Preferences 9783110804294, 9783110159103

Citation preview

Preferences

Perspektiven der Analytischen Philosophie Perspectives in Analytical Philosophy Herausgegeben von Georg Meggle und Julian Nida-Rümelin Band 19

w DE

Cl Walter de Gruyter · Berlin · New York 1998

Preferences Edited by Christoph Fehige and Ulla Wessels

w DE

Walter de Gruyter · Berlin · New York 1998

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability. Library of Congress Cataloging-in-Publication

Data

Preferences / edited by Christoph Fehige and Ulla Wessels. p. cm. — (Perspektiven der analytischen Philosophie ; Bd. 19 = Perspectives in analytical philosophy) Based on a conference held in June 1992 in Saarbrücken, Germany, and in Saarlouis, Germany. Includes bibliographical references and indexes. ISBN 3-11-015007-7 1. Preferences (Philosophy) — Congresses. I. Fehige, Christoph. II. Wessels, Ulla. III. Series: Perspectives in analytical philosophy ; Bd. 19. B105.P62P74 1998 128'.3 —dc21 97-28699 CIP

Die Deutsche Bibliothek — Cataloging-in-Publication

Data

Preferences / ed. by Christoph Fehige and Ulla Wessels. — Berlin ; New York : de Gruyter, 1998 (Perspektiven der analytischen Philosophie ; Bd. 19) ISBN 3-11-015007-7 brosch. ISBN 3-11-015910-4 Geb.

© Copyright 1998 by Walter de Gruyter GmbH & Co., D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Printed in Germany Typesetting: Thomas Fehige, Münster Printing: Arthur Collignon GmbH, Berlin Binding: Lüderitz & Bauer, Berlin Cover design: Rudolf Hübler, Berlin

All I want is to sit on my arse and fart and think of Dante. Samuel Beckett Desires and wants, however intense, are not by themselves reasons in matters of justice. The fact that we have a compelling desire does not argue for the propriety of its satisfaction any more than the strength of a conviction argues for its truth. John Rawls Take any demand, however slight, which any creature, however weak, may make. Ought it not, for its own sole sake, to be satisfied? If not, prove why not. The only possible kind of proof you could adduce would be the exhibition of another creature who should make a demand that ran the other way. The only possible reason there can be why any phenomenon ought to exist is that such a phenomenon actually is desired. William James

Preface Preferences is a collection of essays on the concept and the role of preferences (desires, and the like) in practical reasoning. Ground covered includes welfare, prudence, rational decision making, and all areas of moral philosophy: ethics (applied and not so applied), metaethics, and deontic logic. A special symposium looks at possible preferences and their significance in matters of life and death, including the notoriously thorny question how many people there should be. All the essays are published here for the first time. The book is not just for specialists. We have given it an introduction that, though it may move swiftly, at least starts from scratch; a selected bibliography is also provided. Most of the authors were able to meet in advance, and to present, discuss, and then revise their contributions. But the line has to be drawn somewhere, and authors who receive a reply in this volume were not permitted to adjust their papers in the light of the final version of the reply. The initial exchange took place in Saarbrücken and Saarlouis in June 1992. *

Everybody has been very kind to us. Georg Meggle — selfless and cheerful as usual - co-designed the project and supported it from beginning to end. When we proposed the meeting, we were backed up by Franz von Kutschera and Wolfgang Lenzen. Barbara Schumacher helped prepare and run it. The editors of Perspectives in Analytical Philosophy, Georg Meggle and Julian Nida-Rümelin, have welcomed the book in precisely the form we suggested. The authors have been co-operative and patient throughout. Christopher Abbey and Sean Matthews have given valuable advice, linguistic and otherwise, to many of us. Kornelius Bamberger was able, and kind enough, to convert most of the data that the contributors sent us. Thomas Fehige gave these data a neat, uniform lay-out. Patrick Agsten, Monika Claßen, Franziska Muschiol, Ulf Schwarz, and Valentin Wagner have assisted us, efficiently and in numerous respects; the same holds true of Karin Thom. With this list in chronological order, one important acknowledgement comes last: de Gruyter publishers. Working with Hans-Robert Cram was a pleasure; ditto, at the technical end, with Grit Müller.

viii

Preface

The conference that gave rise to this book was made possible by the financial assistance of: Deutsche Forschungsgemeinschaft (DFG), Fritz Thyssen Stiftung, Ministerium fur Wissenschaft und Kultur des Saarlandes, Universität des Saarlandes, and Vereinigung der Freunde der Universität des Saarlandes. The DFG (research project "Was zählt?") has also funded our own work on this volume. We thank all these persons and institutions for their support. *

We share the belief, now regarded in some quarters as both unsound and oldfashioned, that, in essence, morality is all about welfare, and welfare all about preferences. Some of the contributors to this volume would agree, some would not. With luck, this collection will help advance matters a little. Christoph Fehige and Ulla Wessels Leipzig, January 1998

Contents Abstracts

xiii

Introduction CHRISTOPH FEHIGE AND ULLA

WESSELS

Preferences — an Introduction Preferences - a Short Bibliography

xx xliv

Part I: Preference and Decision ELIJAH

MILLGRAM

Deciding to Desire SYDNEY

SHOEMAKER

Desiring at Will (and at Pill): A Reply to Millgram CHRISTOPH

ANTONELLA

57

BRANDT

The Rational Criticism of Preferences

63

KUSSER

Rational by Shock: A Reply to Brandt J. DAVID

GEORG

78

VELLEMAN

Is Motivation Internal to Value?

88

MEGGLE

Motivation and Value: A Reply to Velleman SVEN

33

CORRADINI

Intrinsic Desirabilities: A Reply to Lumer RICHARD B .

26

LUMER

Which Preferences Shall Be the Basis of Rational Decision?

ANNA

3

103

DANIELSSON

Numerical Representations of Value-Orderings: Some Basic Problems ULRICH

114

NORTMANN

Interval Orders Defended: A Reply to Danielsson

123

χ

Contents

Part II: Preference and Metaethics LENNART AQVIST

Prima Facie Obligations in Deontic Logic: A Chisholmian Analysis Based on Normative Preference Structures UWE

135

BOMBOSCH

The Meaning of "Ought, Prima Facie" and Decision Situations: A Reply to Aqvist F R A N Z VON

KUTSCHERA

Values and Duties WILFRIED

163

HINSCH

Beyond Duty: A Reply to von Kutschera MARCUS

172

WILLASCHEK

Agency, Autonomy and Moral Obligation HILARY

156

176

BOK

Autonomy and Morality: A Reply to Willaschek

204

DAVID WIGGINS

In a Subjectivist Framework, Categorical Requirements and Real Practical Reasons DAVID

GAUTHIER

Subjective Obligation: A Reply to Wiggins ALLAN

212 233

GIBBARD

Preferences and Preferability JULIAN

239

NIDA-RÜMELIN

Goodness and Rational Preferability: A Reply to Gibbard

260

Part III: Preference and Ethics

JOHN

BROOME

Extended Preferences RUDOLF

271

SCHÜSSLER

Wish You Were Me: A Reply to Broome and a Comment on Harsanyi's Extended Preference Theory

288

Contents

xi

RAINER HEGSELMANN

Experimental Ethics: A Computer Simulation of Classes, Cliques, and Solidarity ULRICH

298

KRAUSE

Solidarity among Rational Egoists: A Reply to Hegselmann

321

RAINER W E R N E R TRAPP

The Potentialities and Limits of a Rational Justification of Ethical Norms, or: What Precisely is Minimal Morality?

327

ANTHONY SIMON LADEN

A Hobbesian Choice: A Reply to Trapp CHRISTOPH FEHIGE, RICHARD Μ .

361

H A R E , W O L F G A N G LENZEN, JEFF

M C M A H A N , PETER SINGER, T H O M A S SPITZLEY, AND U L L A W E S S E L S

Symposium on Possible Preferences

367

C H R I S T O P H F E H I G E AND U L L A W E S S E L S

Introduction to Possible Preferences

367

PETER SINGER

Possible Preferences RICHARD M .

383

HARE

Preferences of Possible People

399

W O L F G A N G LENZEN

Who Counts?

406

ULLA WESSELS

Procreation JEFF

429

MCMAHAN

Preferences, Death, and the Ethics of Killing

471

T H O M A S SPITZLEY

McMahan on Psychological Continuity and the Value of Future Goods

503

C H R I S T O P H FEHIGE

A Pareto Principle for Possible People

508

xii

Contents

Notes on Contributors

544

Name Index

550

Subject Index

555

Abstracts

These are abstracts of the papers that receive a reply, not of the replies themselves. The abstracts appear in the alphabetical order of the authors' names; for the contributions to the symposium on possible preferences, see the final abstract.

LENNART A Q V I S T

Prima Facie Obligations in Deontic Logic: A Chisholmian Analysis Based on Normative Preference Structures (page 135)

The paper argues for an analysis of the W. D. Ross notion of prima facie obligation which results from adding a certain Chisholm-style definition to the system G of Dyadic Deontic Logic, supplemented with so-called propositional quantifiers. In the semantics for that system a von-Kutschera-inspired conception of normative preference structures turns out to be of vital importance. U W E BOMBOSCH'S comment, "The Meaning of 'Ought, Prima Facie' and Decision Situations", begins on p. 156.

RICHARD B . BRANDT

The Rational Criticism of Preferences (page 63)

Preferences are rationally criticized if vivid representation of confirmed beliefs will result in a reversal or strengthening. It is universally agreed that plans can be so criticized but not basic preferences for types of events. Define "preference" as "desiring more". Psychologists agree that desire for an event-type is increased if an event-type has been associated with pleasant events in the past - conditioning by contiguity. (The status of bodily needs - like thirst and hunger - is different; such needs are fixed by chemical imbalances in the body.) But many events are pleasant for evolutionary reasons; if they weren't pleasant and hence the pleasant type of event wanted, the individuals would not survive. This connection - pleasant event, being wanted from classical conditioning, and hence preference — opens the way to rational criticism. For reflection on facts can alter preferences when the preference is seen to be a result (1) of inadequate representation of facts, or (2) of influence by temporary motivational states, or (3) of stimulus generalization from abnormal cases, or (4) of overlooking unpleasant facts about the object, or (5) of failure of making discriminations, or (6) as a result of suggestions by teachers, or (7) as a result of false or unjustified factual beliefs. The author suggests we

Abstracts

XIV

say a preference has been rationally criticized if reflection on these defects results in a modification of the preference. A N N A KUSSER S

comment, "Rational by Shock", begins on p. 78.

JOHN

BROOME

Extended Preferences (page 271) Ordinalism is generally taken to imply that interpersonal comparisons of good are impossible. But some ordinalists have argued that these comparisons can be made in a way that is consistent with ordinalism, on the basis of extended preferences. This paper shows that this argument is mistaken, and ordinalism is indeed incompatible with interpersonal comparisons of good. R U D O L F S C H Ü S S L E R S comment, "Wish You Were Me: A Reply to Broome and a Comment on Harsanyi's Extended Preference Theory", begins on p. 288.

SVEN DANIELSSON

Numerical Representations of Value-Orderings: Some Basic Problems (page 114)

Measures of value or preference usually presuppose value or preference relations which are weak orders. Numerical representations of semiorders and of interval orders have to some extent also been considered. It is fairly obvious, however, that value- and preference-orderings often are not, and should not be expected to be, even interval orders. A way of representing partial orders is suggested. ULRICH NORTMANN'S

comment, "Interval Orders Defended", begins on p. 123.

C H R I S T O P H F E H I G E AND U L L A W E S S E L S

Preferences - an Introduction (page xx)

In theories of practical reasoning, we can encounter preferences (desires, and the like) in five places. Two of them are the form and the content of rationality; the other three are the form, the content, and the foundation of morality. This introduction presents the terrain and explains its overall structure; it also pays a brief visit to each of the locations and points out some of the disputes surrounding them. The doctrine of preferentialism and its problems will be a convenient leitmotiv, since it is widely held and employs preferences, and preferences only, on all the five levels. The tour is structured as follows. After a prologue that sketches preferentialism, we will consider the very

Abstracts

XV

concept of a preference (section 1). We will then look at the possible roles of preferences in rationality (section 2), and at the triad of roles they might play in morality (section 3). Finally, there is a selected bibliography.

ALLAN

GIBBARD

Preference and Preferability (page 239)

What does "good" mean? The paper starts with two vague truisms: That goodness is a matter of preferability, and that of two things, the preferable one is the one it is rational to prefer. In his book Wise Choices, Apt Feelings (1990), the author had proposed a theory of what "rational" means; this paper faces two questions: (1) What concept of preference, if any, fits the formula that the preferable of two things is the one it is rational to prefer? (2) How should this formula be filled out: Rational for whom to prefer, when? Classical decision theory treats preference as consisting in ones disposition to choose. Such revealable preference will probably not serve as a good explanatory concept in a scientific psychology, it is suggested, but it may be much the concept that is needed for purposes of defining preferability. Roughly, the preferable of two things is the one it is rational to choose. This needs to be refined, though: Talk of what is preferable to what purports to be neutral among parties to the conversation. Indeed one use of the term "good", prominent among philosophers, treats all humanity as our conversational group. If rational intrinsic preferences need not be impartial, then not all considerations that bear on rational choice need be matters of goodness so understood. Good-making considerations will be those considerations that bear on choices consequentially and neutrally, and goodness will be a matter of how these good-making considerations sum up. JULIAN N I D A - R Ü M E L I N S

comment, "Goodness and Rational Preferability", begins on

p. 2 6 0 .

RAINER

HEGSELMANN

Experimental Ethics: A Computer Simulation of Classes, Cliques, and Solidarity (page 298)

The article deals with two questions: (a) Can relations and networks of solidarity emerge in a world exclusively inhabited by rational egoists, who are unequal and choose their partners opportunistically? (b) If networks of solidarity do emerge in such a world, what do they look like? By means of computer simulations it is shown that networks of solidarity can emerge in such a world. But the networks will show quite distinct features of some class segregation. ULRICH

KRAUSE'S

comment, "Solidarity among Rational Egoists", begins on p. 321.

Abstracts

XVI

F R A N Z VON K U T S C H E R A

Values and Duties (page 163)

It is argued that both deontological and consequentialist principles have their legitimate place in ethics, and that neither kind is reducible to the other. The problem, then, is how to integrate them into a unified system. A simple solution would be to have duties override value considerations, and the discussion centers on the merits and shortcomings of this proposal. WILFRIED HINSCH S

comment, "Beyond Duty", begins on p.

CHRISTOPH

172.

LUMER

Which Preferences Shall Be the Basis of Rational Decision? (page 33)

Theories of rational decision normally distinguish basic and other preferences, using only the former for calculating an agents utility function. The idea behind the distinction is that, on the one hand, a theory of rational decision must allow criticism of at least a part of the agent's actual preferences; on the other hand, so as not to lose touch with the agent s real interests, it must rely on his factual preferences. Different decision theories have declared as basic various sets of preferences, thereby arriving at very different utility functions. Therefore, the question of which preferences shall be basic is of large practical importance. Nonetheless, it has rarely been discussed. The article criticises some standard approaches, but mainly develops criteria for the selection of basic preferences. One of the principles for the selection of basic preferences, for example, is epistemic rationalisation. From these principles, then, 12 conditions of adequacy for the selection of the preferential basis are derived, e.g. taking over only intrinsic preferences, and of these not the single preferences but their underlying criteria. ANTONELLA CORRADINI S

comment, "Intrinsic Desirabilities", begins on p.

57.

ELIJAH MILLGRAM

Deciding to Desire (page 3)

We do not, and cannot, normally come to have desires by simply deciding to have them. It is argued that this is not a contingent fact, and that the explanation for this fact shows a widely held view of practical reasoning to be false. SYDNEY SHOEMAKER'S

comment, "Desiring at Will (and at Pill)", begins on p.

26.

Abstracts

xvii

RAINER W E R N E R TRAPP

The Potentialities and Limits of a Rational Justification of Ethical Norms, or: What Precisely is Minimal Morality? (page 327)

Starting from the insight that, due to certain epistemological peculiarities of 'normative truth', normative statements cannot claim to be objectively (= Ο) true, the paper systematically works out the idea of basing the O-validity of general moral norms on their O-utility rather than on their O-truth. According to this idea any restriction of choice, in an «-person-conflict of interests S, qualifies as O-valid if it fulfills one of the two following criteria: Either compliance to it by at least a specifiable number k of the « individuals in S would make everybody already in each instance o f S better off than norm-free anarchy (= criterion (I), which establishes two classes of unconditionally Ovalid norms each avoiding a corresponding type of trap of prudence), or it would, under certain assumptions of the interacting individuals on the probabilities of the roles taken in their respective lifetime-sequences of situations of type S, increase everybody's utility payoff in the long run (— criterion (II), which establishes three classes of only conditionally O-valid norms). Thus even 'non-veiled' rational egoists refusing to initially concede any rationally unfoundable moral protonorm whatsoever, one that demands some (Harsanyian, Rawlsian, ... ) impartial standpoint in considering an agreement on mutual restrictions of behaviour, will - so it is argued - have to contract on at least these norms in a fictitious original agreement. The latter's extension defines the system Mm;n of minimal morality. Though being far more comprehensive than related approaches to 'morals by agreement' (notably Gauthier's), Mmwill finally be assessed as morally insufficient due to its not containing any compensatory norms. Since some of the latter, according to widespread convictions, are indispensable and since these, at the same time, are not justifiable as O-valid on the basis of whatever brand of veil-free contractarianism, any progamme of founding a satisfactory moral system on mere collective rationality is considered as doomed to fail eventually. A N T H O N Y S I M O N LADEN'S

comment,

"A

Hobbesian Choice", begins on p. 361.

J . DAVID VELLEMAN

Is Motivation Internal to Value? (page 88)

The view that something's being good for a person depends on his capacity to care about it - sometimes called internalism about a person's good - is here derived from the principle that 'ought' implies 'can'. In the course of this derivation, the limits of internalism are discussed, and a distinction is drawn between two senses of the phrase "a person's good". GEORG MEGGLE'S

comment, "Motivation and Value", begins on p. 103.

Abstracts

XV111

DAVID

WIGGINS

In a Subjectivist Framework, Categorical Requirements and Real Practical Reasons (page 212)

In this paper, the author tries to show that Hume, interpreted as a genealogist of morals - not as empiricist, prescriptivist, projectivist, expressivist or error theorist - , can do justice to the moral phenomena that moral philosophers discuss under the heading of the categorical imperative. His position on this matter is compared and contrasted with that of Kant. It is claimed that Hume discusses the real reasons, such as they are, why, regardless of inclination, we should heed the categorical requirements of morality. DAVID GAUTHIER'S

comment, "Subjective Obligation", begins on p.

MARCUS

233.

WILLASCHEK

Agency, Autonomy, and Moral Obligation (page 176)

The paper proposes and, in part, defends an understanding of human agency, autonomy, and moral obligation as integral parts of our concept of a person. Specifically, the first part (sects. 1-12) argues for a causal theory of action in which the acting person plays a central role in the causal history of her actions. The person exercises her causal influence according to normative principles of rationality. That presupposes some independence from her own motivation including the ability to acknowledge or reject parts of it as a basis of her rational decisions. This ability is constitutive of the autonomy of the person. The second part (sects. 13-29) presents an argument to the effect that the concept of autonomy presupposes a general universalist principle of morality. Autonomy involves a distinction between motives that are 'authentic' and motives that are not. This distinction does not rest on a substantive idea of what autonomous action is, but rather on a formal or procedural notion. Nevertheless, it presupposes a normative standard which is different from and largely independent of the motives a person in fact has. This standard can be found in the ideas of impartial benevolence and universal rational consent which inform universalist conceptions of morality. HILARY BOK'S

comment, "Autonomy and Morality", begins on p. 204.

CHRISTOPH JEFF M C M A H A N ,

FEHIGE, RICHARD Μ .

HARE, WOLFGANG

LENZEN,

P E T E R S I N G E R , T H O M A S SPITZLEY, AND U L L A

WESSELS

Symposium on Possible Preferences (page 367)

Sometimes our actions make a difference not just to the frustration or satisfaction of preferences that exist (have existed, or will exist), but to the very question which preferences will exist; so they require us to look not only at actual, but also at possible pref-

Abstracts

xix

erences. These actions, their morality and their rationality, are the topic of the present symposium. Most choices concerning a preferrer's life or death are dramatic and obvious examples of such actions (no life, no preference), and they have come to dominate the discussion of possible preferences, and this symposium as well. Thus, on the more applied level, this is a symposium about the morality of conception and contraception, abortion, population policy and killing, about the value of life and the badness of death. For a guide to this web of issues, see the "Introduction to Possible Preferences" at the beginning of the symposium (p. 367); more information on the various contributions, and on how they relate to each other, is given in the last section (pp. 379-81) of that introduction.

C H R I S T O P H FEHIGE AND U L L A W E S S E L S

Preferences - an Introduction* Abstract: In theories of practical reasoning, we can encounter preferences (desires, and the like) in five places. Two of them are the form and the content of rationality; the other three are the form, the content, and the foundation of morality. This introduction presents the terrain and explains its overall structure; it also pays a brief visit to each of the locations and points out some of the disputes surrounding them. The doctrine of preferentialism and its problems will be a convenient leitmotiv, since it is widely held and employs preferences, and preferences only, on all the five levels. The tour is structured as follows. After a prologue that sketches preferentialism, we will consider the very concept of a preference (section 1). We will then look at the possible roles of preferences in rationality (section 2), and at the triad of roles they might play in morality (section 3). Finally, there is a selected bibliography.

1.

The Concept of Preference

1.1.

General Problems

1.2.

The Taxonomy of Preferences

2.

Preference and Rationality

2.1.

General Problems

2.2.

Rational Decision Theory

3.

Preference and Morality

3.1.

Normative Preferences and the Format of Morality

3.1.1.

Consequentialism

3.1.2.

Deontic Logic

3.2. 3.2.1. 3.2.2. 3.3.

Preference Satisfaction and the Content of Morality Welfare and Preference Satisfaction Distributing Preference Satisfaction Shared Preferences and the Foundations of Morality

Preferences, so the received opinion, are the alpha and omega of practical reasoning: people are rational if they do what they believe will best satisfy their own preferences (more on this in section 2 of this introduction); and people * To find writings about issues raised in the introduction, consult the similarly structured bibliography that follows it. We are grateful to Christopher Abbey, Krister Bykvist, Georg Meggle, and Elijah Millgram for helpful comments. We thank the Deutsche Forschungsgemeinschaft (DFG) for supporting the research project "Was zählt?"; work on this survey was part of the project.

Preferences - an Introduction

XXI

are moral if they do what will best satisfy everybody's preferences (more on that in section 3). Call this type of position preferentialism) Typically, preferentialism is inspired by an enlightened respect for people's autonomy: prejudice aside, it is their own wishes that count. Chacun h sa fa-

ςοη. Popeye, for example, likes eating spinach, and eating spinach conflicts with none of his other wishes. Surely this gives him a reason to do what he believes will provide him with spinach. It would be puzzling for a concept o f rationality to deny this; reason would lack sensitivity, and nobody would want to apply it to their own affairs. Say that Mary, however, does not like spinach, and does not believe it to be conducive to any other end of hers. Then how could there be a reason for her to eat it? It would be puzzling if a concept of rationality required her to eat spinach no matter what she wanted; reason would be dictatorial, and again nobody would want to apply it to their own affairs. Same thing with morals. Given that Popeye wants spinach, the world would, ceteris paribus, be a little better if we gave him some; so that is what we ought to do. And given that Mary does not want spinach, the world would, ceteris paribus, not be better if we gave her any; so we have no obligation to do so. This is how, if we believe preferentialism, both reason and morality correlate with preferences.

1.

The Concept of Preference 1.1.

General Problems

Precisely what it is that preferentialists say, and their opponents deny, will depend on what they mean by "preference", or - for some versions of the theory 1

1

1

«

·

»

(( J

·

5 ,

Because for Wt (A,X)) to be prudentially rational f o r i t is not necessary that Bt(X, G* (A,X j). (See Velleman's autobiography, sect. 6.) And why not Bt(X,G*(Wt(X,A),X))

ν

Bt(X,G*(A,X))

instead? Because the second disjunct would be redundant, viz. implied by the first one. (PROSIM), rational practical goodness in the motivational case, thus turns out to be a special case of subjective goodness. But again, does (PROSIM) really relate to something Velleman himself would like to subscribe to? Minor exegetical problem. Velleman thinks: "Sidgwick's account would identify your future good with what you would desire only after being informed about the motivational changes that you were due to undergo (among other matters)" (sect. 6). However, this follows neither from Sidgwick's own words (as quoted by Velleman) nor from Velleman's paraphrase of them. Motivational changes are (normally) not brought about by our own actions. Hence these changes are not (generally) part of our ARC. Hence our full information as postulated by (POSIM) does not cover these kind of changes.

Motivation and Value: A Reply to Velleman

107

So Vellemans "proposed interpretation" need not "remove" this "problem" (if this is supposed to be "the former problem" referred to in section 6); there is no such problem. Or is "full information" not to be restricted to X's ARC? Then there would be plenty of other problems: full knowledge of all of one's own future decisions, for example? POSICs? We know what "a person's good" means "in its practical sense". It is a POSIM. "Say that a person's practical good is that which ought to be the object of his self-interested motives - what he ought to desire and be inclined to pursue for his own sake". Velleman continues (sect. 8): (W)

"whereas his well-being is that which ought to be the object, more generally, of his self-regarding affect, or self-concern - what he ought to cherish or treasure or be glad about, in the same, self-regarding spirit."

So let's define "in the same, self-regarding spirit", i.e. in direct analogy to (POSIM), what it is to be a POSIC (a Proper Object of Self-interested Caring): (POSIC) A is (at time t) a POSIC for X iff, given that X (at t) is fully informed about X's R and exclusively self-interested, X would (at t) intrinsically and rationally care about A. Note again that Λ being a POSIC does not imply thatX cares for yi; just that X would care i f . . . Questions. Is (POSIC) a correct representation of Vellemans POSICs? What could the repertoire R now consist of? In POSIMs, it had been X's action-repertoire and all the action-consequences. But what could it be in the non-motivational cases of caring? All the As that X could be glad about? Could be glad about, if he knew about them? Any restrictions as to the constitution (of X's person, or species)? Or should the condition of full information be deleted? If X is to be fully informed, then what about the possibility of surprises that Velleman himself says is so important? Well-being and POSICs. Motivational practical goodness being defined via (POSIM), Vellemans statement (W) (quoted above) invites us to (misunderstand it as defining well-being via (POSIC). But the result would be totally inadequate. The practical good is a kind of subjective good, whereas well-being can't be. Well-being is the objective good (of the kind of goodness at stake here). To equate G*(A,X) with vis being a POSIC for X would be just as problematic as to equate Sidgwick's "future good" with a person's future wellbeing. So here is a major difference between practical good and POSIMs on

108

Georg Meggle

the one hand and well-being and POSICs on the other. Is this correct? I hope so. If not, then I have missed the point of Velleman's paper. Caring-rationality? In order to know what it is to be a POSIC, one has to know what the rationality involved, the (rational) ought', is. Velleman's answer is (again) likely to be: What the norms of prudential self-regarding affect would recommend that he self-regardingly care about (or something along these lines). But then, what would they recommend? Here is my general answer, in analogy to (PROSIM). Ct(X, A) is short for "X cares about A at time t": (PROSIC) Q(X,A) is prudentially rational iff Bt(X,

G*(A,X)).

Correct? I don't know. But if so, then we can identify A's being a POSIC with A's being the object of X's rational caring, and so A's being a POSIC is a kind of subjective goodness just as much as practical goodness is. Wanting and caring. Wanting is a special case of caring in Velleman's sense. But we cant just substitute Wt(X,A) for Ct(X,A) and get (PROSIM) out of (PROSIC). What should be done about this? Caring about our desires. Note the following corollary of this reconstruction: (PROSIM) and (PROSIC) together imply that caring about one's own desires is always prudentially rational. I take this to be a virtue of my construction, even if it should fail, which I hope it does not, to be a correct interpretation of Velleman's intentions. Forget about practical goodness? The version of normative internalism Velleman argues for is a version about a relation between value and caring, not about a relation between value and wants (or motivation). Now, why care about the special case at all? Why take so much trouble with respect to practical goodness, POSIMs and PROSIMs, if we are then invited to ignore them?

2.

What Can we Learn from the Stones?

Value = good or bad. For a moment, let us forget about all the above distinctions, make a new start - and use, without further qualifications, the following abbreviations: G{A,X)

for "A is good for X",

109

Motivation and Value: A Reply to Velleman

Ν {A, X) V(A,X)

for "A is negative (bad) for X", and for "A has value for X",

where we might define V(A,X)

as G(A,X)

ν N(A,X),

and N(A,X)

as

G(-iA,X).

is not true for every A and comes to be true for some As and Xs. This is the very question Vellemans paper starts with. But again, his approach is only an indirect one; he starts (in sect. 2) with a quotation from Railton and then asks what we are "entitled to conclude" from it. The quotation runs as follows (numbers inserted by me): Stones and some conclusions.

