oxford: Oxford University Press). Jones, K. (1999) ‘Second-Hand Moral Knowledge’, Journal of Philosophy 96 (2): 55–78. Joyce, R. (2001) The Myth of Morality (Cambridge: Cambridge University Press). Joyce, R. (2007) ‘Morality, Schmorality’. In Paul Bloomfield (ed.), Morality and Self-Interest (Oxford: Oxford University Press). Joyce, R. (2016) ‘Moral Anti Realism’. In Edward  N.  Zalta (ed.), The Stanford Encyclopedia of Philosophy (Winter 2016 Edition), https://plato.stanford.edu/ archives/win2016/entries/moral-anti-realism/. Kant, I. (1785) Groundwork of the Metaphysics of Morals. Killoren, D. (2019). ‘Infinitism about Cross-Domain Conflict’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 14 (Oxford: Oxford University Press). Kirchin, S. (2017) Thick Evaluation (Oxford University Press). Köhler, S. & Ridge, M. (2013) ‘Revolutionary Expressivism’, Ratio 26 (4): 428–49. Korsgaard, C. (1986) ‘Skepticism about Practical Reason’, The Journal of Philosophy 83 (1): 5–25. Korsgaard, C. (2013) ‘The Relational Nature of the Good’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 8 (Oxford: Oxford University Press).

200 References Kratzer, A. (1977) ‘What “must” and “can” must and can mean’, Linguistics and Philosophy, 1 (3): 337–55. Kraut, R. (2007) What is Good and Why: The Ethics of Well-Being (Cambridge, MA: Harvard University Press). Lauinger, W. (2012) Well-Being and Theism: Linking Ethics to God (London: Continuum Publishing). Lewis, D. (1979) ‘Scorekeeping in a language game’, Journal of Philosophical Logic 8: 339–59. Lewis, M. (2020) ‘The New Puzzle of Moral Deference’, Canadian Journal of Philosophy 50 (4): 460–76. Lillehammer, H. (2007) Companions in Guilt: Arguments for Ethical Objectivity (London: Palgrave Macmillan). Lin, E. (2015) ‘Prudence, Morality, and the Humean Theory of Reasons’, Philosophical Quarterly 65 (259): 220–40. Lin, E. (2016a) ‘Monism and Pluralism’. In Guy Fletcher (ed.), The Routledge Handbook of Philosophy of Well-Being (Routledge), 331–41. Lin, E. (2016b) ‘The Subjective List Theory of Well-Being’, Australasian Journal of Philosophy 94 (1): 99–114. Lin, E. (2017) ‘Against Welfare Subjectivism’, Noûs 51 (2): 354–77. Lin, E. (2018) ‘Simple Probabilistic Promotion’, Philosophy and Phenomenological Research 96 (2): 360–79. Lin, E. (2020) ‘The experience requirement on well-being’, Philosophical Studies: 1–20. Lord, E. (2014) ‘The real symmetry problem(s) for wide-scope accounts of rationality’, Philosophical Studies 170 (3): 443–64. Lord, E. (2018) ‘How to Learn about Aesthetics and Morality through Acquaintance and Deference’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 13 (Oxford: Oxford University Press). Lutz, M. (2014) ‘The “Now What” Problem for error theory’, Philosophical Studies 171: 351–71. McDowell, J. (1979) ‘Virtue and reason’, Monist 62: 331–50. McElwee, B.  (2017) ‘Supererogation Across Normative Domains’, Australasian Journal of Philosophy 95 (3): 505–16. McGrath, S. (2009) ‘The Puzzle of Pure Moral Deference’, Philosophical Perspectives 23: 321–44. McGrath, S. (2011) ‘Skepticism about Moral Expertise as a Puzzle for Moral Realism’, Journal of Philosophy 108 (3): 11–137. McLeod, S. K. (2014) ‘Absolute Biological Needs’, Bioethics 28 (6): 293–301. McPherson (2018) ‘Authoritatively Normative Concepts’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 13. (Oxford: Oxford University Press). Mackie, J. L. (1980) Hume’s Moral Theory (Routledge & Kegan Paul). Mackie, J. L. (1990) Ethics: Inventing Right and Wrong (London: Penguin). Manne, K.. (2016) ‘Democratizing Humeanism’. In E.  Lord & B.  Maguire (eds.), Weighing Reasons (Oxford: Oxford University Press).

