Weighing and Reasoning: Themes from the Philosophy of John Broome [Hardcover ed.] 0199684901, 9780199684908

Citation preview

Weighing and Reasoning

Weighing and Reasoning Themes from the Philosophy of John Broome

EDITED BY

Iwao Hirose and Andrew Reisner

1

3

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries # the several contributors 2015 The moral rights of the authors have been asserted First Edition published in 2015 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: 2014947354 ISBN 978–0–19–968490–8 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.

Preface This volume is a collection of essays written by philosophers who have been influenced by the work and thought of John Broome. It is presented on the occasion of Broome’s retirement from the White’s Professor of Moral Philosophy at the University of Oxford. Broome, born in Malaysia in 1947, was originally trained as an economist, reading for his first degree at Trinity Hall, Cambridge, and then for a Ph.D. at MIT. He held a lectureship in economics at Birkbeck College, London. At the invitation of David Wiggins, and under Wiggins’s and Anthony Savile’s supervision, Broome read for an M.A. in philosophy at Bedford College in London University, while in his first year as a lecturer at Birkbeck. Subsequently, he became a reader in economics at the University of Bristol, later being promoted to Professor of Philosophy and Economics. He moved from Bristol to take up the post of Professor of Moral Philosophy at the University of St Andrews. In 2000, he moved to the University of Oxford, where he was the White’s Professor of Moral Philosophy and a fellow of Corpus Christi College until September 2014. Early in his career, Broome worked in mathematical economics, focusing on the analysis of general equilibrium. He also published an economics textbook, The Microeconomics of Capitalism, in 1983. Broome saw economics, especially welfare economics, as a field that was in many respects part of the study of ethics. Accordingly, his research gradually shifted focus towards the clear areas of overlap between economics and moral philosophy. In doing so, he both trod a well worn path and blazed a trail. We sometimes forget that economics originally grew out of moral philosophy. We forget this in part because the interaction between the two disciplines has gradually diminished over time. Broome helped to revive the traditional association of the two fields by bringing together the formal method of economics and the methods and approaches of philosophical analysis. His diverse and extensive work immediately made a significant impact on contemporary moral philosophy. This impact is eloquently illustrated by Derek Parfit in the preface to Reasons and Persons (Clarendon Press, 1984: viii): John Broome was a Visiting Fellow at my College throughout the academic year at the end of which I write these words. Both in written comments, and in very many discussions, he solved very many of my problems, and suggested many great improvements. . . . As the different academic disciplines drift away from their neighbours, it is heartening to find that an economist should be, in his spare time, so good a philosopher.

Broome’s early works on combining economics and moral philosophy are collected in the volume Ethics out of Economics (1999). However, his most important

vi

PREFACE

contribution to philosophical ethics came earlier in the decade with the publication of Weighing Goods (1991). In this seminal work, Broome considers how the utilitarian principle of distribution can be defended by arguments based on the formal notion of separability, familiar from the groundbreaking work of John Harsanyi. This book has become a modern classic, and it constitutes one of the important achievements in philosophical ethics of the last 30 years. His second instalment in the development of the formal axiological approach to moral philosophy was Weighing Lives (2004), in which he extends and reformulates the analytical framework of Weighing Goods to address the difficult problems posed by variable population-size cases. The time between the publication of Weighing Lives and Ethics out of Economics marked a second transition in the central focus of Broome’s work, this time away from the theory of value and normative ethics and towards the subjects of normativity and rationality. His contributions to the study of normativity and rationality have been many. In his first papers on these subjects, two central themes emerge. The first is the importance of thinking about ought rather than reasons, going against the fashion of the time. The second was in paying attention to the scope of normative operators. Although the distinction between detaching and non-detaching operators did not originate with Broome, it was he who helped show the importance of scope distinctions for contemporary normative theorising. In the period from 2005 to 2014, Broome published a series of influential papers on the nature of rationality. He argued that rationality was not closely connected to reasons, making an important case against what had been the prevailing view. During this time he also made significant advances in our understanding of whether rationality is itself normative and in understanding the different bases for practical and theoretical rationality. This period of his work culminated in Rationality through Reasoning (2013), in which Broome brings together his thinking concerning the nature of reasoning, correct reasoning, and the connection between rationality, reasoning, and normativity. Over and above his purely theoretical work, Broome has been philosophy’s most important contributor to the world’s response to the problem of global warming. His first contribution was Counting the Cost of Global Warming (1992). More recently, Broome has published Climate Matters: Ethics in a Warming World (2012). In it, he argues that, while responsibility for global warming mostly falls to governments and institutions, individuals also have an obligation to present and future people to reduce greenhouse gas emissions. He has also dedicated much of his time and expertise in recent years to his role as a lead author on Working Group III of the United Nation’s International Panel on Climate Change. With such a broad legacy, it is difficult to do full justice to Broome’s contributions to philosophy in a single volume. Nonetheless, we feel that the essays collected here provide some sense of the breadth and depth of his influence on the discipline over the last 30 years. As former students of Broome’s, we would be remiss if we did not

PREFACE

vii

take a moment to note specially that in addition to the general influence he has had on the profession, he was, and remains, an important influence on us, his students. A tremendous teacher and supervisor, we owe more to him than we can repay. It is our very great pleasure, along with our contributors, to present this volume to John Broome in celebration of his tenure as White’s Professor of Moral Philosophy at the University of Oxford. IH & AR June 2014 Montre´al

Notes on the Cover Drawings Young John (front): I suppose you could say that John and I met because of the Viet Nam war. We both happened to be in Cambridge, Massachusetts, in October 1969, and joined thousands of others on a march to Boston Common in order to protest against the war. A group of students from the UK had organized a party to round off a day of peace and love, and John and I went along. Evidently it worked for us. The drawing came about as a result of an interminably long discussion between John and a fellow MIT economics graduate student, Ron Smith, which sent me in search of paper and a pencil. The only paper available was a type of lightweight paper known as “onion skin”, which was used for carbon copies. It’s no small miracle that the drawing has survived to this day. Older John (back): The story behind Older John is not half so romantic. It just seemed fitting that having made a sketch of him at the outset of his academic career, I should mark his retirement from full-time academic pursuits with another drawing. What I discovered was that it is much easier to draw a young person than an older one. Ann Broome June 2014 Oxford

Contents Notes on Contributors My Long Road to Philosophy John Broome

xiii 1

Part I. Weighing 1. Liberty, Preference Satisfaction, and the Case against Categories Geoffrey Brennan

11

2. Challenges to the Principle of Personal Good Douglas MacLean

31

3. Metasemantics out of Economics? Anandi Hattiangadi

52

4. Separability Iwao Hirose

61

5. The Social Disvalue of Premature Deaths Hilary Greaves

72

6. Being and Wellbeing Krister Bykvist

87

7. On the Social and Personal Value of Existence Marc Fleurbaey and Alex Voorhoeve

95

8. The Affirmative Answer to the Existential Question and the Person Affecting Restriction Gustaf Arrhenius

110

Part II. Reasoning 9. The Meaning of ‘Darn it!’ Luc Bovens and Wlodek Rabinowicz

129

10. Keeping Things Simple Roger Crisp

140

11. Moral Requirements Michael J. Zimmerman

156

12. Reasons for Broome Jonathan Dancy

177

xii

CONTENTS

13. Normative Conflicts and the Structure of Normativity Andrew Reisner

189

14. Reasons and Rationality: The Case of Group Agents Lara Buchak and Philip Pettit

207

15. Weighing Explanations Stephen Kearns and Daniel Star

232

Index

253

Notes on Contributors G USTAF A RRHENIUS is Director of Institute for Futures Studies, Professor of Practical Philosophy at Stockholm University and co-chair of the Franco-Swedish Program in Economics and Philosophy at Colle`ge d’e´tudes mondiales and the Swedish Collegium of Advanced Study. L UC B OVENS is Professor in the Department of Philosophy, Logic, and Scientific Method in the London School of Economics and Political Science. G EOFFREY B RENNAN holds joint Professorships in the School of Philosophy at the Australian National University, the Philosophy Department at the University of North Carolina at Chapel Hill, and the Political Science Department at Duke University. He is Director of the UNC/Duke PPE Program. J OHN B ROOME is Emeritus White’s Professor of Moral Philosophy and an Emeritus Fellow of Corpus Christi College at the University of Oxford, an Adjunct Professor of Philosophy at the Australian National University, and a Visiting Professor at Stanford University. L ARA B UCHAK is Assistant Professor, University of California at Berkeley. K RISTER B YKVIST is Professor of Practical Philosophy at the Department of Philosophy, Stockholm University. R OGER C RISP is Uehiro Fellow and Tutor in Philosophy at St Anne’s College, Oxford, and Professor of Moral Philosophy at the University of Oxford. J ONATHAN D ANCY is Professor of Philosophy at the University of Texas at Austin and Senior Research Professor at the University of Reading, UK. M ARC F LEURBAEY is Robert E. Kuenne Professor of Economics and Humanistic Studies, Woodrow Wilson School and University Center for Human Values, Princeton University. H ILARY G REAVES is Associate Professor of Philosophy at the University of Oxford and a Fellow of Somerville College, Oxford. A NANDI H ATTIANGADI is Professor of Philosophy at Stockholm University and Pro Futura Scientia Fellow at the Swedish Collegium of Advanced Study. I WAO H IROSE is Associate Professor at the Philosophy Department and the School of Environment, McGill University. S TEPHEN K EARNS is Assistant Professor at Florida State University.

xiv

NOTES ON CONTRIBUTORS

D OUGLAS M AC L EAN is Professor of Philosophy and Adjunct Professor of Public Policy at the University of North Carolina at Chapel Hill. P HILIP P ETTIT is L. S. Rockefeller University Professor of Politics and Human Values, Princeton University, and Distinguished Professor of Philosophy, Australian National University. W LODEK R ABINOWICZ is Professor Emeritus of Practical Philosophy at Lund University, Centennial Professor at the London School of Economics and Political Science, and Long-Term Fellow of the Swedish Collegium of Advanced Study. A NDREW R EISNER is Associate Professor in the Philosophy Department at McGill University. D ANIEL S TAR is Assistant Professor at Boston University. A LEX V OORHOEVE is Associate Professor at the Department of Philosophy, Logic, and Scientific Method in the London School of Economics and Political Science. M ICHAEL J. Z IMMERMAN is Professor of Philosophy at the University of North Carolina at Greensboro.

My Long Road to Philosophy John Broome

The English school system encourages specialization in later years at school. From the age of 14, I studied little apart from mathematics and physics. I aimed to be a physicist, but I was told that to achieve that aim it was better to take an undergraduate degree in mathematics rather than physics. So in 1965, I went to Trinity Hall, Cambridge, to take the Mathematics Tripos. From my undergraduate supervisor in Trinity Hall, Anthony Pearson, I learnt one thing that has stuck with me. Each week he asked me to return the following week with a proof of some theorem or other in analysis. These theorems were all obvious, as are most things in analysis, including several things that are false. I generally returned with a proof along the lines of, ‘Well, it’s obvious’. After some weeks, I came with a proof like that of the intermediate value theorem (which is obvious, true, and not very easy to prove). My supervisor was enraged and thumped on the table. He demanded proper rigorous proofs in the future. This showed me the standards of rigour that are demanded in mathematics, and I tried to meet them ever after. We cannot attain these standards in philosophy, but I try to make my arguments as rigorous as I can. Despite this important lesson, I got bored of mathematics during the first year, and decided to change to some other subject. I think the first year of the course at Cambridge must not have been well designed at that time. I read recently that the same thing happened to Angus Deaton, an economist whom I came to know later. I sometimes think that perhaps I should have overcome the boredom and stuck to mathematics; I was quite good at it. I did not know what to do instead. I made a tour of various departments to see what attractions they had. I investigated philosophy. Since there was no philosopher in Trinity Hall, I was sent to talk to one in the neighbouring college. He said that, if I was thinking of studying philosophy, I would do better to leave the university and work as a road builder for a few years. He said that would put the idea of doing philosophy out of my head. So I decided not to take philosophy. Long afterwards I learnt that he was merely copying what Wittgenstein used to say to prospective students.

2

JOHN BROOME

Instead I chose economics, which welcomed people with mathematical experience. My career as an economist lasted 30 years, from 1966 to the beginning of 1996, when I started my first permanent, full-time job in philosophy. Perhaps I should have worked a few years on the roads instead. At Cambridge, my supervisor in economics was Jim Mirrlees. He was awarded a Nobel Memorial Prize in 1996. He was the first of my Nobel-Laureate supervisors. He suggested I go to MIT to do a Ph.D. in Economics. MIT generously supported me with a fellowship. I arrived there in 1968. This was in the heyday of welfare economics: the branch of economics that deals explicitly with the ethical foundations of valuation and decision-making within economics. A decade later, welfare economics went out of fashion. This was one consequence of the reactionary swing in politics that brought Ronald Reagan and Margaret Thatcher to power. The subject dropped out of the economics curriculum. Now I find that, although I am no longer an economist, I know more about welfare economics than many of the younger economists I have dealings with. For example, I know about such important subjects as the discount rate and the value of equality. Welfare economics is the philosophical branch of economics. Indeed, it is really a branch of applied ethics. It is a sort of formal ethics, and its methods can make a valuable contribution to ethics itself. During my years as an economist, many of my writings in welfare economics were also intended to contribute to ethics. I enjoyed MIT very much. The Economics Department has a reputation for competitiveness, but I found both students and faculty members kind and friendly. Arriving when I did put me into the centre of the student protests against the Vietnam War. The academic year 1969–70 was filled with strikes and protests. On 15 April 1970, I was in a full riot near Harvard Square, dodging the plumes of tear gas by means of fast footwork, and avoiding being clubbed down only by means of good luck. When the National Guard killed four students at Kent State University in May, all faculty and students at MIT went on strike for the rest of the academic year. These events brought me my lovely wife Ann, as she explains in her note above, and they were a radicalizing experience. I dropped out of MIT for a full year. When the Economics Department was kind enough to let me drop back in again, I fondly imagined it would allow me to write a dissertation on William Godwin, the anarchist political philosopher who wrote at the time of the French Revolution. Indeed, it would have allowed me to, had I been able to find enough economics in Godwin to write about. But after a semester of trying I realized I could not. In the end I wrote a thesis on general equilibrium theory. The advantage of this mathematical topic was that, since I was lucky enough to hit on a theorem I could prove, I wrote the thesis very quickly. My main supervisor was Duncan Foley, who originally introduced me to general equilibrium theory. Robert Solow was also on my dissertation committee, and was particularly patient over my struggles with Godwin. He very kindly lent my wife Ann and me his lovely house in Concord for two summers. The third was Paul Samuelson. Solow and Samuelson brought my score of Nobel-Laureate supervisors to three.

MY LONG ROAD TO PHILOSOPHY

3

While registered at MIT, I took some courses in philosophy at Harvard. One was a course on liberalism from John Rawls. I am sorry to say I did not enjoy it much. The other was Stanley Cavell’s course on Wittgenstein. This was thrilling. Cavell was a charismatic lecturer. I watched him apparently struggling at the lectern with profound philosophical difficulties, and was gripped. (True, I did not understand much.) This is why I eventually became a philosopher. I told this story about Cavell when I gave some lectures at Harvard in 2011. Cavell was unwell and not at the lectures. He did not know my story, since he and I had little personal contact when I took his course. But when it was repeated to him, he took the trouble to come to a reception, and I was able to thank him. My means of becoming a philosopher was through David Wiggins, who was visiting Harvard at the time. He invited me to take an MA in Philosophy at Bedford College in London University, where he was the professor. I owe David a great deal for giving me this opportunity. My first job was in the Economics Department at Birkbeck College in London—a new department that was being created by Professor Bertie Hines. Bertie had managed to persuade the college that the department needed to be fully staffed for a full year before the students arrived, to buy the library and design the courses. The result was that I had plenty of time for philosophy that year, and completed the MA in 1973. David Wiggins and Anthony Savile were my main supervisors. While taking the MA under David’s influence I developed a deeper interest in the philosophical foundations of valuation in economics. In particular, I became interested in the value of human life—a subject that involves both economics and philosophy. This started a line of work that was my principal occupation until 2004, when it was finally completed in my book Weighing Lives. This MA was my only qualification in philosophy until Lund University bestowed an honorary degree on me in 2013. It was not enough to give me a broad knowledge of philosophy, and even now I remain ignorant of important subjects, including ancient philosophy. Unfortunately, it was also not a sufficient qualification for getting a job teaching philosophy. I soldiered on as an economist. During the 1970s, life in the Birkbeck Economics Department was fraught. There was conflict between the Marxists and the others. Actually, the conflict was more imagined than real, but we did spend a lot of time in department meetings. I was relieved to get a job as Reader at the Economics Department in Bristol University from the beginning of 1979. Bristol Economics was always harmonious. By the time I moved there, I had embarked on writing my first book. It was a textbook on economics done in the Ricardian style, entitled The Microeconomics of Capitalism. Ricardo’s economics analysed the contribution of different classes to production, and the distribution of income between classes. It was a precursor to Marx’s work. My book was a general equilibrium analysis built on a Ricardian basis. By the time it came out in 1983, the political reaction was well under way. Thatcher had transformed the politics of Britain, and there was no market for a book of this

4

JOHN BROOME

sort. Yet I am pleased I wrote it, even though few people have read it. I remember thinking at the time that writing a textbook was a useful exercise in writing. You have to struggle to put things in the clearest possible way. Unexpectedly, I received a letter from Derek Parfit, suggesting I should apply to visit All Souls College in Oxford. The college made me a Visiting Fellow for the academic year 1982–3. This was a great honour, and a great opportunity to develop my philosophical interests. During my year at All Souls, Derek was engaged in the painful process of producing Reasons and Persons. I had the privilege of watching his thoughts develop as he wrote. Afterwards, he gave me an exceptionally generous acknowledgement in his preface, which gained me credit among philosophers. Derek has benefited me greatly at several points in my life, and I am immensely grateful to him. Next, also out of the blue, Geoff Brennan generously invited me to the Australian National University. I went for six months in 1986. At the time, Geoff, like me, was employed as an economist but had strong philosophical interests. I taught economics during that visit, but it gave me my first contact with the Philosophy Department at the Research School of Social Sciences. Since then I have visited the RSSS with everincreasing frequency. (For a while Geoff was its director.) It is a wonderful place for philosophy, especially during the winters when philosophers from the northern hemisphere congregate there, sacrificing their own summers for the sake of the philosophical conversation in the RSSS tea room. Each time I go, I meet many new and old friends. Geoff introduced me to Philip Pettit during my first visit to Australia, and Philip has been another great philosophical friend to me ever since. Bristol was very tolerant in allowing me leave during the 1980s. It was actually the result of the department’s deliberate policy: we had decided that our means of weathering the Thatcher cuts was to allow people to take unpaid leave whenever they had the opportunity. I took full advantage of the opportunity to develop my experience of philosophy. Almost immediately after getting back from Australia, I went to Princeton in 1987 and 1988. This visit was arranged by Angus Deaton, another friend to whom I am indebted for a great deal of help and support. Angus was in Bristol before moving to the Woodrow Wilson School in Princeton. I worked half-time in the Woodrow Wilson School and half-time in the Philosophy Department for two semesters, and then for one semester full-time in the Philosophy Department. That was my first job in philosophy. During that last semester I taught the main undergraduate course in ethics. I was really not qualified to do so; I had never even taken a course in ethics. I was anxious about taking it on, but Michael Smith, who was working at Princeton at the time, encouraged me. I am pleased he did. I don’t think I taught the course very well, but it made me feel that I might sometime be able to work as a philosopher. Like the RSSS, Princeton is another wonderful place for talking philosophy. At that time, members of the faculty had the habit of attending each other’s graduate

MY LONG ROAD TO PHILOSOPHY

5

seminars, which led to a flourishing philosophical conversation. I went to David Lewis’s seminar on realism and witnessed the power of his intellect. He spoke perfectly coherently, without notes, for sometimes two and a half hours before noticing it was time for a break. By the time of my stay in Princeton, my interest in the value of life had led me to take up decision theory. I was fascinated by a theorem published by John Harsanyi in 1955, which provided some support for utilitarianism on decision-theoretic grounds. The graduate seminars I gave at Princeton were about the connection Harsanyi made between decision-making under uncertainty and the distribution of wellbeing between people. They formed the basis of my second book Weighing Goods. I met Harsanyi in 1988, and got to know him when he visited Bristol in 1991. In Princeton, I learnt a lot more decision theory from Dick Jeffrey, and also from Brian Skyrms, who was visiting at the same time. The first time I met Doug MacLean was during my stay in Princeton. Doug is a philosopher who takes very seriously the connection with economics. We were together at the RSSS in 1993. We talked a lot. I particularly remember a weekend with Doug, his wife Susan Wolf and other philosophers at Geoff Brennan’s house by the New South Wales coast, which has a spectacular view to the sea. Doug remains one of my main sources of excellent philosophical advice. Every two years he brings together a group of moral philosophers for a conference at some nice place in Europe. Through him I have been able to forge friendships with some of the best moral philosophers. Back in Bristol from Princeton, I finished Weighing Goods. It was published in 1991. This book is more like philosophy than economics, and it bought me my ticket into philosophy. Bristol recognized it by letting me move half-time into the Philosophy Department. I was already a frequent visitor at seminars there, where I picked up a lot of philosophy from Adam Morton, Christopher Williams, and others. Also as a consequence of the book, I started to get more invitations from philosophers. Jonathan Dancy invited me to Keele—the beginning of another long friendship in which I have been the beneficiary of a lot of philosophical wisdom. I was very excited to receive an invitation from Sven Danielsson to give the Hägerstro¨m Lectures in Uppsala in 1994. There is a fine tradition of highly analytical, even formal, moral philosophy in Sweden. Sven was a leading figure in this movement, though he was not himself so keen on decision theory. On that first visit I met Wlodek Rabinowicz, who has been a close and generous philosophical friend ever since. I have subsequently spent a lot of time in Uppsala at the Swedish Collegium for Advanced Study. This exceptionally open-handed institution has several times brought together philosophers of similar interests for long periods. One winter I was able to go cross-country skiing with Wlodek almost every afternoon, directly from SCAS. The philosophical conversations tended to slow us down, so we were sometimes out till near dark.

6

JOHN BROOME

Then I was offered a Professorship of Philosophy at St Andrews. I took it up on 1 January 1996. This was John Skorupski’s doing. It was a bold, risky move on his part, for which I shall be eternally grateful. He made me officially a philosopher. Only a few years later I was offered the White’s Professorship Oxford. My career has contained a lot of good fortune, but that was the most fortunate, amazing and exciting thing that has happened in it. I was overwhelmed to learn that Oxford was willing to trust someone who was so far an outsider from Oxford, and even an outsider in philosophy. I still did not feel secure in the role of philosopher. For many years I felt a sort of imposter, concealing my lack of a proper philosophical education. Several of my colleagues still think I am just an economist in disguise. But now I am old enough to be no longer bothered. In any case, once I had a job in the discipline, various processes helped me to absorb it. For one thing, I learnt from and with my students in St Andrews and Oxford. I gained a lot from the time I spent with Iwao Hirose and Andrew Reisner in both places, and from a sequence of many other excellent students. And, though I lack a philosophical education, I can bring to moral philosophy something that many of my colleagues lack. The precise analytical, indeed mathematical, methods of economics can be very useful in the philosophy of value. Some questions of value cry out for them. The value of equality is an example: it is a quantitative matter involving relations among many different people. Moreover, it was explored by welfare economists long before philosophers became interested in it. Much recent work in philosophy about the ethics of equality has suffered from the lack of a grounding in the methods of economics. After leaving my job in economics, I still had work to do within the theory of value. I was still working on the value of life and the value of population until I brought that very long programme to an end in Weighing Lives. But even before finishing that book, I found myself moving further away from subjects that have a close connection with economics, and shifting my attention away from the philosophy of value. I started working in the philosophy of normativity, which is not directly concerned with goodness, but with ought, reasons, reasoning, reason, and rationality. My immediate stimulus was an invitation from Derek Parfit to respond to a paper of his entitled ‘Reasons and Motivation’ at a Joint Session of the Mind Society and the Aristotelian Association. But my interest in these topics started a decade earlier at Princeton, when Michael Smith told me that he found the subject of motivation the most exciting thing in moral philosophy. Or really it started during my MA course in philosophy, when I read Thomas Nagel’s The Possibility of Altruism. The philosophy of normativity has been my principal interest for the last 15 years. It led to my book Rationality Through Reasoning, and I expect this to be my principal interest in the future. But another turn was still to come. When I became a philosopher, I thought I had left behind the actual practice of economics. But in the last few years I find myself

MY LONG ROAD TO PHILOSOPHY

7

engaged with economics again. In 2005, Nick Stern asked me to write something about the ethics of climate change as a small contribution to the work of The Stern Review. I thought I had said my piece on this subject after publishing a book about it in 1992, but, as a consequence of writing this piece for Nick, I soon found myself deeply involved again. I was commissioned to write another book, and I became a Lead Author for the Intergovernmental Panel on Climate Change. Climate change raises important and difficult problems of value. For example, it raises problems about life and death and how to set a value on these things, since climate change is killing many people and will affect the world’s population. Because of the scale and complexity of the problem, the methods of economics have to be used in making the valuations, but these methods need to be founded on an account of what is truly valuable. This can only come from the theory of value within moral philosophy. Moral philosophy is crucial to dealing with climate change. Still, the most important thing I learnt about climate change is a matter of economics. A few years ago, my doctoral supervisor Duncan Foley wrote to remind me of something that should have been obvious to me: since greenhouse gas is an externality, it creates inefficiency. The externality could therefore be corrected without any sacrifice on anyone’s part. Yet the process of international negotiation seems to demand sacrifices from the current generation for the sake of future people. The process gets nowhere because governments will not accept sacrifices on behalf of their people. I think the negotiations should be restructured. Governments should see themselves as negotiating about the distribution of benefits rather than sacrifices. The realization that no sacrifice is required may be the way to unlock the process and move negotiations forward. My road to philosophy seems to have turned in a circle: I am once again learning from the supervisor of my student days in economics.

PART I

Weighing

1 Liberty, Preference Satisfaction, and the Case against Categories Geoffrey Brennan

I Introduction In the introduction to his 1999 collection of essays Ethics Out of Economics, John Broome draws a distinction between categorical and comparative thinking. According to him, ‘Economists . . . think naturally in comparative terms’, whereas ‘Philosophers do not have the same instinct to think comparatively’ (9). Thinking comparatively is one of the ‘areas where the methods of economics can contribute to ethics’: it is one of the things that ‘philosophers can learn from economics’ (1, 9). I think Broome is broadly right, on both descriptive and normative fronts. • Economists are more likely than philosophers to think comparatively. The methods of philosophy proceed naturally in terms of categories—such as true /false, good/bad, right/wrong, permissible/impermissible. Economists are more likely to work with comparative or relative concepts such as epistemic warrant, better/worse, degrees of permissibility, and so on. • And comparative thinking does have some advantages—especially in ethical reasoning, which is the main focus of Broome’s attention in the aforementioned essays. Broome’s point might well generalize to other domains but his primary concern in the context of these essays is the distinction between goodness (a category) and betterness (a comparator). Broome’s argument for betterness over goodness seems to have several strands. One of his important claims is that goodness is fully reducible to betterness—as if betterness is in some structural (or perhaps ontological) sense the more basic concept. There are suggestions that he thinks betterness is less epistemically demanding, since it requires one to specify an intercept term as well as (or instead of?) a slope term in the relevant value function. There seem to be yet other considerations involved in the argument surrounding his main example of thinking about the cost of death.

12

GEOFFREY BRENNAN

Here I shall be concerned with none of these strands, but with what I see as a yet further consideration—one that Broome does not appear to emphasize in this setting,1 but which seems to me important—namely, the different ways in which goodness and betterness frame how to think about aggregating various different ethical components/dimensions/considerations to form all-things-considered judgements. If each of a number of ethically relevant considerations is specified comparatively—that is, in terms of degrees of accomplishment—then one is led to think of all-things-considered judgements as a matter of trade-offs. By ‘trade-offs’, here, I mean specifically marginal adjustments, in which some of one valued thing is given up in order to secure a bit more of some other valued thing.2 If the component elements of overall ethical desirability are specified categorically, the possibility of trade-offs in this sense is occluded: indeed, trade-offs seem to be ruled out by formulation! Now, when Broome observes that economists tend to think comparatively and philosophers categorically, he himself is making a comparative claim: economists are more likely to think comparatively than philosophers are. This does not mean, of course, that philosophers never think comparatively, or that economists always do. Indeed, my main concern in this essay is with what I see as one notable exception to the economist’s comparative propensities. It is an exception that is important, because it lies in an area where economics and moral philosophy intersect. I am referring here to ‘social choice theory’. In my view, social choice theory is best viewed as an exercise in analytic ethics (rather than in economics as such3). There are no empirical claims in social choice theory; no assumptions about how the world works; there is no exchange; there are no relative price effects. Social choice theory is an application of certain mathematical techniques to an issue in aggregation—and the fact that the originators of social choice theory were people who self-identified as economists and occupied jobs in economics departments could be seen as a kind of

1 It is certainly relevant in other work—and specifically to ‘Weighing Goods’ (Broome, 1991), though perhaps for consistency one should talk of ‘weighing betternesses’? 2 I find it interesting that at least some influential philosophers use the term ‘trade-off ’ to refer to ‘choice’. To take a random example, Taurek (1977) in a widely read paper, refers to ‘trade-off situations’ as constituted by choices that involve saving a larger or smaller number of people (or none at all). But there is no ‘trading-off ’ in Taurek’s predicament—at least not as economists would understand the term: Taurek’s situation is simply a matter of choice. Faced with a choice between X and Y, economists would be inclined to think of a ‘trade-off ’ as seeking intermediate positions—a little more of X and a little less of Y, or vice versa—rather than the act of choice itself. 3 I have no instinct to be essentialist about disciplinary divides. On the contrary, I think that those divides have a significantly conventionalist aspect. On the other hand, disciplinary identity is meaningful to the participants in academic settings. As myself one who strays rather indiscriminately across disciplinary boundaries, I am familiar enough with claims such as, That’s not Economics, or, This paper is interesting enough; but it is not philosophically interesting. And such claims are by no means bereft of meaning. Indeed, in some senses, the logic of the disciplinary division of labour of which Economics and Philosophy are a part requires a tolerably clear sense of what specialization is entailed and how disciplinary boundaries might plausibly be demarcated.

THE CASE AGAINST CATEGORIES

13

historical accident4 rather than any comment on the nature of a proper disciplinary division of labour. In any event, in this chapter, I want to illustrate the disadvantages of thinking categorically by appealing specifically to the social choice case. The argument will be mainly focused on Sen’s ‘liberal paradox’, rather than on Arrow’s famous ‘impossibility theorem’, though the cases are analogous with respect to the point of concern here. My point here relates to ‘impossibility theorem logic’ rather than to any specific application. I want to insist that the social choice approach involves categorical rather than comparative reasoning, and that this fact has consequences for the kinds of results—and the interpretation of those results—that emerge from social choice theory. Alternatively put, I want to suggest that if the issues at stake were expressed in terms of a more comparative mode of reasoning, there would be no impossibility theorems as such at all—and that the way we would think about how to respond to the problems revealed by social choice analysis would be rather different. However, because the Arrow theorem is more familiar and both precedes and heavily influences Sen’s formulation, it will be useful first to make some brief remarks about the social choice approach (SCA henceforth) by reference to the Arrow theorem (Section II). I shall then turn to the Sen result and attempt to show, by appeal to a few simple diagrams, what is lost by virtue of SCA’s categorical character (Section III). I shall then go on (in Section IV) to explore the normative issues with which Sen is concerned, and in Section V briefly consider an alternative (appropriately modified) framework associated with the no less famous work of Ronald Coase (1960). Section VI will offer some simple conclusions.

II The Arrow Case The general lineaments of Arrow’s impossibility theorem are well known and do not need to be rehearsed in any detail. However, it will be useful to underline the structure of the argument in order to make salient the features I wish to discuss. The problem Arrow confronts is this. Suppose that the only information available about individual preferences is the rankings of choice options over a given domain. The task is to derive from those rankings an ‘aggregate’ or ‘social’ ranking that has certain normatively desirable properties. Call these properties the ‘Arrow desiderata’ (AD, henceforth). On the face of things, there seems to be nothing especially exceptionable about this kind of normative exercise; but I want to draw special attention to the fact that the normative element is formulated by reference to 4 Part of the historical contingency relevant here is the fact that the doctrine of ‘impossibility of interpersonal utility comparisons’ had a significant impact in Economics—and almost none in Moral Philosophy. The social choice framework focuses on aggregation of rankings precisely because finer-grained interpersonal comparisons of preference are ruled out. Although the point is not perhaps foregrounded in social choice formulations, the impossibility of inter-personal utility comparisons is a core piece of the entire approach.

14

GEOFFREY BRENNAN

desirable properties—to conditions that have to be fulfilled. These conditions are, I believe, categorical in Broome’s sense. I shall denote the various AD by P, T, U, D, I.5 It is helpful here precisely not to describe the content of these desiderata. What I am interested in is the structure of the argument. And in this sense, notice two things: 1. That it is these desiderata that constitute the normative component of the whole exercise. Perhaps this is not obvious. Perhaps one is tempted by the thought that there is normative content in the individual rankings, which the aggregate ranking is required to reflect. But I take it that any normative content in individual rankings derives, within the theoretical construct, purely from the P requirement. It is the P requirement that is supposed to capture the relation between the aggregate ranking and the individual rankings. 2. That the desiderata are expressed in on/off terms. That is, the AD are a set of requirements that either are, or are not, met. There is, to use Broome’s language no specification of betterness in terms of more or less P, more or less T, more or less U, and so on. The AD are stipulations: it is an essential part of the exercise that it would be good if each of the AD were fulfilled, but all of them are formulated as sentences that can be true or false in relation to any particular aggregation procedure. Now, as is well known, the Arrow theorem states that there is no mapping from individual rankings to an aggregate (or social) ranking that satisfies all these requirements—hence the impossibility result. It turns out that if we drop one of the desiderata (any one will do) then a mapping can be found and a resultant aggregate ranking derived. But what are we to make of this result? And what are its implications for practical purposes? The range of responses to these questions is considerable. On the one hand, there are those who think (with Buchanan (1954), say) that the impossibility result is essentially obvious and that no one should ever have thought that the individual attributes of rationality would transfer to the collective case. Others (like Tullock (1967)) think that conflict between the desiderata, though it cannot be ruled out, is sufficiently unlikely to be tolerable. On the other hand, there are those who regard the result as disastrous news for democracy. Consider for example Peter Ordershook (1986: 65): ‘It is not stating the case too strongly to say that Arrow’s theorem and the research that it inspired wholly undermine the general applicability or meaning of concepts such as the “public interest” and “community goals”.’ Or Gerry Mackie’s (2003: 2) remark in a general

5 P for Paretianism; T for transitivity; U for unrestricted domain (i.e. all possible individual rankings are in principle admissible); D for non-dictatorship; I for independence of irrelevant alternatives; and so on. I leave these undefined because it would be a distraction here to become pre-occupied with the conditions or the theorem itself.

THE CASE AGAINST CATEGORIES

15

commentary on US political science: ‘the main intellectual trend in American political science is the view that democracy is chaotic, arbitrary, meaningless and impossible. This trend originated with economist Kenneth Arrow’s impossibility theorem.’ In Mackie’s judgement, the Ordershook position and others like it represent the predominant view. We should perhaps note, specifically in relation to Ordershook’s claim, that the P condition is supposed to reflect the idea of the public interest. Public interest, here, is defined by reference to the satisfaction of individual preferences, and the P condition ensures that individual preference satisfaction is maximized in the sense that the social ranking tracks individual rankings with respect to any options over which the individual rankings are common. The P condition purports to go as far as one can, in a world where interpersonal comparisons are ruled out, to ensure that outcomes that maximally reflect individual preference satisfaction are more highly ranked by reference to the aggregate ranking. It is therefore difficult to see how the Arrow theorem as such shows that the concept of the public interest is ‘meaningless’: after all, one important feature of the Arrow theorem is the demonstration that P can always be achieved (by forgoing one (or more) of the other AD). More generally, consider what the ‘impossibility theorem’ actually says. It shows us that not all the desiderata can be achieved. Something has to give! But what form does the possible give in the logical system take, given the structure of the formulation? The only thing that is available to provide give in the system is releasing the grip of one or other of the desiderata. So the only question we can ask is—which of U, P, T, I, or D should it be? Now of course, when you give up on a condition or property, you give up on it in its entirety. It is as if, whatever value non-dictatorship (or whichever desideratum) is supposed to reflect, that value only registers when the condition is completely realized. This is the force of categories: they have value if they are fully met, but, if they are not fully met, then all bets are off—any instance of non-fulfilment is as good (or as bad) as any other.6 Suppose instead we were to approach the issues that Arrow is concerned with by resolutely adhering to the (Broomean) idea that all normative desiderata ought to be specified in comparative terms. How could this be done? The first step would be to indicate, for each of the requirements imposed on the aggregate ranking, the underlying value that that requirement is supposed to promote. So, in lieu of the ‘unrestricted domain’ requirement, we devise a measure of ‘degree of completeness’, which would have its maximal value when the U requirement was satisfied. Analogously, we might form a metric, P0 , of ‘Paretianism’—a metric that reflects the degree to which a procedure fails to meet the Pareto criterion (where P0 is greater the closer an 6 One might think of categories as corresponding to a discontinuity in value—but with major and minor failures differentially valued. I do not think anything in what follows would be seriously affected by this modification.

16

GEOFFREY BRENNAN

aggregation procedure comes to fulfilling P). An analogous exercise would then be performed, in each case, for transitivity, independence of irrelevant alternatives, and non-dictatorship. Now, we know from the Arrow theorem that the maximal values of U0 , P0 , T0 , I0 , and D0 specified by the respective conditions cannot in general be achieved simultaneously. The question then is: What is the appropriate response to that fact? The economistic answer would seem to be that we will need to search out, and eventually settle on, the optimal trade-off between the values so expressed. Interestingly, extrapolating from what we know of other cases where such trade-offs are required, it seems likely that this optimal trade-off will require giving up on the ideal with respect to all of the desiderata (U, T, etc.). We will, as the economist might put it, make adjustments at all the margins. Corner solutions in which one or more value trumps all others are certainly imaginable, but equally certainly, are not logically required—and probably not especially plausible. Think of a prosaic case. You like (value) apples; and you like (value) bananas. Your ideal level of banana consumption is ten per week. And your ideal level of apple consumption is seven. But you do not have enough money to buy seven apples and ten bananas. What do you do? One thing you do not do is to conceive your decision problem as a kind of ‘impossibility theorem’ in the fruit domain. In settings like this, you consider alternative bundles of bananas and apples that you can afford and take the bundle that looks best to you. This will not normally involve your giving up bananas or apples in entirety. You will not standardly say to yourself: Well, since I can’t have the ideal number of apples/bananas, I’ll have to declare myself as either a banana man or an apple man and just forget about the particular fruit in respect to which I decide not to achieve my ideal. You trade off objects (or dimensions) of desirability at the margin until you find the best outcome. You will normally give up some of both. Why might this simple analogy fail to apply in the case of choice over the optimal aggregation rule? Three possible answers: 1. Choosing an aggregation rule is not like choosing a bundle of fruit. The aggregation rule choice has a universalizable character—it has to apply in common across all persons. It is a public good—not a private good. 2. The desiderata are normative—they purport to reflect ethical desirability; fruit choice is merely a matter of individual tastes. 3. Analytic constraints are not like resource constraints. We can at least imagine that you might have enough money to handle your ideal fruit consumption levels: there is nothing that rules that possibility out—except of course the assumed brute fact of your relative poverty. However, we cannot imagine that a proposition and its negation can both be true. The Arrow theorem expresses an analytic truth. Resource constraints do not have that status.

THE CASE AGAINST CATEGORIES

17

I think each of these answers has some force. They draw attention to respects in which ordinary consumer choice and ethical evaluation are not exactly analogous. But I do not think that any of these totally undermine the comparative approach I have outlined. Part of what follows will be an attempt to support that claim. In particular, I do not think that the formulation of the AD is driven by necessary features of normativity as such. Of course, it may be that some values do have the categorical character that the SCA assumes. If so, this might be one way in which ethical desirability is different from ordinary preference or ‘taste’. But in fact, I think that this suggestion is wrong as a general proposition and shall argue against it. Of course, it might be that ethical desirability is not categorical in general, but that ethical desirability happens to be like this in relation to desiderata for aggregation rules.7 But if that is what social choice theorists believe, then one might have thought that the theory ought to offer some arguments as to why that is so. No such arguments are forthcoming. My claim is that the categorical nature of the AD is just a matter of formulation. It is not so much that the alternative comparative procedure is explicitly rejected: it is rather that the categorical approach is hospitable to the analytic tools to which Arrow and followers are attracted. Now, a social choice theorist might respond at this point that, if I am to urge the comparative approach, it is incumbent on me to indicate how plausible metrics of the various desiderata could be developed. And, of course, that is no mean challenge. In some cases, there are obvious possibilities: so, for example, in the case of the completeness condition, one might measure U0 as the proportion of possible individual rankings that have to be ruled out; and in the case of P0 , welfare economists routinely employ measures of closeness or distance from the Pareto optimality frontier.8 But in other cases, I confess I do not have even poorly developed notions. Nevertheless, what I do insist is that, if we are to offer any plausible answer to the question as to what the ‘best’ aggregation rule is, such metrics need to be developed. On its face, merely indicating which of the various desiderata should be dropped— even showing that different ones might qualify in different circumstances9—is not a generally appropriate approach. For the purposes of the argument here, however, I shall develop the argument in the context of a more restricted example—that of Amartya Sen’s ‘impossibility of a Paretian Liberal’ (1970). The reason for focusing on this case is that—whatever the situation is with transitivity, non-dictatorship, unrestricted domain, and the like—

7 Virginity is often suggested as a natural instance of something that has this character—though, even assuming that virginity is ethically desirable, it is far from obvious that one episode of excessively generous affection is in the same class as a life of total debauchery. 8 Though such measures are themselves not entirely unproblematic. I will discuss the Paretian case in more detail below. 9 See Brennan and Baurmann (2006) for an attempt along such lines.

18

GEOFFREY BRENNAN

the prospects for developing appropriate metrics of liberty and Paretianism seem more tractable; and relatedly, the notion that liberty and Paretianism (or preference satisfaction) do come in degrees has considerable intuitive appeal. As far as I can see, that intuition is all I need to make the points I seek to make. To summarize, I want to insist that the fruit consumption analogy is the right way to think about the trade-offs implied by the Arrow theorem. And I want to suggest two things about this comparative formulation: 1. It suggests that the conflict between the various ADs that Arrow’s theorem reveals is not to be thought of as an impossibility result, but as an optimization problem (and one, moreover, that is in broad terms, quite familiar). The impossibility result does not ‘undermine the general applicability or meaning of concepts like the public interest or community goals’, as Ordershook suggests: it simply shows that community goals come at a cost in terms of other values forgone. This latter is not by any means a negligible result—but it is not an occasion for any kind of conceptual despair. 2. The appropriate response to the optimization predicament that Arrow exposes is not, in general, to choose one of the desiderata to give up on, but rather to examine the terms of trade between the associated values (at the relevant margin). The presumption is that, when we achieve the optimal response we will give up on all the conditions—to some appropriately small extent. In what follows, I want to examine these same issues in relation to Sen’s (1970) claim about the impossibility of a Paretian Liberal.10 I want to ask what Sen’s claim amounts to in principle—and what it tells us about the relation between preference satisfaction (Paretianism) and liberty. My claim is that the impossibility result is unhelpful—that it misleads us into drawing mistaken inferences—and that it does so largely because it follows the SCA by specifying values as categories, rather than in terms of more or less. The Sen setting offers, in short, a neat way of illustrating the inadequacies of categorical ethics.

III Sen’s Agenda and the Dubious Relevance of ‘Impossibility’ Economists have long believed that liberal institutions and economic flourishing go hand in hand. Adam Smith’s ‘system of natural liberty’ was justified by Smith not in terms of the value of liberty as such but because it leads (on Smith’s analysis) to an

10 There is a significant literature on Sen’s ‘paradox’, which is nicely reviewed in Mueller (2003). Most of this literature is, however, second-order for my purposes because it basically accepts the social choice framework in which Sen is here operating. It will also be sufficient to focus on the original paper, rather than Sen’s own subsequent evaluations and restatements.

THE CASE AGAINST CATEGORIES

19

increased wealth of the nation and indeed of the entire community of exchange. And so it is that, in contemporary circles, economists will typically argue for liberty not so much on the basis of the intrinsic value of liberty, but on the basis of maximising ‘preference satisfaction’. Milton Friedman’s popular book Free to Choose (1980) or James Buchanan’s Freedom in Constitutional Contract (1977) exemplify this approach. In this context, Amartya Sen’s claim to show the impossibility of a Paretian Liberal represents a significant challenge to mainstream convictions. The implication of Sen’s result is that there is, in fact, a deep conceptual conflict between liberty and preference satisfaction—one that Sen’s impossibility theorem sought to make plain. There are many questions about Sen’s argument that have surfaced in the (now extensive) literature: whether Sen expresses the notion of ‘rights’ accurately; whether he adequately captures the notion of a specifically liberal rights regime; whether he implicitly assigns (liberal) rights the property of inalienability; and so on. All these questions are, in my view, telling ones. But my basic point here is, as far as I can see, independent of all these issues. I am interested in the question as to what impossibility amounts to. To lay the ground for the discussion, it will be useful to appeal to a simple diagram.11 In Diagram 1.1, I depict along the horizontal axis degrees of liberty (L) and along the vertical axis degrees of preference satisfaction (P). Points further from the origin in a horizontal direction have higher degrees of liberty; points further from the origin in a vertical direction have higher degrees of preference satisfaction. L∗ P

T′ T Q

P∗

S S′

L

O

Diagram 1.1

Clearly, in appealing to such a diagram I am assuming that both Paretianism and liberty come in degrees—that it is plausible to talk in terms of metrics of L and P, rather than of L and P as categories. I need to say a little about this claim in each case—a little but not much, because I think the claim is uncontroversial. 11

I confess to a belief in the pedagogical advantages of ‘simple diagrams’.

20

GEOFFREY BRENNAN

Consider first Paretianism, or as I prefer to call it ‘preference satisfaction’. I take it that there is no problem in treating preference satisfaction as a metric for any individual. Social choice theory routinely concedes as much. Suppose there are four social states (a, b, c, d), so ranked by individual A;12 and these same states are ranked (b, c, d, a) by B; and A and B are the only individuals in society. Then, not only can we declare that a and b are Pareto optimal states, because there is no other state that is unanimously preferred by A and B, we can also declare that c is more preferred than d (even though c is not Pareto optimal). Welfare economics generally purports to offer measures of ‘distance from the Pareto frontier’, and though many such measures serve to smuggle in implicit interpersonal comparability, economists seem to have been happy enough to use them. I shall say a little more about one such approach in the ensuing section. All I wish to emphasize, at this point, is that the Paretian framework does more than allow us to distinguish between optimal and non-optimal points: it allows us to distinguish between points that are closer to being optimal than others. This is, I think, unexceptionable. What of liberty? I think it would be an extraordinarily strong claim to argue that liberty does not come in degrees—that the only distinction of relevance is between arrangements that are truly free and all others. Such a claim would assert, for example, that attempts like Lord Acton’s to trace the history of liberty—its waxing and waning—are conceptually meaningless. The claim would entail that exercises involving comparative measures of Freedom of the kind undertaken by the Heritage Foundation and the Fraser Institute, or of the slightly different kind pursued by Amnesty International, are intrinsically impossible. It would mean that we could not make judgements to the effect that the US today enjoys more liberty than does China, even though the US is arguably not perfectly free. I do not say that it is a simple matter to decide in any case whether freedom is greater or lesser, or what the precise ‘freedom-makers’ that bear on the relevant measure are. But to admit complexity is not to concede conceptual impossibility. In any event, Sen himself implies just such a comparative element when he refers to his own condition—the basis of his categorization—as a minimal condition of liberty. Presumably there could also be a maximal, and a just middling, such condition—and perhaps many gradations within that range. Actually, whether Sen’s condition really is a minimal condition—whether, that is, Sen’s implicit metric of liberty captures reasonable intuitions—is an important issue, both for what I shall argue here and more generally. I shall need to say something about that precise issue later. Here, I just want to establish a matter on which I take it Sen and I are agreed— that a specifically minimal condition is conceptually meaningful. On this basis, I want to underline several points about the limits of the categorical approach. So specify the L and P ‘conditions’ as L* and P* respectively. And depict

12

I.e. a > b > c > d, where ‘>’ stands for ‘is preferred to’.

THE CASE AGAINST CATEGORIES

21

them in Diagram 1.1 as points along the L and P dimensions respectively. I am taking it that those conditions, if met, do indeed capture one aspect of the relevant metric in each case: so, if the condition is satisfied, then the level of liberty (preference satisfaction) lies at or above a threshold level. This allows a categorization of normative space into four zones: zone I where both L and P conditions are met (L  L* and P  P*); zone IV where neither are met (L < L* and P < P*); and the two zones, II and III, where one condition is met but not the other. The point Q is the point at which the L* and P* lines intersect. First, consider points labelled S, S0 and T, T 0 . The representation of liberalism and Paretianism as categories allows us to say that both T and T 0 are better than S and S0 , because the T-points satisfy both conditions and the S-points neither condition. But that representation does not allow us to say that S0 is better than S, or that T 0 is better than T: rather, the categorical formulation registers S and S0 (and T and T 0 ) as normatively equivalent. If liberalism and Paretianism do indeed come in degrees, this judgement of equivalence is clearly a major loss in normative sensitivity! However, put that aspect aside and consider instead the set of feasible social states.13 That set is derived as follows. Each feasible social state is given a score in the L and P dimensions, and so can be represented by a point in (L,P) space. Denote the set of such points Of 0 f. The line f 0 f is the outer envelope of this set, and it is on this envelope that normative analysis should focus, since this shows the highest feasible combinations of P and L. What does an ‘impossibility’ theorem amount to in this setting? In fact, the SCA theorems do more than prove impossibility as such (as I shall shortly indicate). But for a start just focus on ‘impossibility as such’ (IAS): what IAS amounts to is just the property that Of 0 f does not contain Q (and hence no point to the north-east of Q).14 Then the first thing to be noted is that IAS does not logically imply any tension between L and P at all. A simple example is sufficient to show this. So, on this basis, consider Diagram 1.2(a). Here, certainly, Q is infeasible—so impossibility is established. But, equally clearly, there is no conflict between P and L: the social state that secures maximally feasible P will also secure maximally feasible L. So, it doesn’t matter which of L or P we aim to maximize, because the other will be carried along more or less in its wake. Simply put, ‘impossibility’ in and of itself does not imply any necessity to choose between L and P at all.

13

I follow here the convention in the social choice literature that the objects of evaluation are social states. Individual preferences are defined over social states—and social states are the bearers of properties of liberty. 14 That is, the prior completeness of domain condition on preferences does not exclude the possibility that Q (and points superior to Q) will not be a member of the feasible set.

22

GEOFFREY BRENNAN

L∗

(a) P

L∗

(b) P

∗ Q P

Q

P∗

f′ f′ L

f

L∗

(c) P

L∗

(d) P

OP f′

L

f

Q

OP

P∗

Q

OL

f′

P∗

OL

f L

L

f

Diagram 1.2

Is there, then, anything that Sen might appeal to in order to rule out cases like Diagram 1.2(a)? There are, I think, three things. First, there is an issue as to whether Paretianism is by definition feasible. Second, there is the issue about the ‘minimalism’ of Sen’s liberal condition. We shall deal with both these possibilities shortly. But consider briefly a third. I have already mentioned that impossibility theorems within the SCA do more than demonstrate IAS. They do seek to establish that the ends (in this case Liberty and Paretianism) are in conflict; and to do that they need to show that at least some of the relevant conditions are satisfiable separately. And in this case, Sen does want to show that P* and L* are feasible when considered independently: it is only together that problems arise. My appeal to the Diagram 1.2(a) case is not then to make a point against Sen (or SCA), so much as to clarify what, in principle, impossibility as such might and might not tell us. To establish that P and L are in tension, it is necessary to show both that P* and L* are independently feasible, and that P* and L* are not jointly feasible.

Paretian feasibility Paretianism is typically defined by reference to feasibility. That is, each individual has preferences over each feasible social state which can be more or less extensively fulfilled. If the social states in question are indeed feasible, then there will be some (a number) that involve maximal preference satisfaction and these will be the Pareto optimal ones—ones such that we cannot increase preference satisfaction for any one

THE CASE AGAINST CATEGORIES

23

individual without reducing it for some other individual. If Paretianism is understood in such terms, then P* must always be feasible. This would mean that the relevant version of Diagram 1.2(a) would be Diagram 1.2(b)—but not much of substance hangs on this difference: ‘impossibility’ as such would still not imply conflict. In fact, however, there is some equivocation within economics as to what exactly feasibility amounts to in such settings, according to exactly which kinds of constraints are incorporated within the analysis. For example, it has been common to derive (Pareto) optimal conditions for public goods on the basis of aggregate resource constraints, with no formal consideration as to whether any institutional arrangement exists under which the optimal conditions in question could be realized. In a similar manner, the cooperate/cooperate outcome in a single shot prisoner’s dilemma is usually registered as a feasible outcome—even though motivational constraints assumed within the model actually rule it out.15 In other words, the relation between feasibility tout court and Pareto optimality is not always a matter of definition: details of formulation obtrude. Certainly, in plenty of familiar (and perfectly useful) economic models, Pareto optimality is not strictly feasible when all relevant constraints are included: in such settings, what is sought is the best available approximation to P*.

Feasibility of liberalism I have said that an important feature of the Sen liberalism condition, L*, is its claim to be ‘minimal’. If L* is indeed minimal, then the implication is that it will be relatively easy to ensure on its own. And then we can rule out cases like that illustrated in Diagram 1.2(a) (and (b)). It is clear enough what Sen has in mind. He thinks that the relation between liberalism and Paretianism is like that illustrated in Diagram 1.2(c)—where there is a trade-off required between the claims of L and those of P. But as we have shown, demonstrating the infeasibility of Q is insufficient to establish that claim. We require additional argument. Specify the point along f 0 f that is best for P. Call that OP. Now specify the point along f 0 f that is best for L. Denote that OL. The critical issue would appear to be the claim that OP and OL are a long distance apart in terms of the values of P and L they represent. But the repertoire of categories that the social choice framework provides does not readily admit any such remark: the term ‘a long way apart’ is not admissible within the categorical repertoire. So even if we can show that conditions L* and P* can be attained separately, but not simultaneously, we simply cannot distinguish between cases that are very like that in Diagram 1.2(c) and cases that are very like that in, say, Diagram 1.2(d). 15

This is a relevant consideration in the Sen context because the prisoner’s dilemma connects with one of the more plausible modifications of the Sen result.

24

GEOFFREY BRENNAN

It is, in other words, perfectly consistent with Sen’s paradox that L and P are highly complementary on average, in the sense that changes that promote the one are in general likely to promote the other. Even if it were true that there was some conflict between P and L, it could be entirely second-order. It is at this point in the argument that the full force of Sen’s claim that L* is a ‘minimal’ requirement becomes apparent. If L* were not minimal—if the condition were quite demanding—then failure to achieve P* and L* simultaneously would not tell us very much about the extent of any conflict between L and P. In other words, to assess whether Sen’s paradox exposes a deep conceptual tension or something more like a minor intellectual curiosity depends on an assessment of just how minimal the liberalism condition is. And this requires us to investigate the details of Sen’s formulation of L* in somewhat greater detail.

IV Sen: Rights vs Exchange? Sen formulates his liberalism condition as a requirement on the derivation of the aggregate social preference ranking from the underlying individual rankings. Liberalism is then not a matter of freedom of action but of respecting individual preferences in the determination of what is in aggregate desirable. Individuals’ preferences are respected in the following sense: each individual is assigned a pair of social states; the minimal liberal requirement is that individual’s preferences over those social states are ‘respected’ in the overall ranking in the sense that the aggregate ranking tracks the individual’s preferences over those social states. ‘Tracking’ here simply means that if A is assigned states X and Y, and ranks X more highly than Y, then the aggregate ranking will rank X more highly than Y. There is no requirement that X will be ranked above all other social states, so A is not ‘decisive’ over anything except that A’s X/Y preference ensures that A’s less preferred state is not the most highly ranked. Sen does not stipulate what a specifically liberal assignment of social states to individuals for this purpose would look like. He does offer some suggestive remarks in his example of Lady Chatterley’s Lover. But in several respects the example runs ahead of the formal treatment, and it is helpful here to focus specifically on what the analytics show, shorn of gratuitously imported intuitions. Sen’s theorem applies to any such assignment—what I shall call a ‘preference tracking specification’. His argument is that preferences could be such that, given a specific such preference tracking arrangement, the aggregate ranking could rank social state X more highly than Z despite the fact that X is Pareto-dominated. A simple example makes the point. Suppose the preference tracking specification assigns social states X and Y to individual A’s determination, and social states Z and W to B’s determination.16 Suppose without loss of generality that A prefers X to Y

16

Recall that ‘determination’ here means merely that A’s ranking over X and Y must be ‘respected’.

THE CASE AGAINST CATEGORIES

25

and B prefers Z to W. Every aggregate ranking consistent with minimal liberalism must then track the ranking; and it is easy to see that the aggregate ranking must be one of the following six possibilities: X>Z>Y>W; or X>Z>W>Y; or Z>X>W>Y; or Z>X>Y>W; or Z>W>X>Y; or X>Y>Z>W Now, it will be noted that all of those rankings has either X>W or Z>Y. But preferences can be such that both individuals prefer Y to Z and W to X. Again, a diagram may help to clarify. In Diagram 1.3, A’s preferences are shown along the horizontal axis, such that points further to the right indicate higher levels of A’s preference satisfaction; B’s preferences are shown along the vertical axis such that points that are higher indicate higher levels of B’s satisfaction. In Diagram 1.3, A prefers X to Y (X is further to the right); and B prefers Z to W (Z is higher). But clearly both prefer Y to Z and W to X. In short, preferences are such that the liberal preference tracking specification violates the Pareto requirement. B′s preference

Y

Z

W X

A′s preference

Diagram 1.3

The reason for this result is that the preference tracking specification itself does not reflect preference intensity (presumably it reflects liberal principle). As Diagram 1.3 suggests, the choice between X and Y is less important to A than the choice between W and Z: A would prefer to have been assigned determination over W and Z rather than over X and Y. And analogously, B would prefer to have been assigned determination over X and Y rather than over W and Z. But that is contrary to what liberalism requires of the preference tracking specification. Put another way, if, in the arena of action, A and B could have agreed to bring about, say, W rather than X, they

26

GEOFFREY BRENNAN

would have done so. But the Sen formulation has the effect of specifying that any such agreement would be a violation of the requirements of liberalism. In effect, the preference tracking arrangement assigns pairs of social states to individual preferences inviolably. If those preference-tracking arrangements are, in some sense, equivalent to liberal rights (or something akin to liberal rights) those rights are effectively inalienable. But if all such ‘rights’ are inalienable, then it is by no means clear that the liberal condition is minimal—certainly not minimal in all relevant respects. If the preference tracking specification effectively rules out the possibility of trade, then what are we to make of the liberal right to truck and barter on mutually agreed terms? Put a slightly different way, the economics tradition has routinely conceived a (liberal) rights structure as establishing the point of departure for exchange. The Sen formulation of the relation between liberty and preference satisfaction involves the idea that liberal principles are best articulated entirely independently of exchange. More particularly, Sen’s formulation suggests that any and all exchanges, from an initial position of rights that is consistent with the liberal preference tracking specification, would be a violation of liberal principles. But if that is so—if all liberal rights are effectively construed as inalienable—then a conflict between Paretianism and liberalism is hardly surprising. To the extent that exchange is feasible, the whole point of any arrangement that makes rights inalienable is to suppress possible voluntary exchanges in relation to them. So it more or less follows from any formulation in which rights are treated as inalienable that conflicts between the rights structure and maximal preference satisfaction can arise. The only issue, on this reading, is whether it is at all a plausible feature of liberalism to conceive of all liberal rights (even with the very weak ‘preference tracking’ notion of rights that Sen uses) as inalienable. It does not seem so—and in that sense at least, Sen’s minimal liberal condition hardly seems minimal, and possibly not especially liberal. Sen might respond to this reading by conceding that, in those cases where exchange is feasible, exchange is legitimate – but that this does not cover all possible cases.17 In particular, it does not cover cases in which exchange is infeasible (say, for transactions costs reasons). It is not then so much that the liberal preference tracking specification is made inalienable by force of law: it is rather inalienable by force of circumstance. So, imagine a case where exchange is not possible and a rights determination has to be made. In normatively assessing that determination, is it self-evident that the best rights allocation is that which one believes would have emerged from free exchange between the parties (supposing one knew what that was)? Or should one opt for the allocation that best instantiates liberal principles? To pose that question as a live one presupposes that liberal principle and preference 17

This is a significant retreat to be sure; but I cannot see how Sen can retain a more general interpretation.

THE CASE AGAINST CATEGORIES

27

satisfaction can come apart. But in the circumstances in which trade is infeasible there seems to be no reason for thinking otherwise. And I think the answer is not so clear. It is one thing in liberal terms to allow persons to make mutually agreeable arrangements, and another to have arrangements that one would have agreed to thrust upon one. Familiar arguments concerning the authority of purely hypothetical consent clearly bear here; but I do not wish to add to that discussion. What I do want to emphasize is that the conflict between liberal principle and preference satisfaction is a rather restricted one. It deals with two kinds of cases: that where there exists a right that liberal principle insists ought to be treated as inalienable (something that presumably will not be true of anything like all liberal rights); and that where transactions costs are sufficiently high that exchanges cannot occur. In this latter case, a liberal dilemma of a kind can arise: and in the limited scope of that dilemma, Sen’s impossibility theorem indicates the necessity of choice between rival principles.

V Sen vs Coase vs Optimal Compromise? The idea that Sen’s treatment bears mainly on cases where transactions costs are sufficiently high to rule out exchange, suggests a direct connection to a related, and also highly influential, treatment of such issues within economics. The rival treatment I have in mind is that laid out in Ronald Coase’s classic 1960 paper, ‘The Problem of Social Cost’. The connection is that Coase, too, is concerned with the issue of how to allocate ‘rights’ when exchange is costly (and potentially impossible). Coase is concerned with legal rights, rather than some normatively constructed rights conception—but the setting and the issues are in many ways uncannily like Sen’s. Here I want to briefly explore the similarities and differences between the two treatments, partly to draw a contrast between the social choice approach and what might be seen as a mainstream economics rival, and partly to suggest a kind of middle ground at the normative level that I think both treatments obscure. To focus the similarity, imagine a judge18 who is to make a rights determination in a specific case and who looks variously to Coase and Sen—two Nobel Laureates in Economics—for guidance. There are several striking differences between the two approaches. For one thing, Coase is not concerned with possible conflicts between liberalism and Paretianism: for Coase, Paretianism is the only relevant feature! For another, Coase specifies his notion of Paretianism in terms of a metric—namely social cost, which on Coase’s view it is the normatively proper role of the judge to minimize. In this respect, Coase seems little perturbed by issues of interpersonal incomparability: social costs are a simple aggregate measure, and the question of who happens to bear them plays no role in his treatment. The problem facing the Coasian 18

Perhaps one sufficiently influenced by the law-and-economics movement to think that the economics literature might throw light on such a case.

28

GEOFFREY BRENNAN

judge is essentially an informational and analytic one: to work out which among the competing allocations (and definitions) of rights would correspond most closely to that that would emerge from exchange between the affected parties—both in this case and any others sufficiently like it that those others would fall under its precedent. A Sen judge, by contrast, recognizes that there are two normative principles at stake—Paretianism (whether measured by social cost or some more strictly Paretian condition) and substantive liberalism. In lots of cases, these will be in conflict—and the issue for the pure Sen judge is which of these principles to give up on. Reading between the lines, it seems plausible to think that, for Sen, liberalism should trump Paretianism, and hence the Sen judge will typically decide in that way.19 The central argument of this chapter can be captured by observing that there is a third kind of judge, a quasi-Sen judge, who is like the Sen-judge in that she weighs substantive liberal principles in the allocation of rights, but is unlike the Sen-judge in that she does not think in terms of a simple dichotomy between rival normative principles. Instead, she has a single rule—but an appropriately complex one. It involves weighing the stakes in both dimensions. Just how ‘far’ would any given rights allocation be from the liberal ideal? What are the costs in terms of liberty forgone? If you give up a little bit of liberty by shading the rights in social cost minimizing ways, how much will social cost be reduced? In terms of the diagrammatics of Diagram 1.2(d), she will be searching along the frontier between OP and OL, attempting to find the optimal compromise between the conflicting elements of normative desirability. It may well be that she will rarely declare certain rights totally inalienable—though in certain extreme cases she might. In certain other cases, she may not rule out voluntary exchanges, but she might refuse to provide court support in enforcing the terms of such exchanges. In more general cases, considerations of substantive liberal principle will weigh alongside ‘social cost’/preference-related considerations. And in those cases, these rival goods will be weighed and some optimal compromise sought. I say that such a judge is a quasi-Sen judge in the sense that questions of what kinds of rights structures score well on liberal grounds will weigh with her. But she is not a pure Sen judge because she does not give up on social cost considerations in their entirety. In weighing both liberty and social cost considerations, it is unlikely to occur to her that she is doing something especially paradoxical—or something that commits her to holding logically inconsistent propositions. There is surely nothing ‘impossible’ here!

19

Here I am focusing on what I take to be the normative differences. Epistemic (and/or conceptual) challenges—like those surrounding interpersonal comparisons of utility or how the judge is to get access to information about agents’ preferences (including the intensity of those preferences) in order to make the social cost calculation—I am simply setting aside here. Those are, clearly, interesting and important issues. But they do not bear centrally on the point I seek to make.

THE CASE AGAINST CATEGORIES

29

She may not find Sen’s particular formulation of liberalism especially enlightening or useful. After all, Sen’s analytics tell us next to nothing about what a substantively ‘liberal rights structure’ would look like. But she will, at least, accept what I take to be Sen’s basic point—which is that, in at least some cases, liberalism does not fully collapse to preference satisfaction. More relevant for current purposes, however, is that she will depart from Sen in refusing to formulate ethical desirability in strictly categorical terms. She will depart in this respect precisely because she recognizes the need to weigh competing considerations—and to do that properly she will need some measure of the degree to which liberal principle and aggregate preference satisfaction (and doubtless other normative considerations) are secured. In her view, categorical thinking is confused and confusing. And I think she will be right to hold that view.

VI Conclusions In this chapter, I have argued that the social choice approach involves an example of ‘categorical’ as opposed to ‘comparative’ or relative thinking in the normative domain, along the lines of the distinction Broome draws in Ethics out of Economics. I have claimed that, in the social choice context, it is the categorical formulation of the basic norms that gives rise to impossibility theorems rather than anything intrinsic to the issues with which social choice deals. I have illustrated this claim primarily by reference to Sen’s impossibility of a Paretian Liberal, because both Paretianism and liberty naturally invite rendering in comparative terms. The focus of my objection to the categorical formulation here lies in the way in which it frames the question of how to deal with conflicts between different normative considerations. The structure of impossibility theorems—showing both that multiple desiderata cannot be achieved simultaneously, and also that subsets of the desiderata can be—encourages normative analysis to focus on the question as to which logically consistent subset of the desiderata ought to be retained; or equivalently, which desiderata ought to be given up (in their entirety). This is, I believe, in general a misleading way to deal with conflicts among different normative considerations. It contrasts with the idea of trading off considerations at the margin to achieve an optimal compromise. Perhaps the idea of trading off is mistaken in some cases: perhaps some values just are on/off things. But comparative ways of thinking go with a broad optimizing picture. The idea that you cannot have everything is common across optimizing and impossibility theorem approaches. But, in general, the optimizing approach gets the response to that constraint right, it seems to me, while the categorical approach does not. I should note finally that John Broome himself seems not to agree with me that social choice theory is guilty of categorical thinking. I know this from previous conversations with him on the question. I am not sure why he does not agree. It may be that he thinks a metric of ‘number of desiderata satisfied’ is sufficient for SAC

30

GEOFFREY BRENNAN

to qualify as comparative. I think that would be a mistake. The critical feature of the basic theorems of social choice theory is that all the normative work is done by the required features of the relation between individual rankings and the aggregate ranking; and these desired features are expressed as on/off requirements. In that sense, the social choice approach represents the epitome of categorical thinking! Alternatively, it may be that he is misled by the fact that the objects being aggregated—the individual rankings over states of affairs—are themselves comparative. This is true but irrelevant for the ethical aspects of the case. It mistakes the domain of the relevant theorems for the criteria of desirability that do all the normative work. In any event I believe him to be wrong on this aspect of the issue. And the object of this essay is to explain why, and to provide one reason why it matters.

References Brennan, G., and Baurmann, M. (2006) Majoritarian inconsistency, Arrow impossibility and the continuous interpretation: A context-based view. In C. Engel and L. Daston (eds.), Is There Value In Inconsistency? Baden-Baden, Germany: NOMOS, 93–118. Broome, J. (1991) Weighing Goods. Cambridge, MA: Basil-Blackwell. Broome, J. (1999) Ethics out of Economics. Cambridge: Cambridge University Press. Buchanan, J. (1954) Social choice, democracy and free markets. Journal of Political Economy 62: 114–23. Buchanan, J. (1977) Freedom in Constitutional Contract. College Station, TX: Texas A&M University Press. Coase, R. (1960) The problem of social cost. Journal of Law and Economics III: 1–69. Friedman, M., and Friedman, R. D. (1980) Free to Choose. New York: Harcourt-Brace. Mackie, Gerry (2003) Democracy Defended. Cambridge: Cambridge University Press. Mueller, D. (2003) Public Choice III. Cambridge: Cambridge University Press. Ordershook, Peter (1986) Game Theory and Political Theory. Cambridge: Cambridge University Press. Sen, A. (1970) The impossibility of a Paretian Liberal. Journal of Political Economy 78: 152–7. Taurek, J. (1977) Should the numbers count? Philosophy and Public Affairs 6: 293–316. Tullock, G. (1967) The general irrelevance of the general impossibility theorem. Quarterly Journal of Economics 81: 256–70.

2 Challenges to the Principle of Personal Good Douglas MacLean

I My goal in this essay is to examine what John Broome calls the principle of personal good (PPG).1,2 This principle, in conjunction with others, is supposed to explain how the good of individuals is combined to determine overall good. It says that two alternatives are equally good if they are equally good for each person, and one alternative is better than another if it is at least as good as the other for every person and definitely better for someone. In conjunction with the axioms of utility theory, PPG is about the structure of the good, and Broome, like many consequentialist moral philosophers and most economists, aims to be as non-committal as one can reasonably be about the content of the good. But, of course, structure and content cannot be entirely separated, and if one’s moral theory is also consequentialist and committed to the idea that a state of affairs with more good in it is inevitably morally better than a state of affairs with less good in it, then a principle about the structure of the good will have broad substantive moral implications. As I will try to bring out, Broome intends PPG, with one qualification, to reflect the idea that all good is people’s good, and also to reflect the idea that people’s good is individualistic rather than communal. According to this view, what is good for groups is reducible to the aggregate of what is good for the individual members of the groups. But I will argue that the qualification Broome makes to the idea that all good is people’s good is more significant than he seems to acknowledge, and that the idea that all good for people is individualistic is unsustainable.