Now, since V(A,X)

X , the question arises under what conditions V(A,X)

There wouldn't be any values ("good and bad") in "a universe consisting only of stones, for [i] nothing matters to stones, [ii] Introduce some people, and you will have introduced the possibility of value as well". Summary: [iii] "Notions like good and bad have a place [...] only in virtue of facts about [iii.a] what matters, or [iii.b] could matter, to [iii.c] beings for whom it is possible that something matter."

Now, quite a lot is implicit in this position, manifold as it is. Let M(A,X) stand for "A matters to X" (again easily time-relativized by writing Αι instead ο£A) and OA for "It is possible that A". (Let us assume, as Velleman does, that, if M(A,X) is a source of value, then it is so at least ß r X , i.e. for V(A,X) and let us forget about Vellemans "value only" for X . I am not sure whether the problems created by this "only for X" could be overcome by remembering that Vellemans values are to be intrinsic ones. What intrinsic values in relation to mattering and caring are is much less clear than in the case of wanting.) So here are some conclusions from Railton's story: (I)

V(A,X)

M(A,X)

(II)

V(A,X)

0(M(A,X))

(III)

Ο (V(A,X))

^0(M(A,X))

Value only if mattering, cf. (i) and (iii.a). Value only if possible mattering, cf. (iii.b). Possible value iff possible mattering, cf. (ii) and (iii.c).

And since the existence of beings to whom things do matter would suffice not just for the possibility, but also for the very existence of value (which might be already found in Railton's (ii)), we can and, herewith do, strengthen (I) to (IV)

V(A,X)

ο M(A,X)

A has value for X iff A matters

for X .

110

Georg Meggle

Viable conclusions? I have no problems with these conclusions, and could subscribe to all of them. (At least at this level, with no further restrictions on value and mattering. Perhaps things would change if we returned to POSIMs and POSICs. But note that, if M(C(X ,A),A) and if the relevant caring itself is prudentially rational, then the good-version of (IV) would be fulfilled even for A being a PROSIC for X.) My reason for accepting (IV): Just define V (A, X) by means of (IV). To repeat Railtons words: "Good and bad have a place [... ] only in virtue of [... ] what matters". What else could that kind of value (a subjects own good or bad) consist of? If you have any idea, let me know; if not, buy my definition. All the other principles are entailed by that definition. And since principle (IV) is the very core of Railton's statement, I must admit I find his argument' much more convincing than what Velleman tries to add to, or to unfold from, it's premiss.

3.

Derivation

What is Velleman trying to derive from what? The general form of his version of normative internalism is: (ΝΓ)

G(A,X) -> 0(C(X,A))

A is good for X only if it is possible that X cares about A.

(ΝΓ) is, according to Velleman's program, to be derived from the principle (*)

0(A)

Ο (A)

ought' implies 'can'.

But what kind of goodness is G(A,X) in (NI*)? It is (I suppose) not the good "which ought to be the object of his [that is, .Λ" s] self-interested motives", but "that which ought be the object, more generally, of his self-regarding affect, or self-concern" (sect. 8 of Velleman's paper); in other words, the A in G(A,X) is supposed to be a POSIC; to indicate that fact, let us write POSIC(yl,X). Thus, Velleman's relevant normative internalism is this: (NI)

POSIC(AJ>0 -»· 0{C(X,A))

A is for X a proper object of self-interested caring only if it is possible for X to care about A.

Next, we meet Velleman's specification of (*): "Naturally, what ought to be the object of an attitude must be something that can be the object of that attitude" (sect. 8).

Motivation and Value: A Reply to Velleman

111

Truth of(NI). Let us take the special case of wanting; let us take it, that is, that A is a POSIM, and hence a POSIC. Then the motivational specification of (NI) would be: (NI.M)

POSIM(AJ>0

0(W(X,A))

A is for X a proper object of self-interested mattering only if it is possible that X wants that A.

And this principle is true by definition: just check with (POSIM) above. (Mind the full information with respect to ARC; and note that for POSIMs the stronger version, without O, will, in general, not hold; for it to hold, we would need the further premiss that X, in the relevant situation, is prudentially rational, i.e. does want or care about what according to (PROSIM) his believed (or known?) well-being would recommend him to want or to care about.) Now, does the same hold true for the general case (NI)? Have a look at (POSIC), and you will see that precisely that question had not been settled by (POSIM) so far. One of my critical questions was precisely how the relevant repertoire R (as well as full information) is to be understood. Is R (and the postulated information about it) to cover all possible types of caring? Clearly, in order for (NI) to hold, a POSIC must be defined (in full analogy to POSIMs) in the strongest sense of R and full information. But that only makes the questions I raised in the section on POSICs the more pressing. So if (NI) is a kind of Normative Internalism, the relevant possibilities of caring (and wanting) have been put into POSICs (and POSIMs) right from the start. Derivation? Of course, (NI) is a special case of (*), since being a POSIC for X is being something, X (prudentential-rationally) ought to care about according to (PROSIC). But I find this connection, I must admit, a little too trivial to deserve the title of "derivation". 4.

Best Start? And Consequences?

Best start? When looking for a plausible candidate for NI, why should we start with a persons own good? Why not with moral or rational goodness} Because, says Velleman, "[t]he norms of morality and rationality aren't tailored to suit individual tempers; and so we can imagine an agent expressing indifference to morality, and

112

Georg Meggle

perhaps even to rationality, by saying they aren't for him. But a persons own good is indeed tailored to him, and we cannot imagine him saying that his good is not for him" (sect. 1).

Yes, I can. Imagine a person who decides to commit suicide just in order to cause some other person to suffer. She knows (let us assume) that to go on living would be (objectively) best for her, but nevertheless she might react to our reproaching her just by saying "So what?". So, when sticking to a persons good either in the sense of G*(A,X) or of B(X, G*(A,X)), Velleman's assertion ("we cannot imagine him saying that his good is not for him") seems to be false - and again for precisely the reasons that make, as Velleman says quite correctly, the corresponding moral assertion false. (As far as rationality is concerned, I would be glad to find an example of someone who does not just say that he doesn't care about rationality, but who manages to live up to that position (consistently?).) "The norms of morality and rationality aren't tailored to suit individual tempers [...] But a person's own good is indeed tailored to him". Now, it is true that the norms of (morality and of) rationality are norms containing no reference to any particular feature of any particular individual. But neither do the norms that define what counts as a persons own good, if there are any such norms. ((PROSIM) and (PROSIC) at least do not.) But, of course, whatever these norms might turn out to be, it will be trivially true that, whatever will count as good for a particular person according to these norms, will "suit" his or her "tempers". Given her genetic endowment, her education, her living conditions etc., her personal good will comprise certain things; and these things need not be part of the personal good of others. But this kind of relativity to her will be provided by considerations of rationality as well, at least if we think of rationality in its thoroughly subjective sense as defined in rational decision theory. Thus, to conclude, as far as the personal tailoring is concerned, a persons good seems not to be better suited to her "individual tempers" than what it would be rational for her to do. (Note that, though I presented the argument with relation to objective goodness, it would go through just as well for goodness in any subjective sense.) So why not just start with the rational ought' proper? And the corresponding "normative internalism" (There is no rational ought to do f unless f is in the agent's repertoire) would be derivable with equal ease from "'ought' implies 'can'". So Velleman's suggestion, that, when looking for a plausible candidate for NI, we best start with a persons own good, is not conclusive. Not, at any rate, if "best" is to be understood in the strong sense of "better than any other alternative". And surely that was the meaning intended.

Motivation and Value: A Reply to Velleman

113

Consequences? What about the transferability of Vellemans thoughts to metaethics proper, i.e. to the study of "ought" and "good" in the moral sense? This question refers to a problem that Velleman himself (see section 1 of his paper) is well aware of. Now, NI can't be more plausible than it has been made by definition (if it has been or would have been so defined as reconstructed above). But where do we go from there? I have no clue what the transition to metaethics could look like. For, paradigmatically, moral "ought's are about what's good for others.

SVEN

DANIELSSON

Numerical Representations of Value-Orderings: Some Basic Problems

Abstract: Measures of value or preference usually presuppose value or preference relations which are weak orders. Numerical representations of semiorders and of interval orders have to some extent also been considered. It is fairly obvious, however, that value- and preference-orderings often are not, and should not be expected to be, even interval orders. A way of representing partial orders is suggested.

1. In arguments about value, or preference, numbers are used sometimes more and sometimes less seriously. If the value of the alternatives x, y and ζ is indicated by, say, the numbers 8, 4 and 2, the author may have a more or less definite intention concerning how much information these numbers should be taken to convey. And the intention may be more or less ambitious. The figures may be supposed to show, in some well defined way, that χ is twice as good asjy which is twice as good as z. Or the intention may be just vaguely to indicate that χ is more strongly preferred to ^ thanjy to z. Or something there in between. Here I shall stick to problems which occur at a lower level of ambition. I shall assume that the task is just to represent an ordering, designated by "better than" or "preferred to", by an assignment of real numbers. Whether we are dealing with "better than" or with "preferred to" does not make much difference in this context, so I shall use the label "value-ordering" to cover both. Since the task is not to represent differences or ratios of value (though such problems will be touched upon here), such a numerical representation of an ordering is not a measure in any very interesting sense - it tells us nothing that we do not know from the ordering itself1 but it is an indispensable first step towards such a measure. Hence the problems are general problems for the measurement of value and preference. I shall simplify the questions a bit, however, by assuming that the set of alternatives, those things of which some are better than or preferred to some others, is a finite set.

1

From an ordering we usually learn something about differences, of course. If we know, for instance, that * is better than ^ and y better than z, then we can conclude, it seems, that the difference between χ and ζ is greater than that between χ and y.

Numerical Representations of Value-Orderings

115

2. If Ρ is a value-ordering, then the natural idea of a numerical representation of Ρ no doubt is that of a function, say v, which assigns real numbers to the alternatives in question in such a way that: (A)

xPy

if and only if

v(x) > v{y)

The alternative χ is better than the alternative y if and only if the measure assigns a greater number to χ than to_y. The relation "better than" is mirrored by the relation "greater than". That certainly seems very plausible. But the problem is of course that there is such a function if and only if Ρ is a strict weak order, that is, satisfies the following two conditions: (B)

If xPy, then not yPx

(C)

If xPy, then xPz or zPy

while value-orderings often violate condition (C). It is very easy to find examples where χ is better than y while χ neither is better than ζ nor ζ better than y. We represent such a value-ordering by the following figure, where there is an arrow from one alternative to another if and only if the former has the relation Ρ to the latter: χ ζ y Fig. 1 It has been assumed that such situations, at least sometimes, can be taken to show that ζ certainly differs from χ and y, but less than noticeably. It has been assumed that the relation Ρ is mirrored by the relation "greater than" among alternatives in accordance with the formula: (D)

xPy

if and only if

v(x) — v{y) > k

where k is a constant representing the smallest noticeable difference.2 2

Alternatively: xPy if and only if v(x) — v{y) > k. In that case k is rather the greatest nonnoticeable difference.

116

Sven Danielsson

The idea of using just noticeable differences for measurement seems first to have appeared in psychophysics. The object was to study the relations between certain subjective phenomena and their objective correlates, for instance subjective and objective loudness. In that context the subjective difference between two stimuli was assumed to be one J N D (Just Noticeable Difference) when the objective difference could not be made smaller without making the stimuli indistinguishable. If s is subjectively louder than s' and any signal which is objectively just a little louder than s' is - subjectively and as to loudness — indistinguishable from s, then the difference between s and s' is one JND. 3. JND-representations have been much appreciated. The main reason probably is that such representations under fairly simple conditions seem to give very much information. Suppose, for instance, that we have a situation of the structure of: χ

ζ

y

u Fig. 2

Then a JND-representation seems to tell us that the difference between x and y as well as the difference between ζ and u is at least one, but less than two JND, the difference between χ and u more than one but less than two J N D and the differences between χ and ζ, between ζ andjy, and between y and u all greater than zero, but smaller than one JND. This may seem too good to be true. Does the ordering really tell us that much? As was shown by Scott/Suppes (1958) there is a function satisfying (D) if and only if Ρ is a strict semiorder, i.e. satisfies (B) and the conditions (E)

If xPyPz, then xPu or uPz

(F)

If xPy and zPu, then xPu or zPy

but certainly there are lots of semiorders where the JND-interpretation seems not to be plausible at all. Take for instance the following example which was, for somewhat different purposes, suggested by Restle (1961): Someone is comparing the following three alternatives:

Numerical Representations of Value-Orderings

117

x: a trip to California plus an apple, y: a trip to California, z: a trip to Florida. Since he likes apples he prefers χ to y. But he has no definite preference for California over Florida, nor for Florida over California: he is indifferent (in a weak, not necessarily transitive, sense) between y and z. And in the comparison of χ to ζ he does not think that one apple more or less makes any difference - much more important aspects of these alternatives are indefinite - so he is also indifferent between χ and z. The structure of the example is that of figure 1, but the idea that the agent prefers, though less than noticeably, the trip to California plus an apple to a trip to Florida, and prefers, also less than noticeably, the trip to Florida to the trip to California, is not credible. In its original environment "x differs by one J N D from y" has a clear meaning in terms of subjective reactions to variations in the objective dimension. But hardly anyone would suggest that "x is subjectively one J N D better than could be taken to mean that if the objective value-difference between χ and y were made smaller, the subjective value-difference would disappear. In some contexts it may make sense to say that χ is one J N D better than y when no smaller difference than the actual one as to the natural relation on which the value-ordering supervenes would make χ better than y. (If your ranking of cars is based only on speed capacity, you prefer the Maserati to the Ferrari by one J N D if no smaller speed difference would make you prefer the Maserati.) But the explanation of the JND-notion often seems to be possible only in terms of structural conditions on the value-ordering. Restles simple example, however, shows that we cannot assume that the JND-hypothesis is reasonable as soon as the ordering is a strict semiorder, though a function satisfying condition (D) always exists in those cases. The JND-interpretation of semiorders and representations of the type (D) is not the only available one. The JND-terminology suggests that there are differences which are real, though so small that they cannot be directly observed. That is a rather problematic assumption if we are speaking of subjective phenomena, are there really differences in subjective loudness which cannot be heard? The J N D idea suggests that there is a precise reality which can be only imprecisely detected. Another idea is to locate the impreciseness in the object rather than in the subject. In the present context we would then say that the value of the alternative λ- is not precisely v(x) but v(x) ± | . The impreciseness of the measure is k, so to speak, the value of jc is determined within an interval of length k, but this is now not taken to mean that there is in reality a precise value v(x) which cannot be more precisely identified, but that the real value is

118

Sven Danielsson

indefinite within these limits. 3 This often seems far more reasonable than to assume a definite value and an imperfect discrimination power. 4. It may be doubted that the objective' impreciseness hypothesis in this version is more plausible than the JND-hypothesis in the trip and apple case. We shall return to that. Even if it were, it is obvious that the condition (D) is not the general mirroring formula which we are looking for. There are plenty of cases of the structure of: χ y

u

ζ Fig. 3 where the only value-relation is the relation Ρ as here marked by the arrows and condition (E) hence is violated. To take care of such cases Fishburn (1973) suggested another mirroring formula. Instead of a function v, assigning real numbers to the alternatives, he suggested that we use a function assigning intervals of real numbers to the alternatives according to the formula: (G)

/(x) > f ( y ) if and only if

xPy

where f ( x ) and f (y) are intervals of real numbers and the relation > holds between f{x) and f (y) if and only if it holds between every member of f ( x ) and every member of f ( y ) . (If f{x) is a unit set we identify it with its sole member.) Fishburn proved that there is an f satisfying (G) if and only if Ρ is a strict interval order,4 i.e. satisfies (B) and (F). The traditional mirroring condition for semiorders then turns out to be a special case of (G), one where f ( x ) has the same length for every alternative x. 3

4

Is the phenomenon we are studying the vagueness of preference and value? At most a part of it. Notice that the degree of impreciseness is very precisely determined here. With vagueness it is different, there is no clear cut-off point between the clear and the unclear cases of baldness. The label "interval order" is not a very happy one. The similarity to "interval representation" and "interval scale" may give the false impression of a connection. Being accepted in Foundations of Measurement (Suppes et al. 1989), however, "interval order" is here to stay.

Numerical Representations of Value-Orderings

119

And the original mirroring condition for weak orders is the special case when the length is always zero. In terms of imprecise value Fishburn's model accepts that different alternatives can have values which are imprecise to a different degree. So far the Fishburn model means an indubitable progress. Unfortunately it seems to be very easy to find examples where the relation Ρ is not an interval order either: We just extend the trip and apple case with a fourth alternative u\ A trip to Florida plus an apple. The same kind of reasoning as in the original case will then produce the ordering: λ:

u

y

ζ Fig. 4

which clearly does not satisfy (F) and hence is no interval order and hence cannot be mirrored according to the formula (G). And there are many obvious ethical examples. There is a common view on the value of human life, for instance, according to which it is always worse that one person more is killed, but not necessarily worse that a greater number is killed. Given that view the alternatives of figure 4 could be:

x: y: u: z:

Only Peter is killed, Peter and Paul are killed, Only Tim is killed, Tim and Tom are killed.

5. There may be several ways out. One is to switch to the following mirroring formula: (H)

xPy

if and only if

i(xy) > 0

where i is a function assigning intervals, not to individuals but to ordered couples of individuals: χ is better than y if and only if the imprecise difference between χ and y is positive. By the mirroring formula (H) we have reached the end of the road in a certain sense: We can be sure that we shall not have to give up the formula (H) for anything like the arguments that have hitherto forced us to retreat. For no matter which relation Ρ is, there is a function satisfying (H). (We can

120

Sven Danielsson

always choose the characteristic function of Ρ as i, that is assume that i(xy) is 1 if xPy, otherwise 0.) Given (H), different logical conditions on Ρ will correspond to different conditions on /'. Asymmetry, if xPy then not yPx, will correspond to (I)

i{yx) =

-i(xy)

where the subtraction operation — for intervals is defined by — (a,b) = Transitivity will correspond to: (J)

i(xz)

C

i(xy) + i{yz)

where -f for intervals is defined by (a, b) + (c, d) = (a + c, b + d). It can be shown that there is a function i, satisfying (Η), (I) and (J) if and only if Ρ is transitive and asymmetric, that is, if and only if Ρ is a strict partial order? 6. This kind of representations differ from the traditional ones by there being nothing representing directly the value of χ - not even the imprecise value of x, as with the Fishburn functions. If Ρ is at least an interval order, then we can always define a function which represents the value of x: we choose a suitable alternative, say o, as a reference object and then we define the one-place function f by the condition f(x) — i(xo). This function will be a Fishburn function, it will satisfy (G), but it can be shown that this is possible only if Ρ is at least an interval order. Otherwise there will be no suitable reference object o. Hence we seem to be dealing with value-differences which are not differences between values, and that may appear rather strange. Nevertheless this move often makes possible a plausible description of the situation. In the trip and apple case, for instance, it seems reasonable to argue that the subject can recognise clearly a very small value-difference between a trip to Florida plus an apple and just a trip to Florida - the value-difference equals the value of the apple (granted no organic units including the apple). But comparing a trip to Florida to one to California it is not possible to make discriminations which are that fine. It is the differences that appear more or less precise, and this shifting preciseness can not be explained by distributing it as a shifting preciseness of the values of the compared alternatives. 5

Alternatively we may assume a function q assigning intervals representing value-ratios to pairs of alternatives, according to the formula (H') xPy iff q{xy) > 1. The conditions corresponding to a strict partial order then are (Γ) q{xy) — ^ (yx) and (J') q(xz) C q{xy) χ q{yz) with multiplication for intervals defined in the obvious way.

Numerical Representations of Value-Orderings

121

We may be inclined to believe that it does not matter whether we start from a notion of value and then define "better than" in terms of that or start from a relation "better than" and define value in terms of that — the value of χ taken to be the value difference from χ to some chosen reference point, but if the value-ordering is not an interval order we cannot find any alternative which can serve as a common reference point for all the other alternatives: no alternative χ such that for every other y and ζ the value-comparison o f y to ζ can be reduced to the comparisons o f y and ζ to x. 7. Expected utility theory may be a field of application. Consider the following lotteries: A: B: C: D:

£100 with probability .500, otherwise 0. £ 99 with probability .500, otherwise 0. £101 with probability .494, otherwise 0. £101 with probability .491, otherwise 0.

Provided more money is preferable to less, the ordinary von-NeumannMorgenstern theory of expected utility implies that A should be preferred to B, C to D and either A to D or C to B. This follows from three simple assumptions: (a) If you replace a prize in a lottery by a preferable one, the resulting lottery is preferable, (b) If you increase the probability of winning a preferable prize the result is a preferable lottery, (c) The value-ordering of the lotteries is an interval order. To some of us, however, it appears perfectly rational to prefer A to Β and C to D but still be indifferent between A and D and between C and B, and a defense of that position could run something like this: When we are comparing A to Β the principle (a) appears very reasonable and so does the principle (b) when we compare C to D. If the difference is only a difference in the preferability of the prizes we can make very fine discriminations. And if the difference is only a difference in the probability we can make very fine discriminations. But when we change the value of the prizes in one direction, and the probability in the other, so to speak - as in the comparison of A to D and C to Β - then the choice is much more difficult, then the difference is much more imprecise. The fact that we argue that our preferences are sensitive even to a change of .003 in the probability of winning £101, or to the change from £100 to £99 in a fifty-fifty lottery, does not commit us to hold that there is, or would be if we were perfectly rational, some change in the probability which is determinable within an interval of .003 or anything like that, which would correspond to a given change in the amount of money. The crucial principle is neither (a) nor (b) but (c). Even if we consider a set of lotteries which is generated from a set

122

Sven Danielsson

of prizes on which we have preferences which are completely precise in some reasonable sense, we may well accept both (a) and (b) but still deny that our preferences among the lotteries have to form an interval order. In expected utility theory we usually start with weak orders. Assuming just partial orders would not fundamentally thwart the possibility of measuring expected utility essentially along the usual lines. But we would then rather be measuring imprecise utility differences.

References FISHBURN

(1973). Peter Fishbum: "Interval Representations for Interval Orders and Semiorders", Journal of Mathematical Psychology 10 (1973). R E S T L E (1961). Frank Restle: Psychology of Judgment and Choice, New York 1961. S C O T T / S U P P E S (1958). Dana Scott and Patrick Suppes: "Foundational Aspects of Theories of Measurement", Journal of Symbolic Logic 23 (1958). S U P P E S ET AL. (1989). Patrick Suppes, David H. Krantz, R. Duncan Luce and Amos Tversky: Foundations of Measurement, vol. II, San Diego 1989.

ULRICH NORTMANN

Interval Orders Defended: A Reply to Danielsson*

1. The problems Danielsson deals with in his paper center around the question: Can we homomorphically embed a structure (Χ, Ρ) (with Ρ being interpreted as a two-place partial preference-order or value-order on a set X of alternatives x) into a numerical structure in such a way that properties of the latter structure would not force properties upon (X, P) which we think valueorders need not have? (Note that homomorphic embeddability is not understood here as implying the existence of an embedding function which is an injection.) It is to be expected that metrization will tend to give us trouble with respect to cases of indifference (i.e. cases in which a subject of preferences prefers neither χ toy nor to x). An example of an embedding of the type in question is given with any function ν: X —> R plus the relation > on real numbers, when Vx,y £ X :

xPy if and only if v(x) > v{y).

This is the first type of embedding Danielsson discusses (see his sect. 2, formula (A)). Obviously, every such embedding forces upon (X,P) the property labeled (C) by Danielsson: Vx,jy, ζ € X :

if xPy, then xPz or zPy.

Danielsson is right in saying that, as a rule, value-orderings do not comply with (C). Compare Fehige's fashion of displaying the Mere Addition Paradox (Fehige 1998, sect. 6). Fehige's Β is taken to be uncontroversially better than his A + . Yet Β > A seems to be false (from a Narvesonian point of view one might even argue that A > B), and so seems A > A + (since in this special case, A + > A). We must therefore clearly refer to a wider frame than the one given by (R, > ) in order to have a chance of finding embeddings of the desired type. Danielsson continues by considering embeddings into (Pow(R),5), with S being any appropriate relation on the power set Pow(R). The aforementioned * I would like to thank R. Cuplinskas and U . Schäfer for their assistance in correcting my English.

124

Ulrich Nortmann

special case is included in this wider frame by identifying a real number r with the singleton set {r}. Danielssons thesis reads: Even the most plausible' candidates for embeddings of (X, P) into (Pow(R),S) require properties of Ρ which value-orderings need not have, namely (i) they require (X,P) to be an interval order, but (ii), as a rule, value-orderings are not interval orders. Much of what Danielsson says is meant to establish (ii). I want to argue that the case for (ii) is not as strong as it appears in Danielssons paper. As a consequence, I am not as sceptical as Danielsson is about the chances of numerically representing preference-orderings. 2. Before turning to a discussion of (ii), I wish to make some remarks concerning the technical side of Danielssons arguments. Danielsson assumes that the set of alternatives^ is finite (sect. 1). However, he does not give a detailed account of the proofs of the theorems employed; as a consequence, the reader cannot see where the assumption is needed. Some hints would help readers who are interested in the underlying mathematics. As far as I am acquainted with the field, one of the points in which finiteness is needed is the Scott/ Suppes-result mentioned in the beginning of sect. 3. Scott and Suppes prove that "the class of finite semiorders is a theory of measurement relative to the numerical relational system (Scott/Suppes 1958, p. 118). Here, Re stands for R , and » is the relation on R determined by Vx,y :

χ » j y if and only i f * > y + 1.

Saying that the class of finite semiorders is a theory of measurement relative to (R, >>) is essentially the same as saying that finite semiorders are embeddable in (R, 3>). What is more important with respect to Danielssons assertion (i), however, (and what of course is far more easy to see) is the opposite direction: if ( X , P ) is embeddable in ( R , » ) , then P is a semiorder. Let us recall the definition of (strict) semiorders. To begin with, Ρ is called a strict interval order on X if and only if Ρ is asymmetric and satisfies Danielssons (F)

Vx,j/, z , « G X :

if xPy and zPu, then xPu or zPy.

A semiorder is an interval order additionally satisfying (E)

z,ueX

:

if xPy and yPz,

then

xPu

or

uPz.

Interval Orders Defended: A Reply to Danielsson

125

Here, intervals are involved as follows: if ν : X —> R exists so that Vx,y G X : xPy if and only if ν (χ) >k (i.e. v(x) » v{y) with k playing the role 1 plays in the formulation of Scott's and Suppes's embedding result), there also exists a function F , assigning real intervals to members o£X, so that Vx,y G X : xPy if and only if Fix) > F{y) (i.e. inf(^(*))>sup(^))). We can define F(x) as the interval

As Danielsson reports in sect. 4, if ( X , P ) is embeddable in the class of real intervals by means of the above relation, Ρ is an interval order in the sense of the definition. Finally I want to know how seriously Danielsson's remarks on correspondence in his sect. 5 are to be taken. Asymmetry of P, for instance, certainly does not correspond to (I)

i(y,x) =

-i(x,y)

in the sense that asymmetry together with Danielsson's (H) entails (I). This is obvious from Danielsson's own remark that characteristic functions can always be chosen as functions i representing value-orderings Ρ in the sense of (H). In that case, if i(x,y) = {1}, i(y,x) will be {0} and not { - 1 } . 3. So much for technicalities. Let us now have a closer look at Danielsson's arguments connected with the California-Florida-example. This example and an extended version thereof are used in two places of his paper (cf. sects. 3 and 4). In the first case, the example is introduced in support of the assertion: Even if value-orderings are semiorders, they are not in general open to a certain interpretation, i.e. the JND-interpretation (the "Just Noticeable Difference'-interpretation) of semiorders. According to the JND-interpretation, circumstances of the type ->(pcPy) & -i(yPx) (i.e. cases of indifference of a subject of preferences with respect to χ andy) have to be understood as cases where there is possibly a value-difference between χ and y which in any case is so small that it does not exceed a certain threshold (this threshold possibly being fixed for all cases). Even if the example

126

Ulrich Nortmann

does suffice to establish that assertion, we need not worry if our concern is just the metrization ofvalue-orderings, since the JND-interpretation ofsemiorders need not be the appropriate interpretation. If for instance on the one hand we have: not v{x) — v{y) > k, and: not v{y) — v(x) > k, and on the other hand we have v(x) — v(y) = λ with 0 < λ < k, we are by no means compelled to describe the situation as a case where χ is preferred to y even though the preference is minute. Instead, we may confine ourselves to saying that χ is simply not preferred to y. More important is the extended version of the example, since it will be introduced by Danielsson to show that there are preference-orderings which are not semiorders and not even interval orders. The first version of the California-Florida-example runs as follows. Danielsson introduces a supposedly realistic setting in which a person is indifferent concerning the choice between a trip to California (y) and a trip to Florida (ζ); moreover she is indifferent with respect to a trip to CA plus an apple (x), or a trip to FL, whereas she prefers a trip to CA plus an apple to a mere trip to CA. On a JND-reading, our persons indifference with respect to ^ and ζ cannot rest on an assignment ν of. values such that v{y) > v(z). Otherwise we would have to conclude that v(x) > v(z) + k (from the assumption that χ is preferred to y, that is, v(x) > v[y) + k) which means that χ is preferred to z. As a consequence, v{y) < v(z): FL is preferred to CA, but merely to such a small degree that the relevant threshold (k) is not exceeded. As a further consequence, v(x) > v{z), and our person's indifference with respect to χ and ζ has to be described as a case in which χ is preferred to z, but to such a small degree that there is no substantial difference between a: and ζ. Here, Danielsson says that this way of interpreting the situation cannot be accepted. The reason seems to be his opinion that an extra apple cannot have enough weight to reverse a former (slight) preference for FL to CA into a (slight) preference for CA (plus apple) to FL. But even if we were ready to accept this talk of small value-differences not exceeding a threshold, where it would actually seem more adequate simply to deny value-differences (as is suggested by Danielsson himself at the end of sect. 3), we could still argue: if an extra apple is reason enough to make us clearly prefer χ to y, why should an extra apple not have enough weight to change a slight preference for ζ to y into a slight preference for χ to z? I think Danielsson is right in asserting that, in general, preferenceorderings are not semiorders, for I believe that in fact cases as depicted by Danielssons figure 3 are possible:

127

Interval Orders Defended: A Reply to Danielsson

χ

y

u

!