References  201 Markovits, J. (2014) Moral Reason (Oxford: Oxford University Press). Mason, E. (2008) ‘An Argument Against Motivational Internalism’, Proceedings of the Aristotelian Society 108: 135–56. Mill, J. S. (1861) Utilitarianism. Mogensen, A, (2017) ‘Moral Testimony Pessimism and the Uncertain Value of Authenticity’, Philosophy and Phenomenological Research 95 (2): 261–84. Moore, G. E. (1903) Principia Ethica, Revised Edition, T. Baldwin (ed.) (Cambridge: Cambridge University Press, 2000). Moore, G.  E. Ethics [1912], W.  H.  Shaw (ed.) Ethics and ‘The Nature of Moral Philosophy’, (Oxford: Clarendon Press, 2005). Nagel, T. (1970) The Possibility of Altruism (Princeton. Princeton University Press). Nickel, P. (2001). ‘Moral Testimony and Its Authority’, Ethical Theory and Moral Practice 4 (3): 253–66. Nolan, D., Restall, G., & West, C. (2004) ‘Moral fictionalism versus the rest’, Australasian Journal of Philosophy 83 (3): 307–30. Nozick, R. (1974) Anarchy, State, and Utopia (Oxford: Blackwell Publishing). Olson, J. (2006) ‘Review of Stephen Darwall’s Welfare and Rational Care’, Economics and Philosophy 22 (1): 171–77. Olson, J. (2011a) ‘In defense of moral error theory’. In Michael Brady (ed.), New Waves in Metaethics (Palgrave Macmillan). Olson, J. (2011b) ‘Error theory and reasons for belief ’. In A. Reisner & A. SteglichPetersen (eds.), Reasons for Belief (Cambridge: Cambridge University Press). Olson, J. (2014) Moral Error Theory: History, Critique, Defence (Oxford: Oxford University Press). Parfit, D. (1984) Reasons and Persons (Oxford University Press). Parfit, D. (1997) ‘Reasons and motivation’, Aristotelian Society Supplementary Volume 71 (1): 99–130. Parfit, D. (2011) On What Matters: Volume One (Oxford: Oxford University Press). Plato (375 BCE) The Republic. Plunkett, D. & Sundell, T. (2013) ‘Dworkin’s Interpretivism and the Pragmatics of Legal Disputes’, Legal Theory 19: 242–81. Prichard, H.  A. (1912) ‘Does Moral Philosophy Rest on a Mistake?’, Mind 21 (81): 21–37. Railton, P.  (1984) ‘Alienation, consequentialism, and the demands of morality’, Philosophy and Public Affairs 13 (2): 134–71. Railton, P. (1989) ‘Naturalism and Prescriptivity’, Social Philosophy and Policy 7 (1): 151. Railton, P. (2003) ‘Facts and Values’. In his Facts, Values and Norms (Cambridge: Cambridge University Press). Reader, S. & Brock, G. (2004) ‘Needs, Moral Demands and Moral Theory’, Utilitas 16 (3): 251–66. Ridge, M. (2014) Impassioned Belief (Oxford: Oxford University Press). Ridge, M. (2015) ‘Internalism: Cui Bono?’. In G. Björnsson et al (eds.), Motivational Internalism (Oxford: Oxford University Press).