1 For helpful discussion of the issues in this chapter and comments on earlier drafts, I am especially grateful to Geoffrey Brennan, John Broome, Iwao Hirose, Christian List, and Susan Wolf. 2 The principle is defined and discussed in Broome (1991), chs 8–9.

32

DOUGLAS MACLEAN

My own view about morality and moral reasons is not consequentialist. And though it is not my aim in this chapter to describe or defend this view, I should briefly explain part of what it involves, because it expresses some assumptions that are relevant to this discussion. The view begins with the attitude of valuing or caring or regarding as important. To value something is to take oneself to have reasons for acting or responding to it in appropriate ways and for adopting or developing certain positive attitudes. Valuing something, in the sense that I am using this term, does not imply that one believes the thing to be valuable. We can value or care about things for idiosyncratic reasons. Nobody else has much reason to care about my album of family photographs, but I do. To regard something as valuable is to believe that one’s reasons for valuing or caring are reasons that others ought to share. Moral reasons are reasons to act in ways and adopt attitudes that honor, express, and promote what is valuable. Human beings are the only creatures capable of guiding their actions and attitudes in the light of reasons. This fact, in conjunction with the fact that values are fundamentally linked to what we have reason to think, feel, and do, yields a fairly trivial sense in which all values are human values.3 Moreover, if what is valuable is determined by our reasons for caring or regarding as important, these reasons may be grounded in specific facts about our nature or our lives. The importance of this last point will become clearer in remarks I make below. Broome says something that sounds superficially similar to the claim that all values are human values, but it is really quite different. He says that all good is people’s good. This remark occurs in his discussion of one of the qualifications he makes to PPG. He considers an example of a couple that might have a child, but if they do, the child will have a hereditary disease that will make her life painful and unhappy.4 Even if we assume that the existence of this child would make nobody else worse off, it would be better—that is, the general good would be increased—if they did not have this child. But it would not be better for that child not to exist, and it would not be better for anyone else, so this seems to be a counterexample to PPG. To avoid the counterexample, Broome stipulates that PPG applies only when the same people exist in both of the alternatives that are being compared. He claims that this limitation of the principle does not affect its plausibility, however, because, The principle is a particular instance of a general idea that applies universally: that all good is people’s good. The goodness of an alternative depends only on the good of the people who exist in that alternative. To put it another way: good supervenes on people’s good . . . The example [of the child] does not suggest there is any good other than people’s good. When it is applied to comparisons between alternatives that contain the same people, the general idea comes down to the principle of personal good.5 3 If we were to learn that some species of non-humans have and use reasons in the relevant way, then we may count them as persons, and we should say that all values are persons’ values. 4 5 Broome (1991: 166). Broome (1991: 167).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

33

Human values as I understand them can include the belief that some things may be valuable in ways that have no direct connection to the good of people. But the point of PPG is to make explicit that the general good supervenes on people’s good. This raises two questions: What does Broome mean by “people”? And how does he interpret supervenience? I will state my reactions to and criticisms of Broome’s view in response to these questions. I will begin with some comments on supervenience that emphasize the difference between Broome’s view about the good and my view about what is valuable. Then I will examine two further qualifications Broome makes to the principle, which suggest different ways of challenging it. The first qualification is how we understand the scope of “people”. Actually, that is misleading. The issue is about what is included in the scope of individuals that have a good that combines to determine the general good. The second challenge is about how the good of individuals combines to determine the general good. More specifically: Is the general good reducible to an aggregation of individuals’ good, or are there communal or social goods of a kind that do not imply changes in any individual’s good? In Section X of this chapter, I comment briefly on Broome’s view of the value of human life itself. This comment is not strictly relevant to his defense of PPG, but it reveals something about the difference between Broome’s philosophical temperament and my own that may help explain my general reservations about a carefully qualified principle that Broome defends.

II Supervenience is a useful concept for expressing the relationship between facts at different levels of description.6 Someone who is a naturalist but believes in the objectivity of moral values would claim that moral facts supervene on natural, non-moral facts. This means that there cannot exist two distinct possible worlds that are identical in all non-moral respects but differ in some moral respect. Conversely, if two situations are in any way morally different, then they must differ in some non-moral way. Similarly, a physicalist theory of mind would claim that psychological facts about someone’s mental state supervene on the physical facts about the person’s brain and bodily states. There cannot be two people with physically identical brain and bodily states, but different mental states. I believe that values supervene on natural facts about human beings and the world. In other words, I do not believe that the world contains normative, non-natural properties, which could be present in some state of affairs but absent in some other state of affairs that is identical in all its natural properties and relations.

6

My remarks here and in Section IX below borrow heavily and directly from List and MacLean (manuscript).

34

DOUGLAS MACLEAN

But are values reducible to natural facts about human beings and the world? Supervenience and reducibility are different concepts. Supervenience is a metaphysical notion, while reducibility is an explanatory notion. The literature on non-reductive physicalism in the philosophy of mind provides an example of supervenience without reducibility. A physicalist rejects metaphysical dualism; she is committed to the view that human beings are natural, biological, physical entities. But she may also deny that mental states are reducible to, or identical with, physical states. Thus, her theory of mind, insofar as it is an explanatory theory, relies on non-reducible mental concepts. Similarly, a central point in some recent theories of group agency claims that, while all facts about a group agent supervene on facts about its individual members, we cannot always give illuminating explanations of a group agent’s attitudes and actions solely by reference to its members’ attitudes and actions.7 My view about what is valuable also denies reducibility. There are some reasons for valuing an object or a state of affairs that we all ought to accept, and facts about the objects or states, together with facts about human nature, explain why. Where the facts are identical, the same reasons supervene on them. Broome claims that facts about general good supervene on facts about individual good. But his view about the way that the general good supervenes on individual good, as I understand it, also commits him to a claim about reducibility. It isn’t a claim about the reducibility of the good to non-moral concepts, but about the reducibility of the general good to the good of individual people. Let me state Broome’s claim more precisely. States and prospects can both be better or worse for people. It is better for someone to be healthy than to be ill, and it is better for an individual to have a flu shot if it makes it more likely that she will be healthy, and less likely that she will be ill. Utility functions assign numbers to states and to prospects that represent the betterness relation among them. In the 1950s John Harsanyi proved that when certain seemingly weak conditions of rational choice are met, social utility for outcomes and for prospects could be represented as the sum of individual utilities.8 The weak conditions are equivalent to the axioms of expected utility theory, which is a theory of individual rational choice. Broome accepts the axioms, and he argues that when the PPG is added to them, then Harsanyi’s proof implies that general good is the total or aggregation of personal good. The general good thus supervenes on, but is also reducible to, the sum of individual goods. I will say that social facts supervene atomistically on individual facts if the social facts are determined by an aggregation of the individual facts. PPG commits Broome to atomistic supervenience. I will argue below that some social facts are determined jointly by the distribution of individual facts, and thus supervene holistically on

7

See List and Pettit (2011).

8

Harsanyi (1955).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

35

individual facts. My argument against atomistic supervenience is not new. Peter Diamond made this argument against Harsanyi in the 1960s.9 I will examine Broome’s response to Diamond in Section VIII, and I will offer in Section IX some additional examples of holistic supervenience involving the distribution of risk.

III We have already mentioned one of Broome’s qualifications to PPG, which involves the example of bringing an unhappy child into existence. This is an instance of a now familiar puzzle in moral philosophy known as the “non-identity” problem. It arises with actions or policies that determine the identities of people who will exist in the future as well as the circumstances of their existence. The options we face today about how to respond to climate change have this feature.10 The consequences of adopting (or failing to adopt) a climate change policy today are so profound that they will surely determine which individuals will live in the future as well as the climate and environment they will inherit. If this is true, then we cannot benefit or harm future individuals—we cannot make any future individuals better or worse off than they would have been—by either limiting our carbon emissions or doing nothing about them and continuing on with “business as usual.” Depending on what we choose to do, some future people will have better lives in more favorable conditions, or other future people will have worse lives in less favorable conditions. An explanation of our moral responsibility in “non-identity” situations cannot appeal to how we might benefit or harm people. Non-identity situations thus constitute a counter-example to PPG, although, as Broome insists, they do not cast doubt on the general idea behind PPG, which is that all good is people’s good. Broome does not offer a solution to the non-identity problem in Weighing Goods. He does not defend a principle that is compatible with PPG and would explain our moral responsibilities in non-identity situations. I believe we have compelling moral reasons to leave future generations with a better world rather than a worse one, and Broome would surely agree.11 But I will not further pursue the question here of how we explain those responsibilities. Instead, I turn now to Broome’s other qualifications of PPG. One of them addresses the wellbeing of non-human animals. He writes, “One prospect may be better than another without being better for a person, because it is better for an animal. The principle ought really to take people and animals together. For simplicity, though, I shall continue to speak of people only. The inclusion of animals is to be understood.”12 The scope of “people” in the phrase “all good is people’s good” should be

9 11

10 Diamond (1967). Broome explores this issue in detail in Broome (2012: 59–68). 12 See the discussion in Broome (2012: 59–68). Broome (1991: 165).

36

DOUGLAS MACLEAN

modified or extended to include at least all sentient creatures. PPG is thus innocent of the sin that Peter Singer calls “speciesism.” It respects Singer’s principle of equal consideration of interests, which requires us to consider the interests of animals.13 This qualification does not affect Broome’s argument about the structure of betterness or goodness, or his claims about how general goodness supervenes on the good of individuals. But it does open his view to the doubts one can raise about Singer’s principle of equality. Given the vast number of sentient creatures that exist on earth, and the ocean of suffering in which most of them live, there can be little doubt that it would be better, in terms of reducing pain and suffering, not only to eliminate factory farms but also to devote significant resources to eliminating nature and wilderness itself, where so much of the suffering on earth occurs. Including the good of animals within the scope of individuals whose good combines to determine the general good is thus not an insignificant caveat to PPG.14 After addressing the good of animals, Broome offers a further qualification to PPG. He writes, There may be other sorts of good that are independent of the good of people. Perhaps, for instance, the survival of any living species is good in itself, quite apart from any good it may do for people. I do not wish to deny the existence of goods like this . . . Of course, if they are genuine goods, people ought to promote them, even though they are not the good of people. So they compete for resources with the people’s good. Eventually, therefore, they will need to be combined with people’s good to arrive at an overall assessment of good . . . The principle of personal good ought to be prefaced with: “Setting aside goods that are entirely independent of the good of people . . .”15

Does this mean that we should further broaden the scope of “people” in the original formulation of PPG to include not only all sentient creatures but all things or individuals that may have a good or their own? An animal species, after all, is not a sentient creature, and I assume that Broome means to include the survival of various plant species as well. I will call these goods “non-personal,” in order to distinguish them from the good of people, or what I will call “personal goods.” I will leave it to the reader to decide whether to include the good of non-human animals in the class of personal or non-personal goods. The fully qualified principle that Broome defends might be stated like this: Fully Qualified PPG: Setting aside non-personal goods, when we are comparing two alternatives that contain the same people (or sentient creatures), the alternatives are equally good if they are equally good for each person (or sentient creature), and one alternative is better than another if it is at least as good as the

13

See Singer (1979: chs. 2–3). I have developed the argument for thinking differently about the value of non-human animals and defending speciesism in MacLean (2010). 15 Broome (1991: 165–6). 14

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

37

other for all people (or sentient creatures) and better for some individual. When we are comparing alternatives that contain different people (or sentient creatures), the goodness of the alternative depends only on the goodness of the individuals that exist in that alternative. Of course the fully qualified principle buys precision at the cost of elegance, so we will use PPG, as Broome does, to refer to the principle, and we will take the qualifications for granted. The final issue that concerns Broome in his discussion of qualifications is the possibility that some things that are good for people are non-reducibly social or communal. These are possible goods that Broome cannot allow, because they challenge his argument about the structure of the good. In particular, they challenge the claim that general good is reducible to the good of individuals. Following Broome, we will call possibly non-reducible social goods communal goods. I will discuss them below, in Sections VI–IX.

IV How should we think about non-personal goods? Broome’s example in Weighing Goods, as we have noted, is the environmental good of saving an endangered species, but the non-instrumental reasons we have for preserving an endangered species are similar to the non-instrumental reasons we have for preserving and protecting other non-sentient objects, such as cultural artifacts, works of art, beautiful natural objects, and objects of historical significance. Living in a world in which we value and protect these things, appreciating and responding appropriately to them, obviously contributes to the meaning, richness, and value of human life, but not in a way that is necessarily dispersed to individuals. How can we explain the value of such objects? It is implausible to think that the value of non-personal goods supervenes atomistically on the good of individuals. The value of a work of art is not dependent on how much satisfaction people get from seeing or hearing it; nor is it a function of how many people appreciate or like it. A work of art that is complicated and relatively inaccessible can be greater and more valuable than one that is much more popular. And the value of preserving the original Declaration of Independence is not made greater as the population of appreciative U.S. citizens grows. It is not entirely implausible to regard the value of a document like the Declaration of Independence as being social or communal in a non-reducible way, because of its cultural significance, but Broome cannot explain the value of culturally significant objects in this way. And it is less implausible to regard other non-personal goods, such as great works of art, beautiful natural objects, or relics from an ancient civilization, as irreducibly social or communal goods.

38

DOUGLAS MACLEAN

The alternative on Broome’s account is to regard such non-personal goods as independent of the good of people and thus not covered by the theory of the general good. But that response has two problems. First, it leaves out of account not only environmental goods like the value of saving an endangered species, but many different and important values and goods. Second, as Broome acknowledges, if there are such goods, responding appropriately to them demands resources, and these demands will have to be combined with the demands of other values or goods to determine what, all things considered, we ought to do. If Broome excludes these values from the theory of the general good, then, for the purpose of guiding our actions and attitudes, the implications of PPG for determining what is better overall or most good is more limited than most of us would have expected.

V Let us look more closely at environmental values. These come up often as examples in Broome’s writing, and they reveal that he has broad environmental sympathies. So we should ask, if saving an endangered species is a non-personal good, how should we think about its value? One possibility is to deny that environmental values are human values in the sense I described above. This view, which is shared by some environmental philosophers, implies a radically non-anthropocentric moral theory. Peter Singer rejects anthropocentrism when he argues that equal consideration of interests requires us to take animals into account. But some environmental philosophers, for example Kenneth Goodpaster, argue in turn that sentience itself is an arbitrary criterion for determining moral standing.16 Goodpaster claims that any living thing can be benefitted or harmed by our actions and therefore has genuine moral interests—a good of its own—that we should take into account. Singer says that we cannot harm a stone by kicking it down the road because a stone, lacking sentience, has no morally relevant interests. But we can harm a plant by neglecting to water it; we can cause it to die. Goodpaster also suggests that holistic entities like a species or an ecosystem can be interpreted as living things in this sense. Having a life satisfies what Goodpaster calls a regulative principle for determining what entities have a good of their own that qualifies them for moral consideration. But he also acknowledges that our reasons for determining what we ought to do must be guided by separate, operative principles, which tell us how to set priorities and establish tradeoffs. In my view, the operative principles are determined by our reasons for valuing, and they come first. These reasons are necessary to explain regulative moral principles and to establish what objects have a good of their own in a morally relevant sense. This is because the interests of any entity are morally relevant only if they make claims on us as moral agents. As Bernard Williams points out, “If a

16

Goodpaster (1978).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

39

tree has any interests at all, then it must have an interest in getting better if it is sick; but a sick tree, just as such, makes no claim on us.”17 Some of Broome’s remarks suggest that his view is similar to Goodpaster’s. For example, in a recent paper defending PPG, he writes that one reason to doubt the principle is that there are goods other than the good of people, and these must contribute to the general good. The good of nonhuman animals is a clear example. It is also easy to take account of in [Harsanyi’s] theorem. We have only to include nonhuman animals along with people, and extend the principle of personal good to make it a principle of personal and animal good. The theorem remains valid with this amendment. Other sorts of good can be accommodated in the same way: the good of ecological systems, for example, if it really exists.18

Animals have goods of their own, and their good must be taken into account in determining the general good. Similarly, if an ecosystem has a good of its own, then it must also be considered an individual whose good is to be taken into account in determining the general good. Trees and plants have goods of their own. Surely an ecosystem can also be healthy or unhealthy, stable or threatened, in ways that ecologists explain. Does the meaning of “people” in PPG have to expand to include not just animals, but also plants and ecosystems? According to this way of thinking, our reasons for caring for an ecosystem are different from our reasons for caring for a work of art or preserving a historically important building. Paintings and buildings are not alive. Do they therefore not have a good of their own? Is their value strictly instrumental? I find questions like this puzzling but also unhelpful for understanding our moral reasons and commitments. Environmental philosophers like Goodpaster might be influenced by G. E. Moore’s analysis of goodness as a non-natural normative property that we intuitively know belongs to some individuals as well as to some organic wholes.19 This kind of moral intuitionism does not need to claim metaphysical supervenience. But, more likely, environmental philosophers who reject anthropocentrism find their inspiration in writers like Aldo Leopold. In a remarkable essay published after his death in 1948, Leopold claimed that if our moral attitudes would evolve and take account of the discoveries of ecology, we would embrace a “land ethic.” He wrote, “There is as yet no ethic dealing with man’s relation to land and to the animals and plants which grow upon it. Land, like Odysseus’ slave-girls, is still property. The land-relation is still strictly economic, entailing privileges but not obligations.”20 Just as our moral development no longer allows us to deny that “slave-girls” have morally relevant interests that demand our respect, we need now to change our attitude toward land. Leopold describes the

17 20

18 19 Williams (1995: 236). Broome (2015). See Moore (1903). Leopold (1949: 203). Goodpaster’s article begins with an epigram from this essay.

40

DOUGLAS MACLEAN

change he thinks is both necessary and warranted: “It is inconceivable to me that an ethical relation to land can exist without love, respect, and admiration for land, and a high respect for its value. By value, I of course mean something far broader than economic value; I mean value in the philosophical sense.”21 Leopold was not a trained philosopher, but a forester and essayist who became an influential advocate for environmental values. His aim in calling for a land ethic was not theoretical but practical and polemical. His writing nevertheless has had a powerful influence on subsequent generations of environmental philosophers like Goodpaster, who embrace and defend a land ethic as part of their attacks on anthropocentric ethical theories. Another prominent follower of Leopold interprets him as follows: “By ‘value in the philosophical sense,’ Leopold can only mean what philosophers more technically call ‘intrinsic value’ or ‘inherent worth.’ Something that has intrinsic value or inherent worth is valuable in and of itself, not because of what it can do for us.”22 This way of thinking, combined with an environmentalist’s sensibility, can lead to the rejection of anthropocentrism and the insistence that animals, trees, species, and ecosystems all have intrinsic, morally relevant good. Leopold, by the way, also includes rocks, soil, and rivers in a land ethic. But this way of thinking seems false to the spirit of Leopold and, I think, also to Broome’s environmental sensibility. Leopold was specifically concerned with rejecting what he saw as a dangerously narrow economic conception of value. He wanted us to recognize that we have reasons for valuing things that have no commercial value and to value them in ways other than treating them as commodities. The environmental economists and policymakers of his day seemed to regard land merely as raw material and saw its value merely as a means for increasing human welfare. Leopold was trying to open our eyes to a broader range of values that includes the land. Thus, he also wrote, To the laborer in the sweat of his labor, the raw stuff on the anvil is an adversary to be conquered. So was wilderness an adversary to the pioneer. But to the laborer in repose, able for the moment to cast a philosophical eye on the world, that same raw stuff is something to be loved and cherished, because it gives definition and meaning to his life.23

Broome shares Leopold’s rejection of the narrow way some economists and policy analysts think about values. In cost–benefit analysis, for example, environmental economists regard goods or commodities strictly in terms of their effects on human wellbeing. Their tool box now includes concepts like “existence value,” which aims to measure individual preferences for things like knowing that a pollution free body of water or a freely running river still exists. This kind of good is independent of the good of recreational benefits like swimming, boating, and fishing in beautiful areas. But these analysts still interpret values as individual preferences that we can order and quantify by measuring people’s willingness to pay to satisfy them. Broome 21

Leopold (1949: 223).

22

Callicott (1989: 97–8).

23

Leopold (1949: 189).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

41

explicitly criticizes this kind of reductionist approach to fixing the value of nonmarketed environmental goods. He writes, For some goods, [determining individuals’ willingness to pay for them] seems a hopeless project. Some values are so different from the value of commodities that it seems obvious they cannot be measured on the same scale. They are “incommensurable” with marketed commodities, to use a philosopher’s term. For example, global warming will do great damage to the beauty of the Earth, thereby impoverishing people’s lives. This is a great harm it will inflict on future generations.24

Leaving aside the problem of whether it makes sense to call this loss a harm, because of the non-identity problem, Broome’s point is that the beauty of the Earth is not a commodity; its value cannot be captured by the preferences we reveal in our consumer behavior or by cost–benefit analysis. He concludes, in words that echo the spirit of Leopold, that, “It is now widely accepted that economists’ material measures of value miss a great deal of what makes life good.”25 At times, I think, Broome’s environmental sensibility conflicts with his theory. Consider one more illustration that Broome uses to distinguish himself from some economists: People may be prepared to contribute to saving a lake even if they do not benefit from it at all, in any way. I was faced with a question much like this when Greenpeace asked me to contribute to saving the Atlantic from oil exploration. When I decided my willingness to contribute, I thought a bit about the true value of the clean Atlantic, and also about what I ought to contribute given my financial and other circumstances. Greenpeace helped me with the first consideration by telling me about the purity of the area and the whales that live there. It never tried to persuade me that the existence of the unpolluted Atlantic is good for me, and it never occurred to me to think in those terms when I determined my willingness to contribute. Indeed, it simply is not good for me, so far as I can tell, and even if it is, my willingness to contribute has nothing to do with the benefit to me.26

As Broome deliberates about what he ought to contribute to Greenpeace, he does not think about how this act enhances his own good. Does it make any more sense to think that he tries to work through all the incommensurable values to try to figure out how much this act contributes to the general good? Perhaps. But it seems more plausible that the literature from Greenpeace convinced him that this is a worthy cause, and then, as he thinks about his financial and other circumstances, he decides how much this cause matters to him, and he writes a check. Then perhaps he moves to the next solicitation, the one from Oxfam, and continues the same sort of deliberation. And this process, it seems to me, is not just a default mode of reasoning because the calculations about how much a clean Atlantic contributes to the general 24

25 Broome (2012: 136). Broome (2012: 136–7). “Why economics needs ethical theory,” a talk to the British Association for the Advancement of Science, London, 2000, available at . 26

42

DOUGLAS MACLEAN

good are impossibly complex and difficult. Rather, this way of balancing what is valuable and how much it matters to us personally as we make our charitable donations is an exactly appropriate response to our reasons for valuing and acting.

VI The challenge to PPG that worries Broome is the possibility of non-reducible communal goods, and in particular the good of equality. His discussion begins by dismissing a standard utilitarian defense of equality, which appeals to an assumption of the diminishing marginal utility of additional resources. If we assume diminishing marginal utility, which says that a given quantity of resources will produce greater wellbeing in those who are worse off than in those who are better off, then social good will tend to be increased by a redistribution of resources from individuals who are better off to those who are worse off. Utilitarianism, which is concerned only with maximizing the good and is indifferent to how good is distributed, would then advocate policies that push us toward more egalitarian outcomes. But even if we assume diminishing marginal utility, Broome explains, utilitarianism promotes equality only in special, highly implausible circumstances. The utilitarian argument requires not only the assumption of diminishing marginal utility, but also that each individual’s utility function is the same. Otherwise, as Edgeworth pointed out, utility is maximized when each individual’s marginal benefit is equaled. This might be accomplished through redistribution from a worse off individual, who happens to be poorly endowed to convert additional resources into a higher level of wellbeing, to a better off individual, perhaps a “utility monster” that derives outsized pleasure from his outsized house and jet. Edgeworth claimed that equalizing marginal benefits in order to maximize utility does not promote social equality; rather, it is a defense of “aristocratical privilege—the privilege of man above brute, of civilized above savage, of birth, of talent and of the male sex.”27 Broome concludes that this utilitarian argument “is not really an egalitarian argument at all.”28 Many philosophers embrace egalitarianism as a moral principle and regard it as an irreducibly communal good. T. M. Scanlon accepts this view. He writes, “Fairness and equality do not represent ways in which individuals may be made better off. They are, rather, special morally desirable features of states of affairs and institutions.”29 This view is incompatible with PPG, and Broome rejects it. In arguing against communal values, Broome first considers situations in which there is no uncertainty and it is possible to redistribute an increment of wellbeing from someone who is better off to someone who is worse off. By definition, this redistribution will reduce inequality overall, so an egalitarian might claim that this redistribution makes the overall result better (ceteris paribus, of course). Broome 27 29

Edgeworth (1881: 78–9). Quoted by Broome (1991: 177). Scanlon (1978: 99–100). Quoted by Broome (1991: 178).

28

Broome (1991: 177).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

43

correctly points out that even though the sum of good may not be increased by this redistribution, someone—the worse off person—is made better off by the transfer. So claiming that the transfer increases the general good is compatible with PPG. Less convincingly, Broome claims that if someone believes that equality is so important that overall good could be increased even by taking some money from the richest people and burning it, then this view is also compatible with PPG. This “leveling down” may not improve anyone’s welfare, but if it makes society better overall because it makes things more equal, then this change improves the good of the worse off. This strikes me as an implausible claim. Broome does not advocate leveling down, but isn’t it more plausible to claim that if some leveling down made society better, this is not because it improves the good of any individual but, rather, because it makes the distribution of good in the community more equal? Distributive equality, on this view, can compete with the aggregation of individual good to determine best overall. If distributive equality per se is a good at all, then it can offset some overall diminution in individual wellbeing. Distributive equality seen in this way is a good of the state of affairs as Scanlon suggests—a communal good—not a good that belongs to the worse-off individuals in that state.

VII Deeper issues are at stake when we try to reconcile equality with PPG in situations of uncertainty, because the communal understanding of equality favored by philosophers like Scanlon poses a challenge to assumptions of the theory of rational choice. My discussion of this challenge has three parts. In this section, I begin at some distance from equality or PPG with a brief description of the Allais paradox. The reason for starting here is that this paradox makes clear what is at stake in accepting or rejecting the “sure-thing” principle, which turns out to be crucial to defending PPG. In Section VIII, I will discuss the Peter Diamond’s equality-based objection to the sure-thing principle and Broome’s response to that objection. Then, in Section IX, I will introduce a different example that shows some other ways that equal risk distributions can express non-reducible communal values. The Allais paradox compares two gambles. In the first gamble, you may choose A, which gives you $1 million for sure, or B, which has an 0.89 chance of returning $1 million, a 0.10 chance of returning $5 million, and a 0.01 chance of returning nothing. Most people in this situation prefer A to B, presumably because they think it more reasonable to choose $1 million for sure than to take a small risk of gaining nothing in exchange for a chance of winning an even bigger prize. The second gamble involves a choice between C, which has a 0.11 chance to win $1 million and a 0.89 chance to win nothing, and D, which has a 0.10 chance to win $5 million and a 0.90 chance to win nothing. Most people prefer D to C, because the

44

DOUGLAS MACLEAN

Table 2.1.

Prospect

Ball number

A B C D

1–10

11

12–100

$1 million $5 million $1 million $5 million

$1 million $0 $1 million $0

$1 million $1 million $0 $0

chance of winning in either case is small and nearly equal, so it seems more reasonable to choose the gamble with the greater expected gain. The paradox is that, given the probabilities, it is impossible to assign utilities to the outcomes that make the preferences for A over B and D over C consistent. These preferences together violate the axioms of utility theory and are thus irrational according to the theory. To see why, imagine that the outcome will be determined by picking a ball at random from an urn that contains 100 numbered balls. The payoffs are represented in TABLE 2.1. An assumption of the theory of rational choice—the strong independence axiom or the “sure-thing” principle—says that outcomes that will occur regardless of the choice you make (i.e. “sure-things”) should not affect your preferences. The outcome in either gamble that results if any of the balls numbered 12–100 are chosen is not affected by a preference for A or B, or by a preference for C or D, so preferences in both gambles should be determined only by what happens if the ball chosen has a number between 1–11. It follows that the relevant prospects are identical in the two gambles: A equals C, and B equals D. That is why it is impossible to assign utilities to the outcomes that make a preference for A over B consistent with a preference for D over C. One response to Allais’s paradox tries to reconcile the popular preferences with the sure-thing principle.30 This argument claims that the outcomes have not been correctly individuated. Someone who prefers A in the first gamble knows that, if she chooses B and ball #11 is selected, she will deeply regret her choice. Her outcome will be not just that she wins nothing; it will also include decision regret, because she could have won $1 million for sure, but foolishly chose to gamble for the chance to get more. She chooses A in part to avoid the possibility of regret. The prospect in the second gamble is different. The chances of winning in either alternative are small and nearly identical, so someone who prefers to maximize her expected outcome by choosing D is unlikely to regret her choice if she loses. Thus, the outcome of choosing B and losing is different from the outcome of choosing D and losing, as shown in 30

See Jeffrey (1987: 223–36) and Broome (1991: 107–15).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

45

Table 2.2.

Prospect

Ball number

A B C D

1–10

11

12–100

$1 million $5 million $1 million $5 million

$1 million $0 plus decision regret $1 million $0

$1 million $1 million $0 $0

TABLE 2.2. Different values can be assigned to different outcomes, so the pair of preferences A and D are no longer inconsistent. A critic of this response might worry that it solves the paradox by making the theory of rational choice empty or trivial. If there are no restrictions on how we individuate outcomes, then we can distinguish outcomes simply in order to make any preferences consistent with the axioms of the theory, so no set of preferences would be irrational. Broome addresses this worry, but we will not consider his response yet, because a similar issue arises in his argument about equality. My point in describing the Allais paradox is just to illustrate an implication of the sure-thing principle and to suggest that it is not immune to objections or doubts, even in the context of individual rational choice.

VIII A similar challenge to the sure-thing principle is relevant to whether we should understand equality as a communal good or a property of states of affairs. Peter Diamond first made this argument with an example intended to be a criticism of Harsanyi’s theorem about social utility. Imagine that two individuals, P and Q, each need a kidney transplant to survive, and only one kidney is available.31 A physician flips a coin and then determines who will get the kidney, but in one scenario, he decides to give the kidney to P regardless of the result of the coin-flip, and in the other scenario he lets the coin determine who will get the kidney. He gives it to P if the coin lands heads and Q if it lands tails. The two scenarios are represented in TABLE 2.3. Diamond argues that Harsanyi’s theory implies that these two scenarios are equally good. The sure-thing principle tells us that there is no difference in the outcomes of either scenario if the coin lands heads, so the choice should be determined only by what happens in the state in which the coin lands tails. If we assign an equal value to saving either individual, then it follows from Harsanyi’s theory that the outcome of tails in either scenario is equally good. Thus: 0.5U(P lives, Q dies) + 0.5U(P dies, Q lives) = U(P lives, Q dies) 31

Broome (1991: 110–15).

46

DOUGLAS MACLEAN

Table 2.3. States of nature

P People

Q

Heads

Tails

Lives

Lives

States of nature

People Dies

Dies

First scenario

Heads

Tails

P

Lives

Dies

Q

Dies

Lives

Second scenario

Diamond claims that this conclusion violates a requirement of fairness. The procedure for making the decision should give each individual an equal chance at survival. The second scenario is therefore better than the first, because it is fairer, and it is fairer because the procedure for determining the outcome gives each individual an equal chance at survival. The value of equality is expressed by the fair procedure, not the outcomes. Diamond concludes that we should reject the sure-thing principle when it is applied to social choice. He writes, “I am willing to accept the sure-thing principle for individual choice but not for social choice, since it seems reasonable for the individual to be concerned solely with final states while society is also interested in the process of choice.”32 The fairness of the procedure in the second scenario is determined not only by the outcome when the coin lands tails, but also by what would have happened to each individual in the other state, if the coin had landed heads. Although equality is realized here by the fairness of a procedure that gives each individual an equal chance at survival, Broome rejects the idea that this is a communal value and argues instead that it is a good that belongs to or is owned by the individuals in this situation. This leads him to respond to Diamond as some defenders of the sure-thing principle respond to Allais. Broome agrees that Diamond’s second scenario is better than the first, but he claims that the outcomes in the first scenario have not been correctly described. The first scenario is unfair, and the unfairness belongs to Q, so the correct description of the outcomes should reflect this fact, as shown in TABLE 2.4. Here too the cost of defending the sure-thing principle in this way is the risk of making the theory of social choice empty or vacuous by immunizing it against any possible counter-example. Broome’s response to this worry is to insist that only “justifiers” can be used to individuate outcomes differently, and fair treatment is an obvious justifier. Granted, in some cases inequality is bad because it makes individuals worse off. For example, some empirical evidence suggests that inequality in the distribution of incomes adversely affects the health of people at all levels of income.33 If this is true, then the badness of income inequality is at least partly due to the consequences of inequality for the individual members of the community. But the value of a fair

32

Diamond (1967: 766).

33

See Daniels (2008: 83–5).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

47

Table 2.4. States of nature

People

Heads

Tails

P

Lives

Lives

Q

Dies and is treated unfairly

Dies and is treated unfairly

States of nature

People

First scenario

Heads

Tails

P

Lives

Dies

Q

Dies

Lives

Second scenario

procedure that expresses equality in how people are treated is essentially and irreducibly communal. The unfairness in Diamond’s first scenario cannot be seen without reference to all the possible outcomes. Q is not the only individual who is treated unfairly; P also receives unfair treatment, although it is to his benefit. But it is implausible to think that the unfairness belongs to either of the individuals in isolation or to the aggregation of unfairness to both of them. Rather, the unfairness is in the procedure that determines the outcomes for both of them. The value supervenes holistically on the comparison of the different ways P and Q are treated when the possible outcomes of state Heads and state Tails are seen together. The outcome of Heads is identical in both scenarios, but seeing it together with the outcome Tails shows us that the situation in the first scenario is rigged; the coin flip is a sham; the procedure is unfair. Broome defends his individualistic interpretation of the badness of inequality by arguing that we should regard equality and inequality as modal properties of individuals. He writes: Many properties are like this. Many dispositions are, for instance. A wooden ship is inflammable, even if it never burns but instead decays at the bottom of the sea. What makes it inflammable is that if it had been exposed to fire it would have burnt. Still, inflammability is a genuine property of the ship. Furthermore, the property of inflammability supervenes on the nonmodal properties of the ship. The ship could only have failed to have this property if it had been different in some structural or chemical way from how it actually is.34

On Broome’s view, unfairness is a modal property of individual Q because, had the coin landed differently, the outcome for Q would have been the same. This is what makes his situation unfair. The analogy correctly calls attention to one important fact. Whether a ship is inflammable does not depend on what actually happens to the ship. The ship could be inflammable even if it never catches fire but sinks and decays in the sea. Similarly, whether Q is treated unfairly does not depend only on what actually happens to Q. Q may die whether the situation is fair or unfair. But the analogy does not establish the similarity 34

Broome (1991: 114).

48

DOUGLAS MACLEAN

that Broome’s argument requires. The inflammability of the ship supervenes only on the natural properties of the ship, while the fairness or unfairness of the kidney allocation supervenes on the procedure that jointly determines the chances for both of the patients.

IX I can amplify and support this criticism of PPG with a different example involving the distribution of risk.35 Imagine a situation in which a population of three individuals is exposed to a risk of death, and consider the following three scenarios for how this risk might be distributed. Scenario A: Each individual faces a risk of death of one-third, and the risks are mutually independent. Scenario B: The population faces the risk jointly: either all will die or none will, and the risk of death is one-third. Scenario C: Exactly one person will die, while the other two will live. There is an equal probability (one-third) of each individual being the victim. Each individual faces an identical risk in each scenario: a one-third probability of dying and a two-thirds probability of surviving. But other differences can be seen from the social perspective, and the relevant facts about which scenario is better or worse are not just facts about each individual’s prospect in isolation or their aggregation, but include these social or holistic facts. For example, although the expected number of survivors is the same in each scenario, namely two, the variance in the outcomes to the group in the scenarios differs significantly. This difference becomes obvious when we consider the probability distribution over the number of survivors in each scenario, as shown in TABLE 2.5. These distribution patterns can be morally significant in obvious ways. If the community represents a culture, for example, and it is important that the culture survives, then scenario B might be much worse than A and C because of the risk in B that no one will survive. For the same reason, A might be worse than C, because there is also a chance in A that everyone will perish. In this respect, C involves no risk to the group. If instead solidarity among the community’s members is particularly important, then B might be better than the other alternatives. The comrades in B will stick together and survive or perish as a group! I take cultural survival and solidarity in this example to be communal values. Details about the nature of the group, the nature of the risk, and other features of the context can explain why we might regard one scenario as socially better or worse than the others. However, I see no way that the facts of any circumstance that determine which scenarios are socially better or worse can be dispersed to the good of individuals and then aggregated into a social value function. This example shows a different failure of the individualism expressed by PPG. 35

My discussion in this section relies on List and MacLean (manuscript).

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

49

Table 2.5. The probability distributions over the numbers of survivors in Scenarios A, B, and C Number of survivors

Probability

0 1 2 3

Scenario A

Scenario B

Scenario C

1/27 6/27 12/27 8/27

1/3 0 0 2/3

0 0 1 0

X I conclude with a brief comment on the value of a person or a person’s life. Broome’s discussion of the value of life is not directly related to his commitment to PPG, but it reveals something about his philosophical temperament, which is different from my own, and which helps explain the basis of my disagreement with him on equality and other issues. So these comments are, as it were, personal. Broome’s work is filled with original and subtle arguments, which have deeply influenced my own thinking. When I find myself disagreeing with one of his conclusions, I usually come to realize that this is not because I can locate flaws in his reasoning, but, rather, that we approach moral philosophy with fundamentally different assumptions or starting points. I have come to understand my own views better by reflecting on Broome’s work; he has been an important teacher for me. The value of a person’s life, as Broome understands it, is properly viewed from that person’s perspective. In Weighing Lives, he identifies the value of human life with the goodness of the person’s life, and he identifies personal goodness in turn with the amount of wellbeing the life contains.36 In Climate Matters he says simply, “The value of a life is nothing other than the value of the good things it contains.”37 Broome allows for the possibility that what happens outside a person’s life can affect a person’s wellbeing and thus the value of her life. He includes these possible effects, somewhat perversely in his own words, as part of her life for the purpose of talking about its goodness or value. He allows for backward causation in determining personal good, and in this respect his view is Aristotelian. If the good of a person’s life depends in part on the contribution he makes to an area of research, a political cause, or to how he has raised his children, then the good of his life may be determined in part by what happens after he dies. I agree with this part of Broome’s view, but I disagree about whether the value of a person’s life is determined only from a person’s perspective, and whether it is determined only by the good it contains. Broome’s view is not uncommon; I suspect that it is widely shared by economists and other philosophers. 36

Broome (2004: 94–5).

37

Broome (2012: 161).

50

DOUGLAS MACLEAN

There is one way of interpreting wellbeing as the aggregate of good experiences a life contains that would make this conception of the value of a life clearly implausible. Compare the lives of two people born into identical circumstances of affluence. The first person spends his inheritance on enjoyable activities and develops his talents simply to get the most happiness from these activities. The second person gives his wealth away and devotes his life to trying to help improve the lives of people who are living in miserable circumstances. His life is a struggle, and it contains far fewer enjoyable or rewarding events. Nevertheless, if the second person’s dedication to helping the needy is not entirely in vain, most of us would not hesitate in declaring his life to be more valuable than the life of the bon vivant. Now, this simply means that we need a different conception of personal good, a claim that Broome would not deny, but what else can this example tell us? I think it suggests that the value of a person’s life is neither something that is seen strictly from the person’s perspective, nor something that is solely a function of the amount of good it contains. Consider another example. A talented young man volunteers to serve his country as an officer in the army and is called off to war. Faced with a situation in which there is no alternative for saving the lives of the other men and women in his company, he throws himself on an explosive device and takes the full impact upon himself. He is killed, but the others in his company survive. If their survival was in some way critical to the war effort, then his courageous act may have had very broad and good consequences. But what shall we say about the value of his life? I cannot see how his sacrifice could be viewed as valuable from his perspective or interpreted as adding to his good or his wellbeing. He may have had plans and goals that were left frustrated by his death, and he may have cared little for his reputation or for posthumous honors. People who knew him may regard his death as courageous but tragic because there was so much good he might have realized but never will. Nevertheless, it seems to me that his life had great value, but his courage and his tragedy are values that can only be seen from the perspective of others. Broome’s philosophical temperament is reductionist in the following sense. The value of a person’s life is a function of the good things it contains, as viewed from the person’s perspective. Similarly, social value is a function of the good of individuals, and PPG is part of the explanation of how personal goods are combined to determine overall or general good. On this view, value enters the world through the experiences of people. For a non-reductionist committed to the view that all values are human values, value enters the world through human understanding and sentiment, but it can enter at any point and from different human perspectives. The value of a person’s life depends not only on how that life is experienced by the person who lived it but also on how it is seen and evaluated by others. Because we can adopt a social perspective, we can appreciate values like equality and fairness that attach irreducibly to communities or to social states of affairs, often through the processes of decision making that the members of the community agree to use. And because we are able to take an even more detached perspective and see ourselves as products of nature, remarkable organisms that inhabit the Earth and occupy roles in ecosystems, food chains, and energy

CHALLENGES TO THE PRINCIPLE OF PERSONAL GOOD

51

pyramids, we can see value in the complex interactions of the systems of which we are a part. As Leopold put it, “The land ethic simply enlarges the boundaries of the community to include soils, waters, plants, and animals, or collectively: the land . . . [It] changes the role of Homo sapiens from conqueror of the land-community to plain member of it. It implies respect for his fellow-members, and also respect for the community as such.”38 This includes direct and non-instrumental reasons both to promote social equality and to protect the beauty of the Earth.

References Broome, J. (1991) Weighing Goods. Oxford: Basil Blackwell. Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press. Broome, J. (2012) Climate Matters: Ethics in a Warming World. New York: W. W. Norton & Co. Broome, J. (2015) General and personal good: Harsanyi’s contribution to the theory of value. In Hirose, I. and Olson, J. (eds.) The Oxford Handbook on Value Theory. Oxford: Oxford University Press. Callicott, J. B. (1989) The conceptual foundations of the land ethic. In Defense of the Land Ethic: Essays in Environmental Philosophy. Albany: State University of New York Press. Daniels, N. (2008) Just Health: Meeting Health Needs Fairly. Cambridge: Cambridge University Press. Diamond, P. (1967) Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility: Comment. Journal of Political Economy 75: 765–6. Edgeworth, F. (1881) Mathematical Psychics. London: Kegan Paul. Goodpaster, K. (1978) On being morally considerable. The Journal of Philosophy 75: 308–25. Harsanyi, J. (1955) Cardinal welfare, individualistic ethics, and interpersonal comparisons of utility. Journal of Political Economy 63: 309–21. Reprinted in Essays in Ethics, Social Behavior and Scientific Explanation. Dordrecht: Reidel (1976). Jeffrey, R. (1987) Risk and human rationality. The Monist 70: 223–36. Leopold, A. (1949) A Sand County Almanac and Sketches from Here and There. Oxford: Oxford University Press. List, C. and MacLean, D. (manuscript) Irreducibly social risks? List, C. and Pettit, P. (2011) Group Agency: The Possibility, Design, and Status of Corporate Agents. Oxford: Oxford University Press. MacLean, D. (2010) Is ‘human being’ a moral concept? Philosophy and Public Policy Quarterly 30: 16–20. Moore, G. E. (1903) Principia Ethica. Cambridge: Cambridge University Press. Scanlon, T. M. (1978) Rights, goals, and fairness. In Hampshire, S. (ed.) Public and Private Morality. Cambridge: Cambridge University Press. Singer, P. (1979) Practical Ethics, 3rd Edition. Cambridge: Cambridge University Press. Williams, B. (1995) Must a concern for the environment be centred on human beings? In Making Sense of Humanity and Other Philosophical Papers. Cambridge: Cambridge University Press. 38

Leopold (1949: 204).

3 Metasemantics out of Economics? Anandi Hattiangadi

In Ethics out of Economics, John Broome argues for the use of formal economic techniques in philosophical ethics. He says, for instance, ‘economists have developed for their own purposes sophisticated methods of analysis that turn out to be useful in philosophical ethics. They can help not only with questions of practical ethics, but also with fundamental issues in ethical theory’ (Broome 1999: 1). As Broome points out, the formal aspect of utility theory ‘is nothing more than the axioms and the theorems. It can be reinterpreted by substituting in place of preference any other relation that happens to satisfy the axioms’ (Broome 1999: 9). Can formal methods of economics be usefully applied to philosophical issues outside of ethics? In this chapter, I shall argue for the use of certain formal techniques from economics in metasemantics.

Radical Interpretation Metasemantics is the study of the determinants of meaning, or other semantic properties, such as truth conditions, reference, or truth. One central question of metasemantics is whether radical interpretation is possible. Suppose that Mary is a perfectly rational being who knows N1—a statement of all positive non-semantic truths about the actual world, including all of the truths of physics, chemistry, biology, geology, psychology, and so forth.1 (It is a positive non-semantic truth that a particular particle, e, has a negative charge; it is a negative truth that e is not a ghost.) In addition to the positive truths, suppose that Mary knows T—a ‘that’s all’ statement to the effect that nothing more exists than is needed to satisfy N1.2 Now, let S1 be a statement of all the semantic truths about the actual world, that is to say, truths concerning the distribution of such semantic properties as meaning, truth 1 N can include phenomenal truths about the world, such as that a subject has an experience with a particular phenomenal character at a certain time, as well as non-semantic indexical truths, fixing the location of a centre in a world. 2 Any world that satisfies N1T is a minimal non-semantic duplicate of our world. From N1T, Mary can infer that there are no ghosts, but not from N1 alone.

METASEMANTICS OUT OF ECONOMICS ?

53

conditions, truth-values, reference, content, and so forth. S1 is a massive, global interpretation, a statement of how things actually are in all semantic respects. Suppose that a radical interpreter knows all of the non-semantic truths about the actual world— that is, she knows N1T. Suppose, further, that she is perfectly rational and omniscient with regard to all a priori knowable truths. Radical interpretation is possible only if it is possible for such an ideal being to logically deduce S1 from what she knows, namely, N1T together with some a priori truths.3 As an ideally rational being, the radical interpreter is constrained by principles of rationality. If radical interpretation is possible, then these principles of rationality, together with the a posteriori information (N1) and a priori truths must determine the ideal interpreter’s choice of S1 from the set of interpretations X = {S1 . . . Sn}, given her knowledge of N1T.4 Is radical interpretation possible? If it is, then what are the relevant a priori truths governing choice of interpretation? These a priori truths are often characterized as constraints or principles governing the rational selection of interpretations. For instance, one might think that Mary should select the most charitable interpretation in the sense that she should pick the one that maximizes truth in the subject’s system of beliefs (Davidson 1990), or which would render the subject maximally rational, given her experiences (Lewis 1974).5 Alternatively, one might adopt a principle of 3 As I characterize it, radical interpretation is fully reductive, in the sense that the radical interpreter has access to no semantic information whatsoever. This is not always assumed. For instance, Davidson (1990) allows the radical interpreter to have information about what a subject ‘holds true’, that is, what sentences a subject believes to be true. In one version of Lewis’s interpretationism, discussed by Williams (2007), the sentential data includes the sentences that constitute a subject’s ‘global theory’, which naturally suggests that they are sentences that the subject assents to, or is disposed to assent to. But of course such truths are also semantic. 4 I have simplified somewhat in assuming that the radical interpreter must pick a unique interpretation. Some philosophers, such as Lewis, might be inclined to allow the radical interpreter to select a set of interpretations, to allow for vagueness. Suppose S1 assigns a certain precise extension to the English predicate ‘bald’, and S2 assigns an extension that is very slightly different from S1 (e.g. it includes people with one more hair on their heads than are included in the extension of ‘bald’ assigned by S1). If vagueness is a semantic phenomenon, then arguably non-semantic facts will be indeterminate between S1 and S2 (Lewis 1974). In other words, the radical interpreter need not be required to select a unique interpretation. However, the simplifying assumption that the radical interpreter should select a unique interpretation is acceptable because nothing in the argument turns on vagueness. Moreover, one could simply allow interpretations to capture semantic vagueness. For instance, S1 need not assign a precise extension to ‘bald’; it could assign a set of admissible extensions and state that ‘x is bald’ is determinately true iff x is a member of every extension in the set of admissible interpretations, or something to that effect. If all the truths about vagueness are included in interpretations, then the radical interpreter can once again be expected to select a unique interpretation. 5 Strictly speaking, assignments of meaning will be determined more precisely by way of convention. For instance, Lewis argues that ‘the convention whereby a population P uses a language £ is a convention of truthfulness and trust in £’ (2008: 658). This is not an independent, fully reductive criterion, however, since Lewis defines these conventions in terms of subjects’ preferences and beliefs:

To be truthful in £ is to act in a certain way: to try never to utter any sentence of £ that are not true in £. Thus it is to avoid uttering any sentence of £ unless one believes it to be true in £. To be trusting in £ is to form beliefs in a certain way: to impute truthfulness in £ to others, and thus to tend to respond to another’s utterance of any sentence of £ by coming to believe that the uttered sentence is true in £. This, in effect, constrains what is involved in rationalizing a subject’s linguistic behaviour.

54

ANANDI HATTIANGADI

epistemic charity according to which she should maximize a subject’s knowledge (Williamson 2004). Perhaps Mary should select the interpretation that assigns the most ‘natural’ or ‘joint carving’ properties as referents of concepts and expressions (Lewis 1999; Sider 2011). According to one version of Lewis’s view, the perfectly natural properties are those that are referred to by our most fundamental physical theory, and a property F is more natural than G iff the members of F resemble one another more closely in perfectly natural respects than the members of G. Alternatively, one might think that the assignments should best match certain causal relations (Field 1975; Williams 2007). All of these suggestions have some initial plausibility. The trouble is that the constraints that have been suggested have also been shown to be insufficient—at least when considered in isolation. As is well known, for instance, S1 is not the most charitable interpretation. As many philosophers have argued, it is possible to construct a crazy interpretation that is as charitable as S1 but is incompatible with it (Quine 1968; Field 1975; Putnam 1980; Kripke 1982; Williams 2007). The trick is to permute the referents of names in some systematic way and then alter the meanings of predicates so that the changes ‘cancel each other out’ and, at the level of whole sentences or thoughts, truth-values remain unchanged. This means that Mary, appealing to the principle of alethic charity and knowledge of N1T, could not rule out crazy interpretations a priori, and hence could not deduce S1 a priori from N1T. Similar objections have been raised to the appeal to naturalness (Williams 2007), rationality (Eriksson and Ha´yek 2007), and causal constraints (Kripke 1982; Lewis 1999). A natural reaction to these concerns is to consider whether a number of constraints might work in concert. Indeed, this thought may well have been in play all along—in ‘Radical Interpretation’, Lewis puts forward no less than six constraints, and suggests that they work together. Even if one or more constraints is emphasized (such as naturalness in Lewis (1999)), there is usually a compositionality constraint assumed in the background (cf. Field 1975). Sider (2011), in a recent defence of a broadly Lewisian approach to metasemantics suggests that ‘joint-carvingness’ (much like Lewisian naturalness) can be combined with other constraints to determine the intended interpretation. The thought is that Mary is rationally required to select the interpretation that is best overall in meeting a number of individual constraints. Here is where economics comes in. Let’s say that X is the set of alternative interpretations and that each constraint determines a weak ordering of interpretations denoted xRiy (‘x is at least as good as y with respect to the ith constraint’). From Ri, a strict ordering Pi (‘x is better than y with respect to the ith constraint’) and indifference Ii (‘x is equally as good as y with respect to the ith constraint’) can be defined. For instance, the alethic charity constraint will determine a weak ordering of interpretations, where interpretation x is at least as good as y with respect to alethic charity iff the number of beliefs x renders true is greater than or equal to the number of beliefs y renders true. Similarly, the naturalness constraint will determine a weak

METASEMANTICS OUT OF ECONOMICS ?

55

ordering of interpretations such that x is at least as good as y with respect to naturalness iff the naturalness of assignments of referents to lexical items made by x is greater than or equal to the naturalness of assignments of referents to lexical items made by y. And so on, for each constraint. A profile is the n-tuple of individual orderings {Ri}. What does Mary do with this profile? She needs to find a way of aggregating the individual orderings of interpretations so as to determine which interpretation is best overall. She needs to arrive at an overall ordering, xRy, which is a function of the individual orderings—call that function an OIF for ‘overall interpretation function’. Of course, there are countless OIFs. Some rank S1 as best overall; others rank some crazy interpretations highest. Which one to use? Once again, Mary needs some kind of rational basis for picking one such function over the rest. Her predicament is structurally analogous to those investigated by social choice theory.6

Social Choice Theory Social choice theorists investigate functions from individual preferences (Ri) to a social preference (R) over the set of states of the world in the set X. Suppose five committee members wish to elect a chair from their ranks. Each member ranks the five candidates according to her own preferences. Each member’s preferences determine a weak ordering of states of the world, denoted xRiy (x is at least as good as y according to the ith member), and from this, a strict ordering xPiy and indifference, xIiy can be defined. The social choice theorist investigates ways to determine an overall, or general preference ordering, xRy from the individual ones. Once again, there are many social welfare functions (SWFs) from individual to overall orderings. The social choice theorist attempts to find constraints that will rule out unreasonable methods of aggregation yielding only the set of aggregation methods that are reasonable (Sen 1999). Famously, Kenneth Arrow proved that no social welfare function (SWF) could satisfy a small number of seemingly plausible and relatively weak constraints on a SWF (Arrow 1963). First, Arrow assumes that the individual orderings xRiy are transitive (if xRiy and yRiz, then xRiz) and complete—they have preferences over all of the alternatives in the set. The additional constraints are as follows: U (unrestricted domain): The domain of the SWF is the set of all logically possible profiles.

6

Social choice theory, and in particular, Arrow’s theorem, has recently been applied to other domains outside of ethics. Michael Morreau (2010) has, for instance, applied Arrow’s theorem to similarity judgements, particularly in relation to counterpart theory (‘x resembles me at least as much as y overall’). Samir Okasha (2011) has applied them to theory choice in the sciences, where values such as simplicity, scope, fruitfulness (etc.) need to be aggregated into an overall ranking.

56

ANANDI HATTIANGADI

P (weak Pareto): If every individual strictly prefers x to y, then x is socially preferred to y. D (non-dictatorship): There is no individual who dictates the social preference in the sense that if she strictly prefers x over y, then x is socially preferred to y. I (independence of irrelevant alternatives): The social preference between x and y can only depend on individuals’ preferences between x and y. Arrow’s theorem is that, so long as there are at least three alternative states and a finite number of voters, U, P, D, and I cannot be satisfied (Arrow 1963; Sen 1986). As Broome suggests, formal economic theories are nothing more than the axioms and the theorems. We can reinterpret them by substituting in place of voter’s preferences, different relations that satisfy the axioms. In the next section, I will make a prima facie case for applying Arrow’s axioms in the case of radical interpretation, where each individual preference ordering is reinterpreted as an ordering of interpretations based on individual criteria (such as charity, naturalness, rationality, and so forth). Metaphorically, we can imagine the radical interpreter as a committee composed of single-minded individuals ranking interpretations: Charity, who cares only about charity, Natalie, who only cares about naturalness, Radha, who only cares about rationality, and so forth. But that is really just picturesque. It is sufficient that the axioms are satisfied. I will argue in the next section that they are.

Back to Radical Interpretation Arrow assumes that individual preference orderings are transitive and complete. Though there have been reasons to question whether these should be weakened in the case of social choice (Sen 1986), there seems to be no reason to weaken these constraints here. Transitivity seems to be analytic of ‘at least as good as’, ‘strictly better than’, and ‘indifferent between’, and the radical interpreter aims to select the interpretation that is best (i.e. strictly better than the rest) overall. Completeness, once again, might be implausible if the individual orderings are meant to capture the preferences of individuals over infinitely many states of the world, given human cognitive limitations. However, no comparable cognitive limitations plague the radical interpreter, who can work out each individual ordering a priori, given full information about any logically possible Ni together with full information about every logically possible Si. U says that the domain of the SWF is the set of all logically possible profiles. Is it plausible that the domain of the OIF is the set of all logically possible profiles? Initially, this may seem implausible. After all, the radical interpreter is given the non-semantic truths about the actual world, N1, and these, together with a certain criterion, determine an individual ordering. However, this initial appearance is misleading. The reason is that though we are primarily interested in the actual

METASEMANTICS OUT OF ECONOMICS ?

57

world, there are many logically possible worlds with different logically possible Ni. Many non-semantic truths are contingent, and if they had been different, the criteria would determine different individual orderings of interpretations. For instance, consider a world, w, that differs from our world in non-semantic facts about our use of ‘cat’ and ‘dog’, and where ‘cat’ means ‘dog’ and vice-versa. If we let N2 be a statement of all non-semantic truths at w, and let S2 be a statement of the semantic truths at w, the OIF should rank S2 over S1 in relation to w, but S1 over S2 in relation to the actual world. Of course, one might think that all we need is an OIF that works for the actual world. Who cares whether it gets the right results in all logically possible worlds? However, in radical interpretation, the principles governing aggregation must be a priori, and hence, should not depend on knowledge of any contingent a posteriori truths apart from what is contained in Ni. This suggests, then, that the domain of the OIF ought to be any profile determined by the set {NiT}, for all logically possible Ni. A further worry one might have is that some criteria are not independent of one another. For instance, alethic charity orders interpretations by how many true assertions and beliefs are assigned to a subject, and epistemic charity orders them by how much knowledge is assigned. Since knowledge entails true belief, the two orderings cannot vary independently of one another. However, Arrow’s theorem only requires that there are at least three logically independent voters. There are arguably at least three logically independent criteria—such as, for instance, alethic charity, naturalness, and causality. P says that if interpretation x is better than y according to each individual criterion, then x is better than y overall. This seems plausible. The overall ranking of the interpretations should be wholly determined by the individual rankings in relation to the criteria. For, there is nothing else that could be relevant to determining the overall ranking, since all of the relevant truths, both a priori and a posteriori, have been taken into consideration in the determination of the profile. If the overall ranking were not determined by individual rankings alone, then something else (either a falsehood or a randomizing mechanism) would play a role in determining the overall ranking. But this would tell against the overall ranking having a fully rational basis. D says that no individual dictates the social preference. This is clearly plausible in the case of social choice, since we want to make sure that no single voter’s ranking determines the overall ranking. In the metasemantic case, D* says that there is no single criterion whose strict ranking dictates the overall ranking. That is, there is no Pi which is such that for all x, y, if xPiy, then xPy. Is this plausible in this case? It seems to me that it is, though for very different reasons. In the case of radical interpretation, if any individual constraint determined the overall ranking, the wrong result would be determined. This is because each individual constraint, considered entirely in isolation, does not rank the intended interpretation highest. For instance, alethic charity ranks many interpretations higher than S1. For, given human fallibility, there are many sentences that we assent to, or hold true,

58

ANANDI HATTIANGADI

which are false. Now, consider an interpretation, S3, which assigns meanings in such a way that all of our true beliefs come out true, and in addition, some of our false beliefs come out true. For instance, suppose that some subject actually believes that the Earth is flat. Suppose that S3 assigns the Earth to the extension of ‘flat’ in the subject’s language, so that ‘the Earth is flat’ comes out true. The meaning of ‘flat’ may thereby become gruesome (like Goodman’s predicate, ‘grue’), referring to all the flat things and only the flat things with the exception of the Earth. Gruesome though it may be, S3 would be strictly more charitable than S1. If the strict ordering in respect of charity were to dictate the overall ordering, then it would be determined that S3 is true, rather than S1. That S3 might wind up being wildly unnatural or non-compositional is beside the present point, because compositionality and naturalness are further constraints, and we are considering an ordering with respect to alethic charity alone. Similarly, consider an interpretation, S4, that maps every name onto a microphysical particle, and every predicate onto a highly natural property.7 S4 will be strictly more natural than S1. For we speak of many highly unnatural things: of shoes and ships and sealing wax, of cabbages and kings. Once again, S4 might well be less rationalizing, or involve fewer causal relations than S1, but that is neither here nor there, since we are considering a strict ordering by naturalness alone. Similar points can be raised against the epistemic constraint, since we are far from omniscient, against the causal constraint, because sometimes we talk about things that lack causal powers, and so forth.8 Of course, whether every criterion ranks some interpretation above the intended interpretation depends on what, more precisely, the criteria are. Yet I have left this open to some degree. However, if there were a single criterion that ranked the intended interpretation highest, there would be no need to aggregate—that criterion would be all that the radical interpreter would need to deduce S1 from N1. Indeed, if there were such a constraint, the radical interpreter would be ill-advised to aggregate. But then, what of the intuitive plausibility of various forms of charity, naturalness, causality, and so forth? Those considerations would then turn out to be irrelevant to the determination of meaning. Thus, it seems that we cannot let the strict ordering determined by any individual criterion dictate the strict ordering overall, irrespective of the orderings with respect to other criteria. There should be no metasemantic dictators.9 7

See Hawthorne (2008) for examples. One might well worry that this motivation for D is not a priori—i.e. it depends on our knowledge of semantic truths about the actual world. But this need not jeopardize the project. For we can allow that the principles that govern radical interpretation have something to do with our concepts (e.g. the concept of a language, of a person, of communication, etc.). We can thus ‘work back’ from our knowledge of semantic truths to uncover the rational principles governing radical interpretation. 9 Note that D does not rule out various decision procedures which involve a criterion acting as a tiebreaker. Suppose, for instance, that we start by ranking interpretations by alethic charity and then break ties between the highest ranked by appeal to naturalness. This method would not violate D since neither 8

METASEMANTICS OUT OF ECONOMICS ?

59

I says that the overall ranking of interpretations x and y can only depend on the rankings of x and y by the individual criteria. This was intuitively plausible in the case of an election—the social ordering of Smith and Jones should depend on the individual orderings of Smith and Jones, and not on how some third candidate, Brown, is ranked. The same seems to hold in the metasemantic case. In this case, the individual orderings of interpretations x and y are determined by the relevant a priori and a posteriori truths. Since the overall ordering should depend on these truths alone, no further truth, such as how some third interpretation Sk is ranked, should be relevant to fixing the overall ranking. If the ranking of Sk did influence the overall ranking of Si and Sj, the overall ranking would not have a basis in the a priori and a posteriori truths determining the individual rankings of Si and Sj.

Concluding Remarks I have made a preliminary case for the applicability of Arrovian axioms to radical interpretation. If Arrow’s axioms apply, then radical interpretation is impossible, for there is no rational basis for aggregating the orderings determined by individual criteria (and no individual criterion determines the intended interpretation, on its own). That said, there are several ways out of Arrow’s impossibility theorem, and a fuller discussion would consider whether any such exit strategies are available in the metasemantic case. I have already suggested that rejecting or weakening the axioms seems to be unwarranted in the metasemantic case. Another way out of Arrow’s impossibility involves increasing the information contained in the individual rankings.10 Crucially, Arrow does not allow for interpersonal comparisons of utility. Yet, if interpersonal comparisons of utility are possible, then versions of Arrow’s constraints can be satisfied (Sen 1986). A further question, therefore, is whether inter-criterial comparisons are possible. Though many such further questions remain, I hope to have at least shown that some formal techniques from economics can help to provide a way of addressing an issue in metasemantics: whether radical interpretation is possible.

alethic charity nor naturalness would alone determine the overall ordering. For suppose that there is a set of interpretations that is tied for first place by charity. There could still be an interpretation that is ranked below those in that set by charity, yet higher than any in that set by naturalness. If naturalness only works as a tie-breaker, the overall ordering will not in this case be determined by either the individual ordering by charity, or by the individual ordering by naturalness alone. 10 Some suggest that it might be enough to avoid Arrow’s impossibility that the individual rankings are cardinal rather than merely ordinal, as Arrow assumes (see Samir Okasha 2011). However, Sen (1986) points out that this is insufficient to avoid the impossibility theorem.

60

ANANDI HATTIANGADI

References Arrow, K. (1963) Social Choice and Individual Values. 2nd edn. New York: Wiley. Broome, J. (1999) Ethics out of Economics. Cambridge: Cambridge University Press. Davidson, D. (1990) Radical interpretation. In Inquiries into Truth and Interpretation. Oxford: Oxford University Press. Eriksson, L., and Ha´jek, A. (2007) What are degrees of belief? Studia Logica 86(2): 183–213. Field, H. (1975) Noûs 9(4): 375–405. Hawthorne, J. (2007) Craziness and metasemantics. The Philosophical Review 116 (3): 427–40. Kripke, S. (1982) Wittgenstein on Rules and Private Language: An Elementary Exposition, Cambridge, Mass.: Harvard University Press. Lewis, D. (1974) Radical interpretation. Synthese 23: 331–44. Lewis, D. (1999) New work on a theory of universals. In Papers in Metaphysics and Epistemology. Cambridge: Cambridge University Press. Lewis, D. (2008) Languages and language. Reprinted in A. P. Martinich, ed. The Philosophy of Language. Oxford: Oxford University Press. Morreau, M. (2010) It simply does not add up: Trouble with overall similarity. Journal of Philosophy 107(9): 469–90. Okasha, S. (2011) Theory choice and Social Choice: Kuhn versus Arrow. Mind 120(477): 83–115. Putnam, H. (1980) Models and reality. The Journal of Symbolic Logic 45(3): 464–82. Quine, W. V. (1968) Ontological relativity. The Journal of Philosophy 65(7): 185–212. Sen, A. (1986) Social choice theory. In K. J. Arrow and M. D. Intriligator (eds.) Handbook of Mathematical Economics, III. London: Elsevier. Sen, A. (1999) The possibility of social choice. The American Economic Review 89(3): 349–78. Sider, T. (2011) Writing the Book of the World. Oxford: Oxford University Press. Williams, J. R. G. (2007) Eligibility and inscrutability. The Philosophical Review 116(3): 361–99. Williamson, T. (2004) Philosophical ‘intuitions’ and scepticism about judgment. Dialectica 58(1): 109–53.

4 Separability Iwao Hirose

1 Introduction John Broome’s Weighing Goods contains so many issues that value theorists must examine very carefully.1 Amongst others, I want to concentrate on the issue of separability in this chapter. In Weighing Goods, separability plays a central role. There is no doubt about that. At the beginning of Weighing Goods, Broome clearly states that a major goal of his project is “to examine just how successfully the utilitarian principle of distribution can be defended by arguments based on separability, like Harsanyi’s” (Broome 1991: 29). The notion of separability has been analyzed and used in economics (Blackorby et al. 1978; Debreu 1954, 1960; Gorman 1968; Wakker 1988). It is Broome’s Weighing Goods that first used the notion of separability to elucidate the structure and farreaching scope of the utilitarian principle of distribution. By the utilitarian principle of distribution, Broome means the principle such that the general good is represented by the additive function of personal good. The utilitarian principle of distribution is more general than fully fledged utilitarianism. Fully fledged utilitarianism requires a particular interpretation concerning what is personally good. Typically, fully fledged utilitarianism takes net pleasure to be the only thing that counts as personal good. In contrast, the utilitarian principle of distribution takes a person’s personal good to be everything that is good for her. Broome does not assume a particular account of personal good, and hence his notion of personal good is neutral amongst competing accounts of personal good. Thus, what he means by the utilitarian principle of distribution is more general than utilitarianism as such. Broome thinks that separability in two dimensions (people and states of nature) supports the interpersonal addition theorem and the Bernoulli hypothesis, and that the interpersonal addition theorem and the Bernoulli hypothesis imply the utilitarian principle of distribution. Furthermore, Broome thinks that separability in the

1

I thank Karsten Klint Jensen for detailed comments on this chapter.

62

IWAO HIROSE

dimension of time provides an additional support for the Bernoulli hypothesis. Thus, separability plays a key role in establishing the utilitarian principle of distribution. However, it is the notion of separability that poses a fundamental challenge to Broome’s project. The challenge is that separability in the dimension of time may be implausible. Broome supports separability in the dimensions of people and states of nature. If we accept separability in these two dimensions, it may be quite natural to accept separability in the dimension of time. However, in Weighing Goods, Broome is not sure about separability in the dimension of time (Broome 1991: 29). If separability of time is unacceptable, then the separability-based argument for the utilitarian principle of distribution is partly undermined. In this chapter, I will critically assess Broome’s objection to separability of time. I am not persuaded by his objection, but agree with his main claim that separability of time is implausible. In Section 2, I will sketch Broome’s separability-based argument for the utilitarian principle of distribution. In Section 3, I will explain Broome’s objection to separability of time on the basis of the value of longevity. In Section 4, I will argue that the value of longevity can be dispersed successfully. In Section 5, I will explain why separability of time does not fit in Broome’s general analytical framework.

2 Separability-based Argument I will first define the notion of separability. Let R be a weak relation defined on n-component vectors, say (x1, x2, . . . , xn). Call each place in this vector a location, and the term occupying each place the occurrence at that location. Assume that R is reflexive, transitive, and complete, and hence that R is a complete weak ordering. Pick any subset of locations and collect the occurrences at the locations in this subset to form a subvector. Let R* be an ordering of the subvectors conditional on the occurrences at the other locations, and call it a conditional ordering. Then, the notion of separability is defined as follows. A subset of locations is separable under the ordering R if and only if the conditional ordering determined by R* on the subvectors corresponding to this subset is the same, regardless of the constant occurrences at the other locations. An ordering is weakly separable if and only if every location is separable on its own. An ordering is strongly separable if and only if every subset of locations is separable. Suppose that U is a utility function defined on n-component vectors. An ordering is additively separable if and only if it can be represented by a sum of subutilities, i.e. Uðx1 ; x2 ; …; xn Þ ¼ u1 ðx1 Þ þ u2 ðx2 Þ þ … þ un ðxn Þ An ordering is crosscutting separable if and only if the locations are arranged on a rectangular grid, where each row and each column is separable in the ordering. Broome then appeals to two economic theorems about separability. The first theorem is about the relation between strong separability and additive separability.

SEPARABILITY

63

This theorem was established by Debreu (1960) and Krantz et al. (1971). The second theorem is about the relation between crosscutting separability and additive separability. It is known as the overlapping theorem. It was proved by Gorman (1968). First separability theorem: An ordering is strongly separable if and only if it is additively separable.2 Second separability theorem: If an ordering is crosscutting separable, then it is additively separable.

These two theorems play a central role in Broome’s analysis of the utilitarian principle of distribution. Now I will outline Broome’s separability-based argument for the distributive principle of distribution. A large part of Weighing Goods is devoted to establishing separability in the dimensions of people and states of nature. The main result from his analysis of separability of people and states of nature is the interpersonal addition theorem. The interpersonal addition theorem. If the individual and general betterness relations are coherent, and if the principle of personal good is true, then the general betterness relation can be represented by an expectational utility function that is the sum of expectational utility functions representing the betterness relations of individuals. (Broome 1991: 202) The interpersonal addition theorem connects together the aggregation of good across the dimensions of people and states of nature, and tells us that the general betterness ordering is additively separable with respect to people. When this theorem is combined with Bernoulli’s hypothesis about good, we have a remarkable result: the same utility functions supply the weights in aggregation across the dimensions of people and states of affairs. This result means that the interpersonal addition theorem is about the utilitarian principle of distribution. What is Bernoulli’s hypothesis? The thesis is stated as follows. Bernoulli’s hypothesis. One alternative is at least as good for a person as another if and only if it gives the person at least as great an expectation of her good. (Broome 1991: 142, 213) The basic idea of the hypothesis is that if an expectational utility function represents good ordinally, it also represents it cardinally. If Bernoulli’s hypothesis is true, the same utility functions supply the weights in aggregation across the dimensions of time, people, and states of nature, and represent actual quantity of good. Consequently, we have the utilitarian principle of distribution. If Bernoulli’s hypothesis is

2 There are additional conditions that must be met. One of them is continuity. For example, leximin is a strongly separable ordering, but is not additively separable because it is a discontinuous ordering and cannot be represented by a real-valued function.