ζ

Think for instance of two goods G\ and Gi, and of χ as the state in which 7 units of G\ and 5 units of Gj are available: G\\G2

7|5 ' Let y, z, and u be represented by G\\G2

G\\Gj

"6Ϊ3"'

TIT'

, and

G\\Gi

T|T'

. respectlvely

·

Assuming that Gj means especially much to me, the decrease in G\ by 3 units, when comparing χ with u, might just be compensated (or even be slightly overcompensated) through the increase in G j by one unit, so that, as a result, I am indifferent with respect to χ and u. As with u and z, u will at first glance appear preferable; for one missing unit of the highly appreciated G j seems not to be made up for by one extra (7i-unit. ζ however supplies more pairs of goods of the type G\ and of the type Gj. Maybe I am particularly interested in having a brandy after smoking my cigar - without being able to give an exact account of how much pairing means to me in comparison with a decrease or increase in G\ or Gj taken by itself. Such a kind of desire might well make me indifferent with respect to ζ and u. 4. Yet I do not think Danielsson has definitely shown that there are preference-orderings which are not interval orders. Before I turn to the extended version of the California-Florida-example and its application to interval orders, let me make some general remarks on interval orders and Fishburn-functions. Among the cases Danielsson discusses as candidates for embeddings of preferential structures in numerical structures, the most promising involves functions F : X —> Pow(R) such that for χ in X, F(x) is an interval (say, a closed interval) in R and, furthermore, for χ and y in X we have: xPy iff F(x) > F(y). Here > is understood to hold for intervals F(x) and F(y) iff F(x) is fully located to the right of F{y) on the real number

128

Ulrich Nortmann

line. Such functions F are called Fishburn-functions by Danielsson. The idea of representing preference-orderings by Fishburn-functions seems very attractive to me when compared with the standard approach to metrization of preferential structures. If χ is an alternative in X , then there will be a smaller or larger number of ways of realizing x, or of possible states x' in which χ is realized. We can think of χ as the set of all such states x'. In the standard approach it is assumed that there is a subject a of preferences, assigning reals u{x') as definite degrees of utility to elements x' of a given class of states of affairs, and, moreover, assigning probabilities p{x') to those elements. The value of χ is then determined according to the formula (Σ)

Σ

«(*')

·ρ{χ'/χ)

x'ex

It can hardly be denied, however, that any subject we could think of would be unable to assign a definite distribution of probabilities to the states x'. Therefore, we should give way to the possibility that the probabilities p(x') vary within certain boundaries (in mutual dependence of course). The factors w(x') can cause further indefiniteness. Our subject will assess the value of x' in several respects. u(x') will then emerge as a weighted sum in which the summands result from multiplying the degrees of value, in certain respects, with the weights which our subject attaches to those respects. To the same extent that there is no definite attachment of weights, there will also be a variation in the sum. Accordingly, there will be a variation in Σ^Έ* w(x')^>(x'), and the degree of variation will itself vary for different x. This situation is modeled by the concept of a Fishburn-function insofar as such functions assign not a real number, but a real interval to χ as the degree of utility of x. The J N D representation of preference-orderings stands for the less realistic special case of a Fishburn-type representation in which all intervals have one and the same length. Assuming, for instance, that x'Q andy'Q are ways of realizing χ and y, respectively, which our agent a would especially enjoy, she will prefer χ toy only if the value of χ according to (Σ) exceeds the value of y, even when the summation is based, on the one hand, on a probability-distribution being most unfavorable with respect to XQ, and, on the other hand, on a distribution being most favorable with respect tojj/Q. This idea is mirrored in Fishburn's embedding method by the condition: xPy holds if and only if the infimum of the interval assigned to χ is still larger than the supremum of the interval assigned to y.

Interval Orders Defended: A Reply to Danielsson

129

5. So far, Fishburn's representation seems to be what we are looking for. But now Danielsson objects: Fishburn-representability requires ( X , P ) to be an interval order (correct), and preference-orderings are not generally interval orders. In support of the latter claim, Danielsson enriches the California-Florida example by a further alternative u, namely a trip to FL plus an apple, and says (sect. 4) that the "same kind of reasoning as in the original case will then produce the [following] ordering": χ

u

This is slightly misleading, for part of that reasoning was implicitly carried out within the framework of a (constant-length-) interval representation of the case in point, and only questioned the JND-interpretation of indifference. Of course, arguing like that would yield v{u) > ν (ζ) + k> v(y) + k, and therefore u

y

By referring to "the same kind of reasoning as in the original case" Danielsson apparently considers the same assessment to be valid for u and y as he made concerning the former case, namely that "in the comparison of χ to ζ he [that is, our subject of preferences] does not think that one apple more or less makes any difference - much more important aspects of these alternatives are indefinite" (sect. 3). This certainly sounds plausible. In order to have a less bizarre example, I would like to think of more realistic alternatives, such as a trip to California or Florida, by bus or by train, and of a subject a who enjoys going by bus, and is far less enthusiastic about trains. Now, recall formula (Σ) and the ensuing remarks, a may well hope for realizations z' of the Florida trip (by train) including alligator-watching (lacking, of course, a precise estimate for the probability of such a course of events), and may be undecided whether or not to attach considerable weight to this faunal aspect of the trip. As a consequence, there will be realizations of ζ preferred to certain realizations of χ (even though these involve going by bus). Among the latter there might, for instance, be courses of events lacking any equivalent to the faunal highlights of the former (Danielsson (sect. 3): "much more important aspects of these alternatives [such as, for instance, faunal highlights

130

Ulrich Nortmann

or not] are indefinite"). By the same token, we can think of Λ as having no definite preference of u to y (nor o f y to u). So far, Danielsson seems to have established his point. Nevertheless, doubts remain about the conclusiveness of the example. For the assumption that a prefers χ to y (and u to z) is not as harmless as it appears at first glance - if the same sort of considerations as above are applied. There is no problem in thinking of a as preferring bus per se to train per se in a prima-facie-sense ("he likes apples"). But can χ simply be regarded as a ceteris-paribus-variant of y differing merely by the means of transport involved? Replacing train by bus might have drastic and even undesirable effects when combined with certain courses of events however improbable they may be - one cannot simply replace the good (train) by the better (bus) and leave the rest unchanged. As for the Peter-Paul-example, a defender of interval orders can argue as follows. If a does not prefer χ to ζ (even though in ζ a greater number of people are killed) the only possible reason can be that α has an especially close relationship with Peter, but not with Tim or Tom; and, as a consequence, α could even tolerate Tims and Tom's deaths if only Peter were to survive. Such being our subjects inclinations, it seems clear that she will prefer u to^. Danielssons lotteries-example (sect. 7) seems to be meant as a further case for the claim that (rational) preference-orderings need not be interval orders. I think the case need not be accepted either. Our concern is not with propositional attitudes and with the truth-conditions of sentences of the type "I prefer alternative χ to y" - where there are certain linguistic characterizations of alternatives replacing the variables. As I understand the discussion, preferential structures (X,P) are interpreted in an objective or de-re-sense. As a consequence, arguments should not be founded on cases construed in such a way that it would essentially depend on linguistic matters which alternative an agent will choose. Danielssons lotteries A and D could also be described as alternatives with expected utilities 50 and 49.591, respectively. Assuming that characterization, A will be preferred to D (provided that the concept of expected utility is properly understood), and the case will no longer represent an ordering which is not an interval order.

References (1958). Dana Scott and Patrick Suppes: "Foundational Aspects of Theories of Measurement", Journal of Symbolic Logic 23 (1958).

SCOTT/SUPPES

Interval Orders Defended: A Reply to Danielsson FEHIGE

(1998). Christoph Fehige: volume.

"A

Pareto Principle for Possible People

Part II Preference and Metaethics

LENNART A Q V I S T

Prima Facie Obligations in Deontic Logic: A Chisholmian Analysis Based on Normative Preference Structures Abstract: The paper argues for an analysis of the W. D. Ross notion of prima facie obligation which results from adding a certain Chisholm-style definition to the system G of Dyadic Deontic Logic, supplemented with so-called propositional quantifiers. In the semantics for that system a von-Kutschera-inspired conception of normative preference structures turns out to be of vital importance. 1. 2. 3. 4. 5. 6. 7.

Introduction Some Controversial Principles of Monadic Deontic Logic A Theory of Prima Facie Obligations: Some Conditions of Adequacy Choice of Framework: The System G of Dyadic Deontic Logic with Propositional Quantifiers The Operator Ο ' Λ A Chisholmian Definition Added to the System (7„ Gn with Def O f f Meets Adequacy Condition (II) Appendix: Syntax, Semantics and Proof Theory of the System GK References

1.

Introduction

The well-known concept of prima facie duty was introduced into theoretical ethics in this famous passage from Ross (1930), p. 19 f.: "I suggest'prima facie duty' or 'conditional duty' as a brief way of referring to the characteristic (quite distinct from that of being a duty proper) which an act has, in virtue of being of a certain kind (e.g. the keeping of a promise), 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. Whether an act is a duty proper or actual duty depends on all the morally significant kinds it is an instance of."

In the following passage from Broad (1949), p. 552, Ross's contrast, prima facie duty v. duty proper, is replaced by, or explicated as one between compon-

ent obligation and resultant obligation·. "We might compare the claims which arise from various right-tending and wrong-tending characteristics to forces of various magnitudes and directions acting on a body at the same time. And we might compare what I will call

136

Lennart Aqvist

the resultantly right course of action to the course which a body would pursue under the joint action of such forces. Looking at the situation from the point of view of the agent, we can say that each right-tending and wrong-tending characteristic imposes on him a component obligation of a certain degree of urgency; and that his resultant obligation is to make the best compromise that he can between his various component obligations."

In this paper I shall be concerned with the problem of how to incorporate the notion of prima facie oughtness (duty, obligation, right, wrong) into deontic logic, i.e. the discipline dealing in a systematic way with the formal properties of various normative concepts as well as the validity/invalidity of arguments and inferences involving those concepts. I try to handle the problem as follows. After considering (in sect. 2) a number of controversial principles allegedly governing monadic ("categorical", "unconditional", "absolute") normative concepts, I formulate (sect. 3) two conditions of adequacy that the analysis of prima facie oughtness is expected to satisfy, and then (in sect. 4) I choose a logical framework that I believe to be appropriate for those analytical purposes. Sections 5 and 6 will show that our desiderata are met by a theory of prima facie oughtness (obligation) which results from adding a certain Chisholmstyle definition to a certain logical system for dyadic ("conditional", "relative") normative notions, supplemented with so-called propositional quantifiers. Some details of that combined system, called Gn below, are given in the appendix. Two highly important aspects of the resulting theory must now be commented upon, relating respectively (i) to our Chisholm-inspired definition, and (ii) to the role and import of the notion of normative preference that figures in the semantics of our dyadic deontic logic Gn. (i) In two remarkable papers, Chisholm (1964) and (1974), the concept of prima facie duty is connected to that of requirement in inference patterns like the following: (A)

p

(e.g. the fact that I have promised one man to go) occurs.

p

requires that A (= I go). I have a prima facie duty to go.

(B)

q

(= the fact that I have promised another man not to go) occurs

q

requires that Β (= I don't go) I have a prima facie duty not to go

Prima Facie Obligations in Deontic Logic

137

Moreover, Chisholm asserts not only that both arguments are valid logically, but also that one may consistently affirm the premisses and conclusions of both arguments jointly, i.e. without running into contradictions. (As a matter of fact, Chisholm (1974) considers a third argument as well, which involves the highly interesting notion of a requirement being overridden·, I regret being unable in this paper to deal with this central notion and Chisholm's suggestive analysis of it.) Chisholm's own definition of prima facie duty ("oughtness") is in terms of a single ethical primitive '/> requires q (in symbols: pRq). In (1974) he makes an interesting attempt to characterize this basic notion axiomatically, but no semantics (in the modern model-theoretical sense) is given for it. In section 6 below, I argue that Chisholm's pRq'can be interpreted either as Op q or as O^ q (i.e. the fundamental connectives for conditional obligation in the systems G and DFL of dyadic deontic logic, respectively; see Aqvist (1986) and the appendix infra). And it turns out that almost every one of the axiomatic principles proposed by him for his concept of requirement is valid under one or the other (or both) of these interpretations. Upshot: I would like to emphasize here the importance of Chisholm (1964) and (1974) for the technical and philosophical development of dyadic deontic logic as such, as well as for the analysis of the tricky notion of prima facie oughtness. (ii) In our appendix below, we present a model-theoretical semantics for the system GK, the central notion of which is that of a Gn-model in the sense of a structure M=(W,P,>,V) with W a set of possible worlds, Ρ a set of facts or states of affairs, V an assignment to prepositional variables, and, most importantly, where > is a binary relation of (weak) preference among the possible worlds in W. We shall think of > as a relation of normative preference, and of !M as a whole as a normative preference structure, more or less in a sense suggested by von Kutschera in a number of writings (notably von Kutschera 1973, ch. 5, 1974, 1976, ch. 5, and 1982, ch. 1). Let > be the 'strict' counterpart of >. According to the interpretation I have in mind, then, a judgment of the form x> y

("x is strictly better than / ' )

asserts that χ ought (is) to be strictly preferred toy, and should be carefully distinguished from judgments expressing what von Kutschera (1973), sect. 5.1.

138

Lennart Aqvist

calls subjective preferences, which are rather of the form x>ay

("* is stridy better thany according to a")

and assert that, as a matter of fact, the (group of) person(s) denoted by "a" strictly prefers χ to y. As I have tried to show in considerable detail in the appendix to my book Aqvist (1987) and in Aqvist (1986), normative preference structures in essentially the sense just indicated have contributed significantly to recent developments of dyadic deontic logic, and, in particular, to an improved understanding of its semantics. An especially nice application of these normative preference structures is illustrated by the following result, which goes back to Danielsson (1968) and was further elaborated by several writers, including von Kutschera (1974) and myself in the works mentioned above. Against the background of the latter, we can formulate the result as follows. Consider the system DFL as described in § 3 of Aqvist (1986). Then there exists a system PR having primitive logical connectives for such comparative notions as 'better than', at least as good as' and 'just as good as' in its object language, and which is deductively equivalent to DFL under appropriate definitions. Moreover, DFL and PR have the same model-theoretical semantics in terms of normative preference structures serving as models for both systems simultaneously. To illustrate a bit, the Z)/Z-definitions of the PR-connectives > and > are:

and the PR-definitions of the DZ-Z-connectives OdA and PdA are: Of A

(B&cA) > (B & -A)

p f A =df (B&cA) > (B & -vl) A corollary of this result together with that of section 5 infra is that our Chisholmian definition of the operator Opf ("prima facie it ought to be that") could equivalently be formulated as follows in an appropriate framework: Off A =df 3p{p & ( p &cA) >(p&c -Λ)) where prima facie oughtness is explicitly defined in terms of an connective for 'better than'.

object-language

Prima Facie Obligations in Deontic Logic

139

In spite of the technical success of the above intertranslatability result, both Danielsson and von Kutschera seem to worry about intuitive counterexamples to the suggested definitions (see e.g. Danielsson 1968, sect. 3.4, von Kutschera 1974, sect. 6, and 1976, sect. 5.3); however, von Kutschera concludes that the fruitfulness of the discussed notions of preference and obligation is a matter which is difficult to assess in a satisfactory way. I agree. Let me close this somewhat lengthy introduction by pointing out certain respects in which normative preference structures could be made a more flexible, sophisticated and efficient tool in realistic applications of deontic logic and preference theory. Let me do so by way of a quotation from E. J. Lemmon (1965, towards the end of sect. 2): "Perhaps the unease [concerning deontic logic] can be crystallized in this form: so far there have not been enough parameters to deontic operators, and the symbol 'CM' leaves too many questions unanswered - 'it is obligatory that but on whom, to whom is this obligation due and when, how was it incurred and when? For a useful deontic logic, we shall need, I suspect, at least variables for people, variables for individual acts, and variables for times, together with appropriate quantifiers." If we base deontic logic on normative preference structures in the way illustrated in this paper, then, I think, Lemmon's remarks about parameters (or 'indices') to deontic operators apply with equal force, mutatis mutandis, to parameters to preference relations like > in normative preference structures. We should like to know according to whom, according to which authority, and when, it is the case that χ ought to be preferred to y. In short, in realistic applications, the source and the time of coming into force of normative preferences should somehow be explicitly indicated. In my work together with Jaap Hoepelman (Aqvist/Hoepelman 1981) we tried to realize at least part of the above Lemmon program. The resulting framework of deontic tense logic also seems to be a good one for handling and bringing in probability considerations - which are clearly also needed in any realistic applications of preference-based deontic logic. In this manner, we could establish better contact between the fields of deontic logic and decision theory.

2.

Some Controversial Principles of Monadic Deontic Logic

In (1974) and (1980), Brian F. Chellas argues that one should depart from familiar systems of monadic deontic logic by rejecting the following usually

140

Lennart Aqvist

accepted principles: OD*

-.(CW&CK4)

OK

(OA & OB) -). 0(A & 5)

A similar claim is made by Schotch/Jennings (1981); they pertinently characterize OD* as a "no-real-conflicts" principle or a "no-irresolvable-moral-dilemmas" principle, and O K as an aggregation principle. O n the other hand, both Chellas and Schotch/Jennings are concerned about protecting, or saving, the following axiom and rule of inference, respectively: Con(sistency) Mon(otonicity)

-O_L

OA

OB

They also claim that the consistency principle Con somehow expresses (or is a version of) the Kantian dictum that "ought" implies "can", although one would expect the straightforward formalization of this sollen-können principle to be rather as follows (where u M n expresses some appropriate notion of possibility): Kant

OA

MA

Now, Chellas (1980), sect. 6.5, observes that in any normal system for Ο the deontic theses Con and OD* are equivalent in the sense that their biconditional

is provable as a theorem. (By a normal system for Ο we mean, as usual, one that contains all theses of Kripke's Κ with • = O; in particular, every instance of the above schema O K is provable in every normal system for Ο.) Again, since Chellas and Schotch/Jennings are anxious to retain the sollenkönnen principle Con while at the same time just as anxious to reject the "noirresolvable-moral-dilemmas" principle OD*, they have to find some weaker logical basis for monadic deontic logic than Kripke's K . In particular, such a basis must not contain O K as a theorem schema, since O K is responsible for the left-to-right direction of the above biconditional in the following way: sup-

Prima Facie Obligations in Deontic Logic

141

pose that the logical basis contains the schema OK and assume the following: 1.

->0_L

hypothesis (= Con)

2.

OA & O-A

hypothesis

(for some wff A)

(for reductio ad absurdum)

3. 4.

0(A & -A) Ο _L

from 2 by OK from 3 by propositional logic and the rule Mon

5.

-Ό-L

1 reiterated

6.

->(OA & O-A) (for each wff A)

from the deduction 2 - 5 by negation introduction

7.

1 —• 6

from the deduction 1—6 by implication introduction

Then:

where 7 = the left-to-right direction, which is objectionable according to our colleagues. One reason why I wont here explore the details of Chellas's and Schotch/ Jennings's non-Kripkean approaches to deontic logic has to do with the topic of the present volume: however interesting their respective approaches may be, they make no attempt whatsoever to relate the subject of deontic logic to preference theory (at least not explicitly), nor does their work contain any serious attempt to elucidate the interesting notion of prima facie obligation (at least not explicitly). Well, I say "at least not explicitly", for one might certainly argue as follows: the interesting criticisms that have been levelled against the no-conflicts principle OD* and the aggregation principle OK are indeed perfectly valid, if the monadic O-operator is interpreted as expressing the notion of prima facie obligation, whereas they may well cease to be valid if Ο is taken to express some notion of ought', which is of a more 'absolute' kind, like Searle's concept of ought, all things considered' (1978) or, for that matter, the 'toti-resultant' concept of 'ought (to be)' defined already by Chisholm (1964). Remark. In the paper of his we have just mentioned, Searle points out (on p. 85) that the structure of conflict situations is correctly representable by formulas of the following shape (in my notation):

OA&cOB&c^M(A&B)

142

Lennart Aqvist

If so, it is a disturbing feature of many systems of combined deontic-alethicmodal logic that the principle NoConfl

^{OA&cOB&c-iM(A&B))

turns out to be provable as a theorem in these combined systems. Again, we may hold the aggregation principle OK responsible: Suppose that such a combined system has OK as a theorem schema and assume the following: 1.

OA ->· MA

hypothesis (= Kant)

2.

OA & OB & ^M{A & Β)

hypothesis

(for some A, Β)

(for reductio ad absurdum)

Then: 3.

0(A & B)

from the first two conjuncts in 2 by OK

4.

M(A&cB)

from 3 by 1 (= Kant)

5. 6.

->M(A & Β) -.(CM & OB & ^M(A&cB)) (for all A, Β)

from the third conjunct in 2 from the deduction 2 - 5 by negation introduction

7.

1 -» 6

from the deduction 1—6 by implication introduction

Clearly, the upshot of this argument is that the Kantian sollen-können principle - in the presence of OK - implies the objectionable principle NoConfl. Thus, if Kant is provable in a combined system with OK, so is NoConfl. It is interesting to note here that Lemmon (1965), who apparently agrees (p. 45) with Searle (1978) on how to characterize cases of conflicting obligations properly, puts the blame for NoConfl not on OK, but on the Kantian principle. Unlike Schotch/Jennings (1981) and Barcan-Marcus (1980), Lemmon does not consider the alternative of keeping Kant while rejecting OK a more sensible course, in my view.

3.

A Theory ofPrima Facie Obligations: Some Conditions of Adequacy

In the light of the preceding discussion, I now want to state explicitly certain requirements that I expect my theory of prima facie obligations to satisfy, viz. the following:

Prima Facie Obligations in Deontic Logic

I

143

If the monadic operator Ο is read, and interpreted, as "prima facie it ought to be (the case) that", then none of the principles NoConfl, OD* or OK should be valid under that interpretation, whereas such principles as Con, Mon and Kant might well remain acceptable. I think this is a fairly safe conclusion that can be drawn from the arguments of Chellas, Schotch/Jennings, Searle and others. As for the logical framework within which our analysis of prima facie oughtness is to be carried out, we expect it to enable us to test principles and inferences involving that notion for validity and nonvalidity - at least to a reasonable extent. And we entertain the same expectation with respect to the more basic notion of requirement that Chisholm uses in his (1974). There is no doubt about the need for an improved theoretical apparatus within which Chisholm's important insights can be articulated and more clearly formulated - so far I am willing to agree with Thomason (1984), although I find his criticism (p. 154) of Chisholm (1974) not appreciative enough of its positive aspects.

II

4.

Choice of Framework: The System G ofDyadic Deontic Logic with Prepositional Quantifiers

For a rather compact description of the syntax, semantics and proof theory of the system G, see the appendix below. Thus, I intend to base the analysis of prima facie oughtness (obligation) on that system of dyadic deontic logic. En passant, let me just observe here that the system G is an extension (in various ways) of the dyadic calculus D3 of von Kutschera (1974), towards the end of sect. 5; for details, see §§ 26 f. of the appendix to Aqvist (1987). As my account of prima facie oughtness is essentially the one proposed already by Chisholm (see 1964, sect. 10, and 1974, towards the end of that paper), it is important to observe that his definition makes use of the device of quantifying over propositions, facts, states of affairs, events, or what have you. This means that, in order to be able to reconstruct the Chisholm account of prima facie oughtness in the framework of the system G, we have to supplement the expressive resources of G with so called propositional quantifiers. And in turn, these will have to be characterized both semantically (modeltheoretically) and proof-theoretically, say, by means of new axiom schemata and rules of inference added to those of G. Without entering into the details of doing this, let me just state that for present purposes I use the approach to propositional quantification in modal

144

Lennart Äqvist

logic presented by Kit Fine in (1970): this gives rise to a system Gn which, axiomatically or proof-theoretically, results from adding to the axiom schemata and rules of inference of G the schemata 6 through 8 of Fine (1970) as well as his rule Gen of generalization for his universal propositional quantifier. Fine's semantics for this quantifier can then be accommodated to the 'preferential' semantics for G in the straightforward way. For some basic features of the Fine approach, see my appendix below.

5.

The Operator O^: A Chisholmian Definition Added to the System Gn

In the spirit of the Chisholm contributions mentioned above, we now add to the system GK just described the following definition (definitional schema): Def OPf

Off A =df

Ξρ(ρ&ΟρΑ)

where the definiendum is to be read as "prima facie it ought to be that Λ", and the definiens as "there is a (possible) state of affairs ρ such that p actually occurs and ρ requires (that) A" or, more simply, as "there is a requirement for A" (see Chisholm 1964, sect. 3). Alternative readings of the clause "Op A" are available, e.g. "given that p, it ought to be that A". Note that, whereas the definiens uses the characteristic dyadic (binary) O-connective for conditional obligation, the operator 0?f in the definiendum is just a monadic (one-place) one. We can now state the main result of the present paper. Theorem: Interpret the O-operator in the principles OD*, OK, and NoConfl as O ^ in the sense just defined. Then, on that interpretation, those three principles are neither (i) valid according to the semantics for our system Gn, nor (ii) provable as theorems of that system according to its axiomatic proof theory. Proof (after Äqvist 1987, ch. 2, § 7, pp. 69-75). To begin with, let us have a look at the following figure: u o p,-

χ

>

Ο p,q,B

y ο p,q,A

>

ζ ο p ^ B

Fig. 1 The figure represents a set W of possible worlds (or situations), consisting of

Prima Facie Obligations in Deontic Logic

145

four distinct elements u, x,y, z, which are ranked from left to right by a binary relation > of strict preference or strict betterness, as shown by the picture. Thus, u is the optimal ('best') member of W, χ is the 'second best' member of W, and so on. Furthermore, in the spirit of Chisholm (1964), sect. 4, we may adopt the following readings of the formal sentences p, q, A and Β in the diagram above: p: q: A: Β:

I have promised one man to go I have promised another man not to go I go I don't go

Hence, we get this reading for the negated statement ->q: I have not promised 'the other' man not to go Again, these sentences, understood in accordance with the above readings, will now be taken to be true/false at different worlds as shown by the figure: p is true at u, χ, y, ζ and false nowhere in W q is true at χ and y, but false at u and ζ A is true at u and y, but false at *· and ζ Β is true at χ and z, but false at u and y Molecular Boolean compounds of these 'atoms' receive truth-values according to the familiar tables; thus, p&cq ("I have promised one man to go and another man not to go") is seen to be true both at χ and at y. How are we then to handle sentences of the form "0^4" which appear in the definiens of Def O^? Well, by means of the following truth condition, where w is any world in our 'universe' W\ TC

Op A is true at w, iff all the optimal /»-worlds are ^-worlds.