202 References Ridge, M. (2017) ‘Reasons for Action: Agent-Neutral vs. Agent-Relative’. In Edward  N.  Zalta (ed.), The Stanford Encyclopedia of Philosophy (Fall 2017 Edition), https://plato.stanford.edu/archives/fall2017/entries/reasons-agent/. Ridge, M. (2020) ‘Reinventing ethics: Inventing right and wrong’, Journal for the History of Analytical Philosophy 8 (4): 1–20. https://doi.org/10.15173/jhap. v8i4.4000 Roberts, D. (2017a) ‘Thick Concepts’. In Tristram McPherson & David Plunkett (eds.), The Routledge Handbook of Metaethics (Routledge), 211–25. Roberts, D. (2017b) ‘Depending on the Thick’, Aristotelian Society Supplementary Volume 91 (1) (June): 197–220. https://doi.org/10.1093/arisup/akx006 Roberts, D. (2018) ‘Why Believe in Normative Supervenience?’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 13 (Oxford: Oxford University Press). Rosati, C.  (1995a) ‘Persons, Perspectives, and Full Information Accounts of the Good’, Ethics 105 (2): 296–325. Rosati, C.  (1995b) ‘Naturalism, Normativity, and the Open Question Argument’, Noûs 29 (1): 46–70. Rosati, C. (1996) ‘Internalism and the good for a person’, Ethics 106 (2): 297–326. Rosati, C. (2008) ‘Objectivism and Relational Good’, Social Philosophy and Policy 25 (1): 314–49. Rosati, C. (2009) ‘Relational Good and the Multiplicity Problem’, Philosophical Issues 19: 205–34. Rosati, C. (2019) ‘Constitutional Realism’. In D. Plunkett, S. Shapiro, & K. Toh (eds.), Dimensions of Normativity: New Essays on Metaethics and Jurisprudence (Oxford: Oxford University Press). Rosati, C. (2020) ‘Welfare and Rational Fit’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 15 (Oxford: Oxford University Press). Ross, W. D. (2002) The Right and the Good (Oxford University Press). Rowland, R. (2013) ‘Moral Error Theory and the Argument from Epistemic Reasons’, Journal of Ethics and Social Philosophy 7 (1): 1–24. Rowland, R. (2016) ‘Rescuing Companions in Guilt Arguments’, The Philosophical Quarterly 66 (262): 161–71. Rowland, R. (2019) The Normative and the Evaluative: The Buck-Passing Account of Value (Oxford: Oxford University Press). Sarch, A. (2011) ‘Internalism About a Person's Good: Don't Believe It’, Philosophical Studies 154 (2): 161–84. Scanlon, T. (1998) What We Owe to Each Other (Belknap Press of Harvard University Press). Schroeder, M. (2005) ‘Realism and reduction: The Quest for robustness’, Philosophers’ Imprint 5: 1–18. Schroeder, M. (2007) Slaves of the Passions (Oxford: Oxford University Press). Schroeder, M. (2009) ‘Buck-Passers’ Negative Thesis’, Philosophical Explorations 12 (3): 341–7. Schroeder, M. (2011) ‘Ought, Agents, and Actions’, Philosophical Review 120 (1): 1–41. Schroeder, M. (2015) ‘Is Knowledge Normative?’, Philosophical Issues 25 (1): 379–95.