64

IWAO HIROSE

false, the general betterness relation can be represented by a utility function that is the sum of non-linear utility functions representing each person’s betterness relation. Consequently, we have the non-utilitarian principle of distribution such as prioritarianism. Thus, whether or not we can derive the utilitarian principle of distribution from the separability-based argument depends on the truth of Bernoulli’s hypothesis. How does Broome argue for Bernoulli’s hypothesis? He thinks that the interpersonal addition theorem provides a strong ground for Bernoulli’s hypothesis. He also thinks that the intertemporal addition theorem provides additional ground. In Chapter 11, Broome takes on separability of time. He starts with what he calls the principle of temporal good. Principle of temporal good. (a) If two alternatives are equally good for a person at every time, they are equally good for her, and (b) if one alternative is at least as good as another for the person at every time and definitely better for her at some time, it is better for her. (Broome 1991: 225) The basic claim of the principle of temporal good is simple: all good is dated. Then, we obtain the intertemporal addition theorem. The intertemporal addition theorem. If a person’s betterness relation is coherent, and so are all her dated betterness relations, and if the principle of temporal good is true, then the person’s betterness relation can be represented by an expectational utility function that is the sum of expectational utility functions representing her dated betterness relations. (Broome 1991: 226) The intertemporal addition theorem is parallel to the interpersonal addition theorem. These two theorems jointly imply the most general result of Broome’s Weighing Goods. The combined addition theorem. Suppose the general betterness relation is coherent, and so is each person’s betterness relation, and so are all of each person’s dated betterness relations. And suppose that the principle of personal good is true, and so is the principle of temporal good for each person. Then the general betterness relation can be represented by a utility function that is the sum of utility functions representing each person’s betterness relation. And each of these in turn is the sum of utility functions representing the person’s dated betterness relations. (Broome 1991: 226–7) Broome argues that the combined addition theorem supports the truth of Bernoulli’s hypothesis although it does not prove Bernoulli’s hypothesis conclusively.3 According to him, if the premises in the combined addition theorem are all true, then Bernoulli’s hypothesis is true and, consequently, the utilitarian principle of distribution is true. 3 I will set aside the important issue of whether the combined addition theorem really supports the truth of Benoulli’s hypothesis here. See Jensen (1995), Risse (2002), and Weymark (2005).

SEPARABILITY

65

Are all premises in the combined addition theorem defensible? Broome thinks that the weakest part is separability of time. He examines arguments for and against separability of time. However, in Weighing Goods, Broome is inconclusive. He thinks that, although there is some prima facie evidence against it, separability of time can be defended, for example, if some ethical argument is added to the disuniting metaphysics of personhood of Parfit (1984).4 Thus, Broome does not settle the matter in Weighing Goods.

3 Doubt about Separability of Time It is in his Weighing Lives (2004) where Broome puts forward a clear argument against separability of time in the cases of variable population size. He bases this argument on the value of longevity. Longevity is a pattern good. Pattern goods are goods that cannot be attributed to some individual, at some time, in some state of nature. Pattern goods threaten separability unless they are dispersed to some individual, at some time, in some state of nature. If longevity is good and its goodness cannot be dispersed to some time, then longevity contradicts separability of time. Thus, Broome’s argument contains two steps. In the first step, he argues that longevity is good. In the second step, he argues that the value of a person’s longevity cannot be dispersed to the various times of her life. Let me briefly present his argument for the value of longevity. Consider Table 4.1. The four quadrants show four different possible distributions. The cells show the two people’s condition at each time, and Ω indicates that the person is not alive at that time. In A, person p lives through both times and q never lives. In B, p lives at time 1, and q lives at time 2. That is, p replaces q. Similarly, in D, q lives through both times and p never lives. In C, q replaces p. Table 4.1.

Person p Person q

Time 1

Time 2

1 Ω

1 Ω

Person p Person q

Time 1

Time 2

Ω 1

1 Ω

Distribution A Time 1 Person p Person q

1 Ω

Distribution C

Time 2

Ω 1 Distribution B

Time 1 Person p Person q

Ω 1

Time 2

Ω 1 Distribution D

Source: Broome 2004: 107.

4

Economists tend to accept separability of time. See Mirrlees (1982), Schelling (1986), and Strotz (1955). But Deaton and Muellbauer (1980) hesitate to accept separability of time.

66

IWAO HIROSE

Separability of time implies that A is better than B if and only if C is better than D. However, Broome claims that such an implication is counterintuitive, and that A is better than B, while D is better than C. This is because he thinks it better that a person continues to live rather than that she is replaced. People value longevity. But the length of a person’s life is not something that shows up when we look at temporal parts of her life separately. If A is better than B, and D better than C, then separability of time is violated. I am not persuaded by Broome’s argument. As Broome mentions in passing, there are limits to this intuition (Broome 2004: 108). I am not sure that it is better to prolong a 100-year-old person’s life for another 100 years, rather than have a new person live for 100 years. It may be better. But I do not have a clear intuition simply because I cannot imagine what the life of 200 years looks like. I have a clear intuition about prolonging a 40-year-old person’s life for 40 years: it is definitely better than replacement. But, in cases where the alternative includes a life that continues for a long time, I do not have a clear intuition. To be sure, Broome’s point is very modest. If there are some cases in which life extension is better than replacement, that is enough to contradict separability of time. So, if longevity is understood as the longest lifespan we can expect under the normal circumstances at the present, I agree with Broome’s intuition. For example, if each time period in Table 4.1 stretches over 40 or 50 years, I agree with Broome that A is better than B, and that D is better than C. However, when I judge that A is better than B, I have a different explanation. I judge that A is better than B because B includes premature deaths. When someone dies at the age of 100, I do not really feel sorry. In contrast, when someone dies at the age of 60, I feel sorry to some degree. When she dies at the age of 40, I feel sorry with a greater degree. When I say that A is better than B, I find a bad thing for p and q in B, rather than a good thing for p in A. The bad thing is premature death: the death that deprives her of a reasonably expected range of opportunities to pursue her lifetime project. The disvalue of premature death is different from the badness of death in general, or the loss of good that could have been obtained if she had continued to live. I think A is better than B because two people suffer from premature death in B, whereas there is no premature death in A. How many years should one continue to live in order to avoid premature death? I do not have a non-arbitrary answer to this question. But I have a rough answer that is not entirely arbitrary: the life expectancy at birth for Japanese women. In 2012, Japanese women were the group of people who enjoyed the world’s longest life expectancy, living on average for 86 years. There is no biological reason for believing that Japanese women should live longer than any other group in the world. Given a healthy diet, active lifestyle, good medical system, and so on, any group in the world can live for as many years as Japanese women. Thus, I take it as a bad thing if a person lives for fewer years than the expected lifespan for average Japanese women. The Global Burden of Disease Study actually takes this view (Murray et al. 2012). It uses

SEPARABILITY

67

Japanese women’s life expectancy at birth as the benchmark for measuring the global health deficit. So the criteria for non-premature death can be determined in a way that is not completely arbitrary. Broome admits that we value longevity within certain limits (Broome 2004: 108). He does not think, for example, prolonging a 100-year-old person’s life for another 100 years, rather than have a new person live for 100 years, would be a good thing. So Broome does not think longevity is good for its own sake under any circumstances. In this respect, there is no disagreement between Broome and me. The difference is that Broome takes longevity to be good, whereas I take a premature death to be bad.

4 Dispersing the Disvalue of Premature Death Broome thinks that longevity is good, but that the value of longevity cannot be dispersed. The dispersion of pattern goods is a general strategy to defend separability. Broome uses this strategy when he defends the value of equality. Equality of wellbeing is a pattern good. Its value, however, does not show up in some individuals’ personal good. Broome argues that the value of equality can be dispersed and attributed to individuals’ personal good. Here is one way to do it. Inequality is a bad for worse-off people. So the badness of inequality is dispersed across the worse-off people. Thus, the personal good of a worse-off person is given by her wellbeing and the badness of inequality. The general good is then represented by the sum of different people’s personal good, as distinguished from different people’s wellbeing. This way, the value of equality is dispersed, although it is a pattern good. Broome thinks that, even if equality is good, separability of people can be defended by dispersing the badness of inequality. If longevity is a good that does not show up in any discrete time period, why do we not disperse the good of longevity and defend separability of time? In Chapter 7 of Weighing Lives, Broome considers two ways to disperse the value of longevity, but claims that neither way can disperse the value of longevity successfully (2004: 113–16). Is he right? Consider the example Broome uses. Table 4.2 shows two distributions. If we support separability of time and do not accord any dispersed value to longevity, then A is just as good as B. But, according to Broome, if we value longevity, B is better Table 4.2.

p q r s t

T1

T2

1 Ω Ω Ω 1

Ω Ω 1 Ω Ω 1 Ω Ω 1 1 Distribution A

Source: Broome 2004: 113.

T3

T4 Ω Ω Ω 1 1

p q r s t

T1

T2

T3

1 1 Ω Ω Ω

1 Ω 1 Ω Ω 1 Ω 1 Ω Ω Distribution B

T4 Ω Ω 1 1 Ω

68

IWAO HIROSE

Table 4.3.

p q r s t

T1

T2

T3

1  3Æ Ω Ω Ω 1

Ω Ω 1  3Æ Ω Ω 1  3Æ Ω Ω 1 1 Distribution A

T4 Ω Ω Ω 1  3Æ 1

p q r s t

T1

T2

T3

1Æ 1Æ Ω Ω Ω

1Æ Ω 1Æ Ω Ω 1Æ Ω 1Æ Ω Ω Distribution B

T4 Ω Ω 1Æ 1Æ Ω

than A. There are possible ways to disperse the value of longevity. Broome considers two ways. But he claims that neither way is successful. More specifically, these two ways judge that A is better than B. Here, I offer my own method for dispersing the value we assign to longevity. In the previous section, I offered my interpretation of longevity. My interpretation is that it is premature death, rather than longevity, that is good. How can we disperse the badness of premature death? Here is one way. Let Æ be the badness of a period lost from premature death. Furthermore, the badness of premature death is proportional to the periods lost. The younger a person dies, the greater the badness of premature death. Then, the badness of premature death is dispersed equally across periods lived. Table 4.3 shows the temporal personal good of five people that includes the disvalue of premature death. For the sake of argument, I will assume that each period stretches over 20 years, and the reasonably expected duration of life is 80 years. In A, p dies at the end of T1 and loses three periods that could have been lived if she had not died prematurely. So the badness of her premature death is 3Æ, and that is attributed to her life at T1. The same thing is done to q, r, and s. As can be seen, t does not suffer a premature death, and hence does not suffer any badness of premature death. In B, p loses two periods, and hence the badness of her premature death is 2Æ. 2Æ is then divided equally amongst T1 and T2. Thus, the personal good of p at T1 is (1  Æ). The same calculation is performed for q, r, and s. In B, t does not live at all, so there is no disvalue of premature death for t. The sum of personal good in A is (8  12Æ), whereas the sum of personal good in B is (8  8Æ). If we disperse the badness of premature death in this way, B is better than A. Thus, the value of longevity can be dispersed successfully without giving up Broome’s claim that life extension is better than replacement. I think that we should not reject separability of time on the basis of longevity.5 5 Look at Table 4.1 in Section 3. It may be claimed that if my argument is true, then it violates separability because it must be judged that A is better than B, and that D is better than C. However, this is incorrect. Separability of time still holds because it applies to temporal personal good, as distinguished from temporal wellbeing. In this case, a person’s temporal personal good is determined by her temporal wellbeing and the badness of premature death. Thus, separability of time in terms of wellbeing is false, whereas separability of time in terms of personal good is true.

SEPARABILITY

69

5 Temporal Good Makes Sense? I argued that Broome’s objection to separability of time on the basis of the value of longevity is unsuccessful. However, this does not mean that I support separability of time. Nor does it mean that Broome must support separability of time. Actually, I believe that separability of time is implausible. I also believe that separability of time does not fit in Broome’s general analytical framework. Let me explain. One of the salient features of Weighing Goods is that Broome’s analysis concerns the structure of good, not the content of good. Broome does not attempt to identify that which comprises good. In other words, he does not attempt to elucidate “good existing in advance of ethical thinking, which ethical thinking takes as its object but does not itself determine” (Broome 1991: 17). Broome states: Certainly, teleology aims at good. But this need not mean it takes good as an externally determined objective. The pursuit of good may give to ethics, not an objective, but a structure. It may fix the way in which ethical considerations work and how they combine together. It may provide, not a foundation or an objective, but an organizing principle for ethics. (Broome 1991: 17)

Broome intentionally leaves the account of good open and calls it personal good. He then characterizes personal good as a representation of the betterness relation of the form that x is at least as good for the person as y. A person’s personal good is thus everything that is good for her. It does not commit to a particular account of good or the claim that the overall good increases when the amount of an externally determined good increases. The notion of personal good, which is defined by a person’s betterness relation, enables us to skip the whole discussion about the quality and quantity of particular account of good. Before Broome, we had to go through at least two steps of analysis in order to examine the structure of distributive principles. First, we were required to establish the most plausible account of good. Second, we were required to establish the measurability and interpersonal comparability of good. Conventionally, unless we settle these matters, we were not allowed to discuss distributive principles. However, it is very difficult to determine how the quality and quantity of good are balanced. For example, drinking a pint of beer or two increases my wellbeing. But if I drink twenty pints, I become sick. It is not the case that the more beer I have, the better off I am. Whether beer increases my wellbeing depends on not only the property of beer itself, but also the quantity of beer. If we follow the conventional way of analyzing distributive principles, we need to establish a complex relation between the quality and quantity of a particular good. However, Broome’s focus on people’s betterness relation and personal good enables us to skip this problem. All it cares about is whether an alternative is, all things considered, better for the person than another alternative. This is the salient feature of Broome’s analysis that I am talking about.

70

IWAO HIROSE

Table 4.4.

Life A Life B

T1

T2

T3

1 3

2 2

3 1

I believe that longevity is like the quantity of a particular good. I do not think whether or not longevity is good can be determined in advance. Longevity may be good in some cases. It may not be good in other cases. Whether or not longevity would contribute to the good of a person’s life can be determined only when we compare different courses of her complete life. A similar thing can be said about other pattern goods within a person’s life. Many people think that an upward life is strictly better than a downward life, where the total wellbeing is constant. For example, compare two courses of life, which consist of three temporal periods. Table 4.4 shows the temporal distribution of the two courses of life. Intuitively, life A is strictly better than life B. This is because life A is upward whereas life B is downward. Thus, the shape of life is a pattern good, which does not show up in any particular period. However, the goodness of an upward life may be outweighed, all things considered. For example, if life B contained five units of wellbeing at T1, it would not be clear if life A were better than life B. The best way to compare the relative goodness of different courses of life is to weigh up the allthings-considered goodness of these courses of life. To elucidate the structure of distributive principles, all we need to know is whether a course of life is, all things considered, better for the person than another. That is, we only need each person’s betterness relation defined over states of nature, each of which describes how her entire life goes. I believe that longevity and other pattern goods within a person’s life should be part of this sort of all-things-considered judgment.6 Thus, in the end, I agree with Broome that we should reject separability of time. But I disagree with Broome’s particular argument against it. I believe that he should reject separability of time on the basis of his idea of personal good.

References Blackorby, Charles, Primont, Daniel, and Russell, R. Robert (1978) Duality, Separability, and Functional Structure: Theory and Economic Applications. New York: North-Holland. Broome, John (1991) Weighing Goods: Equality, Uncertainty and Time. Oxford: Blackwell. 6 This point echoes the claim of many broadly defined egalitarian philosophers, who take people’s complete life to be the appropriate temporal unit for distributive judgment (Dworkin 1981; Nagel 1991; Rawls 1971).

SEPARABILITY

71

Broome, John (2004) Weighing Lives. Oxford: Oxford University Press. Deaton, Angus, and Muellbauer, John (1980) Economics and Consumer Behavior. Cambridge: Cambridge University Press. Debreu, Gerard (1954) Representation of a preference ordering by a numerical function. In R. Thrall, C. Coombs, and R. Davis (eds.) Decision Processes. New York: Wiley, 159–65. Debreu, Gerard (1960) Topological methods in cardinal utility theory. In Kenneth J. Arrow, Samuel Karlin and Patrick Suppes (eds.) Mathematical Methods in the Social Sciences. Palo Alto, CA: Stanford University Press, 16–26. Dworkin, Ronald (1981) What is equality? Part 2: Equality of resources. Philosophy and Public Affairs 10: 283–345. Gorman, W. M. (1968) The structure of utility functions. The Review of Economic Studies 35: 367–90. Jensen, Karsten Klint (1995) Measuring the size of a benefit and its moral weight on the significance of John Broome’s “interpersonal addition theorem”. Theoria 61: 25–60. Krantz, David H., Luce, R. Duncan, Suppes, Patrick, and Tversky, Amos (1971) Foundations of Measurement, I. San Diego, CA, and London: Academic Press. Mirrlees, James A. (1982) The economic use of utilitarianism. In Amartya K. Sen and Bernard A. O. Williams (eds.) Utilitarianism and Beyond. Cambridge: Cambridge University Press, 63–84. Murray, Christopher J. L., Ezzati, Majid, Flaxman, Abraham D., Lim, Stephen, Lozano, Rafael, Michaud, Catherine, et al. (2012) GBD 2010: Design, definitions, and metrics. The Lancet 380: 2063–6. Nagel, Thomas (1991) Equality and Partiality. New York: Oxford University Press. Parfit, Derek (1984) Reasons and Persons. Oxford: Clarendon Press. Rawls, John (1971) A Theory of Justice. Cambridge, MA: Harvard University Press. Risse, Mathias (2002) Harsanyi’s “utilitarian theorem” and utilitarianism. Noûs, 36: 550–77. Schelling, Thomas C. (1986) The mind as a consuming organ. In Jon Elster (ed.) The Multiple Self. Cambridge: Cambridge University Press, 177–96. Strotz, R. H. (1955) Myopia and inconsistency in dynamic utility maximization. The Review of Economic Studies 23: 165–80. Wakker, Peter P. (1988) Additive Representations of Preferences: A New Foundation of Decision Analysis. Dordrecht, Germany: Kluwer Academic Publishers. Weymark, John A. (2005) Measurement theory and the foundations of utilitarianism. Social Choice and Welfare 25: 527–55.

5 The Social Disvalue of Premature Deaths Hilary Greaves

There is more to be said for rough estimates of the precise concept than precise estimates of economically irrelevant concepts. E. J. Mishan

Many public decisions confront us with the daunting task of placing a monetary value on the badness of a premature human death.1 A given road safety project would reduce road deaths by a well-estimated number and would cost a known amount; we must decide whether it is cost-effective or not. Increasing the overall budget of the National Health Service would increase the number of lives that service could afford to prolong, but at some point we have to stop and save resources for other public policy goals. More globally, cutting greenhouse gas emissions, and/or investing in capture and storage projects, would reduce the number of lives cut short by climatechange-induced disasters by figures that can at least be guessed at; the value placed on the badness of premature deaths can be one of the factors significantly affecting recommendations for how much mitigative action is cost-effective. Many people find the task distasteful. There are two ways of attempting to give intellectual content to this distaste. The first is to assert that human life has a value that is lexically prior to that of mere worldly goods—that is, that given any such choice, the option that saves more human lives is always better, no matter what the cost. The second is to assert that human life is incommensurable with worldly goods—that is, that given a choice between (say) reducing the number of road deaths by three and having an extra £4 million to spend on other public services, neither option is better or worse than the other, and it is also not the case that the two options are equally good: they literally cannot be placed in any common ranking. But in the context of public decision making, the ‘lexical priority’ claim leads to absurd results, 1

I am grateful to John Broome, William MacAskill, and the editors of this volume for comments on earlier drafts of this chapter.

THE SOCIAL DISVALUE OF PREMATURE DEATHS

73

while the ‘incommensurability’ claim is mere refusal to face up to the issue. Rational public decision-making requires that we settle on a number. What, though, should the number be, and which factors are relevant to determining it? Several techniques exist for estimating the monetary value of a life, but none is satisfactory. Early approaches were other-regarding: that is, they focussed on the importance of the lost life to those who remain living. These approaches were roundly criticized in the 1970s, and have since largely been displaced by self-regarding approaches, that is, approaches whose primary focus is the value of the lost life to the person whose death is in question. John Broome apparently agrees that this move from other- to self-regarding estimates is a step in the right direction, but has been particularly vocal in criticizing the ‘willingness-to-pay’-based (WTP) approach that is dominant within the latter tradition. In recent work, Broome sketches a (selfregarding) alternative, based on a variant of the ‘quality-adjusted life year’ (QALY) concept well known to health economists, and intended to keep the merits of the WTP approach while avoiding its drawbacks. The central point of this chapter is that neither an exclusively other-regarding nor an exclusively self-regarding approach has much hope of being even approximately accurate. Here as elsewhere, the value of an event is determined by its effects on everybody, not only by its effects on the person who is intuitively the subject of that event. Here, as elsewhere, the sums of very small effects on very large numbers of other people can be comparable in magnitude to the sums of large effects on the small number of people most obviously affected by the event in question. These observations would be mere pedantry if in fact either self- or other-regarding effects dominated the calculation. But I do not think we can be at all confident that that is the case (in either direction). In particular, when we try to square our thinking about the badness of premature deaths with our thinking about overpopulation, we must at least take very seriously the possibility that other-regarding and self-regarding effects are indeed comparable in magnitude. It follows that neither can safely be ignored. An unfortunate consequence is that the correct calculation is highly complex, and that, therefore, we are far more uncertain about the appropriate monetary figure to place on the disvalue of a premature death than any of the current literature acknowledges. This is a formidable problem for practical decision-making (in particular, in the context of climate change). But we must not pretend that the keys are under the lamppost. The structure of this chapter is as follows. Section 1 briefly reviews the extant approaches to determining the monetary disvalue of a premature death. Section 2 is an exposition of Broome’s critiques of the WTP approach, and of his modifiedQALY suggestion. Section 3 attempts a relatively complete list of the factors that contribute to the badness of a premature death, and comments on which factors are taken into account by which extant approaches. Section 4 contains the main point of the chapter: I argue that plausible thinking about overpopulation entails that neither other- nor self-regarding aspects of the disvalue-of-death calculation are negligible.

74

HILARY GREAVES

Section 5 summarizes, and briefly considers how we might proceed in practice in the light of the issues raised in this chapter.

1 Extant Approaches to Valuing Life The primary other-regarding approaches are based on some estimate of a person’s output. The gross output approach takes the disvalue of a premature death to consist in the loss of the contributions that the deceased would have made to society had he lived out the remainder of his natural life. Average discounted lifetime earnings are used as a crude estimate of this quantity, the rationale being that this figure is ‘what the society is willing to pay the individual (and thus the person’s ‘marginal product’)’ (Cline 1992: 117). A (dramatically different) variant on the gross output approach is the net output approach. Here, the monetary disvalue of a premature death is taken to be given by average lifetime earnings less average lifetime consumption. The gross output approach dominated until the 1970s.2 A curious feature of any other-regarding approach, however, is that it altogether ignores the victim’s interest in living longer. In part due to misgivings about this omission, and in part due to concerns about the shakiness of the above methods’ theoretical foundations (Mishan 1971), they have since largely been displaced by the WTP and (occasionally) hedonic pricing approaches. Hedonic pricing techniques attempt to infer the value people place on a non-market good from their revealed willingness to pay for a suitably related market good. For example, householders’ implicit valuations of clean air can be derived from the difference in house prices between areas that have different levels of air pollution, but are otherwise similar. In the present case, the idea is to derive implicit value-of-life figures from wage differentials: the extra amount a worker must be paid to render him willing to take on a riskier job tells us something about the value he places on his life. Economists generally regard revealed-preference techniques as superior in principle to stated-preference techniques, on the grounds that what people do is a more reliable guide to their true preferences than what they say they would do in as yet unencountered situations. However, revealed-preference techniques often face significant methodological problems. In the example just given, for instance, it has been objected that labour market imperfections, such as unionization, prevent wage differentials from revealing workers’ real valuations of the risk of death. This brings 2 It appears still to dominate in the developing world. A relatively recent report of the Asian Development Bank (2003: 2) remarks, intriguingly, that

Since the late 1980s willingness to pay has increasingly been applied for accident costing in industrialized countries. However, the gross output method has been the most commonly used method in most countries over the past few decades. This method is recommended for developing countries as its primary objective, increasing a country’s wealth, is thought more appropriate to their needs. However, the report does not elaborate further.

THE SOCIAL DISVALUE OF PREMATURE DEATHS

75

us to the currently dominant stated-preference approach to valuing life, by WTP surveys. In such a survey, respondents report the maximum amount they would be willing to pay (or the minimum amount they would be willing to accept) to have a given project go ahead. The total WTP, summed over all affected parties, is then taken to be the monetary value of the project in question. This technique is widely applied to, for example, the valuation of environmental amenities (‘How much would you be willing to pay to avoid having the size of your local nature reserve decreased by 30 per cent?’) There is an obvious problem in applying this technique to the case of human life, that is, the case in which the ‘project’ is causing or averting the death of the respondent: in that case the relevant willingness-to-accept (WTA) figure (at any rate) may well be infinite. The theorist is then threatened with the conclusion that the ‘value of a life’ for the purposes of public-policy analysis is also infinite. But accepting an infinite ‘value of a life’ would amount to granting life lexical priority over ordinary goods; we have already noted that such lexical-priority claims are implausible. The standard response to this is to drop talk of the ‘value of a life’ altogether, replacing it with talk of the value of decreasing risk to life, since WTP/WTA figures for the latter are finite and well-defined. Dividing these figures by the change in risk involved, one obtains the so-called ‘value of a statistical life’ (VOSL). One is then supposed to use the resulting VOSL figures in place of a monetary value for an actual life, bearing in mind that the latter concept is strictly meaningless.

2 Broome on the Disvalue of a Premature Death 2.1 Against willingness to pay Broome has trenchantly criticized the application of WTP methodology to the issue of the value of life. There are essentially two independent issues. The first concerns the consequences of applying the WTP approach in the particular way it is in fact (but need not be) normally practised; the second concerns an unavoidable problem with applying the WTP methodology to the issue of the respondent’s death. The first point arises if the WTP-theorist declines to weight WTP/WTA data by respondents’ marginal utilities of money. This leads to the VOSL approach outlined above. Against this approach, Broome (i) objects to the absence of marginal-utility weightings in cost-benefit analysis in general (e.g. 2004: 262); (ii) points out that the above approach has the implausible conclusion that whether or not we happen to know who will be killed by some project makes a vast difference to the project’s balance of benefits over costs, holding fixed both the facts about who will be killed and our knowledge of the numbers of people killed (1978); and (iii) emphasizes the radical implausibility of holding that it is only risk to life, and not actual life, that has value (2012b).

76

HILARY GREAVES

The need to resort to talk of the ‘value of a statistical life’ is avoided by the alternative approach, which thinks of the ‘costs’ and ‘benefits’ of cost–benefit analysis in terms of utility, rather than money. On that approach, WTP figures are weighted by marginal utilities of money before being summed in the analysis. This avoids the unwanted conclusion that the value of a life is infinite, since the marginal utility corresponding to an infinite WTA is zero (if the WTA is infinite, that is precisely because money is useless to the (selfish) agent in question when she is dead). WTP for reductions in risk of death can then be used to calculate the utility of a life, on the assumption that in determining their WTP for such changes, individuals conform to expected utility theory. This procedure is sound in principle, but in practice the latter assumption is implausible: as is well known, individuals are notoriously irrational in their dealings with risk. It is in response to this sort of problem that Broome suggests abandoning the WTP approach altogether, in favour of an alternative that seeks to elicit valuations from individuals within a framework that imposes certain rationality restrictions. A sketch of Broome’s alternative requires a brief digression into the health economists’ notion of a QALY.

2.2 Quality-adjusted life years and wellbeing-adjusted life years NHS cost-effectiveness decisions are determined by the metric known as the QALY, or ‘quality-adjusted life year’. A given state of health is indexed by a ‘health utility’, supposed to indicate the health-related quality of life enjoyed by a person in that state of health. A health utility of 1 signifies a state of perfect health, 0 death; negative numbers signify states in which the health-related quality of life is worse than being dead. At any stage of a person’s life, he has a certain number of remaining QALYs: the time-integral of his quality of life, over the duration of his remaining life (no matter that this number cannot be known with anything approaching certainty). Healthcare interventions—for example, cancer treatments and eye surgery—are assessed for the number of additional QALYs they are expected to give the patient. The scale is cardinalized in such a way that (surveys and focus group discussions suggest) a typical individual patient would be indifferent between any two futures that held the same number of QALYs for him. For example, if a state of blindness has a health utility of 0.5, a typical individual would be indifferent between 20 years of life in that state of blindness (followed by death) and 10 years of life in a state of perfect health (followed by death).3 Appropriate cardinalization is of course crucial: arbitrary

3

Other methods of cardinalization are also possible: for example, respondents may be asked instead (‘standard gamble’ questions) what risk of immediate death they would be willing to tolerate to have the condition cured, rather than (‘time tradeoff ’ questions) how many years of life they would be willing to sacrifice in order to have the condition cured. For a critical survey of cardinalization methods, see Nord (1992).

THE SOCIAL DISVALUE OF PREMATURE DEATHS

77

monotonic transformations of the scale would radically affect the cost-effectiveness judgments delivered by the QALY approach. With the cardinalization of the QALY scale thus fixed, the idea is that the overall goal of the NHS is to maximize the number of additional QALYs given to the population as a whole. Hence, given a fixed overall budget and subject to additional factors such as the quality of the available evidence for the ‘cost per QALY’ estimate, the NHS will tend to provide all and only treatments whose cost per QALY lies below a certain threshold. The actual value of this threshold is around £20,000–30,000 (National Institute for Health and Clinical Excellence, 2008: }4.2). The key for our purposes is that in principle, the QALY concept can also be used to place a monetary value on premature death. The point is that the associated ‘health utilities’ place changes between the state of being dead and various states of health on a common scale, appropriately cardinalized, with more trivial changes that are easier to value. For example, a lower bound on the value of curing a headache can be obtained from observed willingness to pay for headache-curing medications. Valuations of these more trivial changes can then be scaled up to yield the value of prolonging a (given) life.4 Broome regards this approach to determining the ‘value of a life’ as a great improvement over the WTP approach: whereas the latter takes individuals’ WTP for changes in risk at face value despite the known irrationalities in the responses of ordinary people to decisions involving risk, the former imposes far more structure from the outside (the latter approach ‘applies no theory’ (2004: 263); see also Broome (1994, 1996)). For instance, it correctly imposes the assumption that, other things being equal, it is better to avert someone’s premature death at age 40 than it is at age 80, despite the fact that an 80 year old might be willing to pay just as much for increased safety as a 40 year old. Also, the structure of its questioning techniques forces people to accept the reality of the tradeoffs between quantity and quality of life. Broome therefore regards it as a plausible way of treading the appropriate path between the dual spectres of (i) altogether ignoring the actual preferences of real people and (ii) infecting public policy analysis with the irrationalities and ill-informedness to which survey respondents may be subject. In his Weighing Lives (2004), Broome (however) criticizes the QALY-based approach to valuing life approach for taking only health into account (see also (Broome, forthcoming)). A better approach, he suggests, would adopt the same general structure, but would assign utilities to states of life more widely construed, not to states of health alone. The result would be a utility scale enabling us to quantify the badness of premature deaths in terms of the known badnesses of any of the things that affect wellbeing (so we could use, e.g., the utility jumps corresponding to salary increases to find the money equivalents), rather than only the goodness/badness of Perversely, figures for the ‘value of a statistical life’ derived from the WTP approach have also been used to estimate the monetary value of a QALY, and hence of lesser health benefits (e.g. Mason et al. 2009). 4

78

HILARY GREAVES

changes in health state (and corresponding WTP for medicines). We might call this the WALY approach (‘wellbeing-adjusted life years’). In fact, for the purposes of estimating the self-regarding disvalue of a premature death,5 I do not think that there is anything fundamentally wrong with the unmodified QALY approach. To place a monetary value on the badness of premature deaths, we require an appropriately cardinalized scale that enables us to weigh death–life transitions against some other transitions that are easier to value. But in principle, we only need one easier-to-value transition; and a health-related one (curing a headache) is as good as any other. In practice, of course, robustness is served by applying the QALY-style survey methodology to a larger number of transitions. For the purposes of this work, in any case, the differences between the QALY and WALY approaches are minor.

3 Social Disvalue Let us allow that the QALY/WALY approach constitutes a reasonable methodology for assessing the disvalue of a premature death to the person who dies—the ‘selfregarding’ disvalue of that death. It does not follow that a figure obtained from QALY/WALY surveys is a reasonable figure to use as the ‘monetary value of a life’ in our public policy decisions. The basic point is familiar. In assessing the social value of any event, we need to consider its effects, not only on those most directly and obviously affected, but on everyone. The parties to a transaction are not always those most affected by it, and even if they are, its small effects on large numbers of people may add up to a larger overall contribution to social value than its large effects on the most directly affected parties. Granted that they may, do they in the present case? To begin addressing this question, we must identify the factors that contribute to the total social disvalue of a premature death. In doing so, it will prove useful to separate deprivation factors from transition factors. By (stipulative) definition, deprivation factors are those aspects of disvalue that are the result purely of the absence of the later part of a person’s life; transition factors are those resulting from the transition from the existence to the non-existence of the person in question. The difference is that deprivation factors but not transition factors would still be present if the person in question had never been born at all. In the following list of the components (positive and negative) of the value of a premature death, deprivation factors are labelled ‘D’, transition factors ‘T’:

5 My purpose in this chapter is narrower than Broome’s purpose in his Weighing Lives (2004); he discusses a comprehensive theory of wellbeing, whereas I am only discussing the monetary disvalue of a premature death. I agree that WALYs rather than QALYs are required for the former purpose. I am not sure whether or not Broome thinks they are also required for the latter purpose.

THE SOCIAL DISVALUE OF PREMATURE DEATHS

79

D1 The loss of the benefit that the dead person would himself have derived from his continued life. X1 The pain (physical and/or psychological) that the dying person experiences during the actual process of dying, as compared to the scenario in which the person ceases to exist at the same time but undergoes no pain beforehand.6 T1 Increased fear of death and/or fear of certain activities, and the associated changes of lifestyle, in those who remain living. For example, if Joe dies paragliding, this may cause other people who would otherwise have paraglided to refrain from doing so, thus missing out on enjoyment and life-enriching adventure; if Joe is murdered on the street, this may cause others generally to hide indoors more. D2 The positive effect of the death due to the freeing-up of the resources (food, water, physical space, etc.) that the deceased would have consumed in his remaining time, had he lived a life of normal duration. T2 The negative emotional impact on family associated specifically with bereavement (not with the foregone pleasure in the deceased’s company, for which latter see (T4)). T3 The negative financial and logistical impact associated with the fact that the deceased’s family had organized their affairs on the assumption that the deceased would live significantly longer. This includes, for example, the effects on wellbeing of the family having adjusted to a lifestyle that is no longer sustainable in the absence of a deceased ‘breadwinner’.7 T4 The negative impact on partners due to relationship loss. The key idea here is that, on average, a person in a voluntary long-term partnership is better off (both financially and emotionally) than that same person would have been if single. The effect is in practice likely to be almost entirely transitional: while bereavement is likely to be followed by at least some period of single status, if the deceased had never existed in the first place, most partners would have formed a stable relationship with some other partner in the first place, at roughly the same time as the formation of their actual relationship. D3 The economic contribution that the deceased would have made to society via the remainder of his working life. T5 Economic losses associated with the disruption involved in the employer’s having to replace the deceased with a new employee (e.g. duplicate recruitment and training costs). This list is not intended to be exhaustive.8 But having a relatively complete list on the table throws the incompleteness of each of the extant approaches into sharp relief. 6 This is not really an aspect of the badness of death at all: it could have occurred without the death, and would have been equally bad if succeeded by survival rather than death of the person in question. 7 The family is of course also adversely affected by the loss of the deceased’s income, quite aside from issues of adjustment. However, this aspect of disvalue is accounted for in factor (D3) below. 8 In particular, it omits the knock-on effects due to the fact that averting the premature death of a young person often has the consequence that additional people—the children, grandchildren, etc. of the ‘saved’ person—are brought into existence. These knock-on effects can be net positive or negative—that depends on whether the world is currently under- or over-populated—but in either case, they are potentially very large in magnitude, and may well actually dominate the calculation.

80

HILARY GREAVES

The ‘gross output’ approach (surveyed briefly in Section 1) appears to be an attempt to estimate factor D3. The ‘net output’ approach appears to be an attempt to estimate the algebraic sum of D3 and D2. Which factors the WTP approach includes obviously depends on which factors respondents are taking account when determining their responses, but it seems likely that the majority of respondents will be primarily influenced by factors D1, T2, T3 and D3. (Meanwhile, if the ‘value of a statistical life’ rhetoric is taken at face value, it seems to suggest that only something like our factor (X1) is taken into account: the disutility of fear of death.) Similar remarks apply to the hedonic pricing approach. The QALY and WALY approaches appear to be attempts to measure factor (D1) alone.9 It would, of course, be defensible to use an estimate of (D1) as the social disvalue of a premature death if other factors were negligible in comparison to (D1). Charity suggests attributing the hypothesis that factors other than (D1) are indeed negligible by comparison to advocates of self-regarding approaches. I agree that it is a prima facie highly plausible hypothesis. However, I do not think we can be at all confident that it is correct. The purpose of the next section is to make the case for this scepticism.

4 Death and Overpopulation Many believe that the world is currently overpopulated. The thought is that the planet’s capacity to provide food, water, energy, materials, and simple living space, in addition to current artificial infrastructure, is being stretched far beyond optimum usage by the sheer number of people currently living on the planet, with detrimental consequences both for the long-term capacity of the planet to sustain life and for the quality of life of those who do live in the meantime. This is so to such an extent (the thought continues) that it would be better, all things considered, to reduce world population. Thus campaigners urge, for example, increased spending on family planning programmes, and/or voluntary reductions in family size.10 However, there is of course another way to reduce world population: via the premature deaths of some of those who already exist. But even the aforementioned campaigners do not advocate deliberate reductions in public safety on these grounds. And even those who do not think it is true that it would be good to reduce the birth rate tend to agree that it is consistent to hold that this is good while simultaneously holding that additional premature deaths would be bad. This raises the question of how to render consistent the judgment that additional births would be (on balance) bad with the judgment that additional premature deaths would also be (on balance) 9 A relatively complete approach, driven by the same concern to calculate the full social value as that driving the present work, is sketched in Arthur (1981). Thanks to John Broome for alerting me to this article. 10 Academic discussions of the case for these recommendations is thin on the ground. The campaigners’ views are propagated, for example, via the websites of World Overpopulation Awareness () and Population Matters ().

THE SOCIAL DISVALUE OF PREMATURE DEATHS

81

bad: the background theory cannot be as simple as any transition to fewer people would be an improvement, or any transition to more people would be an improvement. This question forces us to tangle with population axiology. This is in a sense unfortunate, as there is a notorious lack of consensus in population axiology. It is, however, inevitable: if we want to compare the valuation of premature death with the ethics of variable population size, we need to tangle with the ethics of variable population size.11 We cannot proceed further without making some controversial theoretical assumptions, at least for the sake of argument. I will first make the ones that I think are true, and later consider how the arguments are affected if we assume instead the axiology defended in Broome’s Weighing Lives (2004). As above, let us call the quantity captured by factor D1 the self-regarding value of a life: it is the value of the life to the person whose life it is. Define the social value of a given person’s life (in a given context) to be the difference in value to society as a whole (or: value simpliciter) between the state of affairs in which that life is lived and a state of affairs in which the person in question never exists but the wellbeing levels of all other people are (somehow) unchanged. Our controversial theoretical assumption is that the social value of a given life is equal to the self-regarding value of that life. Suppose further, merely for illustrative purposes, that the world is currently at optimum population, in the following sense: on average, the difference in value between the actual state of affairs and a state of affairs in which a randomly selected actual person does not exist and others’ lives go as they would have if the person in question had not existed is zero. It follows that the (positive) self-regarding value of an average life is exactly balanced by the net negative effect of the average person’s existence on other people. But the latter quantity is the sum of other-regarding deprivation factors. Thus, under our current assumption, if the world is currently anywhere near optimum population then other-regarding effects are certainly not negligible. It also follows that, under the same conditions, the social disvalue of a premature death is the (negative of the) sum of transition factors alone; note that none of the extant approaches accords these effects any significant role. If the world is currently overpopulated, of course, then our conclusion about the significance of otherregarding effects, including in particular those relating to the creation of additional people, is only exacerbated. (We will take up the possibility of underpopulation shortly.) Our assumption that the self-regarding value of a life is equal to its social value may sound innocuous: if social value is merely an aggregation of values to individual people, and if no other people are affected, what else can the value of a change that affects one person reasonably be? I think this is in fact correct. It is, however, a

11

In fact the two topics are related independently of such explicit comparisons, since, as noted above (see note 7), premature deaths of people who are not yet past reproductive age affect the future population. This point has been noted many times, not least by Broome (1985). My point in the present section concerns instead the additional connection that remains even if those who die prematurely would not otherwise have gone on to have (additional) children.

82

HILARY GREAVES

controversial piece of population axiology.12 The axiology that satisfies our assumption is total utilitarianism, according to which the value of a state of affairs is given by X VTU ðXÞ ¼ ui ðXÞ ð1Þ i2PðXÞ

(Here, ui(X) is the lifetime wellbeing13 of person i in state of affairs X, and P(X) is the set of people who exist in X.) Our assumption is, however, denied by every other axiology. This is not the place for a comprehensive survey of the alternatives, but let us consider in particular the ‘critical level utilitarianism’ that is advocated by Blackorby, Bossert, and Donaldson (Blackorby and Donaldson 1984; Blackorby et al. 1995; Blackorby et al. 2005), and to which Broome gives cautious approval in Weighing Lives (2004). According to critical level utilitarianism, the value of a state of affairs is given by X VCLU ðXÞ ¼ ðui ðXÞ  aÞ ð2Þ i2PðXÞ

where Æ is some constant (generally assumed to be non-negative). According to critical level utilitarianism, therefore, the social value of an additional life (holding the welfare of existing people equal) is given by the self-regarding value of that life minus Æ. (The constant Æ is variously referred to as the ‘critical level’ (Blackorby et al. 2005) or the ‘neutral level for existence’ (Broome 2004)). Total utilitarianism and critical level utilitarianism are structurally very similar (indeed, the former is a special case of the latter). Accordingly, if critical level utilitarianism is adopted, our reasoning proceeds much as above, with a simple modification. We obtain instead (at optimum population) that the self-regarding value of an average life exceeds the net negative other-regarding effect of that life by a quantity Æ, and that the difference in disvalue between a non-birth and a death is given by the sum of the transition factors plus the critical level Æ. However, advocates of critical level utilitarianism agree that the critical level cannot be too high (on pain of making the ‘negative repugnant conclusion’ too repugnant; see e.g. Broome (2004: 213, 264)). It therefore seems likely that our main message, that other-regarding

12 It is also somewhat controversial even aside from issues of variable population. For example, a value function that applies a concave ‘prioritarian’ transformation to individual’s von Neumann-Morgenstern utilities notoriously incorporates a divergence between what is good for individuals and what is good simpliciter. I have argued elsewhere (Greaves, n.d.) that this is not the right way of formalising the prioritarian intuition. Be that as it may, the issues of population axiology on the one hand, and of which is the correct value function even for cases of fixed population, are essentially orthogonal. My focus here is on the former, and I presuppose a utilitarian solution to the latter for simplicity of exposition. 13 I use the term ‘lifetime wellbeing’ in the same sense as does Broome (2004). Lifetime wellbeing is a measure of how well a person’s life, as a whole, goes for him. It is taken to lie on a ratio scale. There is some subtlety over how to define the zero point. Broome defines it as a life lived throughout in such a state such that, at any moment, death at that moment would be exactly as good as continued life for the person in question. I have no objection to this definition.

THE SOCIAL DISVALUE OF PREMATURE DEATHS

83

effects and transition effects are far more significant to the ‘value of a life’ than the dominant approach recognizes—unless the world is significantly underpopulated— would survive the replacement of total utilitarianism with critical level utilitarianism. We now take up this postponed issue of the possibility of underpopulation. We have noted that if the world is currently near or above its optimum population and if in addition premature deaths are bad, then other-regarding effects in general, and transition effects in particular, form (at least) a very significant part of the calculation of the (dis)value of a premature death. But it is worth noting well that any critical level utilitarianism, including total utilitarianism, will deem the world to be underpopulated in many cases in which untutored intuition would deem the world to be overpopulated. Intuitive thinking notes only that average welfare would be increased, or that some extant problems would be solved, by preventing some extra births; our theories recognize this, but recognize in addition that the lives thereby prevented are lives that would themselves have had positive value. We therefore do not conclude that other-regarding and transition effects do dominate the calculation of the disvalue of a premature death. An alternative way of resolving our tension is, guided by theory, to ditch the intuition that the world is currently overpopulated. If the world is underpopulated, premature deaths are indeed bad, for the simple reason that they reduce instantaneous population (an effect that can be entirely accounted for by factor (D1)), even if the deaths in question are of people who would not otherwise have gone on to have any additional children. The point I wish to press is that this— the underpopulation hypothesis—is the only way of reconciling the hypothesis that (D1) dominates the badness of a premature death with the above population axiologies discussed above. Those axiologies, of course, are not uncontroversial; nothing in population axiology is uncontroversial.14 In particular, the tension between a high disvalue for premature deaths and a low disvalue for non-births could easily be avoided if the principle that Broome dubs ‘the neutrality intuition’ were true. This principle states that there is no value to adding an extra person to the population, except insofar as the addition modifies the wellbeing levels of those whose existence is held fixed. (In other words, the self-regarding value of extra lives has no ‘social’ value.) Against the backdrop of a population-axiological view incorporating this ‘neutrality’ principle, of course it is perfectly consistent to hold that the world is overpopulated—it would be bad to add an extra person—and simultaneously to hold that it is worth a very high price to preserve existing lives, on the grounds that the (positive) self-regarding value of the average life by far outweighs the (negative, and significant but relatively modest) net other-regarding value of that life. The intuitive force of the ‘neutrality’ principle is presumably responsible for the widespread failure to take seriously the 14 I have not had space to do justice here to anything like the full range of possibilities, their relative merits and their respective implications for the present issue. The best account of the state of play in population axiology is Arrhenius (forthcoming).

84

HILARY GREAVES

possibility that other-regarding effects might be comparable in magnitude to the selfregarding disvalue of a premature death. As Broome has repeatedly emphasized, however (2004, 2005, 2012a), that principle, however intuitive, is extremely hard to defend. A related issue is: What is the functional form of the badness of a death as a function of age at death? Any sensible exclusively self-regarding approach has the consequence that the badness of a death is a decreasing function of age of death: the earlier a person dies, the more that person loses by her premature death. This consequence is very plausible when the ages in question are more than (say) 25. But it is at least arguable that a death at age 1 year is worse than a neonatal death, and that a death at age 15 is worse than a death at age 1 year. Approaches that treat otherregarding effects as significant contributors to the overall social disvalue of premature deaths can accommodate these putative facts: if a child has survived to age 15 (for example), he has consumed large amounts of resources (via his sustenance and education) that he has not yet had time to pay back via working life, and the death of a 15 year old will generally cause far more parental grief than the death of a newborn or a 1 year old. This is a further reason for thinking that other-regarding effects are non-negligible.

5 Conclusions Pre-theoretically, it is intuitively highly plausible that the effect of a premature death on the person who dies dominates the calculation of the badness of that death. If that were the case, the WALY approach suggested by Broome would supply a viable method for working out, not only the self-regarding disvalue of a premature death, but also its social disvalue, i.e. its overall disvalue for the purpose of public policy analysis. However, unless the world is currently significantly underpopulated, that pre-theoretically plausible assumption does not remain plausible in the light of either the ‘total utilitarian’ axiology, or the ‘critical level utilitarianism’ that Broome has defended. If those axiologies are even serious contenders, then given approximately optimum population or overpopulation, we are forced to take seriously the whole range of effects D1–D3 and T1–T5. There are then two ways of proceeding. The first is directly to take on the task of quantifying each effect D1–D3, T1–T5 on a common scale, and summing them to obtain a figure for the monetary disvalue of a premature death. The slight advantage of this direct approach is that it enables us at least to avoid making any populationaxiological (and therefore particularly controversial) assumptions in the special case in which future population size is independent of the timing of the death in question. Its disadvantage is that, barring extremely careful methodology, errors in the estimation of each factor are likely to swamp the calculation. The second, alternative, approach takes a detour via population axiology: we first make a judgement of which population axiology is correct and (in the light of that axiology) the value or

THE SOCIAL DISVALUE OF PREMATURE DEATHS

85

disvalue of an extra birth, and we then modify that figure (e.g. by the sum of transition effects) to obtain a figure for the disvalue of a premature death. Once the first step has been completed, this calculation is relatively straightforward. This may well be a superior approach, since the first step must eventually be taken in any case. But there is no escaping the complexity: this second approach simply sweeps that complexity into the independent task of calculating the (dis)value of an additional birth. Given this complexity, it is hard to escape the conclusion that although precise calculations are possible in principle, in practice we are more likely to arrive at a sensible figure for the ‘monetary value of a life’ by sensible judgment than by precise calculation. I will not attempt to venture a guess at a plausible figure. But if the world is currently optimally- or overpopulated, it seems likely that the ‘value of a life’ figure that the arguments of this chapter would sanction is significantly lower than the figures normally used. In that scenario, we have seen, the value of a life to the person living it tends to be cancelled or outweighed by the negative effects of that person’s resource consumption on others; any remaining ‘disvalue of a premature death’ may come only from transition effects; it is reasonable to expect these to be significantly lower. It is of course innocuous to focus on calculations of the self-regarding disvalue of a premature death as one input into a sensible estimate of its social disvalue, always remembering that the result will need to be combined with other quantities before being put to use in practice. The literature suggests, however, that this is invariably not remembered in practice. In that case my conclusions are revisionary, not only of current practice in public policy analysis, but also of the practice that would ensue if, as Broome has suggested, WTP-based valuations of life were replaced with QALY-/ WALY-based calculations. I will take the case of climate change as an illustrative example. It is often remarked—by Broome among others15—that the deaths caused by climate change seem likely to account for a very significant fraction of the overall badness of that phenomenon. But the calculations backing up that claim involve using figures that are actually estimates of the self-regarding disvalue of a premature death as if they were estimates of the social disvalue of a premature death.16 Absent a case for thinking not only that the estimates in question are at least roughly accurate qua estimates of self-regarding disvalue, but also that other-regarding contributions to overall (dis)value are negligible in comparison, we have precisely no reason for 15

See Broome (2000: 956; 2012a: 156). The figures Broome cites in his (2000), cited also by the IPCC’s Second Assessment Report (1996: 195), are from Fankhauser (1995). Fankhauser estimates total climate damages at US$269.6 billion, of which $49 billion, hence 18 per cent, is due to mortality damages. The figure of $49 billion is based on assuming ‘value of a life’ figures of $1.5 million for developed countries, $300,000 for middle-income countries and $100,000 for lower-income countries, these figures in turn being justified by the literature on WTP estimates. 16

86

HILARY GREAVES

drawing any such conclusions. Of course the killing it does may be one of the worst things about climate change, but only in the sense that absent some credible analysis to guide our beliefs, anything may be the case.

References Arrhenius, G. (forthcoming) Population Ethics: The Challenge of Future Generations. Arthur, W. B. (1981) The economics of risks to life. The American Economic Review 71(1): 54–64. Asian Development Bank (2003) Road Safety Guidelines for the Asian and Pacific Region. Manila, Philippines: ADB Publishing. Blackorby, C., Bossert, W., and Donaldson, D. (1995) Intertemporal population ethics: Critical-level utilitarian principles. Econometrica 63(6): 1303–20. Blackorby, C., Bossert, W., and Donaldson, D. (2005) Population Issues in Social Choice Theory, Welfare Economics, and Ethics. New York: Cambridge University Press. Blackorby, C., and Donaldson, D. (1984) Social criteria for evaluating population change. Journal of Public Economics 25: 13–33. Broome, J. (forthcoming) Measuring the burden of disease. In D. Wikler and C. J. L. Murray (eds.) Fairness and Goodness: Ethical Issues in Healthcare Resource Allocation. World Health Organisation. Broome, J. (1978) Trying to value a life. Journal of Public Economics 9: 91–100. Broome, J. (1985) The economic value of a life. Economica 52(207): 281–94. Broome, J. (1994) Structured and unstructured valuation. Analyse and Kritik 16: 121–32. Broome, J. (1996) The value of life and the value of population. Journal of Population Economics 9: 3–18. Broome, J. (2000) Cost–benefit analysis and population. Journal of Legal Studies 29: 953–70. Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press. Broome, J. (2005) Should we value population? Journal of Political Philosophy 13: 399–413. Broome, J. (2012a) Climate matters. New York: W. W. Norton & Co. Broome, J. (2012b) The Philosophy and Economics of the Environment. (Unpublished lectures delivered at the University of Oxford, Trinity, 2012.) Cline, W. R. (1992) The Economics of Global Warming. Washington, DC: Institute for International Economics. Fankhauser, S. (1995) Valuing Climate Change: The Economics of the Greenhouse. London: Earthscan. Greaves, H. (forthcoming) Antiprioritarianism. Forthcoming in Utilitas. IPCC (1996) Climate Change 1995: Economic and Social Dimensions of Climate Change. Cambridge; New York: Cambridge University Press. Mason, H., Jones-Lee, M., and Donaldson, C. (2009) Modelling the monetary value of a QALY: A new approach based on UK data. Health Economics 18: 933–50. Mishan, E. (1971) Evaluation of life and limb: A theoretical approach. Journal of Political Economy 79: 687–705. National Institute for Health and Clinical Excellence (2008) Social Value Judgments: Principles for the Development of NICE Guidance. 2nd edition. London: NICE. Nord, E. (1992) Methods for quality adjustment of life years. Social Science and Medicine 34: 559–69.

6 Being and Wellbeing Krister Bykvist

Morality is in part concerned with what is better or worse for people.1 Other things being equal, it is better to make things better for people, and it is worse to make things worse for them. However, it is unclear how we are to put this person-affecting idea of morality to use in cases where the identity of people is at stake, that is, cases where, depending on what we decide to do, different people will come to exist in the future. In order to apply this idea to these cases, we need to be able to say whether coming into existence can be better or worse for people. But can it really be better for a person to exist than not to exist? John Broome thinks not, and his views on this question, and more generally on the problems of population ethics, have been extremely influential, not just among philosophers but also among economists and policy-makers. In this short chapter, I shall present Broome’s original argument, show that it is not fully convincing as it stands, and then argue that there is a better argument in the vicinity that is more convincing.2 Here is how Broome stated his argument: [I]t cannot ever be true that it is better for a person that she lives than that she should never have lived at all. If it were better for a person that she lives than that she should never have lived at all, then if she had never lived at all, that would have been worse for her than if she had lived. But if she had never lived at all, there would be no her for it to be worse for, so it could not have been worse for her.3

One natural way to spell it out is this: 1. If your existence is better for you than your non-existence, then your nonexistence is worse for you than your existence. 2. If your non-existence is worse for you than your existence, then your nonexistence would be worse for you if you did not exist. 1 I would like to thank Erik Carlson and Anandi Hattiangadi for valuable comments on an earlier draft of this chapter. 2 I endorsed an argument similar to Broome’s in a prior article (Bykvist 2007). So I am in effect arguing against my earlier self in this chapter. 3 Broome 1999: 168.

88

KRISTER BYKVIST

3. It is not the case that your non-existence would be worse for you if you did not exist. So 4. It is not the case that your existence is better for you than your non-existence. Here, (1) seems innocent since it just follows from the fact that ‘worse’ is the converse of ‘better’. Also, (3) seems plausible if we accept that ‘wellbeing entails being’: x is better for S than y only if x, y, and S exist. Hence, if S does not exist, there is a missing relatum, and it cannot be true that x is better for S than y. It is true that some have objected to this idea, but this is not the place to engage with their arguments.4 That leaves us with (2). Why should we accept it? One could claim it is an instance of the following general principle: Counterfactual Support If p is better for you than q, then p would be better for you than q, if p obtained.5 Arrhenius and Rabinowicz (2014: 428) think this principle is the real culprit in Broome’s argument. They point out that it is not logically true that if a relation R holds over states of affairs p, q, and person S, then R would hold over p, q, and S even if p were to obtain. For example, it does not hold if the relation is preference and the states of affairs are your non-existence and your existence and the person is you. However, Counterfactual Support need not be defended in this way. One could claim that this principle holds for value relations. Why? One idea is that it holds because it follows from the following more general principle: Axiological Invariance If p has absolute or comparative value for a person, it would have this value, no matter whether p were to obtain or not. This is not a good way of defending Counterfactual Support, however, because it would rule out that existence can be bad, good, or neutral for a person. If S does not exist, then it is not true to say that her existence is good for S, only that her existence would be good for S, if she existed. But this is then a violation of Axiological Invariance. Another idea would be to say that ‘x is better for S than y’ entails ‘S would be better off in x than in y’, and to be better off in x than in y requires that S exists both in x and in y. But this may seem question-begging, since the opponents could just say that this entailment does not hold in general, but only when S exists in both x and y.6

4

For an interesting argument of this kind, see Fleurbaey and Voorhoeve (this volume, Chapter 7). In Rabinowicz and Arrhenius (2014), this principle (strengthened to a biconditional) is called ‘Subjunctive Connection 1’. In Bykvist (2007), it is called ‘Accessibility’. 6 Adler (2009) makes a similar point. 5

BEING AND WELLBEING

89

So, we are stuck. It is time to take a step back. I think we can avoid the issue of whether Counterfactual Support is true, for there seems to be a better argument in the vicinity of Broome’s own argument that does not rely on this principle. Instead of being concerned with whether a state of affairs that is better for you would be better for you if it obtained, we should ask whether it could make things better for you. To spell this out, it is useful to first consider the opponents’ view. Their positive view is that the better-for relation is defined over a set of abstract states of affairs that can exist without obtaining. That is why they think it is perfectly coherent to say that my existence is better for me than my non-existence when I exist, since my existence and my non-existence are both states of affairs that co-exist with me, even though only the former state of affairs obtains. But what exactly do we mean when we say that abstract entities such as states of affairs have value? In order to answer this question, it is useful to think about what it is for properties such as bravery and courage to have value. If courage is good, then it is possible for people to be good (at least to some extent) in virtue of exemplifying courage. If bravery is good, then it is possible for people to be good (at least to some extent), in virtue of exemplifying bravery. The same holds for all other virtue properties. The slogan is: a good character trait could ‘rate a person a plus’. As Danielsson, drawing on Chisholm and Sosa, has suggested, this can be generalized to the value of states of affairs.7 If a state of affairs p is good, then it is possible for something (‘the universe’ as Chisholm and Sosa think of it) to be good (at least to some extent) in virtue of exemplifying p. In slogan form: a good state of affairs could ‘rate the universe a plus’.8 So, on this understanding of the values of abstract entities, if a property or state of affairs has value, then it is a possible value-maker of something. Applied to wellbeing, this means that if a state of affairs p is good for you, then it is possible for something to be good for you (at least to some extent) in virtue of exemplifying p. (Mutatis mutandis for ‘bad for’ and ‘neutral for’.) The slogan is: states of affairs that are good for you could rate the universe a plus for you. They are possible good-for-you makers. Note that the account does not say that when a state of affairs or property has a certain degree of value it must always make the things that exemplify the state of affairs or property (the exemplifiers) have this exact degree of value. It only says that good states of affairs or properties could make the relevant exemplifiers (or some parts or aspects of them) have some degree of goodness. So, we even allow for the possibility that a good state of affairs or property could make the exemplifiers bad to

7

Chisholm and Sosa 1966; Danielsson 2001. In fact, I think this idea can be generalized even further to the normative status of actions. If an abstract act-type, such as lying or killing, is wrong, then it is possible that some act-token of this type is pro tanto wrong in virtue of exemplifying the type in question. 8

90

KRISTER BYKVIST

some extent. Perhaps your feeling pleasure is good, but, when you feel pleasure because you engage in sadistic torture, the situation is made bad, not good. Furthermore, nothing is said about how the overall value of an exemplifier is calculated; we have only said that it is possible that a good state of affairs or property would make some part or aspect of an exemplifier have some degree of goodness. It is left open how this degree of goodness should be weighed against other values when calculating the overall value of the exemplifier. If we accept this understanding of the values of abstract states of affairs, we have a quick argument against the idea that existence can be better for you than nonexistence. 1. If your existence is better for you than your non-existence, then your nonexistence is either good, bad, or neutral for you. 2. If your non-existence is good for you, then it is possible that something is good (to some extent) for you in virtue of exemplifying your non-existence. 3. If your non-existence is bad for you, then it is possible that something is bad (to some extent) for you in virtue of exemplifying your non-existence. 4. If your non-existence is neutral for you, then it is possible that something is neutral for you in virtue of exemplifying your non-existence. 5. It is not possible that something is (to some extent) good, bad, or neutral for you in virtue of exemplifying your non-existence. So 6. It is not the case that your existence is better for you than your non-existence. Here, (1) follows from the plausible assumption that if x is better for you than y, then either (a) both x and y are good for you, but y is less good for you, (b) both x and y are bad for you, but x is less bad for you, (c) x is good for you and y is neutral for you, (d) x is good for you and y is bad for you, or (e) x is neutral for you and y is bad for you.9 Next, (2), (3), and (4) follow from the definition of value-for as defined over abstract states of affairs. Finally, (5) follows from the assumption that not just wellbeing but also ill-being and neutral-being each entails being, which seems plausible since, if ‘good for’ picks out a relation between people and states of affairs, then it is difficult to see why ‘bad for’ and ‘neutral for’ would not do the same. Rabinowicz and Arrhenius (2014: 432) acknowledge that one possible objection to their view is that ‘the world is not made any better by the existence of such

9 The argument is robust enough to survive some tweaking of this premise. Perhaps there are indeterminate values for people, things that are neither good, bad, nor neutral for people, but still valuable for them. Certain disjunctive states, such as your being happy or your feeling indifference, come to mind. However, even for these states of affairs it holds that it must be possible that something is good, bad, or neutral for you in virtue of exemplifying them. In particular, note that it is possible that the world (or some part thereof) would be good for you, if it exemplified your being happy or your feeling indifference in virtue of exemplifying your being happy.

BEING AND WELLBEING

91

states’.10 But they point out that it would not help to say that it is the obtaining of one state of affairs that is better for a person than the obtaining of another, since often the compared states of affairs cannot both obtain. For example, we cannot say that the obtaining of my existence is better than the obtaining of my non-existence, for the obtaining of one of these states necessarily excludes the obtaining of the other. This problem can be evaded, they think, by insisting that the very states of affairs themselves, and not just their obtaining can have value. Now, the account I have given does not assume that we compare the values of obtaining states of affairs (if the obtaining of state of affair p is supposed to be identical to the fact that p obtains).11 It only assumes that if one state of affairs p is better for a person than another q, p has some value (good, bad, neutral), such that it is possible that something is valuable (good, bad, neutral), in certain respects and to some extent, in virtue of exemplifying p. Of course, to have a complete account of comparative value I need to say something about what it means to say that something would (or could) be better for a person if it exemplified state of affairs p rather than state of affairs q, where p and q are incompatible. But this, as Rabinowicz and Arrhenius point out, is an instance of a general problem in axiology. The general problem has to be addressed whenever we consider claims about the comparative value of incompatible features. 10 Arrhenius and Rabinowicz (2014: 432). In a footnote, they thank Erik Carlson for pressing this point, but do not say exactly what his point was. In conversation, after I had written this chapter, Carlson told me that his idea was that they have to deny the natural idea that a state of affairs p is better for S than another incompatible state of affairs q just in case S’s life in the nearest possible world where p obtains is better than S’s life in the nearest possible world where q obtains. This idea is similar to mine, but there are some crucial differences. First, Carlson does not ground his idea about wellbeing comparisons in the more general idea that abstract features, such as states of affairs, are good in the sense that they are possible good-makers of things. Second, I do not take a stand on whether lives must be assigned value, I only claim that valuable states of affairs must be able to make some something valuable to some extent. Third, if p is better for me than q, then Carlson would say that my life in the closest p-world would be better than my life in the closest q-world, whereas I am saying something much weaker: that p must be either good, bad, or neutral (or indeterminate), and that p could make something good, bad, or neutral (or indeterminate) to some extent. In fact, I do not think that Carlson’s proposal is acceptable as it stands, since p can be better for me than q even though p would not make my whole life better, if it obtained. There might be some state of affairs r, which would obtain no matter whether p or q obtained, and which is such that p in combination with r is not better for me than q in combination with r. A possible case might be p = my feeling pleasure, q = my feeling displeasure, and r = my being such that I only take pleasure in harming others. In short, Carlson’s proposal does not take seriously the possibility of organic unities. 11 The obtaining of p could be understood in other ways. For example, it could be understood as a second-order state of affairs: the state of affairs of p’s obtaining. But this cannot be what Rabinowicz and Arrhenius have in mind, since such second-order states of affairs can themselves exist without obtaining, and thus p’s obtaining and q’s obtaining can co-exist even if p and q cannot co-obtain (cf. the second order propositions that p is true and that q is true can co-exist even if p and q cannot be true together). An alternative option is to identify the obtaining of p with the concrete occurrence of p. This would lead to problems if we also said that p is better for a person than q just in case the concrete occurence of p is better for S than the concrete occurence of q. For when p and q are incompatible there can only be (at most) one concrete occurrence to take into account. But there is no need to consider the world in a state in which both p and q are exemplified. What we need to compare is the world in two different states, one in which it exemplifies p and another in which it exemplifies q.

92

KRISTER BYKVIST

For example, we need to address this problem when we consider claims about the comparative value of incompatible character traits, for example, You would be a better person if you were brave rather than cowardly. Indeed, the problem has to be addressed whenever we consider cross-world comparisons such as, I would have been less fat if I had been eating healthy food rather than chips and hamburgers. Since the problem is not specific to the issue at hand but pertains to all cross-world comparatives I think we can ignore it here.12 Arrhenius and Rabinowicz allow for the possibility that there are entities that are valuable for a person even though they could not contribute any value for this person in worlds in which they exist. These values are therefore necessarily evaluatively inert for a person. If I am right, there cannot be any such values. We all agree that if a character trait cannot contribute any positive value to a person, it cannot be a valuable character trait. The same holds, I contend, for wellbeing: if a state of affairs cannot contribute any value for you, then it cannot have any value for you. In its most general formulation the idea is that if a property or state of affairs cannot contribute any value to the things that exemplify it, then it cannot have any value.

Guardian Angel Arguments At this point, the opponents could turn the tables and present the following interesting argument for the possibility of existence being better for you than non-existence. 1. If your existence is good for you, then someone who only cares about your wellbeing (your guardian angel) would prefer your existence to your nonexistence. 2. If your guardian angel prefers your existence to your non-existence, then your existence is better for you than your non-existence. So, 3. If your existence is good for you, then your existence is better for you than your non-existence.

12 One option is to reduce value comparison to one of degrees of value. Applied to character traits, this account says (very roughly) that bravery is better than cowardice just in case the degree of value a person would have if she exemplified bravery is m and the degree of value the person would have if she exemplified cowardice is n, and m > n. The idea is that degrees of value exist in all (relevant) worlds, and the claim m > n is therefore also true in all (relevant) worlds. This account is just an extension of the popular degreetheoretic account of cross-world comparisons: to say that x would be F-er if p than if q is to say (very roughly) that if p were the case, the degree of F-ness of x would be m, and if q were the case, the degree of F-ness of x would be n, and m > n.

BEING AND WELLBEING

93

I think (2) is questionable. It is true that what your guardian angel prefers is what is better for you in cases where your existence is not at stake and her preference is about states that have a defined value for you. However, from the fact that your guardian angel prefers your existence to your non-existence, it does not follow that existence is better for you. Suppose that your existence is good for you and, as I have been arguing, your non-existence is neither good, bad, nor neutral for you. What do we expect my guardian angel to prefer? My existence to my non-existence, of course, since in a choice between something that is good for me and something that lacks any value for me, she should prefer what is good for me. In general, in a choice between a good x and a y that lacks any value (is neither good, bad, nor neutral), it is reasonable to prefer x to y. So, you can care about goodness and prefer x to y without x being better than y.

Primitive Absolute Wellbeing? I have to deny the general principle that if p is good for you, then p is better for you than not-p, since I do not want to say that if existence is good for you, then existence is better for you than non-existence. One may worry that this means that I will have to accept primitive absolute wellbeing. But this is not so. For I could say that (for all possible states of affairs p): • p is indifferent in value for S iff p = S’s being F, for some F, and S’s being F has the same value for S as S’s being not-F. (Note that S’s being not-F ascribes a negative property to S, being not-F.) • p is good for S iff p is better for S than something that is indifferent in value for S. • p is bad for S iff p is worse for S than something that is indifferent in value for S. • p is neutral for S iff p has the same value for S as something that is indifferent in value for S. To have well-behaved absolute values (e.g. nothing can be both good and bad for you), I also have to assume that: • if p is indifferent in value for S and q is indifferent in value for S, then p and q have the same value for S.13

13 Johan Gustafsson (2013) has recently suggested a counterexample to this principle, and Carlson (2014) has suggested an alternative account of value that avoids the problem. I am not (yet) fully convinced that the counterexample is genuine. But, if it is, I am willing to go along with Carlson’s proposal, which, applied to wellbeing, says (very roughly) that what is indifferent for a person is that which, when added to any other valuable state, does not make a difference in value for the person (i.e. things with and without the added state are equal in value for the person). Note that Carlson’s account also reduces absolute values to comparative ones. In order to avoid problems with non-existence, I would add that what has value for S must be a state of affairs of the form S’s being F, where F can be a negative property.

94

KRISTER BYKVIST

So, if your existence is good for you, then your existence is better for you than some state of the form your being F, which is such that your being F has the same value for you as your being not-F.

Concluding Remarks While I have my doubts about Broome’s argument, I agree with him that a person’s existence cannot be better or worse for her than her non-existence. The reason why this is so is better seen if we consider what it is for an abstract state of affair to have value. An abstract state of affairs has a certain value only if it can make something have this value. Since my non-existence cannot make things have any value for me, it lacks value for me. Now, this means, of course, that I cannot rely on a personaffecting morality in all cases. Here Broome and I face similar problems. We need to somehow extend, amend, or supplement our person-affecting moral principles so that they apply to cases where the identity of people is at stake. That this is no easy task has been shown by Broome himself in his important work on population ethics.14 But that it is an extremely urgent task is beyond doubt. For example, in order to decide what we should do with climate change we need to know how much weight we should give to the fact that we may prevent the existence of a huge number of valuable future lives by not taking sufficiently radical measures.