Here, we mean by a p/A-world any world in W at which pjA is true, and by "optimality" we mean "optimality under the preference relation >" (or, even better, "under the weak preference relation > that matches >"). Remark. A neater, more general way of expressing the truth condition T C is this: Let C be any wff of Gn and let M={W,P,>,V) be any G^-model in the sense of our appendix below. Then, set [C]** = {

x e

w

:M,x\=C}

146

Lennart Aqvist

i.e., [C]^ is the truth-set or extension in fM of the wfF C in the precise sense of

the set of worlds in W at which C is true in iM. Again, for any subset X of W, define opt(JQ = {x e X : (for all y € Χ) χ > y] We then give the following improved formulation of TC, which applies to all G^-wffs C, D: TC*

M,x\= OcD iff o p t ( [ C ] m ) Q [ D ] m

Let us now go back to figure 1. In the technical jargon of our appendix infra that figure can be said to amount to a specification of the following G^-model i W = (W,P,>,V),

where:

W

=

{u,x,y,z}

Ρ >

= the power set of W, i.e. the set of all subsets of W = { uu, xx,yy, zz, ux, uy, uz, xy, xz,yz} (where the ordered pairs are represented just by concatenation)

V(p)

=[p] = W

V{q) V(A) V(B)

=[q] = {*,y} =[A) = {u,y} =[B] = {x,z}

(where the assignment V is here construed as a map from the set of propositional variables into the set Ρ of states-of-affairs or propositions, and the index has been dropped in the notation "[ ]"). Appealing to the above definition of the function "opt", we readily verify that opt[/>] = opt W = {«} op%] = o p t { x j } = {*}. Finally, we claim that the model just specified is such that the negation of each of the three principles NoConfl, OD* and OK is true at the point y (= the actual' world in our model), when Ο — O ^ and when the sentences A, Β are interpreted as indicated. Ad NoConfl. We have y e [p] and opc[/>] = {«} C {u,y} = [A], hence O^A is true at y by our truth conditions and Def . Similarly, we have y G [q] and opt[^] = {χ·} C {x,z} — [5], so O^Β must be true at y by those truth conditions and Def Opf. Finally, we have [A & Β] = [Α] Π [5] = 0 so that, by

Prima Facie Obligations in Deontic Logic

147

the truth condition for Μ etc. (see the appendix), ->Μ(Α&Β) is true a t y as well. Q.e.d. Ad OD*. The verification is immediate from the preceding argument by virtue of the fact that [B) = W — [A] so that, in our model, OPf^A is true zty along with C0A. Q.e.d. Ad OK. We have already shown that A and Opf Β are both true at y in our model; we now show that, contrary to OK, the wff ->0*f (A&cB) is true at^ as well. Suppose it isn't, i.e. that Opf (A &B) is true at y. Semantically, by our truth conditions and Def O p f, this means that some state of affairs X in the set Ρ is such that7 e X and opt(X) C [A] η [5] = 0 . Since e Χ,Χ φ 0 whence, by LimAss, opt(X) Φ 0 and so cannot be a subset of the empty set. Hence Opf (A & B) cannot be true 21 y, so its negation must be true aty. Q.e.d. The points just established prove clause (i) of our theorem, i.e. that none of the three principles under discussion are valid according to the semantics of Gn. And clause (ii), that none of them are provable in the axiomatics of Gn, follows immediately from their non-validity in GK by virtue of the soundness of that axiomatics relative to this semantics (which is readily established). The proof of the theorem is complete. Corollary. The theory of prima facie oughtness (obligation) obtained by adding Def Opf to GK is adequate in the sense of meeting the requirement (I) in section 3 supra. This result is, of course, immediate from the above theorem. It can also easily be shown that, for Ο = O p f, the principles Con, Mon and Kant remain provable and valid in that theory. We also verify that the axiom ON

OT

which is equivalent in GK to the rule of proof O-nec

A

—— OA

remains valid in the theory, when Ο = O p f. Note that ON is rejected by Chellas (1974) and (1980), sect. 6.5. Remark. Suppose that in the definiens of our Chisholmian definition of "OpfA" we replace the clause "OpA" by "Of A", where " O ^ " is the

148

Lennart Aqvist

Danielsson-van-Fraassen-Lewis operator defined below in the appendix. The resulting definition can be seen to be equivalent in GK to the original one; the proof is left as an exercise to the reader. We devote the next section to the task of showing that our theory is also adequate in the sense of meeting requirement (II) of section 3 supra.

6.

Gn with Def O^Meets Adequacy Condition (II)

From the description of the system Gn in the appendix infra it is obvious that, at least in general, our testability' condition (II) is satisfied. I now want to substantiate this claim somewhat more in detail by testing for -validity certain Chisholm (1974) principles that, according to him, govern his proposed notion of requirement. Let me briefly summarize my findings in the following Theorem: Mostly sticking to the Chisholm (1974) notation, consider three lists of principles: (i)

(ii)

(iii)

Opfs

(A)

ρ & pRs

(B)

q&cqRrdcN(r —» ->s) ->· Opf^s

(Al)

pRq

3x3y{xRy)

(A2)

pRq

N(pRq)

(A5)

(pRs&qRs)

{pyq)Rs

(A6)

(pRq & pRs)

pR(q&cs)

(A7)

(p vq)Rs ->• (pRs ν qRs)

(T5)

pRq ->((/»& s)Rq ν {ρ & ~^s)Rq)

(A3)

pRq

M(p & q)

(Tl)

pRq

Μρ

(T2)

pRq ->· Mq

(T3)

pRq

(T4)

(pRq&CsR-iq)

^{pR^q)

Then the following results hold true:

H(p&Cs)Rq)v^({p&s)R^q))

Prima Facie Obligations in Deontic Logic

(i)

149

The laws (A) and (B) in the first list are provable and valid in Gn with Def , whether Chisholm's pRq is interpreted as Op q or as O^ q. The principles in the second list are all provable and valid in Gn, when the form pRq is interpreted as our Op q. The principles in the third list are all provable and valid in (χπ, when the form pRq is interpreted as our Odp q.

(ii) (iii)

Proof. Exercise. Note that the interesting laws (A5) and (A7) are identical under the interpretation pRq = Op q to (T29) and (T30) in section 14 of Aqvist/Hoepelman (1981). Remark. (A4)

Chisholm (1974) also asserts as axiomatic' the principle 3x3y3z{M(x &cy) & xRz

ScyR^z)

Clearly, for pRq — Opq (or = Op1 q), (A4) is not provable, nor valid in Gn. On the other hand, (A4) is certainly consistent and satisfiable in GK (as is easily shown, using the (/„-model defined in the preceding section). Again, the following principle: (C)

(p&cq&c pRs & qR^s)

(A4)

is certainly both provable and valid in Gn, by the logic of propositional quantification.

7.

Appendix: Syntax, Semantics and Proof Theory of the System Gn

As we remarked in section 4 above, our system GK is the result of adding socalled propositional quantifiers to the system G of dyadic deontic logic; G is presented and discussed both inAqvist (1986), § 2, and in Äqvist(1987), §23. Thus, the only essential difference of Gn from G consists in the presence in Gn of two propositional quantifiers, which are absent from the primitive expressive resources of my original system G. This difference is then reflected in the semantics and axiomatics of the two systems as follows. From the description of the semantics for Gn given below we gather that G^-models (or normative preference structures for Gn) are ordered quadruples whose second, characteristic item is a set Ρ of states of affairs' (or propositions'); this characteristic item is altogether missing in the semantics for G (see again Aqvist 1986, § 2). Furthermore, from the description of the axiomatic system GK infra we learn that Gn has a rule of universal

150

Lennart Aqvist

generalization for the universal propositional quantifier, which is, of course, missing in the axiomatics for my original system G; similarly, GK has four axiom schemata A0-A3 governing the universal propositional quantifier, which are not forthcoming in the axiomatics of G. Having made these observations, we now proceed to give a detailed description of the language, semantics and proof theory of the system GK, which is the main concern of this appendix. The language of the system Gn has, to begin with, a countable set of propositional variables p,q,r..., the propositional constants Τ (verum) and _L (falsum), the Boolean sentential connectives -> (negation), & (conjunction), ν (disjunction), —> (material implication) and ο (material equivalence). In addition to these familiar expressive resources, the language of Gn has the following characteristic deontic and modal operators·. Ο (for conditional obligation), Ρ (for conditional permission), Ν (for universal necessity) and Μ (for universal possibility). Finally, the language of Gn has two quantifiers., V (for universal propositional quantification) and 3 (for existential propositional quantification). The set of wffs (well formed formulae) of our ^-language is then defined recursively in the usual way (see Aqvist 1986, § 2, and Fine 1970). For instance, the clause for the propositional quantifiers will read: if A is a wff and ρ any propositional variable, then VpA and 3ρ A are wffs. Note also that we write OßA/PßA to render the ordinary language locution "if B, then it ought to be that /4'7"if B, then it is permitted that A", and that we thus deviate from the current notation 0(A/B) /P{AjB) for two reasons: (i) our present notation is paranthesis-free, and (ii) the way to read it is from left to right, and so not the other way around. We now adopt the following semantics for GK. By a Gn-model (or a normative preference structure for Gn) we mean any ordered quadruple

M=(W,P,>,V) where: (i) (ii) (iii)

W (W is a non-empty set of possible worlds'). Ρ (states of affairs, propositions) is a non-empty set of subsets of W that is closed under formulas in the sense of Fine (1970), sect. 1.1. > is a binary transitive relation on W ("is at least as good as") satisfying the following, somewhat controversial condition: LimAss. Every non-empty subset of W has at least one >-maximal element (see Aqvist 1986, § 2).

Prima Facie Obligations in Deontic Logic

(iv)

151

V (assignment, valuation) is a function from the set of propositional variables into the set P.

We can now tell what it means for any wffyl to be true at a world χ in W in a (jjt-model Μ — (W,P, >, V); in symbols: ίΜ,χ ^ A. The recursive definition starts out with obvious clauses for atomic wffs like 94.,χ 1= ρ iff χ e V(p)

(for any propositional variable p);

94,χ 1= Τ; not 94,χ 1= ±. Then it goes on to state the familiar clauses for molecular wffs having Boolean connectives as their principal sign. Again, more interestingly, we lay down the following clauses for molecular wffs having as their principal signs our deontic and modal operators as well as the propositional quantifiers: for any wffs C and D, 94,X\=

OCD

iff

optflq5*) C

[D]M

(for explanation, see sect. 5 above, the formulation TC*) Λί,χ 1= PCD iff o p t ( [ q ^ ) η [DF* ψ 0 94,χ 1= NC iff for each y in W : 94,y 1= C 94,χ \= MC iff for somey in W : 94,y 1= C 94, χ \= VpC iff 94',x\=C for all G^-models 94'= (W,P,>, V') such that V(q) = V'(q) for all propositional variables q distinct from p; 94,x\= 3pC iff 94',χ t= C for some G^-model 94' = satisfying that restriction.

(W,P,>,V')

Finally, we say that a wff A is GK-valid iff 94, χ Ν A for all -models 94 and for all x in W. This completes the description of our semantics for the system GN.

Turning next to the proof theory of Gn, we now propose a Hilbert style axiomatization. Thus, the axiomatic system Gn is determined by one rule of

152

Lennart Aqvist

inference, two rules of proof and a finite bunch of axiom schemata, as follows:

AA Β

Rule of inference

M P (modus ponens)

Rules of proof

Nec (necessitation for N)

— >

Β

A

NÄ A

Gen (generalization for V)

VpÄ

(A quick way of bringing out the distinction between a rule of inference and a rule of proof is this: in order for a rule of proof like Nec, or Gen, to be applicable in a deduction, its premise is required to be provable, whereas no corresponding restriction is imposed on the applicability in deductions of a rule of inference in the proper sense, like MP.)

Axiom schemata: (aO)

All tautologies (over the language of GN)

(al)

PbA ^Οβ-^Α

(a2)

Ob(A->C)->(

(a3)

OBA

(a4)

ΝA

(a5)

S5-schemata for N , M (i.e., M A -h· ^N^A,

ObA NOBA

OBA

N(A ΝA

(αΟ)

Ν (Α

- > ΟΒ C)

ο

Β)

{NA

-> NNA,

Β)

(0AC

NB),NA

and

MNA

A, A)

ο 0 5 C)

(al) (a2) (a3)

MA^(OAB^PAB)

(o4)

( 0 ^ ( 5 -> C)

(AO)

3/>Λ ο -Hp^A

(Al)

VpA(p)

(A2)

Vp{A 5 )

(A3)

vi

O^C)

^4(5), where 5 is any wff free for ρ in Λ(/>) (V/Λ

V/>5)

V/vl, where ρ is not free in Λ

Prima Facie Obligations in Deontic Logic

153

Here, (Al) = axiom schema 8 in Fine (1970), sect. 1.2; (A2) = his schema 6 and (A3) = his schema 7. As usual, the set of Gn-provable wffs is the smallest set of wffs S such that (i) every instance of (each of) (a0)-(a5), (α0)-(α4) and (A0)-(A3) is in 5, and (ii) S is closed under the rule of inference MP as well as under the rules of proof Nec and Gen. Soundness Theorem: Proof.

For all wffs A: if A is (/„-provable, then A is Gn-valid.

Straightforward and left as an exercise.

The question whether, conversely, all -valid wffs are G^-provable remains open. An affirmative answer would amount to the completeness of the axiomatic system Gn just presented. Already in the introduction to the present paper we spoke of a system DFL of dyadic deontic logic, which is known to be deductively equivalent — under appropriate definitions - both to the system G (Aqvist 1986, § 3) and to the system PR (see § 5 of that paper). DFL itself - so called in honour of Danielsson, van Fraassen and D. K. Lewis - is discussed in Aqvist (1986), § 3, and, more extensively, in Aqvist (1987), § 28. For our present purposes it is enough to know the following about DFL: its basic vocabulary is like that of the system G with the sole exception that two dyadic connectives Od^ and respectively replace Ο and Ρ as characteristic primitive deontic operators. And, from the standpoint of G and Gn, the meaning of these new connectives is easily explained by the following definitions, which might be added to G (and to CJ: Def Odfl:

0 f D =df PcΤ & OcD

Def Pdfl·.

P f D =df Oc 1 ν PCD

On the basis of these definitions we can derive truth conditions for wffs having OdA and P d fi as their principal sign. These conditions are stated relative to normative preference structures (for G or Gn, as the case may be) and, informally, amount to the following (cf. the formulations TC and TC* of sect. 5 supra): OQD is true at χ in Μ just in the case when the set of optimal C-worlds (according to M) is a non-empty subset of the extension (in 94.) of Z); and, dually, PQ'D is true at χ in 9/C just in the case when either there are no optimal C-worlds (according to 94) or the set of optimal C-worlds has a non-empty intersection with the extension (in 94) of D.

Lennart Aqvist

154

Clearly, then, O* is a stronger conditional obligation-operator than O, and PdA is a weaker conditional permission-operator than P.

References (1986). Lennart Aqvist: "Some Results on Dyadic Deontic Logic and the Logic of Preference", Synthese 66 (1986). A Q V I S T (1987). Lennart Aqvist: Introduction to Deontic Logic and the Theory of Normative Systems, Naples 1987. A Q V I S T / H O E P E L M A N ( 1 9 8 1 ) . Lennart Aqvist and Jaap Hoepelman: "Some Theorems About a 'Tree' System of Deontic Tense Logic", in Hilpinen (1981). B A R C A N - M A R C U S (1980). Ruth Barcan-Marcus: "Moral Dilemmas and Consistency", Journal of Philosophy 77 (1980). B R O A D ( 1 9 4 9 ) . C. D . Broad: "Some of the Main Problems of Ethics", in Feigl/Sellars

AQVIST

(1949).

(1974). Brian F. Chellas: "Conditional Obligation", in Stenlund (1974). C H E L L A S (1980). Brian F. Chellas: Modal Logic, Cambridge 1980. C H I S H O L M (1964). Roderick M. Chisholm: "The Ethics of Requirement", American Philosophical Quarterly 1 (1964). C H I S H O L M (1974). Roderick M. Chisholm: "Practical Reason and the Logic of Requirement", in Körner (1974); also in Raz (1978). D A N I E L S S O N (1968). Sven Danielsson: Preference and Obligation, Uppsala 1968. F E I G L / S E L L A R S (1949). Herbert Feigl and Wilfrid Sellars (eds.): Readings in Philosophical Analysis, New York 1949. F I N E (1970). Kit Fine: "Prepositional Quantifiers in Modal Logic", Theoria 36 (1970). G A B B A Y / G U E N T H N E R (1984). Dov M. Gabbay and Franz Guenthner (eds.): Handbook of Phibsophical Logic, vol. II, Dordrecht 1984. H I L P I N E N (1981). Risto Hilpinen (ed.): New Studies in Deontic Logic, Dordrecht 1981. K Ö R N E R (1974). Stephan Körner (ed.): Practical Reason, Oxford 1974. V O N K U T S C H E R A (1973). Franz von Kutschera: Einfuhrung in die Logik der Normen, Werte und Entscheidungen, Freiburg 1973. V O N K U T S C H E R A (1974). Franz von Kutschera: "Normative Präferenzen und bedingte Gebote", in Lenk (1974). V O N K U T S C H E R A (1976). Franz von Kutschera: Einfuhrung in die intensionale Semantik, Berlin 1976. V O N K U T S C H E R A (1982). Franz von Kutschera: Grundlagen der Ethik, Berlin 1982. L E M M O N (1965). E. J. Lemmon: "Deontic Logic and the Logic of Imperatives", Logique et Analyse 8 (1965). L E N K (1974). Hans Lenk (ed.): Normenlogik, Pullach (near Munich) 1974. RAZ (1978). Joseph Raz (ed.): Practical Reasoning, Oxford 1978. CHELLAS

Prima Facie Obligations in Deontic Logic

Ross

155

W. D. Ross: The Right and the Good, Oxford 1 9 3 0 . SCHOTCH/JENNINGS ( 1 9 8 1 ) . Peter K . Schotch and Raymond E. Jennings: "NonKripkean Deontic Logic", in Hilpinen ( 1 9 8 1 ) . SEARLE ( 1 9 7 8 ) . John R. Searle: "Prima Facie Obligations", in Raz ( 1 9 7 8 ) . STENLUND ( 1 9 7 4 ) . Sören Stenlund (ed.): Logical Theory and Semantical Analysis, Dordrecht 1974. THOMASON (1984). Richmond H. Thomason: "Combinations of Tense and Modality", in Gabbay/Guenthner (1984). (1930).

U W E BOMBOSCH

The Meaning of "Ought, Prima Facie" and Decision Situations: A Reply to Aqvist

1.

Introduction

In his instructive paper on prima facie obligations, Lennart Aqvist suggests a formalization of "ought, prima facie" within deontic logic that is inspired by Chisholm's analysis of this notion. Aqvist presents two very reasonable conditions of adequacy that any such formalization should fulfill and goes on to prove that his own suggestion meets the conditions. Here is the upshot of what I shall do in my reply: Section 2 will briefly sum up how Aqvist embeds "ought, prima facie" as into deontic logic. Section 3 will examine this " O ^ " within the context of decision situations. We will see that Aqvist s "Off", although on the right track, lacks a natural interpretation in what I think is the formal counterpart of a decision situation: a model of deontic logic. In order to repair this shortcoming, section 4 will develop a new formalization of "ought, prima facie", which remains faithful to the spirit ofAqvist's approach and in particular meets Aqvist s conditions of adequacy. 2.

Semantics of the System Gn

First, we specify the language L of Aqvist's system Gn. The primitive symbols of L are: (i) (ii) (iii) (iv)

countably many prepositional variables p, q,pOip\ ,p2) • - -> the logical connectives ->, &, the modal operator • and the dyadic deontic operator O, the "ought, prima facie"-operator O p f.

Definition 2.1 (of the L-formulas): (a) (b) (c)

Every propositional variable is a formula. If A is a formula, so are -A, Π A and OPf{A). If A, Β are formulas, so are A & Β , Οβ{Α).

"Ought, Prima Facie": A Reply to Aqvist

157

A structure !M for the now given language L is a quadruple M={W,P,!], then [jDo XI]. The person who utters this, according to my metatheory, is expressing his acceptance of a norm that is conditional in form. The norm tells how to go from norms for intrinsic preferences to a normative directive for action. 4.6. I am suggesting, in short, that we regard formal decision theory as aspiring to be a part of the theory of practical rationality, but only a part. It makes claims about instrumental rationality, and it makes claims about the formal characteristics of rational intrinsic preferences. We should see it, though, as making no claims about the substantive content of rational intrinsic preferences. In particular, it makes no claims about whether it is rational to have intrinsic preferences for or against certain kinds of acts: lies, say, or harmings.4 Nor, I've said, is decision theory all of the theory of instrumental rationality. Since one acts with various kinds of information and in various kinds of ignorance, the general theory of practical rationality would have to encompass theoretical rationality. It would have to include a theory of rationality in belief. Now as with intrinsic preferences, decision theorists have a standard line on rational degrees of belief, namely personalisnr. that any assignment 4

T h e term "instrumental" is sometimes used in a w a y that rules o u t intrinsic preferences f o r acting in certain ways. I mean to imply n o such restriction in m y o w n use o f the term. As I conceive the subparts of the theory o f practical rationally, w h e t h e r it is rational t o prefer, intrinsically, such things as n o t lying is a question in the t h e o r y o f the substantive o f preferences. It is no part o f decision theory.

rationality

248

Allan Gibbard

of degrees of belief to propositions is rational so long as it satisfies the standard axioms of probability. As with intrinsic preferences, though, a part of this claim could be rejected and classical decision theory would lose none of its interest. The interest of the theory depends on its claiming that if degrees of belief are rational, then they will satisfy the standard axioms of probability. Much of the interest of classical decision theory lies in its arguments to this conclusion. The converse, though - the personalism' — we could reject, and still find the arguments decision theorists devote most of their effort to none the less interesting. 5 4.7. I interpret classical decision theory, then, as consisting in three related kinds of norms: (1) Norms that rule out all combinations intrinsic preferences that lack certain formal characteristics. (2) Norms that rule out all combinations of degrees of belief that lack certain characteristics (specified by the standard probability axioms). (3) Norms for action that are doubly hypothetical: they tell how to get to norms for action from (i) norms for intrinsic preference and (ii) norms for belief. In decision theory so conceived, norms for preferences will be included, but not for intrinsic preferences. The preferences that figure in action are often non-intrinsic: I prefer, say, grading papers to reading a novel, and going to the grocery store to grading papers, and so end up going to the store - but my preference won't be for the trip to the store taken in itself. These preferences won't be purely intrinsic. Again, decision theory's norms for non-intrinsic preferences will be doubly conditional: they will say how to go from norms for intrinsic preferences, and from norms for belief, to norms for decision and for non-intrinsic preferences. Decision theory, read in this way, leaves open the question of what things are intrinsically good: It leaves open the question of what it is rational to prefer, intrinsically, to what. It deals in norms for preference, including 5

An exception to this is de Finetti's central work (1937) showing that in important cases, people with quite different prior probability ascriptions come to converge, as evidence accumulates, in the probabilities they ascribe to propositions. Still, given any fixed corpus of evidence, some prior probability functions will give highly divergent posterior probabilities. We may regard some wide range of prior probability ascriptions as within the range of reason, and still conclude that on certain matters, all reasonable people with the evidence must agree. We can't, though, claim that any prior probabilities whatsoever would be within the range of reason, or else we can never claim that there is now enough evidence that all rational people would now have to agree.

Preference and Preferability

249

norms of formal coherence for intrinsic preferences. Its norms, though, don't tell you substantively what, intrinsically, to prefer to what. They don't tell you such things as whether to prefer a history in which you will be less happy but accomplish more, to a history in which you will be happier but accomplish less. They don't tell you whether intrinsically to prefer a history because in it, on the present occasion, you keep a promise rather than breaking it.

5.

Preferences as Explanatory

5.1. What, in decision theory, are preferences' supposed to be? If we find an answer, we can ask whether preferences in this sense fit a metatheory of good. Are they what we need for a theory of what terms like "better" or "preferable" mean? By the "better" of two alternatives, I proposed, we mean the one it is rational to prefer. Should the word "prefer" in this formula be taken in a sense that classical decision theory provides? 5.2. In the decision theoretic tradition, preferences are revealed by action. Perhaps we should even say that preferences don't figure independently in the theory, that its norms for preference really amount to norms governing dispositions to action. 6 What the norms of decision theory demand is no more than a kind of coherence in an agent s dispositions to action. All talk of 'preferences' in the theory is eliminable. Whether or not this is right, though whether or not preferences in decision theory boil down to dispositions to action - the theory does depict a tight relation between the two. Given a choice between two alternatives, classical decision theories assume, the agent will choose whichever alternative he prefers to the other. Given a choice among more than two alternatives, a person will choose the alternative he most prefers, or one of them in case of ties. Thus whether or not preference is fully revealed in dispositions to action, dispositions to action always do reveal aspects of a person's preferences. A choice reveals that none of the unchosen alternatives was preferred to the one that was chosen. 5.3. In this, the decision theorist's notion of preference may be at odds with more ordinary notions of preference. True, common sense does take preference and choice to be closely linked. Often we do take it that a person will ^ Whether this is right would raise complex questions of the interpretation of various formulations of classical decision theory. Savage ( 1 9 5 4 , 1 9 7 2 ) treats actions as functions from 'states' of the world to 'consequences'; this seems far removed from an action as something one can decide to perform. For a reading of Savage's framework, see Gibbard (1986a).

250

Allan Gibbard

choose what he most prefers. Some turns of phrase, though, go against this. If asked whether I prefer staying home and studying or going to the movies, I might be a little puzzled, and find I had been asked the wrong question. "I'd prefer going to the movies, but I've got to stay home and study", I might reply. We don't think of our motivations to do what we know we've got to do as straightforward cases of preference. (And of course when the got to' is moral, something like this leads to Kant's controversial distinction between an 'inclination' (Neigung, Triebfeder) and the motive of duty (Pflicht, Bewegungsgrund).) Now I think that common usage in this matter may have a point. If we try to use preferences as a psychological, explanatory notion, the decision theorist's characterization may not do very well. Motivations are of diverse kinds. They can be grounded in emotions, in cravings and appetites, in the maintenance of self-esteem, in the social pressures of one's circumstances, and in the acceptance of norms. A good psychology of human motivation would presumably extend this list and revise it; I don't claim to know how human motivations are best classified. I do suspect, though, that the diversity of this list would remain. Wouldn't our evolutionary history have left us with layer upon layer of motivations and inhibitions, some older and some newer in humanity's phylogeny? 5.4. Perhaps even so, a person's motivations could be weighed against each other, much as vector forces resolve in physics. Then we could still speak of the 'valence' of each motivation, and thus regard each motivation as a 'preference' of a certain strength and in a certain direction. Preferences, we could say, add up, in a vector-like way, to yield one's preferences all told, and one's preferences all told determine action. Each preference might have its own special character, but still they can be weighed against each other on a single scale. Speaking of motivations in this way seems almost irresistible. What other schema do we have at our command that could even remotely do justice to the complexities of motivations? Even to speak of different 'motivations', as I am doing, seems already to be buying into a picture like this. And there are reasons to expect that the picture might be accurate, more or less. In the first place, the selection pressures that shaped human motivational mechanisms fit a decision-theoretic model nicely — hence the success of game theory in population genetics.7 This doesn't mean that the human brain itself will amount to a computer that does computations of classical decision theory. Human predicaments - and human social predicaments especially - will be too complex

7

See, for instance, Maynard Smith (1982).

Preference and Preferability

251

for such computations to be possible, unless they proceed from oversimplified, misleading models of the world. We have to expect the brain to code for motivational heuristics that will depart from the dictates of classical decision theory, but still do pretty well by decision-theoretic standards. 8 Often weighing mechanisms make for good heuristics, but sometimes they may not.

5.5. It isn't clear, then, whether talk of preferences' can be anything but loose. Such loose talk serves ordinary purposes well at times - though at other times, I would claim, it serves them badly. (It doesn't cope well, in my experience, with such puzzles to common sense as depression, procrastination, coping with status insecurities in social encounters,... and the list might well be extended to fill the page.) An adequate psychology of motivation, I've been suggesting, might not have a place for preference in any sense that parallels a decision theorist's regimentation of the term. Indeed even in its own terms, classical decision theory doesn't offer us a concept of preference that will serve purposes of psychological explanation in general. Classical decision theory tells us how to glean the preferences of a rational agent - rational in the form of his dispositions to action — from his dispositions to action. For the decision-theoretically rational agent, preferences are tied to action in a tight system. Decision theorists admit, though, that people can violate their precepts of rationality, and they don't tell us what an irrational person's preferences are.9 All this leaves open the possibility that a good psychological theory of motivation would employ a systematic concept of preference that was closer to ordinary usage than is the decision theorist's concept. Perhaps — but I have my doubts. Why think the ordinary, loose notion of preference can be given a tight, scientific application to human beings as we actually are?

8

9

See Nisbett/Ross (1980) for a wide-ranging treatment of psychic 'heuristics' in this sense. Examples in the psychology of decision include Kahneman's and Tversky's models for choices among gambles (1979), and their talk of "framing" decision problems (Tversky/Kahneman 1981). These heuristics violate rationality as depicted by classical decision theory. Still, these psychologists argue, the methods may work well, for the most part, on problems that classical decision theory could not deal with tractably. Ramsey thought that decision-theoretic rationality is a kind of consistency in preference and belief. "If anyone's mental condition violated these laws, his choice would depend on the precise form in which the options were offered him, which would be absurd" (1931, p. 182). I take it, then, he thinks there is no clear fact of the matter what a person's preferences are if his dispositions to choose among gambles violate decision theoretic rationality.

252

Allan Gibbard

6.