References  203 Schuppert, F. (2013) ‘Distinguishing basic needs and fundamental interests’, Critical Review of International Social and Political Philosophy 16 (1): 24–44. Sennet, A. (2016), ‘Ambiguity’, The Stanford Encyclopedia of Philosophy (Spring 2016 Edition), Edward  N.  Zalta (ed.), . Sharadin, N. (2015) ‘Problems for Pure Probabilism about Promotion (and a Disjunctive Alternative)’, Philosophical Studies 172: 1371–86. Sharadin, N. (2016) ‘Checking the Neighborhood: A Reply to DiPaolo and Behrends on Promotion’, Journal of Ethics and Social Philosophy 11: 1–8. Sidgwick, H. (1981) The Methods of Ethics 7th edition, originally published, 1907 (Indianapolis: Hackett Publishing Company). Sliwa, P. (2012) ‘In Defense of Moral Testimony’, Philosophical Studies 158 (2): 175–95. Smart, J., & Williams, B.  (1973). Utilitarianism: For and Against (Cambridge: Cambridge University Press). Smith, M. (1994) The Moral Problem (Oxford: Blackwell). Snedegar, J. (2013) ‘Reason claims and Contrastivism about Reasons’, Philosophical Studies 166 (2): 231–42. Sobel, D. (2019) ‘The Case for Stance Dependent Reasons’, Journal of Ethics and Social Philosophy 15 (2): 146–74. Stalnaker, R. C. (2002) ‘Common ground’, Linguistics and Philosophy 25: 701–21. Stanley, J. (2002) ‘Making it articulated’, Mind & Language 17: 149–68. Stratton-Lake, P. & Hooker, B. (2006) ‘Scanlon versus Moore on goodness’. In T. Horgan & M. Timmons (eds.), Metaethics After Moore (Oxford University Press). Street, S. (2009) ‘In Defense of Future Tuesday Indifference: Ideally Coherent Eccentrics and the Contingency of What Matters’, Philosophical Issues, Vol. 19, Metaethics: 273–98. Streumer, B. (2013) ‘Can We Believe the Error Theory?’ Journal of Philosophy 110 (4): 194–212. Streumer, B. (2017) Unbelievable Errors: An Error Theory About All Normative Judgments (Oxford University Press). Stroud, S. (1998) ‘Moral Overridingness and Moral Theory’, Pacific Philosophical Quarterly 79: 170–89. Stroud, S. (2003) ‘Weakness of Will and Practical Judgement’, in S.  Stroud & C. Tappolet (eds.), Weakness of Will and Practical Irrationality (Oxford: Clarendon Press), 121–46. Sturgeon, N. L. (1985) ‘Moral Explanations’. In James Rachels (ed.), Ethical Theory 1: The Question of Objectivity (Oxford University Press). Suikkanen, J. (2005) ‘Reasons and Value—in Defence of the Buck-Passing Account’, Ethic Theory and Moral Practice 7: 513–35. Sumner, L. W. (1996) Welfare, Happiness, and Ethics (Oxford University Press). Svavarsdóttir, S. (1999) ‘Moral Cognitivism and Motivation’, The Philosophical Review 108: 161–219. Swartzer, S. (2015) ‘Humean Externalism and the Argument from Depression’, Journal of Ethics and Social Philosophy 9 (2): 1–16.

204 References Sylvan, K. (2018) ‘Knowledge as a Non-Normative Relation’, Philosophy and Phenomenological Research 97 (1): 190–222. Theunissen, N. (2018) ‘Must We Be Just Plain Good? On Regress Arguments for the Value of Humanity’, Ethics 128 (2) (January): 346–72. https://doi.org/10.1086/694273 Thomson, G. (1987) Needs (London & New York: Routledge & Kegan Paul). Thomson, G. (2005) ‘Fundamental Needs’. In S.  Reader (ed.), Royal Institute of Philosophy Supplement 80 (57): 175–86. Tiffany, E. (2007) ‘Deflationary Normative Pluralism’, Canadian Journal of Philosophy 37 Supplement [vol. 33]: 231–62. Van Weelden, J. (2019) ‘On Two Interpretations of the Desire-Satisfaction Theory of Prudential Value’, Utilitas 31 (2):137–56. Väyrynen, P. (2006) ‘Resisting the buck-passing account of value’, Oxford Studies in Metaethics: Volume 1 (Oxford: Clarendon Press), 295–324. Väyrynen, P. (2013) The Lewd, the Rude and the Nasty: A Study of Thick Concepts in Ethics (Oxford University Press). Ventham, E. (2019) ‘Reflective Blindness, Depression and Unpleasant Experiences’, Analysis 79 (4): 684–93. Viebahn, E. & Vetter, B. (2016) ‘How many meanings for “may”? The case for modal polysemy’, Philosophers’ Imprint 16 (10). Way, J. (2011) ‘The symmetry of rational requirements’, Philosophical Studies 155 (2): 227–39. Wedgwood, R. (2004) ‘The Metaethicist’s Mistake’, Philosophical Perspectives 18: 405–26. Wedgwood, R. (2006) ‘The Meaning of “Ought”’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 1 (Oxford: Clarendon Press), 127–60. Whiting, D. (2015) ‘The Glass is Half Empty: A New Argument for Pessimism about Aesthetic Testimony’, British Journal of Aesthetics 55 (1): 91–107. Whiting, D. (forthcoming) ‘Aesthetic Reasons and the Demands They (Do Not) Make’, Philosophical Quarterly. Wierzbicka, Anna. (2004) ‘Happiness in cross-linguistic & cross-cultural perspective’, Daedalus 133: 34–43. doi:10.1162/001152604323049370 Wiggins, D. (1987) ‘Claims of Need’. In his Needs, Values, Truth: Essays in the Philosophy of Value (Oxford: Basil Blackwell). Wiggins, D. & Dermen, S. (1987) ‘Needs, need, needing’, Journal of Medical Ethics 13 (2): 62–8. Williams, B. (1981) ‘Internal and External Reasons’. In Moral Luck (Cambridge: Cambridge University Press), 101–13. Wodak, D. (2017) ‘Expressivism and Varieties of Normativity’. In R. Shafer-Landau (ed.), Oxford Studies in Metaethics: Volume 12 (Oxford: Oxford University Press), 265–93. Wodak, D. (2019) ‘Mere formalities: fictional normativity and normative authority’, Canadian Journal of Philosophy 49 (6): 828–50. Wodak, D. (forthcoming) ‘Approving on the basis of Normative Testimony’, Oxford Studies in Metaethics.