References Adler, M. D. (2009) Future generations: A prioritarian view. The George Washington Law Review 77(5–6): 1478–520. Arrhenius, G., and Rabinowicz, W. (2014) The Value of Existence. The Oxford Handbook of Value Theory. Oxford: Oxford University Press. Broome, J. (1999) Ethics out of Economics. Cambridge: Cambridge University Press. Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press. Bykvist, K. (2007) The benefits of coming into existence. Philosophical Studies 135 (3): 335–62. Carlson, E. (2014) ‘Good’ in terms of ‘better’. Noûs, published online 3 March 2014, DOI: 10.1111/nous.12061. Chisholm, Roderick M., and Sosa, Ernest (1966) On the logic of ‘intrinsically better’. American Philosophical Quarterly 3: 244–9. Danielsson, S. (n.d.) Intrinsic value. Unpublished paper. Gustafsson, E. (2014) Neither ‘good’ in terms of ‘better’ nor ‘better’ in terms of ‘good’. Noûs 48(3): 466–73.

14

Broome (2004).

7 On the Social and Personal Value of Existence Marc Fleurbaey and Alex Voorhoeve

Introduction It is certainly better for all of us that John Broome exists, and it would have been worse for us if he had never existed.1 This alone would have given anyone with the power to decide whether John Broome would exist a good reason to ensure his existence. But supposing, as we hope is true, that Broome’s life is also good for him, in the sense that he enjoys high lifetime wellbeing, is it therefore also better for him that he exists and would it have been worse for him if he had never existed? In this chapter, we argue that these questions should be answered affirmatively. We also explore the relevance of our answers to issues in population ethics.

1 The Alleged Absurdity of the Claim that Existence is Better Than Never Existing Broome (1999, 2004) has developed an ethics of population which does not involve any comparison, from the perspective of a person’s self-interest, between her existence and her never existing. Whether it is better to bring a person into existence than not to bring her into existence is only considered in terms of social welfare (or social goodness, more generally). Broome argues that such social evaluation should take the following form. If a potential individual would have a level of wellbeing equivalent to what we shall call the ‘population-value indifference level’ of wellbeing (often referred to as the ‘critical level’), then her coming into existence with this level of wellbeing would be a matter of indifference from the perspective of social value. If she would have a level of wellbeing in excess of this level, then her existence would increase the value of the population, and so make the world a better place. If this 1

We thank Richard Bradley, Susanne Burri, and Nils Holtug for comments. Alex Voorhoeve’s work was supported by the British Arts and Humanities Research Council through grant AH/J006033/1.

96

MARC FLEURBAEY AND ALEX VOORHOEVE

individual would instead have a level of wellbeing below this level, then her existence would make the world worse. This approach therefore has no need for the notion of a life that is better for an individual (considering only her wellbeing) than her not coming into existence would be for her. Broome offers the following argument for eschewing the notion of a life that is better for the person than her never coming into existence is for her. It cannot be better for a person to exist than never to exist, because this would mean that it would be worse for her if she never came into existence, a statement that is deemed absurd because, Broome claims, there would be no one for whom the latter would be worse. As he writes: it cannot ever be true that it is better for a person that she lives than that she should never have lived at all. If it were better for a person that she lives than that she should never have lived at all, then if she had never lived at all, that would have been worse for her than if she had lived. But if she had never lived at all, there would have been no her for it to be worse for, so it could not have been worse for her. (1999: 168; emphasis in original)

As a consequence, in Broome’s view, the notion of ‘a life worth living’ cannot be used in the sense of comparing a life of a particular quality to never existing. Rather, it only refers to the value of extending life (i.e. making a given life last longer, without considering the problem of the creation of a new life), and should really be called a ‘life worth continuing’. In the literature, this argument is often decomposed into the following basic elements (where a description of who exists and at what level(s) of wellbeing is called a ‘social situation’):2 1. Social situation A is better than social situation B for a person P if and only if B is worse than A for P.3 2. If A is better than B for P, then A would be better than B for P if it obtained.4 3. a. A person who never exists has no wellbeing. b. Nothing can be better than anything else for a person who never exists.5 On the basis of these conditions, the argument runs as follows. From (1) and (2), it follows that if a particular life is better for P than never existing, then never existing would be worse for P than that life if P never came into existence. From (3), it follows that never existing could not be worse for P than anything if P never came into

2 Throughout this argument, for simplicity, ‘better than’ stands for ‘better than (worse than)’ with all needed adjustments to the corresponding formulation, and similarly ‘worse than’ stands for ‘worse than (better than)’. 3 Bykvist (2007) refers to this premise as ‘converse’. 4 Bykvist (2007) refers to this premise as ‘accessibility’; Arrhenius and Rabinowicz (2010) as ‘subjunctive connection’. Bykvist (2013) refers to it as ‘counterfactual support’. 5 Bykvist (2007) refers to this as ‘actualism’; Bykvist (2013) as ‘wellbeing implies being’.

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

97

existence. Hence it cannot be the case that there is a life that is better for P than never existing.

2 Others’ Responses to the Argument Several authors who endorse the argument remark that while it rules out comparative judgments about whether a given life is better or worse for a person than his never existing, it permits non-comparative claims about the good and ill done to a person by causing her to exist with a particular level of wellbeing. For example, Krister Bykvist (2007, 2013) endorses the aforementioned three conditions but holds that a life may be good for an individual even though it is not better for her than never existing. Similarly, Elizabeth Harman (2004) argues that one can harm someone by creating her if her life contains particular bad features, where it is not necessary for the existence of such harm that leading the life in question is worse for her than never existing. However, many authors seem uncomfortable with the absence of any comparative evaluation of existence vis-a`-vis never existing. They therefore believe that we should try to relax at least one of the three premises. One possibility is proposed by Melinda Roberts (2003: 168–9), who rejects (3a) and (3b), when she assimilates never existing to a zero level of wellbeing, implying that in some sense a person who never exists has a level of wellbeing. However, we believe Roberts is wrong to reject (3a), since we do not think that it makes sense to assign a level of wellbeing to a never-existing person. After all, wellbeing implies being. A different approach is taken by Nils Holtug (2001: 374) and Gustaf Arrhenius and Wlodek Rabinowicz (2010), who argue that premise (2) is too strong. What is needed for the claim that existence in social situation A with a high level of wellbeing is better for a person P than his never existing in B is a three-part relation between A, B, and P. This relation can exist, they claim, only if the person exists. Since, when P exists, all three relata exist, it can make sense to assert that A is better for him than B. But, contrary to (2), one cannot conclude from this that it would have been worse for P if he had not existed, because in that scenario, one of the elements in the relation in question would not exist. As Arrhenius and Rabinowicz put it: A triadic relation consisting in one state [leading a life of a particular quality] being better for a person than another state [never existing] cannot hold unless all three relata exist. Now, the states in question are abstract objects and can indeed be assumed to exist even if they do not actually obtain. However, if [as is assumed] a person is a concrete object, . . . then the relation could not hold if [the] person weren’t alive, since [one of the relata] would not exist. Consequently, even if it is better for [the person] to exist than not to exist . . . [contrary to premise (2)] it doesn’t follow that it would have been worse for [the person] if [he] did not exist, since one of the relata would then be absent. (2010: 405–6)

98

MARC FLEURBAEY AND ALEX VOORHOEVE

As a consequence, they propose to weaken (2) as follows to make an exception for cases in which P does not exist in one of the two states being compared: 20 . a. If P exists in both A and B, then A is better for P than B if and only if B would be worse for P than A, if B obtained. b. If P exists in A but not in B, then A can be better for P than B although B would not be worse for P than A, if B obtained. (2010: 407)

3 Why It Is Not Absurd to Claim that Existence Is Better Than Never Existing In contrast with the aforementioned authors, we accept (1), (2), and (3a), but propose to revise (3b) in such a way that the resulting principle is weak enough to be consistent with the claim that a good life can be better for a person than never existing. As we have seen, Holtug (2001) and Arrhenius and Rabinowicz (2010) accept (3b) because they want to avoid assertions of the following kind about someone who did not come into existence but who would have had a good life, had he existed: It is worse for this person not to exist. They also want to avoid assertions of the following kind about someone who exists with a high level of wellbeing: It would have been worse for her had she never come into existence. The reason seems to be that they want their analysis to be consistent with the following principle: No properties of the never-existent. An individual who never exists cannot have any properties, not even the relational property of something being better or worse for her. (Holtug 2001: 370) We shall now argue that this principle should be rejected because it is too strong. In certain circumstances, there is a clear appeal to the thesis that a never-existing person has no properties. For example, the thesis is true when the comparison is made exclusively between social situations in which the person never exists. But the thesis seems to us to be false when the comparison involves a counterfactual situation in which the person would exist with definite, identifying characteristics. Suppose we consider a never-existing person who could have existed in a counterfactual situation in which this person would have had a great life. Such assumptions give properties to the never-existing person we are thinking about. For one, this person has the property of ‘having a great life in the counterfactual situation’. More generally, this person has all the characteristics that are assigned to her in our description of her situation in the counterfactual state. This is just a straightforward generalization of common discussions of counterfactual states. When we talk about what would have happened if Pope Francis had married, there is no metaphysical conundrum about the identity and characteristics

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

99

of Jorge Maria Bergoglio in this counterfactual state. Similarly, when we talk about what would have happened if Vincent van Gogh would have had a fourth sister (with particular identifying characteristics) instead of his brother Theo, we imagine a person who is different than an existing person without making this hypothetical person a mysterious entity. By a very minor extension, we can talk meaningfully about hypothetical persons who never exist and who are not variations or replacements of existing persons, but whom we can nevertheless richly describe. We can talk about the additional children that Nelson would have had with Emma Hamilton if he had not died at Trafalgar (describing them in a manner that would uniquely identify them), and imagine the different lives they could have had depending on various counterfactual assumptions. There is no greater mystery in establishing the truth conditions for the statement that it is worse for Nelson’s hypothetical son not to have been born (supposing he would have had a great life) than in establishing the truth conditions for the statement that it is worse for the Pope not to have married (supposing he would have had a good marriage). In ordinary discourse, a person is not just a concrete entity who exists in the actual world. Instead, it is the set of possible descriptions associated with the same identity in all the counterfactual states we care to describe. This set may include states in which the person never exists. The literature seems to have failed to see the difference between persons as concrete objects and persons as they figure in ordinary discourse. A concrete object has properties only in the states in which it exists, and one cannot say that a person would be taller in a state in which she does not exist than in the current state in which she does. But when we discuss possible people, the topic of our debate is not only concrete persons. Instead, it is persons as they are considered by evaluators of possible worlds, evaluators who assess these worlds from the perspective of these persons’ interests. For an observer who compares two different possible worlds, in one of which a person exists and in one of which she never exists, it makes sense to compare these worlds for the sake of this person. For a discourse about a person and how different states can be ranked for her sake, it does not even matter whether the person is purely fictional. The only requirement for meaningful comparative evaluation of states from the perspective of a person’s interests (i.e. as better or worse for her) is that the description of every state in which she exists includes information on her level of wellbeing in that state and the description of every state in which she never exists mentions the fact that she does not come into existence.6 In sum, suppose that in situation A, person P would have a good life, and that in B, he would never exist. Accepting (1) and (2), it makes sense to hold not merely that if 6 Arrhenius and Rabinowicz (2013), very briefly (in a footnote), and Bykvist (2013), more extensively, consider the possibility that merely possible persons have a weak form of existence. They rightly reject the idea that such objects can have something like a level of wellbeing. However, they do not seriously consider the possibility that if a merely possible person has well-defined wellbeing levels in counterfactual states, this is sufficient to make the comparison between the person’s lives in these states and his never existing meaningful in states in which he does not exist.

100

MARC FLEURBAEY AND ALEX VOORHOEVE

A were to obtain, then A would be better for P than B, but also that, if B were the actual situation, then his never existing would be worse for him than A. If B were actual, then the fact that this person does not come into existence in the current situation is compatible with the fact that in the counterfactual situation there would be a person with characteristics that are sufficiently definite to make possible the latter comparison. Note that, as mentioned, we endorse (3a). Since a person who never exists has no wellbeing, the statement, It is worse for her never to exist than to exist, cannot mean that in the former case, she has a lower level of wellbeing than in the latter. But one can sensibly hold that a particular life can be better for a person than never existing without assigning a level of wellbeing to never existing. It is sufficient that there is a level of wellbeing, when existing, that is deemed equivalent to never existing. (Call this the ‘personal-value indifference level’ of wellbeing; it is often referred to as the ‘neutral level’.) Then, we submit, enjoying a greater wellbeing than this level implies that a person’s life is better for her than never existing. It also implies that her never existing would be worse for her than existence. Our proposal is, therefore, to replace (3) with the following, weaker condition. 30 . a. A non-existing person has no level of wellbeing; and b. Nothing can be better than anything else for a person who exists in none of the (actual and counterfactual) states under consideration. The combination of (1), (2), and (30 ) is compatible with the idea that existence at a high level of wellbeing is better for P than non-existence, and that this is true whether or not he actually exists. It is also compatible with holding that existence at a very low level of wellbeing is worse for P than non-existence and that this is true whether or not he exists.

4 A Difficulty for Other Views As we shall now explain, this lack of dependence of the truth of such judgments on the actual state of the world gives our proposal an advantage over the proposal advanced by Holtug and by Arrhenius and Rabinowicz. Consider a not-yet-existing, potential person P who would have an awful life if he were to come into existence and whose existence depends on our decision. On the view we propose, it would be worse for P to come into existence and it would be better for him if he never came into existence. This is just to say that, if we were acting solely for his sake, we would not create him. By contrast, the view proposed by Holtug and by Arrhenius and Rabinowicz holds that if P were to come into existence, then his existence would be worse for him and that if he did not come into existence, then his never existing would be neither better than, nor worse than, nor equally good as existence for him. These judgments alone do not determine how one should rank not-yet-existing, contingent

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

101

person P’s existence and his never existing when considering only P’s interests. To do so, their view requires further assumptions. Arrhenius and Rabinowicz appeal to the figure of potential, contingent P’s ‘guardian angel,’ who must make decisions only for P’s sake. Of this angel, they write: if [P’s] guardian angel has a choice between bringing her charge into existence with negative welfare [and] not bringing [him] into existence at all, she would choose the latter. Moreover, if the guardian angel had the choice between bringing her charge into existence with a positive welfare [and] not bringing [him] into existence, she would choose the former. . . . [I]t seems reasonable to say that in the choice between bringing P into existence with negative welfare [and] not bringing [him] into existence at all, one ought to prefer the latter for P’s sake. Likewise, in the choice between bringing P into existence with positive welfare [and] not bringing [him] into existence at all, one ought to prefer the former for P’s sake. (2010: 410–11)

They also propose that we take the expression ‘one ought to prefer never existing to existing at a low level of wellbeing for P’s sake’ simply to mean ‘never existing is better for P than existing with a low level of wellbeing’ (2010: 411). Jointly, these proposals have unappealing implications. For they imply the following two statements about P’s situation: 1. When P does not yet exist, but might be brought into existence, the situation in which he exists is worse for him than the situation in which he does not exist, and the latter is better for him. 2. If P is not brought into existence (and he can no longer be brought into existence), then his never-existing is no longer better for him than his existence. This shift in the circumstances under which the statement ‘P’s never existing is better for him than his existence at a very low level of wellbeing’ is held to be true is very peculiar. If never being brought into existence is better for P now (when he does not yet exist, but can be brought into existence), then how can it fail to remain better for him once an irrevocable decision has been made not to bring him into existence? Of course, an analogous problem arises for Arrhenius and Rabinowicz’s view if the not-yet-existing, contingent P would have a wonderful life. On their view, it would then be true that: 3. When P does not yet exist, but might be brought into existence, the situation in which he exists is better for him than the situation in which he does not exist and the latter is worse for him. 4. If P is irrevocably not brought into existence, then never existing is no longer worse for him than existence. Again, such a shift seems to us very implausible. Holtug (2001: 375 ff.) takes a different approach, but nonetheless seems to face the same problem. According to Holtug, we ‘extrinsically harm’ a person by failing to bring him into existence rather than causing him to exist if and only if (a) bringing

102

MARC FLEURBAEY AND ALEX VOORHOEVE

him into existence would have been intrinsically better for him and (b) bringing him into existence would not intrinsically harm him. As a consequence, Holtug holds that if not-yet-existing, contingent P’s life would be a good one, then it would harm P not to bring him into existence and it would benefit P to bring him into existence. He adds that, on his view, ‘it is difficult to resist the claim that, everything else being equal, we ought to cause [P] to exist’ (384). The reason that Holtug finds this claim difficult to resist, we surmise, is that, on his view, if one takes only the interests of potential, contingent P into account and so chooses for P’s sake, one ought to prefer P’s existence to his never existing. If one then makes the further, natural assumption that ‘preferring P’s existence to his never existing for P’s sake’ is just to say that one takes P’s existence to be better for him than his never existing, then Holtug’s view leads to the same conclusion as Arrhenius and Rabinowicz’s view. When P does not yet exist and his existence depends on our choice, then the state in which P is brought into existence with a high level of wellbeing is judged to be better for him, but if we choose not to bring him into existence, and this choice is irreversible, then it is no longer considered worse for P that he was not brought into existence. In sum, we have argued that views that reject (2) in favour of (20 ) face a dilemma. They must either refuse to rank potential, contingent P’s existence against P’s never existing for P’s sake, or give up the natural assumption that ranking them for his sake is just to rank them as better or worse for him. The view we have proposed does not face this dilemma.

5 Implications for Population Ethics Why is it so important whether one can compare existence to never existing from the point of view of the affected person’s interests? The stakes are often traced to the following principle. Person-affecting principle (PAP). A social situation cannot be better than another if it is not better for someone. Suppose this principle is correct. Also suppose (contrary to our arguments) that one cannot compare existence to never existing from the perspective of a person’s interests. Then one faces the unpleasant prospect of being unable to compare social situations with totally disjoint populations. This is so even when social situation A is obviously better than B, because, say, everyone who exists in A (call them the x-people) has a very high level of wellbeing and everyone who exists in B (the y-people) has an awful life. Accepting that existence can be better for a person than never existing seems a way out of this conundrum. If one adopts Holtug’s and Arrhenius and Rabinowicz’s proposal, then the PAP is compatible with the claim that A is better than B, because it is better for the x-people to exist with a high level of wellbeing rather than never to exist. However, as Arrhenius and Rabinowicz (2013: 15) note, when paired with their proposal for ranking existence vis-a`-vis never existing, the PAP is still too restrictive.

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

103

Consider the choice between A, which contains only x-people who lead wonderful lives, and C, in which the x-people are as well off as in A, but which also contains y-people who have awful lives. Since, on Holtug’s and on Arrhenius and Rabinowicz’s view, A is not better for anyone (it is not better for the x-people because they are equally well off in C, and it is not better for the y-people, since they do not exist in A), it follows from the PAP that A is not better than C. By contrast, on our proposal for ranking existence vis-a`-vis never existing, these two difficulties for the PAP do not arise: A can be said to be better than B because it is better for the x-people that they exist than that they never exist, and A can be said to be better than C, because it is better for the y-people that they are not brought into existence to lead awful lives. Nonetheless, in combination with our view, the PAP might still be regarded as too strong, because it rules out several seemingly reasonable views in population ethics, including Broome’s theory supplemented with a notion of a personal-value indifference level that is below the population-value indifference level. To illustrate: suppose that A consists entirely of well off x-people, while D consists of these x-people at the same level of wellbeing plus y-people at a relatively low level of wellbeing which is barely above the personal-value indifference level. Since, on our view, D is better for the y-people and worse for no one, the PAP holds that A cannot be better than D. But if, as is typically supposed, the population-value indifference level is significantly higher than the personal-value indifference level, then A is better than D. Paired with our view, the PAP would therefore rule out such a variant of Broome’s view. In sum, when paired with our view, the PAP is too restrictive, because it rules out reasonable views like a variant of Broome’s view. But one would also like to say more than the PAP says. Our diagnosis of some of the PAP’s shortcomings is this: insofar as the motivation for the PAP is that it is wellbeing that ultimately matters, this motivation is ill-served by a principle that focuses on the identity of the wellbeing bearers. As a starting point for a different basic principle, we therefore propose the obviously correct claim that a social situation with only one person is better than another social situation with only one person if wellbeing is greater in the former. We propose the following way of extending this to situations with a fixed number of people. It is enough to find a one-to-one mapping from one social situation to the other so that each individual in the former is better off than her counterpart in the latter. This is the Suppes–Sen principle, which combines the Pareto principle for a given population size with anonymity. Suppes–Sen principle (SSP). A social situation A is better than B if they have the same population size and the level of wellbeing is better at every anonymized position (rank in the distribution of wellbeing), from the worst-off to the best-off position.7

This is equivalent to Parfit’s (1984: 360) Principle Q. While we find this principle attractive, we should note that, as Parfit mentions, it rules out some so-called ‘claim-based’ views of population ethics. On these 7

104

MARC FLEURBAEY AND ALEX VOORHOEVE

Can we extend this principle to comparisons of social situations with different numbers of people, without ruling out reasonable views in population ethics? This is indeed possible as follows: Variable-population SSP. A social situation A is better than B if there are two situations A0 and B0 such that: a. A0 is as good as A and B0 is as good as B; b. A0 and B0 have the same population size; and c. the level of wellbeing in A0 is higher at every position than in B0 . This principle does not commit us to anything particular about the value of increasing population size. It is, for instance, satisfied by total utilitarianism,8 average utilitarianism,9 and ‘critical-level’, or (as we would call it) ‘population-value indifference-level’, utilitarianism.10 One may note that for fixed populations, the PAP is weaker than the Pareto principle because it never requires a comparison to be made in a certain way (it only precludes certain relations). A similar variant can be formulated for the variable population SSP: Weak variable population SSP. Social situation B cannot be better than A if there are two situations A0 and B0 such that: a. A0 is as good as A and B0 is as good as B; b. A0 and B0 have the same population size; and c. The level of wellbeing in A0 is higher at every position than in B0 . Finally, we note that the Suppes–Sen principle involves interpersonal comparisons. Can one formulate a principle which, like Pareto and the PAP, involves only intrapersonal comparisons, but which applies to different populations, and does not imply the notorious Mere Addition principle? (This principle holds that raising the wellbeing of incumbents and adding new people just barely above the personal wellbeing indifference level always improves the social situation.) This seems impossible. In views, people do not have a claim to be brought into existence, but they do have a claim to have their quality of life improved if they exist. Now consider a choice between state E, in which Charles comes into existence with 60 utils, Dan with 80, and Edward never exists, and state F, in which Charles has 79 utils, Dan never exists, and Edward exists with 60 utils. (Assume that a life with more than 0 utils is better for a person than never existing.) Then no one has a claim against F, whereas Charles has a strong claim against E. On some claim-based views, Charles’s claim should override impersonal goodness. Parfit (2010: }78) argues that such views are therefore problematic; Otsuka (n.d.) offers a defence of such views. Indeed, there seems to be a tension between the Suppes–Sen principle and claim-based views even in fixed-identity cases (see Voorhoeve (2014) for an example). A0 (B0 ) can have a single individual whose utility is the total utility in A (B). A0 (B0 ) can have a single individual whose utility is the average utility in A (B). 10 A0 (B0 ) can have a single individual whose utility above the population-value indifference level is the total utility in A (B) minus the population-value indifference level multiplied by the population size of A (B). 8 9

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

105

particular, it is impossible to sensibly compare disjoint populations without making interpersonal comparisons.

6 Challenges to Our View The conjunction of our view on the value of existence and the variable-population SSP does not, as far as we can see, have counterintuitive implications for population ethics. However, the idea that existence at a high level of wellbeing is better for a person than never existing may be thought to have such implications when paired with other apparently attractive principles. First, if one accepts our view and the Pareto principle, then the Mere Addition principle follows: if adding an individual does not have negative effects on others, then it is better to add an individual to the population so long as his level of wellbeing is in excess of the personal-value indifference level. If one also assumes inequality aversion, then one cannot avoid the infamous Repugnant Conclusion, which holds that a sufficiently populous social situation in which everyone’s quality of life is just barely in excess of the personalvalue indifference level is better than a social situation with a smaller, uniformly welloff population.11 Second, as we discuss elsewhere in detail (Fleurbaey and Voorhoeve n.d.), the view that existence with a high level of wellbeing is better than never existing may have unpalatable implications when it is paired with a common view about fairness. To illustrate, suppose that Fiona and Georgina both exist. Matters are so fixed that precisely one of them will, due to an untreated illness, have a life at the personal-value indifference level, which we can set at 0 utils (see note 7). The other will get a treatment which will give her a good life. If Fiona is treated, she will have a lifetime wellbeing of 70 utils; if Georgina is treated, she will have 69 utils. You must choose the probability p that the treatment will go to Fiona, with 0  p  1. (Georgina’s chance of receiving the treatment is therefore (1  p).) In this case, a plausible view of fairness requires that you give them both equal chances of treatment.12 Next suppose that neither Fiona nor Georgina exist and that precisely one of them will come into existence. If Fiona exists, her wellbeing will be 70 utils; if Georgina 11

Once adding persons just above the personal-value indifference level is deemed acceptable, then redistributing between a very large number of such additional people and the incumbents can create a very large population with an equalized level of wellbeing that is arbitrarily close to the personal-value indifference level. Note that a similar problem arises for Arrhenius and Rabinowicz’s view when combined with a version of the Pareto principle which they propose, namely: Subjunctive Weak Pareto (SWP): If social situation A would be better than B for everyone who would exist if A were to obtain, and for everyone who would exist if B were to obtain, then A is better than B. As Arrhenius and Rabinowicz (2013) note, their view and SWP together imply a variant of the Mere Addition principle, and therefore, assuming inequality aversion, entail the Repugnant Conclusion. 12 See, for example, Broome (1990).

106

MARC FLEURBAEY AND ALEX VOORHOEVE

exists, her wellbeing will be 69 utils. You must choose the probability p that Fiona will exist. If existence is better than non-existence and if fairness applies to these benefits as it would to benefits to already existing individuals, then fairness requires you to equalize their chances of existence. However, it is counterintuitive that you would be acting unfairly to Georgina in this case if you minimized her chance of existence and maximized Fiona’s chance of existence.13 It seems to us that these challenges reveal that the good that one could do for a contingent person by creating her may not always have the same moral force as the good one could do for an already existing person (or a person whose future existence is determined independently of our choices). Naturally, this difference in moral force requires explanation, which we cannot offer here.14 But if there is indeed such a difference, then one should not simply extend the Pareto Principle to variablepopulation cases; nor should one apply standard ideas about the fairness of equal chances to a benefit for already existing people to cases in which the benefit is coming into existence with a high level of wellbeing.

7 A Challenge to Welfare-based Approaches to the Value of Existence So far, we have followed the literature in focusing only on the wellbeing (understood as what is in the person’s self-interest or what is of prudential value for him) that is enjoyed in the envisioned life in order to determine the personal or social value of a person’s existence. But this is a highly limiting assumption, because it ignores people’s views on how values besides wellbeing determine the value of their existence. Consider a man whose life consists mostly of struggles and suffering and who has few pleasures and achievements. Suppose a neighbour says to him, Your wellbeing is so low that your life is not worth living. This man can reasonably feel insulted by this remark. He may sensibly regard his existence as of great value, even though he would much prefer being spared his trials. He may, for example, have acted well towards others, intelligently pursued noble aims (even though he did not achieve these aims), and responded to adversity with fortitude. And he may, quite sensibly, believe that this makes his life a valuable one, albeit one with a very low level of wellbeing. (Or, if he is a religious man, he may believe that his existence is valuable because it is part of God’s plan.) Symmetrically, a villain who repents of his wrong-doing at the end of his days may have a dim view of the value of his existence even though standard measures of lifetime wellbeing would put him well above the personal-value 13 Bykvist (2013) expresses similar concerns about the idea of valuing never existing as equivalent from the perspective of the person’s interests to living at wellbeing level of zero. If a never-existing, but at one time possible, person were treated just like an existing person with a wellbeing of zero, then this would imply that there is a staggering and heretofore unrecognized amount of inequality. 14 Some have argued that the explanation is that a person cannot be wronged by an action if she would never exist if that action would be performed (Roberts 1998; Vallentyne 2000).

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

107

indifference level and even though, as chance would have it, the consequences of his crimes for others’ wellbeing were minor. What these examples suggest is that wellbeing is not the only consideration that matters in the evaluation of a person’s life relative to his never existing. Whether a life is worth living or not from a person’s reasonable15 comprehensive moral perspective is a deeper question than the question of whether his wellbeing exceeds a particular level. Comparing existence to never existing from the viewpoint of the personal value of individual wellbeing may, at least on some people’s reasonable, comprehensive views, be just as beside the point as comparing Matisse to Ce´zanne by the size of their paintings.16 In sum, there is a mismatch between the evaluation of existence in terms of individuals’ comprehensive moral views and the welfare-based evaluation of existence. Individuals can answer the existential question on the basis of different philosophical and religious values, which cannot be exhaustively accounted for by a suitable notion of wellbeing. These observations might suggest the following argument in favour of an approach that, like Broome’s, avoids appealing to the notion of a life worth living. Social welfare evaluation relies on a population-value indifference level that is defined in terms of wellbeing only. From the perspective of social welfare evaluation, a judgment that a person’s wellbeing is below the population-value indifference level and therefore lowers social welfare does not imply that his existence is bad for him, or bad in terms of a more comprehensive set of values. It just means that social welfare is improved only when new members are above his level. Being told that one’s life is not worth living is very different from being told that it is bad for the social distribution of welfare. It is therefore more legitimate to evaluate the contribution to social welfare of additions to the population in terms of individual welfare only than to judge the value of an individual’s existence (versus his never existing) in terms of his welfare only.

However, obviously, this defence of Broome’s approach is fragile and observations similar to the preceding ones could be used to raise an objection against his theory. Why make social evaluation so distant from the assessments that people will make on the basis of their reasonable, comprehensive moral views? And, why shouldn’t one feel insulted when the criterion for social evaluation assesses one’s existence negatively even though one deems one’s own existence valuable in spite of one’s low level of wellbeing? In sum, both an exclusively welfare-based assessment of whether an individual’s life is worth living and an exclusively welfare-based assessment of whether an individual’s existence improves the distribution of wellbeing are problematic.

We here employ the term ‘reasonable’ in Rawls’s (2005) sense. A similar problem occurs for the notion of a ‘life worth continuing’ when wellbeing is defined in such a way that it may run against the individual’s own judgment (based on her comprehensive moral view) about whether her life is worth continuing. 15 16

108

MARC FLEURBAEY AND ALEX VOORHOEVE

One may object to this conclusion in two ways. First, one may reject the idea that individuals’ evaluations of their existence are generally very different from wellbeing evaluations. This is an empirical counter-argument which may be correct, but which does not address the philosophical problem.17 Moreover, even if only a handful of individuals reject an exclusively wellbeing-based approach to the value of existence, it remains an open question as to how to take their perspective into consideration. Another possible response is to broaden the measure of the value of an individual’s life that one uses in assessing whether an individual’s existence adds personal or social value. The idea would be to move beyond wellbeing narrowly construed as selfinterested or prudential value. On a broadly defined preference-and-value approach, the value of life could be determined by whether people think their life is worth living, taking full account of their reasonable, comprehensive moral views. One would then develop a notion of individual and social value that would incorporate the diversity of values of the members of the population. There are two difficulties with this solution. First, unless we adopt the abovementioned idea that benefits that come from being brought into existence have a different moral force than benefits to existing people, relying on people’s views on the value of existence would suggest that it is good to add people to society when they think their life is worth living in this broader sense (and no incumbent is affected). A version of the Mere Addition principle will therefore obtain, as will a variant of the Repugnant Conclusion. This problem would become even more pressing if there were people whose wellbeing (narrowly construed) is very low but who nevertheless believe strongly in the value of their existence. One may then obtain a variant of the Repugnant Conclusion in which a sufficiently large number of people who all have an arbitrarily low level of wellbeing (narrowly construed) but who all believe their existence is valuable has to be declared better than a smaller, well-off population. The second difficulty is that this approach suggests that one should use a lower population-value indifference level of wellbeing (narrowly construed) for creating lives among populations who strongly believe in the value of their existence, rather than for creating lives among populations who have no such beliefs. It would be strange to evaluate social situations with such a mechanical treatment of diverse philosophical views about the value of existence! In performing social evaluations, could it be right to assign greater value to the creation of individuals in certain sects whose members believe that their own existence fulfils God’s plan, rather than to the creation of equally well-off individuals in other segments of society with a less grandiose view of the value of their existence? The practice of leaving it to genitors to decide whether to have children typically leads to larger families wherever the 17 The argument also does not address the objection that an individual who evaluates his existence solely in terms of his own wellbeing and who has a decent level of wellbeing just below the population-value indifference level may still find his existence judged to be a social bad, even if it is bad for no one.

ON THE SOCIAL AND PERSONAL VALUE OF EXISTENCE

109

value of existence is deemed greater. But should the social evaluation criterion take these views on board and really adopt different population-value indifference levels of wellbeing for different sects? There is a clear dilemma here for liberals. Either one takes account of the value of existence as assessed by the members of the population—and then the social evaluation of variable population choices will, unappealingly, depend on sectarian views— or only ordinary wellbeing considerations are taken on board, as in Broome’s approach, and the project is potentially divorced from what really matters to some people.

References Arrhenius, G., and Rabinowicz, W. (2010) Better to be than not to be? In H. Joas (ed.) The Benefit of Broad Horizons: Intellectual and Institutional Preconditions for a Global Social Science. Leiden, The Netherlands: Brill, 399–421. Arrhenius, G., and W. Rabinowicz (2013) The value of existence. ms. (extended version of Arrhenius and Rabinowicz, 2010). Available at . Broome, J. (1990) Fairness. Proceedings of the Aristotelian Society, New Series 91: 87–101. Broome, J. (1999) Ethics out of Economics. Cambridge: Cambridge University Press. Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press. Bykvist, K. (2007) The benefits of coming into existence. Philosophical Studies 135: 335–62. Bykvist, K. (2013) Being and wellbeing. Paper delivered at the Wellbeing and Preferences workshop, Colle`ges d’Etudes Mondiales, Paris. Fleurbaey, M., and Voorhoeve, A. (n.d.) Priority or equality for possible people? Unpublished manuscript. Harman, E. (2004) Can we harm and benefit in creating? Philosophical Perspectives 18: 89–113. Holtug, N. (2001) On the value of coming into existence. The Journal of Ethics 5: 361–84. Otsuka, M. (n.d.) How it makes a moral difference that one is worse off than one might have been. Unpublished manuscript. Parfit, D. (1984) Reasons and Persons. Oxford: Clarendon Press. Parfit, D. (2010) On What Matters, I. Oxford; New York: Oxford University Press. Rawls, J. (2005) Political Liberalism. 2nd edition. New York: Columbia University Press. Roberts, M. (1998) Child versus Child-maker: Future Persons and Present Duties in Ethics and the Law. Lanham, MD: Rowman & Littlefield. Roberts, M. (2003) Can it ever be better never to have existed at all? Person-based consequentialism and a new Repugnant Conclusion. Journal of Applied Philosophy 20: 159–85. Vallentyne, P. (2000) Review of Child versus Child-maker: Future Persons and Present Duties in Ethics and the Law by Melinda Roberts. Noûs 34: 634–47. Voorhoeve, A. (2014) How should we aggregate competing claims? Ethics 125: 64–87.

8 The Affirmative Answer to the Existential Question and the Person Affecting Restriction Gustaf Arrhenius

I Introduction The person affecting restriction, put as a slogan, states that an outcome can only be better than another if it is better for someone (and, since ‘worse’ is just the converse of ‘better’, an outcome can only be worse than another if it is worse for someone).1,2 The existential question concerns whether existence can be better or worse than nonexistence for a person. According to the affirmative answer to this question, existence can indeed be better or worse than non-existence for a person. In this chapter, I shall discuss the implications of the restriction and the affirmative answer to the existential question for population ethics.3 Hence, the chapter is an investigation into what one could call ‘analytical existentialism’.4

1 I would like to thank John Broome, Krister Bykvist, Christian List, Iwao Hirose, Karim Jebari, Wlodek Rabinowicz, and Orri Stefa´nsson for their very helpful comments. Thanks also to the audience at the Wlodek Rabinowicz Symposium, Dept. of Philosophy, Lund University, 25 April 2014, and to the Colle`ge d’e´tudes mondiales for being such a generous host during some of the time when this chapter was written. Financial support from Riksbankens Jubileumsfond, SCAS, and Fondation Maison des sciences de l’homme through the Franco-Swedish Program in Economics and Philosophy is gratefully acknowledged. 2 See Temkin (1993a: 248) and (1993b: 290) for a similar formulation. The term ‘person affecting restriction’ was introduced by Glover (1977: 66) but see also Narveson (1967). 3 The first three sections of this chapter draws heavily on Arrhenius and Rabinowicz (2010, forthcoming (a)) and Arrhenius (2009, forthcoming). 4 Christian List and Wlodek Rabinowicz have suggested that ‘Scandinavian existentialism’ would be a more fitting term given the dominance of Scandinavian contributions to this topic. In light of John Broome’s important contributions to the field, however, I think this term would be misleading (setting aside the possibility of declaring Broome an honorary Scandinavian).

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

111

II The Person Affecting Restriction The person affecting restriction has a strong intuitive appeal and it has been suggested that it is presupposed in many arguments in moral philosophy, political theory, and welfare economics.5 Moreover, several theorists have suggested that the counterintuitive implications in population ethics of so-called ‘impersonal’ welfarist theories arise because such theories violate this restriction. This applies in particular to Derek Parfit’s well-known repugnant conclusion, which is entailed by classical utilitarianism.6 One can of course interpret the restriction in many ways and some interpretations are actually sufficiently weak to make them perfectly compatible with impersonal welfarist theories such as classical utilitarianism.7 However, here I am interested in a strong reading of the restriction, which is also the most widely discussed one and the one that is thought to help us with the counterintuitive results of impersonal welfarist theories. This version stresses the individualist aspect of value by claiming that axiology is essentially person comparative: The person affecting restriction. If an outcome A is better (worse) than B, then A is better (worse) than B for at least one individual in A or in B. In cases involving only the same people in the compared outcomes, this restriction is quite straightforward and, I surmise, widely accepted by theorists with welfarist inclinations.8 In comparisons between outcomes involving different people, however, and in particular in cases involving people whose existence is contingent on our choices, the restriction becomes ambiguous. An outcome A is better than B for John if he has a higher welfare in A as compared to B. We can assume that much. But what if John exists in outcome A but not in outcome B? Is A then better or worse than B for John? In other words, what is the correct answer to the existential question? Depending on the answer to this question, the restriction has very different implications regarding how to morally evaluate different possible futures.

See Temkin (1993a: }9.4, 1993b). See Parfit (1984: 388) (for my formulation of this conclusion, see Section V of this chapter). For an overview of the counterintuitive implications of impersonal welfarist theories, see, e.g., Parfit (1984), Arrhenius et al. (2010), and Arrhenius (2000a, forthcoming). 7 See Arrhenius (2003a, 2009, forthcoming). 8 Three qualifications: (a) The label ‘person affecting’ might be misleading, since many theorists would, sensibly, weaken the restriction so as to also cover other sentient beings. Cf. Holtug (1996). (b) Since the person affecting restriction is formulated without a ceteris paribus clause, value pluralists are not likely to accept it since it leaves little room for other values apart from welfarist ones. Clearly, one might embrace non-welfarist values such as virtue, reward in accordance to desert, beauty, variety of natural species, etc. (for a discussion of value pluralism in connection with the restriction, see Arrhenius (2003a, 2009, forthcoming). Here I shall, however, only discuss implications of the restriction in cases where one can assume that non-welfarist values are not at stake. Hence, the arguments below also apply to the ceteris paribus version of the restriction, that is, the version that is of interest also to the value pluralists. (c) Certain welfarist theories are ruled out by the restriction already in the same people cases, such as some extreme versions of welfarist egalitarianism (Arrhenius 2009, forthcoming). 5 6

112

GUSTAF ARRHENIUS

The most popular answer to the existential question is probably the negative one: Existence cannot be better or worse than non-existence for a person. Thus, for example, Derek Parfit (1984), John Broome (1999), Krister Bykvist (2007a), and others have worried that if we take a person’s life to be better for her than nonexistence, then we would have to conclude that it would have been worse for her if she did not exist, which is clearly absurd: nothing would have been worse or better for a person if she had not existed. This argument is eloquently stated by Broome: it cannot ever be true that it is better for a person that she lives than that she should never have lived at all. If it were better for a person that she lives than that she should never have lived at all, then if she had never lived at all, that would have been worse for her than if she had lived. But if she had never lived at all, there would have been no her for it to be worse for, so it could not have been worse for her.9

The negative answer to the existential question is bad news for the restriction since it will then have clearly unacceptable implications.10 For example, consider the ‘future bliss or hell case’ shown in Diagram 8.1.

x

y

x

z A

B

Diagram 8.1

The width of each block in the diagram represents the number of people in the population, and the height represents their lifetime welfare. This welfare is positive (or, as we could also put it, people have lives worth living) when the block is above the horizontal line, and negative when the block is below the line.11 Assume that we can 9

Broome (1999: 168, emphasis in the original). Note that this argument, if correct, would also work against the idea that existence could be worse for someone than non-existence. See also Buchanan et al. (2000: 234), Heyd (1988: 159–61, 1992: 124–5), Narveson (1967: 61), and Dasgupta (1995: 383) for similar arguments in favour of the negative answer to the existential question. 10 I discuss this at length in Arrhenius (forthcoming). 11 Let’s call a component of life that neither makes this life better nor worse for the person living it a neutral welfare component. A hedonist, for example, would typically say that an experience which is neither pleasurable nor painful is neutral in value for a person and as such neither increases nor decreases a person’s welfare. We shall say that a life has neutral welfare if and only if it is equally good for the person living it as a neutral welfare component is for a person, and that a life has positive (negative) welfare if and only if it has higher (lower) welfare than a life with neutral welfare. The above definition can of course be combined with other welfarist axiologies apart from hedonism, such as desire and objective list theories. Two remarks: first, these definitions of lives with neutral, positive, and negative welfare do not require comparing a life with non-existence, and thus don’t prejudge the existential question. Secondly, we actually do not need an analysis of a neutral welfare in the present context but rather just a criterion, and the criterion can vary with different theories of welfare. For a discussion of alternative definitions of a neutral life, many of which would also work fine in the present context, see Arrhenius (2000a, forthcoming: chs. 2, 9) and Bykvist (2007b). See also Broome (1999, 2004) and Parfit (1984: 357–8 and Appendix G).

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

113

either see to it that all the people in the future have excellent lives (the y-people in outcome A) or that they have hellish lives (the z-people in outcome B). Assume further that these two possible future populations are of the same size but consist of different people, and that these two outcomes are equally good for us, the currently existing x-people. Since the y- and z-people do not exist in both outcomes, the negative answer to the existential question implies that outcome A is neither better nor worse for the y- and z-people as compared to B. Moreover, the two outcomes are equally good for the xpeople. Hence, according to the person affecting restriction, A cannot be better than B since it is not better for any individual (nor is of course B better than A). In other words, if combined with the negative answer to the existential question, the person affecting restriction implies that these outcomes are either equally good or incomparable in value. But that is clearly the wrong diagnosis of the future bliss or hell case. Rather, outcome A is clearly better than outcome B. So with the negative answer to the existential question, the person affecting restriction has to go. Nils Holtug (1996, 2001), Melinda Roberts (1998, 2003), Matthew Adler (2009), and Rabinowicz and I (2010, forthcoming(a)) have defended an affirmative answer to the existential question. We shall not enter this intriguing and not yet resolved debate here, but we shall assume that any reasonable defence of the affirmative answer to the existential question avoids the absurd conclusion that worries Broome and others. Hence, even if a person’s life is better or worse for her than non-existence, it does not follow that it would have been worse or better for her if she did not exist.12 Given this affirmative answer to the existential question, one might hope that one can retain the person affecting restriction as the link between ‘better’ and ‘better for’. It does seem plausible to claim that, to the extent we only focus on welfare, an outcome cannot be better than another outcome without being better for someone. While the restriction yields very counterintuitive implications when combined with the negative answer to the existential question, might it avoid such implications given an affirmative answer? And can it help us with the counterintuitive implications of impersonal welfarist theories? To which I shall now turn.

12 This matter is discussed at length in Arrhenius and Rabinowicz (2010, forthcoming(a)). The simple answer to why the absurd conclusion doesn’t follow is that a triadic relation consisting in one state (having a certain life) being better for a person p than another state (non-existence) cannot hold unless its three relata exist. Now, the states in question are abstract entities and thus can be assumed to exist even if they do not actually obtain. Consequently, the triadic relation in question can indeed hold as long as also the third relatum, person p, exists. However, if persons are concrete objects, which is the received view, a person exists only insofar as she is alive. Consequently, even if it is better for p to exist than not to exist, assuming she has a life worth living, it doesn’t follow that it would have been worse for p if she did not exist, since one of the relata, p, would then have been absent. What does follow is only that non-existence is worse for her than existence (since ‘worse’ is just the converse of ‘better’), but not that it would have been worse if she didn’t exist.

114

GUSTAF ARRHENIUS

III The Restriction and the Affirmative Answer Firstly, it should be noted that coupled with the affirmative answer, the restriction, as it is usually stated (and as I have stated it above), does yield counterintuitive implications. Consider for instance the case set out in Diagram 8.2.

x

x

z A

B

Diagram 8.2

Assume that we can either see to it that all the people in the future have excellent lives (the x-people enjoying very high positive welfare in outcome A) or that some of them have excellent lives (the x-people in outcome B) but most of them they have hellish lives (the z-people in outcome B), that is, very negative welfare. Clearly, outcome A is superior to outcome B. However, suppose that outcome A is the one that actually obtains. The person affecting restriction implies, counter-intuitively, that A is not better than B, since—as things actually are—there exists no one for whom A is better than B. The added zpeople in the hypothetical outcome B, for whom A would have been better, do not actually exist. Intuitively, however, if A would have been better than B had B obtained (and B is not better than A if A obtains, as in our case), then A is better than B irrespective of whether A or B obtains.13 The reason for this failure of the restriction is that the betterness relation between outcomes does not require the actual existence of the affected persons. Persons enter as relata in the triadic ‘better for’-relation and therefore must exist for that relation to obtain, but they are not relata in the dyadic betterness relation that obtains between outcomes. This contrast between the triadic and the dyadic relations of betterness explains why the person affecting restriction cannot be correct as it stands. To solve problems like this, Holtug has suggested that we should weaken the restriction by adding a disjunctive element.14 We shall formulate this weaker version as follows: The subjunctive person affecting restriction. If an outcome A is better than B, then A would be better than B for someone that would exist if either A or B were to obtain.

13 This counterfactual invariance of the dyadic betterness relation is possible only because its relata (outcomes) can be assumed to exist even if they do not obtain. By contrast, the triadic relation of ‘better for’ can only satisfy a weaker condition of counterfactual invariance: if A would have been better for p than B if B obtained, then A is better for p than B even if B does not obtain, provided that p exists. 14 Holtug (2004).

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

115

In the case discussed above, the second disjunct of this weaker restriction is applicable since, if B were to obtain, then A would be better than B for the z-people. Hence, A can be better than B according to this restriction. Clearly, it is this disjunctive version of the restriction that we should consider given the affirmative answer to the existential question. One might worry, however, that this restriction does not have much bite. If one compares two outcomes A and B, then this restriction will not exclude any rankings of A and B as soon as A contains some persons with positive welfare that do not exist in B, and B contains some persons with positive welfare that don’t exist in A.15 The appearances are misleading, however. The subjunctive restriction does have considerable force, irrespectively of whether it is coupled with a positive or negative answer to the existential question. For example, it rules out all welfarist theories which imply that a mere addition of lives with positive welfare can make a population worse. Prominent examples of such theories are average and critical-level utilitarianism.16 To see this, consider the case presented in Diagram 8.3 (dashes indicates that the block in question should be much wider than shown, that is, the population size is much larger than shown).

A

AÈB

Diagram 8.3

According to average utilitarianism, A[B is worse than A since the average welfare is lower in A[B as compared to A. According to critical-level utilitarianism, the contributive value of a person’s life is her welfare minus a positive critical level and the value of a population is calculated by summing these differences for all individuals in the population. Assuming that the B-people are below the critical level, critical-level utilitarianism reaches the same verdict as average utilitarianism. The addition of B-people has a negative contributive value given that their welfare is below the critical level. However, A[B would not be worse than A for anyone, irrespectively of whether A or A[B were to obtain. Consequently, the subjunctive person affecting restriction implies that A[B cannot be worse than A and thus rules out theories such as average and critical-level utilitarianism. 15 I am here assuming that given the affirmative answer to the existential question, it is better for someone to exist with positive welfare than not to exist at all (more on this below). 16 For the latter theory, see Blackorby et al. (1995, 1997, 2005) and Blackorby and Donaldson (1984). For a discussion of both theories, see Arrhenius (2000a,b, forthcoming, ch. 3 and 5).

116

GUSTAF ARRHENIUS

So the subjunctive restriction has considerable force. Nevertheless, although close at hand given the subjunctive restriction and the affirmative answer to the existential question, we are not yet forced to say that A[B is better than A (or equally as good as A). The subjunctive restriction is compatible with theories that declare these outcomes incommensurable, that is, A[B is neither at least as good as A, nor worse than A. One example is a version of critical-level utilitarianism suggested by Blackorby et al. Instead of using one critical level, they propose an interval of critical levels when comparing populations of different size. The interval of critical levels is assumed to be between zero and a positive welfare level Æ. The idea is that a population A is better than another population B if and only if A is better than B for all critical levels in the interval. If A is better than B for only some critical levels in the interval, and B is better than A for some other critical levels, then A and B are incommensurable.17 According to this theory, incomplete critical-level utilitarianism, A and A[B in Diagram 8.3 are incommensurable since, for some critical levels, the former is better than the latter, and for some other (low) critical levels, the reverse is true.

IV Subjunctive Weak Pareto Consider, however, the following condition: Subjunctive weak Pareto. If A would be better than B for everyone who would exist if A were to obtain, and for everyone who would exist if B were to obtain, then A is better than B. This condition is a version of the standard weak Pareto condition (formulated in terms of ‘better for’) adjusted for the affirmative answer to the existential question. It seems an irresistible condition given that it can be better or worse for someone to exist than not to exist and given the present setting in which we disregard other values apart from welfare. Consider now the outcomes presented in Diagram 8.4 below. Assume that A and A0 consist of the same people, namely the x-people who enjoy very high welfare in both outcomes but higher in A0 as compared to A. The B-people have very low but positive welfare. Again, A and A0 [B are incommensurable according to incomplete critical-level utilitarianism since, for some critical levels, the former population is better than the latter, and for some other (low) critical levels, the reverse is true.

17 See Blackorby et al. (1997: 216–19, 226). That the critical levels consists of all numbers between zero and a positive welfare level is not part of Blackorby et al.’s definition of incomplete critical-level utilitarianism, but they assume this in their discussion of it. See also Blackorby et al. (2005: 219–21, 248–52). For a similar approach, see Broome (2004: 180 ff.) and Rabinowicz (2009).

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

x

A

117

x

A’ÈB

Diagram 8.4

Suppose, however, that A is the case. Then A0 [B is better than A for all the people that exist since the x-people enjoy higher welfare in A0 as compared to A. Assume instead that A0 [B is the case. Since the B-people have positive welfare, A0 [B is better for them than A given the affirmative answer to the existential question. As before, A0 [B is better than A for x-people. Thus, A0 [B is better than A for everybody who would exist irrespective of whether it is A or A0 [B that obtains. It follows from subjunctive weak Pareto that A0 [B is better than A. Hence, the affirmative answer to the existential question in conjunction with subjunctive weak Pareto rules out all theories which imply that A0 [B and A are incommensurable, such as incomplete critical-level utilitarianism. A possible rejoinder here is that we should not only have a critical level for ‘better’ but also for ‘better for’.18 Above, we assumed that given the positive answer to the existential question, it is better for a person to exist with positive welfare than not to exist at all, which indeed seems plausible. However, one might deny this by severing the relation between ‘positive welfare’ and ‘better for’ by claiming that even if a person would enjoy positive welfare, it might not be better for her to exist than not to exist. Rather, in some interval of welfare levels it would be neither better nor worse for a person to exist than not to exist but instead incommensurable in value for her. In other words, below a certain critical welfare level, but above the neutral welfare level, the negative answer to the existential question holds true. As long as the definition of a life with positive welfare does not imply that it is better for a person to exist with positive welfare than not to exist at all, this is a conceptual possibility.19 Given this approach, and assuming that the welfare of the B-people in Diagram 8.4 is below the critical welfare level, it would no longer follow from the affirmative answer to the existential question that A0 [B is better than A for the B-people. Hence, subjunctive weak Pareto would not imply that that A0 [B is better than A, and it would still be open to us to claim that these two outcomes are incommensurable in value.

18

I’m grateful to Wlodek Rabinowicz for pressing me on this point. Recall that the definition of lives with neutral, positive, and negative welfare I suggested in note 11 above did not require the comparison of a life with non-existence and thus did not prejudge the existential question and whether it is better for a person to exist with positive welfare than not to exist at all. 19

118

GUSTAF ARRHENIUS

There are, unfortunately, two problems with this proposal. First, given the affirmative answer to the existential question, the most attractive and congenial definition (or criterion) of a life with positive welfare is in terms of ‘better for’: A life has positive welfare if and only if it is better for a person to exist with such a life than not to exist at all. Given this definition, there is no conceptual space for claiming that a person would enjoy positive welfare but that it might not be better for her to exist than not to exist. Secondly, in combination with the restriction, this proposal yields violation of the following condition, which is, I believe, as uncontroversial as it gets in population axiology: The egalitarian dominance condition. If A is a perfectly equal population of the same size as population B, and every person in A has higher welfare than every person in B, then A is better than B, other things being equal. Consider two same-sized mutually disjoint populations A and B in which everyone enjoys positive welfare albeit below the critical welfare level. Assume that A is a perfectly equal population and that every person in A has higher welfare than every person in B. Accordingly, egalitarian dominance ranks A as better than B. However, since the welfare of all the A-people are below the critical welfare level, A would not be better than B for the A-people if it were to obtain (nor, of course, would B be better than A for the A-people if it came about). Rather, A and B are incommensurable in value for the A-people. Since A would not be better than B for anyone, irrespectively of whether A or B were to obtain, it follows from the subjunctive person affecting restriction that A cannot be better that B, a clear violation of egalitarian dominance.

V The Repugnant Conclusion In light of the above results, one might suspect that rescuing the person affecting restriction by adopting the affirmative answer to the existential question will make it hard to avoid the well-known counterintuitive implications of classical utilitarianism, such as the repugnant conclusion.20 This is true, I fear. Consider the following very weak inequality aversion condition: The inequality aversion condition. For any triplet of welfare levels A, B, and C, A higher than B, and B higher than C, and for any population A with welfare A, there is a larger population C with welfare C such that a perfectly equal population B of the same size as A[C and with welfare B is at least as good as A[C, other things being equal.21 20 As I have argued elsewhere, making population ethics more ‘person affecting’, so to speak, does not suffice to save it from counterintuitive implications (Arrhenius 2009, forthcoming). 21 The ceteris paribus clause in the formulation is meant to rule out that the compared populations differ in any axiologically relevant aspect apart from individual welfare levels.

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

119

Another way of stating the inequality aversion condition is that, for any welfare level of the best off and worst off, and for any number of best off lives, there is a (much) greater number of worst off lives such that it would be at least as good to have an equal distribution of welfare on any level higher than the worst off, other things being equal. It is a very weak egalitarian condition since it can be satisfied by a theory which demands that the total welfare must be greater for a population with perfect equality to be better than an unequal population of the same size. Moreover, it is also compatible with principles that give much greater weight to the welfare of the best off as compared to the welfare of the worst off. For example, a theory which requires that to compensate for one person falling from twenty to ten units of welfare, a hundred people have to be moved from zero to ten units, is compatible with the inequality aversion condition. In that sense, its name is a bit misleading since it is compatible with quite non-egalitarian theories. Roughly, inequality aversion only rules out theories that imply that we should always or sometimes give some kind of ‘lexical priority’ to the best off. A simple example of the former theory is ‘Maximax’: maximize the welfare of the best off (we shall below briefly discuss another more subtle theory that violates inequality aversion).22 Now, consider the populations shown in Diagram 8.5 below. A and A0 [B are the same populations as in Diagram 8.4. The C-people have very low positive welfare but higher that the B-people. The size of C is the same as A0 [B.

x

A

x

A’ÈB

C

Diagram 8.5

According to inequality aversion, there is a size of the B-population such that C is at least as good as A0 [B. As we saw above, it follows from subjunctive weak Pareto and the affirmative answer that A0 [B is better than A. It follows by transitivity that C is better than A. Hence, the affirmative answer to the existential question together with subjunctive weak Pareto and inequality aversion yield the repugnant conclusion: 22 It should be noted that inequality aversion is applicable even if only ordinal measurement of welfare is possible. It only presupposes that lives can be ordered by the relation ‘has at least as high welfare as’. Here is an example of a principle which only presupposes ordinal measurement of welfare and satisfies this condition: If the worst off make up at least 99 per cent of a population, then it would be better to have an equal distribution of welfare on any level higher than the worst off. Maximax, mentioned above, is an example of an ordinal principle that violates inequality aversion.

120

GUSTAF ARRHENIUS

The repugnant conclusion. For any perfectly equal population consisting of people with very high positive welfare, there is a population consisting of people with very low positive welfare which is better, other things being equal.23 Hence, saving the person affecting restriction by adopting the affirmative answer to the existential question comes at a high price: we cannot avoid the repugnant conclusion, an implication of paradigmatically impersonal theories such as classical utilitarianism. Moreover, this is an implication that many find highly counterintuitive, and especially those that have embraced the restriction. One might find the dialectic here a bit hard to follow: How can the above derivation be a problem for the restriction when we did not use it in the derivation? It is true that in this derivation of the repugnant conclusion, we did not make use of the restriction, but of subjunctive weak Pareto. The latter principle is very plausible given that we accept the affirmative answer to the existential question. So it is this affirmative answer, rather than the restriction, that makes the above derivation of the repugnant conclusion possible. However, the affirmative answer is needed to make the restriction plausible, as was shown by the future bliss or hell case. On the other hand, the above derivation also shows that, given the affirmative answer, rejecting or weakening the restriction will not suffice to avoid the repugnant conclusion. Of course, a defender of the restriction and/or the affirmative answer might here decide to reject the inequality aversion condition instead and in that way block the derivation of the repugnant conclusion, perhaps by an appeal to some kind of superior goods that putatively would be lost in C as compared to A0 [B.24 An obvious drawback of this move is, of course, that it is very counterintuitive to reject the compelling inequality aversion condition. Most theories imply, and most theorists endorse, much stronger inequality aversion conditions. Moreover, the inequality aversion condition can be derived from an even more intuitively compelling condition, the non-elitism condition, which roughly says that there is at least some very small decrease in welfare for one of the best off persons which can be compensated for by an increase in welfare for at least some (possibly much greater) number of the worst off people, to the effect that the involved people enjoy the same level of welfare.25

23 See Parfit (1984: 388). This formulation is more general than Parfit’s apart from that he does not demand that the people with very high welfare are equally well off. Although it is through Parfit’s writings that this implication of classical utilitarianism became widely discussed, it was already noted by Henry Sidgwick (1907: 415), before the turn of the century (but he didn’t claim that it was ‘repugnant’, however). For other early sources of the repugnant conclusion, see Broad (1930: 249–50), McTaggart (1927: 452–3), and Narveson (1967). 24 For a detailed discussion of superior goods and the repugnant conclusion, see Arrhenius (2005; forthcoming) and Arrhenius and Rabinowicz (2005, forthcoming(b)). 25 For an exact statement of this condition and the derivation of inequality aversion from it, see Arrhenius (2000a, 2001, 2003b, 2011, forthcoming).

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

121

Finally, it is not at all apparent that C must involve less superior goods than A0 [B since what might explain the lower welfare level in C does not have to be a smaller amount of superior goods but that there is more pain and suffering in the C-lives which to a sufficient extent outweighs the superior goods to make their welfare very low.26 Nevertheless, the implausibility of rejecting inequality aversion is actually not the main problem with this move in the present dialectic. Rather, the problem is that it still would be true that it is not an appeal to the restriction that is solving the problem here. Rejecting inequality aversion and non-elitism is equally open to those who reject the restriction and defend impersonal welfarist theories. So it would not be the restriction, after all, which is helping us with avoiding the implications of impersonal welfarist theories. On the contrary, rescuing the restriction by adopting the affirmative answer limits our options drastically since, given the latter answer, we are forced to give up inequality aversion and non-elitism or subjunctive weak Pareto, or accept the repugnant conclusion—all very unattractive options.

VI The Person Affecting Restriction and Population Ethics Reconsidered The derivation of the repugnant conclusion in the preceding section relied on subjunctive weak Pareto. Perhaps a defender of the restriction and/or the affirmative answer to the existential question could take this derivation ‘as a reason to be cautious with seemingly irresistible conditions such as Subjunctive Weak Pareto’.27 I have no suggestion, however, how one could undermine the intuitive strength of subjunctive weak Pareto given the affirmative answer to the existential question. In any case, even if such a reason could be found, we would still run into problems with the repugnant conclusion as long as we hold on to the restriction. As we shall see, there is no population axiology that avoids this conclusion in a satisfying manner and which also satisfies the subjunctive person affecting restriction, inequality aversion, and egalitarian dominance.28 Hence, in this demonstration, subjunctive weak Pareto is not involved, but rather it is the restriction itself that plays a leading role.

26 One can also show that the affirmative answer to the existential question together with subjunctive weak Pareto and another weak condition yield the negative repugnant conclusion: For any population consisting of people with any very negative welfare, there is a population consisting of people with only slightly negative welfare which is worse, other things being equal. For a discussion of this conclusion, see Carlson (1998: 297 ff.; he calls it the ‘Reverse Repugnant Conclusion’ or ‘RRC’) who ‘find[s] RRC very difficult to accept’. Since this conclusion might only involve bad welfare components, an appeal to superior goods to avoid it seems to be a non-starter. 27 Arrhenius and Rabinowicz (forthcoming(a)). 28 This also shows, contrary to what many seem to have believed, that the restriction will not help us with the repugnant conclusion irrespective of what answer one gives to the existential question (that is, even when coupled with the negative answer or some other possible answer).

122

GUSTAF ARRHENIUS

Let me first introduce the following condition to the effect that the repugnant conclusion is false: The quality condition. There is a perfectly equal population consisting of people with very high positive welfare which is at least as good as any population consisting of people with very low positive welfare, other things being equal. Avoidance of the repugnant conclusion implies that there is at least one population with very high welfare which is at least as good as or incommensurable with all larger populations with very low welfare. The quality condition is logically stronger than avoidance of the repugnant conclusion since the former rules out axiologies that satisfy the latter by implying that at least one population with very high welfare is incommensurable with all populations with very low positive welfare, although none is at least as good as all such populations. This way of avoiding the repugnant conclusion is quite unsatisfactory, however, and does not capture most people’s intuition about the conclusion, I surmise. Perhaps it can be reasonably believed that some populations with very high welfare are incommensurable with some populations with very low welfare, but that some of them are incommensurable with all larger populations with very low welfare seems, given that other things are equal, counterintuitive. Now consider the outcomes presented in Diagram 8.6 below. Populations B, C, and D are three perfectly equal populations with very low positive welfare. There is higher welfare in B than in C, and higher in C than in D. A and E are two perfectly equal same-sized populations consisting of the x-people with very high positive welfare. The x-people are even better off in E as compared to A. Populations B, C, and E[D consist of the x- and y-people and are thus all of the same size. We can assume that A and B are two populations satisfying the quality condition such that A is at least as good as B. Moreover, we can stipulate that D is of such a size that the inequality aversion implies that C is at least as good as E[D. According to egalitarian dominance, B is better than C since there is perfect equality in B and everyone is better off as compared to in C.

x

x

y A

EÈD

x+y

x+y

C

B

Diagram 8.6

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

123

Since A is at least as good as B, and B is better than C, it follows by transitivity that A is better than C. Similarly, since C is at least as good as E[D, it follows that A is better than E[D. However, as the x-people in E are better off than the x-people in A, and the D-people have positive welfare, it follows from the subjunctive person affecting restriction, irrespective of which answer we give to the existential question, that A cannot be better than E[D since A is not better for anyone in A or E[D. In other words, E[D is not worse than A. Hence, the assumption that there is an axiology which satisfies all the adequacy conditions entails a contradiction, namely that A is better than E[D and that E[D is not worse than A. Thus, there is no population axiology which satisfies the subjunctive person affecting restriction, the egalitarian dominance, inequality aversion, and the quality condition.29 In other words, contrary to what many seem to have believed, the restriction will not help us with the repugnant conclusion but rather make it very hard to avoid it.

References Adler, M. D. (2009) Future generations: A prioritarian view. The George Washington Law Review 77(5–6): 1478–520. Arrhenius, G. (2000a) Future Generations: A Challenge for Moral Theory. Uppsala, Sweden: University Printers. Arrhenius, G. (2000b) An impossibility theorem for welfarist axiologies. Economics and Philosophy 16(2): 247–66. ¨ sterberg’s population theory has in common with Plato’s. In Erik Arrhenius, G. (2001) What O Carlson and Rysiek Sliwinski (eds.) Omnium-gatherum. Uppsala Philosophical Studies 50. Uppsala, Sweden: Department of Philosophy, Uppsala University, 29–44. Arrhenius, G. (2003a) The person affecting restriction, comparativism, and the moral status of potential people. Ethical Perspectives 10(3–4): 185–95. Arrhenius, G. (2003b) The very repugnant conclusion. In Krister Segerberg and Rysiek Sliwinski (eds.) Logic, Law, Morality, Uppsala Philosophical Studies 51. Uppsala, Sweden: Department of Philosophy, Uppsala University: 167–80. Arrhenius, G. (2005) Superiority in value. Philosophical Studies 123: 97–114. Arrhenius, G. (2009) Can the person affecting restriction solve the problems in population ethics? In M. Roberts and D. Wasserman (eds.) Harming Future Persons. Aldershot, UK: Ashgate. Arrhenius, G. (2011) The impossibility of a satisfactory population ethics. In H. Colonius and E. Dzhafarov (eds.) Descriptive and Normative Approaches to Human Behavior. Advanced Series on Mathematical Psychology. Hackensack, NJ: World Scientific Publishing Company. Arrhenius, G. (forthcoming) Population Ethics. Oxford University Press. Arrhenius, G., and Rabinowicz, W. (2005) Millian superiorities. Utilitas 17(2): 127–46. 29 I am grateful to Orri Stefa´nsson for discussion of this matter. He suggested a similar argument given full comparability between populations. The above demonstration did not presuppose full comparability (which is preferable since, as we have seen, some people have the intuition that A and E[D might be incomparable in value and this is implied by some theories such as incomplete critical level utilitarianism).

124

GUSTAF ARRHENIUS

Arrhenius, G., and Rabinowicz, W. (2010) Better to be than not to be? In The Benefit of Broad Horizons: Intellectual and Institutional Preconditions for a Global Social Science. Leiden, The Netherlands: Brill, 399–414. Arrhenius, G., and Rabinowicz, W. (forthcoming(a)) The value of existence. In The Oxford Handbook of Value Theory. Oxford University Press. Arrhenius, G., and Rabinowicz, W. (forthcoming(b)) Value superiority. In The Oxford Handbook of Value Theory. Oxford University Press. Arrhenius, G., Ryberg, J., and Tännsjo¨, T. (2010) The repugnant conclusion. Stanford Encyclopaedia of Philosophy, . Blackorby, C., and Donaldson, D. (1984) Social criteria for evaluating population change. Journal of Public Economics 25: 13–33. Blackorby, C., Bossert, W., and Donaldson, D. (1995) Intertemporal population ethics: Critical-level utilitarian principles. Econometrica 65: 1303–20. Blackorby, C., Bossert, W., and Donaldson, D. (1997) Critical-level utilitarianism and the population-ethics dilemma. Economics and Philosophy 13: 197–230. Blackorby, C., Bossert, W., and Donaldson, D. (2005) Population Issues in Social Choice Theory, Welfare Economics, and Ethics. New York: Cambridge University Press. Broad, C. D. (1930) Five Types of Ethical Theory. London: Routledge and Kegan Paul. Broome, J. (1999) Ethics out of Economics. Cambridge: Cambridge University Press. Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press. Buchanan, A., Brock, D. W., Daniels, N., and Wikler, D. (2000) From Chance to Choice: Genetics and Justice. Cambridge: Cambridge University Press. Bykvist, K. (2007a) The benefits of coming into existence. Philosophical Studies 135: 335–62. Bykvist, K. (2007b) The good, the bad, and the ethically neutral. Economics and Philosophy 23: 97–105. Carlson, E. (1998) Mere addition and two trilemmas of population ethics. Economics and Philosophy 14: 283–306. Dasgupta, P. (1995) An Inquiry into Well-being and Destitution. Oxford: Oxford University Press. Glover, J. (1977) Causing Death and Saving Lives. London: Penguin Books. Heyd, D. (1988) Procreation and value: Can ethics deal with futurity problems? Philosophia 18: 151–70. Heyd, D. (1992) Genethics: Moral Issues in the Creation of People. Berkeley, CA: University of California Press. Holtug, N. (1996) In defence of the slogan. In W. Rabinowicz (ed.) Preference and Value: Preferentialism in Ethics (Studies in Philosophy). Lund, Sweden: Lund University, 64–89. Holtug, N. (2001) On the value of coming into existence. The Journal of Ethics 5: 361–84. Holtug, N.(2004) Person-affecting moralities. In J. Ryberg and T. Tännsjo¨ (eds.) The Repugnant Conclusion. Essays on Population Ethics. Dordrecht, The Netherlands: Kluwer, 129–61. McTaggart, J. M. E. (1927) The Nature of Existence. Cambridge: Cambridge University Press. Narveson, J. (1967) Utilitarianism and new generations. Mind 76: 62–72. Parfit, D. (1984) Reasons and Persons. Revised reprint, 1991. Oxford: Clarendon Press. Rabinowicz, W. (2009) Broome and the intuition of neutrality. Philosophical Issues 19: 389–411. Roberts, M. A. (1998) Child versus Childmaker: Future Persons and Present Duties in Ethics and the Law. Lanham, MD: Rowman and Littlefield.

THE AFFIRMATIVE ANSWER TO THE EXISTENTIAL QUESTION

125

Roberts, M. A. (2003) Can it ever be better never to have existed at all? Person-based consequentialism and a new repugnant conclusion. Journal of Applied Philosophy 20: 159–85. Sidgwick, H. (1907) The Methods of Ethics. 7th edition, 1967. London: Macmillan. Temkin, L. S. (1993a) Inequality. Oxford: Oxford University Press. Temkin, L. S. (1993b) Harmful goods, harmless bads. In R. G. Frey and C. W. Morris (eds.) Value, Welfare, and Morality. Cambridge: Cambridge University Press, 291–324.

PART II

Reasoning

9 The Meaning of ‘Darn it!’ Luc Bovens and Wlodek Rabinowicz

When vain desire at last and vain regret Go hand in hand to death, and all is vain, What shall assuage the unforgotten pain, And teach the unforgetful to forget? Dante Gabriel Rossetti, The One Hope

1 The einmal-ist-nicht-keinmal Game A bookie picks a random natural number x from 0 to 1,000,000 under a uniform distribution and constructs an urn that has precisely x white balls and (1,000,000 – x) black balls. You are allowed to pick a random ball from this urn. If it is a white ball you get $1. If it is a black ball you get $0. Suppose that you are a risk-neutral person. It is easy to calculate that you would be willing to pay 50 cents for this game. It is also easy to calculate that your credence that there are less than 50 per cent white balls in the urn is just slightly lower than 0.50. Now suppose that, before drawing the fateful ball from the urn which will determine your winnings, you get a chance to sample the urn. You draw a ball out of the urn, and it is black. You put it back in the urn. How much would you be willing to pay now to play this game? And what is your credence that there are less than 50 per cent white balls in the urn? There is a German saying that once is never—einmal ist keinmal. There is a wisdom embodied in this saying which suggests that sampling a single black ball from a 1,000,000 ball urn should not make much difference to how much a riskneutral person is willing to bet and to her credence that there are less than 50 per cent white balls in the urn. But in probability theory, appearances may deceive. Indeed, some simple calculations show that a risk-neutral person should now only be willing to pay about 33 cents for this game and that her credence that there are less than 50 per cent white balls in the urn is close to 0.75. (See appendix.)