Revealable Preference and Good

6.1. I have been asking whether a general concept of preference would figure in a good psychology of human motivation. It might not, I've been suggesting. One might think, in consequence, that preferability likewise must be suspect. If there will often be no precise fact of the matter what a persons preferences are, can we talk with any precision about what it is rational him to prefer? 6.2. The answer might be that we can. Mightn't a fully rational person have definite preferences in the decision theorists' sense, even though people in general don't? Determinate preferences might be a rational attainment, not an automatic consequence of being human and wanting things. One dictate of rationality, then, would be to attain preferences in the decision-theorists' sense. Preferability, then, might still be a matter of the preferences it is rational to have - with preference taken in the decision theorists' sense, read off from dispositions to action. For purposes of giving the meaning of "better" or "preferable", the decision theorist's characterization of preference might be just what we need - or at least it might be on the right track. We may not need to know what preferences are in general to explain the terms "better" or "preferable" in terms of preference. Let's explore this line. Norms for preference govern one's dispositions to action; for a fully rational agent, actions grow out of a coherent set of preferences. Perhaps norms for preference govern more than dispositions to action, but they govern that at least. Could we say, then, something like this: When I say that X is better than Υ, I mean that if offered a choice between Y and X, it is rational to pick X? A person who chooses X when Y is available, though, shows only that he doesn't prefer Y to X. He doesn't show that he strictly prefers X to Y. We need to distinguish preference and indifference, and this can be an embarrassment to revealed preference theory. For the approach to value concepts that we're exploring, though, this is no great problem: We don't need to know what "indifference" means to explain terms like "good" and "better". That two alternatives are equally good, we can say, means that given a choice between them, it is rational to choose either one. Our metatheory is now looking much like Hare's (1952). To call something "good", Hare says, is to commend it, and a commendation is a kind of imperative: One is prescribing, "Choose it!"10 Now as Hare presented his theory, it was puzzling what kind of imperative this was supposed to be. It seems, 10

Hare (1952), chs. 8 f. The prescription, Hare says, must be universal. This means it applies to all situations that are like this one in their universal characteristics. See Gibbard (1988)

Preference and Preferability

253

though, to be an imperative of advice, of informed rationality.11 Now according to my own metatheory, such imperatives express the speakers acceptance of norms. The person, then, who says that X is better than Y is expressing acceptance of norms that say: If you have a choice between X and Υ, choose X and don't choose Y. The person who says that X and Y are equally good is expressing acceptance of norms that, for the case of having a choice between X and Y, both permit choosing X and permit choosing Y. The speaker in this case is issuing not a normative mandate, but a normative permission. 6.3. This can't yet be right, though. What it makes sense to choose can depend on who you are, whereas what's better doesn't. Imagine, say, that the theory of rationality you accept is hedonistic egoism: If a person can choose in full knowledge between two possible histories, you think, it always makes sense for him to choose the history that includes more net pleasure for himself. The rational choice for the knowledgeable agent is that one that brings him himself the greatest balance of pleasure over displeasure. Suppose, then, there's more pleasure for Jim in history X, and more pleasure for Judy in history Y. As a hedonist, you will tell Jim to choose X and Judy to choose Y or that's what you'll advise if you're being sincere. Sincere or not with Jim and Judy, you can form your own opinion about what it is rational for Jim to do and for Judy to do. That amounts to deciding for yourself, "If I'm in Jim's shoes and offered the choice, let me choose X and not Y, whereas if I'm in Judy's shoes and offered the choice, let me choose Y and not X". From what in all this can we glean your opinion of which history is better? O f course you will think that X is better for Jim and Y is better for Judy. And indeed as an egoist, you seem to have no use for a notion of which history is better 'from the point of view of the universe'. No one ever rationally takes that point of view, you would claim, and so nothing ever hinges on the question. A utilitarian like Sidgwick — a hedonistic universalist in his theory of rationality - has a conception of one possible history's being better than another, and he has great use for this conception. You, though, as a hedonistic egoist with regard to rationality, have neither the concept nor the use for it. 6.4. Could we try, then, defining not "better" from the point of view of the universe, but "better for" a given person? X is better than Y for a person A, for an interpretation o f Hare on the universal prescriptions. 11

I interpret Hare's prescriptions as expressions of preferences; see Gibbard ( 1 9 8 8 ) . I'm not sure whether that was Hare's view in ( 1 9 5 2 ) ; see esp. ch. 1 and sect. 2 . 2 . Imperatives o f advice, I claim, are not expressions o f preference: a burglar who gave sincere advice would advise us to keep our doors locked, but prefer us not to.

254

Allan Gibbard

let's try saying, if, were A to be given the choice between X and Y, it would be rational for him to choose X and not for him to choose Y. Even this, I think, won't capture the views of people with non-hedonistic views of the rational grounds for choice - or of many such people. A person has special reason, it is widely thought, to keep his own promises and cultivate his own virtue. These reasons are intrinsic; they don't just depend on one's being in the best position to see to it that one's promises are kept, and they don't just depend on the advantages to oneself of keeping one's promises. Keeping one's promises, then, is rationally required, but not because so doing is good for one. Or at least there is some rational weight to be given to keeping one's promises, a weight that does not entirely stem from ways in which keeping one's promises can be good for one. We can't then, characterize what is good for a person by what it is rational for the person to choose. Indeed I suspect there is no clear, general concept of something's being good for someone — no concept that could be shared by people who span a common range of views on the grounds for morality and rationality. The notion of being good for someone may lend itself to regimentation only in the context of fairly simple theories like hedonism. It is rational, some people think, to seek accomplishment for its own sake, in scholarship, in the arts, and in other realms. We care about the genuine esteem of other may have for us, and this concern may be a rational one to have. We care about being worthy of their esteem, and this concern too may be rational. If I think all these preferences rational, does this involve thinking that accomplishment, esteem, and worthiness are all goodfor a person? My own ear doesn't clearly answer yes or no. 6.5. The concept of being intrinsically better, though - not better for, but just plain better — seems far more definite. How does it work? There is a kind of impartiality built into the concept: Not that it requires an impartiality that is universal: fans of the Michigan Wolverines football team can exclaim together about what's good and what's bad as a game progresses, without bringing in the standpoint of anyone else. But the talk of good and bad is still neutral among participants in the conversation. The same talk would be tendentious in dealings with people from Ohio. It couldn't be regarded as seriously defensible, or if it were, the grounds would have to be far different. Likewise I can think to myself about good and bad developments as, say, I compete for a job - without having to decide whether my getting the job is best on an impartial view of things. My conversational group has then shrunk to one - though with luck I may be able to rope in a few friends to share my good news and bad. They can't, though, be equally friends of another competitor, or

Preference and Preferability

255

I'll have to qualify my language or make the conversation awkward for them. Suppose, though, we take all humanity into the conversational group we're addressing. Will our talk of intrinsic good then have any meaning? It will, perhaps, if we are neutralists in our view of rational goals: if we think that everyone's rational aims should be to promote the same states of affairs. 6.6. What, though, if we accept a common-sense diversity of intrinsic aims as rational? I'm inclined to this hypothesis: Goodness is what our rational aims have in common. Goodness, in other words, is what everyone has some reason to promote if he can, whatever other rational grounds for choice apply to him in particular.12 A characteristic goes to make a history good to the degree that it gives everyone reason to choose that history, if offered a choice. Talk of good and bad, then, weighs up whatever reasons for action take a particular form: of being consequentialist and neutral. The reasons are consequentialist in that they are reasons to promote, if one can, the existence of states of affairs of certain kinds. They are neutral in that they are reasons for everyone to promote those states of affairs, if they can, regardless of their relations to those states of affairs. How would this work with a deontologist? The deontologist will probably think that it isn't always rational to act to produce the best expected consequences. For by definition, he thinks that the morally right act isn't always the one with the best consequences. It would be hard to see the force of this doctrine if he nevertheless thought that the act with the best expected consequences is always rationally required. Being morally right may not be the same thing as being rational, but there'd better be a connection. If the moral rightness of an action has no bearing on its rationality, it would be hard to see why moral rightness mattered. Will our proposed definition of good, then, capture something a deontologist might mean by the term? I think it will. I can't speak for all deontologists, but let's imagine one as plausibly as we can — Donna, let's call her. Donna thinks that many things bear on which act it is rational to perform. 12

I don't take having a reason to do something to mean anything that is definable in terms of the person's intrinsic preferences alone. (Indeed I'd better not, given that I don't think the concept of preference, as it figures in psychology, isn't in good enough shape to do the job.) Rather, I take an expressivistic, non-cognitivistic view of talk of reasons: A person w h o calls R a reason for person A to do X expresses his acceptance of a norm that, as applied to A's situation, tell A to treat consideration /fas weighing in favor of doing A'. See Gibbard (1990), pp. 160—4). This allows for substantive characterizations of reasons in terms of preferences: it might be, for all I'm saying, that indeed C constitutes a reason for A to do X just in case if A were calm and fully knowledgeable, then C would motivate Λ toward doing X (cf. Brandt 1979). I'm saying, though, that no such formula gives what the term "a reason" means.

256

Allan Gibbard

Many are tied to the agent: whether the act would constitute breaking a promise the agent has made, whether it benefits those who have benefited the agent, and so on. Still, some other considerations weigh impersonally. There may be things that everyone has some reason to promote. There may be features the course of history might have - things that might happen or not - with this characteristic: that for anyone at all, that an act would have that as a consequence rationally weighs in favor of the act. These, Donna can say, are the things that are good. The rationality of an act, then, isn't entirely determined by the impersonal goodness of its consequences. It is determined by a variety of considerations, many tied to the agent. But that X is good means that for any agent, the fact that an act will have X as a consequence weighs toward the act's being rational. Donna can accept that there are side-constraints on rationality. A consideration may weigh toward performing an act, and yet the act may be rationally precluded by other considerations. Suppose, say, that killing the joyless old widow for her gold would make for happiness. This consideration will weigh in favor of committing the murder, and weigh in favor of everyone else's promoting the act. Still, Donna might say, side constraints against murder flatly rule the act out - rationally as well as morally. Impersonal goodness as a consideration can be trumped. Still, it weighs in the absence of trumps, and that's what it means to say it is a consideration of goodness. 6.7. What, though, of another deontologist who thinks that side constraints are all there is? Don (as I'll call him) thinks there are side constraints one must not transgress. Apart from them, he thinks, one is free, rationally and morally, to pursue whatever one wants. Can he mean by "good", then, what I'm saying he must? I'm inclined to think he won't - he won't, that is, have any place in his thinking for the grandiose concept 'good from the point of view of the universe.' Other common uses of the term will fit reasonably well in his mouth. He'll be able to use the word "good" when talking with people who share his rationally permitted wants in the matter at hand. He'll talk, in effect, as he would with a group who all cheer for the same team. Still, such things as war, flood, and pestilence among far-flung strangers won't be bad, in his system, unless he and his conversational group all want them not to occur. I have an unsympathetic view of such a doctrine. Most of us, I think, take more seriously the question of what's worth wanting. We might well agree that the answer depends in part on ones pursuits and enthusiasms, one's attachments, and one's sense of self - along with one's tastes, cravings, and susceptibilities. But we won't think that any preferences whatsoever would be rational ones to have, in whatever strength one had them. We won't think

Preference and Preferability

257

that any preference whatsoever is one it would be rational to lack. We won't even think that any preference would be rational so long as it weren't a preference for something that violates moral side constraints. Perhaps someone who found appeal in Don's view, then, would be more resourceful, and find something else for Don to mean by "good". I myself find it plausible that he wouldn't need any notion of a good that's neutral among people.

7.

Summary

I started out with the thoughts that goodness is a matter of preferability, and that, of two things, the preferable one is the one it is rational to prefer. In my book I gave a theory of what "rational" means, and so in this paper I faced two other questions: What concept of preference, if any, fits the formula that the preferable of two things is the one it is rational to prefer? How do we fill out this formula: Rational for whom to prefer, when? Classical decision theory treats preference as revealed in choice, as consisting in one's dispositions to choose. I was skeptical about such revealable preference as a good explanatory concept in a scientific psychology. The decision theorist's concept, though, may be pretty much what we need for purposes of defining preferability. Roughly, we can try saying, the preferable of two things is the one it is rational to choose. I then worked on refining this characterization. Talk of what's preferable to what is neutral among parties to the conversation, and we might have a use that treats all humanity as our conversational group. Not everything that bears on rational choice need then be a matter of goodness. Good-making considerations will be whatever considerations bear on choices in a special way: consequentially and neutrally. Goodness will then be a matter of how these goodmaking considerations stack up. I've been concerned, in this paper, with questions of meaning: What does it mean to say that one thing is preferable to another? I haven't tried to answer substantive questions of what makes one thing preferable to another, of what qualities make a thing intrinsically good or intrinsically bad. Those are immense questions, not to be quickly resolved. I've been trying not to answer them, but to figure out what they mean. Answering such questions, I've been saying, would consist in deciding what to choose, on what basis, for the widest variety of actual and hypothetical circumstances.

258

Allan Gibbard

References AYER

( 1 9 3 6 ) . Alfred J. Ayer: Language, Truth and Logic, London 1936. ( 1 9 7 1 ) . Robert Binkley, Richard Bronaugh and Ansonio Marras (eds.): Agent, Action and Reason, Oxford 1971. ( 1 9 4 6 ) . Richard B. Brandt: "Moral Valuation", Ethics 56 (1946).

BINKLEY ET

BRANDT BRANDT DABONI

AL.

( 1 9 7 9 ) . Richard B . Brandt: A Theory of the Good and the Right, Oxford 1979. E T A L . ( 1 9 8 6 ) . Luciano Daboni, Aldo Montesano and M. Lines (eds.): Recent Devebpments in the Foundations of Utility and Risk Theory, Dordrecht 1986.

Jon Elster and Aanund Hylland (eds.): The Foundations of Social Choice Theory, Cambridge 1986. E W I N G ( 1 9 3 9 ) . A. C. Ewing: "A Suggested Non-Naturalistic Analysis of Good", Mind 4 8 ( 1 9 3 9 ) . F E I W E L ( 1 9 8 7 ) . George R . Feiwel (ed.): Arrow and the Foundations of the Theory of Economic Policy, New York 1987. D E F I N E T T I ( 1 9 3 7 ) . Bruno de Finetti: "Foresight: Its Logical Laws, Its Subjective Sources", in Kyberg/Smokler ( 1 9 6 4 ) ; English translation of "La prevision: ses lois logiques, ses sources subjectives", Annales de l'Institut Henri Poincari 7 ELSTER/HYLLAND

(1986).

(1937).

(1986A). Allan Gibbard: "A Characterization of Decision Matrices that Yield Expected Utility", in Daboni et al. (1986). G I B B A R D (1986B). Allan Gibbard: "Interpersonal Comparisons: Preference, Good, and the Intrinsic Reward of a Life", in Elster/Hylland (1986). G I B B A R D ( 1 9 8 7 ) . Allan Gibbard: "Ordinal Utilitarianism", in Feiwel ( 1 9 8 7 ) . GIBBARD

( 1 9 8 8 ) . Allan Gibbard: "Hare's Analysis of'Ought' and its Implications", in Seanor/Fotion (1988). G I B B A R D ( 1 9 9 0 ) . Allan Gibbard: Wise Choices, Apt Feelings, Oxford 1990. G I B B A R D ( 1 9 9 2 ) . Allan Gibbard: "Reply to Blackburn, Carson, Hill, and Railton", Philosophy and Phenomenological Research 52 (1992). H A R E ( 1 9 5 2 ) . Richard Μ. Hare: The Language of Morals, Oxford 1952. H A R E ( 1 9 7 1 ) . Richard M. Hare: "Wanting: Some Pitfalls", in Binkley et al. ( 1 9 7 1 ) . H A R E ( 1 9 8 1 ) . Richard M. Hare: Moral Thinking, Oxford 1981. J E F F R E Y ( 1 9 6 5 ) . Richard C. Jeffrey: The Logic of Decision, Chicago 1965. K A H N E M A N / T V E R S K Y ( 1 9 7 9 ) . Daniel Kahneman and Amos Tversky: "Prospect Theory: An Analysis of Decision under Risk", Econometrica 4 7 ( 1 9 7 9 ) . K A H N E M A N / T V E R S K Y ( 1 9 8 1 ) . Daniel Kahneman and Amos Tversky: "The Framing of Decisions and the Psychology of Choice", Science 211 (1981). K Y B E R G / S M O K L E R ( 1 9 6 4 ) . Henry Kyberg and Howard E. Smolder (eds.): Studies in Subjective Probability, London 1964. M A Y N A R D S M I T H ( 1 9 8 2 ) . John Maynard Smith: Evolution and the Theory of Games, Cambridge 1982. GIBBARD

MILL

( 1 8 6 1 ) . John Stuart Mill: Utilitarianism, London 1863; first publ. in 1861.

Preference and Preferability

259

(1903). G. E. Moore: Principia Ethica, Cambridge 1903. M O O R E (1942). G. E. Moore: "An Autobiography", in Schilpp (1942). VON N E U M A N N / M O R G E N S T E R N (1953). John von Neumann and Oskar Morgenstern: Theory of Games and Economic Behavior, third ed., Princeton 1953. N I S B E T T / R O S S (1980). Richard Nisbett and Lee Ross: Human Inference, Englewood Cliffs, NJ, 1980. R A M S E Y (1931). Frank P. Ramsey: The Foundations of Mathematics and Other Logical Essays, London 1931. SAVAGE (1954). Leonard J. Savage: The Foundations of Statistics, New York 1954. SAVAGE (1972). Leonard J. Savage: The Foundations of Statistics, second, revised ed., New York 1972. S C H I L P P (1942). Paul Arthur Schilpp (ed.): The Philosophy of G. E. Moore, La Salle, 111., 1942. S E A N O R / F O T I O N (1988). Douglas SeanorandN. Fotion (eds.): Hare and Critics, Oxford 1988. S T E V E N S O N (1937). Charles L. Stevenson: "The Emotive Theory of Ethical Terms", Mind 46 (1937). S T E V E N S O N (1944). Charles L. Stevenson: Ethics and Language, New Haven 1944. MOORE

JULIAN

NIDA-RÜMELIN

Goodness and Rational Preferability: A Reply to Gibbard

1.

Introduction

Aristotle's Nicomachean Ethics begins with the sentence: "Πάσα τέχνη και πάσα μέθοδος, ομοίως δέ πράξίς τε και προαίρεσις, αγαθού τινός 'εφίεσθαι δοκεί· διό καλώς άπεφήναντο τάγαθόν, ού πάντ 1 έφίεται."1 Allan Gibbards "Preferences and Preferability" begins in a similar way: "What things are good? Deciding this, I suggest, consists in deciding what to prefer to what." (Sect. 1.1.) Aristotle's and Gibbards theories are both metatheoretical: both are concerned with the question how the term "good" can be understood adequately On the other hand, these two metatheoretical approaches differ in the respective immediate material implications ensuing from them: in Aristotle's theory an inherent value is ascribed to certain forms of πράξις; Gibbard, on the other hand, commits himself to a consequentialist theory of rational preferences, notwithstanding the fact that in his paper and more explicitly in his book (Gibbard 1990) there are some interesting modifications to the orthodox view. !n fact Aristotle hints at the point that in my view is fatal to any consequentialist theory including Gibbards: "διαφορά δέ τις φαίνεται των τ ε λ ώ ν τα μεν γάρ είσιν ένέργειαι, τα δέ παρ' αύτάς έργα τινά."2 But before we come to this point, I should make one remark about Gibbards normexpressivist theory of rational preferability.

1

2

1 0 9 4 a l - 3 . There is no really adequate translation of this short piece of text into modern language because the four crucial concepts ( τ έ χ ν η , μ έ θ ο δ ο ς , π ρ α ξ ί ς , π ρ ο α ί ρ ε σ ι ς ) have no synonyma in English. One translation is: "Every art and every investigation, and similarly every action and pursuit, is considered to aim at some good. Hence the good has been rightly defined as 'that at which all things aim'." NE 1094a3—5. One translation is: "Clearly, however, there is some difference between the ends at which they [all things] aim: some are activities and others result distinct from the activities."

Goodness and Rational Preferability: A Reply to Gibbard

2.

261

A Remark on Norm-Expressivism

Gibbard takes goodness to be rational preferability. One can agree to this without accepting his expressivist characterization of rational preferability. Gibbard (1990) argues for norm-expressivism by referring to psychological (or metapsychological) assumptions, by explaining norms in the context of evolutionary biology and by giving an account of objectivity and consistency claims within the context of norm-expressivism. It is not my task here to comment on this book, even if it is difficult to comment on the paper without commenting on the book, but one argument against norm-expressivism (and other forms of internalism) should be mentioned. In Gibbard s terminology intrinsic goodness refers to events ("things that happen", sect. 2.1 of his paper). "Good things are desirable" and "something is desirable if it would be reasonable to desire it" (sect. 2.2), if a desire for it would be rational. In this way questions of goodness are brought down to questions of warranted preference. Gibbard understands this as a theory of the meaning of the term "good" (cf. sect. 3.2). The reduction of good' to 'warranted preference' suggests a cognitivist interpretation insofar as the term "rational" or "warranted" points at giving reasons for actions. But Gibbard tries to prevent the cognitivist drift by elaborating an expressivist interpretation: To call a preference "warranted" is to express the acceptance of norms that say to have the preference. For ordinary language (and ordinary understanding) there is a difference between the claim that a preference is rational and the expression of being in a specific mental state constituted by an acceptance of the norms which guide the respective preference. "This preference is rational" is a statement that demands an explication of its reasons, and there is no way to get rid of the demand by referring to a subjective acceptance of certain specific norms. A third-person-expressivism leads inevitably to emotivism in the usual sense. If the statement of a person A: "I assume the preference prj to be rational" could be reported as "A accepts the norms n\,...,n„, and, under the given circumstances, prj is the only preference that fulfils n\,...,n„", then we are forced into a reflective attitude regarding the first person utterance: "I assume the preference pr x to be rational" is equivalent to "I accept the norms «!,...,»„, and, under the given circumstances, pr x is the only preference that fulfils « ι , . . . , n„ . This in fact is traditional emotivism. Now imagine a situation in which a person Β denies the validity of the statement "prj is rational" by referring to different norms n\,..., n'm, accepted by Β that are not fulfilled by pr j. In this situation a dialogue of the following kind could develop: A refers to his normative belief system (including «1,..., n„), and Β refers to his normative belief system (including n\,..., n'm),

262

Julian Nida-Rümelin

each in order to show t h a t p r j is rational or, respectively, that/»r, is irrational. Case 1:

Case 2:

A and Β come to the result that n\,...,n„ and n\,...,n'm are congruent. So one of them, A or B, must be guilty of logical inconsistency and one of the two statements will be disproved logically. A and Β come to the result that n\,...,n„ and n\,...,rim are not congruent and that indeed pr x fulfils n \ , . . . , n „ , but not n\,...,n'm. As a rule this will evoke another discussion about the question whether there are other norms that belong to both normative belief systems of A and Β and that at the same time would be incompatible either with n \ , . . . , n „ o r with n \ , . . . ,ri m .

On the basis of such a modification of the accepted norms, A and Β could either (case 2a) finally come to the same judgment about the rationality of prl, or (case 2b) still cling to different norms and therefore also insist on different judgments about the rationality of pr j. But even in this second case both will be in perfect agreement about one point: A and Β both know that not more than one of the two opinions, "pr j is rational" or "pr^ is irrational", can be true. Neither of them can come to the conclusion that both statements, "prx is rational" and "pr{ is irrational" can be true. If this (metatheoretical) exposition (based on the logics of language alone) is adequate, then the (metatheoretical) thesis that "A states: 'pr^ is rational'" is equivalent to "A accepts certain specific norms, pr j fulfils these norms and A expresses his feeling of acceptance of these norms" cannot be adequate. 3

3.

Rules, Dispositions and the Intrinsic Good

Rule following has no genuine place in traditional decision theory. If the question is about which rules of behaviour to establish, e.g. by legal sanctions which everybody who does not follow the rule is subjected to, then this is one kind of decision which certainly can be judged by the Bayesian criterion of maximizing expected utility. If rationality is understood as accepting certain rules (and following them), then a tension arises between decision theoretic rationality and this kind of rationality. One could object that following the 3

In his book, Gibbard tries to give an account of normative objectivity and especially the claims of consistency in normative judgments which are compatible with norm-expressivism. If these arguments succeed, then only a version of the error theory of normative objectivity would be left. T h e meanings of certain normative concepts would presuppose a mistaken theory of normative objectivity. Gibbard seems to try to avoid such a radical conclusion; cf. Mackie (1977), ch. 1, and the discussion in Honderich (1985).

Goodness and Rational Preferability: A Reply to Gibbard

263

Bayesian criterion is itself following a rule, and, since the Bayesian criterion is a rule that can be derived from a set of weaker rules, acting rationally in the decision theoretic sense is following these rules too. This is certainly right, but it is hardly conceivable that it is those rules only that Gibbard thinks of when he takes practical rationality to be constituted by accepting certain rules. Since neither the paper nor the book says anything about the possible contents of rules constituting practical rationality, it is difficult to decide whether these rules can be in accord with the Bayesian criterion traditionally understood. The rule of optimizing the consequences of ones acts (or decisions), i.e. to maximize expected utility, is (trivially) compatible with logically weaker rules like transitivity or connectivity, but the rule of optimizing the consequences of one's acts, which constitutes the core of traditional rational choice theory, is not compatible with almost every other normative rule of interpersonal cooperation and intrapersonal coordination, like pursuing certain personal projects, keeping promises, or cooperating in Prisoners Dilemma situations. Even if one integrates metapreferences4, it is impossible to reconcile cooperation and maximizing expected utility given the traditional interpretation. 5 The Axelrod results cannot be used as a counter-argument for two reasons: the assumptions under which cooperation in iterated Prisoner's Dilemmas is maximizing are quite unusual and rarely hold in everyday life;6 and - more importantly - even when all these assumptions are met, maximizing by cooperation is still parasitic upon conformity to basic normative rules7 which is not warranted by expected utility maximizing in the standard consequentialist interpretation. Things look different if one gives up the standard consequentialist interpretation of the Bayesian criterion and turns to an, as I call it, radically coherentist interpretation. 8 Then the utility function is nothing but a numerical representation of a coherent preference relation — coherence in the sense of Ramsey's (1931) axioms. A radically coherentist interpretation, though, entails fundamental changes to the ordinary interpretation of decision theoretic models. It is e.g. certainly not compatible with the interpretation Savage (1954) gave for his model of rational choice. A radically coherentist interpretation does not allow for interCf. Sen (1974), and in a broader context Frankfurt (1971), also George (1984), NidaRümelin (1991). I argued in Nida-Rümelin (1993) for the incompatibility of the Bayesian criterion in the standard consequentialist interpretation with following rules of intrapersonal cooperation and intrapersonal coordination. 6 E.g. an extremely low probability of the next interaction being the last one. ^ For example, rules are constitutive for communication. 8 There is not enough space here to spell this out; for a sketch see Nida-Rümelin (1993), §51. 4

264

Julian Nida-Rümelin

preting consequences in the ordinary sense. Gibbard's general theory of practical rationality fits well into a radically coherentist interpretation of rational choice. But a radically coherentist interpretation of decision theory constitutes a new, non-consequentialist paradigm of rational choice. To make this obvious, I mention only one startling implication: cooperation in genuine one-shot Prisoners Dilemmas would no longer be necessarily irrational, contrary to all standard interpretations of decision theoretic models. I think one cannot have both: an instrumental-consequentialist notion of rational choice and warranted preference on the one hand, and the constitution of practical rationality by certain accepted norms on the other. The norm-expressivist theory of practical rationality that Gibbard developed does not allow a return to the traditional theory of rational choice. Like Gauthier (1986), 9 Gibbard seems to use dispositions as a bridge between norm-constituted rationality and rational choice in the traditional sense. It is not clear whether the role of dispositions here is the same as in Gauthier's theory of morals by agreement. 10 But since in ordinary circumstances we cannot decide to adopt certain dispositions, I do not see how this bridge can ever be stable. Gauthier identifies practical rationality with utility maximization at the level of dispositions to choose, but in what sense does one choose dispositions? Dispositions become objects of choice only under special and rare conditions. But even if one concedes that dispositions can be chosen, I think this bridge is of no help because I hold it to be an essential attribute of a rational person that he is relatively free from dispositional determinants. The ideal rational agent is free to choose that singular action that has the best reasons in its favour. He is not forced to choose a rational disposition first and then to act in accordance with that disposition. Certainly, it is one criterion of rational choice that a single action is chosen only if it fits into a type of behaviour that meets certain rules. But if dispositions do not bind the rational agent, then the rational agent who acts in accordance with certain rules does deliberately not maximize — since the primary object of choice is the single action. If one allows for rule following to be intrinsically desirable in at least some cases of morally relevant rules, then the foundationalist account that takes the intrinsic goodness of certain states or processes as consequences of action as the starting point of practical rationality cannot survive. This tension between the intrinsic value of rule following (Aristotle's ένέργεια) and the instrumentalist

9 10

See also Gauthier/Sugden (1993). Gibbard (1990) is not explicit about this, either.

Goodness and Rational Preferability: A Reply to Gibbard

265

conception of practical rationality might become more evident if we look at Donna's point of view.

4.