References  205 Worsnip, A. (2018) ‘Eliminating Prudential Reasons’. In M. Timmons (ed.), Oxford Studies in Normative Ethics: Volume 8 (Oxford: Oxford University Press), 236–57. Xenakis, J. (1957) ‘A Mistaken Distinction in Ethical Theory’, Philosophical Studies 8: 69–71. Ziff, P. (1960) Semantic Analysis (New York: Cornell University Press).

Index For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on occasion, appear on only one of those pages. Alexandrova, Anna  7, 92–107 ambiguity and polysemy  64, 84, 86n.49 amoralist 138n.20 anti-alienation constraint  140–2, 175n.17, 178–83 anti-realism, see realism and anti-realism Aristotle  76n.29, 123n.9 Arneson, R.  172n.10 aspectualism 104–7 attributive goodness  35–6, 53n.46, 67–70 Baker, D.  14n.6, 39n.18 Bedke. Matt  14n.7, 48n.37, 52n.45, 150n.17 Behrends, Jeff  61, 64n.7, 71n.23, 79 Bradford, Gwen  54n.48 Bradley, Ben  176, 183n.37, 184n.39, 185 Broome, John  131n.13 Brunero, John  131n.13 buck-passing theory of value  14, 16n.11 Campbell, Steve  7, 108–19 Chrisman, M.  86n.50 circumscriptionism, see Alexandrova, Anna companions in guilt arguments  150–9, 153n.23 conceptual pluralism  108 contextualism, see Alexandrova, Anna Copp, David  8, 39n.18 Cowie, Chris  31n.47, 52n.45, 152n.22, 153n.24, 154n.28, 155n.30

Crisp, Roger  5n.11, 18n.17 Cuneo, Terence  41n.24, 52n.45, 151n.21, 152n.22, 153n.23, 169n.3 Dancy, Jonatham  14n.7, 169n.5 Darwall, Steve  41, 149n.12, 156n.33, 171n.8, 173n.13 Davidson, D.  123n.9 descriptivism and anti-descriptivism  176n.21, 187 developmentalism, see Kraut, Richard differential realization, see Alexandrova, Anna Diogenes syndrome  29–30 disagreement  40–2, 51n.43, 109–18, 157–8 Dorsey, Dale  50n.41, 141nn.23, 25, 179nn.26, 28, 182n.34 Dowell, Janice  41n.23 egoism 19 Eklund, M  165n.54 Elstein, D.  164–6 end-relationalism 70–7 Enoch, David  5, 14n.5, 23n.27, 169n.3, 193n.6 epistemic normativity, see normativity error theory meta-normative 35 moral  45, 150–64, 192 prudential  17, 18n.17, 52–4, 109, 150–64, 156n.31, 192