130

LUC BOVENS AND WLODEK RABINOWICZ

Where does one go wrong in underestimating the import of a single observation? One does not realize that the minimal information provided by a single draw of one black ball radically changes our credence distribution over the objective chances of winning the game. Obviously, the credence that the urn contains no black balls now goes to zero, but this is by no means the only thing that happens. Posterior credences of all the possible cases in which the urn contains fewer black balls than white balls decrease, whereas posterior credences of all the cases in which it contains more black balls than white balls increase. It is a matter of a global all-pervasive change in credences. Consequently, drawing a single black ball makes a huge difference to our expectation and to our credence about the number of balls in the urn. Considering this drastic effect of a single draw we call this the einmal-ist-nicht-keinmal game—the once-is-not-never game. Now let us return to the game without sampling—let us call this the urn game. Suppose that you can play either the urn game or a game of coin-toss, a game which involves tossing a fair coin and which pays $1 if heads and $0 if tails. Suppose you choose to play the urn game and you lose. ‘Darn it!’ you exclaim. What do you mean by this? That is our subject in this chapter. The relevance of this issue is that we would like to understand the nature of regret after making uncertain choices. We understand ‘uncertain choices’ in the following way. These are choices whose outcomes teach us something about the chances of a win or a loss. Hence, there is a difference between our prior credence that things may turn out well or may turn out poorly and our posterior credence that things might have turned out well or might have turned out poorly. Such choices are different from risky decisions, for example choosing the game of coin-toss with a fair coin—whether we win or lose, the chance of winning the game of coin-toss is known beforehand. There is no difference between our prior and posterior credence.

2 Twelve Types of Regret When I ask you for a gloss on your expletive (‘Darn it!’) after losing the urn game there are two things you might say. Either you might say, I should not have played the urn game; I should have gone for the coin-toss instead!, or you might say, I could have done so much better! That is, your regret may focus on your action of choosing the urn game rather than the coin-toss game. Let us call this action regret. Or it may focus on the poor outcome that you got. Let us call this outcome regret. There are many terms in the neighbourhood of ‘regret’ that are not quite synonymous. For example, one might say that one regrets not having done otherwise, but one is disappointed that things did not turn out otherwise. This may be so, but what we are after is something more generic: the negative retrospective emotion that is the root cause and the root expressive content of the expletive ‘Darn it!’ This expletive is fitting both when we realize that we made the wrong choice as well when a coin did not come up the desired way. To start off, we will use ‘regret’ as a blanket term to

THE MEANING OF ‘ DARN IT !’

131

cover this emotion. We will draw some further distinctions when the need arises. But first we are going to provide a taxonomy of different kinds of comparisons that might elicit the generic emotion in question. a. Three Types of Action Regret. As to action regret, I could still push you a bit further to expand on your gloss. Again, there are two things you might say. First, you may point to the fact that, knowing what you know now, it is clear that the coin-toss game had a higher expectation than the urn game, namely 50 cents rather than 33 cents. Let us call this expectation action regret. Second, you might say that, knowing what you know now, your credence is high that the chance of winning the coin-toss game was higher than the chance of winning the urn game. Indeed, you have 0.75 credence that there are less than 50 per cent white balls in the urn. Let us call this probabilistic action regret. There is one more way of assessing action regret which our set up does not permit. In our set up, actions and states are not independent. As a matter of fact, if we chose to play the urn game, the coin will not be tossed and thus neither the state of heads nor the state of tails in the coin-toss game will materialize. (And vice versa: if we play the coin-toss game, neither the state of drawing a white ball from the urn nor the state of drawing the black ball will materialize.) When actions and states are independent, then it makes sense to compare the outcome that you did get to the outcome that you would have got if you had chosen differently, assuming that the state which is actualized is kept fixed. For instance, suppose that my actions are predicting heads or predicting tails, whereupon a fair coin is being tossed and I win a $1 if and only if I correctly called the outcome. If I predict heads, and tails comes up, then I can compare my $0 with the $1 if I had predicted tails. Let us call this outcome action regret. We measure expectation action regret by means of the posterior expectation of the urn game minus the expectation of the coin-toss game. We measure probabilistic action regret by means of the posterior credence that the chance of winning the cointoss game was higher than the chance of winning the urn game. b. Nine Types of Outcome Regret. While there is no room for outcome action regret in the urn game, due to the dependence of states on actions, it is still possible to talk about pure outcome regret in this case: when you lose the urn game, you may well be disappointed with the outcome. As to such outcome regret, indeed, you could have done better—but in what way could you have done better? Well, there are six possibilities: You may point to your loss and compare it to: i. the win that you could have had doing what you did, i.e. playing the urn game; ii. the win that you could have had had you chosen to do differently, i.e. playing the coin-toss game; iii. the prior expectation of what you did, i.e. of playing the urn game; iv. the posterior expectation of what you did, i.e. of playing the urn game;

132

LUC BOVENS AND WLODEK RABINOWICZ

v. the prior expectation of what you could have done, i.e. of playing the coin-toss game; vi. the posterior expectation of what you could have done, i.e. of playing the cointoss game. Here, (v) and (vi) are not extensionally distinct from each other, since the prior and the posterior expectation of what you could have done are the same: you did not learn anything about the unchosen action. Indeed, even if you had played the cointoss game, your prior and posterior expectation for that game would have been the same: a win or a loss would not have affected your expectation of this game. Furthermore, in comparing your loss to the win that you did not get, it might also be natural to compare this loss to the best win that you could have had, be it from the chosen or the unchosen action. Correspondingly, in measuring your loss by comparing it to the expectation, it might be natural to compare this loss to the best prior or posterior expectation, be it from the chosen or the unchosen action. In our case, the best posterior expectation is the expectation of the unchosen action, the coin-toss game, since the posterior expectation of the urn game has gone down to 0.33. Table 9.1 summarizes all these different types of outcome regret. We will name these types of outcome regret by concatenating row names and column names, for example, chosen-action/win outcome regret, and so on.

3 Uncertainty Aversion and Uncertainty Proneness Krähmer and Stone (2013) provide an explanation for uncertainty aversion that can be framed in terms of our analysis of regret. Suppose that you are offered a choice between 50 cents in hand or playing the urn game. You are uncertainty averse and take the 50 cents. What might explain your refusal to play? You might reason as follows. There is a 50–50 probability of winning or losing. Suppose I lose. Then I will look back on my choice and say: I chose to play a game that has a posterior expectation of 33 cents rather than doing the smart thing and opting for the 50 cents for sure. I now have zero cents and so my loss, as compared to the expected value of the smart action, is 50 cents. Let this be a measure of my regret. Suppose I win. Then I will look back on my choice and say: I chose to play a game that has a posterior expectation of 67 cents. That was a smart thing to do. In addition, Lady Luck smiled on me. I have $1 in my hands and so my gain, as compared to the Table 9.1. Nine types of outcome regret

Chosen action Unchosen action Best

Win

Prior expectation

Posterior expectation

(1
0) = 1 (1
0) = 1 (1
0) = 1

(0.50
0) = 0.50 (0.50
0) = 0.50 (0.50
0) = 0.50

(0.33
0) = 0.33 (0.50
0) = 0.50 (0.50
0) = 0.50

THE MEANING OF ‘ DARN IT !’

133

posterior expectation of the smart action, is 33 cents. Let this be a measure of my joy. If I choose to play the urn game, then (i) the chances that I will experience regret or joy are 50–50 and (ii) my anticipated regret is greater than my anticipated joy. Hence I refuse to play the urn game.1 This is a clever suggestion, but it requires a very particular way of framing our anticipation of feelings of regret and joy. Regret is measured as best/posteriorexpectation outcome regret. I ask myself: How much worse did I do than the posterior expectation of the action that has the best posterior expectations? The answer is: I have $0, and that is 50 cents less than the 50 cents which is my best posterior expectation. I could have taken the 50 cents in hand rather than playing a game with a posterior expectation of 33 cents. Correspondingly, joy is measured as best/posterior-expectation outcome joy. In the case of joy, I ask myself: How much better did I do than the posterior expectation of the action that has the best posterior expectation? The answer is: I have $1, and that is 33 cents more than the posterior expectation of 67 cents of playing the urn game. But if we measure regret relative to the best thing that I could have done from an ex post perspective, then should we not measure joy relative to the worst thing that I could have done from an ex post perspective? Should I not say: I am so joyful that I did not take the 50 cents in hand—that would have been a stupid thing to do since the urn game has a posterior expectation of 66 cents? And hence I should frame my joy in terms of a gain from 50 cents to $1. And then my anticipated regret and anticipated joy are equal, and the argument for abstaining from the urn game crumbles. Krähmer and Stone may respond that this way of framing the choice problem may well be reasonable, but the fact of the matter is that some people are uncertainty averse. Krähmer and Stone are merely providing a behavioural explanation of uncertainty aversion by pointing to some feature of the psychology of uncertainty averse people. But then, could we extend this behavioural explanation for uncertainty proneness? Building on our analysis, we can indeed provide an explanation of both uncertainty attitudes. Note the terms ‘joy’ and ‘regret’. They are not quite antonyms. So far we have let ‘regret’ cover the widest range of emotions and the term is indeed used broadly in the English language. But we could distinguish between two negative emotions: an emotion of sorrow (sadness, disappointment, . . . ) which is responsive to bad luck in the chosen action and an emotion of regret proper, so to speak, which is responsive to the realization that things might have been better with unchosen actions. Similarly, we could distinguish between two positive emotions: an emotion of joy (gladness, being thrilled, . . . ) which is responsive to good luck in the chosen action and an emotion of relief which is responsive to the realization that things might have been 1 Krähmer and Stone apply this kind of reasoning to the famous Ellsberg’s paradox. They use their model to explain why agents tend to prefer risky choices to the uncertain ones. Unlike in uncertain choices, in risky choices the posterior expectation does not increase in case of a win, which explains why the expected joy from a win is higher in these cases; so much higher that it offsets the expected regret from a loss.

134

LUC BOVENS AND WLODEK RABINOWICZ

much worse with unchosen actions. ‘Sorrow’ and related expressions (‘sadness’, ‘disappointment’, . . . ) are then antonyms of ‘joy’ (‘gladness’, ‘being thrilled’, . . . ), whereas ‘regret’ in this more restricted sense is an antonym of ‘relief ’. Much more can be said in the way of conceptual analysis of terms denoting retrospective emotions, but this will do for our purposes. Sorrow and joy are responsive to Lady Luck’s frowning or smiling upon us. Now, from an ex post perspective, to what extent did Lady Luck frown upon us in the urn game when we lose? She presented us with $0 in a game with a posterior expectation of 33 cents. This difference of 33 cents is a measure of our sorrow, which equals chosen-action/posterior-expectation outcome regret in our scheme. Similarly, the extent to which Lady Luck smiled upon us when we have a win is the difference between the $1 that we won and the posterior expectation of 67 cents, or 33 cents. Regret and relief are responsive to features of the unchosen actions. A measure of my regret is the difference between the $0 resulting from my loss compared to the expectation of 50 cents, had I chosen for the money in hand, or (50 cents
$0) = 50 cents. This is unchosen action/expectation outcome regret in our scheme. A measure of my relief is the difference between the $1 resulting from my win compared to the expectation of 50 cents had I chosen for the money in hand, that is ($1
50 cents) = 50 cents. We can now let uncertainty attitudes be determined by our anticipated retrospective emotions in decisions under uncertainty. If an agent anticipates joy (in case of a win) and regret (in case of a loss), then the negative emotion of regret measured at 50 cents trumps the positive emotion of joy measured at 33 cents, and hence she will be uncertainty averse.2 If an agent instead anticipates relief (in case of a win) and sorrow (in case of a loss), then the positive emotion of relief measured at 50 cents trumps the negative emotion of sorrow measured at 33 cents and she will be uncertainty prone. Let us try to make this psychology plausible. Suppose that we interview uncertainty averse and uncertainty prone persons. In cases in which a win or a loss affects our posterior expectation of the game, an uncertainty averse person would justify her choice of taking the money in hand as follows: See, if I win, I would not be all that excited—I would feel that my win was very much in the cards. But if I lose, I would kick myself, because I could instead have opted for an action whose posterior expectation was so much better than the outcome I have got now. An uncertainty prone person would say: See, if I lose, then I would not feel that bad—I would just feel that there was not much of a win in the cards anyway. But if I win, I would be pleased that I did not choose the option that would have kept me down at a much lower posterior expectation.

2

Our analysis is slightly different from Krähmer and Stone’s. They measure joy by comparing the value of a win to the posterior expectation of the action that has the highest posterior expectation. We measure joy by comparing the value of a win to the posterior expectation of the chosen action. These measures are extensionally equivalent in the case at hand, since the chosen action turns out to have the highest posterior expectation if we win, but they are not intensionally equivalent.

THE MEANING OF ‘ DARN IT !’

135

Now, as a behavioural explanation, we do not take issue with the reasonableness of these responses. However, we can do an empirical test of the plausibility of the explanation. To start with, using choice problems like the ones described above, and relying on the suggested explanation of uncertainty aversion and uncertainty proneness, we can identify among our subjects the ones who are uncertainty averse and the ones who are uncertainty prone, Now, consider a one-ball urn version of our game: the bookie picks a random natural number x from 0 to 1 under a uniform distribution and constructs a one-ball urn that has precisely x white balls and (1
x) black balls. That is, its contents are one white ball or one black ball, depending on the bookie’s random choice. Note that the posterior expectation of this game is 1 upon a win and 0 upon a loss. Suppose that we offer a subject a choice between the urn game with 1,000,000 balls or the urn game with a single ball. Now if the explanation of uncertainty aversion that has been suggested above is correct, then an uncertainty averse person should pick the 1,000,000 balls urn game: his regret upon a loss is fixed at 50 cents in both games, but his joy upon a win is 0 in the single-ball urn game and 33 cents in the 1,000,000 balls urn game. An uncertainty prone person should pick the single-ball urn game: her relief upon a win is fixed at 50 cents in both games, but her sorrow upon a loss is 0 in the single-ball urn game and 33 cents in the 1,000,000 balls urn game. We do not know whether this empirical test will hold up. If it does not, then the analysis of uncertainty aversion and uncertainty proneness along Krähmer and Stone lines is questionable. In case preparing an urn game with 1,000,000 balls in an experimental setting would be too difficult and an urn game with a single ball too contrived, no matter: we can instead offer the subject a choice between an urn game with fewer balls and an urn game with more balls. On the Krähmer–Stone-type modelling, the uncertainty averse person will prefer the more-balls urn game, whereas the uncertainty prone person will choose the fewer-balls urn game: sensitivity to uncertainty decreases as the number of balls increases. This is a rather strange implication of the model, to say the least, and one might doubt that it will stand up to experimental testing. There is a temptation to say that uncertainty neutrality is explained by being the anticipation of sorrow and joy or by the anticipation of regret and relief. In both cases the anticipated emotions balance each other out. But there is no need to do that. Relying on Broome’s (1991) account of the goodness of an action, we can just provide a rational choice explanation of risk neutral choice. Broome argues that the goodness of an action equals its expected goodness. Goodness is linear in money in the context of this chapter. Hence the rational action is the action that maximizes the expectation of the game. Since the expectations of the certain choice (i.e. 50 cents), the risky choice (i.e. the single-ball urn game), and the uncertain choice (i.e. the multiple-ball urn game) are all the same, a rational person should be indifferent between any of these options. For Broome, the goodness of an action is insensitive to the distinction

136

LUC BOVENS AND WLODEK RABINOWICZ

between certainty, risk, and uncertainty: only the expected goodness matters. Hence this is a rational choice explanation of uncertainty neutrality. By contrast, the explanations of uncertainty attitudes in terms of anticipated regret, relief, sorrow, or joy are behavioural; they carry no normative weight.

4 Regret Avoidance Therapy What do we say to a person who is overcome by regret? If we cannot convince her to give up on this emotion, we may try to steer her towards a kind of regret that is minimal for the problem at hand by framing the problem in a particular way. Consider the following examples. Suppose that we choose to play the single-ball urn game rather than the coin-toss game and we lose. Then all types of regret can kick in. But a therapist will tell us: What you didn’t choose is water under the bridge—forget about that; as to what you did choose, there was no way that you could have won; it simply was not in the cards (or, in casu, in the urn). Hence, there is no reason to regret anything. In this case, the therapist tries to convince us that the relevant way to experience regret is through chosen-action/posterior-expectation outcome regret. And then he points out that there is no regret to be had of this kind. It is the fatalistic antidote to regret. Things went as they went and, though we might not have known at the time, it was never in the cards for them to go differently. Suppose that we have an urn game in which we know that there is an equal number of black and white balls in the urn. We play this urn game rather than the coin-toss game and we lose. All kinds of regret can kick in. But now a therapist will tell us: The action that you did choose was perfectly reasonable. What else could you have done? Play the coin-toss game? That game had exactly the same expectation as the game you chose to play and you can rest assured that it did not have a higher chance of a win. Hence there is no reason to regret anything. In this case, the therapist tries to convince us that the relevant way to experience regret is through expectation action regret or probabilistic action regret. There is no regret to be had of this kind. This is the random-world antidote to regret. What we could have done was just as much a crap shoot as what we did do. Hence there is no reason to regret anything. The therapist teaches the agent how to focus regret assessment so that it disappears in the situation at hand. He teaches regret avoidance. The agent may follow. But the agent may resist and be regret seeking, by turning the therapist’s advice topsy-turvy. In the one-ball urn case, a regret-seeking agent will focus on expectation action regret or probabilistic action regret—which she does have reason to have. She will say to the therapist: Yes, but it would have been so much better to play the coin-toss game! In the equal-number-of-black-and-white-balls urn case, a regret-seeking agent will focus on chosen-action/posterior-expectation outcome regret and chosen-action/win outcome regret—which she does have reason to have. She will say to the therapist:

THE MEANING OF ‘ DARN IT !’

137

Yes, but think of the expectation of my choice and of the win that I could have had in the game that I played! Now depending on the nature of the chosen and unchosen action, one therapeutic strategy may be more successful than another. But often we don’t know what the true nature of our choices is. Then life is a Rashomon—after the 1951 Kurosawa movie in which a crime and its aftermath are recalled from differing points of view—we can frame our actions in a particular way that permits us to deal with regret. There are two such frames. Either we may frame our choices as choices between actions that have random outcomes—it’s all a crap shoot. This corresponds to Aristotle’s open future. The point of such framing is to steer us away from action regret. If things go wrong, there is no reason for regret: if we had chosen differently, things might have gone equally wrong! The accompanying virtue is epistemic modesty—we should not pretend to know how things might have gone if we had acted otherwise. Or we may frame our choices as choices between actions that are fated—actions with pre-ordained outcomes. This corresponds to the Stoic conception that the future is pre-determined. The point of such framing is to steer us away from outcome regret relative to the chosen action. If things go wrong, there is no reason for regret: things could not have gone any better—it just wasn’t in the cards! The virtue in this case is serenity—we should learn to accept what could not have been otherwise. It is curious that opposing positions on this deep metaphysical question concerning the status of future events can be called in as antidotes to retrospective emotions in response to misfortunes resulting from our past choices.

Appendix Let n be the number of balls in the urn, with x white balls and (n
x) black balls for x = 1, . . . , n. My credence of drawing a white ball, conditional on the proportion of white balls being x/n equals x/n. My credence of the proportion of white balls being x/n equals 1/(n + 1). Let W be the event of drawing a white ball and let Prop(White) be the proportion of white balls in the urn. The following calculations are based on simple algebra and the formulas for the sum of the first natural numbers and the sum of the squares of the first natural numbers.3

3

That is, P Pn nðnþ1Þ and nx¼1 x2 ¼ nðnþ1Þð2nþ1Þ : x¼1 x ¼ 2 6

138

LUC BOVENS AND WLODEK RABINOWICZ

My credence in W equals cðWÞ ¼

n  X x  x c W│PropðWhiteÞ ¼ c PropðWhiteÞ ¼ ¼ n n x¼0 n X x 1 1 1 ðn þ 1Þn 1 ¼ ¼ x¼ x¼0 n ðn þ 1Þ ðn þ 1Þn x¼0 ðn þ 1Þn 2 2

Xn

ð1Þ

A white ball yields $1 and a black ball $0; hence the expectation of the game is 50 cents. If n is odd, my credence that there are less than 50 per cent white balls in the urn is   Xðn
1Þ  x  ðn þ 1Þ 1 1 2 c PropðWhiteÞ < 0:50 ¼ ¼ c PropðWÞ ¼ ¼ x¼0 n 2 ðn þ 1Þ 2

ð2Þ

If n is even,   Xn
1  x  n 1 n 1 2 c PropðWhiteÞ < 0:50 ¼ ¼ ð3Þ c PropðWÞ ¼ ¼ x¼0 n 2 ðn þ 1Þ ðn þ 1Þ 2 which approaches 1/2 as n ! 1. We calculate the posterior credence (c*) of the proportion of white balls being x/n, upon having drawn a black ball with replacement (i.e. upon learning that B). By conditionalization and Bayes Formula:   x x  c* PropðWhiteÞ ¼ ¼ c PropðWhiteÞ ¼ jB ¼ n n  1      x x x 1
c BjPropðWhiteÞ ¼ n c PropðWhiteÞ ¼ n n nþ1 ð4Þ ¼ 1=2 cðBÞ We now calculate my posterior credence of drawing a white ball upon learning that B. This is the sum for x ranging from 0 to n of the products of the posterior credences of drawing a white ball conditional on the proportion of white balls being x/n and the posterior credences that there are x white balls in the urn:     Xn * W│PropðWhiteÞ ¼ x c* PropðWhiteÞ ¼ x c* ðWÞ ¼ c ð5Þ x¼0 n n Clearly  x x ð6Þ c* W│PropðWhiteÞ ¼ ¼ n n Hence, from (4), (5) and (6),   1  n n n 1
nx nþ1 X X X x 2 c* ðWÞ ¼ ðn ¼ x
x2 Þ ¼ 1 2 ðn þ 1Þ n n 2 x¼0 x¼0 x¼0

THE MEANING OF ‘ DARN IT !’

  2 nðn þ 1Þ nðn þ 1Þð2n þ 1Þ 1 ðn þ 1Þ
n ¼ 2 n ðn þ 1Þ 2 6 3 n

139 ð7Þ

which approaches 1/3 as n ! 1. Furthermore, if n is odd,

  1  x   Xðn
1Þ   ðn
1Þ 1
X n nþ1 x 2 2 c* PðWÞ < 0:50 ¼ c* PropðWhiteÞ ¼ ¼ ¼ 1 x¼0 x¼0 n 2  2 Xðn
1Þ x 1 ðn
1Þ ð3n þ 1Þ 2 1
¼1
x¼0 ðn þ 1Þ n 4 n 4n

ð8Þ

which approaches 3/4 as n ! 1. If n is even,   1  n x 2
1Þ 1
  X n
1   ðX n nþ1 x ð Þ 2 c* PðWÞ < 0:50 ¼ ¼ c* PropðWhiteÞ ¼ ¼ x¼0 n 1=2 x¼0 2 Xðn2
1Þ  x n 1 ðn
2Þ ð3n þ 2Þ
¼ 1
¼ x¼0 ðn þ 1Þ n ðn þ 1Þ 4 ðn þ 1Þ ð4n þ 4Þ

ð9Þ

which also approaches 3/4 as n ! 1.

References Broome, J. (1991) Weighing Goods. Oxford: Basil Blackwell. Krähmer, D., and Stone, R. (2013) Anticipated regret as an explanation of uncertainty aversion. Economic Theory 52: 709–28.

10 Keeping Things Simple Roger Crisp

We aim at simplicity and hope for truth. Nelson Goodman, Problems and Projects

1 Reasons and Parsimony In his deeply thoughtful and thought-provoking book Rationality Through Reasoning (2013),1 John Broome aims to answer several important questions, including: What are reasons? What is their relation to ought, and to rationality? Is there a logic of ought? What is rationality? Is rationality normative? How is it connected to our process of reasoning? What is the process of reasoning? What is practical reasoning in particular? When is reasoning correct? (4)

In this chapter, I shall concern myself primarily with the first five of these questions, concerning reasons, ought, and rationality, though what I say will have implications for the last four concerning practical reasoning.2 Consider the following case. Fruit-bowl. You are slightly hungry and have been offered a choice of one fruit from a bowl. It contains apricots, bananas, and cherries. You find the taste of cherries unpleasant, and that of apricots more pleasant than that of bananas. What is the best account of this apparently uncomplicated situation? According to what we might call the simple view (SV), the following claims hold: 1. Reason. You have a reason to eat the apricot, a reason to eat the banana, and a reason not to eat the cherries. 2. Pro tanto/overall. These reasons are pro tanto, not overall, reasons. 1

All unattributed page references in the text are to this book. I am grateful to Robert Audi, Ralf Bader, Derek Parfit, and the editors for helpful comments on and discussion of previous drafts. 2

KEEPING THINGS SIMPLE

141

3. Favouring. Each reason counts in favour of eating, or against eating, the relevant fruit.3 Pro tanto reasons, then, can be seen as properties of actions that count in favour of, or against, acting in some way.4 In this case, the properties in question are primarily those of the goodness (that is, the goodness for you, or the pleasantness) of eating the apricot and of eating the banana, and the badness of eating the cherries. 4. Degrees of favouring. Pro tanto reasons favour, or count against, acting to different degrees, and hence can be said to have different weights. In this case, you have stronger reason to eat the apricot than to eat the banana, and no reason to eat the cherries. 5. Overall reason. Any agent has weightiest, strongest, or overall reason to perform any action maximally favoured by her pro tanto reasons, the weight of this overall reason consisting only in the combined weight of the relevant pro tanto reasons. In Fruit-bowl, if other things are equal, you have overall reason to eat the apricot. SV is stated throughout in terms of reasons. On the view that reasons themselves are fundamental—which we might call the primacy of reasons thesis (PRT)—the simple view is itself the most basic account of Fruit-bowl. According to PRT, reasons are fundamental in at least two senses. First, the Socratic question ‘how should one live?’ is, as Bernard Williams says (1985: 4, 19), ‘the best place for moral philosophy to start’ and means ‘ “how has one most reason to live?”’.5 Williams speaks of reasons for living. The concern we have with how to live is, I suggest, primarily a practical one. We want to know how we should live because we are interested in shaping our lives through action. The basic question in ethics, then, is: which reasons for action do we have?6 Second, the notion of a reason is ‘primitive’, as T. M. Scanlon puts it: 3 In this paper I shall speak only of favouring at certain points, though the points carry across also to counting against. 4 This claim could be stated also in terms of facts, such as the fact that eating the apricot will be pleasant. Nor is any distinction required between properties of or facts about actions, on the one hand, and outcomes, on the other. If it is claimed that the outcome of F-ing is the reason to F, this claim can be restated in terms of the property of or fact about F-ing that it will issue in this outcome. 5 I am taking Williams’s claim here at face value. It may be that his talk of reasons is best understood non-normatively: see Parfit (2011: 2.433–48; 452–3). 6 I am grateful to Michael Smith for conversations which enabled me to see this. It might be thought that PRT is not neutral between certain substantive ethical views, and that this throws doubt on the claim that reasons are primary. Consider, for example, the view that morality is rights-based, and that rights, unlike reasons, are not to be weighed against other considerations. This view, like other such views, can be stated in terms of reasons: rights give an agent overriding reasons—that is, reasons that always outweigh other considerations. See note 11, below. Another issue is whether, in addition to reasons to act, there are reasons to have certain aims, attitudes, or emotions (Parfit 2011: 1.31). Because of the importance of the immediately practical question, I am inclined to see any such reasons as secondary. It may also be that they are not best understood in terms of favouring (for example, something’s being dangerous makes the response of fear, if anything, appropriate, and it is not obvious that some response’s being appropriate counts in favour of it).

142

ROGER CRISP

Any attempt to explain what it is to be a reason for something seems to me to lead back to the same idea: a consideration that counts in favor of it. ‘Counts in favor how?’ one might ask. ‘By providing a reason for it’ seems to be the only answer. (Scanlon 1998: 17; see Parfit 2011: 1.31; Broome 2013: 54)

Note that primitiveness is not the same as emptiness. This suggestion about favouring may be helpful to those who have not yet thought about what a reason is, or have formed some other view of the nature of reasons—for example, that reasons can be analysed in terms of some other more basic notion, such as good, ought, or rational requirement. Scanlon’s point is that the notion of a reason cannot be explained in terms of anything more fundamental. A principle of parsimony applies to philosophical explanations and elucidations as in all areas of human enquiry. This principle is important not only in answering philosophical questions, but in formulating those questions themselves. We should begin with questions which are fundamental (fundamental questions, then, are primary), and parsimony requires that the concepts we use in asking these questions should themselves be as basic and as few in number as possible. In this way, philosophy can address what really matters to us, and not be side-tracked into unnecessary discussion of superfluous concepts. As we have seen, the Socratic question is primarily about reasons for action, and, at least in a case like Fruitbowl, an answer can be given in which reasons play the basic normative role. Broome shows some sympathy for SV, though he would deny certain elements, expand on others, and add further claims. He denies PRT outright, claiming that ought rather than reason has normative primacy. The notion of a reason, that is to say, is explained in terms of that of ought (46), and ‘ought’ is the most fundamental normative concept (4, 11). Further, normative reasons are not best understood just in terms of favouring: the notion of explanation is also required. Accepting these claims would require, as well as giving up PRT, serious revision of SV 3 (Favouring). In the following section, I shall question the need to introduce the notion of explanation, and in Section 3 I shall argue against introducing the notion of ought, except in so far as it is shorthand for what an agent has most reason, or reason overall, to do. SV 3 also includes a claim about goodness. In Section 4, I claim that we can operate with SV’s account of what is good in outcomes, and do without Broome’s technical notion of prospective goodness or value. In Section 5, I shall turn to SV 4 (Degrees of favouring), and argue that we should not draw any distinction of the kind proposed by Broome between favouring on the one hand, and weighing on the other. In the last two sections of the chapter I shall argue against two possible additions to SV. Section 6 considers Broome’s position on ‘centrality’, and argues that introducing reasons or oughts with their source in the principles of rationality is unnecessary. In Section 7, I turn to Broome’s discussion of objective and subjective oughts.

KEEPING THINGS SIMPLE

143

I criticize his own arguments against subjective oughts, but accept that, properly understood, there is a good reason to avoid them in normative theory.

2 Reasons, Explanation, and Ought In his chapter on ‘Reasons’, Broome first distinguishes normative reasons from explanatory and motivating reasons (46–7). The latter explain why a person does something, whereas a normative reason explains why a person ought to do something.7 He then clarifies his conception of explanation (48–9). When I say that X explains Y, I am saying simply that Y is so because X is so, or that X makes Y so. He also suggests, plausibly, that there is, of any explanandum, ‘really one big explanation’, parts of which we may pick out depending on our background knowledge, interests, and so on. Broome then goes on to explanations of the fact that someone ought to F (49–51). Suppose that in some case the explanation of why you ought to F is X. Then, Broome says, X is the reason why you ought to F, because, in one sense of ‘reason’, ‘the reason why’ just means the same as ‘the explanation why’. This reason—as explanatory—is not normative (because, as we might put it, explanations give us reasons why, not reasons for). In an important passage, Broome continues: However, in ‘X is the reason why you ought to F ’, the ‘reason why’ is so closely attached to the normative ‘ought’ that the two tend to slide into each other. We slide from ‘X is the reason why you ought to F’ to ‘X is the reason for you to F ’, meaning exactly the same thing by it. The ‘reason why’ (meaning explanation) bumps into the normative ‘ought’, yielding a normative sense of ‘a reason’ that combines the meaning of both. In this sense, a reason for you to F can be defined as an explanation of why you ought to F. So we have a reason defined in terms of ought and explanation.

Broome then offers a definition of what he calls a ‘pro toto’ reason:8 Definition. A pro toto reason for N to F is an explanation of why N ought to F. And he glosses this as follows: ‘a pro toto reason for you to F is something that makes it the case that you ought to F ’.

7 Broome speaks also of reasons to believe, hope, and so on. Note also that Broome’s formulation of the distinction might appear to be giving primacy to reason over ought, through the former’s explaining the latter. Why this is not Broome’s view will be explained below. 8 I take this notion to be very close to, and perhaps identical with, that of an overall reason. Broome’s definition of a pro tanto reason is:

Definition. A pro tanto reason for N to F is something that plays the for-F role in a weighing explanation of why N ought to F, or in a weighing explanation of why N ought not to F, or in a weighing explanation of why it is not the case that N ought to F and not the case that N ought not to F. (53)

144

ROGER CRISP

This definition, so glossed, seems too broad (see Kearns and Star 2008: 49). Return to Fruit-bowl. Part of the explanation of why you ought—why it is the case that you ought—to eat the apricot is causal: your physiology is such that eating that fruit rather than the cherries will cause you to experience pleasure. But it would be odd to claim that your having this physiology is part of your reason to eat the apricot: that reason consists in the goodness—that is, the pleasantness—of eating it. With a different physiology, you might well have no reason to eat the apricot; but that is because that physiology makes it possible for you to experience pleasure. In general, what we seek in answer to the Socratic question is an abstract account of which are the ultimate reason-giving properties. One of these is pleasantness; it is not acting in such a way as to cause certain physiological events in oneself. Consider one of Broome’s own examples. Mr Reed’s being skilled at dentistry, he claims, is an explanation of why you ought to visit him, and hence a reason for doing so. But the ultimate reason-giving property of the action of visiting Mr Reed is the same as eating the apricot in the fruit-bowl case: that it will make your life better for you, that is, advance your wellbeing. Mr Reed’s being skilled at dentistry helps to explain why it is that you ought to visit him; but you have no ultimate reason to visit someone skilled at dentistry.9 That is, the property of your action that it is a visiting of someone skilled at dentistry is not an ultimate reason-giving property. Some facts that explain why someone ought to F, then, are not plausibly thought of as reasons for that person to F. But others are—because they are indeed reasons. If we restrict our attention to this latter group, then I see little for the proponent of SV to object to in Broome’s claim that, in certain contexts, ‘X is the reason why you ought to F’ is equivalent to ‘X is the reason for you to F’ (except for its use of ‘ought’, which we shall come to shortly). Asking for a normative explanation in such contexts is, in effect, asking for an account of the reasons one has to act. But this has the result that Broome’s account, when appropriately focused, is very close to SV. Consider an analogue of Scanlon’s claim about reasons and considerations that count in favour: Any attempt to explain what it is to be a reason for one to F seems to lead back to the same idea: an explanation of why one ought to F. What kind of explanation?, one might ask. The kind that provides a normative reason, seems to be the only answer.

In other words, offering a normative explanation of why someone ought to F and stating the normative reason for someone to F are very similar. And since nothing is gained by introducing the highly philosophically contested notion of explanation into our account, parsimony counts at this point in favour of SV—and in particular SV 3—against Broome’s view.

9 It might be said that the ultimate reason is the complete reason, and so includes the property of Mr Reed’s being skilled at dentistry. But this is to ignore the fact that some elements of the complete reason are more ultimate than others.

KEEPING THINGS SIMPLE

145

3 Ought Consistently with SV, the notion of ought may be used as shorthand for ‘has most reason to’, and this poses no threat to the primacy of reasons thesis.10 Broome admits that this position ‘makes sense’, if it uses the notion of ‘normative attraction’ to elucidate how whether or not one has most reason to F depends on the quantitative aggregation of normative attraction (64). But Broome has two arguments for preferring his position, which gives primacy to ought. The first is that his view is more ‘metaphysically economical’, since he does not need the notion of ‘normative attraction’. But it can plausibly be argued that normative attraction is merely another way of expressing the idea of normative weight: a consideration has greater normative attraction to the extent that it has greater weight. And if weight is the same as favouring (as I shall argue in the next section), then SV is, in this regard, no less parsimonious than Broome’s view.11 Broome’s second objection is that the ‘normative attraction’ view cannot handle cases in which you have most reason to F or ought to F, but there is no weighing explanation of why you so ought, and hence the notion of aggregation of reason cannot be put to work. Broome’s example is evidentialism, according to which it may be that you ought not to have obviously contradictory beliefs, but this is not explained by the weighing of reasons. There are several strategies available to the defender of SV here. First, one might insist that, if it is to play the part in practical reasoning that it seems to play, the reason not to have contradictory beliefs must have some determinate weight. You might always have such a reason, but in some case it might be outweighed by, say, the practical advantages to you of having such beliefs. In the case Broome is imagining, the evidentialist will be understood to be claiming that either the reason not to have contradictory beliefs outweighs any other reasons on the table (it may even be lexically weighted so that no other reasons could ever outweigh it), or that it turns out that there are no other competing reasons present. A second, and I think more successful, strategy would be to keep epistemic and practical reasons apart. ‘Believing’ is not best understood as an action, let alone the kind of action under voluntary control which is the central focus of the Socratic question and hence normative ethics in general. Epistemic reasons may lie behind certain practical reasons. For example, it may be bad, in itself, for me to be in some state I have epistemic reason not to be in, such as that of believing a contradiction. If so, then I have reasons to avoid believing

10 I am inclined to think that in any other sense (for example, in the context of the moral ought or the ought of etiquette), reason is always prior to ought, in that the question, Do I have a reason to do what I ought?, can always be asked, while the question, Ought I to do what I have a reason, or most reason, to do?, seems philosophically questionable. Normative ethics ultimately seeks reasons, not oughts. 11 Note that SV also avoids the deontic language of requiring, permitting, and so on. Cf. Broome 2013: 109–10.

146

ROGER CRISP

contradictions, and may, for example, check some of my central beliefs about the world against each other for consistency. But these reasons will have a practical weight like any others. Broome allows the evidentialist to say ‘with propriety’ that you have most reason not to have contradictory beliefs, as long as she is clear that this notion of ‘most reason’ is not ‘built from the concept of reason through aggregation’. And, he suggests, if it is not so built, it must be an independent concept: If ought is identified with most reason, it is not defined in terms of reason at all. Once again, reason seems to have no useful place in the structure of normativity. (64–5)

But it is not clear why the only concept of reason available must be straightforwardly aggregative or additive. If the evidentialist claims that you have most reason, or a decisive reason, not to have contradictory beliefs, it would be somewhat odd to deny that you have a reason not to have contradictory beliefs (though the claim that you do does not provide further information). The procedure for arriving at overall ought or ‘most reason’ judgements may not be straightforwardly aggregative;12 but presumably there is some procedure or other, and this procedure will be enough to establish the link between reason and most reason that Broome seeks. Further, if we are indeed permitted to translate ‘ought’ into ‘most reason’, there seems to be no work being done by ought independently of reason. That is, we can translate Broome’s definitions as follows: Definition. A pro toto reason for N to F is an explanation of why N has most reason to F. Definition. A pro tanto reason for N to F is something that plays the for-F role in a weighing explanation of why N has most reason to F, or in a weighing explanation of why N has most reason not to F, or in a weighing explanation of why it is not the case that N has most reason to F and not the case that N has most reason not to F. This of course again brings Broome’s account much closer to SV. Broome accepts the view that a reason is a consideration that counts in favour of something, but he believes his view has the advantage that it specifies the relevant sort of counting in favour (54). But as this section and the previous one have shown, the specification Broome offers, once appropriately narrowed, ends up close to SV, though unnecessarily complicated by the addition of the notions of ought and explanation.

12

See for example p. 127, where Broome discusses a view according to which moral requirements dominate other requirements, and contrasts it with a view expressed in terms of weighing. But note that the domination view can be expressed equally effectively by saying that moral requirements always outweigh other requirements. In general, for reasons of parsimony, if one is willing to describe any case in terms of weighing, one should at least attempt to describe all such cases in the same terms.

KEEPING THINGS SIMPLE

147

4 Favouring and Weighing In Fruit-bowl, SV captures the notion of overall reason to F in terms of favouring: the pleasantness of eating the apricot favours eating it more strongly than the lesser pleasantness of eating the banana or the cherries. According to SV, this same idea can also be stated metaphorically in terms of ‘weight’ (see Kearns and Star 2008: 42): your overall reason to eat the apricot is weightier than that of your reason to eat the banana, and you have no reason to eat the cherries (and of course a positive reason not to eat them). Broome, however, sees the two notions as different, and uses the metaphor of weight in defining the idea of a pro tanto reason (see note 8 above).The advocate of SV may immediately object that we cannot grasp the ‘for-F ’ role in a weighing explanation except through a prior understanding of the notion of favouring (54–5). Broome first responds to this objection by noting that a pro tanto reason can be defined without reference to the ‘for-F’ role: ‘[a] reason . . . is something that has a weight, where the weights of reasons combine in some way to determine whether or not you ought to F.’ But, as we have seen, the metaphor of weight can itself plausibly be seen as just another way of expressing the notion of favouring. To the extent that some consideration has more weight in a weighing explanation of some action, that consideration more strongly favours that action. So it is not that the ‘for-F ’ role itself depends on the prior notion of favouring: it is, in effect, the same as favouring. Broome then goes on to claim that ‘[t]he “for-F” role can be identified from the structure of the explanation itself ’: In a weighing explanation of why you ought to F, the for-F role is the winning one, and that is how it can be identified. Similarly, in a weighing explanation of why you ought not to F, the for-F role is the losing one. In either case, we can identify this role without calling on any prior understanding of counting in favour.

A similar point can be made again in response to Broome (see Kearns and Star 2008: 43–4). In both the winning and the losing cases, the ‘for-F ’ role is played by those considerations that favour F-ing. Winning just is favouring most strongly, and losing is favouring less strongly than other considerations. In other words, SV is not being unparsimonious in speaking of favouring as well as weighing, since these are both potentially helpful ways of seeking to elucidate a single idea.

5 Objective and Prospective Oughts SV could be developed as a form of consequentialism, according to which the reason you have to eat the apricot is that this will produce the best outcome. Broome calls this ‘outcome consequentialism’ (37). Another version of consequentialism, Broome suggests, takes the consequence of your doing an act to be the ‘prospect’ for the world

148

ROGER CRISP

that would result from your doing it, a prospect being ‘a portfolio of possible outcomes, each associated with a probability’. According to prospect consequentialism, prospects have ‘degrees of goodness’. ‘Goodness’ here can be seen as a technical term. Consider Fruit-bowl. The goodness in this case consists in the pleasure of eating the apricot. If eating the apricot is the best prospect, but on your way to the fruit-bowl you have a heart attack and die, that prospect has added nothing good to your life or to the history of the world. Prospective goodness, then, is equivalent to the value of real goodness multiplied by probability. Broome uses an example based on objective chances to illustrate the difference between outcome and prospective consequentialism (37–41). In his example, you have £35, just enough to buy your ticket home one evening. You also have the opportunity to place £1 bets on a roulette wheel. You have 35/37 chances of winning £36 overall, and 2/37 chances of losing your money. If you win, you can use your extra £1 to buy a chocolate bar. The number 20 will come up, so outcome consequentialism implies that, rather than not betting, you ought to bet on any set of numbers including 20. If we assume that not betting—which will get you home, though without chocolate—is the best prospect, then according to prospect consequentialism, you ought not to bet. Partly because his book is about rationality (22), Broome introduces the notion of practical rationality to explain the ‘central’ ought (which I shall discuss in the following section), and offers in particular the following principle: Enkrasia, very roughly. Rationality requires of you that, if you believe that you yourself ought that you F, you intend that you F. (23) Broome notes that Enkrasia applies to the central ought and the prospective ought, whereas it does not to the objective ought. Further, he is doubtful about the very genuineness of the objective ought, since it is not ‘directly practical’. This is partly because Enkrasia does not apply to it. Also, on G. E. Moore’s account of objective consequentialism, you can never know what you ought to do. Broome goes on: An ought that is not supposed to be directly practical is misleading. We mostly take ought to be practical, and in particular we assume Enkrasia applies to it. I would prefer to reject this ought altogether, and so take outcome consequentialism to be false.

By ‘practical’ here, Broome means ‘action-guiding’ or ‘practically useful’. It is now a standard view in ethics, especially consequentialist ethics, that there is a distinction between truth and usefulness. The objective ought is certainly practical in that it applies to actions rather than anything else, and claims about it may be both true and important, yet of little or no use to us. Broome uses the first person plural as well as singular in this passage, and I read him as stating his own preference, not offering an argument. The question is which ought is central, and it cannot be assumed in

KEEPING THINGS SIMPLE

149

advance that the central one is that to which Enkrasia applies. Indeed, as I shall argue below, the fact that principles of rationality can be seen as subordinate to reasons suggests that this is not a good criterion for centrality. Broome then goes on to admit that the objective ought can be defended by hindsight. When 20 comes up, you might think that you ought to have placed a bet on that number. But, he claims, even if there is a genuine objective ought, it is ‘certainly not the central ought’. Again, however, there appears to be no argument here, independently of the assumption that Enkrasia provides the criterion for centrality. It is more parsimonious to assume a single source of normativity, and Fruit-bowl suggests that this source is objective reasons rather than prospects. (It is hard to see how one could easily give up on the idea that actually eating the apricot, independently of any prospect, is good.) But what about G. E. Moore’s concern? Consider Broome’s case, and imagine that you apply a prospective consequentialist analysis and refrain from betting. Let us imagine also that you know that, had you decided to bet, you would not have placed a bet on 20 (you had a terrible time when you were twenty and this leads you to see it as an unlucky number). Then, after the spinning of the wheel, hindsight enables you to see that you used the right procedure. And, thinking about other relevantly similar cases in your life, you see that again, overall, prospective consequentialism (and its subordinate rules) has been the decision-procedure you had most reason to adopt. But there is no reason to apply it in itself, and it may be that in future you can see that you had a reason (unavailable to you) to reject it and adopt some other procedure.13

6 The Central Ought and Rationality As we have seen, Broome believes that ‘ought’ is the fundamental normative concept (11). There are different kinds of ought, and the one that plays this fundamental role he calls the ‘central’ ought (11, 22–5), and as we just saw it is important to Broome that Enkrasia applies to it. On SV, the central ought is that equivalent to ‘overall reason to F’: this is what is being asked for in the Socratic question. For Broome too, the central ought is indeed the unqualified or overall ought (26–9), and it is determined by normative ‘requirements’. These requirements may have different sources, such as morality, prudence, or rationality. Broome seems to introduce the idea of requirements partly to avoid narrowing the scope of such determining considerations to those that can play a part in weighing explanations. As we have seen that any such consideration be made to play such a part, we could speak just of

13 I am assuming that propositions about the contingent future have a truth-value. If they do not, the assumptions underlying outcome consequentialism will have to be restated. But I see no reason why this could not be achieved.

150

ROGER CRISP

pro tanto reasons here. So the SV-theorist can agree in this respect with Broome about which ought is central. Disagreement might arise, however, over the introduction of the notion of practical rationality. First, recall the primacy of reasons thesis. If we are answering the Socratic question, we will be interested in practical rationality only in so far as it provides reasons for action. If we return to Fruit-bowl, it is easy to see how Enkrasia might help you. Imagine that it is a hot day, and taking the trip to the kitchen to fetch your apricot will be slightly tiring. But you know that the pleasure of eating the apricot will be worth the pain of fetching it. So you believe that you ought—that is, have overall reason—to eat the apricot. But you cannot bring yourself to move. Then you reflect on Enkrasia, and, not wanting to be practically irrational, you go to fetch your fruit. But Enkrasia will not always be so helpful (see Southwood 2008: 14–15, 21). Consider a case in which you are under the delusion that your house is on fire and that you can fly to safety from a top-storey window. Here Enkrasia might result in your death. A more ordinary case might be a version of Fruit-bowl in which, through some minor error of memory, you believe that you will enjoy eating the cherries more than either the apricot or the banana. Here again, if reflection on Enkrasia will strengthen your will, such reflection will make you worse off. But perhaps we have ultimate reasons to live by the principles of practical rationality. These may be overridden by other reasons, as in the cases I have described. Broome believes this to be the case, though he cannot provide any argument for his view (193, 204). Lack of argument is not a problem if the claim is plausible in itself. To me, it seems not to be. If we restrict ourselves to a single-person case, the only reason I can see to accept that one always has a reason to abide by Enkrasia, or to seek to abide by Enkrasia, is that so doing will advance one’s wellbeing. But I am unable to accept that practical rationality is an element of wellbeing, let alone an element lexically prior to all others. It might be claimed that practical rationality constitutes a moral virtue (Kolodny 2005: 553–4). Again, I find this unpersuasive, except in so far as practical rationality might advance the wellbeing of the agent (or indeed others) and so be a means to a genuinely valuable end. How might the SV-theorist explain principles such as Enkrasia? One plausible way would be to see them as analogous to common-sense moral principles as understood in most versions of utilitarianism (cf. Kolodny 2005: 512, 543–4). They are helpful rules of thumb, which in many cases will enable us to do what we have independent reason to do. But they are not foolproof, and have no normative weight in themselves. Of course, principles such as Enkrasia seem to have normative weight in themselves (see Southwood 2008: 11–12). But in this respect they are analogous to common-sense moral principles such as: P: The fact that some action is the keeping of a morally unproblematic promise counts in its favour.

KEEPING THINGS SIMPLE

151

Many welfarist consequentialists—especially act utilitarians—believe that promisekeeping is valuable only in so far as it advances wellbeing, and that we can explain the development of common-sense morality, which consists of many principles such as P, as itself a wellbeing-promoting institution. Broome himself is at least sympathetic to a version of consequentialism (36–41; see Broome 2004: 41–2), so it is a little surprising that he does not seriously consider a consequentialist account of the requirements of rationality as derivative and that he believes rationality to be nonderivatively normative, especially when he tries so hard to find a good argument for its non-derivative normativity and admits his own failure to find one. This is exactly the structure of thinking that leads many consequentialists to see the principles of common-sense morality as derivative. Broome objects also to the attempt to justify rationality indirectly, as a disposition which leads its possessor to satisfy rational requirements (199–201). He begins by finding problematic the idea that you ought to have a rational disposition because it is part of the best means of your achieving much of what you ought to achieve. Broome admits the plausibility of this claim, but sees the derivation in question as requiring a principle that correctly specifies how normativity is ‘transferred’ from ends to means. And Broome knows of no such principle. Broome is right to find the means-end relationship hard to explain. Return to Fruit-bowl. You have reason overall to eat the apricot. But to eat the apricot requires taking the correct means, including picking it up. But on the face of it you have no reason in itself to pick the apricot up. If you do pick it up, and then fail to eat it, you cannot defend yourself by saying, Well, at least I did something I had reason to do. What we must recognize is that reason-giving properties can adhere to courses of action, or actions described broadly as opposed to narrowly. Broome’s problem arises from his separating ends from means, and then—rightly—finding it hard to explain how normativity is ‘transmitted’ from ends to means. Standing in front of the fruit bowl, you have a reason to eat the apricot, and that reason is the goodness consisting in the pleasantness of your eating it. But you also have a reason to pickup-and-eat the apricot, and that property is exactly the same: that it will be good for you. For the sake of argument, assume that normative egoism is true. Then, at any time, you have strongest reason to take any course of action which will maximally advance your wellbeing. There is no need to account for the transmission of normativity from ends to means, since single courses of action can be described which include both. Having assumed the transmission of normativity from ends to means, Broome then suggests that the next step in the attempt to justify rationality’s normativity indirectly is to show that you have a reason to satisfy each individual requirement of rationality. But, he claims: ‘You ought to F; if you F, your Fing will cause you to G; so you have a reason to G’ is not a valid pattern of deduction. This is obvious. Suppose you ought to take some drug to cure your

152

ROGER CRISP

serious disease. Suppose the drug has the side effect of causing you to feel unsteady. It does not follow that you have a reason to feel unsteady. (210)

First, we should note that the indirect strategy does not require showing that the agent has a reason in the case of every actualization of a rational disposition. There will, as Broome notes, be cases in which one has reason to be irrational. Nevertheless, isn’t the indirect strategy still open to Broome’s objection even if we restrict ourselves to that subset of actualizations that lead to one’s doing what one has reason to do? Consider: You ought to have a rational disposition; if you have a rational disposition, your having this disposition will cause you to satisfy certain requirements of rationality; so you have a reason to satisfy these requirements. Again, however, we can draw distinctions in a different way to avoid Broome’s objection. Continue to assume normative egoism. Then: You have overall reason to pursue any course of action which will maximally advance your wellbeing; any such course of action will consist in your developing a rational disposition, which will lead you to satisfy the requirements of rationality. In other words, describing the course of action as itself involving the satisfaction of the requirements of rationality avoids the logical error Broome rightly points out in the indirect strategy he criticizes.

7 Objective and Subjective Oughts In his third chapter, Broome considers various further alleged distinctions between different kinds of oughts, the first being that between objective and subjective oughts. Broome attributes the following view to Henry Sidgwick: Sidgwick’s View. If you believe you ought objectively to do something, you ought subjectively to do it. The reference Broome gives is to p. 207 of Sidgwick’s Methods (1907).14 It is unlikely, however, that in this passage Sidgwick is offering his own opinion. He is speaking of the view of ‘moralists who adopt the Intuitional Method’: it would, I conceive, be universally held that no act can be absolutely right, whatever its external aspect and relations, which is believed by the agent to be wrong. Such an act we may call ‘subjectively’ wrong, even though ‘objectively’ right.15 14 As Broome notes, Sidgwick himself stated his view in terms of rightness and wrongness rather than ought. 15 Note the interesting contrast with Kolodny (2005: 515, n. 6): ‘It seems generally accepted that merely believing that one has conclusive reason to X does not give one reason to X.’

KEEPING THINGS SIMPLE

153

This sentence follows the sentence in which Sidgwick mentions the intuitionists, a sentence which ends in a colon. So Sidgwick probably means, ‘universally held [by intuitionists] . . .’.16 As we can see, the intuitionists’ view includes a third kind of ought, not mentioned by Broome: what you ought to do ‘absolutely’ or ‘completely’. Subjective rightness consists in the agent’s believing her act to be right, objective rightness in that act’s being right for reasons independent of the agent’s beliefs.17 The two kinds of rightness, when present, lead to an act’s being absolutely right (that is, overall right because both subjectively and objectively right). Though Sidgwick does not use the terminology, objective and subjective oughts can be seen as pro tanto, with the absolute ought being an overall ought that emerges from the combination of an objective and a subjective ought. Consider a case in which I keep my promise to you, thinking that this is the right thing to do. An intuitionist will claim that what makes this action absolutely right is (a) its being the keeping of a promise (that is, its being objectively right, or what I objectively ought to do) and (b) my believing it to be right (that is, its being subjectively right, or what I subjectively ought to do). In other words, subjective rightness and wrongness contribute to absolute rightness and wrongness. So the view might be stated as: The Intuitionists’ View. If you believe you ought to do something, to some extent you ought to do it.18 Subjective rightness, that is to say, is a genuine form of rightness, on a par with objective rightness. Broome says that the only argument he can find for Sidgwick’s view (as he states it) is the following: Suppose you believe you ought to do something, but you do not do it. Intuition tells us you are going wrong in some way; you are not being true to your beliefs. So it seems that, if you believe you ought to do something, in some sense or other you ought to do it. But the second ‘ought’ in that sentence cannot have the same sense as the first. If it did, it would mean your belief is infallible: whenever you believe you ought to do something, you ought to do it. That cannot be so. The two ‘ought’s must therefore have different senses, which we may call ‘objective’ and ‘subjective’ respectively.

As we can now see, this cannot be an argument for the Intuitionists’ View. For that view does suggest that, in a sense, ought-beliefs are ‘infallible’: if I believe that I ought to F, then I ought to F. But this seems paradoxical only because the first ought is

Derek Parfit also pointed out to me the significance in the passage of the phrase ‘we may call’. In the first edition of the Methods (1874: 182), and in other writings, Sidgwick equates the subjective/ objective distinction with that between what the agent believes and what is really the case. According to the intuitionist view under discussion here, subjective rightness is a kind of real rightness. 18 The claim that to some extent you ought to do something—a pro tanto ought—is odd, but sense can be made of it through an analogy with the straightforward idea of a pro tanto reason. 16 17

154

ROGER CRISP

overall, while the second is pro tanto. If I believe that I ought to F, then my doing Fing will be to some extent the right thing to do.19 What then is the argument for the Intuitionists’ View? Why should we distinguish subjective and objective rightness, or subjective and objective oughts? I think the intuitionists would have claimed that their view needs no argument, since it is plausible in itself. Their view is close to what Broome calls ‘the Intuitive Premise’ in the argument for his version of Sidgwick’s view, according to which: Intuitive Premise. If you believe you ought to do something, but do not do it, you are going wrong. This statement is also elliptical. A fuller statement of what the intuitionists believe is: Intuitive Premise*. If you believe you ought to do something, but do not do it, you are—to some extent—going wrong. If one denies this claim, then one will be able to state one’s view without drawing the distinction between objective and subjective rightness or oughts.20 But Broome has provided no argument against the intuitionist position and so no argument against distinguishing between subjective and objective rightness, or against distinguishing between subjective and objective rightness or oughts on the one hand, and absolute or overall rightness or oughts on the other.21 Nevertheless, because I find the Intuitive Premise* implausible, I agree with Broome’s conclusion and would not advocate the inclusion within SV of the claim that believing that you ought to act in some way gives you any reason to do it. To conclude. Concepts are much more dangerous than most philosophers appear to believe. Even the ones we need are hard to elucidate and make precise, and the contingencies of history have left us with many which, though natural and familiar, are unnecessary; in philosophy, we would often be better off without them. In this chapter, I have argued that SV and PRT are preferable to Broome’s view, which gives pivotal roles to the notions of ought and of explanation. But this is not to say that

19 So the intuitionist can accept the Narrow-Scope View stated by Broome on p. 32: ‘If you believe you ought to do something, you ought to do it’, if that view is seen as an elliptical statement of the view that believing that you ought overall to do something makes it the case that, to some extent, you ought to do it. 20 Sidgwick himself says that his main topic must be objective rightness (1907: 208), and his key definition of utilitarianism (411) is in terms of objective rightness. In a paper on Martineau (1887: 38), Sidgwick appears to accept that it is a necessary condition of our approving of some choice that it be thought right by the chooser. But here we must remember his view that praise and blame themselves are subject to utilitarian calculation (1907: 428). Elsewhere (1905: xxxviii; see 333), he is concerned to assess the effects of agents’ beliefs about the rightness and wrongness of their own actions. I am inclined to think that, if pressed, Sidgwick would have rejected the intuitionists’ notion of subjective rightness, on the plausible ground that believing that you have a reason to F is not any kind of ultimate reason in itself for Fing. (See the reference to Kolodny 2005 in note 15, above.) 21 Note that the pro tanto/overall distinction also provides an answer to Broome’s argument against Sidgwick’s View that it leads to a conflict of oughts (33). Broome is right that allowing overall oughts to conflict is implausible. But conflicts between pro tanto oughts are commonplace.

KEEPING THINGS SIMPLE

155

thinking hard about these notions is unhelpful. I believe that the degree of philosophical convergence on positions close to SV and PRT over the last half-century may constitute a major advance in normative ethics (an advance anticipated by earlier philosophers, Sidgwick in particular).22 But it is too early to tell; for the time being, it is important to continue to consider competing proposals, and there is none more significant than that of John Broome.

References Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press. Broome, J. (2013) Rationality Through Reasoning. Chichester, UK; Malden, MA: Blackwell. Crisp, R. (2006) Reasons and the Good. Oxford: Clarendon Press. Kearns, S., and Star, D. (2008) Reasons: Explanations or evidence? Ethics 119: 31–56. Kolodny, N. (2005) Why be rational? Mind 114: 509–63. Parfit, D. (2011) On What Matters. Oxford; New York: Oxford University Press. Scanlon, T. M. (1998) What We Owe to Each Other. Harvard, MA: Belknap. Sidgwick, H. (1874) The Methods of Ethics. 1st edition. London: Macmillan. Sidgwick, H. (1887) Idiopsychological ethics. Mind 12: 31–44. Sidgwick, H. (1905) Lectures on the Ethics of T. H. Green, H. Spencer, and J. Martineau. London: Macmillan. Sidgwick, H. (1907) The Methods of Ethics. 7th edition. London: Macmillan. Southwood, N. (2008) Vindicating the normativity of rationality. Ethics 119: 9–30. Williams, B. (1985) Ethics and the Limits of Philosophy. London: Fontana.

22 In addition to Parfit and Scanlon, see also the work of Joseph Raz and John Skorupski, and the first two chapters of Crisp (2006).

11 Moral Requirements Michael J. Zimmerman

In a series of influential papers, John Broome has conducted an extensive and incisive inquiry into the nature of normativity, during the course of which he has developed a distinctive position on the nature of requirements and associated phenomena, such as reasons and ‘oughts’.1 He has focused much of his attention on the requirements of rationality, but he has also extended his findings to requirements stemming from other sources, such as prudence and morality. Although there is a great deal in what Broome has to say about moral requirements that I accept and find illuminating, there are certain points at which I believe his account could be improved, and it is on these points that I will tend to concentrate here. My chapter will be divided into four main sections. First, I will lay out Broome’s own use of two key terms, ‘require’ and ‘ought’. Second, I will explain my own use of these terms. (Our uses differ in certain respects. In some cases, the differences are purely terminological; in others, they reflect what are or may be substantive disagreements between us.) Third, I will present my account of the nature of moral requirements. Finally, I will discuss certain implications of this account, paying close attention to those that cast doubt on Broome’s view of normativity, particularly with regard to the question of under which conditions an unconditional moral requirement may be detached from a conditional moral requirement.

1 Broome’s Use of ‘Require’ and ‘Ought’ In one of his earlier (1999) papers on normativity, Broome uses ‘require’ in an unconventional way. He has since revised his usage; nonetheless, it will be worth taking a look at his earlier account, since it helps lay out the territory nicely. Broome begins with ‘ought’, saying that, although grammatically it takes an infinitive, it is useful to construe it as governing a proposition. Thus, for example, instead of saying, You ought to relax, we could as well say, You ought that you relax, if only grammar permitted our doing so. To soften the oddness of saying this, we can 1

Many thanks to Andrew Reisner and Bas van der Vossen for comments on a previous draft.

MORAL REQUIREMENTS

157

say, You ought to see to it that you relax, where the phrase ‘to see to it’ is to be regarded as semantically idle. Such ‘ought’-statements can thus be understood as having the form ‘Oq’, where q is a proposition on which ‘O’ operates. Broome notes that, if it is the case that Oq, there will be some fact that makes this be the case. Where p is the proposition that this fact obtains, he uses the phrase ‘p oughts q’ to capture the pertinent normative relation between p and q, a relation that he calls a determining relation. One consequence of this relation between p and q is the following: p!Oq (where ! is the material conditional). Broome calls this consequence the logical factor of the determining relation between p and q. Something similar can be said about reasons—normative reasons, that is: reasons that consist in considerations in favor of some proposition. We can read ‘Rq’ as ‘You have a reason (to see to it) that q’. Again, if it is the case that Rq, there will be some fact that makes this be the case. Where p is the proposition that this fact obtains, Broome uses the phrase ‘p reasons q’ to capture the pertinent normative relation between p and q, a relation whose logical factor is the following: p!Rq. The main difference between ‘O’ and ‘R’ is this. Reasons are pro tanto, and it is therefore possible both that Rq and that Rq. ‘Oughts’, however, are not pro tanto; it is not possible both that Oq and that Oq. Broome does not in fact make this claim explicitly in his earlier paper (1999), but I believe that it is implicit in what he says. Broome then makes a very important point: we must distinguish between its being the case that p!Oq or its being the case that p!Rq on the one hand (situations in which the operators have narrow scope) and its being the case that O(p!q) or its being the case that R(p!q) on the other hand (situations in which the operators have wide scope).2 It is here that he introduces his unconventional use of ‘require’. Just as the fact that p!Oq is the logical factor of the relation that obtains between p and q when p oughts q, so too the fact that O(p!q) is the logical factor of some relation between p and q. Broome calls this relation the relation of normative requirement and expresses it thus: p requires q. (In similar fashion, he uses the expression ‘p recommends q’ to express the relation whose logical factor is the fact that R(p!q).) Before explaining why I say that this use of ‘require’ is unconventional, let me just note that there seems to me to be an asymmetry between p’s oughting q and p’s requiring q that Broome overlooks. This asymmetry has to do, not with how their respective logical factors differ (something to which Broome pays a good deal of attention), but rather with how the relations of which these logical factors are a consequence differ. When it is true that p!Oq, the relevant relation between p and q consists in some fact (represented by p) making it the case that Oq. As it is often put (but not by Broome): the fact represented by p grounds the fact that Oq. When it is true that O(p!q), however, the relation between p and q of which the fact that O(p!q) is the logical factor must be of a different nature. For, just as, if it is true that

2

A relatively early statement of this point is to be found in Chisholm (1963: 33–4).

158

MICHAEL J. ZIMMERMAN

Oq, there must be some fact that grounds the fact that Oq, so too, if it is true that O (p!q), there must be some fact that grounds the fact that O(p!q). Now, the logical factor of this determining relation will look something like this: r!O(p!q). Since the fact that O(p!q) is distinct from the fact that r!O(p!q), the respective underlying relations are also presumably distinct. In particular, it seems that we cannot understand ‘p requires q’ as referring to some grounding relation between some fact and the fact that O(p!q). I am therefore uncertain just what relation it is that it does refer to and also uncertain why Broome calls it a determining relation. In any case, Broome himself (1999) notes that his use of ‘require’ differs from others’ use of this term. In particular, he contrasts his use with Chisholm’s, which, he says, expresses the determining relation whose logical factor is the fact that p!Rq. If I understand him correctly, Broome has since abandoned his earlier use of ‘requires’ and adopted Chisholm’s—with one qualification. Elsewhere (Broome 2007: 3), he says: ‘a requirement on you to F is normative if and only if it constitutes a reason for you to F.’ I take this to mean that p normatively requires q iff p reasons q—which is precisely the use of ‘requires’ that Broome attributes to Chisholm (Broome 1999: 401, n. 2). The qualification is this: whereas Chisholm implicitly assumes that all requirement is normative, Broome explicitly denies this. Some sources of requirement, he says, have no normative force at all, in the sense that they have no bearing on what you ought to do. For example, Catholicism requires you to abstain from eating meat on Fridays, and yet this may have no bearing on whether you ought not to eat meat on Fridays (Broome 2007: 3).3

2 An Alternative Use of ‘Require’ and ‘Ought’ What matters in the end, of course, is not so much what terms one uses but rather that one make clear how one is using them. Whatever terms one uses, it is important to acknowledge and accommodate the distinctions between the propositions that Broome (1999) seeks to convey by means of the pairs of expressions: ‘p oughts q’ and ‘p!Oq’; ‘p requires q’ and ‘O(p!q)’; ‘p reasons q’ and ‘p!Rq’; and ‘p recommends q’ and ‘R(p!q)’. In this chapter, I am especially concerned with certain concepts that may suitably be expressed in terms of ‘require’ and ‘ought’. Let me now try to make 3 In saying that Broome has come to adopt (this qualified version of ) Chisholm’s use of ‘requires’, I am oversimplifying what he says, since he distinguishes between this ‘source’ sense of ‘require’ and another sense that he calls the ‘property’ sense. Since both he and I are primarily concerned with the source sense of the term, I will not pursue this point. Another point should be noted here, and that is that Broome oversimplifies when he attributes to Chisholm the view that p requires q just in case p reasons q. Chisholm allows for the possibility of requirement in cases in which reasons are not implicated (whereas Broome, as far as normative requirement goes, does not). For example, at one point (Chisholm 1974: 3) Chisholm gives the example of a curve requiring a particular completion, and he doesn’t confine this to cases in which anyone might be in a position, and so have a reason, to effect this completion. But this is a minor matter. It seems clear that Chisholm would agree that, if p reasons q, then p requires q.

MORAL REQUIREMENTS

159

clear how I intend to use these terms, so that you can get a fix on the relevant concepts. As I see it, Broome’s later use of ‘requires’ is much closer to its normal use in everyday language than is his earlier use, and it is for this reason that I have called the latter unconventional. Be that as it may, I will adopt a variant of the former and say that p normatively requires q relative to S at t iff p constitutes a binding reason for S at t to see to it that q. This calls for some explanation. First, like Broome, I believe that some sources of requirement have no normative force. Just what sources lack such force is debatable, but both he and I agree that morality is not among them (Broome 2004: 4–5). Since my focus in this chapter is on moral requirements, I will henceforth drop the qualifier ‘normative’ when talking of requirements, but it should be regarded as implicit. Second, requirements, and so reasons too, are relative to persons (S) and times (t). It may be that p is a reason at one time for me to see to it that q but not a reason at another time for me to do so or a reason at any time for you to do so. Third, given that reasons are pro tanto, the correlation of requirements with reasons means that requirements are also pro tanto. Broome’s later use of ‘requires’ endorses the pro tanto nature of requirements, but his earlier use (according to which p requires q only if O(p!q)) repudiates it, since he regards, and has always regarded, ‘oughts’ as not being pro tanto. Fourth, like Broome, I take normative reasons to be considerations in favor of some proposition, but some such reasons seem to me non-binding and hence not to constitute requirements. I am not sure just how to read Broome on this matter. His claim that a requirement is normative iff it constitutes a (normative) reason commits him officially to denying what I have just said, and yet he gives examples which suggest sympathy with my view. He reports that he once advised a guest that he ought to try a mangosteen, on the grounds that mangosteens taste delicious. He immediately adds that he did not think that his guest was obliged to try a mangosteen (Broome 2004: 39–40).4 Broome’s advice seems sensible to me. I would say that, whereas obligation involves reasons that are binding, the ‘ought’ that Broome has in mind in his example does not.5 Given his official statement, Broome appears committed to saying that his guest was nonetheless required to try a mangosteen. That strikes me as incorrect, precisely because the kind of reason at issue does not factor into any obligation his guest may have had. Now as to ‘ought’: here I will part company with Broome in two ways. First, I will use ‘ought’ and ‘require’ interchangeably. More precisely, I will say that you ought to see to it that q iff, for some p, p requires you to see to it that q. In so saying, I am of course ignoring the non-binding use of ‘ought’; that is, I am treating ‘ought’ as equivalent to ‘obligated’. Moreover, I am supposing that there are two 4 5

Cf. Chisholm (1964: 152–3). I assume that, in this context, ‘obliged’ means the same as ‘obligated’.