Donna

Gibbard s theory of goodness contains three parts: (1) Intrinsic goodness as rational preferability common to every person, where the constituting preferences are not instrumental. The coherence axioms of decision theory are relevant only as a limiting condition. (2) The theory of rational belief, which accords with standard probability axioms. (3) Preferences concerning actions, where actions are instrumentally rational regarding (1) under the conditions regarding (2), whereas "instrumentally rational" is defined by the coherence axioms of rational decision theory. I suppose Donna will disagree with (3), which takes practical rationality as instrumental. Gibbard takes this into consideration (cf. sect. 6.6 of Gibbard's paper), but in a way I think Donna should not accept. Those preferences which Donna considers to be morally demanded, i.e. those that are in accordance with deontological side constraints, certainly violate the Bayesian criterion in the standard consequentialist interpretation. The reason is - as Gibbard himself notes — that she prefers, in some cases, an act that does not violate a moral rule, even if the consequences of this act are worse than that of another act that would violate the moral rule. But if we give up this consequentialist interpretation of the Bayesian criterion, then there is no reason to assume that Donna would violate the coherence axioms of rational choice. But if this is the case, we can attribute a numerical function over the set of alternatives which represents Donnas preference relation (the 'utility function of traditional decision theory). This works only if we take Donna to be a reasonable Kantian who is not a rigorist agent insofar as she takes consequentialist weakening of rule following into account 11 . The utility function representing Donnas preference relation, including her morally motivated preferences, though, is an unusual one. The domain of this function cannot be the set of states of affairs in the usual sense. The elements of the domain must inter alia contain information about which moral 11

By "consequentialist weakening" I mean the following: One can always imagine a situation in which following a moral rule would have negative consequences to such an extent that not to follow the rule is morally demanded. Donna, we might assume, is one of these reasonable Kantians who accept this kind of consequentialist weakening of deontological rules.

266

Julian Nida-Rümelin

rules would, or would not, be violated. To apply the monotonicity and continuity axiom to her preferences, we must include probabilities for violation and non-violation of specific moral rules. The utility function constituted in such a way would be a measure for Donnas account of the good, including moral and extramoral aspects. The differentiation between the intrinsic and the instrumental good would be obsolete. Certainly Donna's behaviour could not be described anymore as instrumentally rational in the usual sense, since the type to which various actions belong is relevant when she chooses among them. But if actions are chosen because of their specific characteristic qualities (i.e. types they belong to), this cannot, as far as I can see, be brought into accordance with Gibbard s notion of instrumental rationality. So it seems that Gibbard cannot have both: Characterization of practical rationality as instrumental by means of decision theory on the one hand and the universal validity of the coherence axioms of decision theory on the other hand. Either Donna is right and practical rationality is not (always) instrumental and Gibbard cannot criticise her for not being in line with rational decision theory, or Gibbard is right that practical rationality is (always) instrumental and he cannot prove it by the intuitive plausibility of the coherence axioms of rational decision theory. Obviously, then, it seems to be necessary to make a clear distinction between coherence of individual preferences on the one hand, and a theory of rational motivation on the other, telling that actions should be chosen with regard to their factual (or probabilistic) consequences for the state of the world. The latter might imply the former, but certainly not the other way round. Donna, as a reasonable Kantian, will fulfil the coherence axioms, but she will not be prepared to subordinate her choice of actions to the criteria of instrumental rationality. The utility function that can be ascribed to Donna on the grounds that she fulfils the coherence axioms of decision theory is (according to the concept of revealed preferences) nothing else than the quantitative representation of her preferences regarding actions. But to conclude that Donna chooses her actions in order to maximize the value of this function is a non sequitur. Donna chooses her actions partly to conform with certain moral rules, and partly (within the realm of a morally indifferent sphere not restricted by deontological rules) with regard to 'pragmatic imperatives', and in some borderline cases she has to decide whether to choose between respecting pragmatic or deontological aspects. The quantitative representation thus does not yield any theory of rational motivation. A fortiori it is not true that due to her conformity to the coherence axioms of decision theory Donna has turned into a consequentialist agent. This is implied by the fact that the range of the utility function attributed to Donna does not comprise states of the world in

Goodness and Rational Preferability: A Reply to Gibbard

267

the usual sense. The elements of the range are determined irreducibly by parameters like "Donna obeys a certain specific deontological rule".

5.

Conclusion

An adequate theory of practical rationality must contain at least two elements. The first deals with the question which action is right regarding a given structure of interaction and given goals the agents are aiming at. The second deals with the question of reasonable goals. The first of these elements cannot be exhausted in an instrumental-consequentialist conception, since an adequate definition of what is practically reasonable requires norms of cooperation of different degrees of generality. What rational human agents prefer is not independent of the cooperation norms they accept: therefore, the intrinsic good is determined by deontological restrictions, too. The good has neither a logical nor a motivational primacy over the right, but the moral right is not independent from the intrinsic good either. Gibbard meets this conception half way by interpreting preferences as expressing accepted norms, but he does not so all the way. This may seem justified because Gibbard tried to concentrate on metatheoretical questions. But if there is no adequate first order theory of practical rationality which accords with this second order account, then there remains a problem to be solved.

References ARISTOTLE

(NE). Aristotle: Nicomachean Ethics·, references are to Bekker's academy edition, Berlin 1831. F R A N K F U R T ( 1 9 7 1 ) . Harry G. Frankfurt: "Freedom of the Will and the Concept of a Person", Journal of Phibsophy 68 (1971). G A U T H I E R ( 1 9 8 6 ) . David Gauthier: Morals by Agreement, Oxford 1986. G A U T H I E R / S U G D E N ( 1 9 9 3 ) . David Gauthier and Robert Sugden (eds.): Rationality, Justice and the Social Contract, New York 1993. G E O R G E (1984). David George: "Metapreferences in Reconsidering Contemporary Notions of Free Choice", Journal of Social Economics 11 (1984). G I B B A R D ( 1 9 9 0 ) . Allan Gibbard: Wise Choices, Apt Feelings., Cambridge, Mass., 1990. H O N D E R I C H ( 1 9 8 5 ) . Ted Honderich (ed.): Morality and Objectivity, London 1985. K Ö R N E R ( 1 9 7 4 ) . Stephan Körner: Practical Reason, Oxford 1974. M A C K I E ( 1 9 7 7 ) . John L. Mackie: Ethics, Harmondsworth 1977. N I D A - R Ü M E L I N ( 1 9 9 1 ) . Julian Nida-Rümelin: "Practical Reason or Metapreferences? An Undogmatic Defense of Kantian Morality", Theory and Decision 30 (1991).

268

Julian Nida-Rümelin

(1993). Julian Nida-Rümelin: Kritik des Konsequentialismus, second ed., Munich 1995; first ed. first publ. in 1993. R A M S E Y (1931). Frank P. Ramsey: The Foundations of Mathematics and Other Logical Essays, second ed., London 1954; first publ. in 1931. S A V A G E (1954). Leonard J. Savage: The Foundations of Statistics, second ed., New York 1972; first publ. in 1954. SEN (1974). Amartya Sen: "Choice Ordering and Morality", in Körner (1974).

NIDA-RÜMELIN

Part III Preference and Ethics

JOHN

BROOME

Extended Preferences* Abstract: Ordinalism is generally taken to imply that interpersonal comparisons o f good are impossible. But some ordinalists have argued that these comparisons can be made in a way that is consistent with ordinalism, on the basis o f extended preferences. This paper shows that this argument is mistaken, and ordinalism is indeed incompatible with interpersonal comparisons o f good. 1. 2.

Introduction Ordinalism

3. 4. 5. 6. 7. 8.

Extended Preferences and Interpersonal Comparisons o f Good Harsanyi's Causal Argument T h e Causal Determination of Extended Preferences T h e Causal Determination o f Good Ordinalism and Goodness for a Person From Individual Orderings to Interpersonal Orderings? References

1.

Introduction

Many economists have adopted a doctrine known as "ordinalism". Ordinalism insists that we can only know about people s good by means of our knowledge of peoples preferences. Many ordinalists believe their doctrine implies we cannot know how one persons good compares with another's. But some have resisted this conclusion; they have argued that we can make interpersonal comparisons of good in a way that is consistent with ordinalism. They think a particular class of preferences can constitute a basis for comparisons between the good of different people. They have in mind peoples preferences between very widely defined alternatives, each of which consists of a way of life together with the personal characteristics of a person who lives that life. These are called "extended preferences". This paper reveals a flaw in the argument that extended preferences can be a basis for interpersonal comparisons of good. It shows that interpersonal comparisons really are inconsistent with ordinalism. * I have greatly benefitted from discussions and correspondence on the subject o f this paper with John Harsanyi, Susan Hurley, Serge-Christophe Kolm, Brian Skyrms, and, particularly, Hans-Peter Weikard. The research for this paper was funded by the Economic and Social Research Council under grant R 0 0 0 23 3334.

272

John Broome

Section 2 of this paper explains ordinalism in more detail. Section 3 describes the notion of extended preferences, and specifies a condition extended preferences must satisfy if they are to serve the purpose they are meant for: everyone must have the same extended preferences. Section 4 quotes an argument of John Harsanyi s that is intended to show everyone will indeed have the same extended preferences. Section 5 explains that Harsanyi s argument contains an error. There is an alternative approach to interpersonal comparisons implicit in some ordinalist writings, including Harsanyi's. This second approach is not usually clearly distinguished from the approach through extended preferences, but actually it does not depend on extended preferences. Section 6 describes it. Section 7, however, shows it is inconsistent with ordinalism. Section 8 argues that this approach is anyway otiose. My aim in this paper is only to prove that interpersonal comparisons of good are inconsistent with ordinalism. What conclusion should we draw? I draw the conclusion that ordinalism must be incorrect, since interpersonal comparisons of good are clearly possible. But in this paper I shall not try to prove as much as that.

2.

Ordinalism

I shall take ordinalism to be the conjunction of these three claims: 1. The preference-satisfaction theory of good. If a person has a high-grade preference for one alternative over another, then it would be better for the person to have the first alternative rather than the second. 2. Given a pair of alternatives, we can (sometimes at least) know whether a person has a high-grade preference for one of them over the other. 3. Our knowledge of people's high-grade preferences is the only way we can come to know anything about how good it would be for a person to have some alternative. By a "high-grade preference" I mean a preference that passes some test of quality: it is rational and well-informed, or something of that sort. It does not matter for my purposes precisely what the test is, though in section 7 I shall mention something it cannot be. Some ordinalists may not insist on any test at all; for them, any preference is high-grade. From now on in this paper, I shall only speak of high-grade preferences, and the word "preference" is always to be understood as referring to a high-grade preference.

Extended Preferences

273

I am treating ordinalism as a theory about what we can know, and not as a theory about meaning. Many authors take ordinalism to include some verificationist view about meaning: perhaps the view that a statement about peoples good has no meaning unless we can know whether or not it is true. But this paper is not concerned with meaning, and I shall leave verificationism aside.

3.

Extended Preferences and Interpersonal Comparisons of Good

One conclusion that has often been drawn from ordinalism is that we cannot compare one person's good with another's. The argument is this. One person i's preferences will tell us whether one alternative is better for i than another. Another person j s preferences will tell us whether one alternative is better for j than another. But no one's preferences will tell us whether one alternative is better for i than another alternative is for j. Not all ordinalists have accepted this conclusion. Some have argued that preferences can indeed allow us to compare one person's good with another's. It depends on what sorts of alternatives the preferences are amongst. Preferences amongst cheeses will not do it, but these ordinalists say extended preferences, as they call them, will. An extended preference is a preference between extended alternatives, and an extended alternative consists of a way of life paired together with particular personal characteristics. For instance, one extended alternative is to live the life of an academic whilst possessing a thirst for knowledge and modest material needs. Another is to live the life of an academic whilst possessing an insatiable desire for wealth. No doubt we have preferences amongst such alternatives; I prefer the first of the two I have just described to the second. These are extended preferences. It has been argued that extended preferences can give us grounds for interpersonal comparisons of good (for instance, in Arrow 1977 and Harsanyi 1977, pp. 57-60). How would this work? Suppose I have an extended preference for living person i's life with i's personal characteristics over living j's life with j's personal characteristics. According to the preference-satisfaction theory, it would be better for me to lead i's life with i's personal characteristics than to lead j's life with j's personal characteristics. This suggests that /, living her life, is better off than j, living hers. In this way an extended preference seems to allow a comparison between one persons good and another's. In this case my preference seems to have allowed a comparison between i's good and j's. This is not a convincing argument without some extra support. The fact that John Broome has an extended preference for i's life and characteristics

274

John Broome

over j's life and characteristics cannot be sufficient evidence that i is better off than j . What if someone else had the opposite extended preference to mine? Even if i herself has this same extended preference, that would not be sufficient evidence; what if j had the opposite one? Evidently, if extended preferences are to be the basis for interpersonal comparisons of good, the extended preferences of different people must coincide to some extent at least. Those authors who rely on extended preferences as a basis for interpersonal comparisons of good generally insist that everyone will have the same extended preferences, provided extended alternatives are construed widely enough. 1 We must make sure that our extended alternatives are a full specification of all those features of lives and personal characteristics that anyone can possibly have preferences about. Given that, these authors claim we must all have the same preferences about such very widely extended alternatives. Why should this be? It certainly seems implausible. I myself prefer to live the life of an academic, with my own academic characteristics, even in the conditions allotted to academics in contemporary Britain, to being a financial analyst with the characteristics of a financial analyst living in the conditions allotted to financial analysts. A financial analyst, with her different values, would no doubt have the opposite preference. So her extended preferences will differ from mine. Both our preferences may well be high-grade: they may be rational, well-informed and so on. The reason I have mine is that an academic has some slight chance of making a worthwhile contribution to knowledge. I recognize that, if I were a financial analyst, with all the characteristics of a financial analyst, I would not then value knowledge as I do now. Nevertheless, I do value knowledge, and that is why I prefer to be an academic. The different values of academics and financial analysts lead us to have different extended preferences. Or so it certainly seems. Appearances, then, are against the claim that different people must have the same extended preferences. Are there, on the other hand, any arguments in favour of this claim? I know of one. It is based on the causes of preference. It is spelt out in most detail in Harsanyi (1977), pp. 57-60, but other authors have used it too.2

1 2

See the references in note 2. The argument appears in rudimentary form in Harsanyi (1955), pp. 17 f. in the reprinted version, and independently in Tinbergen (1957), p. 501. It also appears in Kolm (1972), pp. 7 9 f.

Extended Preferences

4.

Harsanyi's Causal

275

Argument

Harsanyi s argument is difficult to interpret, and I do not want to misrepresent it. To reduce the risk of unfairness, I shall quote almost all of it in this section, before going on to explain and criticize it in section 5. Harsanyi (1977), pp. 58 f., says: "If all individuals' personal preferences were identical, then we could ascribe the same utility function U to all individuals and could always make interpersonal utility comparisons in terms of this common utility function U. Moreover, all zwterpersonal utility comparisons could be reduced to inrrapersonal utility comparisons. If we wanted to know whether a given apple would give more utility to Peter (who has just had a heavy meal) than to Paul (who had only a very light meal), we could simply ask whether Peter himself would derive more utility from an apple after a heavy meal or after a light meal. Of course, in actuality different individuals often have very different personal preferences and very different utility functions. But the possibility of meaningful interpersonal utility comparisons will remain, as long as the different individuals' choice behaviour and preferences are at least governed by the same basic psychological laws. For in this case each individual s preferences will be determined by the same general causal variables. Thus the differences we can observe between different people's preferences can be predicted, at least in principle, from differences in these causal variables, such as differences in their biological inheritance, in their past life histories, and in their current environmental conditions. This means that if Peter had Paul's biological makeup, had Paul's life history behind him, and were currently subject to Paul's environmental influences, then he would presumably have the same personal preferences as Paul has now and would ascribe the same utility as Paul does now to each particular situation. Let Pj again denote individual j s subjective attitudes (including his preferences), and let Rj denote a vector consisting of all objective causal variables needed to explain these subjective attitudes denoted by Pj. Our discussion suggests that the extended utility function Vi of each individual i should really be written as V/ = Vi[Aj,Rj] rather than Vj[Aj,Pj). Written in this form, the utility function Vi = V^Aj.Rj] indicates the utility that individual i would assign to the objective position Aj if the causal variables determining his preferences were Rj. Because the mathematical form of this function is defined by the basic psychological laws governing people's choice behaviour, this function Vi must be the same for all individuals i, so that, for example,

Vh[Aj,Rj] = Vi[Aj,Rj}

276

John Broome for each pair of individuals h and i. In the special case in which h = j, we can write ViiA^Rj] = V^Rj)

=

Uj(Rj)

That is, individual i (or individual h) would have the same preferences and would assign the same utility to any objective situation Aj as individual j now does, if the causal variables determining his preferences took the same value Rj as do the causal variables determining j's preferences. In other words, even though the ordinary' utility functions Ui and Uj of two individuals i and j may be quite different, their extended utility functions V{ and Vj will be identical. This is so because, by the definition of the causalvariables vectors Rj and Rj, all differences between the two functions Uj(Aj) — Vi[Ai,Ri] and Uj(Aj) — Vj[Aj,Rj\ must be attributed to differences between the vectors /?, and Rj and not to differences between the mathematical form of the two functions Vj and Vj. Yet, if the two individuals have the same extended utility function V,· = Vj = V, then we are back in a world of identical utility functions. Hence individual i will be able in principle to reduce any /«impersonal utility comparisons that he may wish to make between himself and individual j to an intozpersonal utility comparison between the utilities that he is in fact assigning to various situations and the utilities he would assign to them if the vector of causal variables determining his preferences took the value Rj (which is the value that the vector of these causal variables takes in the case of individual j). For example, if I want to compare the utility that I would derive from a new car with the utility that a friend would derive from a new sailboat, then I must ask myself what utility I would derive from a sailboat if I had taken up sailing for a regular hobby as my friend has done, and if I could suddenly acquire my friend's expert sailing skill, and so forth."

5.

The Causal Determination of Extended Preferences

Harsanyi s argument relies on the notion of a utility function, and cannot be expressed without it. So I need first to explain this notion. If a person has preferences over some range of alternatives, and if these preferences conform to a number of conditions, it is possible to represent them by a function U, which is called a utility function. U assigns a value called a utility to each of the alternatives. To say U represents the preferences means that U assigns higher utilities to alternatives that are preferred. More precisely: the utility assigned by U to one alternative is at least as great as the utility assigned by U to another if and only if the first alternative is preferred or indifferent to the

Extended Preferences

277

second. This is how Harsanyi (1977), pp. 31 f., explicitly defines "utility". Some of his expressions in the passage I have quoted suggest he may also have in mind another meaning for "utility", and I shall come to that in section 6. But for the moment I shall stick with this one: utility is the value of a function that represents preferences. Harsanyi calls a utility function that represents extended preferences an extended utility function, and I shall do the same. One technical point about utility functions. There is not just one utility function that represents a persons preferences; many do. If a function U represents someone's preferences, then another function U' will also represent her preferences if and only if U' is an increasing transform of U. U' is defined to be an increasing transform of U if and only if, for any pair of alternatives A and B, U'(A) > U'{B) if and only if U(A) > U(B). For instance, if U' is the function ^ , obtained by dividing U by some positive constant a, then U' is an increasing transform of U, and it will represent the same preferences as U. A person j has preferences about ways of life, which may be represented by a utility function UJ. Different people have different preferences, and / will have preferences represented by a different function Ui. But there is a causal explanation of why each person has the preferences she has. Let R stand for the causal variables that determine the form of a persons preferences (her upbringing, friends, sporting ability and so on). A person who is subject to causal influences R will have preferences that can be represented by a utility function £/#. Any person subject to the same influences R will have the same preferences, so there is no need to index the function UR by the name of the person whose utility function it is. This function assigns a utility UR(A) to ways of life A. An alternative notation is U (.A;R). I separate A and R by a semicolon rather than a comma for a reason that will appear. The function U is a universal function, the same for everybody. It represents the preferences that a person subject to causal influences R will have about ways of life A. It certainly does not represent preferences of any sort, belonging to anybody, about the combination (A, R) of a way of life and causal influences. I hope this is obvious, but an example may help to make it more so. Suppose people have preferences about opportunities to contribute to knowledge, /, and money, m. Suppose a persons preferences can be represented by a utility function that has this particular form: Ua(l, m) = alog/ + (1 — a) logm. Not everyone has exactly the same preferences, because the parameter α (which lies between 0 and 1) differs from person to person. There is a causal explan-

278

John Broome

ation of why a person has the particular value of α she has. Let us suppose her α is determined by the age when she was weaned. In fact, let α be equal to w, the age at which the person is weaned (everyone is weaned before one). A persons utility function Uw(l, m), or in an alternative notation U(/, m\ w), is: U(l,m;w)

- wlog/ + (1 - w)\ogm.

This function represents the preferences that a person weaned at w has about / and m. It does not represent anyone's preferences about w. It may be that no one even has any preferences about when she is weaned. The semicolon separates the objects of preference / and m from the cause of preference w. Objects of preference and causes of preference have quite different roles in the utility function. There is a formal way of demonstrating this point. Let m) = log/ +

-

logm.

U'a is obtained by dividing Ua by a, which, as a parameter of the function, is a constant. U^ is therefore an increasing transform of Ua. So U^ represents the preferences of a person with parameter a, just as Ua does. But, as a causal matter, α is equal to the person's age of weaning w. So the preferences about I and m of someone weaned at w can be represented by the function U'(l, m\ w) — log/ + ^

logm.

U'(l,m;w) is an increasing transform of U(l,m;w) if the arguments of the function are taken as I and m. But if /, m and w are all taken as arguments, then U'{1, m\ w) is not an increasing transform of U(/, m\ w). Consequently, these two functions cannot represent the same preferences over the three variables /, m and w together. This shows they are not representing preferences over these three variables at all. Return to the general case. U (.A; R) as I defined it is a universal function, the same for everyone. But it is not a universal utility function representing preferences about A and R together. In the argument I quoted, Harsanyi calls Vi[Aj,Rj\, which is my U(A\R) expressed in his notation, an extended utility function. He implies that it represents extended preferences over ways of life A and causal variables R together. But it does not. Harsanyi hoped to exhibit universal extended preferences, the same for everyone. But he fails to do so. There is a complication. The things that have a causal influence over people's preferences may also be things that people have preferences about.

Extended Preferences

279

For instance, people have preferences about the friends they have, and their friends influence their preferences. Harsanyis separation between 'objective position' A and causal variables R is too sharp. I must reformulate my point to take this complication into account. Let A be a variable that stands for ways of life. Let Ρ stand for personal characteristics. Then (A,P) is an extended alternative as I originally defined it: a way of life lived with particular personal characteristics. People have extended preferences about these extended alternatives. Let us make sure that A and Ρ are defined broadly enough to include anything that anyone has a preference about. (It does not matter what is included in A and what in P, so long as everything comes into one or the other.) A person j has extended preferences about (A, Ρ) that can be represented by an extended utility function Vj{A,P). I have given Vj the index j because as yet we have no reason to think j s extended preferences are necessarily the same as anyone else's. Now, there are causal variables R that determine peoples extended preferences. A person who is subject to causal variables R will have extended preferences VR{A,P). An alternative notation is V(A,P\R). Since anyone subject to the same causal variables will have the same extended preferences, there is no need to index this function by j. V is a universal function. Many of the variables in R will be things that people have preferences about. So they will also appear in A or P. Indeed, let us now go back and enlarge A and Ρ to make sure that they include, not only anything that anyone has a preference about, but also anything that has any causal influence on people's preferences. (A, P) is now a very comprehensive specification of a life and a person living that life. In V(A,P\R) all the variables contained in R will now also appear in A or P. Still, R is not redundant as an argument of the function. Although all the variables in R will also appear in A or P, it will be different values of the variables in each case. R contains the values of the causal variables that a person is actually subject to. {A,P), on the other hand, contains the values of the variables that the person contemplates as objects of her preference. V(A,P;R) is a universal function. But it does not represent universal extended preferences. It does not represent any preferences at all over A, Ρ and R taken together. It represents preferences over A and P, but many different preferences, one set of preferences for each value of R. So, even after taking account of the complication, we have not found universal extended preferences. Serge-Christophe Kolm (1972), pp. 79 f., says: "At bottom, all individuals have the same tastes, the same desires. Without doubt, this assertion requires explanation. If two persons have preferences which appear to differ, there is a reason for

280

John Broome

this, there is something which makes them different from each other. Let us place this 'something' within the object of the preferences which we are considering, thereby removing it from the parameters which determine the structure of these preferences. T h e preferences of these two persons defined in this way are necessarily identical. We may carry out this operation in the case of any society: namely, the operation of placing in the object of preferences everything which would cause differences between the preferences of different members of society. An identical preference of all members of this society obtained in this way is called a fundamental preference' of the members o f this society. It is a property which describes the tastes and needs o f the 'representative individual' o f this society. If this society includes all human beings, then that which discerns this c o m m o n preference is at bottom 'human nature'." 3

Kolm evidently has an idea like Harsanyi s. The causes of people s preferences may indeed also be objects of their preferences. But by including causes amongst objects of preference, we do not stop them from being causes. We do not remove them from the parameters which determine the structure of peoples preferences. They retain their role as causes, and they may cause one person's preferences to be different from another's. My preferences are influenced causally by the life I lead. Since I am an academic, I have preferences that differ from the preferences of a financial analyst. The life I lead is also amongst the objects of my preferences: I prefer the life of an academic to the life of a financial analyst. But simply recognizing that the cause of my preferences is also an object of my preference is not going to make my preferences identical to a financial analyst's. No doubt the financial analyst has the opposite preference to mine. It is just not true that, at bottom, all individuals have the same tastes and desires. Kolm seems to think that, by treating the causes that act upon me as objects of my preference, I can somehow withdraw myself from their influence. But I cannot escape from my own causal situation. The causal argument was offered as a demonstration that everyone will have the same extended preferences. It fails.

6.

The Causal Determination

of Good

What then, is the attraction of the causal argument? Why have so many authors put forward an argument that is so plainly mistaken? I suspect they have 3

I take this translation from Rawls (1982), p. 174.

Extended Preferences

281

confused it with another argument that is not mistaken. They have been looking for grounds for interpersonal comparisons of good. And there is a truth about causation that is germane to interpersonal comparisons of good. It is true, but not germane, that if two people were in the same causal situation they would have the same preferences. It is also true, and germane, that if two people were in the same causal situation, they would be equally well off. This latter truth might possibly give us a route to interpersonal comparisons of good. It would allow us to move from comparisons of good for a single individual to comparisons of good between individuals. Suppose we are able to determine, for a single person i and for any pair of alternatives A and B, whether A or Β is better for i, or whether the two are equally good for her. Suppose, that is to say, we can put all the alternatives in order according to their goodness for i. Now suppose the alternatives we are dealing with are very broadly defined. They are extended alternatives, including both ways of life and personal characteristics, and also including all the causal variables that could have an influence on how well off a person is. One of these alternatives could not possibly be any better or worse for i than it would be for anyone else. Take two of these extended alternatives, Ε and F . Would Ε be better for one person, say j, than F would be for a different person, say ki This would be so if and only if Ε would be better for i than F would be for i. It would be so, that is to say, if and only if Ε comes higher than F when the alternatives are ordered by their goodness for the single person i. When we order the alternatives by their goodness for i, therefore, we are also producing an interpersonal ordering. Provided our alternatives are defined sufficiently widely, an ordering by goodness for a single person is also an interpersonal ordering. Simply, it is an ordering of the alternatives by their goodness. If I were a financial analyst living the life of a financial analyst, subject to all the causal influences that determine how well off a financial analyst is, then I should be exactly as well off as anyone else would be if she occupied that position. The same is true for the alternative of being an academic in the causal situation of an academic. Therefore, if it is better for me to live the life of an academic, it would be better for anyone. The life of an academic would, simply, be better. So, if we can discover an ordering of widely extended alternatives by their goodness for any single person, we can make interpersonal comparisons of good: we can know whether one person is better off than another. This conclusion is only conditional. If we can discover an ordering for a single person, we can make interpersonal comparisons. How we can discover an ordering for a single person is another matter, which I shall come back to in sections 7 and 8.

282

John Broome

I suspect this possible route to interpersonal comparisons has been in the minds of several of the economists who have more explicitly followed the route of extended preferences. It is difficult to be sure, because of the ambiguous use of the word "utility" that is common in economics. (See Broome 1991.) I defined "utility" in section 5 as the value of a function that represents preferences. This is the definition most often made explicit. But economists also often speak of a persons "utility" when they mean the person's good. In his discussion of "extended sympathy" as a basis for interpersonal comparisons of good, Kenneth Arrow (1977), p. 224, says: "We may suppose that everything which determines an individual s satisfaction is included in the list of goods. Thus, not only the wine but the ability to enjoy and discriminate are included among goods. [ . . . ] If we use this complete list, then everyone should have the same utility function for what he gets out of the social state."