208 Index examples (first appearance of) Agnes’ friendship  23 Alexis and the Aliens  127–8 Balanced Belinda  26 Christine’s career  15 Cult Colin  25 Deliberating Denise  121 Deliberating Denise (alternate)  125 Deliberating Dugald  136 Deluded Delphine  116 Depressed Dave  25 Doomsday device  79 Donald (landlord)  168 Educating Emily  25 Electrician 104 Hayward’s tippler  27–8 Helen’s holiday  15 Hard to reach Hillary  116 Mandeep’s job offer  138 Masha and Good Samaritan etc  93–4 Simple Stan  30 Ursula’s upgrade  15 Virkam and Julia  21 experience machine  41–2 everyday imprudence, inattention to  28–9 Finlay, Steve  4, 7, 70–7, 79, 169n.4 Foot, Philippa  13n.2, 41n.24, 76n.29, 147n.6, 169n.3 Foucault, M.  156n.31 Geach, P.  67n.12 Gert, J.  34 good for attributive view (GFA)  67–70 different senses  59–62, 81 Hurka’s scepticism  19–20 Moore’s scepticism  18–19 translation 63–4 Gregory, Alex  14n.7, 33, 131n.12 Haji, I.  34 Hampton, Jean  48n.37, 150n.17 Hanson, Louise  35n.9, 169n.5

Hawkins, Jennifer  141n.23, 182n.34 Hayward, Max  27–32 Heathwood, Chris  49n.39, 153n.24, 154, 177n.24, 183n.38 Hooker, Brad  171n.8 Horgan & Timmons  157 Hume, David  17n.13, 27n.36, 31n.47 Humean / Humeanism theory of reasons  129, 139 theory of motivation  129–30, 139–40 Hurka, T.  19–20, 164–6 Ingram, S.  160n.41 Johnson King, Z.  14n.9 jokes  43n.28, 80n.40, 163n.50 Joyce, Richard  50, 156n.32, 161, 162n.49, 169n.4 Kirchin, Simon  165n.54 Korsgaard, C.  74n.27 Kratzer, A.  7, 86–90 Kraut, Richard  77–80 Lauinger, William  185n.42 Lillehammer, Hallvard  52n.45, 151n.20 Lin, Eden  23nn.27, 28, 25n.31, 174n.15, 177n.24, 187n.43 Lord, Errol  43n.30, 44n.31, 131n.13 McElwee, Brian  29n.42 McPherson, T  14n.6 Mackie, J. L.  45–6, 52n.45, 151n.18, 161 Manne, K.  23n.27 markers of normativity, see normativity Markovitz, J.  23n.27, 31n.47 meta-prudential questions  1, 3, 189 meta-normative theory/theorising  35, 186–8, 192–3 Moore, G. E.  18–19, 19n.19, 169n.3 moral facts  168–70 moral realism and anti-realism  160n.41, 168–70

Index  209 moral twin earth, see disagreement, Horgan & Timmons moral scepticism, see error theory, why be moral motivation and irrationality  122–4, 134–5 and moral judgement  132n.15, 139–40, 141n.25, 193 and rational requirement  130–5 conative states  129–30 illicit intuitions  127–8 pragmatic explanations  125–7 prudential judgements and motivation  120–2, 135, 140 must, see needs, needs claims as modal claims Nagel, T.  17n.12, 21n.24 naturalism  48–52, 168–70, 192 needs needs claims as modal claims  86–90 supposed different types of needs  83–5 non-cognitivism  35, 132, 139–40, 176n.21, 187n.43, 191 and cognitivism  139–40, 191 hybrid con-cognitivism  140n.22 non-naturalism  35, 168–9 normativity aesthetic  24, 35n.9, 43 clubs, etiquette, fashion, and laws  35–6, 51, 53–4, 123–4, 190 epistemic  4n.6, 151–9, 153n.23 markers of  35–44 moral 35–44 pluralism 8 Olson, Jonas  5, 34, 42n.25, 50n.41, 52n.45, 150n.17, 153n.24, 156n.32 One-theory-to-rule-them-all 6, 187, 192–3 ought all-things-considered  9 paradox of hedonism  29n.41 Parfit, D.  21n.24, 25n.32 parity strategy  35–7