160

MICHAEL J. ZIMMERMAN

kinds of obligation. First, there is pro tanto obligation, where the (binding) reason that is constituted by p is liable to being overridden or counterbalanced by some contrary reason, r, that requires you to see to it that q. (If there is such an r, then the requirement imposed by p is merely pro tanto, as is the associated ‘ought’.) Second, there is all-things-considered obligation, where the reason that is constituted by p is conclusive or decisive, in that it overrides any such r. (Elsewhere (2004), Broome calls such a reason a ‘perfect’ reason.6) As I have noted, Broome does not use ‘ought’ in this liberal way; he restricts it to those cases in which, the weight of reasons being such as to favor seeing to it that q, there is a decisive reason to do so.7 That is, he uses ‘ought’ only in the sense of ‘ought all things considered’. I see no reason to use the term so restrictively. Philosophers and laypeople alike often use it in the more liberal way that I will adopt, and I see no reason to eschew such usage, as long as one is clear to distinguish ‘ought pro tanto’ from ‘ought all things considered’. Second, there is the question of how to distinguish moral ‘oughts’ from other kinds of ‘oughts’. Broome calls morality a source of requirements and ‘oughts’, and the metaphor seems apt. Perhaps he would agree that, if you ought (all things considered) to see to it that q and morality is the source of this fact, then the ‘ought’ is a moral one; that is, you ought morally to see to it that q. But so saying leaves an important question unsettled, one that I find difficult to articulate. One way of putting the question is this. When you ought morally to see to it that q, is the ‘ought’ internal or external to morality? I expect that, as it stands, this question is not very clear, so let me try to explain. Suppose that you have various normative requirements to see to it that q, that these requirements issue from various sources, A, B, and C, and that each of these requirements overrides any contrary requirements. (A may be morality, B may be prudence, C may be . . . well, I will leave that open.) Then, Broome would say, you ought to see to it that q. I have suggested that he might agree that you ought A-ly to see to it that q, B-ly to see to it that q, and C-ly to see to it that q. Now, how many ‘oughts’ are at issue here? There seem to me to be (at least) three reasonable answers to this question: one, three, and four. According to the first answer, each of A, B, and C contributes to the fact that there is an ‘ought’ that applies to you to see to it that q, one that is external to and transcends the sources from which it emanates. We might say that, on this view, you ‘ought period’ or ‘ought full stop’ to see to it that q, and that this is the only kind of ‘ought’ there is. To say also that you ought A-ly, B-ly, and C-ly to see to it that q is not to say that there are other ‘oughts’ that apply to you beyond ‘ought period’; it is just another, and perhaps somewhat misleading, way of saying that the fact that you ought to see to it that q has its source in A, B, and C. 6

Broome 2004: 34 ff. This is actually an oversimplification, since Broome allows for cases in which what one ought to do does not involve a ‘weighing explanation’. (See Broome 2004: 36 ff.) I will ignore this complication here. 7

MORAL REQUIREMENTS

161

The second answer is quite different. According to it, each of A, B, and C issues in its own ‘ought’. You ought A-ly, B-ly, and C-ly to see to it that q, but there is no point of view that transcends these sources and that issues in an ‘ought period’. The third answer is a compromise between the first two. According to it, each of A, B, and C issues in its own ‘ought’, but they also combine to issue in an ‘ought period’. What I have just said may still be pretty unclear, but I hope it will do. As I understand him, Broome embraces the first answer,8 whereas I am inclined toward the second. But I think this difference between us need not detain us. I have said that my concern in this chapter is with the nature of moral requirements, and, although I believe the question of which (if any) of the three answers just recorded to accept is an important one, we need not settle the matter here. We can finesse the problem simply by stipulating that, henceforth, all cases in which you are required to see to something are cases in which morality is the only source of this fact.

3 An Account of the Nature of Moral Requirements So far, whatever differences between Broome and me that have emerged are minor. Some are purely terminological (e.g. whether to use ‘ought’ to express pro tanto requirement), while others can be put to one side (e.g. whether there are non-binding reasons, or whether there is a distinctly moral ‘ought’). More substantial differences will suggest themselves, however, once I have given my account of the nature of moral requirements. I will flag these as and when they arise. The account will be very sparse, containing only those details necessary for deriving the implications to be discussed in the next section. (I have provided further details elsewhere.9) On my view, you can be morally required, whether all things considered or merely pro tanto, to do something only if you have a choice about whether to do it. The choice you make will be one that concerns not just the immediate future but also the future beyond that; the choice will set you on a certain path and cut you off from other paths you could have gone down. Of course, in almost all cases somewhere down the path you will arrive at another juncture where you will have another choice to make, one that will set you on a certain sub-path and cut you off from other subpaths. And so on. Another, familiar way of capturing this idea is to say that, whenever you make a choice, you access or actualize some possible world and thereby render other possible worlds forever unactualized. 8

Cf. Thomson 2008: chs. 10–11. See Zimmerman 1996: chs. 2, 4, 2008: chs. 2–3. The account that I give there of the structure of overall (i.e. all-things-considered) moral obligation is very heavily indebted to the account given in Feldman (1986: ch. 2). The account is basically the same in both works, except that in the later work I advocate what I call a ‘prospective’ view of obligation. This difference between the accounts is not relevant to the present discussion. The account of prima facie (i.e. pro tanto) obligation that I give in the earlier work is much more complicated than the one I give in the later work. What follows reflects the later, simpler account. 9

162

MICHAEL J. ZIMMERMAN

We may think of possible worlds as maximal propositions. This fits well with Broome’s construal of ‘ought’. Suppose that you ought to relax. According to Broome, this amounts to its being the case that you ought (to see to it that the proposition) that you relax (is true). Let us abbreviate this as: Oq. (Let me stress that, in keeping with the foregoing discussion, ‘Oq’ is here to be understood to mean that you have a moral obligation, either all things considered or merely pro tanto, to see to it that q.) On my view, it can be the case that Oq only if both Cq and Cq, that is, only if you can both see to it that q and see to it that q. Moreover, it can be the case that Cq only if there is a possible world, w, such that w is accessible to you (i.e. Cw) and q is true at w (i.e. w entails q). Now, suppose that you have a choice between relaxing (q), eating (r), and exercising (s), such that Oq but Or and Os. Then, as Broome notes, there must be some fact that makes it the case that q is to be chosen (on that occasion) rather than either r or s. As I will put it, this fact renders q (on that occasion) ‘deontically superior’ to each of r and s. Moreover, this can be the case only if any accessible world in which (on that occasion) you either see to it that r or see to it that s is such that there is some deontically superior accessible world in which (on that occasion) you see to it that q and do not either see to it that r or see to it that s. In brief (and roughly but, I think, sufficiently precisely for present purposes): necessarily, Oq iff any accessible q-world is deontically inferior to some accessible q-world. Again, this account encompasses both all-things-considered and pro tanto moral obligation. In the case of all-things-considered obligation, whether one world is deontically superior to another will be a function of all the propositions that are true at those worlds that have a ‘deontic bearing’ on your choice. (Different substantive moral theories will give different accounts of which propositions have such a bearing. Classical utilitarianism will identify propositions having to do with pleasure and pain as the relevant propositions. Classical Kantianism will fix on propositions having to do with the universalizability of maxims or respect for persons. And so on.) In the case of pro tanto obligation, however, deontic superiority will be a function of only some such propositions. Let me use W. D. Ross’s well-known example to illustrate.10 Suppose that you have made a promise to meet someone (p) that requires you pro tanto to meet him (q), but that en route you run across the victim of an accident whose condition (r) requires you pro tanto to tend to his injuries (s). Suppose, further, that these requirements conflict and that your obligation all things considered is to tend to the injuries and miss the meeting. Then we may say that q is deontically superior to s relative to the reason constituted by p, s is deontically superior to q relative to the reason constituted by r, and s is deontically superior to q relative to the totality of the (binding) reasons that pertain to your choice.

10

W. D. Ross 1930: 18.

MORAL REQUIREMENTS

163

We may of course also say the following: necessarily, Oq iff any accessible q-world is deontically inferior to some accessible q-world; and, necessarily, O(q& r) iff any accessible (q& r)-world is deontically inferior to some accessible (q& r)-world. A corresponding account can be given of what I will call conditional moral obligation proper. We often have occasion to say something along the lines of, You ought to see to it that q if p. Now, this could be construed in a number of different ways. We have already seen two of these ways: p!Oq, and O(p!q). What I have in mind by ‘conditional moral obligation proper’ is a third way, which I will represent as follows: O(q/p). And my account of such obligation is this: necessarily, O(q/p) iff any accessible (q& p)-world is deontically inferior to some accessible (q& p)-world.11 I will give illustrations of such obligation in due course.

4 Implications of the Account In this section I will discuss several implications of my account of the nature of moral requirements. (I will leave it to you to judge whether these implications are assets or liabilities.) Let me begin, however, by noting two non-implications of this account. First, my account does not imply that consequentialism, in any standard sense of the term, is true. All substantive moral theories, whether consequentialist or not, provide an account of when and why it is that one option is to be chosen over, or preferred to, another, and it is simply this kind of preferability that I call deontic superiority. Second, my account does not imply that supererogation is impossible. If you have a choice between two options each of which it would be morally permissible for you to choose, then neither option is deontically superior to the other. If, nonetheless, one of these options would be supererogatory and the other not, then the former would indeed be superior to the latter in some way, but not deontically.12 Let me now turn to some implications that my account does have. A. One obvious implication of my account is that ‘ought’ implies ‘can’, that is: hðOq ! CqÞ

ð1Þ13

I have not attempted to spell out the nature of the relevant ‘can’, nor will I, except to say that it is supposed to signify a kind of personal possibility that goes beyond mere logical or physical possibility and is roughly equivalent to the agent’s having both the capacity to do something and the opportunity to exercise that capacity. Many philosophers accept that ‘ought’ implies this sort of ‘can’, but of course some philosophers deny this. Broome may be among the deniers. At one point he says 11 This account is heavily indebted to the account given in Feldman (1986: ch. 4), which is itself indebted to that given in Lewis (1973: 100 ff.). 12 See Zimmerman 1996: ch. 8. The non-deontic superiority will involve reasons that are non-binding. 13 ‘□’ signifies metaphysical necessity.

164

MICHAEL J . ZIMMERMAN

this: ‘a psychological failing is often not enough to excuse you from an ought. Even if you cannot bring yourself to give up cream cakes, still it may be that you ought to give them up’ (2001: 188). I have defended the thesis that ‘ought’ implies ‘can’ at length elsewhere. In that discussion, I did not consider a case involving the inability to give up cream cakes, but I did discuss similar cases. And here is what I have to say in response.14 First, my account is an account of ‘ought morally’. I acknowledge that there are ‘oughts’ that have their source in something other than morality and which do not imply ‘can’. It may be that Broome has some such ‘ought’ in mind in his example, for he immediately goes on to add: ‘Even if you cannot bring yourself to believe you are mortal, still you ought to believe it.’ It would be odd to think of the kind of psychological failing at issue here as a moral failing. Second, however, some people may of course still regard the failure to give up cream cakes in particular as a moral failing, at least on some occasions, even if it is something that on those occasions you can do nothing about.15 So saying is in fact consistent with my view, if the failure is one that you could have done something about. We should distinguish between the time, t, at which an agent is obligated and the time, t 0 , of the action that that agent is obligated to perform. In some cases t may be the same as t 0 ; if so, we may call the obligation in question ‘immediate’. But in other cases t will precede t 0 ; in such cases, we may call the obligation ‘remote’. When I say that ‘ought’ implies ‘can’, what I mean, more precisely, is this: necessarily, you ought at t to see to it at t 0 that q only if you can at t see to it at t 0 that q. This means that, although immediate obligation requires immediate performability, remote obligation only requires remote performability. Still, remote obligation is a form of obligation, the failure to fulfill which can constitute wrongdoing. If on Tuesday you cannot resist your craving for cream cakes, but you could on Monday have forestalled its irresistibility, then your succumbing to this craving on Tuesday may well be a moral failing on your part. Third, though, some may regard your succumbing to this craving as a moral failing even if it is something that it was never in your power to resist. Even this is consistent with my view, as long as your succumbing is regarded as a failure to live up to some moral ideal (whose associated ‘ought’ is non-binding and so does not express a moral requirement) as opposed to a moral obligation. It may be that Broome has this kind of moral failing in mind. But it also may well be that he rejects (1). If so, we have arrived at what I will call Substantive Disagreement #1.

14

Cf. Zimmerman 1996: 1–2, 90–3, 96 ff. I am assuming that it is indeed strictly true that you cannot give up cream cakes. If it is merely (merely!) difficult for you to resist their allure rather than impossible for you to do so, then my account does not imply that you have no moral obligation to do so. Of course, the distinction between difficulty and impossibility is itself difficult to pin down. 15

MORAL REQUIREMENTS

165

B. Another implication of my account is this: hðOq ! O  qÞ

ð2Þ16

Broome notes correctly that some sources may issue conflicting requirements (Broome 2007: 22), but in private correspondence he has indicated that he is inclined to think that normative requirements that stem from the same source cannot conflict in the manner presented in (2). Since (2) concerns moral requirements, which are normative, I suspect that Broome accepts it. C. Another implication of my account is this: hf½Oq&  Cðq&  r ! Org

ð3Þ17

Broome challenges a similar claim (Broome 2007: 19–20), and in this case, I believe, he would indeed apply the challenge to moral requirements as well as other kinds of requirements. He cites Ross’s paradox as a reason for rejecting the claim and, although he couches his discussion in terms of what prudence requires, it is easily re-cast in terms of what morality requires. Here, then, is Substantive Disagreement #2. The challenge is this.18 Suppose that you ought (morally) to post a letter. Now, you cannot post a letter without either-posting-it-or-burning-it. From (3) it follows that you ought therefore to either post the letter or burn it. But this implies in turn that, if you burn the letter, you do something that you ought to do. Since this implication is unacceptable, (3) must be rejected. Here is my response.19 Sentences of the form, You ought to see to it that either q or r, are ambiguous. One reading is this: Oðq _ rÞ

ð4Þ

This, I submit, is not what such a sentence typically means. On the contrary, what is typically meant is this plus the claim that it is morally up to you whether to see to it that q or to see to it that r, as long as you see to at least one of these. That is: Oðq _ rÞ&  O  q&  O  r

ð5Þ

The two extra conjuncts make a big difference, of course. Now, regarding (4) in particular: if it is the case that Oq, then, it seems plain to me, it is also the case that O (q_r); equivalently, if it is the case that O(q), then it is the case that O(q & r). From this it of course does not follow that Or (although this is an implication of (5)), let alone that Or. It is therefore highly misleading to say that ‘you do 16 Regarding pro tanto obligation, I need in fact to state matters more carefully, as follows: it cannot be the case both that p requires you to see to it that q and that p requires you to see to it that q. (Cf. Chisholm 1974: 7.) After all, it is perfectly possible both that p requires you to see to it that q and that r requires you to see to it that q. Similar qualifications apply to some of the other formulas below. 17 Actually, the claim that I accept is this: □{[Oq & C(q& r) & Cr] ! Or}. This is because, as indicated in Section 3 above, I believe that Or implies Cr. (See Zimmerman 1996: 69–70.) We need not worry about this qualification here. 18 19 The original source is A. Ross (1941). Cf. Zimmerman 1996: 72–3.

166

MICHAEL J. ZIMMERMAN

something that you ought to do’ if you burn the letter, since this strongly suggests that it is the case that Or. A second challenge to (3) concerns the Good Samaritan paradox. Since Broome does not mention this challenge, I will not address it here. I have done so elsewhere.20 There is a third challenge to (3) that I do want to address here, though, since I have not addressed it before and it is a challenge to which Broome himself has alluded in discussions that I have had with him. But I will postpone addressing it to section (F) below, once I have dealt with certain related preliminary matters. D. Let me now turn to the matter of detaching an unconditional moral requirement (of the form ‘Oq’) from a conditional moral requirement (of the form ‘O (p!q)’). There are three types of detachment that I will address: factual, deontic, and forced. i. Factual detachment is expressed in the following formula: hf½p & Oðp ! qÞ ! Oqg

ð6Þ

Much of what Broome has to say about requirements concerns the distinction between narrow-scope statements of the form ‘p!Oq’ and wide-scope statements of the form ‘O(p!q)’. In a number of places he points out that, although we should accept hf½ p & ðp ! OqÞ ! Oqg

ð7Þ

it does not follow that we should accept (6). This is of course absolutely correct. But Broome also makes a stronger claim, namely, that we should reject (6). I am sure that he is right about this. Here is one argument for rejecting factual detachment.21 Suppose that there are three candidates for office, Tom, Dick, and Harry. Let t be the proposition that you vote for Tom, d be the proposition that you vote for Dick, and h be the proposition that you vote for Harry. Suppose that, deontically, t is superior to d, which in turn is superior to h. Then a number of facts follow. One is this: Ot

ð8Þ

Oð t ! dÞ

ð9Þ

Another is this:

Suppose now that, although Tom is the best candidate, you have a personal animus toward him and refuse to vote for him. Thus:

20 21

See Zimmerman 1996: 73–5. Cf. Zimmerman 1996: 116–17. The example is borrowed from Jackson and Pargetter (1987: 77–8).

MORAL REQUIREMENTS

167

t

ð10Þ

Od

ð11Þ

Should we therefore conclude

Certainly not. As I say elsewhere: To say that you ought to vote for Dick just because you won’t vote for Tom is to make things either impossibly difficult for you or ridiculously easy. It makes things impossibly difficult if the claim that you ought to vote for Tom is retained; for, on the assumption that you cannot both vote for Tom and vote for Dick, you would then be faced with a moral dilemma (and, whatever one thinks about the theoretical possibility of such dilemmas, surely no one thinks that they are so easily generated . . . ). It makes things ridiculously easy if the claim that you ought to vote for Tom is abandoned; for you cannot change the fact that you ought to vote for Tom simply by not voting for him.22

We should therefore reject factual detachment. (Or, to put matters more cautiously and precisely: we should therefore reject factual detachment for non-subsidiary moral obligation. I will clarify and defend this qualification in section (F) below.) Note, moreover, that such rejection is validated by my account, since it certainly is not the case that voting for Dick is somehow magically transformed from being part of a deontically second-rate accessible world to being part of a deontically first-rate accessible world by your refusal to vote for Tom. Even though you will not vote for Tom, you still can and still ought to do so. Your refusal to vote for him does not make it the case that you ought to vote for Dick. On the contrary, it remains the case that you ought not to do so. ii. Deontic detachment is expressed in the following formula: hf½Op & Oðp ! qÞ ! Oqg

ð12Þ

Broome considers and rejects a very similar claim (Broome 2007: 17–19). His discussion has to do with the requirements of prudence, but it can easily be re-cast in terms of the requirements of morality. Hence I infer that he rejects deontic detachment as well as factual detachment. We have now arrived at Substantive Disagreement #3, for I part company with Broome on this matter. My account validates (12). Given that it is the case that Op, then, on my account, any accessible p-world will be deontically inferior to some accessible p-world. Moreover, given that it is the case that O(p!q) (i.e. that O (p& q)), then any accessible (p& q)-world will be deontically inferior to some accessible (p& q)-world. Jointly, these two facts imply that any accessible qworld will be deontically inferior to some accessible q-world.

22

Zimmerman 1996: 117.

168

MICHAEL J. ZIMMERMAN

Why does Broome reject deontic detachment? Consider a case of a sort first introduced (as far as I know) by Roderick Chisholm23 and raised by Broome himself in discussions that I have had with him. Let v be the proposition that you visit your mother in hospital and c be the proposition that you call ahead to tell her you are coming. Suppose that Ov

ð13Þ

Oðv ! cÞ

ð14Þ

Oc

ð15Þ

and

Deontic detachment of course yields

But now suppose that it is also the case that Oð v ! cÞ

ð16Þ

since, after all, it would be cruel—it would add insult to injury—to dash your mother’s expectations by calling ahead and not following through. And suppose, finally, that, because you are lazy, v

ð17Þ

Given both (16) and (17), shouldn’t we reject (15) after all and therefore reject deontic detachment? I grant that, faced with (16) and (17), accepting (15) may seem odd, at least at first, but I submit that we should do so nonetheless. Notice that Broome himself is not willing to claim that Oc

ð18Þ

Inferring (18) from (16) and (17) would be to endorse factual detachment, which Broome rejects—and appropriately so, for reasons just discussed. But if (18) is to be rejected and (15) too, what remains? Presumably this: Pc & P  c

ð19Þ

where ‘P’ denotes permissibility. But (19) is decidedly odd. Consider its first conjunct, Pc

ð20Þ

If (15) is unacceptable, how is (20) any more acceptable? If the cruelty of calling ahead and not following through is sufficient under the circumstances to make it the case that you are not obligated to call ahead, it would also seem sufficient to make

23

Chisholm 1963: 34–5.

MORAL REQUIREMENTS

169

it the case that you are not permitted to call ahead. Now consider the second conjunct of (19), Pc

ð21Þ

If (18) is unacceptable, how is (21) any more acceptable? If laziness on your part is insufficient under the circumstances to make it the case that you are obligated not to call ahead, it would also seem insufficient to make it the case that you are permitted not to call ahead. No, (19) will not do, and since, as Broome himself says, (18) also will not do, we must accept (15) after all. I recognize that accepting (15) might still sit uneasily with you, though. I will try to mitigate this unease in section (F) below. iii. Forced detachment is expressed in the following formula: hf½ C  p & Oð p ! qÞ ! Oqg

ð22Þ

Once again, my account validates this claim. Given that it is the case that Cp, then p will be true at every accessible world. Moreover, given that it is the case that O (p!q), then, on my account, any accessible (p& q)-world will be deontically inferior to some (p& q)-world. Jointly, these two facts imply that any accessible q-world will be deontically inferior to some accessible q-world. It is my understanding from private correspondence that, although there may be contexts in which Broome is willing to accept forced detachment, he is not prepared to accept it in all. I believe, for example, that he would not accept that you ought to eat exactly one cream cake on the grounds that you cannot bring yourself to eat none and that, if you do not eat none, you ought to eat just one. If so, we have reached Substantive Disagreement #4. Notice that, if the moral ‘ought’ did not imply ‘can’, there would seem to be good reason to reject (22). To see why, consider the following principle: hf½O  p & Oð p ! qÞ ! Oqg

ð23Þ

This principle is clearly unacceptable. Suppose, for example, that you ought not to injure your neighbor but that, if you do injure him, you ought to injure him only mildly. We should not infer that you ought to injure him mildly. On the contrary, you ought not to injure him at all. But now suppose that you cannot avoid injuring your neighbor, although you can avoid injuring him severely. (He has stepped out suddenly in front of your car. You have time to swerve and give him only a glancing blow, but unfortunately you cannot miss him altogether.) If ‘ought’ didn’t imply ‘can’, then perhaps it would still be true that you ought to avoid injuring your neighbor, in which case we should presumably once again deny that you ought to injure him mildly—even if it is also true that, if you do injure him, you ought to injure him mildly. But, as I have said above, I take it that the moral ‘ought’ does imply ‘can’. In the illustration just given, I believe we should indeed say that you ought to injure your

170

MICHAEL J. ZIMMERMAN

neighbor mildly (rather than severely). This is precisely what forced detachment implies and, on the assumption that ‘ought’ implies ‘can’, I can see no good reason to reject such detachment. On the contrary, it seems intuitively very plausible. E. I want now to consider the distinction between statements of the form ‘O(p!q)’ and statements of the form ‘O(q/p)’. Broome does not couch his discussion of conditional obligation in terms of the latter, but I think that in many, perhaps most, cases he would be better advised to do so. This fact constitutes Substantive Disagreement #5. As I have noted above, my account says the following. First, necessarily, O(p!q) iff any accessible (p& q)-world is deontically inferior to some accessible (p& q)world. Second, necessarily, O(q/p) iff any accessible (p& q)-world is deontically inferior to some accessible (p& q)-world. Since every (p& q)-world is a (p& q)world, but not vice versa, the following holds: O(q/p) only if, but not if, O(p!q). The fact that what I call conditional obligation proper holds in a proper subset of the set of cases in which the kind of conditional obligation that Broome discusses holds has some interesting consequences. Here are some. i. First, as is easily verified, factual detachment fails, while deontic detachment and forced detachment hold, for statements of the form ‘O(q/p)’ just as they do for statements of the form ‘O(p!q)’, and for the same reasons. ii. Second, the two kinds of conditional obligation diverge in significant ways in some cases. For an example, return to the case of Tom, Dick, and Harry. Suppose that your options are to be ordered, deontically, as shown in Diagram 11.1.

1. t

& ∼d & ∼h

2. ∼t & d

& ∼h

3. ∼t & ∼d & h 4. ∼t & ∼d & ∼h Diagram 11.1

This diagram represents the following facts, among others. First, you can vote for each of but at most one of Tom, Dick, and Harry. Second, any accessible t-world is deontically inferior to some accessible t-world. Third, any accessible (t& d)-world is deontically inferior to some accessible (t& d)-world. And fourth, any accessible (t& d& h)-world is deontically inferior to some accessible (t& d& h)-world. In light of these facts, my account implies the following: Ot

ð8Þ

Oðd=  tÞ

ð24Þ

Oðh=  t &  dÞ

ð25Þ

MORAL REQUIREMENTS

171

All of this seems right. You ought to vote for Tom; if you do not vote for Tom, then you ought to vote for Dick; if you do not vote for either Tom or Dick, then you ought to vote for Harry. However, as noted above, my account also has the following implication: Oð t ! dÞ

ð9Þ

So the question arises whether the claim that, if you do not vote for Tom, then you ought to vote for Dick, is better represented by (24) or by (9). The answer is clear, I think: (24). Why? Because (9) follows from (8) alone. (Or rather, (9) follows from (8) conjoined with hf½Oq &  Cðq &  rÞ ! Org

ð3Þ

Since Broome rejects (3), he might not be impressed by this argument in favor of (24) over (9).) But we surely should not say that, just because you ought to vote for Tom, it is therefore also the case that, if you do not vote for Tom, then you ought to vote for Dick. Another way of putting this point is to note that a further implication of (8) (conjoined with (3)) is this: Oð t ! hÞ

ð26Þ24

But we should deny that, if you do not vote for Tom, then you ought to vote for Harry. My account of conditional obligation proper validates such a denial, since, as Diagram 11.1 indicates, it is not the case that any accessible (t& h)-world is deontically inferior to some (t& h)-world. iii. Finally, my account of conditional obligation proper provides an answer to a puzzle posed by Broome regarding a certain kind of asymmetry (Broome 2007: 35–7). He notes that wide-scope requirements are symmetric, in the sense that they accommodate the contraposition of logical equivalents. Thus, for example, Oðp ! qÞ

ð27Þ

Oð q ! pÞ

ð28Þ

is equivalent to

Yet in some cases it seems correct to affirm a statement of the form of (27) and yet to deny the corresponding statement of the form of (28). Here is such a case. (It is not a case that Broome discusses since, in keeping with the theme of this chapter, it concerns moral requirements, whereas, when he poses his puzzle, Broome does so in terms of the requirements of rationality.) You are in a pastry shop and you have the following options, among others: to take a certain item (a cream cake, perhaps) and to pay for that item. Let t be the proposition that you 24

This echoes a point made in Chisholm (1963: 34).

172

MICHAEL J. ZIMMERMAN

take the item and p be the proposition that you pay for it. Suppose, now, that your options may be more fully represented by the deontic ordering shown in Diagram 11.2.

1. t &

p

2. t & ∼p 3. ∼t &

p

4. ∼t & ∼p Diagram 11.2

Then my account implies that Oð p=tÞ

ð29Þ

Oð t=  pÞ

ð30Þ

Oðt=  pÞ

ð31Þ

But it does not imply that

On the contrary, it implies that

Here is a very short story that provides partial support for the deontic ordering just given and the statements of conditional obligation proper derived from it: your child is in desperate need of food, for which you can afford to pay.25 F. I said above that we should reject factual detachment for non-subsidiary moral obligation. I believe that it nonetheless holds for subsidiary obligation. Let me explain. Consider once again the case of Tom, Dick, and Harry, represented in Diagram 11.1. Since ‘d’ is featured on line 1, the diagram accurately reflects the fact that Od

ð32Þ

I want now to qualify this claim by saying that the diagram accurately reflects the fact that you have a primary obligation not to vote for Dick. That is: O1  d

ð32′Þ

Saying this allows for the possibility that you have a non-primary, that is, a subsidiary obligation to vote for Dick. How so? Well, if you do not vote for Tom, then you ought to vote for Dick. Suppose now that you do not vote for Tom. Should we then infer that Od

25

I first presented this case in Zimmerman (1996: 123).

ð11Þ

MORAL REQUIREMENTS

173

Earlier, in agreement with Broome, I answered this question by saying, ‘Certainly not’, and I gave my reasons for saying this. I stand by those reasons, but in fact my answer was too sweeping. Here is what I really want to say. From your failure to vote for Tom, we may detach the subsidiary obligation to vote for Dick. In this case, the subsidiary obligation is, more precisely, a secondary obligation, since ‘d’ is featured on line 2 of Diagram 11.1. More precisely still, you have (a) a secondary conditional obligation to vote for Dick if you do not vote for Tom. That is: O2 ðd=  tÞ

ð24′Þ

And if (but only if ) you do not vote for Tom, you then indeed have (b) a secondary unconditional obligation to vote for Dick. That is: O2 d

ð11′Þ

Why say this? Well, first, despite the very good reasons given above for rejecting the claim that you are obligated to vote for Dick simply in virtue of not voting for Tom, there is nonetheless, I believe, considerable pressure to affirm this very claim. By distinguishing between primary and subsidiary obligation, we can have our cake and eat it too. We can deny (11) (understood as the claim that O1d) and still affirm (110 ). And, second, this way of framing matters helps account for the fact that, if you do not vote for Tom and do not vote for Dick, then there is a sense in which you do a double wrong; you compound the primary wrongdoing involved in not voting for Tom. This account of subsidiary obligation is of course very sketchy. I have provided further details elsewhere.26 Still, even in the absence of these details, we can see how it can help mitigate the unease that might have lingered after our discussion of deontic detachment and the case involving your ailing mother. That case may now be more perspicuously, even though still partially, represented as shown in Diagram 11.3.

1. c

& v

2. ∼c & v 3. ∼c & ∼v 4. c

& ∼v

Diagram 11.3

In light of this diagram, we may now affirm all of the following:

26

O1 n

ð13′Þ

O1 c

ð15′Þ

O3 ð c=  nÞ

ð16′Þ

Zimmerman 1996: 128 ff.

174

MICHAEL J. ZIMMERMAN

Given factual detachment for subsidiary obligation, (160 ), in conjunction with n

ð17Þ

O3  c

ð18′Þ

warrants also saying The availability of (180 ) should, I think, help overcome any hesitancy you may have had to affirm (150 ). (180 ) helps explain how it is that, if indeed you do call ahead and then fail to follow through, you will have compounded your wrongdoing—you will, as I put it before, have added insult to injury. My account of subsidiary obligation also handles a challenge to hf½Oq &  Cðq &  rÞ ! Org

ð3Þ

that I did not discuss earlier, but which I believe Broome would endorse. Consider a variation on the case involving your ailing mother in which for some reason you cannot visit her without calling ahead. Given that you are in fact not going to visit her, it may, as before, seem doubtful that you ought nonetheless to call ahead; yet this is precisely what (3) implies (since you ought to visit her and cannot do so without calling ahead). The solution is straightforward. The diagram that is applicable to this variation of the case—Diagram 11.4—is the same as Diagram 11.3 with its second line removed. In light of Diagram 11.4, we may now affirm all of the following:

1. c & v 2. ∼c & ∼v 3. c & ∼v Diagram 11.4

O1 n

ð13′Þ

O1 c

ð15′Þ

O2 ð c=  nÞ

ð16′′Þ

Given factual detachment for subsidiary obligation, (1600 ), in conjunction with n

ð17Þ

O2  c

ð18′′Þ

warrants also saying 00

As before, the availability of (18 ) should help overcome any hesitancy you may have had to affirm (150 ).

MORAL REQUIREMENTS

175

G. I wish, finally, to address a distinction that has been much discussed in recent years and which Broome has himself addressed on a number of occasions. I will introduce it by quoting a well-known passage written by A. C. Ewing (1948: 121): it is clear that a person may make a mistake and decide that he ought to do A, though A is really wrong. In that case he clearly ought to do A, therefore he ought to do what is wrong, that is, what he ought not to do. Self-contradiction can only be avoided if we suppose that ‘ought’ is being used in two different senses here.

The two senses of ‘ought’ that Ewing is concerned with are often called the ‘objective’ and ‘subjective’ senses of the term. This passage raises two important questions. First, is it really the case that, if you believe that Op, then it is in some sense true that Op? Second, if it is in some sense true that Op, is this sense different from the sense in which it is in fact the case that Op? As I understand him, Broome would answer these questions with ‘Yes’ and ‘No’, respectively.27 I reject this combination of answers. We have thus arrived at Substantive Disagreement #6. (As before, there is the complication that Broome is primarily concerned with the requirements of rationality, whereas I am concerned with the requirements of morality.) If, as Ewing claims, the answer to the first question is ‘Yes’, then the answer to the second question must also be ‘Yes’. Let me explain. Suppose that you believe that Op but in fact it is the case that Op. How could Ewing’s view that, if you believe that Op, then it is the case that Op, be reconciled with the fact that Op, without ‘O’ being used in different senses? The answer, according to Broome, is straightforward. Ewing’s view may be rendered thus: OðBOp ! pÞ

ð33Þ

BOp

ð34Þ

Suppose now that

Since factual detachment fails, we cannot infer Op

ð35Þ

and so there is nothing that conflicts with the fact that Op. All that we may infer from (33) is that you ought not to both believe that Op and not see to it that p. This is an interesting position, one whose plausibility may seem only to be enhanced if we invoke deontic detachment. For then, from (33) and the fact that Op, we can infer that O(BOp). Since in general it would seem best not to hold false beliefs, this inference may seem welcome.

27

See Broome 1999: 404–5; 2005: 322 ff. In the latter work, Broome qualifies his position by claiming that what is ‘primarily’ required is not bodily actions but states of mind.

176

MICHAEL J. ZIMMERMAN

The problem is, though, that deontic detachment can run in the opposite direction, and so this response does not dispose of Ewing’s problem. Imagine a case in which, perhaps because the consequences of believing that Op would be so good, it is true that OðBOpÞ

ð36Þ

Then (35) would follow from (33) after all, and that is unacceptable.28 So, too, with forced detachment. Imagine (as it is easy to do, since beliefs would seem seldom to be in our control) a case in which it is true that  C  ðBOpÞ

ð37Þ

29

(35) would once again follow.

References Broome, J. (1999) Normative requirements. Ratio 12: 398–419. Broome, J. (2001) Normative practical reasoning. Proceedings of the Aristotelian Society 75(suppl.): 175–93. Broome, J. (2004) Reasons. In R. Wallace et al. (eds.) Reason and Value. New York: Oxford University Press, 28–55. Broome, J. (2005) Does rationality give us reasons? Philosophical Issues 15: 321–37. Broome, J. (2007) Requirements. In T. Rønnow-Rasmussen et al. (eds.) Hommage a` Wlodek. (http://www.fil.lu.se/hommageawlodek). Chisholm, R. M. (1963) Contrary-to-duty imperatives and deontic logic. Analysis 24: 33–6. Chisholm, R. M. (1964) The ethics of requirement. American Philosophical Quarterly 1: 147–53. Chisholm, R. M. (1974) Practical reason and the logic of requirement. In G. E. M. Anscombe and S. Ko¨rner (eds.) Practical Reason. Oxford: Blackwell, 1–17. Ewing, A. C. (1948) The Definition of Good. London: Routledge and Kegan Paul. Feldman, F. (1986) Doing the Best We Can. Dordrecht, Germany; Boston, MA: Reidel. Jackson, F., and Pargetter, R. (1987) Two puzzles about conditional obligation. Philosophical Papers 16: 75–83. Lewis, D. (1973) Counterfactuals. Cambridge, MA: Harvard University Press. Ross, A. (1941) Imperatives and logic. Theoria 7: 53–71. Ross, W. D. (1930) The Right and the Good. Oxford: Clarendon Press. Schroeder, M. (2009) Means–end coherence, stringency, and subjective reasons. Philosophical Studies 143: 223–48. Thomson, J. J. (2008) Normativity. Chicago, IL: Open Court. Zimmerman, M. J. (1996) The Concept of Moral Obligation. Cambridge; New York: Cambridge University Press. Zimmerman, M. J. (2008) Living with Uncertainty. Cambridge; New York: Cambridge University Press.

28

Cf. Schroeder 2009: 227. The same point would of course hold if (33) were replaced with O(p/BOp) (330 ) and so switching from the former to the latter would not solve the problem. 29 Cf. Schroeder 2009: 227. Once again, switching from (33) to (330 ) would not solve the problem.

12 Reasons for Broome Jonathan Dancy

Chapter 4 of Broome’s Rationality through Reasoning (forthcoming) is about reasons. Broome has a lot to say about reasons, and he says it in the service of a general view that reasons are not as important in normative theory as many, myself included, have tended to suppose. I will comment on this at the end of my chapter, but my main concern is to sort out where I stand on the relation between Broome’s functional definition of a pro tanto reason as a consideration that plays a certain role in the weighing explanation of an ought, on the one side, and on the other a more standard account which takes as basic some notion of favouring, of counting in favour of, or perhaps just the distinction between for and against. Broome claims, effectively, that his definition in terms of weighing explanations does offer an account of the distinction between for and against. And this would certainly be worth having if it can be achieved. But I remain of the opinion that it cannot.

1 Pro Toto Reasons A pro toto reason to F is what explains why one ought to F, if one ought. Such a reason might be a quite complex object, not every part of which would ordinarily be considered to be a reason of any sort to F. But whenever one ought to F, there will be an explanation of that fact, and that explanation specifies what Broome calls a pro toto reason. I am not convinced that this notion of a pro toto reason is a helpful one. It is true that if one ought to F, there must be some reason to F. But this reason would be an ordinary sort of reason, the pro tanto sort with which we are more familiar. It is also true that if one ought to F, there must be some explanation of that fact, and that explanation will (often) take us beyond simple specification of a pro tanto reason. But the notion of a pro toto reason is a normative notion, and the notion of the explanation of an ought is a non-normative notion. So there is a certain disconnection here. The notion of the explanation of an ought is a non-normative notion because the notion of explanation is not normative. (There can, of course, be better or worse

178

JONATHAN DANCY

explanations, but that is nothing to the point.) We get an explanation when we have a ‘because’, and the notion of a ‘because’ is not normative. ‘Because’s simply explain, and what explains a normative fact need not itself be normative, nor need the relation of explanation be normative. Another way of giving the explanation of an ought is to specify what makes it the case that one ought. But this notion of ‘making the case’ is not normative either, as we can easily see when we realize that there are plenty of makings of cases in non-normative contexts; what makes it the case that this table is not square is the fact that one side is longer than another. Nothing normative about that. A pro tanto reason, by contrast, is a pro tanto reason to F. And this is a normative notion, just because it contains the normative phrase ‘reason to F’. So is the notion of a pro toto reason to F a normative one? Broome’s answer to this question is peculiar: We slide from ‘X is the reason why you ought to F’ to ‘X is the reason for you to F’, meaning exactly the same thing. The ‘reason why’ (meaning explanation) bumps into the normative ‘ought’, yielding a normative sense of ‘a reason’ that combines the meaning of both.’ (52)

He does admittedly betray a certain unease about this ‘picturesque etymology’, but insists that the sense he attributes to ‘the reason for you to F ’ is ‘undoubtedly genuine’. It would be hard to be satisfied with this. I do not see how he can both say that the notion of an explanation why you ought to F is not normative, that the notion of the (pro toto) reason for you to F is normative, and that these are the same notion. But I would say that the notion of the reason why you ought to F, while it may be identical with the notion of what explains why you ought to F, is still distinct from the notion of the reason for you to F. The first two are non-normative, and the last is normative. And there is no such thing as a pro tanto reason why one ought to F, even though there is such a thing as a pro tanto reason for you to F. None of this matters very much yet. But it may turn out to matter later when we come to explicating the sense of a pro tanto reason, to which I will shortly turn. But before I do that, I want to remark on a distinction that has concerned me for over 10 years but which has received little attention. This is the distinction between two normative relations, the favouring relation and the right-making relation. The distinction I have in mind is not based on the fact that one of these is contributory and the other overall. It is that, even though one and the same feature is capable of standing on the left-hand side of both of these relations, the right-hand sides are very different. What is favoured is acting (or at least responding) in a certain way, not the rightness of so acting, which cannot be favoured at all. (It cannot be favoured because the only things that can be favoured are responses, and the rightness of an act is not a response at all.) But the right-hand side of the right-making relation is the rightness so made. The relation involved is the general making-relation, the right-hand side of which is in this case the rightness of the action, not the fact that it is right. Certain features of the situation give the action a certain property, that of rightness.

REASONS FOR BROOME

179

Now this ‘giving’ is not itself an explanatory relation, even if in asking for explanations we are often satisfied by being told what it was that gave the object the quality concerned. But it is not a normative relation either. Myself, I would say it is metaphysical; it is how I have always conceived the relation I call resultance. (Not all would want to go down that route.) The relevance of all this is that Broome is hoping to give an explication of the favouring relation in terms of explanatory relations to oughts. But these two sorts of relations have very different structures, and I find it hard to think that one of them might really be the other in disguise.

2 Pro Tanto Reasons Broome offers us what he calls a functional definition of a pro tanto reason. This comes in two stages. First he gives his account of a pro tanto reason. Then he gives an account of a reason to F—what he calls ‘the for-F role’. Pro tanto reasons are either for or against, and something has to be said about that distinction. I suggested in the previous section that Broome’s pro toto reasons are not really reasons to F; they are at most reasons why one ought to F. But the notion of a pro tanto reason is definitely normative. As I would put it, pro tanto reasons play a contributory role in the construction of an ought. There is a kind of normative thrust coming up from below, and a combination of such thrusts can succeed in making an action one that we ought to do. This is the home ground of normativity. But Broome wants to persuade us that the whole story can be understood in terms of a special sort of explanation, which he calls a ‘weighing explanation’. Broome writes, the fact that you ought to F is the explanandum, and analogous to the fact that [in a mechanical weighing] the balance tips left. A reason for you to F is analogous to an object in the left-hand pan, and a reason for you not to F is analogous to an object in the right-hand pan. (54)

Here the choice seems to be between F-ing and not F-ing, and the reason is analogous to an object in a pan. The whole thing is expressed beautifully in the standard model of the scales of justice, as on the statue above the Old Bailey in London. So let us look in a little more detail at how these weighing explanations work. In a mechanical weighing explanation, what is to be explained is that the pan on the left goes down, and the explanation is given by attributing weights to the objects in the two pans, such that the combined weight of the objects in the left-hand pan is greater than that of the objects in the right-hand pan. In a normative weighing explanation, what is to be explained is the fact that of two alternatives F and G, F is what you ought to do, and you ought not to G at all; it is not that you ought to F more than you ought to G. And the explanation is achieved by attributing greater normative weight to the considerations that are reasons to F than to those that are reasons to G. So far, all this might seem entirely innocuous. Of course we do not really know

180

JONATHAN DANCY

what we mean by talking of the weight of a consideration that is a reason. It is not as if the one that is weightier is more of a reason. It is a stronger, or weightier reason. (One might similarly say that it is not as if one man who is stronger or weightier than another is more of a man; he is just stronger or weightier.) Being a reason is not a matter of degree. But what it is for a reason to be strong, or weighty, is not yet given. Now, I wonder whether we should already agree that being a reason is not a matter of degree. Certainly those who think of reasons as considerations that stand in a relation of favouring or disfavouring to a certain response may allow that the considerations themselves are not matters of degree—any more than are the objects in the two weighing pans—but still insist that the favouring or disfavouring does come in degrees. So, on their view, it does make sense to say that one consideration is more of a reason than another. It is, say, more of a favourer because it favours more. But as Broome explains his analogy, what is weighed is in the one case an object and in the other a consideration. We do not weigh the weight of an object and we do not weigh the normative weight of a consideration. The normative weight of a consideration is not what is in the metaphorical pan. This is what Broome says about a weighing explanation of why you ought to F: The explanation must include one or more pro tanto reasons, either for you to F or for you not to F. Each of these reasons must be associated with something that is identified as its ‘weight’. The reasons and their weights play a characteristic role in the explanation. The role is that the weights of the reasons on each side are combined in some way, and whether or not you ought to F is determined by the combined weights on either side. (54–5)

He concludes: This gives me enough material for defining a pro tanto reason. I have described a normative weighing explanation, and the characteristic role that pro tanto reasons play in one. A pro tanto reason is whatever plays this role in a normative weighing explanation. (55)

The idea here must be that being a reason is playing this role in an explanation (and on this point Broome is a sort of realist about explanation, thinking of the explanation as there to be given, not in terms of the giving of it). It is not that a reason is the feature or consideration that plays that role; though we say that her need for help is a reason for us to help, what we mean is that her need for help plays the role of a pro tanto reason in its contribution to a weighing explanation of whether we ought to help or not. Now it seems to me certainly true that there are weighing explanations of whether we ought to help or not. And I also agree that if a consideration is a reason in the case, it will contribute to such an explanation. The main question for me is whether this account of what it is to be a pro tanto reason in terms of contributions to a certain explanation is an account of what it is to be a reason. One might well feel wary on this

REASONS FOR BROOME

181

point, because of the way in which things work in mechanical weighing explanations, which are our model. The objects in the pans have a certain weight, and their presence is part of a weighing explanation of why the left-hand pan goes down. Now is it that they play that role in the explanation because of the weight that they have, or is it that they have the weight they have because of the role their presence plays in a certain sort of explanation? I would have thought it is the former. We have an independent conception of weight, in the light of which we can understand that the relevant object, placed in a pan, will contribute (because of its weight) to an explanation of why the pan did or did not go down. So, though mechanical weight can be characterized functionally, in terms of its contribution to weighing explanations, this does not mean that there is no independent conception of weight that is appealed to by those who peddle weighing explanations. All that we know is that if an object has weight, appeals to its presence in one pan will play a certain role in weighing explanations of why one pan goes down and the other up. I would say the same about the notion of a pro tanto reason. Considerations that are such reasons are able, because of that, to play a certain role in a normative explanation—that is, in the explanation of certain normative facts such as whether we ought or ought not to act in a certain way. But the feature that enables them to play that role is exactly the feature of being a reason, of which we have had as yet no substantial account. So what we are being offered is really a consequence of what we are interested in, as so often with functional explanations.

3 The for-F Role Consider now the following ‘definition’ of the for-relation: a consideration is for—or counts for, or counts in favour of—a certain response if it is able to play a certain role in a weighing explanation of why that response is appropriate, inappropriate, or neither appropriate nor inappropriate. Similarly for the against-relation: a consideration is against a certain response if it is able to play a certain role in a weighing explanation of why that response is appropriate, inappropriate or neither appropriate nor inappropriate. (I use the notions of appropriateness and inappropriateness here to indicate that there is nothing special about ought.) It seems to me that nobody could deny these claims. The only question is what sort of definition they amount to. Broome is clearly sensitive to this issue. He writes: Still, although my definition picks out one sense of ‘counts in favour’, you might think that [my definition] relies on our prior understanding of counting in favour in this sense. It defines a reason as something that plays the for-F role in a weighing explanation, and you might think that this role is nothing other than counting in favour of F. You might suspect that we could not identify the for-F role except through a prior understanding of counting in favour. That is not so. Notice first that a pro tanto reason can be defined without any reference to the for-F role. A reason is anything that plays the less specific charactistic role I described in a

182

JONATHAN DANCY

weighing explanation: it is something that has a weight, where the weights of reasons combine in some way to determine whether or not you ought to F. The doubt is about my definition of a reason for you to F specifically. That definition refers to the for-F role in a weighing explanation. But it still does not rely on a prior understanding of counting in favour. The for-F role can be identified from the structure of the explanation itself. (56–7)

One worry is that we are defining the for-relation in exactly the same way as we are defining the against-relation. The only way to tell the difference between them, since for and against play structurally identical roles in relation to explanations of appropriateness and inappropriateness, is to appeal to an independent understanding of that difference (which I would say we all have). There is no way of understanding against-F as for-not-F. These are not the same role, even if whatever plays one also plays the other. So I have expressed two worries about Broome’s suggestion that we understand the notion of a reason in terms of two distinct notions, those of ought and of weighing explanations. First I suggested that the favouring relation, which relates considerations to responses, is distinct from the right-making relation, which goes from those same considerations to oughts—effectively, to overall normative facts. Second, I have suggested that there is no way, within the purely structural constraints to which Broome restricts himself, to generate the non-structural distinction between for and against. There is a third worry, which is that, as well as having a polarity, reasons have a strength, and it is not obvious that Broome can make sense of that. It might appear that he can, because of the analogy with weight. The different objects in the pan have different weights, some heavier, others less heavy. But this ignores the fact that we have an independent conception of weight, which is not to be identified with the ability to contribute to weighing explanations. According to Broome, there is no independent conception of normative weight, the weight of a reason. There is only the ability to contribute to a weighing explanation of an ought. But is it possible to make more, or less, of a contribution to a weighing explanation? Which of the objects in the pan makes more of a contribution to the explanation of why that pan goes down? The answer is not necessarily the heaviest object in the pan. Suppose that we have two 1 kilo weights, one in each pan, and a 10 gram weight in the left-hand pan. The explanation of the fact that the left-hand pan goes down will not refer primarily to the heavier object in the pan. And similarly there is no sense to the idea that one reason contributes more than another to a weighing explanation of an ought.

4 Reasons and Explanations There is a fourth worry also, which is that Broome thinks of the notion of a pro tanto reason as normative only because of the normativity of what it is used to explain. It

REASONS FOR BROOME

183

has no inherent normativity, one might say. It gets its normativity from the normativity of an ought. So is the notion of a pro tanto reason an essentially explanatory notion? Broome suggests that it is, since what it is to be a reason (I restrict myself to the pro tanto from now on) is to be able to contribute in a certain way to an explanation of a certain sort. Myself, I doubt it, but one needs to distinguish between (a) what can be used to explain, (b) what cannot be used without contributing to an explanation, and (c) what is essentially explanatory. As I suggested earlier, uses of ‘because’ are essentially explanatory. To say ‘p because q’ is essentially to offer a (perhaps partial) explanation of p. This is so whatever p is—whether it is an action, an event or anything else. The same is true of ‘the reason why p is that q’. To say that his reason for doing F was that q, is also necessarily to offer something that is a partial explanation of his doing F, but I would say that the notion of an agent’s reason for acting is not itself an explanatory notion in the way that ‘because’, or ‘the reason why’ are. Similarly, I suggest that ‘cause’ is not itself an explanatory notion in the way that ‘because’ is, at least not if we conceive of causation as the cement of the universe rather than thinking of a cause, as Hobbes did, as that given which it cannot be understood but that the event should occur. But it is still true that when one says that one event is caused by another, one is necessarily offering a partial explanation of the latter. The notion of a pro tanto reason is even further removed from explanation. And this is just as well, because otherwise we would be unable to make adequate sense of the relation between reasons and oughts.

5 Reasons and Oughts In this area there are three possible views. The first view is that the two notions of a pro tanto reason and of an ought are distinct, but connected—a sort of non-reductive local holism, one might say, because one can have neither without the other. The second view is that the notion of a reason is to be understood in terms of some relation to an overall ought. The third view is the other way around, that the notion of an ought is to be understood in terms of some relation to pro tanto, or contributory reasons. I prefer the terminology of contributory because of the way in which the phrase ‘pro tanto’ introduces what for me are unwanted implications of generalism and atomism. ‘Pro tanto’ cannot escape its etymology in my mind, since it used to mean ‘as far as this goes’, ‘to the extent that’, or ‘for as much as’, as in the prayer ‘Lord, for as much as without thee we are not able to serve thee’. And the phrase ‘as far as this goes’ seems to extract a consideration from its context, with the suggestion that we can tell what it does by considering it all alone. I accept, however, that one can perfectly well use the notion of the pro tanto so long as one remembers the dangers of this implication and resolves to abjure them. So, since this is a chapter about Broome, I will continue to talk his language.

184

JONATHAN DANCY

Of these three views, Broome promotes the second, and I prefer the third. These views can still be holistic, in the sense that they can allow that one could not have one notion without having the other, but they are both reductive. In this section I discuss the comparative merits of these two approaches. Broome’s view is a variant of W. D. Ross’s notion of the relation between prima facie duties and duties proper. Mine is a version of H. A. Prichard’s response to Ross on that topic. (Prichard wrote, in a letter, ‘I used to believe Ross’s view until he wrote it down.’) Broome’s view is a variant of Ross’s because of the ways in which Ross tried to explicate the relation between his prima facie duties and what he called duties proper. Ross had various shots at this. His overall view is that prima facie duties are not duties at all, but something related in a special way to duty. In his first book The Right and the Good (1930) he tried two ways of explicating that ‘special way’. The first of these was what I take to have been his official view at the time, which is that a prima facie duty is ‘the characteristic (quite distinct from that of being a duty proper) which an act has, in virtue of being of a certain kind, . . . of being an act which would be a duty proper if it were not at the same time of another kind which is morally significant’ (1930: 19). I have made several distinct criticisms of this attempt in the past, but the one that concerns me here is that being a prima facie duty should be able to be a matter of degree, in the sense that such duties can be more or less binding, pressing, incumbent, stringent, or urgent; these are all terms used by Ross, and he also uses the phrase ‘more of a duty’, to which he has no right, since his official view is that duty proper is not a matter of degree. Regardless, the point is that the sort of isolation test offered in the quotation above cannot make any sense of prima facie duty being a matter of degree. An act cannot have more or less of ‘being an act which would be a duty proper if it were not at the same time of another kind which is morally significant’. What is more, Ross fails to establish that the normative ought, the duty proper, is made out of normative forces coming up from below. But this is what he is trying to show. He holds that prima facie duties are characters that an act can have and that in the right circumstances can make that act a duty. To do this he needs to show what sort of a normative role is played (or, as one might put it, what sort of normativity is enjoyed) by a prima facie duty, whether it is or is not successful in generating a duty proper. And this he fails to do. In response to this difficulty, some writers point to a second attempt of Ross’s, which emerges when he considers the analogy between mathematical attributes and moral ones. The equality of the two angles [of an isosceles triangle] is a parti-resultant attribute. And the same is true of all mathematical attributes. It is true, I may add, of prima facie rightness. But no act is ever, in virtue of falling under some general description, necessarily actually right; its rightness depends on its whole nature and not on any element in it. The reason is that no mathematical object (no figure, for instance, or angle) ever has two characteristics that tend to give it opposing resultant characteristics, while moral acts often (as everyone knows) and

REASONS FOR BROOME

185

indeed always (as on reflection we must admit) have different characteristics that tend to make them at the same time prima facie right and prima facie wrong. (1930: 33–4)

There is a scholarly debate about which version is intended by Ross as the official version, the isolation test or the appeal to tendencies. But this does not make much difference in the great scheme of things, because the appeal to tendencies cannot be got to work either. Tendencies to be right can be stronger or weaker, and so a tendency to be right is a matter of degree, which helps; but a tendency of this sort is not itself normative. The normativity of the tendency comes entirely from the normativity of the rightness that acts of this sort tend to have. Further, it is just not clear that an unfulfilled tendency is the right sort of thing to generate residual duties, and it should be clear if an unfulfilled tendency had its own normativity. The conclusion is fairly drastic. It is that Ross failed to make any sense of the relation between prima facie duties and duties proper. And this is an important matter, because it means that what is perhaps Ross’s most important idea in ethical theory is a failure. That idea was that duties, or oughts, are built out of normative contributions coming from below. This is what Broad thought of as one of the great advances in ethical theory in the last century. But Ross’s official account of how this works, of duties and of the features that generate them, was a failure. So what of Broome’s account, in terms of contributions to weighing explanations of oughts? This account is analogous to Ross’s in the way in which it tries to understand the pro tanto in terms of some suitable relation to an overall ought. The question has to be whether pro tanto reasons, as he understands them, have the sort of normativity that they need if they are to be able to generate normative oughts. In addressing this question, we might think it would help to get hold of Broome’s conception of an ought. But this is harder than one would expect, because he gives no account of oughts. He takes it that we have a general concept of ought which we use in various contexts, and he identifies the particular ought that interests him by contrasting it with others that are not his immediate concern. So the ought that he is concerned with is a normative, owned, overall ought. And I think it fair to take it that this ought is not a matter of degree. It is verdictive, in Stratton-Lake’s sense. One cannot say ‘you ought to do this but you ought more to do that’. No ought is stronger than any other ought, though one may perhaps be subject to greater censure if one fails to do one thing that one ought to do than if one fails to do another such. So, I ought to finish this paper because it is already late, and I ought to support my sister in her struggles looking after our aging mother, and if I fail to do the latter I am not only more liable to censure but also liable to more censure than if I fail to do the former. So in this sense some oughts are more incumbent than others, but there is no suggestion that degree of appropriate censure comes with degree of ought. By analogy, guilt is not a matter of degree; one cannot ask whether a person guilty of murder is more guilty than a person guilty of shoplifting, even if some guilt attracts

186

JONATHAN DANCY

more Draconian punishment than other guilt. At most we might say that one has more to feel guilty about than the other does. We might say that one feels more guilty than the other, or should feel more guilty, but this does nothing to show that their guilts vary accordingly. This Broomean ought is normative. The normativity of a pro tanto reason to F is generated only by the way in which the presence of the relevant consideration is part of what explains the fact that one ought to F. Is this enough to establish the normativity of the reason to F? I cannot see that it is. Broome is in a spot similar to Ross’s spot. He has the normativity of ought (for Ross, duty proper), and he wants to establish the normativity of the things whose presence explains that ought. But he has announced that explanation is not itself normative, even if what is explained is normative, and he has no independent method of establishing the normativity of the explainers. Broome writes, ‘The combined weight of the reasons for you to F exceeds the combined weight of the reasons for you not to F. That is what makes it the case that you ought to F’ (54). But his only model of normativity is the normativity of ought, and making the case is not a normative relation. So it seems impossible to establish the normativity of the reason using only the tools that Broome is willing to countenance. In material that has only recently come to light, H. A. Prichard writes: To maintain that we ought to do that of two actions which we ought to do more is to be involved in a contradiction, because it is to imply both that while the one act is a duty the other is not, and that the latter is a duty though to a lesser degree than the former. Putting this otherwise, if we think that of two acts there is a greater obligation to do one of them, we cannot go on to think that we ought to do that action without implying that there is no obligation to do the other in any degree whatever. Consequently we gain nothing by maintaining the existence of degrees of obligation in addition to absolute obligation. (2002: 79)

This remark is made against Ross, but it applies to Broome as well. For Broome, the normativity of an ought is basic, and other normativities are established, if at all, in relation to that. What we are looking for, then, is something that is like the normativity of ought but not just the same—like that, but not that. But what explains an ought does not thereby acquire a subordinate form of normativity, derived from the normativity of what it explains. Matters look quite different if we take the alternative tack, starting from the normativity of a reason as basic and building the normativity of an ought out of that. Obviously the easiest way of doing that is to understand an ought as a ‘most reason’, and that is the way that I am tempted to go.1 (see again Dancy forthcoming) But I could, I suppose, imagine a more conciliatory position which does not reduce

1

I say more about this issue in my ‘More Right than Wrong’, forthcoming.

REASONS FOR BROOME

187

the ought to a ‘most reason’, but still explains the normativity of the ought in terms of its relation to, or derivation from, the reasons that stand as its ground. Such a view might hold that reasons are in the business of trying to make the action they support (or are reasons for) the action that one ought to do, and that an ought is what you get when such an attempt is successful. Reasons are not themselves right-makers, but some reasons are right-makers, those that succeed in their proper business. (You can tell from the way I have expressed this third position that I do not really think it can be got to work.)

6 Ownership of Reasons Broome ends his chapter on reasons with the following remark: Philosophers of normativity are mainly interested in the normative state of the world. They want to know people’s normative properties and relations. For instance, they want to know when a person ought to do something. But in recent decades many philosophers of normativity have concentrated on reasons, which are not normative properties of people but things that explain the normative properties of people. Many reasons are non-normative facts or . . . nonnormative predicates. Attending particularly to reasons can distract you from the normative properties you aim to study. (72)

This contrast between oughts, understood as properties of people, and reasons, which are not properties of people, deserves some comment. The idea that oughts are properties of people reminds me of Prichard’s notorious argument in his ‘Duty and Ignorance of Fact’ (reprinted in Prichard 2002), in which he ended up believing that rightness is a property of actions and ought is a property of agents. This enabled him to maintain that whether an agent ought to do the action is dependent not on the facts but on the agent’s beliefs, while whether the action was right or wrong is dependent on the facts. I do not recommend this conclusion, and Broome is not guilty of it. He is, however, guilty of a false contrast here. Is it true that reasons are not properties of people? As I see it, a reason is a consideration standing in a normative relation to a response. But it is not a reason for the response to happen; it is a reason for someone or other (specified) to respond in that way. There are no reasons waiting around for someone to attach themselves to, for an owner. So a reason is a three-place object. It consists in a normative relation between a consideration, a person, and a way of responding. Where there is a reason, there will be some consideration calling for a certain sort of response from some specific individual who has that reason. In this sense, reasons are no less ‘owned’ by individuals than are Broome’s oughts. Any temptation to suppose otherwise is presumably due to our constant (and perfectly harmless in its own way) talk of the consideration itself as the reason.

188

JONATHAN DANCY

References Broome, J. (forthcoming) Rationality through Reasoning. Oxford: Oxford University Press. Dancy, J. (2002) Prichard on duty and ignorance of fact. In P. J. Stratton-Lake (ed.) Ethical Intuitionism. Oxford: Oxford University Press, 229–47. Dancy, J. (forthcoming) More right than wrong. In R. Johnson and M. Timmons (eds.) Kantian Reflections on Morality, Law, and Society: Critical Essays on the Philosophy of Thomas E. Hill, Jr. Oxford University Press. Prichard, H. A. (2002) Moral Writings; H. A. Prichard, ed. J. MacAdam. Oxford: Clarendon Press. Ross, W. D. (1930) The Right and the Good. Oxford: Clarendon Press.

13 Normative Conflicts and the Structure of Normativity Andrew Reisner

1 Introduction When I began my graduate studies in 1997, what might be dubbed, ‘the normativity revolution’ was already well under way.1 In the wake of influential work by Jonathan Dancy, Thomas Nagel, Derek Parfit, Joseph Raz, and Bernard Williams, people studying practical philosophy had become increasingly focused on the general phenomenon of normativity, de-emphasizing to some degree more traditional and narrower inquiries into specific normatic topics, such as morality and prudence. The principal unit of currency in these debates was a reason. By the time I had dipped a toe in the water as a researcher, the assumption that the study of normativity was the study of reasons had rooted itself so deeply amongst practical philosophers that one might barely have thought to question it. It is thus interesting, and from a historical point of view important, to note that many of the advances in present thinking about normativity arose from John Broome’s striking suggestion that progress had been stifled by an excessive focus on reasons. At first by fixing our attention on the importance of normative relations best captured with oughts, and later through his discussion of requirements, Broome did much to reshape our fundamental understanding of the normative world.2 One lesson to take away from Broome’s work is that, if one has an ought, one does not need a reason. To put it another way, oughts are stronger than reasons. If one knows what one ought to do,3 then one knows what normativity requires of oneself; merely 1 I am very grateful to John Broome and Michael Zimmerman for valuable feedback on this chapter. I should also especially like to thank Bruno Guindon, whose acuity as an interlocutor and writer on closely related subjects has improved my understanding of many of the issues touched on in this chapter. 2 The initial suggestion that we focus on oughts was made in Broome (1999). His work on requirements can be traced through a number of papers, but it is most highly developed in Broome (2007, 2013). 3 As I do in other places in this chapter, I am using ‘do’ somewhat incorrectly as a universal verb. Here it is to be understood to include under its scope the having of various attitudes, including beliefs and feelings, in addition to the performance of actions.

190

ANDREW REISNER

knowing that one has a reason or reasons to do something leaves it open that normativity nonetheless requires one to do something else. In this chapter, I shall be considering a new lesson that we might learn from Broome’s insight into the centrality of ought: our views about certain features of ought constrain importantly what views are available to us concerning reasons.4 In particular, they constrain the possible relations between particular kinds of reasons— for example, reasons to act, believe, or feel—and the sources of each of those kinds of reasons.5 The features of ought that are of special interest here are those that concern whether there are genuine conflicts amongst final oughts, or whether such conflicts are only apparent. I shall call conflicts amongst final oughts ‘normative conflicts’, and the standard for the occurence thereof is the mutual non-satisfiability of two or more final oughts.6 I shall conclude that the most common view about the relation between types of reasons and their sources commits one to the existence of genuine normative conflicts. I shall also suggest, tentatively, that philosophers inclined towards rejecting the possibility of genuine normative conflicts may be pressed to consider universal pragmatism about reasons.

2 Assumptions There remain many unresolved questions about the nature of oughts and of reasons. This chapter is not the place to resolve them. I am therefore adopting the expedient of working with the views that I take to be correct. How much, and how many, of the arguments in this chapter hinge on adopting these views about reasons is difficult to assess. My suspicion is that the details of the background views about reasons adopted here are of little consequence to the overall discussion. I shall assume an ordinary version of reasons externalism: the view that facts or considerations provide normative reasons to do, feel, and believe things. These facts and reasons are not, or are not for the most part, dependent on an individual’s mental states or motivations.7 I have also worked under the assumption that there is what I shall call a ‘rock bottom’ or ‘final’ ought. The English word ‘ought’ no doubt has varied applications;8 I am only interested in one of them, and I can only gesture at it, since it is not analysable. It is the most basic normative use of the word ‘ought’, the one that applies 4 The constraint is symmetrical; views about the relevant features of reasons will also constrain possible views about oughts. 5 I draw the use of ‘source’ in this paper from Reisner (2004). Its use here is broadly similar to how it is used in Broome (2007, 2013), with the difference that I am relating sources directly to reasons, whereas Broome relates them first to requirements and via requirements to reasons. 6 I leave it open as to the correct type of mutual non-satisfiability to be employed in the condition, but I am assuming that the analytic, metaphysical, or conceptual impossibility of satisfying two or more oughts will suffice. 7 See Parfit (2001) for the standard account of externalism about reasons. 8 Broome (2007) and Cariani (2013) discuss the uses of ‘ought’ in great depth. A standardly used, but problematic, semantics for ‘ought’ is given by standard deontic logic. See Forrester (1996: ch. 2).

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

191

to agents, and the one that figures in the rational requirement to intend to do what you believe you ought to do.9 I shall remain neutral about whether the relevant facts or considerations are themselves reasons, or whether they provide reasons. However, I am assuming that reasons, at least in most cases, determine what it is that one ought to do, believe, feel, and so on.10 I am assuming cognitivism about sentences using normative language (‘reason’, ‘ought’) and also cognitivism about normative judgements. Concerning the metaphysics of reasons, I am assuming some form of non-naturalist realism. It is not clear how much hinges on either of these assumptions. What is important is that these assumptions are all compatible with a wide variety of what may be called ‘normative architectures’. Normative architectures are different ways reasons and their sources can relate to each other. The kind of constraints that different views about normative conflicts put on a theory of reasons, and therefore on the overall theory of normativity, are at minimum architectural constraints. Not all architectures are compatible with all views about the possibility of normative conflicts.

3 Conflicts One simple way of thinking about reasons is that they are the normative entities that aggregate or combine in various ways to yield oughts. The kind of ought I have in mind is that of the final normative operator: the all-things-considered or final ought. It is the ought that is at normative rock bottom: there is no further, more fundamental ought or other normative notion. This does not preclude the possibility that there is more than one type of final ought at rock bottom. There may be a single ‘just plain ought’,11 or there may be a number of domain specific oughts—ought to believe, ought to do, ought to feel, and so on—that are jointly at rock bottom. Normative conflicts occur when there are two or more final oughts that are not mutually satisfiable. Normative conflicts can be classified by whether they are intra-domain conflicts or inter-domain conflicts. By ‘domain’, I mean that of, for example, practical normativity (what one ought to do, intend, etc.), theoretical normativity (what one ought to believe), affective normativity (what one ought to feel), and any other category of this kind. Intra-domain conflicts occur when competing reasons within a domain (putatively) generate mutually unsatisfiable oughts. One way to generate intra-domain

For a defense of this way of individuating the final ought, see Broome (2005, 2013) and Wedgwood (2007). 10 I am unsure whether this is consistent with efforts to analyse reasons in terms of oughts. See Broome (2005) and Kearns and Star (2008) for examples of such analyses. I am assuming no such analysis here. 11 I take the phrase from McLeod (2001). 9

192

ANDREW REISNER

conflicts is to consider (putative) conflicts between different kinds of reasons for a particular propositional attitude. For example, there might be mutually unsatisfiable differences between what an agent ought to desire, given her state-given reasons, and what she ought to desire, given her object-given reasons.12 Object-given reasons for desires are normally thought to be given by the goodness of the object of the desire.13 State-given reasons for desires are given by the goodness of having particular desires. For example, an evil demon will punish you unless you do not have a desire to help those in need. That gives you, suppose, an overwhelmingly strong state-given reason not to desire to help those in need. That helping those in need is, suppose, a very good thing to do gives you an overwhelmingly strong object-given reason to desire to help those in need. According to your state-given reasons, you ought not to desire to help those in need; according to your object-given reasons, you ought to desire to help those in need. Assuming that there are both state- and object-given reasons, there is an (at least putative) intra-domain normative conflict in this case.14 Inter-domain normative conflicts occur when what you ought to do15 in one domain precludes you from doing what you ought to do in another domain. Here is an example. Suppose that the evidence strongly suggests that p is the case. On a number of common views about theoretical reasons, one ought to believe p. Suppose, however, that believing p will make you depressed and will interfere with your life. And, further, suppose that you foreseeably will not rely on whether p is the case for any consequential decisions. On a number of common views about practical reasons, you ought to cause yourself not to believe p (by reading lots of anti-p propaganda, by only hanging around with others who are agnostic about p or who disbelieve it, etc.).16 ‘Cause’ is a success verb, so if you do what you ought to do (cause yourself not to believe p), you will not believe p. However, you ought to believe p. So, you cannot do what practical normativity says that you ought to do and believe what theoretical normativity says that you ought to believe. This is an example of an inter-domain conflict. Putative normative conflicts of both varieties are interesting for a number of reasons. It is tempting, at least, to formulate questions about these conflicts by employing a final ought, although of course one may do so under other descriptions. For example, one might ask, ‘Ought I to do17 as practical rationality says that I ought, 12

The classic discussions of this distinction may be found in Parfit (2001) and Piller (2006). It is this feature of object-given reasons for desire and other pro-attitudes that underpins fitting attitude analyses of value. 14 This example assumes that state-given reasons cannot be reduced to object-given reasons, but examples can be generated without this assumption. For a defense of the non-reducibility of state-given reasons, see Reisner (2009). Parfit (2001) and Skorupski (2011) defend the reducibility of state-given reasons to object-given reasons. 15 I use ‘do’ here as a universal verb, so it is meant to include beliefs and feelings. English lacks a universal verb, so one has to be stipulated for this purpose. 16 For examples, see Parfit (2001) and Skorupski (2011). 17 Here ‘do’ is being used as a universal verb. 13

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

193

or as theoretically rationality says that I ought?’ This suggests that we at least possess the concept of a final ought that is not pegged to a particular domain. At the same time, in asking whether there are genuine conflicts, we raise the possibility that there is more than one final ought, each pegged to a particular domain. The pluralistic view about final oughts is very common.18 People whom I shall call ‘normative separatists’ (or ‘separatists’) think that there are distinct final oughts for each domain of normativity, for example, one of theoretical reason, one of practical reason, and one of affective reason. Within any single domain, it is open to the separatist to think either that there are intra-domain conflicts, or that there are not. Normative separatism is not the only possible response to putative inter-domain conflicts. One could be what I shall call a ‘normative non-separatist’ or ‘non-separatist’. Non-separatists think there is one, single final ought that ranges over all the normative domains. There is more than one possible response for the non-separatist to putative normative conflicts at the inter-domain level. Non-separatists can accept that more than one of the single, final, inter-domain oughts can obtain and conflict, or that all putative normative conflicts are apparent.