What does Arrow mean when he says everyone should have the same utility function? If he is thinking of a utility function as representing a person's preferences, he must mean that everyone has the same preferences - in this case the same extended preferences. As I say, that is false. If, on the other hand, he is thinking of a utility function as measuring a person's good, he may mean that each extended alternative is as good for one person as it is for another. As I say, that is true, and leads to a different route to interpersonal comparisons of good. I am inclined to think Arrow means to follow this second route, but I am not perfectly sure. The ambiguity of "utility" is acute in Harsanyi's writings. It is the main reason why the argument I quoted in section 4 is so hard to interpret. When Harsanyi says "if Peter had Paul's biological makeup, had Paul's life history behind him, and were currently subject to Paul's environmental influences, then he would presumably have the same personal preferences as Paul has now and would ascribe the same utility as Paul does now to each particular situation", I think he is probably making two quite different points. The first is that if Peter had the same biological makeup and so on as Paul, he would have the same preferences as Paul. The second is that if Peter had the same biological makeup and so on as Paul, then any particular situation would be exactly as good for Peter as it would be for Paul. The first of these points led Harsanyi to the argument that is most explicit in the passage I quoted: the argument that everyone must have the same extended preferences. This argument is mistaken, as I explained in section 5. The second point is submerged through most of the passage but surfaces at the end, when Harsanyi wants to compare the benefit a friend would get from a

Extended Preferences

283

new sailboat with the benefit he himself would get from a new car. He says, "I must ask myself what utility I would derive from a sailboat if I had taken up sailing for a regular hobby as my friend has done." Harsanyi plans to ask himself, not what preferences he would have if he were in his friend's causal situation, but how well off he would be if he were in his friend's causal situation and had bought a new boat. He knows that if he, Harsanyi, were in his friend's causal situation and had bought a new boat, he would be exactly as well off as his friend would be if he had bought a new boat. So, on this occasion, the route he is taking to interpersonal comparisons is not via extended preferences. He is taking the alternative route I have described in this section.

7.

Ordinalism and Goodness for a Person

It is not surprising that this second route to interpersonal comparisons is not made very explicit in the writings of ordinalists. It is inconsistent with ordinalism. The route goes like this. First, we must have extended alternatives ordered by their goodness for one person. We know that an extended alternative is equally as good for one person as it is for any other person. So the ordering for the single person will also be an interpersonal ordering. But how do we find an ordering for the single person to start with? The answer cannot be consistent with ordinalism. According to ordinalism, we should have to find the single person's ordering through her preferences. The ordering we are interested in is the ordering of extended alternatives by their goodness for the person. We could find it through the person's extended preferences if the preference-satisfaction theory of good (see sect. 2) were true. This theory implies that the ordering of extended alternatives by their goodness for a person coincides with the person's extended preference ordering. But by now I am able to say with confidence that the preference-satisfaction theory is false, at least when applied to extended preferences. If it were true, one persons extended preference ordering would coincide with the ordering of the alternatives by their goodness for the person. That is to say, it would coincide with their ordering by, simply, their goodness. Consequently, it would coincide with everyone's extended preference ordering. So the preference-satisfaction theory implies that everyone has the same extended preferences. But people do not all have the same extended preferences. Therefore, the preference-satisfaction theory is false. If follows that we cannot use a person's extended preferences to find an ordering of extended alternatives by their goodness for that person. We cannot find this ordering in a way consistent with ordinalism.

284

John Broome

The argument in the previous paragraph depends on my assertion that people do not all have the same extended preferences. Up to section 5 , 1 took seriously the claim that everyone necessarily has the same extended preferences, despite the counterexample of the academic and the financial analyst. But now, after the collapse of the causal argument, I need no longer take this claim seriously. I am happy to use the clear fact that extended preferences differ — for instance, that different people have different preferences between the life of an academic and the life of a financial analyst — as evidence against the preference-satisfaction theory of good. The preferences of all of us are determined by our causal situation, and in the world as it is, our preferences are caused to differ. I come out preferring to be an academic, and an analyst comes out preferring to be an analyst. To be sure, the preference-satisfaction theory is concerned only with highgrade preferences (see sect. 2). If either my preferences or the analysts were not high-grade, our differing preferences would not confute the preferencesatisfaction theory. But both our preferences might be high-grade by any reasonable test. We might have thought long and hard about the relative merits of the two alternatives, scratching our ears, applying our best mental powers and using all the information available. The difference between us is our values: our different causal situations cause us to have different values. But that is no reason to think one of us irrational or ill-informed. It might be that one of the alternatives that face us is better than the other as a matter of objective fact. I doubt it, but suppose it is for a moment. Then either I or the analyst has got it wrong: one of us prefers an alternative that is objectively worse. Let it be the analyst. Could we say the analyst's preference fails to be high-grade on that account? Could our test of quality for preferences be that they must be in line with objective goodness? That would be inconsistent with the epistemology of ordinalism. It would imply that we could not know whether a preference between two alternatives was high-grade unless we first knew which of the alternatives was better. According to ordinalism, however, we could know nothing about the goodness of any alternative unless we already knew some preferences and knew that those preferences were high-grade. There is no way to break into this circle. So, even if one of our preferences is objectively wrong, it cannot follow that it is not a high-grade preference. A high-grade preference cannot be defined in such a demanding way. We have to conclude that the preference-satisfaction theory of good is false when applied to extended preferences. This is no serious problem for ordinalism itself; ordinalists can perfectly well decline to apply ordinalism to the arcane domain of extended preferences. But it does mean that there is no

Extended Preferences

285

route to interpersonal comparisons of good that is consistent with ordinalism. Granted their own assumptions, those ordinalist who deny the possibility of interpersonal comparisons are right.

8.

From Individual Orderings to Interpersonal Orderings?

In section 6,1 suggested we might be able to move from a ordering of extended preferences by their goodness for a single person to an interpersonal ordering. In section 7 , 1 explained that this idea is inconsistent with ordinalism because we could not derive a single person's ordering from preferences. Is there any other way we could find out a single persons ordering, in order to discover an interpersonal one? What must we do to order extended alternatives by their goodness for a person? The task has two parts. First, we must do some theoretical work in ethics to discover what a person's good consists in. Then we must do some empirical work to see how much of this good is delivered by each particular alternative. The nature of the empirical research we shall need to do depends on what conclusion we have reached at the first stage, in deciding what a person's good consists in. For instance, suppose we conclude that ethical hedonism is correct, and a persons good consists in having good feelings. Then we shall have to do some psychological research into how good are the feelings that different ways of life produce. One way of proceeding would be to take a subject, put her into various causal situations, and ask her which ones make her feel better or worse. This would give us an ordering of the alternative situations by their goodness for the subject. However, the types of alternative we are interested in are whole lives together with the personal characteristics of the people living them. No experiment would allow us to put one subject into a range of alternative situations of this sort. Instead, our empirical research will require us to compare the feelings of a person living one life with the feelings of a different person living another life, to see which feelings are better. But many people - most ordinalists amongst them - are sceptical about the possibility of comparing the feelings of different people. If they are hedonists, this will lead them to doubt the possibility of interpersonal comparisons of good. However, they might think they can overcome the difficulty by undertaking an imaginary experiment rather than a real one. Instead of causing an actual subject to lead various lives, they would try to imagine themselves leading various lives, and see how they feel when they do. That way, they could find out which lives would be better or worse for themselves: they could order lives by their goodness for

286

John Broome

themselves. From this personal ordering, they could move to an interpersonal one. So this is one putative way of finding an order for a person and deriving an interpersonal one from it. But this procedure is dubious for at least two reasons. First, the imaginary experiment itself is dubious. You are supposed to imagine yourself leading a different life from your actual life, with different personal characteristics, and find out how good you feel by observing yourself. However, if the characteristics are very different from your actual characteristics, it would actually be impossible (not just causally impossible, but even metaphysically impossible) for you to have those characteristics, because anyone with those characteristics would not be you. So the situation you are supposed to imagine in this experiment is impossible. Just how you are supposed to conduct this imagined impossible experiment is obscure. Probably the best you could do is imagine a person, not particularly yourself, leading the life and possessing the characteristics you are interested in. You could estimate how good that person would feel. Whatever ordering by goodness you derive from this estimation would not particularly be a personal ordering by goodness for yourself. Yet that is what you are supposed to be finding. The second reason why the procedure is dubious is that it assumes the dubious doctrine of ethical hedonism. I doubt that hedonism would be the right conclusion to draw from our ethical investigation into what a persons good consists in. I think it is more likely that a persons good consists in a number of more traditional good things: health, having friends, material comfort, varied and interesting experiences, and so on. The most difficult part of our work is likely to be the theoretical job of deciding what these goods are, and how they weigh against each other. After that, we shall come to the empirical investigation into how much of the goods each life delivers to the person who leads that life. But if the goods turn out to be overt things like the ones I mentioned, this may not be so difficult. Hedonists may find the empirical task difficult because they believe the good things in life are feelings, which are often supposed to be covert. But overt goods are more easily detected. I should like to offer two particular speculations about the conclusions we are likely to come to. I think one conclusion is likely to be that there is no objective fact as to whether the life of an academic is better or worse than the life of a financial analyst. Different goods, such as wealth and opportunities to contribute to knowledge, are surely incommensurable to some extent, from an objective point of view. Consequently, there is room for the analyst and me to have different extended preferences, without either of our preferences being open to objective criticism.

Extended Preferences

287

A second conclusion is likely to be that nothing can be gained in our research by starting from the good of an individual and then moving to interpersonal comparisons. The question is simply which lives are better than others, and nothing can be gained by asking first which would be better than others for a particular individual. Indeed, this is a peculiar question in the first place.4 Since it would not have been possible for me to live a life far different from my actual life, it is peculiar to ask how good such a life would be specifically for me. All we can sensibly ask is, simply, how good the life would be. Interpersonal comparisons are unlikely to be a real problem, because our investigation of good is likely to be interpersonal from the start. Section 6 mentioned a possible route to interpersonal comparisons through the goodness of lives for individuals; I am now suggesting this route is redundant.

References ARROW

( 1 9 7 7 ) . Kenneth J. Arrow: "Extended Sympathy and the Possibility of Social

Choice", American Economic Review: Papers and Proceedings 6 7 ( 1 9 7 7 ) . B R O O M E ( 1 9 9 1 ) . John Broome: " 'Utility'Economics and Philosophy 7 (1991). H A R S A N Y I ( 1 9 5 5 ) . John C. Harsanyi: "Cardinal Welfare, Individualistic Ethics, and Interpersonal Comparisons of Utility", in Harsanyi (1976); article first publ. in 1 9 5 5 . H A R S A N Y I ( 1 9 7 6 ) . John C. Harsanyi: Essays on Ethics, Social Behavior, and Scientific

Explanation, Dordrecht 1976. ( 1 9 7 7 ) . John C. Harsanyi: Rational Behaviour and Bargaining Equilibrium in Games and Social Situations, Cambridge 1977. K O L M ( 1 9 7 2 ) . Serge-Christophe Kolm: Justice et iquite, Paris 1972. R A W L S (1982). John Rawls: "Social Unity and Primary Goods", in Sen/Williams (1982). S E N / W I L L I A M S ( 1 9 8 2 ) . Amartya Sen and Bernard Williams (eds.): Utilitarianism and HARSANYI

Beyond, Cambridge 1982. ( 1 9 5 7 ) . Jan Tinbergen: "Welfare Economics and Income Distribution", American Economic Review: Papers and Proceedings Al (1957).

TINBERGEN

4

Susan Hurley impressed this point on me.

R U D O L F SCHÜSSLER

Wish You Were Me: A Reply to Broome and a Comment on Harsanyi's Extended Preference Theory*

1. The idea of using extended preferences as a solution to the problem of interpersonal utility comparisons is old. It has been present whenever somebody imagined how he would feel in the shoes of somebody else. The technical concept of an expected preference structure, however, is comparatively new. In the following I will discuss John Harsanyi's version of it and John Broome's critique. Harsanyi's theory is probably at the core of the extended preferences approach and Broome's critique of Harsanyi seems to be the centerpiece of his paper. Broome and Harsanyi agree on the definition of an extended preference structure as an ordering in which the characteristics of persons and events in their lives are jointly evaluated (cf. Broome, sect. 3). Extended preferences tell us e.g. that it is better to be a healthy millionaire than a sick pauper. The example is trivial but suffices to show that we sometimes evaluate positions in life (being a millionaire or a pauper) and goods (good or bad health) in combination. Most of the time this implies a heavy dose of empathy (cf. Arrow 1977, p. 224). Empathy, however, is a crooked yard-stick. Would you know how much an Australian aboriginal values his totem or the President of the United States of America a haircut? It is probably not fortuitous that we feel safer with trivial comparisons than with more relevant ones. Extended preferences which are derived from empathy will very likely vary considerably between observers. Broome returns to this point several times in his paper but he never doubts that empathy is the main road to interpersonal comparisons of utility. He probably regards any alternative approach as a priori unpromising. Harsanyi, in contrast, abandons empathy and embraces scientific explanation in his causal argument. This leads to a mismatch between Broome's critique and Harsanyi's argument which has to be clarified before we can proceed to more general questions. Broome's critique centers on Harsanyi's causal argument * I would like to thank Hartmut Kliemt for his extended, insightful comments in the margins of the second to last version of this paper.

Wish You Were Me: A Reply to Broome

289

(cf. Harsanyi 1977, p. 51; Broome, sect. 4). Harsanyi assumes that, for individual i, an ordinary extended utility function Vt(A,P) exists in which A denotes positions in social situations and Ρ the subjective attitudes of individuals. Finding out which subjective attitudes someone has is notoriously difficult, but Harsanyi believes that knowledge of psychological laws and of the causal formation of attitudes will help us determine this issue. Therefore, the parameter Pi in individual i's extended utility function Vt can be replaced by a causal parameter R/ which specifies the causal conditions under which i came to possess her attitudes. This means that we can rewrite individual z"s extended utility function using Vi(Ai,Pj) = Vi(Ai, R,j. Broome (sect. 5) points out that the two roles of causal constraints as arguments and as parameters in utility functions should not be confused. This is, of course, true. It is less clear, however, why Broome takes Harsanyi's causal argument as target and not immediately extended preference theories in general. Since just replaces P; one could easily generalize Broome's admonition by stressing that subjective attitudes can also assume the different roles of arguments and parameters in utility functions. As the reader may check, Broome's argument can be generalized in this way. Nevertheless, we have to examine Broome's arguments against Harsanyi as they stand. Let us first talk about an individual's personal utility function U;(A). T h e causal argument tells us that this function is influenced by causal factors Ri, but individual i (at least in the basic scenario of extended preference theory) has no preferences ranging over counterfactual causal histories of her life. Since the causal argument equates individuals with their causal lifehistories, UI(A) could be indexed by RI and rewritten as UR.(A) or U(.A\RI) with a semicolon to indicate that /?; is not an argument of U . Broome prefers to drop any reference to individual i altogether and to write UR(A) and U (Λ;./?). But then how do we know that U is the personal utility function of individual i if R is a variable which ranges over possible life-histories? So much for Broome's notation, now to a problem with Harsanyi's. In the long passage quoted by Broome, Harsanyi (1977, p. 59) states that: (E)

V^Rj)

= Vj[AjtRj]

=

UjiRj).

Rj appears here as argument in the utility function Uj{.) although it should appear as index. This may explain Broome's warning not to confuse the two. Nevertheless, I plead for a mild verdict. From my point of view, putting R in an argument's place constitutes a notational mistake or a printing error that can be corrected without deep consequences for Harsanyi's theory. It follows from the context that the personal utility function in (E) should read Uj{Aj)

290

Rudolf Schüßler

instead of Uj(Rj). Harsanyi (1977, p. 55) maintains e.g. in connection with his Axiom 2° (called the "Principle of Acceptance") that:

Uj(Aj) = Vi(Aj,Pj), for all Λ,·. The Principle of Acceptance forces any evaluator of individual j s utility in situation Aj to follow j s personal preferences. Equation (E) should be a restatement of this principle with Rj replacing Pj in Vj. Broome, however, regards Harsanyis mistake as quite serious and after some remarks about the parametrization and transformation of utility functions he finally equates Harsanyis extended preference function Vt(Aj,Rj) with his U{A\R). This move is highly problematic. Of course, j's preferences as following from i s extended utility function Vl and js preferences as indicated by j s personal utility function have to be identical by virtue of the Principle of Acceptance. But this identity does not extend to the level of utility functions. Only after the problem of scale for utility functions has been solved would both functions have to be identical too. Let us nevertheless assume the identity of utility functions for the sake of the argument here, although I will soon criticize this assumption. Even in this case we can establish only an equivalence between Vi(Aj,Rj) and U(A;Rj) because Harsanyis extended preference function refers to j and so must its personal counterpart. Dropping any reference to particular individuals results in V(A, R) = U (A; R) but there is no principle in Harsanyi (1977) which entails this equation. I mention this formal point only because to ignore it is to ask for conceptual trouble. U (A; R) is no extended preference function and it takes causal values as parameters, whereas the extended utility function V takes them as arguments. There is no confusion involved in this difference, as long as U and V are distinct functions, which they are. 2. Up to this point there is no indication of a substantial inconsistency in Harsanyi but maybe things look different if we add the assumption that individuals have preferences for some life-histories rather than for others. In this case, causal variables enter as arguments into the personal utility functions of these individuals. Since the parameters of these functions also follow from causal histories, we arrive at personal utility functions of the type U (A+-,R), where A+ denotes a set of alternatives A which has become enlarged by lifehistories. Broome (sect. 5) mentions only a second possibility of liberalizing the causal argument. He maintains that an individual s extended utility function will be no less subject to causal parametrization than her personal utility function. Instead of ascribing an extended utility function to some individual j we can therefore just as well represent j through a causal history and write

Wish You Were Me: A Reply to Broome

291

V(A,P\R). Again I can see no reference to j or her particular life-history in Broome's notation but let's skip that point. It may be expected that the causal histories of individuals will differ and Broome takes this to prove that Harsanyi is wrong in believing that all people will have the same extended utility function. In my view, Broome's extension of Harsanyi's theory is misguided. First, why does Broome retain the subjective attitudes Ρ as arguments of V while Harsanyi switches to their causes /?? This can mislead the reader to believe that empathy still plays a noticeable role in Harsanyi's argument. In this case, an observer could indeed hardly abstract from his own causal background which is responsible for his capacity for empathy. But in Harsanyi's argument all observers adopt the personal utility functions of others as basis for their extended preferences in keeping with the Principle of Acceptance. They do not evaluate the situation of other persons according to their own taste and causal background. Harsanyian extended utility functions are derived 'in a detached way' from psychological laws. Therefore the life-history of, e.g., an academic does not interfere with his determination of a financial analysts utility as long as it does not reduce his cognitive abilities, and as long as psychological laws are publicly known. Psychological laws play the same role for Harsanyi as laws play in science everywhere. They serve as basis for explanation and prediction regardless of the life-history and personal preferences of a scientist. On these grounds the Principle of Acceptance tells all observers to derive the same thing, namely individual i's personal utility function as blueprint for their extended evaluation of i's position. This leaves us with the problem that i's personal utility function will contain evaluations of several life-histories if we follow Broome in extending the set of alternatives. In consequence, i's adopted personal preference for say j's position could differ from j's own preference. Of course, this can happen only if individual i fails to evaluate foreign positions according to the Principle of Acceptance. From Harsanyi's point of view, however, it is clear that the Principle of Acceptance informs all positional evaluations. A Harsanyian extension of the set of personal alternatives would therefore come to the same evaluations as the corresponding extended utility function. Again, Broome is wrong if he attributes inconsistency to Harsanyi. The best target for attacks on Harsanyi's position is probably its premises. Broome's repeated contention that real humans evaluate the positions of other individuals idiosyncratically seems to reflect this point because it undermines the plausibility of the Principle of Acceptance. If we are ordinalists we have to respect human preferences as they are. In consequence, Harsanyi' Principle of Acceptance cannot be acceptable for ordinalists, if real humans do not em-

292

Rudolf Schüßler

ploy it in their evaluation of alternative life-histories. This argument leads us to problems of ordinalism which I will soon discuss. For the moment it may suffice to note that Harsanyi's causal argument is not refuted if it is shown that Harsanyi is wrongly regarded as radical ordinalist or erroneously thinks of himself as radical ordinalist. The Principle of Acceptance is indisputably present in his work and has to be respected as basis for an evaluation of Harsanyi's theories. Moreover, it is doubtful whether the idiosyncratic evaluations envisaged by Broome can conflict with Harsanyi's kind of positional evaluations at all. O n closer inspection our usual evaluations of positions we might assume under possible different histories of the world have the form: "How would I (as non-millionaire) feel if I were a millionaire?" This question refers to a blend of the life-histories of say individual i, the non-millionaire, and j, the millionaire. Harsanyi's theory, however, asks how a millionaire feels as millionaire and thus for j s evaluation of j s life-history. Even if we introduce 'blended' positions into our set of alternatives it does not follow that we will observe divergent evaluations of the same position. As long as personal identity corresponds to a unique set. of personal characteristics it will always be clear whose preferences count in a certain position. And if this person's preferences would change after a change of position we can say that she is badly informed about how she would feel if her life were different. Of course, an individual's thoughts about being a millionaire can deviate from his prior expectations after he really has become a millionaire. This is a problem of empathy with oneself and a very insufficient basis for attacks on Harsanyi since it is assumed quite generally in utility theory that all individuals have well-informed preferences and do not value any good less after they have received it. Such things occur in reality and they create problems for welfarism but they do not impugn the extended preferences project in particular. 3. Nevertheless, Harsanyi's causal approach has other weak spots. In order to arrive at a unique and unequivocal extended utility function for all individuals it is necessary that the utility units of all personal utility functions become definite and known. Otherwise, different observers could still weigh extended alternatives differently. Harsanyi (1977), p. 57, acknowledges this problem with respect to von-Neumann-Morgenstern (= NM) utility functions in the paragraph which directly precedes his exposition of the causal argument. Therefore, the causal argument should be regarded as an attempt to solve the problem of scale for NM-utility-functions and to introduce a cardinal measure of utility. It is controversial whether cardinal utility functions should be assumed for human agents but let's grant the assumption for the

Wish You Were Me: A Reply to Broome

293

sake of the argument because it sheds light on a weakness of Harsanyi's position. NM-fiinctions represent ordinal preferences over risky prospects but they determine an individual's cardinal utility function for choices under certainty only up to monotonic transformations (cf. Schoemaker 1982, p. 533). T h e latter functions, however, are of far greater concern than the former for the extended preference approach because NM-functions always measure utility and attitude towards risk together. Extended preference comparisons, in contrast, just want to establish the value of a form of life for somebody and not whether someone would like to win it in a lottery. Harsanyi s causal argument fails to take account of this fact. Fixing the scale of an NM-function does not tell you how much someone would value a form of life if it were allotted to him. Given different attitudes towards risk the same NM-fiinction is compatible with different cardinal utility functions for choices under certainty or alternative allotments. What Harsanyi should have assumed is therefore that the knowledge of life-histories allows us to determine cardinal utility functions for all individuals directly. This calls into doubt whether the spirit of Harsanyi's causal approach is still compatible with the ordinalist gospel (cf. Broome, sect. 2). Ordinalists believe that only the ranking of alternatives is ours to know but not the absolute intensity of our preferences. The causal argument, in contrast, invokes psychological laws to specify the form and units of personal utility functions. How could psychological laws do this? Unfortunately, Harsanyi gives his readers no hint at an explanation. An obvious first hypothesis might be that psychological laws can help determine the intensity of psychic states, which would lead to a resurrection of hedonistic utilitarianism. Measuring psychic states implies that a correlation between psychic and physiological states is established. Nevertheless, decision making depends on cognitive processes to a high degree and these processes can result in deliberate neglect of an individual s own physiological satisfaction. The same holds true for external, objective conceptions of 'the good'. There can be fully conscious ascetics who do not worship 'fun', or lucid nihilist suicides, or well-informed free-thinkers who do not care for 'the good' as long as they are not free to define it for themselves. T h e stipulation of some 'good' provides no sufficient basis for welfare judgements as long as we take an individual's own choices as final evidence for the ranking of her welfare. Measurable material quantities cannot even serve as yard-stick for the calibration of preference structures, since the ordering of the former need not be preserved in the behaviorally effective ranking as exhibited by an individual's preferences. Any satisfactory version of the causal argument therefore has to explain the formation of human beliefs and desires in general. The respective psycho-

294

Rudolf Schüßler

logical laws have to be strong enough to tell us which beliefs human beings hold relative to different life-histories and how to quantify the strength of desires whose satisfaction cannot be expressed in material quantities. Today any appeal to such strong psychological laws is mere science fiction. The individuation of beliefs and unrestricted desires is a notoriously dense and dark philosophical jungle and the observation of choice-behavior is the only way I can conceive of cutting a path into it. The observation of choice-behavior, however, can specify utility up to an increasing transform at best. This message is written on square one of utility theory and most expeditions into the heart of welfare's dark jungle return to square one with bare hands sooner or later. Progress can only be made if we travel with lighter baggage and abandon the ordinalist principle of respect for an individual's own choice. For this reason, Broome (sect. 8) seems right when he maintains that the causal argument strengthens an objectivist theory of the good rather than ordinalism. But it is less clear whether Harsanyi should be classified as a ordinalist paragon entangled in inconsistency. Harsanyi himself is probably to blame for this reading which is shared by well known critics (Broome, Sen). But the reading is not mandatory: Weymark (1991), p. 307, suggests that Harsanyi is better understood as an undercover objectivist and genuine utilitarian with his true identity surfacing in scattered passages of his writings. On the other hand, many signs of ordinalist orthodoxy can be found there as well. If Harsanyi really is having a love affair with objectivism it is certainly not of the ordinary type. The next paragraph will try to show how these seemingly inconsistent strands of interpretation can be woven together to form an intelligible picture. 4. The foregoing considerations assume a strong explanatory role for Harsanyi's causal argument. There is, however, another interpretation which classifies it as part of a normative claim. Take Gibbard's example (1986, p. 188) of the value of Christmas for a Christian and for a Muslim. It can be assumed prima facie that a Christian will enjoy Christmas more than a Muslim. The Muslim, however, might feel at a much higher level of emotional intensity so that even his weakest joy is of higher absolute value than the best Christian Christmas enthusiasm. This point has its well-known technical counterpart in the underdetermination of scale and intercept of the Christian's and the Muslim's utility functions. Gibbard maintains that we can defuse this underdetermination with an argument of insufficient reason. As long as we have no reason to assume different scales and levels of utility we should suppose homogeneity of scale and intercept across individuals on moral grounds. In the case of the Christian and the Muslim we indeed know of no relevant differences in the setup of the persons we want to compare. We just know that a Christian

Wish You Were Me: A Reply to Broome

295

normally has a higher preference for Christmas than a Muslim. Hence it is quite reasonable to assume that Christmas has more objective utility to Christians than to Muslims. Harsanyi seems to have a similar argument in mind, when he maintains that we should assign equal values to the utility units of different individuals on grounds of principles of justice and equal regard for human needs (cf. Weymark 1991, p. 314). Psychological laws, on the other hand, can supply us with reasons to give up the assumption of individual unit equivalence. Therefore, Harsanyi s causal argument can be seen as attempt to delimit the range of an unproblematic equal treatment of individuals. This gives double impact to the intersubjective explanatory role of the notion of a scientific "law", since it also helps to avoid the pitfalls of misguided empathy. Unfortunately, Harsanyi fails to state clearly how he wants us to understand his argument. I will, here, use the most charitable interpretation. All this considered, we obtain the following result: causal arguments can help us to specify a range of equal respect for individuals where equal utility units are assigned for normative reasons. Despite Broome's objections nothing is formally wrong with Harsanyi's concept of an extended preference structure. Therefore we can nourish the hope that the fundamental problems of interpersonal comparisons of well-being can be solved if we share Harsanyis premises. Harsanyi emerges as a kind of mitigated ordinalist. Against objectivist material theories of the good he postulates the primacy of an individual's actual preferences. The scale of the resulting utility function is determined by objectivist normative arguments of equal treatment. These arguments are formal in spirit so that I arrive at the bipartite characterization of Harsanyi's position as "material ordinalism cum objective moral formalism". 5. My defense of Harsanyi should not be mistaken as sign of agreement. Although I tried to show that Harsanyi's arguments are stronger than Broome and other critics are ready to admit, I am sceptical about the whole extended preference approach. Let me briefly sketch some of my reservations. A strong rationale for constructing theories of interpersonal welfare comparisons is that common sense compels us to compare welfare interpersonally in politics and everyday life. Therefore, say the utility theorists, we should try to base our comparisons on theoretically analysed, rational arguments instead on unreflected intuition. This argument looks impressive. But let us remember the problems of real-life welfare judgements. Even if we can stumble upon examples where interpersonal comparisons of welfare seem sound this does not mean that we have a sound way of making such comparisons. It is possible that an excessive confidence in our capacity to distinguish the correct cases for

2 9 6

Rudolf Schüßler

interpersonal comparisons may lead to an unbearable amount of unjustified judgements together with a certain number of justified ones. This is not just an abstract possibility but part of the real problem of paternalism. Many of us would not want to endow even a benevolent individual or collective agent with the power to make interpersonal welfare comparisons, precisely because we do not trust anybody's judgement in such matters, especially since power can corrupt people. Even the most superficial study of history books shows what can go wrong. It is therefore not unreasonable to abstain from interpersonal comparisons of welfare wherever we can and to put theoretical claims to very hard tests. This warning also applies to the seemingly philanthropic argument from principles of equal respect. Rights to equal respect are the crown jewels of enlightened humanitarianism but historically they very often came into existence as weapons against arbitrary acts of states or their agents. In extended preference theory, however, they play a different role. The argument of insufficient reason authorizes administrators of distribution policies to treat individuals alike even against their own will if no relevant objective differences between the individuals can be established. Note that a preference against certain stipulations of the scale of personal utility functions cannot itself be part of an individual's personal preferences. Such a preference would create inconsistencies in the system of preference theory. Harsanyi's brand of egalitarianism supports claims of the state against the individual and not vice versa, as in the best part of the enlightenment tradition. This observation should help cool down equal respect activists and show that even a normative approach to extended preferences is anything but uncontroversial.