Continuities between moral and prudential discourse  3–4 Plato  115n.22, 146–7 Prichard, H. A.  146n.1, 147 prudential discourse  1 directive prudential discourse  1, 13, 59, 81–91 evaluative prudential discourse  1, 13, 59–81 inattention to  2 normativity of  33–5, 191 see also meta-prudential questions prudential error theory, see error theory prudential judgement internalism, formulations of prudential judgement internalism (PJI)  120 prudential judgement internalism* (PJI*)  129 prudential judgement internalism** (PJI**)  133 prudential judgement internalism (PJI 2)  135 prudential judgement internalism (PJI 3)  137 prudential judgement internalism (PJI 4)  139 prudential judgements and motivation, see motivation prudential judgements are normative (PJN)  7, 35 prudential supererogation  29n.42 prudential value, see well-being prudential value matters (PVM)  6, 13, 31, 147 types of objections to PVM distinguished 16–17 objection from agent-neutral reasons 20–2 objection from Humeanism  27–32 Railton, Peter  2, 29n.41, 141, 169n.3, 178–9, 187n.43 rationality, see motivation reactive attitudes  39–40

210 Index realism and anti-realism  168–87, 191–2 attitude-independence constraint  169–71 moral 186–8 reasons aesthetic  24, 32n.49 agent neutral vs agent relative  16–17, 20–1 categorical vs hypothetical  16–17, 23, 150n.16 for action or attitude  13, 22, 31–2, 31n.47 motivating vs normative or justifying 14 normative 13–14 prudential  13–16, 31n.47, 159–64 reduction epistemic to alethic properties  153–5 prudential to non-normative properties  48–52, 157–9 revisionary vs hermeneutic metaethical views 159–64 Ridge, Mike  21n.24, 37n.14, 47n.36, 140n.22 Roberts, Debbie  165, 165n.54 Rosati, Connie  18n.18, 60, 65, 124n.10, 141n.24, 171n.8, 175n.17, 178n.25, 179, 179n.26, 181–2, 187n.43 Ross, W. D.  122n.6, 169n.3 Rowland, R.  52n.45 Sarch, Alex  116n.24, 140–2, 178n.25 Scanlon, T. M.  16n.11 Schroeder, Mark  4n.6, 14nn.8, 9, 17n.13, 23n.27, 29n.44, 49nn.38, 40, 70n.20, 86n.50 self-regarding moral duties  149n.13 Shafer-Landau, Russ  41n.24 Smart, J.J.C  187n.43 Smith, M.  120n.2, 140n.21 Sobel, David  25n.32, 169n.5 Snedegar, J.  29n.44 Street, S.  25n.32

Streumer, Bart  5, 24n.30, 34, 35n.8, 52n.45, 53n.47, 150n.17, 156n.32, 160n.41 Stroud, Sarah  50n.41, 123n.9 Sumner, L. W.  180–3 testimony pure vs impure; direct vs indirect reliance upon  43 moral and prudential  42–4 thick concepts, reductive and non-reductive view of  164–6 threshold dependence, see Alexandrova, Anna Tiffany, E.  18n.18 utilitarianism 21n.23 Van Weelden, Joey  174n.14 Väyrynen, Pekka  14n.7, 165n.54 Ventham, E.  25n.32 Way, Jonathan  131n.13 Wedgwood, Ralph  86n.50, 141n.25 well-being desire-based  25–6, 26n.35, 45–8, 174, 174n.14 first vs second order views  170–8 hedonism 171n.9 objective list theory  25–6, 175–7 perfectionism 25–6 realism vs anti-realism  183–6 subjective vs objective theories  183–6 Whiting, Daniel  32n.49, 44n.31 Williams, B.  23n.27 why be moral?  146–50, 149n.15 why be prudent?  146–50 Wodak, Daniel  14n.6, 33 Worsnip, Alex  13–14, 24n.29, 27–32, 147n.8 zeugma 64 Ziff, P.  62 interests vs outcomes  70