4 Conflicts and the Structure of Normativity There are at least two distinct ways of thinking about putative normative conflicts, at a meta-normative level and at a first-order level. At the meta-normative level, one might adopt normative architectures that do, or alternatively do not, allow for normative conflicts. If a normative architecture does not allow for normative conflicts, then one must provide first-order solutions to the conflicts.19 If it does allow for normative conflicts, then no first-order solution is required. A complete theory of the architecture or logical structure of normativity would have to tell us what sorts of things there are reasons for, such as beliefs, desires, actions, intentions, emotions, and so on. It would also have to tell us something about where those reasons came from: the sources of normativity—evidence, benefit, and so on.20 We can ask whether each different type of thing for which there are reasons or oughts has a different source of normativity. And, we can ask whether the sources of normativity for one domain of normativity may also serve as sources for other domains of normativity. We would need to know whether or not (and if so, how) the reasons and oughts that arise in each domain of normativity (the practical, the theoretical, and so on) could be weighed against or otherwise compared with the reasons and oughts from other domains of normativity. 18 For examples, see Buchak and Pettit in this volume (Chapter 14); Hieronymi (2005); Parfit (2001); and Skorupski (2011). 19 For example, one must say something about how to weigh the reasons that contributed to the apparently conflicting oughts, or one must come up with a theory about which apparent ought actually comes up trumps. For a worked out version of the former approach in the epistemic sphere, see Reisner (2008). 20 For more on the sources of normativity, see Broome (2013), Guindon (n.d.), and Reisner (2004).

194

ANDREW REISNER

Looking at different possible architectures helps us to think about whether to accept normative separatism or normative non-separatism. Normative separatists believe that, when the ought of theoretical normativity and the ought of practical normativity conflict, there is no further question as to whether one ought to comply with the theoretical or the practical ought. The practical ought and the theoretical ought are both final oughts; there is no more basic ought that subsumes them. Normative separatists need not deny that practical and theoretical normativity have various internal structural affinities with one another, but they must maintain that they are not logically related in a way such that the demands of different domains of normativity are comparable with respect to a more basic ought. Nonseparatists are in a position to ask first-order questions about the possible resolvability of inter-domain conflicts. Before continuing with the analysis of possible normative architectures, it is useful to say something very briefly about sources of normativity. While a partial analysis of a source of normativity is given in }4.1, it is worth noting in advance that the notion is not wholly unproblematic. A source of normativity may be intuitively understood as that in virtue of which a reason relation holds, or a particular kind of explanation of why a fact is a reason of a particular type. For example, the fact that there is a seminar today is a reason for me to believe that it is not the weekend. That fact is a reason for that belief, because that fact is evidence for the contents of the belief. The source of normativity for the reason is evidence, or perhaps truth. Sometimes it might be difficult to say why something is a reason for something else, even in cases in which intuition tells us that it is. Suppose Loretta’s windowsill is unadorned and has no plant on it. The fact that Loretta’s windowsill does not have a plant on it is a reason for you to give her a plant. What is the source of normativity? It might be beneficence, as Loretta will be happier if she has a plant on her windowsill. It could be beauty, as the windowsill will not look very nice without a plant on it. Or, perhaps, there is nothing much more to say about why the fact that Loretta’s windowsill does not have a plant on it is a reason for you to give her a plant. At least, I am not certain what the correct thing is to say in this case. One possibility is that the fact that Loretta’s windowsill is unadorned is (or provides) three distinct reasons for me to give her a plant. These reasons will be distinguished not by their relata—they are in each case the fact that Loretta’s windowsill is unadorned, me, and giving her a plant—but the source that grounds this relation. Despite difficulties with individuation, I see little reason to doubt that there are sources of normativity. Two examples may help. Strict evidentialists think that reasons for belief only issue from evidence or perhaps truth.21 Thus, truth or evidence is the source of reasons for belief. Strict consequentialists about reasons for action 21

See Adler (2002); Kelly (2002); Shah (2006), and Shah and Velleman (2005) for examples of strict evidentialism.

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

195

believe that reasons for action issue only from features of the goodness of states of affairs that will (or are expected to) result from the action in question.22 They believe that goodness is the source of reasons for action. What the notion of a source of normativity lacks in clarity, I hope is made up for by its intuitiveness.

4.1 Possible normative architectures 1: Basic and derivative normativity In order to begin exploring normative conflicts, two notions may be introduced. They are the notions of basic normativity and derivative normativity. Value theorists distinguish between final and derived value: some things are valuable in-and-ofthemselves23 and others are valuable in virtue of their—often instrumental— relationship to something that is itself finally valuable. I shall offer a similar distinction for normativity, one between basic and derivative normativity. It will also be helpful at this juncture to say something about the logical structure of ought. Ought is an operator that operates on propositions. We can analyse any ought sentence, at least when ought is serving its role as the basic normative operator, as saying ‘Ought p’, where p is some proposition. Initially, it might look unlikely that ought governs a proposition, as the normal English usage of ‘ought’ has it taking an infinitive. ‘Ought that’ is not a normal locution in English, and we normally expect operators that take propositions to be followed by a ‘that’ rather than an infinitive. There are locutions in English, which in many contexts mean the same thing as ought sentences do, that take ‘that’. Here are two examples. One may say, It is fitting that he goes the store, or one may say, It is meet that Achilles honours his friends. More importantly, the claim that infinitives do not express propositions is sometimes a mistake, as in some uses infinitives contain tacit subjects. There need be nothing conceptually different between the locutions: He ought that he goes to the shop, and, He ought to go to the shop, since ‘to go to the shop’ in fact has a tacit subject, which is ‘he’. However, there is an advantage to using the non-grammatical ‘ought that’ over the grammatical ‘ought to’, and that is the ease with which one can make the tacit subject of the sentence that ought governs explicit. There are other good reasons for using ‘ought that’ in this context, the primary one being that it allows for a convenient way of expressing an ought governing a conditional sentence, such as ‘He ought that if he goes to the store, then he buys some eggs’.24 In normal English usage, an ought governing a conditional is attached to the consequent of the sentence, ‘If he goes to the store, then he ought to buy some eggs’. Unfortunately, this is the identical locution used in English to express the case where the ought is attached to the consequent. Ordinary English has no grammatical means for distinguishing between oughts that govern only the consequent and those

22 A version of this view can be generated by combining Broome’s teleology and consequentialism. See Broome (2004). 23 It may be misleading to say ‘in-and-of-themselves’, as there may be final, but extrinsic, value. 24 For further discussion of this form of ought sentence, see Broome (2013) and Reisner (2004).

196

ANDREW REISNER

that govern whole conditionals. Further, it may seem at first that we can understand the sentence, I ought to go to the shop, as logically expressing: Ought that I go to the shop. That expression is incomplete, because the ought operator needs a subject. To provide a more accurate ought sentence, we can use what I shall call an ‘O-form’ sentence: I ought that I go to the shop. In the O-form sentence, I ought that I go to the shop, the subject of the ought and the subject of the proposition that it governs are the same. Nevertheless, the subjects are logically distinct, and the O-form sentence allows one to make the distinction explicit. It is in principle possible in the grammar of an O-form sentence for the subject of the ought to be different from the subject of the proposition that it governs, although the plausibility of such cases will not be discussed here.25 Although oughts and reasons have different logical structures, they are similar in three respects that are relevant here. First, they both index to an agent: A ought that A do x, and f is a reason for A that A do x. Second, we can type both oughts and reasons by what they are oughts or reasons for. A reason is a reason for belief (a theoretical reason), just if the relation is of the form: fact f is a reason for agent A that A believe b. And an ought is a ‘belief ought’ just if the O-form sentence is of the form: A ought that A believes b. When the expression ‘type of ought’ is used in this chapter, it refers to whether the O-form sentence is an action sentence, a belief sentence, a feeling sentence, and so on.26 ‘Type of reason’ is used in the same manner. Although both oughts and reasons govern propositions, in the remainder of the chapter I shall sometimes discuss reasons for action or for belief rather than reasons for propositions concerning actions or propositions concerning beliefs. I do this because the sentences that employ the full propositional form quickly become cumbersome. It will be useful to say a bit more about what a source of normativity is.27 It is perhaps misleading to suggest that normativity has a source. The use of the word ‘source’ might wrongly be interpreted to imply that normativity flows from some interesting metaphysical construction in the way light does from the sun. A better analysis will appeal instead to a ground in virtue of which an ought fact or reason relation obtains. Assume that the fact that Mary is drowning is a reason for James that he save her. One can ask what sort of reason this is. It is a reason for an action, that James saves Mary. One can also ask why the fact that Mary is drowning is a reason for James that 25 I am sceptical that the concept of ought admits of a divergence between its subject and the subject of the proposition that it governs, but this is not a matter of logical or grammatical form. 26 Note that I am using O-form sentences to make explicit the structure of the correlating proposition. The ‘just if ’ used here is not meant to exclude belief-ought sentences that are superficially different in form. 27 My notion of a source of normativity is close to what Broome has in mind in Rationality through Reasoning (2013). I do not think that it is quite the same, however. I am not committed to there being a general relation between requirements and sources, nor to the picture that Broome provides concerning the relationship amongst reasons, requirements, and sources. For an interesting alternative picture to Broome’s, see Guindon (n.d.). See also Michael Zimmerman’s chapter in this volume (Chapter 11).

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

197

he saves her. One might answer by saying that in these circumstances, Mary would benefit from being saved by James. That it benefits Mary in these circumstances is the explanation of why the fact that Mary is drowning is a reason for James to save her. Consider a case involving a reason for belief. The fact that it is Tuesday is a reason for me to believe that yesterday was Monday. Here, what makes this fact a reason for me to believe that yesterday was Monday is that its being Tuesday metaphysically necessitates that (provided there was a yesterday) yesterday was Monday. In both of these examples, there is a ground in virtue of which some fact is a reason for something. In the case of the action above, it is that it is beneficial to Mary (or just generally) for James to save her when Mary is drowning. In the belief example, it is that the fact that is the reason is evidence for the contents of the belief. Not all sources of reasons will be quite as simple to identify as benefit or evidence. There may be an unlimited number of normative sources or very few. Radical pragmatists, for example, might think that benefit is the only source of normativity for any type of reason.28 A radical pluralist might think that each reason has a slightly different source. The possible relations of sources to reasons will be discussed below. We can now move on to discuss basic and derivative normativity. Basic normativity is normativity that does not derive from its object’s role as a means to, in promoting, or as a component of some other normative end. Derivative normativity is any normativity that is not basic. As an example, consider a simple theory of normativity that tells you that you have reason to do whatever will make you happy. Looking at a case involving reasons, the fact that eating sweets will make you happy is a reason for you to eat sweets. The normativity of that reason is basic. Eating sweets is one of the things that makes you happy. Saving money, on the other hand, does not make you happy. However, in order to buy sweets, you must have saved some money. So, there is a derivative reason to save your money. The fact that saving money will allow you to buy sweets, which in turn will make you happy, is a reason for you to save your money.

4.2 Possible normative architectures: Combinations of views about basic normativity and types of reasons This section discusses the ways in which sources of basic normativity can combine with different types of reasons and oughts. Normative architectures are defined along two axes. The first axis divides between normative monism and normative pluralism. Normative monism and pluralism each reflect a claim about the number of sources of basic normativity. Normative monism is the view that there is only one source of basic normativity. So, a radical pragmatist, who thought only goodness provided reasons of any type, would be a normative monist. Normative pluralism is the view that there is more than one source of basic normativity.

28

This seems to be Stephen Stich’s view. See Stich (1993).

198

ANDREW REISNER

The other conceptual axis contains reasons specialism and reasons generalism. This distinction requires some care in specifying, as there are some species of reasons generalism that look like reasons specialism. Reasons specialism is a view about the conceptual connection between sources of normativity and types of reason. Reasons specialism holds that for each type of reason—for example, for action propositions, belief propositions, and so on—the source or sources of normativity from which it issues have an appropriate and in some way distinctive conceptual link with that type of reason. For example, evidentialists may think that it is a conceptual truth that reasons for belief are always evidential reasons and never reasons deriving from the goodness of believing something.29 The reasons specialist holds that this is a conceptual truth about reasons for belief. By way of contrast, a generalist might still hold that there are no goodnessderived reasons for belief, but that this is not a conceptual truth. Reasons generalism is the negation of reasons specialism. It holds that it is conceptually possible for any source of normativity to issue any type of reason, even if it is the case that there are no actual instances of a particular type of reasoning issuing from a particular source of normativity. This distinction is more easily discussed in context and is elaborated on more fully in }4.2.2. One preliminary difficulty with the distinction is that there are some reasons generalist views that will not have certain basic sources of normativity applying to all types of reasons, and there could be a reasons specialist view in which all sources of normativity are bases for all types of reasons. As an instance of the former, it seems possible that while there could be goodness-based reasons for belief, it is less clear whether there are any evidence-based basic reasons for action, other than perhaps for speech acts.30 Whether such a position is genuinely a reasons generalist one depends on why there is not a match between a particular basic source of normativity and a type of reasons. This matter requires more explanation, which will be provided in }4.2.2. 4.2.1 architecture 1: normative monism and reasons generalism Having presented these two axes, it is now possible to discuss the various possible architectures of normativity. The first architecture is normative monism and reasons generalism (NMRG). The normative monism part says that there is only one source of normativity. The reasons generalism part says that all sources of normativity in principle can apply to all types of reasons. An example of a theory that conforms to NMRG is radical pragmatism. A radical pragmatist thinks that the only reason for anything, be it a belief, an action, or a feeling, is that the thing in question will lead to an increase in overall goodness. 29

See Adler (2002); Kelly (2002); Shah, (2006). Michael Zimmerman suggests in correspondence that there may well be evidential reasons for action. Speech acts may well be governed by evidential reasons. Actions which are not technically speech acts, but which serve communicative functions may also be governed, or partially governed, by evidential reasons. 30

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

199

NMRG has powerful resources for resolving putative inter-domain conflicts, for example, putative conflicts between the requirements of theoretical reason and those of practical reason. To see why, consider what the normative world would be like if the only source of normativity were goodness. A fact would be a reason to do or believe something for an agent only when that fact made it so that the performance of the action or the holding of the belief would increase total goodness. Furthermore, we could assign to each reason a weight based on how much better the world would be if the action proposition or belief proposition for which there was a reason obtained. In this circumstance, figuring out what we ought to do or what there is most reason to do would be a matter of weighing reasons deriving from the same source. On this very simple theory, how much reason there is depends entirely on how good the outcome is of doing or believing what there is a reason to do or believe. One ought to believe what there is most reason to believe if that produces a better result than doing what there is most reason to do, and vice-versa. This is not to say that NMRG does not admit any troubling cases. Presumably, as there is sometimes thought to be incommensurability, or more importantly, incomparability, in value, there might also be incomparability amongst reasons deriving from a single source of normativity. This incomparability might occur, if the source of normativity were value, and the reasons that issued from it kept the same relational structure as the underlying value source. Instances, if there are any, of genuinely incomparable values would issue in incomparable reasons. Despite this difficulty, there would at least be no mystery about which feature to compare when evaluating different types of reasons; all types of reasons would have the same source of normativity. To find out what one finally ought to do, one would weigh up all reasons with respect to their single source of normativity. If all reasons came from goodness, then we just have to look at how to weigh up goodness-based reasons. 4.2.2 architecture 2: normative monism and reasons specialism The second possible normative architecture is normative monism and reasons specialism (NMRS). The normative monism part says that there is only one source of normativity and the reasons specialism part says that each type of reason is distinctively conceptually connected to its sources of normativity. NMRS may be a less plausible normative architecture, because it may rule out the possibility of there being certain types of reasons. Consider a form of NMRS in which the sole source of normativity is evidence. While the link between evidence and belief has much plausibility, it is less clear what the conceptual relationship would be between there being evidence for something and one’s having reason to do it.31 Of

31

The strength of this worry depends on one’s scepticism about there being evidential reasons for action.

200

ANDREW REISNER

course, this problem applies to NMRG, too, but not all sources of normativity will prove equally problematic for NMRG as they do for NMRS. Looking at why some cases are more difficult for NMRS than they are for NMRG provides an opportunity to elucidate the distinction between reasons generalism and reasons specialism. Reasons generalism and reasons specialism distinguish two different conceptual connections between sources of normativity and types of reasons. Reasons specialism says that each type of reason has a source or sources of normativity that have an appropriate conceptual connection to it. The constraints on these connections could vary from theory to theory, but the important contrast is with reasons generalism, which in principle allows any source of normativity to be a basis for any type of reasons. It may be discovered a posteriori that some sources of normativity in fact generate no reasons of a certain type, but this is not because reasons generalism excludes a priori that source of normativity from being a basis for a particular type of reason. As has already been discussed, one view that NMRG can accommodate easily is radical pragmatism. While one might need to make the case for pragmatism, one does not need to show that there is some particular type of connection between belief and goodness that underwrites goodness’s being a reason for belief. In an NMRS architecture, however, it is not sufficient just to show that goodness is the source of normativity. It also must be shown that goodness connects in a conceptually appropriate way with belief, so that there can be goodness-derived reasons for belief. If this connection does not hold, then in an NMRS architecture, there will be no reasons for belief that derive from goodness.32 What is also interesting to note is that NMRS and NMRG can, with the same source of normativity and the same reason types, both yield the result that there are no reasons of certain types. This point is made clearer by an example. A normative monist could claim that the only source of normativity is evidence. If that normative monist were a reasons specialist, then, by definition, evidence could only be a basis for a type of reason that connected with it in the conceptually correct way. Most plausibly, reasons for belief connect appropriately to evidence. At the same time, there is no such clear relationship between truth and action. In this NMRS theory, there would be no reasons for action, because evidence lacks the kind of conceptual connection with action that it has with belief. One might get the result that there was only one type of reason, too, in an NMRG theory that takes the only source of normativity to be evidence and which holds that there are only two types of reasons: reasons for belief and reasons for action. Evidence

It is worth being a bit more precise about locutions like ‘what one would need to show’. In the context of discussion here, it is a stipulation of a generalist view that all sources of normativity can apply to all types of reasons. So, from a theory-internal perspective, there is no need for a further explanation as to why goodness can be a source for reasons for belief. One might need, independently, to show that generalism is true. 32

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

201

can serve as a basis for reasons for belief in an NMRG theory, as in an NMRG theory the source of normativity can in principle be a basis for any type of reason. However, an NMRG theorist might discover a posteriori that there are no evidential reasons for action. So, although there is nothing in the structure of the NMRG theory to rule out the possibility that evidence could serve as a reason for an action, there are in fact no evidential reasons for action. While the NMRG theory and the NMRS theory yield the same result with these inputs—that the only source of normativity is evidence and that there are only reasons for belief—they yield that result for very different reasons. 4.2.3 architecture 3: normative pluralism and reasons generalism Normative pluralism is the view that there is more than one source of normativity. Normative pluralism and reasons generalism (NPRG) is the view that there is more than one source of normativity and that all sources of normativity can in principle serve as bases for all types of reasons. Consider the normative universe in which there are two sources of normativity, goodness and evidence, and also two types of reasons, those for action and those for belief. If this normative universe has an NPRG structure, then it is open conceptually for there to be evidential reasons to act and also to believe, as well as goodness-based reasons to act or believe. It is not entirely clear how strong NPRG’s logical resources are for dealing with competing normative claims amongst different types of reasons. Initially it may look as though NPRG has powerful resources for resolving putative normative conflicts, as it shares the main advantage of NMRG, namely that all types of normativity apply to all reasons. This may be an optimistic assessment of NPRG. We can look first at an example of a theory compatible with NMRG, radical pragmatism, as a contrast case, in which possible inter-normative conflicts are just possible conflicts between goodness-based reasons. Such conflicts are addressable by whatever means are available for weighing only goodness-based reasons, even when weighing up reasons to believe against those to act.33 In our toy version of NPRG, on the other hand, one faces the more daunting prospect of explaining how to weigh up evidential reasons for belief against goodness-based reasons for action. We may think that all reasons that come from goodness have enough in common to find ways of coping with putative conflicts between goodness-based reasons for action and goodness-based reasons for belief. However, it is not at all clear what to do with putative normative conflicts, whether they be within belief, within action, or between the two, that involve reasons with different sources of normativity. We might expect a theory of goodness or a theory of evidence to be structured in such a way that they can manage its own problems, so to speak, but it is not clear why a theory of either would have much to say about how its own object relates in a non-derivative way to the object of the other. 33 Consider a situation in which one has reasons to cause oneself to believe p, but one only has reasons to believe not p. Because ‘cause’ is a success verb, one’s reasons to act and one’s reasons to believe cannot both be satisfied.

202

ANDREW REISNER

Our toy NPRG theory requires a method for weighing or comparing goodnessbased reasons and evidence-based reasons, if it is to provide a solution for putative normative conflicts involving reasons generated from both normative sources. There are at least two roads that NPRG might go in attempting to deal with possible conflicts where the reasons involved have different normative sources. The first looks the more promising one. That road is to insist, against worries about comparability, that instead there is nothing unexpected about the idea that a theory of normativity tells us how to relate the plurality of normative sources. One expectation of a theory of value is that it tells us how a plurality of goods are related to each other, even if it only says that they are incommensurable. Likewise, a theory of evidence might be expected to tell us how a plurality of different sorts of considerations affect the likelihood of a proposition’s being true.34 Analogously, the theory of normativity tells us how all the sorts of normative sources relate to each other, and in that way is like any other theory that has a plurality of ontologically basic properties that are apt for comparison or some form of aggregation. The second road takes us in the opposite direction and arrives at the view that there is something wrong with expecting a theory of normativity to do as well at resolving the problem of comparing reasons with different respective normative sources as does a theory of value at addressing how to compare different values. On this view, it seems that for different sources of normativity to be comparable, there has to be something in virtue of which they can be compared. I do not mean this as the start of a third man argument. Rather the worry is that the very idea that we could compare two different sources of normativity suggests that either one is derivative of the other, and hence not basic, or that they are both derivative of a third source of normativity. Because both of these options are ruled out by stipulation from NPRG, this road threatens to lead to the conclusion that there are genuine normative conflicts. The compatibility of an NPRG theory with the view that there are no genuine normative conflicts depends largely on the plausibility of making brute comparisons amongst reasons arising form different sources. 4.2.4 architecture 4: normative pluralism and reasons specialism The final normative framework discussed here is the one that commonly is taken for granted in the literature, although not specifically under this description. This is normative pluralism and reasons specialism (NPRS).35 Normative pluralism says

34

For an interesting discussion of the complexities of assessing evidence, see Achinstein (2003). A particularly detailed version of NPRS is set out in Skorupski (2011). It is the view that seems to be implicitly adopted in Parfit (2001), and indeed by philosophers who adopt strict evidentialism about normative reasons for belief or who argue that the solution to the so-called ‘wrong kind of reason problem’ is to deny that there are reasons of the wrong kind at all. 35

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

203

that there are a plurality of sources of normativity, and reasons specialism tells us that the sources of normativity must have the right sort of conceptual link with a type of reason, in order to serve as a source for that type of reason. The position I have earlier identified as ‘normative separatism’ is a variety of NPRS, as it says that in principle only a certain source of normativity, for example, evidence, has the right kind of conceptual link with a particular type of reason, for example, reasons for belief, for each type of reason. A caveat is required in saying that normative separatism is a species of NPRS. It is possible to have a mixed normative architecture, one in which some types of reasons must bear the appropriate special conceptual connection with their sources of normativity, while other types of reasons do not require this special connection. A normative architecture might say that as far as reasons for belief go, reasons specialism is true. On the other hand, as far as other types of reasons are concerned, there need not be any special conceptual connection between the reasons and the source of normativity from which they issue. For the sake of simplicity, only pure NPRS will be discussed. On NPRS views, the domains of normativity are distinct. Theoretical normativity and practical normativity are two in some respects analogous, but non-interacting, domains. This claim may sound strong, given that it is possible in principle to have the same source of normativity be a basis for different types of reasons. The unrelatability comes from the requirement that a source of normativity have a distinctive conceptual connection with the type of reason it serves as a basis for. Benefit, for example, might be a source of normativity for both action and feeling. That doing x is beneficial is a reason to do it. It may also be a reason to feel good about yourself for doing it. However, the conceptual connections between acting and benefit and between feeling good about oneself and benefit are understood to be quite different, as acting concerns promoting benefit and feeling good about oneself involves responding to the fact that someone or something has been benefited by your actions. Given this, NPRS lacks the logical resources to say anything at all about putative conflicts between our reasons for actions and our reasons for belief, because they are wholly different sorts of things ex hypothesi. Asking what it is that one ought to do (in the universal sense of ‘do’ stipulated earlier), when one cannot act as one ought to act while also believing what one ought to believe, is, on NPRS, a meaningless question. The upshot of NPRS is that there are no possible resolutions to inter-normative conflicts, because there is no unified domain of normativity that subsumes both theoretical and practical normativity. This is normative separatism. There is an important issue that has not yet been discussed, one which relates to reasons specialism. This is a discussion of what constitutes an appropriate conceptual connection. Reasons specialism requires there to be an appropriate conceptual connection between a type of reason and the source or sources of normativity from which

204

ANDREW REISNER

it derives. There are many possible conceptual connections, and merely identifying that reasons specialism requires the appropriate one is not very illuminating. It is not clear how informative the account of an appropriate conceptual connection can be.36 I have used ‘distinctive’ to denote whatever that connection is. Consider reasons for thick pro-attitudes. Admiring is a good example. The source of normativity that most obviously has a close conceptual connection to admiring is admirability. We might be able to pick out some common features of things that are admirable, but admirable itself is a normative concept: it applies to the sort of thing that one ought to admire and does not obviously reduce to a cluster of other concepts.37 It may very well be a case by case question as to whether or not a particular thing is admirable. There may be no possibility of giving general principles, except as rough guides. As an example of a thin normative concept, it might generally be thought that the fact that an action will produce some benefit is a reason to do it. Here there is some degree of mystery as to why benefit has the right kind of conceptual connection with actions per se, as opposed to with anything for which there can be a reason. In general, that something would produce beneficial results seems to count in favour of it. Yet, this claim is not supposed to be fully general. On the one hand, when beneficial outcomes are the result of an action, there is an appropriate conceptual connection between benefit and one’s reason to act. On the other hand, when the beneficial outcomes are the result of having a belief, there is supposedly not the correct conceptual connection between one’s reason to believe something and its being beneficial to do so. An important task for the separatist or any reasons specialist is to explain two things. First, what these appropriate conceptual connections are, and, second, why they are significant.

5 Conclusion In working on the theory of normativity, there is a methodological balance between trying to work on the structure of the theory as an abstraction, seeing what follows for substantive claims about oughts and reasons, and trying to decide what is true about oughts and reasons in order to determine the structure of the theory. Working from either direction, or indeed both, it is of central importance to see how substantive claims about oughts and reasons limit (or are limited by) the general structure of a normative theory. I have tried to show that one important limiting factor in thinking about the structure, or as I have called it, ‘architecture’ of a theory of normativity is what view one takes on normative conflicts. There is a longstanding debate about whether there

36 Difficulties with specifying the conceptual connection can be seen from the discussion of ‘correctness’ in Danielsson and Olson (2007). 37 I am nonetheless sympathetic to Hurka’s account of thick value concepts, in which the -able concepts are analysable. See Hurka (2003).

NORMATIVE CONFLICTS AND STRUCTURE OF NORMATIVITY

205

are normative conflicts, although it has traditionally been played out only with respect to conflicts between prudence and morality, or with respect to the question of whether there are tragic moral dilemmas. Authors have been willing to entertain a variety of positions to avoid normative conflicts, but it is often unclear whether they are only concerned to avoid intra-domain conflicts or both intra-domain and interdomain conflicts. Intra-domain conflicts make the most sense on reasons pluralist views. There is less reason to expect there to be genuine intra-domain conflicts on reasons specialist views, since it is a question of comparing one kind of reason from a single source with another reason of the same kind from a single source. If one holds that there are intractable intra-domain conflicts, then it looks likely that one is committed to a pluralist architecture. The converse does not hold, however, as one may think that there are no genuine intra-domain conflicts on either specialist or generalist architectures. If one holds the view that there are no genuine inter-domain normative conflicts, then NPRS views (and the separatism they entail) are ruled out. Normative conflicts occur, when there multiple final oughts cannot be mutually satisfied. As the examples above show, this can happen quite easily with NPRS. A worry is that any ranking amongst the types of oughts (practical, theoretical, affective, etc.) would be hard to come by unless there was an appeal to another more basic ought that subsumed the others. To make this appeal would be to deny normative pluralism. Normative pluralism, and the normative separatism that follows from it, are incompatible with a view of normativity that denies the existence of genuine inter-domain conflicts. Conversely, putative inter-domain conflicts are most easily resolved in principle if there is a single source of normativity that applies to all reasons. If there are strong independent arguments in favour of the view that there are no genuine inter-domain normative conflicts, then they may give us some reason to take a view like radical pragmatism about normative reasons seriously. The answers to the important questions of whether there are genuine normative conflicts and, if so what kind, rise and fall together with what turns out to be the true architecture of normativity. Amongst other insights, this shows us that one of the more widely held views concerning reasons, normative separatism, is not compatible with the common intuition that there are no genuine conflicts amongst final oughts, whereas the latter view fits well with one of the least widely held views about reasons: radical pragmatism.

References Achinstein, P. (2003) The Book of Evidence. Oxford: Oxford University Press. Adler, J. (2002) Belief ’s Own Ethics. Cambridge: Bradford Books. Broome, J. (1999) Normative requirements. Ratio 12(4): 398–419. Broome, J. (2004) Weighing Lives. Oxford: Oxford University Press.

206

ANDREW REISNER

Broome, J. (2005) Does rationality give us reasons? Philosophical Issues 15(1): 321–37. Broome, J. (2007) Requirements. In Hommage a` Wlodek, . Broome, J. (2013) Rationality through Reasoning. London: Blackwell. Cariani, F. (2013) Ought and resolution semantics. Noûs 47: 534–58. Danielsson, Sven, and Olson, Jonas (2007) Brentano and the buck-passers. Mind 116 (463): 511–22. Guindon, B. (n.d.) Sources, reasons, and requirements. Forrester, James W. (1996) Being Good and Being Logical. Armonk, New York: M. E. Sharpe. Hieronymi, P. (2005) The wrong kind of reason. The Journal of Philosophy 102: 437–57. Hurka, Thomas (2003) Virtue, Vice, and Value. New York: Oxford University Press. Kearns, S. and Star, D. (2008) Reasons: Explanations of evidence? Ethics, 118, 31–56. Kelly, T. (2002) The rationality of belief and some other propositional attitudes. Journal of Philosophical Research 110: 163–96. McLeod, O. (2001) Just plain ought. The Journal of Ethics 5: 269–91. Parfit, D. (2001) Reasons and rationality. In D. Egonsson, J. Josefsson, B. Petersson, and T. Rønnow-Rasmussen (eds.) Exploring Practical Philosophy. Burlington, VT: Ashgate, 17–39. Piller, C. (2006) Content-related and attitude-related reasons for preference. Philosophy 59 (suppl.): 155–218. Reisner, A. (2004) Conflicts of Normativity. Doctoral Thesis, Oxford University. Reisner, A. (2008) Weighing pragmatic and evidential reasons for belief. Philosophical Studies 138: 17–27. Reisner, A. (2009) The possibility of pragmatic reasons for belief and the wrong kind of reason problem. Philosophical Studies 145: 257–72. Shah, N. (2006) A new argument for evidentialism. The Philosophical Quarterly 56: 481–98. Shah, N., and Velleman, J. D. (2005) Doxastic deliberation. The Philosophical Review 114(4): 497–534. Skorupski, J. (2011) The Domain of Reasons. Oxford: Oxford University Press. Stich, S. (1993) The Fragmentation of Reason: Preface to a Pragmatic Theory of Cognitive Evaluation. Cambridge: Bradford Books. Wedgwood, R. (2007) The Nature of Normativity. Oxford: Oxford University Press.

14 Reasons and Rationality The Case of Group Agents Lara Buchak and Philip Pettit

Introduction John Broome has been very influential in arguing for a distinction between requirements of rationality—for example, the requirement to be consistent—and reasons: that is, objective reasons of the kind that evidence provides for belief (Broome 1999, 2001a, 2001b, 2004, 2005).1 One of the many debates that he has opened up, then, bears on how the requirements of rationality relate to reasons and whether in particular there is any reason to fulfill those requirements. Niko Kolodny has argued that there is not, defending an error theory about requirements of rationality. His claim is that the only genuine requirement is sensitivity to reasons and it is a mistake to think that there are independent requirements of rationality (see Kolodny 2005, 2007a, 2008a, 2008b). After an exchange with Kolodny on the matter (Broome 2007; Kolodny 2007b), Broome’s (2013: 213) current view is that, while the requirements of rationality may have an independent claim on us, it remains an open question whether this is so. We do not try to resolve this difference of view in our chapter. What we want to do rather is to bring some independent material to bear on the issue, using recent work in the theory of group agency. It turns out that Kolodny’s error theory, when applied to group agency, runs into conflict with some well-known results in that theory and we think that exploring the ways in which the conflict can be removed may be of interest. However the conflict is resolved, it will require a revision either to the error theory or to standard views of group agency.

1 Agency Let an individual agent be a system that has certain purposes or goals and certain representations or beliefs—in particular, representations that are reliably responsive 1

Authors’ names are in alphabetical order only.

208

LARA BUCHAK AND PHILIP PETTIT

to evidence—and that acts reliably for those purposes according to those representations: that is, in a way that promises success if reality corresponds to the representations. An agent may occasionally fail to form representations in an evidentially reliable way or to act for its purposes in an executively reliable manner but such failure ought to be more or less unusual. And even if such failure does occur, the agent ought at least to be capable of recognizing the failure and so far as possible putting it right. On the face of it, there are two types of requirements on agents. The first are requirements of reason or “reasons-sensitivity”, the second requirements of rationality or “coherence”. Requirements of reasons-sensitivity concern the relationship between what the world is like—how it presents itself as being—and which attitudes one should hold. For each attitude, there are reasons specific to that attitude that dictate whether one should hold it. In the case of belief, evidence for p provides a reason in favor of believing p, though there might be contrary reasons against believing p. If one’s evidence gives one conclusive reason to believe that p, then one ought to believe that p. In the case of action and intention to act, the perceived attraction or desirability of realizing A might give one a reason to do A, and again there might be contrary considerations that give one reason not to do A. If one has conclusive reason to A, then one ought to intend A.2 Requirements of coherence concern, not the relationship between what the world is like and what attitudes one should hold, but rather the relationship among one’s attitudes themselves. One ought not believe p and at the same time believe not-p. One ought not at the same time believe p, believe p entails q, and believe not-q. One ought not at the same time intend A and believe that B is a necessary means to A, but fail to intend B. That forming attitudes on the basis of attitude-specific reasons is a requirement of agency seems clear: an attitude ought to respond to the reasons that bear on whether one should hold it. More specifically, it ought to respond to the features of its object that make the attitude appropriate, not to features of the state itself that make holding it attractive or desirable (see Parfit 2001). Thus the belief that p ought to form on the basis of whether or not the evidence supports p, not on the basis of whether it is

2 These claims are controversial. In the case of belief, some argue that there are cases in which one ought to believe p for non-epistemic reasons: see, for example, discussions in Booth (2012), Marusˇic´ (2012), and Reisner (2008, 2009). In the case of intention, some have proposed that there are cases in which one has conclusive reason to not-A but nevertheless has conclusive reason to intend A rather than not-A. A classic example is Kavka’s (1983) toxin puzzle, but more mundane examples appear in Morauta (2010). The primary focus of both discussions is cases in which one being in the state of believing, or in the state of intending, furthers one’s goal—in contrast with cases in which one has evidence that the belief is true or in which the intended action furthers one’s goal. However, in these cases, the way in which being in the relevant state furthers one’s goal is not that being in the relevant state furthers the goal of coherence. Put succinctly: these cases concern an agent’s response to facts about what the world is like, not the relationship among attitudes of a given type that the agent has. Thus, those who reject our above assumptions should still find something of interest in our argument.

REASONS AND RATIONALITY

209

profitable or prudent to believe that p. The features of the world that provide reasons fit the aim of the attitude, as we might put it. Belief aims at truth, and a belief is more apt to be true insofar as one has evidence for it; intention aims at what is most attractive or desirable among available options, and an intention is more apt to achieve that aim insofar as it tracks perceived attraction or desirability.3 It is less obvious why agents ought to be coherent, as Kolodny points out. For example, when we consider the case of belief, it is not obvious that a coherent agent will be more apt to have true beliefs than an incoherent agent; coherence is as easily attained among false beliefs as among true. Similarly, for intention, it is not obvious that the agent with coherent intentions will be more apt to fulfill his desires or obtain the good. Nevertheless, we do all seem to feel the pull of coherence requirements: we think they ought to guide our behavior, and we think that in complying with them we have done something right. Let us assume with Broome and Kolodny—and indeed with common sense—that we use both reasons-sensitivity and coherence to guide our attitudes. The puzzle, then, is to explain either how these two sets of requirements can coexist or to explain how one set of requirements (or apparent requirements) reduces to the other: most plausibly, how the (putative) requirements of coherence reduce to the requirements of reason. In pursuing this project here, we start from Kolodny’s arguments in the case of individuals and show how they lead to problems in the case of group agents. We do not try to connect our suggestions to Broome’s (2013) more recent work.

2 Individual Coherence This chapter will focus only on belief, though the parallel issue for intention is also worth considering. Kolodny argues that the requirements of coherence for belief fall out of the requirements of reason, not in the straightforward sense that one has a reason to be coherent as such, but in the sense that requirements of coherence are a byproduct of requirements of reason: they are not genuine requirements in themselves, but the genuine requirements of reason make it possible to explain why we think in error that they are. Let us consider his argument that focuses on the putative coherence requirement of non-contradiction:4 Non-contradiction (N). One is rationally required (if at t one believes p, then at t one does not believe not-p).

3 Does Broome (2013) agree? While he thinks that there are distinct requirements of rationality, the evidential and enkratic conditions that he mentions on page 98, in discussing ‘Scanlon’s condition’, correspond well to our conditions here. He thinks that these need qualification but that they hold ‘under suitable interpretations’. 4 This primarily follows Kolodny (2007a), and a similar argument can be found in Kolodny (2008b). All named principles in this section are Kolodny’s.

210

LARA BUCHAK AND PHILIP PETTIT

Notice that (N) has a wide-scope character. It ought to be, according to (N), that one does not believe p at t or that one does not believe not-p at t; the requirement bears on the disjunction, not on the disjuncts it contains (Broome 1999). The puzzle, Kolodny explains, is that (N) seems to us like a rational requirement but that we lack a good argument for why this is so: it seems that we have no reason to conform to the dictates of (N). However, if we can explain the seeming force of (N) by appeal to principles about reasons, then we can resolve the puzzle by concluding that (N) is not a genuine requirement. Recall that reasons concern the relationship between an agent’s evidence and her beliefs. Kolodny points out that an evidential situation can tell three ways with respect to p. First, it could be the case that the evidence supports p strongly enough that reason dictates one ought to believe p, and that the evidence fails to support notp strongly enough, so that reason dictates one ought not to believe not-p. Second, the reverse might hold. Or, finally, the evidence could be such that it fails to support p strongly enough and fails to support not-p strongly enough, so that reason dictates one ought not to believe either proposition. That these possibilities are exhaustive means that no evidential situation gives one conclusive reason to believe p and at the same time gives one conclusive reason to believe not-p; there will be conclusive reason to believe only one, or to believe neither. The following principle, in which the disjunction is inclusive—the “or” has the sense of “and/or”—sums up the lesson: Reason patterns (R). In any given situation, either it will be the case that (one is required by reason not to believe p), or it will be the case that (one is required by reason not to believe not-p). (R) is easily confused with (N), so if (R) explains all the data about why we might be inclined to accept (N), then we can reject (N) but also explain why we mistakenly thought it was correct. So, we might say that the data we thought pointed to a separate requirement of structural coherence are explained by a structural feature of the norms of reason. Unfortunately, as Kolodny points out, this is too simple: (R) does not explain all the data that we thought justified (N). According to Kolodny, the data that we thought justified (N) are that we seem to accept the following two principles: Violation claim about non-contradiction (VN). If one believes at t that p and believes at t that not-p, then one violates some requirement. Satisfaction claim about non-contradiction (SN). Suppose that one believes p and believes not-p. If one either ceases to believe p, or ceases to believe not-p, then one thereby satisfies some requirement that one would not satisfy if one continued both to believe p and to believe not-p. (R) can certainly explain (VN): if one is correctly responding to reasons, then one can never believe at t that p and believe at t that not-p, since one of these two beliefs must violate the dictates of reason. So it seems like we can explain (VN) without

REASONS AND RATIONALITY

211

appeal to (N). But what about (SN)? The problem is that one satisfies some requirement by giving up one of two conflicting beliefs even if one gives up the wrong belief: the belief unjustified by one’s evidence. Therefore, one can satisfy the requirement outlined by (SN) without satisfying the requirements that reason places on responding to one’s evidence. However, there is another way to explain how satisfying the requirements of reason has (SN) as a byproduct. Kolodny points out that if one finds oneself believing both p and not-p, and one realizes this, then (because one accepts (R)) one has a “second-order” reason to believe that one either believes p without sufficient reason, or one believes not-p without sufficient reason. One will respond to this either by concluding of one of these two beliefs that one lacks sufficient reason to hold it, in which case one must revise that belief on pain of defying one’s own judgment; or by not reaching a conclusion about which belief(s) one lacks sufficient reason for, in which case one must suspend judgment on pain of defying one’s own judgment. It is irrational to defy one’s own judgment in this way, as codified by: Believed reason (BR). If one believes at t that reason requires one to have attitude A, then one is rationally required to form or sustain, going forward at t, on the basis of this belief, A, and if one believes at t that reason does not permit one to have A, then one is rationally required to revise or refrain from forming, on the basis of this belief, going forward from t, A, and if one is deliberating at t, in response to a live doubt, whether reason permits one to have A, but has not yet concluded that it does, then one is rationally required to revise or refrain from forming, going forward from t, A. So, in revising one of the contradictory beliefs, even in revising the wrong one, one fulfills a second-order requirement of reason in conjunction with (BR).5 In passing, it may be worth remarking the following. Suppose that you realize that you believe p and you believe not-p, recognize that this shows you that as yet you do not have sufficient reason to believe p or do not have sufficient reason to believe notp, but on reflection cannot find any new evidence or any reason to re-construe your existing evidence. In that case the fact of your incoherence is going to give you reason to suspend belief on whether or not p. We mention this point now because it will be relevant to discussion later in the chapter. Thus, Kolodny concludes, (N) does not describe a real requirement, but the fact that we are inclined to think it does can be explained by real requirements: the

5 An argument that (N) follows directly from the requirements of reason is due to Reisner (2011). His argument rests on the assumption that if evidence gives one reason not to believe p& q, then evidence gives one reason not to (believe p and believe q). We follow Kolodny in taking the requirements of reason to pertain to particular beliefs, rather than to conjunctions of beliefs.

212

LARA BUCHAK AND PHILIP PETTIT

requirements of reason. In other words he defends an error theory about the coherence requirement of non-contradiction that (N) formulates. And although (N) is only one of the purported requirements of coherence, presumably a parallel dialectic shows that other purported coherence requirements can be explained by the requirements of reasons-sensitivity. The coherence requirement we will employ in later argument is conjunction: one ought not at the same time (believe p, believe q, and believe not-(p& q)). The parallel argument in this case would again invoke a structural feature of evidence-based reasons: that one’s evidence cannot simultaneously give one conclusive reason to believe that p and conclusive reasons to believe that q, but also give one conclusive reason to believe not-(p& q). Thus, so the argument would go, finding out that one is in the incoherent state gives one second-order reason to revise one’s judgments, and reason requires that one adopt the attitude that one thinks is correct during and after the revision process. We recognize that conjunction is a more controversial requirement than noncontradiction, because of paradoxes like the preface paradox and the lottery paradox (see Kolodny 2007a: 253–7). For example, someone who holds that belief is credence above a certain threshold would deny that conjunction is a requirement of coherence precisely because evidence lacks the relevant structural feature: assuming the threshold is low enough, evidence can sometimes simultaneously justify credence above the threshold for p, q, and not-(p& q). Suppose a theory sets the threshold for belief at 0.6 and that you give a credence of just that threshold level to p and to q. That will mean that the credence you should give to p& q, assuming they are independent, is 0.36, and the credence you should give to not-(p& q) is 0.64. Thus you ought not believe p& q and you ought to believe not-(p& q). While recognizing that our discussion will have less to say to those who deny conjunction, we note two things in mitigation. First, all that we need for the discussion here is that if you believe each of two atomic propositions, you ought not to believe the negation of their conjunction—which is a much weaker assumption than that needed to generate the preface and lottery paradoxes. Second, the discursive dilemma (which will be discussed in the next section) can be run for modus ponens and other inference rules, in addition to conjunction; since the group impossibility results are stated in terms of judgments about conjunctions of propositions, it makes things simpler to focus on conjunction. Kolodny’s argument concerning the relationship between reasons and coherence, if it is correct, supports three important claims about individual rationality. First, coherence itself is not a rational requirement: there is no reason to be coherent as such. Second, we only think coherence is a requirement because being incoherent is evidence that you have failed to satisfy a genuine requirement and because noticing one’s own incoherence gives one a reason to revise one’s judgment in some way that will make one coherent. And finally, if we look after reasons, coherence will look after itself.

REASONS AND RATIONALITY

213

3 Group Coherence Reducing the purported norms of coherence to those of reason seems plausible in the case of individual agents. But can a similar argument be run in the case of group agents? We think of group agents as collections of people who organize their relationships and roles—presumably, on a jointly intentional basis—in such a way that they mimic individual agents (see French 1984; List and Pettit 2011). They form representations of their environment that are responsive to how that environment presents itself, they endorse purposes that they aspire to achieve together, and they act in a way that advances those purposes according to those representations: the individuals who are given the role of acting for the group are constrained to meet this condition. Or at least they do so in general and, when they fail, they are responsive in general to correction. Such group agents may utilize many different procedures, some inclusive of all members, some more hierarchical in character, in determining what it is they endorse, pursue, and enact. And so examples of group agents range from the democratic association or club to the corporation or university, to the oligarchical church or state. As with individual human beings, there seem to be two types of requirements on group agents: those of reasons-sensitivity, and those of coherence. A group forms its beliefs—its judgments or opinions—in a way that responds to the evidence: presumably, the evidence available through its eyes and ears, that is, in the judgments or opinions of its members. And just as reasons-sensitivity requires that the beliefs of individuals should be responsive to evidence, so it requires that the beliefs of a group agent should be responsive to the beliefs of its members: they are the “evidencechannels” that mediate the relationship between the group’s attitudes and the world. But, moving from reasons to coherence, it is also essential for group agency that the entity sustained by its members should be capable of responding to challenges of incoherence. It will not function well as an agent, and will be incapable of forming contracts with other individual or group agents, if it is not organized in a way that guards against the formation of incoherent attitudes or that makes it sensitive, at the least, to evidence of incoherence. Let us consider the suggestion, then, parallel to Kolodny’s suggestion with individual agents, that coherence requirements for group agency are not genuine requirements and that their appeal can be explained away as a byproduct of the requirement of sensitivity to reasons. In pursuing this line, we will focus on conjunction, as already advertized: that is, the requirement that you ought to believe the conjunction of what you believe. This focus is convenient for our purposes, as will become apparent, and it does not significantly reduce the interest of the observations we make. Here are the relevant claims for the group case. The “coherence” claim is as follows: Group-conjunction (GC). A group agent is rationally required (if at t the group believes p and believes q, then at t the group does not believe not-(p& q).

214

LARA BUCHAK AND PHILIP PETTIT

Again, the data justifying our acceptance of (GC) seem to be: Group violation claim about conjunction (GVC). If a group believes at t that p and believes at t that q and believes at t that not-(p& q), then the group violates some requirement. Group satisfaction claim about conjunction (GSC). Suppose that a group believes p and believes q and believes not-(p& q). If the group either ceases to believe p, or ceases to believe q, or ceases to believe not-(p& q), then the group thereby satisfies some requirement that it would not satisfy if it continued to believe p and to believe q and to believe not-(p& q). Note, again, that in the case of (GSC), the group satisfies some requirement regardless of which belief it drops. In order to explain away (GC), showing that it is not a real requirement, we need to explain both of these other claims, (GVC) and (GSC), that constitute the data for (GC). The parallel move to that made in the individual case would be to explain (GC) by relying on the following claim about (group) reasons: Group reasons pattern (GR). In any given situation, either it will be the case that (the group agent is required by reason not to believe p), or it will be the case that (the group agent is required by reason not to believe q), or it will be the case that (the group agent is required by reason not to believe not-(p& q)). Take now a group in which the members play more or less equal parts, at least in the formation of attitudes. Presumably, for such a group to be required by reason to believe p is for the profile of member judgments or opinions to give the group conclusive reason to believe p. Here, the profile of individual opinions plays the role in the group case that evidence plays in the individual case: group votes in favor of p are evidence for p in the group case (they make it more likely that p), just as perceptual evidence and the like are evidence in the individual case. Thus, the suggested explanation requires that no profile of individual judgments or opinions—no pattern in group evidence—could give the group conclusive reason to believe p, conclusive reason to believe q, and conclusive reason to believe not-(p& q). Unfortunately, the discursive dilemma presents a problem for this suggestion (see Pettit 2001a, 2001b: ch. 5).6 Let us assume to begin with that what it is for an opinion profile to give the group conclusive reason to believe a proposition is for the majority to believe that proposition. Under that majoritarian assumption, there are some situations in which the group does have conclusive reason to believe each of the three relevant, inconsistent propositions. Consider a group with three members, facing the task of passing judgment on p, q, and not-(p& q), where the opinions of the members are as shown in Table 14.1. Note that the opinions of each of the group members are 6

The discursive dilemma is a generalization of the doctrinal paradox in legal theory (see Kornhauser and Sager 1993).

REASONS AND RATIONALITY

215

Table 14.1.

Member 1 Member 2 Member 3

p

q

not-(p& q)

Yes No Yes

No Yes Yes

Yes Yes No

consistent. And yet majority rule implies that the group has conclusive reason to believe p, to believe q, and to believe not-(p& q); each proposition gets two out of three votes. Thus, (GR) is false when we take the requirements of reason for groups to correspond to what the majority of individual group members believe. Furthermore, unlike in the individual case, if a group finds itself believing p, q, and not-(p& q), then noticing that it has these beliefs will not give the group second-order reason to believe it holds one of them without sufficient reason: that is, without the majority believing all three propositions. So (GSC) cannot be explained by the requirement that a group be responsive to its members, if being responsive to its members means believing what the majority of its members believe. This case illustrates the basic sort of conflict between the theory of group agency and the view that requirements of coherence can be explained away in terms of the requirements of reasons-sensitivity. But it is obvious that the case is relevant only given special assumptions: in particular, the assumption that responsiveness to reasons means conformity of the kind illustrated to the majority opinion of the membership. It turns out, however, that the conflict remains in place even when we broaden those assumptions. A number of recent results in the theory of judgment aggregation combine to support the following more general claim (see List and Pettit 2002; List and Polak 2010). If the members of a group have to make judgments on logically interconnected propositions like p, q, and p& q, then even when the judgment sets of each member are complete and consistent, there is no satisfactory function, majoritarian or otherwise, for generating a group judgment on each proposition from the member judgments on that proposition. The set of group judgments generated on such a bottom-up, proposition-by-proposition basis is liable not to be complete or not to be consistent. Or if the function for generating group judgments escapes that problem, then it is bound to breach some independently plausible constraint. It may treat some individuals as special, for example, perhaps giving one of them the status of a dictator; it may breach anonymity, as the constraint is sometimes known. Or it may treat some propositions as special, perhaps letting votes on those propositions dictate the group view on any proposition that follows from them; it may breach systematicity, as it is sometimes called, or more broadly, independence (for an overview, see List and Pettit 2011: ch. 2).

216

LARA BUCHAK AND PHILIP PETTIT

The upshot of these theorems in the theory of judgment aggregation is that a group agent cannot rely on a bottom-up, proposition-by-proposition way of generating a complete and consistent set of judgments on any interconnected set of propositions from the judgments of its members. And that is to say, to return to the conflict that concerns us here, that as long as we want our voting procedure to guarantee completeness, there is always liable to be a profile of individual opinions among group members that gives the group conclusive reason to believe p, conclusive reason to believe q, and conclusive reason to believe not-(p& q). It appears, then, that there is no way of explaining away the requirement of group coherence (GC) as a byproduct of purportedly the only real requirement on a group agent: that it should act on the reasons, satisfying the group reasons pattern (GR). Let the group be responsive to member votes on each proposition, as required by reasons-sensitivity, and it may well generate an incoherent set of group attitudes. It seems, then, that the claims about group agents and conjunction analogous to those about individual agents and non-contradiction at the end of the previous section are all false. We cannot explain away the intuition that coherence itself is a rational requirement—that there is reason to be coherent as such. Incoherence is not evidence that a group agent has failed to satisfy a requirement concerning the relationship between the opinions of its members and its group judgment. And it is not the case that if we look after reasons, coherence will look after itself. In a word, satisfying coherence is not a byproduct of satisfying reasons-sensitivity.

4 Group Agency One reaction to the discursive dilemma, and to the theorems in judgment aggregation, might be to think that group agency is impossible and so that the problem raised for analogues of Kolodny’s claims is spurious. It is important, however, to recognize that the theorems do not show that it is impossible to have group agents. They show only that if group agents are to form, then that cannot be on the basis of the bottom-up, proposition-by-proposition aggregation of member judgments into group judgments. We saw earlier that if a group is to constitute an agent then it must have purposes and representations—in particular, representations that are more or less reliably responsive to evidence—and must act so as to execute those purposes reliably, using the representations to guide its behavior. It must in general be evidentially and executively reliable. Or at the least it must display a sensitivity to failure, being able to acknowledge criticism for failures and being ready, when possible, to put them right. If a group is to constitute an agent in this sense, then it had better have a way of generating judgments or opinions that are more or less coherent, since incoherent judgments will cause it to seize up in some situations for action and will present it in any case as an entity with which it is impossible for other agents, individual or

REASONS AND RATIONALITY

217

corporate, to do business. Yet not only should those judgments be coherent; they should also be reasons-sensitive, in particular sensitive to evidence: in the case of the group, sensitive to the evidence provided by the judgments of its members. We have seen that it cannot hope to secure the desired coherence if it seeks to form its judgments on the basis of a bottom-up, proposition-by-proposition method of letting them reflect member judgments—if it seeks to be reasons-sensitive in that particular manner. So how then ought a group to proceed? Abstractly, what it must do is to make sure that as its judgments form in response to member judgments, they constitute a coherent whole. It must register the possibility that the judgments produced in a bottom-up, proposition-by-proposition way are liable not to make a coherent whole. And it must introduce a top-down, feedback procedure for checking on how far coherence is emerging and for inhibiting the bottom-up generation of judgments when it is likely to lead to incoherence. This abstract requirement may be concretely implemented in any of a number of ways, but the most straightforward and salient involves recourse to a straw-vote procedure (List and Pettit 2011: ch 2). Under a straw-vote procedure, the group would take a majority vote on each issue—or indeed a vote of any kind—as it comes up for resolution. But it would allow this bottom-up, proposition-by-proposition method to generate only candidates for group endorsement in judgment, not the final judgments themselves. After each vote it would check on whether or not the result would lead it into incoherence, as in the example of our group of three voting on whether (p& q); it would introduce top-down feedback about where the bottom-up procedure is leading. And given the threat of incoherence, it would isolate the minimal set of inconsistent propositions and decide as a group—there is no algorithm for how it might do this—to reject one member of that set; this might involve revising a past judgment or rejecting the judgment to which the most recent vote would lead it. Thus in the case of our sample group, it would lead the group to reject p or to reject q or to reject not-(p& q). Such a straw-vote procedure would not only revise the bottom-up aspect of the standard approach in judgment aggregation; it would also revise that approach in its proposition-by-proposition character. Did the group decide to hold that p, that q, and that (p& q), then its judgment that (p& q) would not be a majoritarian function of the judgments of members; after all, two of the members judge that not-(p& q). The procedure would lead the group to form a judgment on at least one proposition in a way that treats that proposition as less special than others; in the case envisaged, it would allow its judgment on (p& q) to be dictated by member judgments on p and on q, not by member judgments on that proposition itself. In technical terms, it would breach systematicity and independence. There are many other ways in which a group might allow the judgments of members to make an impact on its own judgments, thereby honoring reasonssensitivity, broadly speaking, yet still remain capable of achieving coherence in its

218

LARA BUCHAK AND PHILIP PETTIT

judgments overall. Thus it might let different subgroups decide on different sorts of propositions, for example, but authorize a particular subgroup to ensure that the propositions supported all constitute a consistent whole; this key subgroup might be capable of reversing the proposals of other subgroups or of sending back the proposals for reconsideration. Or it might authorize all the subgroups to debate with one another in the event of incoherence threatening. Or it might constrain the subgroups so that nothing can pass into effect until coherence is achieved.7 And so on; the possibilities are legion (List and Pettit 2011: ch 3). There will be no problem so long as the group as a whole has some procedure in place to ensure top-down feedback and correction in order to guard against incoherence. Could this top-down procedure be mechanized, thereby removing the need for active consideration of feedback by members or subgroups? The most obvious proposal would be to put a mechanism in place, should that be possible, to detect whether any vote introduces incoherence with past judgments and, in the event that it does, to reverse that most recent vote. But this proposal would mean that the group could never revise its past judgments and that what it came to believe would depend on the path it followed in considering issues. Such path-dependence would run against the spirit of reasons-sensitivity and argue against any proposal of that kind. It looks very likely that any group agent that hopes to achieve coherence in its attitudes—specifically, in the judgments or opinions it forms—must allow for active consideration of top-down feedback by its members. The upshot of these observations is that if a group agent is to form, then its members must not only pay attention to what reasons require, as they form their individual judgments and vote accordingly; they must also pay attention independently to how far sensitivity to those reasons is likely to lead them into incoherence as a group. What they do as a group has to be the product of their seeking on the one hand to meet the requirement of reasons-sensitivity and on the other to make sure that they fulfill this requirement only to the extent that that this allows them to achieve coherence in their group judgments. This result reinforces the upshot of our considerations in the last section. We saw in the last section that meeting the requirement of reasons-sensitivity does not ensure fulfillment of the requirement of coherence. We have seen in this section that the project of group agency, in a natural way of modeling it, positively demands that the members of a group agent should treat the requirement of coherence as important in its own right, indeed sufficiently important to argue for downgrading the requirement of reasons-sensitivity.

7 Or it might go quite a different path and introduce ‘distance-based aggregation functions’ which like the procedures described would also reject bottom-up, proposition-by-proposition determination of group attitudes (List and Pettit 2011: 57).

REASONS AND RATIONALITY

219

5 Escape Route 1: Group Agents Should be Practically Restricted We have seen that Kolodny’s claims for the requirement of reasons-sensitivity in relation to the requirement of coherence run into problems when extended in a natural way to the case of group agents. In this and the following two sections we look at three different ways in which those claims might be sustained in face of the threat. Each of these escape routes is associated with a debunking thesis about group agents. Putting them in order of increasing strength, the first holds that group agents should be practically restricted to an extent that removes the problem; the second that group agents should be epistemically restricted in way that removes the problem; and the third that, despite appearances, groups cannot constitute agents, or not at least agents in any proper sense. We turn now to the first of these theses. In outlining the means whereby a group agent might aggregate the judgments of members and yet avoid incoherence, we made the assumption that when the group confronts member votes that would support a set of incoherent judgments according to majority rule, it ought not to hesitate about forming judgments on the matters in question. Assuming that it has to form judgments one way or the other, we argued that the group ought to modify the judgments that member votes would support so that they constitute a consistent set. And we maintained that in doing this, it should take its guidance from the requirement of coherence, allowing this to combine with the requirement of reasons-sensitivity—in effect, the requirement of sensitivity to member votes—to determine what set of judgments it ultimately endorses. But it should be obvious that there is an escape route available at this very point for those like Kolodny, who want to argue that the only real requirement is that of reasons-sensitivity. They can say that the proper response for a group to take in face of the fact that majority rule is apt to lead to incoherent judgments is to use another rule, which in the case above will recommend suspending the group judgment on each of the propositions at issue. Let the member votes provide evidence (according to majority rule) that p, that q, and that not-(p& q), as in our example, and the line proposed is that the group should simply suspend judgment on each of those issues, because majority rule is too weak to capture the requirement of reasons-sensitivity, and so the group does not in fact have strong enough reason to believe any of the three propositions. There is no doubting the availability of this response for those who wish to stick with the claim that the only real requirement on any agent, individual or corporate, is reasons-sensitivity. But the cost associated with the response is to deny the usefulness, in effect, of incorporation and group agency. On the assumption that the members of any plausible group are likely to disagree in their judgments on particular propositions, even when they are each presented with the same evidence, the possibility of their facing discursive dilemmas under even very strong voting rules is ineradicable. And that means that a group agent that prioritizes reasons-sensitivity in

220

LARA BUCHAK AND PHILIP PETTIT

the manner recommended and that wants a procedure that guarantees that the discursive dilemma will not arise is likely to face many situations where it cannot make up its mind and must settle for neither accepting nor rejecting conflicting propositions. At first, this may not seem so bad: after all, most of us suspend judgment on a range of propositions we confront as individuals. Why not settle for the same line here? When people organize themselves as a group agent, they generally do so with a view to achieving purposes that are unavailable to them individually; these may be the purposes of a political party, a business corporation, a trade union, a voluntary association, a commission of inquiry, or an evangelical church. And that means that the judgments they wish to form as a group will almost all bear on theoretical questions that their purposes require them to resolve or on practical questions to do with how to prioritize those purposes, whether there are opportunities available for pursuing them and what are the best means to adopt in their pursuit. Thus the permanent possibility of not being able to form judgments on the issues the group confronts is likely to make for a serious vulnerability. Not only does the typical group agent endorse the pursuit of purposes that this possibility would jeopardize; it also has aspirations to being able to make contractual and other commitments that the possibility would undermine. Like individuals, group agents make promises and enter contracts with other agents, individual and corporate, and their credibility in this role is bound to be important for meeting the goals of their members. But if the fulfillment of any contract depends on the fortuity of the group’s not facing a discursive dilemma in processing relevant information—if fulfillment is not more or less robustly guaranteed—then the group is going to seem like a very unattractive partner in exchange. There are conditions that a group might satisfy such that under one of those conditions the group would have a robust capacity to form judgments on problematic propositions.8 But no such condition is itself likely to be robustly satisfied. To take the most obvious example, if the members were unanimous in their judgments on relevant propositions, then a bottom-up procedure in which the group judgment on any proposition is fixed by the member judgments on that proposition would work fine.9 But the chances that the members of a group could be effectively held to a

8

See List and Pettit (2011: 51–2) on unidimensional alignment. More generally, a similar result will hold if there is a certain supermajority in favor of each proposition: specifically, if more than (k 1)/k of its members believe that proposition, where k is the number of propositions to be voted on. See List and Pettit (2002: 106), who also cite List (2001). Imposing this requirement with our group of three would mean requiring that the group should believe a proposition only if more than two-thirds of its members agree to it, i.e. only if there is unanimous assent. This supermajority requirement will coincide with the unanimity requirement in any case where there are at least as many propositions as group members. 9

REASONS AND RATIONALITY

221

unanimity requirement without reducing their capacity to form judgments on the propositions they have to decide as a group must be very low indeed. The upshot of this discussion is that if group agents are to be effective systems for discharging their goals, thereby meeting the rationale for forming group agents in the first place, then they cannot be subject to the practical restriction that the purely epistemic use of coherence would impose. Those who are intent on maintaining that reasons-sensitivity is the one and only requirement on the formation of beliefs, and more generally of attitudes, can certainly stick to their position, availing themselves of this first escape route. But the cost they have to pay is that of rejecting the feasibility of group agents, or at least the feasibility of group agents that can claim to be more or less apt to reach a judgment on every issue relevant to their purposes.

6 Escape Route 2: Group Agents Should Be Epistemically Restricted The second escape route available to those who prioritize the requirement of reasonssensitivity, dismissing the requirement of coherence, holds that the votes of its members should not be allowed to count as evidence (except possibly derivatively) for what it ought to believe; they do not represent the demands of reason on the judgments or opinions it ought to form. And so the picture of group agency sketched in Section 3, in which member votes are the primary data to which the group must respond in making judgments, ought to be ruled out. Group agents should be not be given a license to invest the votes of members with such an evidential status; they ought to be epistemically more restricted. This approach comes in two forms, the first more moderate, the other more radical. The moderate form maintains that the evidence at the disposal of a group for the truth of a proposition ought not to be equated with the on–off votes of members on that proposition, but rather with the degrees of belief, ranging between 1 and 0, that members invest in it. The idea is that when members report just “yes” or “no” votes on each proposition they consider, they do not thereby furnish the group with the evidence that ought to determine the beliefs it forms. Rather they provide it with a second best: on–off votes that lose much of the information that would be available to the group, did it have access to their degrees of belief. Rather than learning that members vary in their degrees of belief in a range between 0.3 and 0.7, for example, the group envisaged forms beliefs that are shaped by an impoverished version of that information. When it responds to the votes of members, then, it should not be thought to be tracking the reasons that ought to determine its beliefs, only some highly standardized counterparts of such reasons. This first complaint suggests that group agents of the type traditionally envisaged are infeasible, or at least that they are not properly sensitive to the demands of evidence and reason. That pessimistic position would be enough to serve as an escape

222

LARA BUCHAK AND PHILIP PETTIT

route from the challenge posed for the view that there is no genuine requirement of coherence on agents: that it is merely a byproduct of the requirement of reasonssensitivity. It means after all that the group agents that seemed to pose a problem for that view are not really fitted to do so: they are not sensitive to evidence proper. It is natural to ask at this point whether there is a way in which group agents might be led to act on the basis of the degrees of belief that their members hold. But it turns out that the proposal faces serious problems. One problem is that there are well-known difficulties associated with aggregating degrees of belief, as there are difficulties in aggregating on–off beliefs, albeit not perhaps difficulties of the same seriousness (see Genest and Zidek 1986; Russell et al. (forthcoming)). A second problem with the proposal is even more striking. This is that in the vast majority of cases members cannot put fine-grained credences at the disposal of the group, offering them as inputs to a suitable aggregation function. By one analysis (Harman 1986), this is due to the fact that individuals do not have access to their degrees of belief at any fine level of grain. By another, it is due simply to the practical infeasibility of retrieving individual degrees of belief and using them in the formation of a group’s beliefs. But in any case it seems to tell quite strongly in favor of rejecting the idea that a group agent might operate by responding to the degrees of belief of its members when these must be stated explicitly and precisely. The second, more radical version of the claim that group agents ought to be epistemically restricted denies that the reports or votes of members, whether reflective of on–off judgments or degrees of belief, constitute the evidence on the basis of which a group ought to form its beliefs. The idea is that what reason requires of a group is that it should form its beliefs, not on the basis of the outputs from individuals in response to their own evidence, but on the basis of that very evidence itself: on the basis, in short, of the inputs that individual members confront. Recall that requirements of reasons-sensitivity concern the relationship between how the world presents itself and which attitudes one should hold. While evidence must be mediated through the votes of individuals in order to be accessed by the group, the demands of reasons-sensitivity require a group agent to respond appropriately to the inputs confronting its members, not to the voting outputs for which those individuals are responsible. Those evidential inputs bear directly on the truth or falsity of any relevant proposition, where the voting outputs of members have only a derivative bearing on that issue. The first approach suggested that a group agent that operates on the model of Section 4 does not avail itself of all the evidence it has at its disposal. This second approach, by contrast, would suggest that such a group agent does nothing wrong, rather that the model has been incorrectly applied to the phenomenon of reasonssensitivity, because we have incorrectly characterized what group agents are doing when they seek to take account of the votes of their members. According to our earlier characterization, the group agent purportedly tracks the demands of reason or evidence when it takes account of the votes of its members because those votes

REASONS AND RATIONALITY

223

constitute the evidence. However, according to the approach under consideration, in tracking the votes of its members, that group agent is not directly tracking the relevant evidence, but rather factors that tend to serve as indicators of what the evidence actually is. Support for this position comes from noticing that group members seek to bring their votes in line with how the world appears to them: the votes are meant to be an indication of what the evidence is, not the evidence itself. Does this re-characterization of the demands of reasons-sensitivity show that coherence need not play more than an epistemic role? Assume for simplicity that the same evidence is at the disposal of each member of a majoritarian group and that the votes support the incoherent triad: p, q, not-(p& q). On the proposed view of the evidence relevant for the group agent, this means that in parallel to the individual case, the group is entitled to conclude that, on some proposition, a majority of individuals must be failing to respond to the reasons the evidence provides, even if each member is individually coherent and the group is responding to the votes of its members. Thus, the group is failing to respond to the reasons on at least one proposition. As conceived within the proposal, the group agent should be concerned, not with the votes of members as such, but only with the evidence at the disposal of members that leads them to vote in this way. And if each member is exposed to the same evidence, and yet a majority vote in the manner that gives rise to a discursive dilemma, then on this view that means that there is at least one proposition on which a majority are failing to respond properly to the evidence at their individual disposal. Following this procedure, a group might be entitled to reject the majority view on one or other of the conflicting propositions, provided it could find independent evidence for determining the proposition on which the majority are mistaken. And in that case the procedure might show us a way for a group to deal with discursive dilemmas, form a group agent of the kind envisaged in Section 4, and yet not have to give the incoherence of the majority views anything other than an epistemic role. It would give it the role of indicating that there is some proposition on which a majority of the members are failing to respond to the evidence that is available in common to them, as in the individual case. Those who maintain this response argue that incoherence should not play the practical role envisaged in our account, guiding the group on the matter of how far it should respond to member votes—how far it should be reasons-sensitive—in actually forming beliefs on the propositions at issue. They say that incoherence should play the purely epistemic role that Kolodny ascribes to it in the case of the individual agent. It should serve to indicate to the group that it has a second-order reason in the case of each of the conflicting propositions to maintain that it believes that proposition without sufficient reason. In our simple example, it believes that p without sufficient reason, or it believes that q without sufficient reason, or it believes that not(p& q) without sufficient reason. It is hard to be optimistic, however, about this proposal. First of all, the proposed procedure would work for sure only in the special case where the evidence at the

224

LARA BUCHAK AND PHILIP PETTIT

disposal of members is the same. Let that assumption be lifted and it will no longer be clear that an incoherence in majority voting indicates that a majority are responding improperly to their evidence. The incoherence may simply be the result of the fact that different members confront different bodies of evidence; and yet it may be that those bodies of evidence, considered as a whole, would resolve the conflict. Second, it is highly optimistic to think that even in the simplified case where all members confront the same evidence, there will be a way for the group agent to determine the particular proposition on which a majority of members are failing to respond appropriately. In both cases, the group will not have evidence sufficient to conclude that it believes one of the three conflicting propositions without sufficient reason. And so the response supported by reasons-sensitivity is that of suspending judgment, and the problems for this approach for group agents in particular were noted in the previous section. These considerations suggest that those who adopt this escape route will have to abandon the idea that forming group agents is a feasible project, or at any rate that forming group agents that are evidentially reliable systems is a feasible project. Defenders of this radical proposal, like defenders of the moderate proposal—and like defenders of the idea that group agents should be practically restricted—must treat group agents as seriously deficient in comparison with the individual agents who make them up.

7 Escape Route 3: Group Agents Are a Fiction The first escape route suggests that group agents should not be allowed to act on the basis of votes that would give rise to a discursive dilemma, only on a more demanding basis, implying in effect that group agents should be practically restricted in a way that would make them infeasible. The second escape route, which comes in two versions, suggests a parallel result. Arguing that group agents should not be allowed to treat on–off votes properly as evidence (either as non-derivative evidence or as evidence at all), they imply that group agents should be epistemically restricted in a way that would also make them infeasible. Both of these escape routes suggest that group agents provide no real challenge for the view that the requirement of coherence is merely a byproduct of the requirement of reasons-sensitivity. They suggest that the group agents that might be formed on the template described in Section 4 cannot be a source of challenge for that view. The third escape route we should mention is even more radical than the other two. It argues that the group agents that might be formed on the Section 4 model do not count as agents at all, or not at least as agents in more than a fictional sense. This escape route is potentially more interesting than the other two. Those escape routes start from a thought that is close to the view that reasons-sensitivity is the only appropriate requirement on agents. The thought behind these routes is that group agents should not be allowed to act on on–off votes, or to treat on–off votes as

REASONS AND RATIONALITY

225

evidence, or to think that responding to on–off votes articulates the demands of reasons. And the motivation for that thought is presumably that if such votes are taken as articulating those demands, then that will mean that tracking reasons may fail to ensure coherence. Thus those escape routes may seem to presuppose the very thesis they are designed to preserve. No such difficulty would attend the third, for it argues on quite independent lines against the very reality of group agents. The argument against the reality of such agents may be that groups cannot be agents at all or that they cannot be agents in a real as distinct from a fictional sense. The argument that they cannot be agents at all, which abounds in some economics textbooks, and in some legal discussions, routinely cites the observation that group agents come into existence in virtue of contractual relations among their members, as if that in itself meant that they could not also count as agents. This approach views commercial companies—and by extension other forms of group agent—as sites of contracts akin to the market, which are distinguished only by the fact that the nexus of contracts between parties is much more pervasive than in markets more generally. One commentator formulates the core idea as follows: “the Nexus of Contracts Theory . . . treats the company as little more than a collective noun for the web of contracts that link the various participants, which include shareholders, management, employees and creditors. The function of company law is thus conceived of as the facilitation of the parties’ bargains” (Grantham 1998: 579). The assumption in this approach seems to be that since group agents are made out of the same material as markets—individuals in contractual relations with one another—they have no more claim to be agents than markets themselves have. But while markets are certainly not agents—while it is mere metaphor to speak of what the markets think or expect—corporate bodies certainly are (List and Pettit 2011). Unlike markets, they fit the functional characterization of agency given earlier. They have discernible purposes and representations such that in general they reliably pursue those purposes in accordance with reliably formed representations. They may occasionally fail, as we all fail, but when they do so they are generally ready to recognize the failure and even, if appropriate, to make amends. To say that they are not agents on the grounds that they are composed out of individuals in contractual relationships with one another would be as unconvincing as saying that you and I are not agents because we are composed out of cells in biological relationships with one another. What makes a system an agent is not the stuff out of which it is composed but the purposive-representational function that it is capable of discharging. Group agents can function in a distinctively agential way, embracing goals, forming views about how best to realize those goals, and pursuing the goals on the basis of those views. And moreover they can make commitments to other agents, individuals or corporate, by speaking in a self-committal way about what they hold or what they will do. It might be said that there are some clearly non-agential systems, such as the heating system in a building, that nonetheless fit the template for agency: they act

226

LARA BUCHAK AND PHILIP PETTIT

reliably for a goal on the basis of a reliable representation, as the heating system maintains the temperature of a building above a certain level on the basis of registering any fall below the threshold. Perhaps group agents are like that, it may be objected; perhaps they fit the template but still fail in an intuitive manner to count as agents. The suggestion is absurd, for two reasons. One, the group agent can commit itself in words, which the heating system cannot (Pettit 2014b). And two, there are indefinitely many ways in which a given goal or representation may be realized in a group, whereas the system realizes its goal and representation on the sole basis of a simple mechanical construction; thus, the language of goal and representation is extravagant in the latter case, but not in the former. Turning now to a second form of this third escape route, group agents may be said to be agents but not agents proper: not agents in anything more than a fictional sense. The best interpretation of this line on group agents, which goes back to Thomas Hobbes (1994 [1651]: ch. 16), holds that a group agent comes into existence by fiction insofar as a pre-existing agent or agential system is recruited to an extra role over and beyond the role of enacting and speaking for their or its own attitudes (see Pettit 2008: ch. 5, 2014a; Skinner 2009). In Hobbes’s picture, a group agent may come into existence in either of two ways. It may exist in virtue of a particular individual being authorized or accepted by the members of the group as speaking for them: this authorization will involve a readiness on the part of members to act as required in order for the commitments made by that individual in the name of the group to be fulfilled. Alternatively, a group agent may exist in virtue of the fact that an independent, more or less mechanical agential system like a majoritarian committee is authorized or accepted in the same way by the members of the group. The idea is that in neither case is the group agent anything other than a fiction. It makes commitments, by courtesy of its spokesperson, and it generally lives up to those commitments, holding by the beliefs avowed and keeping the promises made. But it does so in a way that is wholly parasitic on an independent, pre-existing agent or agency: hence that agent or agency can count as the real agent, with the group counting only as an agent by fiction. The main examples of group agents that operate on the model of Section 4 are not agents by fiction in this sense. They rarely have a single, dictatorial spokesperson who is entitled to speak for all members, and that case may be regarded as irrelevant. And they cannot operate on the basis of recruiting an independently agential committee— a committee that mechanically generates statements of purpose and representation fit to be authorized by a group agent—for reasons with which we are now familiar. A mechanical procedure like majority voting is not a reliable source of statements that the members of a group agent might endorse, being ready to act as the commitments embodied in those statements require. Any sort of committee, majoritarian or otherwise, that uses a bottom-up procedure to generate candidate commitments is liable, as Hobbes failed to see, to produce statements that are inconsistent with one another.

REASONS AND RATIONALITY

227

This means that in order for a group agent to come into existence, or at least a group agent that does not operate via a dictator, its members have to construct that agent from scratch, using procedures with a top-down as well as a bottom-up component in order to produce statements of purpose or representation that can mediate the commitments of an agent. This means that the agent constructed will be a distinct agent from its members—there will be no pre-existing agent with which it might be identified—even though it is composed entirely out of those individuals and their relationships. Once established as a group agent, the body in question may serve in a Hobbesian fashion as a spokesperson that the members of a wider group can authorize, thereby constituting themselves as an agent by fiction. But in order for such parasitic forms of group agency to materialize, there have to be group agents that exist and function in a non-parasitic way.

Conclusion We do not think that the considerations mustered over the last three sections establish definitively that the case of group agents undermines the case for treating the requirements of reasons-sensitivity as the only real requirements on agents. Nor do we believe that the escape routes described necessarily are the only avenues that those attached to that view might explore. We hold only that the case of group agents provides new data in light of which to consider the argument for and against an error theory about the requirements for coherence. But suppose we decide that the escape routes for the error theory really are closed, maintaining that the group agents cannot be dismissed as relevant and that they do operate in a way that gives coherence requirements a real, guiding role. What in that event should we conclude about the error theory? One potentially attractive line is to hold that the need for group agents to rely on the coherence requirements, using their non-satisfaction to guide its attitudes, teaches us that a corresponding lesson holds with individual agents too and that the error theory is not quite right, after all (at least in the case of conjunction).10 When I as an individual look to the evidence for and against a particular proposition, I have to look to what my eyes and ears tell me, to what I learn by testimony, and to what accumulated experience and theory support. But these channels of evidence may speak to me with many voices, as the votes of members speak with many voices to a group. Just as group agents have to listen to the voices of members in constructing the view they form on any proposition, so individual agents often have to listen to

10 It is worth noting that Kolodny does not claim that the error theory will explain all purported coherence norms (Kolodny 2007a: 253–7), and conjunction may be among those to which it does not apply. Those who take the view that the individual error theory does not apply to conjunction may consider the case of group judgment to be support for their view.

228

LARA BUCHAK AND PHILIP PETTIT

various voices—the different sensory, testimonial, and memory-based channels of evidence—in determining the view that they ought to take on a proposition. And the suggestion would be that in forming the required view, both group agents and individual agents have to rely on coherence in a way that does not merely amount to treating incoherence as additional evidence about what they have reason to believe. They have to silence the evidence available at any moment in support of a proposition when they check and find that it would give them an incoherent profile of views. And they then have to consider how to resolve the conflicting bodies of evidence that occasion the incoherent views. This suggestion, which turns on an epistemic analogy between the position of individual and group agents, may be buttressed by consideration of a further analogy on the practical side. This is that just as group agents are going to fail the rationale for establishing them unless they are prepared to form beliefs in the absence of conclusive evidence, so something similar holds of individuals. Those who defend an error theory about the requirement of coherence suggest that in the absence of conclusive evidence for believing p, or for believing not-p, we individual agents ought to suspend judgment. But this may well look like an unworkable constraint. If as individuals we were to form beliefs only in the presence of conclusive evidence—only when the facts spoke to us in an unambiguous way—then we would have to suspend belief on a great range of matters. As individuals we would be in much the same position as group agents that were required to form beliefs only when a unanimity constraint is satisfied. Like such group agents, we would be unimpeachable in epistemic terms. But, like such groups, we would also be rendered more or less ineffective as centers of agency. We would be incapable of making up our minds on so many matters that we would be deprived of the beliefs essential for resolving most of the choices that arise in ordinary life: most of the choices that the pursuit of our goals requires us to make. Nor is that all. We not only need to have a rich set of beliefs in order to be able to pursue the purposes that are important for us as individual and group agents. We also need to be able to speak for such a set of beliefs in establishing relationships with one another and in presenting ourselves to others as agents with whom they can do business: agents they can expect to be willing to stake out positions on various issues and to prove generally reliable in acting in the manner associated with those positions. Suppose we were to shrink from forming opinions in the absence of more or less conclusive evidence. Or suppose we were to form such beliefs without practical guidance from the requirement of coherence. In any such event we would prove to be less than fully conversable, as we might put it; less than fully capable of relating to one another reliably in avowing beliefs or intentions or in promising to perform one or another action. These two analogies are less than perfect, of course, and even if we reject the escape routes described in the last three sections, they do not make an irresistible case for applying the lesson of group agents to individuals. When we try as individuals to

REASONS AND RATIONALITY

229

make sense of various evidential inputs in order to determine our view on a proposition, we do not employ a rote procedure of tallying the “votes” of our individual sources of evidence. Rather, we consider how reliable each source is apt to be in the given case, and how the deliverances of one source bear on the deliverances of the other. And, unlike the group case, in which degrees of belief would need to be explicitly expressed in order to be used by the group to determine the group’s opinion on the proposition, we can take degrees of support offered by various evidential sources into account in a largely automatic, subpersonal manner. Furthermore, the second-order evidence that we are mistaken in some judgment is easier to employ, since as individuals we often have a clear view about where mistakes are more apt to occur. By contrast, the group agent has no privileged vantage point from which to adjudicate the views of the various members, or to judge the relative reliability of members on a particular question. As we do not say that the escape routes available to those who hold an error theory about the requirement of group coherence are definitively closed, then, so we do not say that even if they are closed, that means that the error theory should be given up on all sides: not just for groups but for individuals too. All we claim to have established is that there is an interesting tension between the error theory of coherence, on the one side, and the theory of group agency on the other. The two theories are associated with different literatures and different traditions and what we hope to have shown is that it may well be profitable to bring them more closely into contact.

References Booth, A. R. (2012) Epistemic ought is a commensurable ought. European Journal of Philosophy 21(2). Broome, J. (1999) Normative requirements. Ratio 12: 398–419. Reprinted in J. Dancy (ed.) Normativity. Oxford; Malden, MA: Blackwell, 2000, 78–99. Broome, J. (2001a) Are intentions reasons? And how should we cope with incommensurable values? In C. Morris and A. Ripstein (eds.) Practical Rationality and Preference: Essays for David Gauthier. Cambridge: Cambridge University Press, 98–120. Broome, J. (2001b) Normative practical reasoning. Proceedings of the Aristotelian Society 75(suppl.): 175–93. Broome, J. (2004) Reasons. In R. J. Wallace, M. Smith, S. Scheffler, and P. Pettit (eds.) Reason and Value: Themes from the Moral Philosophy of Joseph Raz. Oxford: Clarendon Press, 28–55. Broome, J. (2005) Does rationality give us reasons? Philosophical Issues 15(1): 321–37. Broome, J. (2007) Wide or narrow scope? Mind 116(462): 359–70. Broome, J. (2013) Rationality through Reasoning. Chichester, UK; Malden, MA: Wiley Blackwell, Oxford. French, P. A. (1984) Collective and Corporate Responsibility. New York: Columbia University Press.

230

LARA BUCHAK AND PHILIP PETTIT

Genest, C., and Zidek, J. (1986) Combining probability distributions: A critique and an annotated bibliography. Statistical Science 1(1): 114–35. Grantham, R. (1998) The doctrinal basis of the rights of company shareholders. Cambridge Law Journal 57: 554–88. Harman, G. (1986) Change in View. Cambridge, MA: MIT Press. Hobbes, T. (1994) Leviathan, E. Curley (ed.). Indianapolis, IN: Hacket. Original work published 1651. Kavka, G. (1983) The toxin puzzle. Analysis 43(1): 33–6. Kolodny, N. (2005) Why be rational? Mind 114(455): 509–63. Kolodny, N. (2007a) How does coherence matter? ProcMieedings of the Aristotelian Society 107 (1): 229–63. Kolodny, N. (2007b) State or process requirements? Mind 116(462): 371–84. Kolodny, N. (2008a) The myth of practical consistency. European Journal of Philosophy 16(3): 366–402. Kolodny, N. (2008b) Why be disposed to be coherent? Ethics 118(3): 437–63. Kornhauser, L. A., and Sager, L. G. (1993) The one and the many: Adjudication in collegial courts. California Law Review 81: 1–59. List, C. (2001) Mission Impossible? The Problem of Democratic Aggregation in the Face of Arrow’s Theorem. D.Phil. Thesis in Politics, Oxford University. List, C., and Pettit, P. (2002) Aggregating sets of judgments: An impossibility result. Economics and Philosophy 18: 89–110. List, C., and Pettit, P. (2011) Group Agency: The Possibility, Design and Status of Corporate Agents. Oxford; New York: Oxford University Press. List, C., and Polak, B. (2010) Symposium on judgment aggregation. Journal of Economic Theory 145(2): 441–66. Marusˇic´, B. (2012) Belief and difficult action. Philosophers’ Imprint 12(18). Morauta, J. (2010) In defence of state-based reasons to intend. Pacific Philosophical Quarterly 91(2): 208–28. Parfit, D. (2001) Rationality and reasons. D. Egonsson, J. Josefsson, B. Petersson, and T. Ronnow-Rasmussen (eds.) Exploring Practical Philosophy: From Action to Value. Burlington, VT: Ashgate. Pettit, P. (2001a) Deliberative democracy and the discursive dilemma. Noûs 11(suppl.): 268–99. Pettit, P. (2001b) A Theory of Freedom: From the Psychology to the Politics of Agency. Cambridge: Polity; Oxford: Blackwell. Pettit, P. (2008) Made with Words: Hobbes on Language, Mind and Politics. Princeton, NJ: Princeton University Press. Pettit, P. (2014a) Group agents are not expressive, pragmatic or theoretical fictions. Erkenntnis 79. Pettit, P. (2014b) How to tell if a group is an agent. In J. Lackey (ed.) Essays in Collective Epistemology. Oxford: Oxford University Press. Reisner, A. (2008) Weighing pragmatic and evidential reasons for belief. Philosophical Studies 138: 17–27. Reisner, A. (2009) The possibility of pragmatic reasons for belief and the wrong kind of reasons problem. Philosophical Studies 145: 257–72.

REASONS AND RATIONALITY

231

Reisner, A. (2011) Is there reason to be theoretically rational? In A. Reisner and A. SteglichPetersen (eds.) Reasons for Belief. Cambridge; New York: Cambridge University Press, 34–53. Russell, J., Hawthorne, J., and Buchak, L. (forthcoming) Groupthink. Philosophical Studies. Skinner, Q. (2009) A genealogy of the modern state. Proceedings of the British Academy 162: 325–70.

15 Weighing Explanations Stephen Kearns and Daniel Star

The primary goal of John Broome’s new book, Rationality Through Reasoning (2013), is to outline and defend an account of reasoning that makes it clear how it is possible to actively become more rational by means of engaging in reasoning.1,2 In the process Broome finds it necessary to also provide his own accounts of ought, reasons, and requirements. We will focus here on the account of reasons. This is not the first time we have done so. In an earlier paper (Kearns and Star 2008),3 we contrasted Broome’s account with our own favored account of reasons (reasons as evidence).4 Although there are some differences between the views defended in the relevant chapters of Broome’s book (chs. 3 and 4) and the draft manuscript and earlier papers that we used as the basis of our discussion in our earlier paper, these do not, for the most part, substantially affect our earlier arguments. It is clear that in articulating an alternative account of reasons we were also heavily influenced by Broome, so we are particularly grateful to have this opportunity to contribute a piece to a Festschrift for him.5 In a response to us and some other critics, Broome (2008) presented some challenges for our account of reasons, but did not address our criticisms of his own account (and we responded, in turn, to Broome’s challenges in Kearns and Star 2013). Here we will first provide updated versions of our earlier concerns, since they mostly still seem pertinent. We will then turn to provide a fresh response to his account of reasons that focuses on the notion of a weighing explanation. On Broome’s account, pro tanto reasons are facts cited in weighing explanations of what one ought to do; facts that have weights. It is not clear 1

We would like to thank the editors of this volume for providing helpful feedback. All numbers in parentheses that are not clearly references to other works should be taken to be references to this book. 3 From now on, we refer to this paper without mentioning authorship (in other words, any references to 2008 where an author is not mentioned are to this paper). 4 The central claim of which is: Necessarily, a fact X is a reason for an agent N to F if and only if X is evidence that N ought to F. 5 Judith Jarvis Thomson (2008) also has an account of reasons that is similar to ours, and she does not appear to have been influenced by Broome. We did not know about Thomson’s views on this topic at the time of writing and publishing our early papers (2008, 2009). 2

WEIGHING EXPLANATIONS

233

to us what the idea that pro tanto reasons have weights really amounts to. While recognizing that a simple analogy with putative non-normative weighing explanations involving physical weights initially seems helpful, we argue that the notion of a weighing explanation, especially a normative weighing explanation, does not ultimately stand up to scrutiny.

1 Reasons as Normative Explanations, or Parts of Normative Weighing Explanations According to Broome there are at least two types of normative reasons: pro toto reasons and pro tanto reasons. The first of these is defined as follows: A pro toto reason for N to F is an explanation of why N ought to F. (50)

Broome immediately adds “a pro toto reason for you to F is something that makes it the case that you ought to F,” and he clarifies what he has in mind when referring to explanations by saying, “An explanation need not be full or complete, and what counts as an explanation may depend on the context” (50). He provides the following examples: Suppose you ought to visit Mr Reed. The explanation might be that he is the best dentist around. If that is indeed the explanation, it is a pro toto reason to visit him. In a different context, the explanation might be that you ought to visit the best dentist around. This would be a pro toto reason for you to visit Mr Reed, in that context. We now have two alternative pro toto reasons, but they do not compete with each other. (50)6

Like all of us working on reasons, Broome is well aware that there are clearly nonnormative uses of the word “reason”; in particular there are general “explanatory reasons”: “A ‘reason’ in this sense is more or less synonymous with ‘an explanation of ’” (46). Given that this is a clear sense of “reason,” should we not conclude that pro toto reasons are simply non-normative explanatory reasons (of normative facts), since their structure would seem to require this categorization? Broome responds to this question by saying “The ‘reason why’ (meaning explanation) bumps into the normative ‘ought,’ yielding a normative sense of ‘a reason’ that combines the meaning of both” (50). He immediately admits this is a somewhat “picturesque etymology,” but we do not object to this element of his view. On reasons as evidence, it might also be objected that the reasons we take to be normative are not really normative (as distinct from oughts), but we claim that they “inherit” their normativity from oughts. This is less picturesque, but probably not much more illuminative, for all that. It is a little unclear whether “context” here refers to decision situations themselves, or contexts in which decision situations are described, or either. We take it he means either. 6

234

STEPHEN KEARNS AND DANIEL STAR

The second class of normative reasons that Broome is interested in analyzing are pro tanto reasons. These are reasons that appear to be more or less weighty, and might favor an action that we actually ought not do (pro toto reasons never do this). Broome offers the following “functional definition”: A pro tanto reason for N to F is something that plays the for-F role in a weighing explanation of why N ought to F, or in a weighing explanation of why N ought not to F, or in a weighing explanation of why it is not the case that N ought to F and not the case that N ought not to F. (53)

In order to understand this definition, we need to understand what a weighing explanation is, and what a for-F role is. Broome approaches normative weighing explanations by first providing an example of a non-normative weighing explanation that involves a mechanical scale with items of different weights placed on each side of it. When considering such an example, we can explain why the balance tips left (say) by referring to the combined weights of the object on the left and the combined weights of the object on the right and by noting that the combined weight on the left exceeds that on the right (52). In the case of normative weighing explanations, it is the weight of reasons we are considering, rather than the weight of physical objects, and we are looking for an explanation of what an agent ought to do. As for the “for-F role,” its presence in Broome’s definition is what makes the definition a functional one. This means that, in effect, “a pro tanto reason can be defined without any reference to the for-F role . . . it is something that has a weight, where the weights of reasons combine in some way to determine whether or not you ought to F” (54). Weights, then, are more fundamental than for-F roles. In passing, we have three questions about Broome’s account of reasons. These are not really objections, so much as things we would like to see clarified. The first question we have is whether every weighing explanation that you ought to F (used to locate pro tanto reasons) is also specifying a pro toto reason? Broome says, “The combined weights of the reasons for you to F exceed the combined weights of the reasons for you not to F. That is why you ought to F” (52).7 The “that” here seems to refer back to the whole of the preceding sentence (or, more carefully, the fact that would be picked out in an accurate filling out of the schema provided in the preceding sentence), and we have been told that every X in a true statement of the form “X is why you ought to F” is a pro toto reason. Second, it seems to us that, on Broome’s analyses, every pro tanto reason that aligns with an ought (i.e. in every case where there is pro tanto reason to F and where the relevant agent ought to F) is also ipso facto a pro toto reason, and that this is true even when the pro tanto reason in question is quite weak and would not be enough to

7 On the same page Broome also says, “We often say you ought to do something because the balance of reasons is in favour of your doing it. When an explanation takes this form, I shall call it . . . a ‘normative weighing explanation’.”

WEIGHING EXPLANATIONS

235

determine that one ought to F in the absence of other reasons that align with the ought in question. Is this right? This entailment would certainly not be the case if the notion of a pro toto reason were identical to the notion of a complete reason, but, as just noted, these notions are quite separate. Why do we think Broome might be committed to thinking that even many very weak pro tanto reasons that happen to align with oughts must be pro toto reasons? This seems to follow from something he says when discussing facts about explanations that he takes to apply to his own account of pro toto reasons: Although I shall not try to describe the nature of the explaining relation, I do need to say something about the individuation of explanations. Suppose Joanne broke a slate a while ago, and as a result the roof leaks. It rained last night, and today the carpet is wet. When we enquire why the carpet is wet, you might say the explanation is that it rained last night. I might say that it is that Joanne broke a slate. Someone else might say that it is that the roof leaks. We respectively make these statements: “The explanation of why the carpet is wet is that it rained last night”, “The explanation of why the carpet is wet is that Joanne broke a slate”, and “The explanation of why the carpet is wet is that the roof leaks”. Read literally, no two of our statements can be true together. Still, our explanations are not rivals . . . there is really one big explanation of why the carpet is wet. It is a complex fact that includes as parts all the separate facts the three of us described. Each of us is picking out a part to stand in for the whole. We call it the explanation because it is standing in for the one big explanation. . . . Which part we pick out will depend on our context: our background knowledge, our interests in the matter and so on. (48–9)

Notice that it is a feature of the example Broome uses that certain facts that count as explanations fail to be sufficient conditions for the obtaining of the fact that they explain (the missing slate will not make the carpet wet in the absence of bad weather). Consider the following example. Suppose Jake has promised to meet a friend in ten minutes, and is walking to meet him (we include this fact merely in order to ensure we are talking about a case involving pro tanto reasons). Suppose Jake comes across a person needing urgent medical assistance, which will require a trip to an emergency ward. Fortunately she is lying next to a Mercedes that has been left unlocked with keys still in the ignition. Jake judges that, despite the promise to his friend which he will not be able to fulfill if he drives to the hospital, he ought to drive her to the hospital. A bystander might correctly say, The explanation of why he ought to drive her to the hospital is that she needs urgent medical assistance. A different bystander might correctly say, The explanation of why he ought to drive her to the hospital is that there is an unlocked Mercedes parked nearby. If these claims are both appropriate (and we think they are, since, as Broome indicates, what is left out can depend on background knowledge),8 then there seem to be two possibly problematic consequences: (1) the fact 8 It might be thought that, “The explanation of why he ought to drive her to the hospital is that there is an unlocked Mercedes parked nearby,” would not be a true sentence expressed in, or about, the imagined context. It is important to bear in mind Broome’s point about background knowledge. Imagine the

236

STEPHEN KEARNS AND DANIEL STAR

that there is an unlocked Mercedes parked nearby gets to count as a (pro toto) reason, even though many would class it as a mere enabler;9 and (2) the fact that there is an unlocked Mercedes parked nearby gets to count as a pro toto reason, even though it clearly is not a sufficient condition in this context for it being the case that one ought to drive to the hospital. Now let us alter the example a little to make a third point. Suppose the person encountered is clearly merely somewhat unwell and probably should see a doctor soon, rather than in obvious need of urgent medical care, and let us suppose also that Jake has always wanted to drive a Mercedes, and let us also suppose that the law of the land happens to say it is legal to borrow unlocked cars in situations such as the one he is in, and that he will impress some friends standing nearby, and so on. This is intended to be a sketch of a case where none of the reasons that favor driving the car to the hospital are individually strong enough to justify doing so, but together do justify doing so (there presumably will be such cases, so we need not worry that our example has become a little outlandish). Then it will turn out that (3) none of these pro toto reasons is really sufficient to make it the case that Jake ought to drive the car. One response Broome might have to these concerns is to say that we have misunderstood his proposal: the reason “The explanation why . . .” statements can be piled on top of each other without fear of contradiction is because each such true proposition expressed is really extensionally equivalent (for they purposely refer to the complete explanation, merely by explicitly highlighting one part of the explanation). The worries just expressed do not seem to get off the ground if this is true, but it means that for every ought that can be explained using a pro toto reason, there is, strictly speaking, only one pro toto reason, and that is the complete explanation. Admittedly, this is a reasonable way of reading the quotation we provided immediately above, but we are not convinced that this is the best way of explaining why explanatory statements of one and the same fact can often be piled on top of each other. More crucially, it does not fit well with the definition of a pro toto reason as “an explanation of why N ought to F” (50, our emphasis), and with the quotations we provided at the beginning of this section—they explicitly state that there is often more than one pro toto reason in play in relation to any one ought fact, and that pro toto reasons need not be complete explanations. Our third question is related to our second. Broome claims that “the relation of explaining is the relation of making so” (48), but it seems to us that there is

following dialogue: He ought to drive her to the hospital.—How could that be the case? He left his car at home.—Oh, but the fact that there is an unlocked Mercedes nearby is what explains why he ought to drive her to the hospital. Of course, a complete explanation would include something about the fact that driving to hospital is the best means available, but that does not render the last sentence in the conversation false. In any case, we could use a different example to make our points here. 9 Full disclosure: it seems we defenders of reasons as evidence also have to accept that enablers are reasons, since they are evidence concerning what one ought to do (see Fletcher 2013 and Setiya 2014). We do not think this is too bad, actually, but many would disagree with us about this.

WEIGHING EXPLANATIONS

237

something to be said for the view that none of the weak reasons in favor of going to the hospital in the second, altered case above individually make it the case that one ought to drive to the hospital, even though together they do (and even though they might be cited in true and informative “The explanation why he ought . . .” statements). It seems plausible to us that the relation of “making it the case that” should be reserved for a stronger metaphysical relation (perhaps a grounding relation) than the mere explanatory relation, which does seem somewhat more interest-relative, hence somewhat less metaphysically robust. The question is: how metaphysically robust are explanations meant to be on Broome’s account? This is an important question, because, given Broome’s account of reasons, it directly bears on the question of how metaphysically robust reasons are meant to be.

2 The Objections of “Reasons: Explanations or Evidence?” Our earlier paper provided seven key objections to Broome’s account of reasons. We take some of these to be stronger than others. Here is a brief restatement of our key objections. (In most cases, we have had to omit aspects of, or variations on, these objections; aspects and variations that tend to strengthen them.) i. The deliberative role of reasons. We take it to be a truism about reasons that they play an essential role in actual deliberation with respect to answering the question, What ought I do? (2008: 40–2). Typically, one works out what one ought to do in the situation one finds oneself in by considering the reasons that apply to one in this situation; one generally does not first know what one ought to do, and only then ask oneself what reasons are applicable. On the other hand, explanations, or parts of explanations, of any facts are things we reason our way towards only after we have an explanandum in sight. The view that reasons are parts of explanations of ought facts, whilst not strictly inconsistent with the view that practical reasoning typically starts with a grasp of specific reasons in combination with ignorance about specific ought facts (explanans on Broome’s account) and works towards conclusions about what specifically ought to be done (explanandum on Broome’s account), certainly prima facie conflicts with this truism. At the very least, we are owed an account of what explains the truism (or, alternatively, convincing grounds for rejecting it). We will see this objection is also connected to another objection, (v) below. ii. The “for-F role” in normative weighing explanations. It might be thought that Broome fails to provide a non-circular analysis of pro tanto reasons, since one might suspect it is impossible to understand what a “for-F role” is independently of an understanding of what it is for facts to count in favor of acts (and beliefs, etc.). This is one place where what Broome says in his book is more straightforward and

238

STEPHEN KEARNS AND DANIEL STAR

satisfactory than the manuscript we were responding to in our earlier paper (2008: 42–4). As noted above, he now takes the “for-F role” to simply stand in for the weight of the relevant reason involved in any case. One might still ask: is it possible to understand the idea that normative reasons have weights, prior to an understanding of counting in favor? Broome clearly thinks it is (54). We are not so sure about this. Of course, in one way we can obviously understand “weight” independently of “counting in favor,” and that is in the common, non-normative use of “weight,” as when we consider physical scales. But the physical examples are meant to be metaphorical. Might we come to understand normative weight via first understanding physical weights and then making use of a metaphor? Perhaps this might work when we are considering the adoption of a theory of reasons, but this seems like a highly questionable hypothesis when considering how actual ordinary agents come to an appreciation of the weight of reasons early in life (say). Broome might object that it does not matter for his purposes whether agents actually come to understand that reasons have weights prior to understanding that they count in favor; what is crucial is that agents could do so. Perhaps they could; we are just not so sure of this. iii. The mystery of the weight of reasons. In any case, we now think Broome’s way of explaining what a “for-F role” is leads to a more important objection, and that is that Broome leaves it utterly mysterious what it is for a reason to have a weight (2008: 44–5). One way of putting this objection is to consider his account of reasons in relation to both the primitivist account endorsed by T. M. Scanlon and Derek Parfit (according to which no informative analysis of “reason” is possible) and our own analysis of reasons. At first blush, it may seem that Broome’s account of reasons improves on the primitivist account: after all, he seems to offer an analysis where the primitivist account embraces a pessimism about the possibility of being able to do so. But if we are now to take the concept of normative weight as basic, it is not clear that Broome’s analysis is really more satisfactory than the primitivist analysis.10 Defenders of reasons as evidence, on the other hand, may be able to explain weights in other terms. We will say more about weights in the last section of the chapter.

10 It might be objected that reasons as evidence does not really fare particularly well on this front, since it leaves ought unanalyzed (we have suggested that we can analyze evidence in terms of epistemic probability). Our objection is that Broome leaves both ought and weights unanalyzed. Defenders of reasons primitivism sometimes think they can at least explain oughts in terms of reasons. Regarding weights, Scanlon (2014: 111) rejects the view that weights are real, independent properties of reasons: “I conclude that there is no non-normative coin or normative coin, in terms of which the strength of reasons, considered on their own . . . can be expressed. . . . The strength of a reason is an essentially comparative notion, understood only in relation to other particular reasons” (he goes on to add that his approach to understanding the strength of reasons is “top-down,” working from general principles that express normative relations concerning sufficient and conclusive reasons). Reasons primitivists (and others) often use the term “strength” rather than “weight,” but we take these terms to be roughly equivalent when attributed to reasons; at least, we see no particular lessening of the element of mystery when one word is used rather than the other (and, absent the adoption of something like Scanlon’s “top-down” approach, reasons primitivists are not in a better position than Broome, in this respect).

WEIGHING EXPLANATIONS

239

iv. Ambiguity concerning “reason,” or defining “reason” disjunctively. This objection can be simply stated (2008: 45–6). According to Broome, there are two types of normative reasons: pro toto reasons, which are explanations of ought facts, and pro tanto reasons, which corresponds to parts of normative weighing explanations. Broome may be happy with there being two senses of (normative) “reason,” but there do not seem to be any independent grounds for thinking that (normative) “reason” is ambiguous in this way. An alternative to postulating an ambiguity here is to combine both definitions of “reason” that Broome provides in order to form a disjunctive definition (2008: 35); the problem then might be said to be that such a disjunctive definition, combining reference to both weighing and non-weighing explanations, is not sufficiently unified. It certainly does not appear to cut nature at its joints (something one might hope for with definitions). v. Too few reasons. Broome explicitly rejects the idea that facts that are (merely) evidence that you ought to F are, or provide, reasons to F, even when they are conclusive evidence that one ought to F (51). Here, it is fair to say, Broome is in agreement with many philosophers working on reasons. And when asked to assess, for instance, the claim that the fact that a minister accepted a bribe is a reason for him to resign alongside the claim that the fact that a reliable newspaper has reported that a minister ought to resign is a reason for him to resign (or similar pairs of claims), we have found audiences often intuitively judge that the first claim is true and the second false (in the relevant circumstances). Broome clearly shares this intuition, but he does not present much of an argument for it being correct (one may take his claim that evidence that one ought to F does not explain why one ought to F as a crucial step in an argument here, but this argument would rely on his account of reasons being correct, and, in any case, as we point out in (vi), he seems to say elsewhere that evidence in general does very often explain why one ought to F, so it is not clear why evidence that one ought is special in this respect). We agree with Mark Schroeder (2007: 92–7) that negative existential intuitions about reasons are often misleading, and that much can be done to dispel such intuitions by focusing on conversational pragmatics.11 We also think that there is a serious danger that the denial that evidence that one ought can provide reasons for action will lead to a too few reasons problem (2008: 11 On Schroeder’s Humean account of reasons, there is a reason for you to eat your car (it contains iron). The argument that Schroeder provides appeals to Gricean maxims; the defender of reasons as evidence might hope to also do this, but the precise form of the argument will need to take a different form (Schroeder relies on the reasons in question being very weak, but we cannot rely on that assumption, for a start). Even so, assuming Schroeder’s argument is successful, it is very odd to think that the fact that your car contains iron is a reason to eat your car while the fact that a reliable book says you ought to eat spinach (one might imagine the author knows but does not say that spinach contains iron) is not a reason to eat spinach. Some philosophers worry that allowing for as many reasons as we do creates a counting problem (as it would unjustifiably license the continual adding together of the weights of reasons that are not independent of each other); we deny that there really is a problem of this kind (see Star 2011: 93–4, forthcoming).

240

STEPHEN KEARNS AND DANIEL STAR

48–9; 2009: 233–4). From the perspective of fundamental normative ethics, conceived of as a project that attempts to articulate and defend fundamental ethical principles, many of the mundane facts that we commonly take to be reasons are not really fundamental explanations of what we ought to do. If hedonistic utilitarianism were true (for instance), then the fact that an act would produce pleasure would be fit to be part of a fundamental explanation of why one ought to do that act, but the fact that an act would be the fulfilling of a promise would not be part of such an explanation. Assuming one thinks there is a point to the project of normative ethics (as we do), it is not at all attractive to suppose that it is only the features mentioned in fundamental explanations that get to count as reasons, since this would make too many of our ordinary reason claims false (which, as well as being very counterintuitive, would not fit well with the deliberative role of reasons discussed above).12 Of course, Broome need not think that it is only facts cited in fundamental explanations of ought claims that get to count as reasons, but this is precisely why we wonder why he is so sure that evidence that one ought to F cannot nonfundamentally explain why one ought to F. If a reliable book says one ought to eat cabbage, why can that not non-fundamentally explain why one ought to eat cabbage? After all, there does seem to be something remiss about not following the advice, even if it turns out a reliable book (or a reliable doctor, say) is wrong on this occasion (and there are cases where not following testimony that provides evidence concerning what one ought to do seems morally wrong, e.g. if a reliable doctor claims a drug will save a child’s life and requires the consent of a parent). Furthermore, we take it that testimony can provide reasons for action in virtue of the fact that it can reliably indicate what one ought to do. Assuming Broome does not wish to claim that only (parts of) fundamental normative explanations provide reasons, general non-fundamental facts will not invariably coincide with fundamental reason facts, but will instead merely reliably indicate the presence of such facts. From this perspective, facts like the fact that a newspaper has reported a minister ought to resign and facts like the fact that a minister accepted a bribe seem to be on a par.13

12 To be clear: this is not meant to be an objection to utilitarianism. Rather, we are simply saying that the utilitarian can and should adopt an account of reasons that allows the fact that an act would be the fulfilling of a promise to count as a reason. Star (forthcoming) argues that reasons as evidence best elucidates the relation between fundamental and derivative reasons in a way that respects both ordinary virtue and normative ethical theory. 13 The difference in the extent to which each of these facts reliably indicates that he ought to resign is a difference in degree only. Depending on the details of one’s favorite normative ethical theory, one might need to describe a fairly outlandish case to have it come out that accepting a bribe on this occasion is not accompanied by a fundamental ethical reason, but it will be possible to do so, unless the ethical theory recognizes bribes as fundamental reasons, or something always entailed by bribes, e.g., perhaps, deception, but then we could just use some other example following the same general recipe.

WEIGHING EXPLANATIONS

241

vi. Ought and evidence. Broome’s denial that evidence provides reasons seems to be in tension with his account of ought. Here we must ignore many subtle features of Broome’s discussion of ought, and cut to the chase: he takes the central ought (the ought that the rational requirement of Enkrasia applies to) to be the prospective ought, and not the objective ought (40–4). Broome takes it that it is easiest to illustrate the difference between these oughts by focusing on situations where consequentialist moral claims at a very general level might apply; here the objective “ought” (Broome is not sure that there really is such an ought) will be determined by actual outcomes, independent of evidence, but the prospective ought will be determined by prospects, which are dependent on the probabilities of outcomes. Broome thinks that the school of thought that the probabilities in question are to be explained in terms of available evidence gets the right results much of the time, but, for reasons that are fairly complex, it ultimately needs to give way to the main alternative school of thought, namely that the probabilities in question are purely subjective (i.e. the probabilities that Bayesian subjectivists focus on). At no point is there a parallel to his blanket refusal to allow that evidence that one ought to F can be a reason. He concludes, “Prospective oughts . . . are subjective only within limits. Your subjective probabilities must be constrained by the evidence . . . but evidence does not always determine all of the probabilities. It may leave a gap, and if it does Bayesians fill the gap with subjective probabilities” (44, emphasis added). To simplify in a way that we hope Broome will not mind (he does, after all, seem very sympathetic to the evidencebased account of probability): to a considerable although not complete extent, evidence will determine what we ought to do. Now our criticism is fairly easy to state: given that Broome believes that, to a large extent anyway, evidence explains what one ought to do, it seems that his claim that evidence that one ought to F cannot be a reason is unwarranted, given that reasons are, on his own account, parts of explanations of why one ought to F. vii. Fundamental ought facts. The following seems like a very plausible claim: “if one ought to F, then there is a reason to F ”; otherwise, an agent might do something which they rightly take to be something they ought to do at the same time as thinking there is no reason at all to do it! It also seems reasonable to suppose that some ought claims or principles are fundamental, in the sense that the truth of these claims is not something that can be further explained. Whether the correct fundamental ethical theory is consequentialist or non-consequentialist, it seems likely that the principle(s) at the foundation of this theory may lack further explanation. If it is also the case that reasons are parts of explanations concerning what agents ought to do, then it will not be true of fundamental oughts that “if one ought to F, then there is a reason to F” (2008: 53). We now think that Broome can avoid this objection by utilizing the machinery of requirements (provided in ch. 7) to hold on to “if one ought to F, then there is a reason to F,” while holding that there are no unexplained oughts. On his view,

242

STEPHEN KEARNS AND DANIEL STAR

fundamental ethical principles are best cast in terms of normative requirement claims, rather than ought claims. For the sake of simplicity, assume moral requirements override other types of normative requirements (we assume this solution will also work with more complex theories of the way normative requirements of different types interact). Whenever one is morally required to F, this requirement— for example, to respect autonomy, or, say, to maximize total expected welfare—will be a pro toto reason to F; or, in other words, the requirement will explain why one ought to F.14

3 Weighing Explanations As we mentioned above, it is clear that Broome thinks that the important class of reasons generally referred to as pro tanto (i.e. all reasons that may be weighed) are to be analyzed by reference to weighing explanations. It is thus crucial that, as long as we think that there are indeed pro tanto reasons (an assumption that seems very safe to hold to), there actually must be some genuine weighing explanations. We will now present a new line of argument for the conclusion that there are, in fact, no genuine weighing explanations. This may appear to be a rather strong conclusion, so it is important that we proceed by first paying careful attention to Broome’s account of weighing explanations.

Mechanical weighing explanations Broome begins his discussion of weighing explanations in general by describing mechanical weighing explanations: How does a mechanical weighing explanation go? Suppose a balance tips to the left. A typical weighing explanation of why it does so will go like this. There is at least one object in the lefthand pan, and there may be one or more objects in the right-hand pan. Each object has a weight. The combined weights of the objects in the left-hand pan exceed the combined weights of the objects in the right-hand pan. That is why the balance tips left. (In the theory of mechanics, the word “weight” refers to a force, which is a vector. I am not using it that way, but colloquially, to refer to a scalar magnitude.) (52)

Broome then contends that this is analogous to weighing explanations of normative facts: Suppose you ought to F. If there is a weighing explanation of why, it takes an analogous form. There is at least one reason for you to F, and there may be one or more reasons for you not to F. Each reason has a weight. The combined weights of the reasons for you to F exceeds the combined weights of the reasons for you not to F. That is why you ought to F. In this analogous

14

It is no accident that Broome is careful to note in the earlier chapter on reasons that the facts that are pro toto reasons may themselves be normative facts (51).

WEIGHING EXPLANATIONS

243

explanation, the fact that you ought to F is the explanandum, and analogous to the fact that the balance tips left. A reason for you to F is analogous to an object in the left-hand pan, and a reason for you not to F is analogous to an object in the right-hand pan. (52)

Broome’s examples of normative weighing explanations Broome provides two main examples of supposed weighing explanations of normative facts. The first runs as follows: Suppose you ought to bring some wine to the party, and suppose the explanation of why is this. You promised to bring some wine, and you ought to keep your promises unless significant harm will result from doing so. In this case, suppose the only harm that will result is the cost to you in money and inconvenience of buying the wine, and this is not significant. (55)

Broome provides the following gloss on why this is a good candidate for being a weighing explanation: In favour of your bringing some wine is your promise. Against is the cost. Your promise plays the for-bringing-wine role in the explanation. It is therefore a pro tanto reason to bring wine. The cost plays the against-bringing-wine role. It is therefore a pro tanto reason not to bring wine. The promise is a reason of greater weight than the cost. (56)

Here is Broome’s second example, and his reasons for considering it a weighing explanation: Suppose you are choosing between Montreux and Marrakesh as places to visit. Suppose that you ought to choose Montreux, on the grounds that it is a pleasant resort and not so far away as Marrakesh, though less exotic. The explanation of why you ought to visit Montreux is a weighing one. In favour of visiting Montreux are its pleasantness and proximity. Each plays the for-Montreux role in the explanation. They are pro tanto reasons for you to visit Montreux. Against visiting Montreux is the exoticness of the alternative, Marrakesh. That plays the against-Montreux role. It is a pro tanto reason for you not to visit Montreux. The reasons for visiting Montreux outweigh the one for not doing so. (56)

Broome uses this example to illustrate his distinction between particular reasons and general reasons: The explanation could be filled out in more detail. Montreux's pleasantness consists in its beautiful views, cosy restaurants, opportunities for lake cruises and other pleasant features. Its nearness means your journey will add less carbon dioxide to the atmosphere. On the other hand, the exoticness of Marrakesh consists in teeming bazaars, belly-dancing and opportunities for camel rides. One reason for visiting Montreux is the beautiful views. Another is the cosiness of the restaurants. A reason for not visiting Montreux is that going instead to Marrakesh will allow you to take a camel-ride. And so on.

244

STEPHEN KEARNS AND DANIEL STAR

(56)

Particular reasons figure in weighing explanations that are simply more detailed spellings-out of the weighing explanations which feature general reasons. The former explanations are not rivals of the latter, and the particular reasons to F do not combine weights with the general reasons to F to produce a stronger case for F-ing.

Some other explanations Consider the following examples of explanations of non-normative facts, not owed to Broome: • A bridge collapses. Here are some facts that may figure in the explanation of this fact. Though the bridge was well-designed and constructed, built from suitable materials and regularly (and competently) maintained, it was overloaded with cars and lorries while being buffeted by particularly strong winds. • Here is another example. A person dies from cancer. He succumbs to the illness because it is a particularly aggressive strain of cancer and the person’s immune system is compromised. He dies despite the fact that the cancer was noticed early on, and the person is treated with often-effective radiotherapy and chemotherapy. • A third example runs as follows. Someone loses a 100 m sprint Olympic final. She loses because she faces a tough opponent (who wins) and is not very good at handling extreme stress. She had a good chance of winning, however, because she is capable of running faster than her opponent, she prepared well for the race, and she was more motivated to win than anyone. In the above cases, explanations are offered for various non-normative facts. Interestingly, these explanations are structurally similar to Broome’s own examples of supposed weighing explanations of normative facts in at least four ways. First, in our examples, we can identify facts that, in some sense, play the “forthe-explanandum” role and those that play the “against-the-explanandum” role. Consider the first example. The fact to be explained is that a certain bridge collapses. Some of those facts mentioned in the explanation clearly contribute to the obtaining of this fact—the bridge is heavily weighed down with vehicles; the wind is extremely high. These facts can be considered the “for the bridge collapsing” facts. The other facts mentioned (that the bridge is well designed, constructed, and maintained, etc.), play the “against the bridge collapsing role”. These facts might typically explain why the bridge stays up. They are also facts that create an expectation that the bridge will stay up, and which thus create the need for an explanation of why it does not. Similar points apply to the other two examples. The cancer’s aggressiveness and the man’s compromised immune system play the for-dying role in the explanation of his death, while the early diagnosis and excellent medical care play the against-dying role. The athlete’s speed and well-preparedness play the against-losing role, and the

WEIGHING EXPLANATIONS

245

nervousness plays the for-losing role. The parallels with Broome’s examples are obvious—he too identifies for- and against-roles for the facts that constitute explanations of normative facts. Second, there is also a sense is which the for- (or against-) facts may “win.” The sprinter does in fact lose the race, so the for-losing facts “win out” over the againstlosing facts. We may, if we like, express this by saying that the for-losing facts outweighed the against-losing facts (though we explore below how useful it really is to think of these explanations as weighing explanations). Again, this closely resembles Broome’s examples. Third, the explanations of our non-normative facts frequently concern quantities of various kinds, and these quantities often make a difference to which fact obtains. The for-dying facts in the second example include the aggressiveness of the cancer and the weak immune system of the man. The for-collapsing facts in the first example concern the strength of the wind and the weight of the load on the bridge. This parallels Broome’s examples. The against-buying-the-wine facts in his first example concern the cost of the wine and the inconvenience that buying it causes. Both inconvenience and cost come in degrees (a cost may be low, high, reasonable, or exorbitant), and their degrees may affect whether one should buy the wine or not. The for-choosing-Montreux facts in Broome’s second example include Montreux’s relative proximity, the degree of which plays a crucial role in Broome’s weighing explanation. Fourth, Broome’s distinction between general and particular reasons finds a parallel in our examples. Take the third case above. We may fill out the explanation for the sprinter’s loss in more detail: The sprinter’s training regime, as recommended by top fitness experts, consisted in 90 minute workouts that concentrated on improving her explosive power and maintaining her core strength. Her personal best in previous competitions was 0.1 seconds better than her strongest competitor. Furthermore, she very much wanted to win the race to honor the memory of her father, who was also a sprinter and had died recently. On the other hand, she has consistently found that the more prominent the competition in which she races, the more she feels the pressure. In the past this has led to false-starts. In this (highly prestigious) race, she overcompensates and has a slow start. Though she gains on the eventual winner near the end of the race, she cannot reach her and comes second.

The above explanation simply spells out the original explanation. The above explanation of the loss does not rival our first explanation. Nor do the more detailed facts combine weight with the less specific facts to create stronger explanations.

Are our examples weighing explanations? We have seen that the examples we give above are similar in various ways to Broome’s examples of supposed normative weighing explanations. This raises the question of whether our examples are weighing explanations. That is, are they suitably analogous to mechanical weighing explanations?

246

STEPHEN KEARNS AND DANIEL STAR

To help us discover whether they are, let’s consider what Broome says about some essential elements of normative weighing explanations: These are some features that are essential to a weighing explanation of why you ought to F. The explanation must include one or more pro tanto reasons, either for you to F or for you not to F. Each of these reasons must be associated with something that is identified as its “weight”. The reasons and their weights play a characteristic role in the explanation. The role is that the weights of the reasons on each side are combined together in some way, and whether or not you ought to F is determined by the combined weights on either side. (52)

The important points are that the facts involved in weighing explanations have weights (both for the explanandum and against it), and these weights combine with each other in such a way that the facts that favor the (true) explanandum outweigh the facts against it.15 The major disanalogy between weighing explanations, as Broome describes them, and our explanations is that, though the facts in our explanations often involve quantities of certain types—quantities, furthermore, that play crucial roles in these explanations (as described in the previous section)—the facts involved do not have anything readily identifiable as weights. Consider our second example (the same general points apply to our other examples). The aggressiveness of the cancer, the weakness of the immune system, the earliness of detection, and the excellence of the treatment all play a role in explaining the death of the patient. However, the quantities involved in the fordying facts do not together produce a combined weight that outweighs the quantities involved in the against-dying facts. (If they did, presumably such a special quantity would be of great interest to medical science.) Rather than having weights that combine and are then set against each other, the for- and against-dying facts in the explanation of the man’s death play different roles.16 The job done by the for-dying facts in explaining the man’s death is fairly clear. Cancer, particularly aggressive strains, greatly increase the chances of death, especially in patients with compromised immune systems. The exact physical mechanisms of how this happens are relatively well understood. The for-dying facts really do explain (bring it about, cause, etc.) the man’s death.

15 A careful reader will notice that we are using the language of facts when talking of reasons, and that Broome views this as a perhaps optional “regimentation”. However, he also believes this regimentation is “harmless” (48), and he generally sticks to it, so we take it that there is nothing problematic with focusing on the regimented version of his view throughout. 16 It might be claimed that insofar as medical science makes use of data gathered concerning the probabilities of various outcomes occurring, given various conditions, it makes use of data that can be weighed. This may be so, but understanding weights in terms of probabilities leads to our conception of reasons as evidence, and cannot help Broome here (assuming he wishes to reject our account of reasons, as we believe he does).

WEIGHING EXPLANATIONS

247

What of the (losing) against-dying facts? What is their role in the explanation of the man’s death? We think there at least four main roles. First, such facts do play a background causal role in the man’s death (just as do the fact the man was born, that he eats a certain diet, and that he is subject to the Earth’s gravitational pull). These facts concern various physical processes. Such processes, and their interaction with the processes involved in the for-dying facts, help bring about the man’s death. (Question: Is this enough to make such facts causes of the man’s death? Answer: We don’t know. If they are causes, they are to the same extent that the man’s birth and attraction to the Earth are. They are certainly not salient causes. They also presumably do not bear the following counterfactual relation to the man’s death: were these facts to be false, the man’s death would not have occurred. Nor indeed, do these facts increase the probability of the man’s death relative to a suitably chosen alternative scenario.) Second, the against-dying facts (or facts of the same type) very often do explain why someone does not die of cancer. Many people do survive cancer thanks to early detection and suitably administered chemotherapy and radiotherapy. Part of what makes these facts against-dying facts is precisely that similar facts do prevent people dying, and, indeed, these very facts could have prevented the man from dying. They are potential explainers. Third, the against-dying facts play an epistemic role—they lead us, when considered alone, to increase our credence that the man will live. They constitute evidence that he will. In providing such evidence, the against-dying facts create a backdrop that helps us understand the explanation of the fact that the man dies. We understand that the threat posed to him by the cancer, given his excellent medical treatment, and such like, was (for example) particularly severe, or of a certain rare type. Fourth, and relatedly, the against-dying facts create the need for an explanation of the man’s death. Given that such facts can explain the man’s continuing to live, and that they constitute evidence that he will do so, this provides a context in which the fordying facts can explain why he died. In certain contexts, some facts do not cry out for explanation. These might be facts that we would expect to obtain in the normal course of events (this is not to say, of course, that such facts lack explanations, but simply that we do not go out of our way to look for them). We feel the need to explain a fact when we are provided with information that makes that fact surprising or expected to some extent. The against-dying facts in our example help provide such a context. To generalize then: in our examples of explanations, the against-explanandum facts play four roles in such an explanation: (1) they provide background causal facts; (2) they can explain (cause, bring about, etc.) the negation of the explanandum (and similar facts often do explain their explananda); (3) they provide evidence for the negation of the explanandum; (4) they create a need for explanation of the explanandum. Conspicuous by its absence is the idea that the against-explanandum facts have weights that combine together to compete with the combined weights of the forexplanandum facts.

248

STEPHEN KEARNS AND DANIEL STAR

If our explanations are not weighing explanations, then neither are Broome’s In the last two sections, we defended the following claims: (1) Broome’s examples of (supposed) normative weighing explanations are similar in many ways to our own examples of non-normative explanations; (2) our explanations are not weighing explanations. It does not follow from this that Broome’s explanations are not weighing explanations, but it is at least rather suggestive of it. Indeed, we think this suggestion withstands scrutiny. First, the for-F and against-F facts in Broome’s normative explanations play broadly the same four roles as the for-explanandum and against-explanandum roles in our explanations. Such roles do not call for the idea of weights that such facts have and that combine or outweigh the weights of other facts. Given this, unless there is some further reason to attribute such weights to Broome’s for-F and againstF facts, we already have an adequate (if rough) account of how such facts figure in the type of normative explanations that Broome highlights. Appealing to the weights of such facts is superfluous. Take Broome’s first example (the same points apply to his second example too). That you promised to take the wine, and that your doing so is not terribly onerous, rather straightforwardly explain why you ought to do so. The role of these for-takingthe-wine facts is to explain why (make it the case that; ground the fact that) you ought to take the wine. The against-taking-wine-facts that Broome considers are the (modest) cost of the wine and the (slight) inconvenience buying it causes you. These facts play an explanatory background role—they, together with the for-taking-thewine facts and many other facts (such as the continuing existence of off-licenses and your own birth), are part of a scenario which suffices for the fact that you ought to take the wine. The against-taking-the-wine facts too can explain, in different scenarios, why it is not the case that you ought to take the wine. They also serve as evidence that it is permissible not to take the wine (and even, perhaps, evidence that you ought not to take the wine), and they create a context in which an explanation of why you should indeed take the wine is called for. In short we can explain the role of these facts in Broome’s explanations without invoking the idea that they have weights that combine or compete with the weights of other facts.17 Second, just as there is no clear way of combining the various quantities involved in the facts in our non-normative explanations into one quantity analogous to a weight, there is no clear way of translating the various quantities involved in the facts in Broome’s normative explanations into one quantity analogous to a weight. Consider Broome’s second example. The pleasantness of Montreux and its proximity do not plausibly combine into one quantity—a weight—that can outweigh the weight 17 By this we mean that it is possible to give a plausible account of Broome’s examples of normative explanations, according to which the facts that Broome says explain why you ought to bring the bottle, or go to Montreux, do their explanatory work without possessing weights. Of course, Broome’s own account of how these facts explain normative facts does appeal to weights.

WEIGHING EXPLANATIONS

249

that is (somehow) determined by the exoticness of Marrakesh and thus make it the case that you ought to holiday in Montreux. Clearly such quantities do play an important role in explaining what you ought to do, but they do not do so via a generic one-size-fits all quantity.18 Might Broome here appeal to value as being such a quantity? It is not clear he can. First, Broome is open to the possibility that value should be understood in terms of reasons, having defended this option in print previously, and thus might not wish to appeal to value to help analyze reasons (57–8). Second, given the possibility of there being incommensurable values (something Broome is also explicitly open to), there is no singular quantity of value that can be straightforwardly measured. Third, and most importantly, value is a quantity assigned to actions and consequences of actions, not to the facts that explain why these actions ought to be performed. Given this, value cannot be the quantity called “weight” that plays the roles Broome assigns it.

Where to go from here? We have argued that Broome’s normative explanations are not weighing explanations, as they are more akin to our non-normative explanations, which are not weighing explanations. What kind of responses are open to Broome? We think there are at least four. First, Broome may concede that his examples are not weighing explanations, but this is merely because he chose the wrong examples. Broome gave himself this wriggle-room by saying that he does “not wish to assert that any particular weighing explanations are genuine” (55). He may insist that there are indeed normative weighing explanations and that reasons can be analyzed in terms of them. The problem with this approach is that Broome’s examples are paradigm cases of normative explanations (this is, no doubt, why Broome chose them). If even these types of explanations do not count as normative weighing explanations then it is doubtful that any do. Second, Broome may say that while his cases are indeed analogous to ours, our cases are indeed (non-normative) weighing explanations. The for-explanandum facts in our cases really do possess weights that combine and are then set against the combined weight of the against-explanandum facts. But, again, aside from this idea being implausible on its face, it is superfluous given the roles we have pointed out for the for- and against-explanandum facts. What would we be adding to our understanding of such explanations by insisting upon a relatively obscure quantity possessed by these facts that is highly analogous to weight?

18

It is true that we sometimes assign exact degrees to entities in formal contexts (such as expected value, or degrees of credence) without actually being in an epistemic position to measure such degrees, or perhaps even when we doubt there are such exact degrees (rather than something fuzzier that we are modeling). The issue here, though, is not implausible metaphysical exactness, nor our epistemic limitations. Rather, we doubt there is anything useful to measure at all (however inexact or inaccessible).

250

STEPHEN KEARNS AND DANIEL STAR

Third, Broome may simply deny that his cases and ours are analogous. While we have tried to show that they are, such issues are complex and relevant similarities and differences between the cases are not immediately obvious. Perhaps Broome’s forand against-F facts do not play the roles we attribute to our for- and againstexplanandum facts. We describe our case that they do above. More plausibly, Broome might reply that appeal to weights, while superfluous to account for the roles of the facts that figure in our explanations, are not superfluous to account for all the roles of the facts that figure in his. While his for- and against-F facts may indeed play the roles we suggest, they also play roles that include their having weights. We concede that Broome’s fact may play more roles than the facts that figure in our non-normative explanations. For example, it might be that the against-going-toMontreux facts (i.e. the for-going-to-Marrakesh facts) help explain why it is somewhat regretful if you do not go to Marrakesh, or why going to Marrakesh would be a generally fine plan, or that going to Montreux is not ideal, even if something you ought to do. What we doubt is that such roles require us to assign weights to the facts that figure in Broome’s normative explanations. It strikes us that everything Broome wants these facts to do can be done without their having weights. Fourth, and lastly, Broome may concede that his explanations, like ours, are not weighing explanations, but that we are still able to analyze reasons with respect to them. That is, on this proposal, a reason to A is a fact that plays the kinds of roles we have highlighted in this chapter. To go this route, however, would be to give up the idea that reasons have weights that can combine, compete, outweigh, and so on. This, we believe Broome, would concur, would be to give up too much. We take none of these options. Rather, we somewhat tentatively conclude that there are no normative weighing explanations and thus that it is not possible to analyze reasons with reference to them. Like Broome, we think the idea that reasons have weights is crucial to understanding the nature of reasons. In previous work (Kearns and Star 2008, 2009, amongst others), we pointed out that evidence also has weights. We hold out hope that we can understand the fact that reasons have weights as resting on the fact that reasons are a kind of evidence (and that evidence has weights). For our latest thinking on this see Kearns (forthcoming) and Star (forthcoming). Part of our reason for thinking this is precisely our skepticism concerning the supposed weights to which Broome appeals. If there are no such weights, we all have to look elsewhere.

References Broome, J. (2008) Reply to Southwood, Kearns and Star, and Cullity. Ethics 119: 96–108. Broome, J. (2013) Rationality through Reasoning. Chichester, UK; Malden, MA: Wiley Blackwell. Fletcher, G. (2013) A Millian objection to reasons as evidence. Utilitas 25: 417–20. Kearns, S. (forthcoming) Bearing the weight of reasons. In Errol Lord and Barry Maguire (eds.) Weighing Reasons. Oxford University Press.

WEIGHING EXPLANATIONS

251

Kearns, S., and Star, D. (2008) Reasons: Explanations or evidence? Ethics 119: 31–56. Kearns, S., and Star, D. (2009) Reasons as evidence. Oxford Studies in Metaethics 4: 215–42. Kearns, S., and Star, D. (2013) Weighing reasons. Journal of Moral Philosophy 10: 70–86. Scanlon, T. M. (2014) Being Realistic about Reasons. Oxford: Oxford University Press. Schroeder, M. (2007) Slaves of the Passions. Oxford; New York: Oxford University Press. Setiya, K. (2014) What is a reason to act? Philosophical Studies 167: 221–35. Star, D. (2011) Two levels of moral thinking. Oxford Studies in Normative Ethics 1: 75–96. Star, D. (forthcoming) Knowing Better. Oxford University Press. Thomson, J. J. (2008) Normativity. Chicago, IL: Open Court.

Index Adler, Matthew 113 agency 207–8 group 34, 207, 213, 215–21 aggregation 12, 14, 16–18, 33–4, 43, 55, 63, 145–6, 215–17 Allais paradox 43–5 anthropocentrism 38–40 Aristotle 137 Arrhenius, Gustaf 88, 90–2, 97–102 Arrow, Kenneth 56–7 Arrow’s impossibility theorem 13–18, 56–7, 59

evidence 192–4, 197–203, 208–14, 216–17, 221–4, 227–9, 233, 237–41, 247–8, 250 evidentialism 145 Ewing, A. C. 175–6 existence 32, 35–6, 41, 78, 81–3, 87–94, 95–100, 108–9, 110–14 explanation: complete 236 normative 144, 181, 233, 240, 248–50 weighing 145–7, 149, 177, 179–82, 185, 232–5, 237, 242–8

Bernoulli hypothesis 61–4 Blackorby, C. 82, 116 Bossert, W. 82 Broad, C. D. 115 Buchanan, James 19 Bykvist, Krister 97

fairness 42, 46, 48, 51, 105–6 fittingness 195 Friedman, Milton 19

Chisholm, Roderick 89, 158, 168 Coase, Ronald 13, 27 consequentialism: outcome vs prospect 147–9 credence 129–31, 137–8, 212, 222, 247 Dancy, Jonathan 5, 189 Danielsson, Sven 5, 89 Debreu, Gerard 63 deliberation 237, 240 deontic superiority 162–3 detachment: deontic 167–8, 170, 173, 175–6 factual 166–8, 170, 172–5 forced 169–70, 176 Diamond, Peter 35, 43, 45–7 discursive dilemma 212, 214, 216, 219, 220, 223–4 Donaldson, D. 82 duty: prima facie vs duties proper 184–6 Edgeworth, Francis 42 egalitarian dominance condition 118, 121–3 egalitarianism 42–3, 119 Enkrasia 148–50, 241 equality 2, 6, 36, 42–3, 45–7, 51, 67, 119; see also egalitarianism distributive 43 social 42, 51

good: communal 37, 42–3, 45 comparative 88, 91–2, 97, 99, 111 -for 31, 61, 63–4, 69, 90 general 32–4, 36–9, 43, 50, 61, 67 instrumental 39, 195 intrinsic 19, 40 non-personal 36–8 of animals 36, 39–40 of average life 74, 81–3 of states of affairs 30, 45, 51, 63, 89–91, 195 overall 12, 31, 38, 43, 51, 69 pattern 65, 67, 70 personal 31–4, 36, 50–1, 61, 63–4, 67–70 private 16 prospective 142, 148 public 16, 23, 31 social; see good, communal Goodpaster, Kenneth 38–40 Gorman, W. M. 63 grounding relation 158, 237 Harman, Elizabeth 97 Harsanyi, John 5–6, 34–5, 39, 45, 61 Hobbes, Thomas 183, 226 Holtug, Nils 97–8, 100–3, 113–14 incommensurability 41, 72–3, 116–18, 122, 199, 202, 249 independence of irrelevant alternatives 16, 56 inequality aversion condition 105, 118–22 interpersonal addition theorem 61, 63–4 intertemporal addition theorem 64 intuitionism 39

254

INDEX

Kolodny, Niko 207, 209–13, 216, 219, 223 Krähmer, D. 132–3, 135 Krantz, David 63 land ethic 39, 40, 51 Leopold, Aldo 39–41, 51 Lewis, David 5, 54 liberalism 21, 23–9 liberal paradox 13 liberty 18–22, 26, 28–9 lottery paradox 212 Mackie, Gerry 14, 15 Moore, G. E. 39, 148, 149 Nagel, Thomas 6, 189 non-contradiction 209–10, 212, 216 non-dictatorship 15–17, 56 non-identity problem 35, 41 normative: architectures 191, 193–5, 197–9, 203 conflicts 29, 190–3, 195, 201–5 domains 193 monism vs pluralism 197–9, 201–2, 205 separatism vs non-separatism 193–4, 203, 205 normativity 6, 17, 146, 151, 156, 179, 183, 183–5, 189, 197, 202 of rationality 151 basic vs derivative 151, 195, 197 sources of 149, 193–9, 200–4; see also reasons, sources of structures of 193, 204 Ordershook, Peter 14, 15, 18 ought: central 148–9, 241 final 190–4, 199, 205; see also ought, overall fundamentalism 149 kinds of 149, 152; see also requirements, sources of objective 148–9, 154, 241 overall 146, 149, 153, 160–2, 183, 185, 191 owned 185, 187 prospective 147–8, 241 subjective 142–3, 152–3 ‘ought’ implies ‘can’ principle 163–4, 170 Parfit, Derek 4, 6, 65, 112, 189, 238 Paretianism 15, 18–23, 26–9 Pareto principle 15, 25, 56, 103–6, subjunctive weak 116–17, 119–21 person-affecting principle 87, 102, 110–11, 113, 118, 120 subjunctive 114–15, 118, 121, 123 possible worlds 33, 57, 99, 161–2 preface paradox 212

preference: individual 13, 15, 24, 26, 40, 55–6 satisfaction 15, 18–22, 25–7, 29; see also Paretianism social 24, 55–7 Prichard, H. A. 184, 186–7 principle of temporal good 64, 69–70 prioritarianism 64 probability 48–9, 105–6, 129, 132, 148, 241, 247 quality-adjusted life years (QALY) 73, 76–8, 80, 85 Rabinowicz, Wlodek 5, 88, 90–2, 97–8, 100–2, 113 radical interpretation 52–4, 56–9 radical pragmatism 198, 200–1, 205 rational choice theory 34, 43–5, 135–6 rationality 6, 14, 53–4, 56, 76, 140, 142, 149–52; see also requirements, rational group 213–16; see also agency, group practical 148, 150, 192 Raz, Joseph 189 reasoning 6, 13, 49, 82, 192 practical 6, 73, 145–6, 148, 237 reasons: as evidence 232–3, 238 as explanations 142–7, 177–81, 183, 186, 233–7, 240, 250 conclusive 160, 208, 210, 212, 214–16, 228, 239; see also reasons, decisive; reasons, pro toto; reasons, overall contributory 178–9, 183; see also reasons, pro tanto decisive 146, 160 epistemic 145 explanatory 143, 183, 233 externalism 190 for action 141–2, 150, 194–6, 198, 199–201, 203, 239–40; see also reasons, kinds of; reasons, practical for belief 194, 198, 200–3; see also reasons, kinds of fundamentalism see reasons, primitivism kinds of 190, 192 motivating 143; see also reasons, explanatory normative 142–3, 157, 159, 190, 205, 233–4, 238–9; see also reasons, pro tanto; reasons, pro toto overall 140–1, 143, 147, 149–50, 152 perfect 160; see also reason, pro toto practical 145, 192–3, 199 primitivism 141–2, 238 pro tanto 140–1, 143, 156–7, 150, 153–4, 157, 159–62, 177–83, 185–6, 232–5, 239, 242–3, 246

INDEX

255

pro toto 143, 146, 177–9, 233–7, 239, 242; see also reasons, overall responsiveness 215 sensitivity 207–9, 212–13, 215–19, 221–4; see also reason, requirements of sources of 197 specialism vs generalism 198–200, 202–4 state-given vs object-given 192; see also radical pragmatism theoretical 192–3, 196 weight of 141–5, 145, 160, 179–82, 186, 199, 232–4, 238, 242–9 reductionism 31, 33–4, 37, 41–2, 51 regret 44, 130 types of 130–2, 136–7 under uncertainty 132–6 repugnant conclusion 82, 105, 108, 111, 118–22 requirements: agential 208 moral 156, 159, 161, 163–7, 171, 175, 242 of reason 208–9, 211–12, 215, 222, 227 rational 142, 148, 151, 191, 207, 210, 212, 216, 241 sources of 158–60 wide- vs narrow-scope 157, 166, 171, 210 right-making 178, 182 Roberts, Melinda 97, 113 Ross, W. D. 162, 184–6 Ross’s paradox 165

Singer, Peter 36, 38 Smith, Adam 18 social choice theory 12–13, 20, 29–30, 46, 55 Sosa, Ernest 89 speciesism 36 Stone, R. 132–3, 135 Stratton-Lake, Philip 185 Suppes-Sen principle 103–4 supervenience 33–5, 39

Scanlon, T. M. 42–3, 141–2, 144, 238 Schroeder, Mark 239 Sen, Amartya 13, 17–20, 22–7, 29 separability 61, 63, 67 of people 62–63, 67 of states of nature 62–3 of time 62, 65–70 theorem 63 Sider, Ted 54 Sidgwick, Henry 152–5

value see good value-for see good, -for

testimony 227, 240 trade-offs 12, 16, 18, 23; see also marginal adjustments transitivity 16–17, 56 uncertainty aversion vs. proneness 132–3, 135 utilitarianism 42, 150 act 151 average 104, 115 classical 111, 118, 120, 162 critical level 82–4, 115, 116, 117 hedonistic 240 total 82–4, 104 utilitarian principle of distribution 61–3 utility: diminishing marginal 42 expected 76, 115 social 34, 45 theory 31, 34, 44, 52, 76

wellbeing 35, 40, 42–3, 49–50, 67, 69–70, 76–9, 81–3, 88–90, 92–3, 95–8, 100, 102–9, 144, 150–1 primitive absolute 93 Williams, Bernard 38, 141, 189 willingness-to-accept 75 willingness-to-pay 41, 73–5, 77