References Kenneth J . Arrow: "Extended Sympathy and the Possibility of Social Choice", American Economic Review 6 7 ( 1 9 7 7 ) . E L S T E R / H Y L L A N D ( 1 9 8 6 ) . Jon Elster and Aanund Hylland (eds.): Foundations of Social Choice Theory, Cambridge 1986. E L S T E R / R O E M E R ( 1 9 9 1 ) . Jon Elster and John E. Roemer (eds.): Interpersonal Comparisons of Weil-Being, Cambridge 1991. G I B B A R D ( 1 9 8 6 ) . Allan Gibbard: "Interpersonal Comparisons: Preference, Good, and the Intrinsic Reward of a Life", in Elster/Hylland (1986). H A R S A N Y I ( 1 9 7 7 ) . John C. Harsanyi: Rational Behavior and Bargaining Equilibrium in Games and Social Situations, Cambridge 1977. S C H O E M A K E R ( 1 9 8 2 ) . Paul Schoemaker: "The Expected Utility Model: Its Variants, Purposes, Evidence and Limitations", Jounal ofEconomic Literature 20 (1982).

ARROW ( 1 9 7 7 ) .

Wish You Were Me: A Reply to Broome

297

(1991). John Weymark: "A Reconsideration of the Harsanyi-Sen Debate on Utilitarianism", in Elster/Roemer (1991).

WEYMARK

RAINER

HEGSELMANN

Experimental Ethics: A Computer Simulation of Classes, Cliques, and Solidarity

Abstract: The article deals with two questions: (a) Can relations and networks of solidarity emerge in a world exclusively inhabited by rational egoists, who are unequal and choose their partners opportunistically? (b) If networks of solidarity do emerge in such a world, what do they look like? By means of computer simulations it is shown that networks of solidarity can emerge in such a world. But the networks will show quite distinct features of some class segregation. 1. 2. 3. 4. 5.

Introduction How Can Solidaric Relations Be Modelled? How Can Opportunistic Choices of Partners Be Modelled? Some Experiments A Research Perspective: Modelling Moral Dynamics by Cellular Automata References

1.

Introduction

There are a wide variety of questions of moral philosophy. In answering these questions one can rely upon various philosophical methods, like logical analysis, theories of rational decision-making etc. I want to argue that computer aided simulation may also be a useful tool for the moral philosopher. The tool is useful because the problems addressed in moral philosophy are frequently characterised by complex dynamics. Very often the only way of tackling those is using simulations. Using merely intuition and paper and pencil methods one cannot come to grips with complex dynamics. However, as soon as a computer model of the relevant structures is formed, experiments can be carried out and one may hope to gain some understanding of what is going on. I shall illustrate the usefulness of computer simulations for moral philosophy subsequently by studying a paradigm case. The simulation will be concerned with the following two questions: (1) Can relations and networks of solidarity emerge in a world exclusively inhabited by rational egoists who are unequal regarding their natural endowments, who can and must choose their partners themselves and will do so — in coherence with the assumption of rational egoism — opportunistically? (2) If networks of solidarity do emerge in such a world, what do they look like·!

Experimental Ethics

299

Both questions are too vague to allow for a direct answer. It is necessary to state them more precisely and to solve some modelling problems before they can be tackled by means of simulation. This in turn raises problems which can be grouped around two preliminary methodological issues: How should solidaric relations be modelled? How can opportunistic choices of partners be modelled? In what follows I will, firstly, deal with each of these two preliminary questions. This will clarify the central assumptions which form the basis of my simulation. Secondly, I will make use of my 'solidarity simulator' in order to perform some experiments that can serve as paradigm examples of what can be done with the methods introduced here.

2.

How Can Solidaric Relations Be Modelled?

I shall try to characterise solidaric relations by a particular game to which I will refer as the solidarity game. Figure 1 shows the extensive form of the game. The solidarity game is a two-person game. It is a fundamental characteristic of the game that both players will become dependent on help with a certain probability, p\ or pi- The probabilities remain constant throughout the game. (1 -p\) and (1 — pi) are the probabilities that player 1 and player 2 respectively will not become needy. The first move of the game is chosen by nature (player 0): According to the probabilities p\ and p2 it is decided by chance whether both, none, only player 1 or only player 2 will be in need of help. So

it is possible that neither of them is in need of help or that both are at the same time. In the first case neither needs help while in the second neither is able to help. Whoever is in need of help himself cannot help the other. On the other hand a player not in need of help has to decide whether he wants to help. This decision, to help or not to help, is a move of the game. Players who are not able to help cannot make a move. Thus depending on the move of player 0 either player 1 or player 2 or neither can make a move. The payoffs are specified as follows: To get help is of utility S ("to be Saved") whereas not to get help yields utility D ("Drowned") to a player who needs help. 1 It should obviously be the case that S > D. A player who is not in need of help gets the payoff Μ ("Move on") if he does not help the other, and Η ("//elp") if he does. Since it takes some effort to help and we assume our players to be egoists one should expect that Μ > Η . It is helpful 1

Using these variable names for mnemonic reasons may be misleading in one respect: W h o ever drowns in real life cannot join the game again, whereas a player of the solidarity game, even if he does not succeed in getting help, can play on in the following periods.

300

Rainer Hegselmann Both in need of help/none can help

D D

T\ Τϊ None in need of help

(1 ~ p\)

Μ Μ

(1 ~ pi) Help

Only player 1 in need of help pi

S Η

Player 2 Not to help

(l-/>2)

D Μ

Help Only player 2 in need of help Pi

Η S

Player 1 Not to help

(l-/»i)

Μ

D

Fig. 1: The solidarity game

Solidaric (cooperate) Non-solidaric (defect)

Solidaric (cooperate)

Non-solidaric (defect)

Ri

Si r2

T2 Pi

Ά Pi

Pi

Table 1: The solidarity game in its strategic form

to characterise the game also by its strategic form as shown in table 1. The columns and the rows of the table show the strategic options from which each player has to choose. To help if one can and the other is in need of it is called the solidaric strategy. Not to help under the same circumstances is called the non-solidaric strategy. The expected utility, which corresponds to the payoffs Tt, Rn Pt, 5, of the solidarity game, can be calculated as follows ( i , j = 1,2 and i φ j): (A)

Ti = Ri = Pi = Si =

(1 - pi) (1 - Pj)M + piPjD+pj( 1 - Pi)M+pl( \ - pj)S (1 -Pi){\-p^M+piPjD + pji 1 -pi)H + pt{ 1 - Pj)S (1 — pi) (1 -pj)M+piPjD + pj(l-pi)M+pi(l-pj)D (1 -pt)( 1 -pjW + pipjD + pji 1 — pi)H + pi(l-pj)D

The solidarity game turns into a prisoner's dilemma (PD) if and only if the

Experimental Ethics

301

characteristic condition on the ordering of the payoffs (B)

7} > Ri > Pi > Si

(i = 1 , 2 )

is fulfilled. Reverting to (A) and (B) it is easy to prove Lemma 1: The solidarity game is a PD iff Pi{\-pj){S-D)>pj{\-pi){M-H) (i,j = 1,2 and ίφ j). Referring to ( S — D) as solidarity profit and (Μ — Η) as solidarity cost the following proposition may be formulated: The solidarity game is a prisoner's dilemma iff the expected solidarity profit is greater than the expected solidarity cost. In other words: The solidarity game is a prisoner's dilemma iff mutual solidarity would be profitable to both players. This means that the solidarity game even though it is not always a prisoner's dilemma will be one in all interesting cases. I will refer to lemma 1 as the PD-condition. The prisoner's dilemma is the classical example of a social trap: Both players act in an uncooperative way if they choose dominant strategies which lead to a maximin point and the only Nash-equilibrium of the game. Non-cooperative behaviour or defection of both players is the game's only plausible solution2 and just that solution is Pareto-inefficient because both players could be better off if only both had chosen cooperation. Applying this to a solidarity game with PD-conditions fulfilled we have to state: Rational egoists will not be solidaric even if mutual solidarity is profitable for both. Both players know that cooperation by both sides forms a superior solution of the game. But both also know that it is an even better solution for them if the opponent cooperates and the actor defects. Both know that the other knows this fact too and so they have no choice but defection. It is a well-known fact3 that PD games, if the basic game is iterated, can have 'cooperative'equilibria which result from supergame strategies supporting cooperation in all the basic games. In this sense PD games are cooperatively solvable, a result which, of course, must in principle be applicable to solidarity games. Such cooperative supergame equilibria exist if the game is iterated indefinitely and if the payoffs of future games are not discounted too much. Let a be the discount parameter (with 0 < α < 1). Then the payoff of a game in period t will be discounted with the value of. If the future payoffs are exponentially discounted this way the sum of all discounted payoffs converges 2

3

Any other solution does not meet the elemental requirement of being self-supporting and self-enforcing. Cf. Taylor (1976) and (1987), Axelrod (1984) and Friedman (1986).

302

Rainer Hegselmann

and thus the payoff for the whole game or the supergame payoff is well defined and finite. Cooperative equilibria in this model only exist if the discount parameter exceeds a specific threshold value. Using a different though formally equivalent interpretation, which I prefer, one can define the discount parameter as the probability of a further game. Under this interpretation one expects with probability 1 to observe a finite number of plays of the basic game. But the supergame's end or last round of play is uncertain. If the payoffs of future periods are weighted with the probability of reaching them and if the supergame's payoff is again defined as the sum of the weighted payoffs of the basic games, cooperative supergame equilibria shall exist if on each round of play the probability for a further round of play is sufficiently high Which kinds of supergame strategies make cooperative solutions possible? Obviously a strategy of unconditional cooperation (ALL C) cannot be an equilibrium strategy against itself, because unconditional defection (ALL D) will always be a better answer to ALL C. On the other hand, an equilibrium is reached if both players choose ALL D, but that equilibrium is not a cooperative one. For the existence of cooperative solutions it is necessary that supergame strategies be conditional. The strategies must allow for the prescription of different responses depending on the choices of their opponent and thus be capable of retaliation. Two well-examined strategies which have this property are TIT-FOR-TAT and TRIGGER.5 Friendly TIT-FOR-TAT plays cooperative in the first game and afterwards always acts the same way the opponent did in the previous move. So cooperation is answered with cooperation and defection is answered with defection. TIT-FOR-TAT (TFT) is a very conciliatory strategy because, even after a long time of the opponents defection, a TFT player will be ready for cooperation if only the opponent plays cooperative for one time. On the other hand TRIGGER (TR) is a totally unforgiving strategy: It will punish the opponents first defection with never-ending defection. But TR begins the supergame friendly by playing cooperatively and continues in this way until the opponents first defection. If one tries to apply findings about equilibria in iterated standard PD directly to the solidarity game an interesting complication will arise: In determining the threshold values of a* for the iterated standard PD it is assumed that after each period both players are able to recognise whether the opponent did cooperate or defect. The same supposition is not plausible if applied to the 4 5

The two interpretations can, of course, be combined. Cf. Friedman (1986), Taylor (1976).

Experimental Ethics

303

solidarity game: Supergame strategies are plans of behaviour that are made for the whole sequence of the iterated basic game and thus determine how to behave in each specific basic game. Consequently each player knows whether he will help a needy other player if helping is possible. So each player i knows whether he plays solidarically. But what about the opponent j of a player P. Is it possible for him to recognise whether the behaviour of player i is solidaric in a specific game if either none of them was in need of help, both of them, or player i only? In what follows I will not assume such a capability. I will assume instead that only in cases of unilateral emergencies it is observable whether the other one is cooperating or defecting. Lemma 2: An iterated solidarity game has a cooperative solution based on TR or TFT strategies if and only if for both players it holds that 1 . at > ^"rT = a; l-pj(l-pi)+pi(l-pj)jr§l (.i,j= 1,2 and / φ j). I will refer to lemma 2 as the COOP-condition.6 After all that has been said, the preliminary questions of the first complex seem to be sufficiently clarified. When I talk about solidaric relations or solidaric solutions subsequently I refer to iterated solidarity games which are stable enough to make cooperative equilibria viable — under the conditions prevailing at a given time. According to my first central question, we are concerned with solidaric relations among unequals. The solidarity game offers a very natural way of modelling inequalities: Individuals are unequal in that they need help with different probabilities. Pursuing this approach I will divide the set of players of the solidarity game into nine different risk classes. Risk class 1 will need help with probability 0.1, risk class 2 with probability 0.2 etc. Solidarity costs and profits I will assume to be equal for all individuals independently of their risk classes. Thus inequality will solely be modelled in terms of membership of different risk classes. This is not as restrictive as it might seem. Even though solidarity costs and profits do not directly vary over different risk classes there will still be differences between the individuals: These differences stem from the solidarity costs and profits that can be expected as an effect of engaging in solidaric relations with them.

6

But notice that there are many other equilibria also. The proof of lemma 2 only exists as a manuscript (Hegselmann 1991).

304

Rainer Hegselmann

3.

How Can Opportunistic Choices of Partners Be Modelled?

The basic ideas of my model of opportunistic choices of a partner are: -

The individuals live in a two-dimensional cellular world. The individuals have different numbers ofneighbours of different risk classes. Not all the places in the world are occupied. Sometimes the individuals are offered migration options which allow migrations within specific bounds. They use these options in order to search for a position which is as attractive as possible. The individuals are endowed with cognitive skills that enable them to identify the risk classes of other individuals. They get some information about a part of the world around them, and they are intelligent enough to 'draw some conclusions' from that information.

-

-

These basic ideas need some more explanation. 3.1. The 'Geometry of Social Structure. The individuals live within a twodimensional cellular world. It is possible to define different neighbourhood concepts for those worlds. I will assume here that the neighbourhood consists of the adjacent cells in the north, south, east and west. This type of neighbourhood is a so-called von Neumann neighbourhood7 Each individual can in each game period, and at each of his flanks, become dependent on help. Therefore each can be confronted with a maximum offour problems in each period. I will assume that each individual simultaneously plays a solidarity game with each of his four neighbours, each game being independent of the others. On the basis of this assumption it becomes viable to construct a solidarity game out of dyads as its building blocks. If engaging in solidaric relations is strategically rational it is attractive to have all the neighbour cells occupied, since no help at all can be expected from an unoccupied cell. I will take a world of 2 1 x 2 1 cells as a basis of the solidarity game. I assume that the world is a torus. Thus the neighbours of individuals from the western 'border' of the world are those who are situated on the eastern 'border'. Northern and southern border are treated analogously. 3.2. The 'Social Primeval Soup'. The initial state of affairs from which the searching for solidarity partners and for attractive social positions begins, will be created by chance. All the risk classes will contain the same number of indi7

In case of a so-called Moore neighbourhood the cells which diagonally border on the central cell belong to its neighbourhood also.

Experimental Ethics

305

viduals, who are spread out over the 2 1 x 2 1 cells at random. The total number of individuals is chosen such that many cells remain vacant. So an initial chaos or some kind of social primeval soup is formed where individuals live in positions created by chance with more or less (possibly no) neighbours of different kinds. 3.3. Migration Options and Game Periods. A random generator allocates migration chances. For this purpose the cells are visited one by one and migration options are awarded with a certain probability. If an individual gets the chance of migration he may migrate but is not forced to. A lottery specifies the order in which the individuals can use their migration options. Thus the migrations are sequential. Migration is possible only if there is a vacant destination cell; no-one can be ousted from his social position. Because the migration is sequential, there is no competition for attractive positions. Migrations correspond to the principle "first come, first served". After the social primeval soup' has been created in the very first period, migration options are awarded to each individual in each period of the game with an exogenously set probability. Afterwards the migration options are sequentially executed according to the results of a lottery. The new configuration of individuals after the migrations have been executed is called the social situation of this game period. This is the basis for statistical analyses of the changes of neighbourhood structure, of payoff development, etc. 3.4. Generally Accessible Information and Intelligence. I will assume that individuals can instantly tell each other's risk classes and know their own class, too. Each is informed about solidarity costs and profits. Furthermore each knows the probability of being offered a chance to migrate within a specific period. The individuals should be intelligent enough to draw some conclusions from this information. The probability of getting migration options is common knowledge and is used to calculate a pessimistic proxy for being neighbours and therefore having a further round of play with the next period. This probability is nothing else but the discount parameter interpreted in terms of probability. In other words: Knowing the probability of getting a migration option enables individuals to calculate a lower limit of the discount parameter of their world. This is a lower limit because probably not every option will be used, for instance if the actual social position is very attractive. I will assume that the individuals take this lower limit as a pessimistic proxy of the level of stability of their neighbourhood or as their discount parameter. Because every individual knows solidarity costs and profits, each can calculate those risk classes with which — relative to the individual's own class — mutual

306

Rainer Hegselmann

solidarity is profitable and will lead to an equilibrium because neighbourhood relations are expected to be sufficiently stable. In other words: Each individual knows the risk classes that have members with whom mutual solidarity can be possible and profitable. Furthermore each knows the specific payoffs resulting from solidaric relations with the members of all the risk classes, because they only depend on values which are generally known. 3.5. Partners of Different Attractiveness. The payoffs received from solidaric relations with partners of different risk classes are different. This will become particularly clear if we regard the solidarity surplus Δ,·. Let that term refer to the difference between the payoffs from mutual solidarity and from non-solidaric behaviour on both sides. It will be given by the relation: (C)

Δ, = />,·( 1 - Pj)(S - D) - ρj{ 1 - Pi) C Μ - H ) .

Therefore the higher the solidarity profits, and the lower the solidarity costs, the higher will be the solidarity surplus. For a member of a given risk class the solidarity surplus will be maximised by a partner in as good a risk class as feasible. The individuals are intelligent enough to know this. Thus the individuals have to take into account two points: On the one hand a partner should be of a risk class that is as good as possible, on the other hand mutual solidarity must be profitable for both sides, and an equilibrium. Obviously this imposes a constraint on the choice of a partner. A partner of a very good risk class is attractive to members of bad classes because they can get a higher solidarity surplus. But this combination would not be profitable to the other partner and so is out of the question. Thus if you live among rational egoists you might not be well-advised trying to make friends with the strongest. All these conclusions, which can be easily drawn from their common knowledge, are known to all individuals, who are realists in evaluating their opportunities. They know that the search for best partners is restricted by the fact that solidaric relations must be profitable to both sides. As a corollary of the foregoing considerations we can state that within the world studied here, one-sided, non-reciprocal solidarity is not possible: Either the conditions for mutually profitable solidaric relations are fulfilled, in which case both partners will know it, and behave solidarically. Or they are not fulfilled, in which case the individuals will also know, and have no reason for solidaric behavior. 3.6. Best and Worst Social Positions. Against the background of attractiveness of partners, depending on their risk classes on the one hand and on the

Experimental Ethics

307

assumptions concerning the 'geometry' of interaction structures on the other hand, it is easy to characterise a best social position: The best social position is a location where an individual is surrounded by partners chosen from the best risk class whose members are willing to engage in solidaric relations with him. In a position like this the solidarity surplus would be at its maximum. On the other hand there are worst socialpositions, too. These are locations where an individual is surrounded by empty cells or by individuals with whom he cannot engage in profitable solidaric relations. From such neighbours, or empty cells, one could expect no help in case of emergency and therefore would receive the payoff Pi at each of the corresponding flanks.8 Thus empty cells are as bad as useless neighbours. At the worst social positions the solidarity surplus Δ; is zero. 3.7. Satisfied and Dissatisfied Individuals. At each stage of the game best social positions will be scarce: not (or no longer) available, or at least difficult to find. In view of such scarcity, I will assume the following: The individuals know the maximum solidarity surplus that they could receive relative to their own risk class - four times the solidarity surplus resulting from solidaric relations to partners of the best reachable risk class. An individual will be satisfied with a position, however, which offers some fraction of the maximum solidarity surplus this individual could possibly receive. I will call this fraction (which can be chosen as a parameter of the simulation) minimum level. The minimum level can be characterised by a percentage lying between 0%, which is equivalent to the worst social position, and 100%, which is equivalent to the amount at the best social position. Whoever reaches the minimum level will be called satisfied, while an individual who does not reach the desired minimum level will be called dissatisfied. Thus the risk class is not the only difference between individuals; they can also be classified as satisfied or dissatisfied, a classification which is independent from their risk classes. A distinction between satisfied and dissatisfied individuals is made mainly because it allows for a simple way of modelling migration strategies, where an individuals status quo payoff is taken into account. Satisfied and dissatisfied individuals will use their migration options in different ways. For convenience I shall assume that dissatisfied individuals will seize every migration opportunity offered to them. From the positions within their reach they will of course choose the best. But they will switch to another location even if the new one is worse than the old. Moving to a worse position temporarily is regarded as 8

Notice that here the payoff Pi - as given by the equation from (A), see sect. 2 above - does not depend on pj because it is: /} = (!— pi) Μ + pjD.

308

Rainer Hegselmann

a cost of searching. The dissatisfied pay these costs because they hope for the chance of getting to a better position from this one. On the other hand satisfied individuals will not make use of an opportunity to migrate unless the new position is not worse than the actual one. 9 3.8. Limits ofMobility and Knowledge. I will assume that individuals can move only within a sector of their world. Each individual forms the centre of a square sector, which moves along if the individual migrates. Thus starting from a specific point there is no point of the world which cannot be reached in principle. However, not all points can be reached with one move. The freely adjustable size of the sectors, which is the same for all individuals, will be called migration window. As can be seen in figure 2, the migration window sets a limit to the number of cells an individual can move north, east, west or south. The double-lined sector shows a migration window of 5 x 5 with an individual of risk class 3 (indicated by a circle) in the centre. When forming their judgement as to whether a position is attractive, individuals must be able to assess how well off they might be at alternative positions. I will assume that within the limits of the migration window individuals are able to identify vacant positions, and at the same time can anticipate how well off they would be in alternative neighbourhoods. Since cells on the borders of the migration window have neighbours outside, information about adjacent cells of this sector must be available. Thus the information window is a bit larger than the migration window. A very attractive social position for an individual belonging to risk class 3 in figure 2 could be the extreme northwest of its migration window, since this location is surrounded by members of a good risk class. Moving to this position would be attractive if individuals of risk class 2 were ready to engage in solidaric relations with members of risk class 3.

9

Obviously the mechanism is similar to the satisficingprocedure Simon (1955) suggested for explicating the idea of bounded rationality. One could say that in this simulated world the individuals are confronted with two basic strategic problems: (a) With regard to a position's specific neighbourhood they have to decide which neighbours they want to have solidaric relations with, (b) If they get a migration option they have to decide whether they want to migrate, and where they should go. The first problem is handled within the orthodox frame of game theory. The second strategical problem is a hypercomplex problem. Hypercomplexity of situations has always been a motivation for working with models of bounded rationality. To handle the second strategical problem I fall back upon one such model. This means that there is no underlying unified and well-elaborated concept of rationality here.

309

Experimental Ethics

8

2 2

7 2

2 2

5

2 8

®

1

4

Fig. 2: The migration window

4.

Some Experiments

In the following I will describe three experiments. They differ with regard to the probabilities with which the individuals get migration options. Notice that the possibility of functioning solidaric relations between members of different risk classes is affected by that probability. All other parameters are kept constant. Initial Conditions: Payoffs:

Saved = 5

.Drowned = 1

Move on = 7

Η dp = 6

Probability for getting a migration option: 0.05 =>· a + = 0.903 (first experiment) 0.10 a + = 0.810 (second experiment) 0.15 =» a + = 0.723 (third experiment) Individuals per risk class: 35 315 individuals total / 136 empty sites Minimum level = 50% Use of migration options: (a) dissatisfied individuals decide for a best alternative location; (b) satisfied individuals move only if the alternative location is at least as good as the given one;

310

Rainer Hegselmann

(c) individuals knowing that they cannot find partners at all move

randomly. Interactions: von Neumann neighbourhood Migration window: 1 1 x 1 1 World: 2 1 x 2 1 (torus)

Figures 3 to 5 show the evolved structures of societies after 1000 periods. They all started with a primordial soup. Interpreting the figures one should know: -

-

White cells are empty cells. Round cells are dissatisfied (i.e. get less than the assumed minimum level of 50% ) or they are cells which under the given condition do not have any chance of finding partners for mutual solidarity. Filled squares are satisfied. Short white lines connecting two individuals indicate functioning solidarity relations between them. Different patterns and grey levels represent different risk classes according to the legend at the bottom of the figures.

Even a superficial inspection shows that a well-ordered support network has emerged from an initial chaos. So one can state that networks of solidarity and mutual support can evolve even in a world of rational egoists who are differently endowed by nature and who choose their partners opportunistically. By closer inspection one will find more interesting details. In the first experiment members of class 1 establish solidaric relations mostly among themselves. However, support relations with members of class 2 exist as well. This latter class is living around the areas occupied by class 1. Quite often we find members of class 3 just behind those of class 2, behind them members of class 4 and 5. Class 9 could find partners only in class 8 and 9. The members of those classes are living more or less on the edge of the networks. Segregation is less strongly developed by those in the middle of the risk classes than those in the extremes. In the second experiment the members of the best and the worst risk class establish support relations only among themselves. In different areas we find closed networks of members of those classes. Members of class 2 seem to be cores of networks often surrounded by members of class 3. Around them we find members of class 4 and 5. It is now class 6 and 7 that are living on the edges of the networks. And it is again towards the middle of the risk classes that segregation is less dramatic.

Experimental Ethics

Class:

1

2

3

311

4

5

6

7

8

9

Fig. 3: First experiment after 1000 periods and 9420 migrations

In the third experiment members of class 1 and 9 could not find partners at all. It is now class 2 and 8 whose members are establishing support networks only among themselves. Often members of class 3 are a nucleus for networks in which they are surrounded by members of class 4, around them class 5, then 6. The members of class 7 are strongly separated, living partly as neighbours of class 6, partly among themselves. For an understanding of what is going on, one should ask two questions: 1. Members of which classes can successfully establish support relations among one another? 2. How attractive are the different partners? As to the first question an answer is given by figure 6, which visualises the PD- and the COOP-condition. The horizontal axis represents probabilities of becoming needy, i.e. p\ and pj- The vertical axis indicates the threshold

312

Rainer Hegselmann

• Class:

9

9

9

9

1

2

3

4

5

Η

•

w

m

m

6

7

8

Fig. 4: Second experiment after 1000 periods and 7020 migrations

a + for probability of stability which makes support relations an equilibrium according to lemma 2. If there is no a + -value, then condition (1) is not met, which is equivalent to the assumption that at least for one of the players mutual support would not pay. Looking at figure 6 and asking when functioning support relations are possible at all, then we can state: -

The risk classes must not be too far apart. Middle risk classes may be further apart than classes at the extremes.

With regard to the probability of stability which is necessary for mutual support the following is valid: -

The further apart risk classes are, the higher the demands on stability.

313

Experimental Ethics

nmmmmm.

mmnm

ΦΦΦ^ΦΦΦΦΦ Class:

1

2

3

4

5

6

7

8

9

Fig. 5: Third experiment after 1000 periods and 14 709 migrations -

The demands on stability rise as the risk classes of pairs of potential partners move towards the extremes.

The last point - a direct consequence of lemma 2 - may be surprising at first glance. The explanation is that two individuals of extremely good or bad risk classes are facing a problem which members of middle classes only have to a much lower degree: It is very likely that members of two extremely good risk classes will not need help. Members of two extremely bad risk classes will probably be in need of help simultaneously and thus neither of them will be able to help. In both cases the situation of unilateral emergency will rarely occur. Therefore the expected utility of mutual support decreases at the extremes. At the same time it becomes more difficult to find out whether the other one is cooperating or defecting. Thus a probability of stability making

314

Rainer Hegselmann

Probability α + a c c o r d i n g to t h e COOP-condition

Fig. 6: PD- and COOP-condition for (S - D) j{M

Risk class

Quite bad

Quite bad

COOP-condition tends not to be fulfilled.

Middle risk

Middle risk

-H)=4

Quite good C O O P - and PD-condition tend not to be fulfilled.

C O O P - and PD-condition can comparatively easy be fulfilled.

Quite good

C O O P - and

COOP-condition

PD-condition tend

tends not to be

not to be fulfilled.

fulfilled.

Table 2: Problems and possibilities of pairing risk classes

support relations for middle risk classes viable might not be sufficient for pairs of extreme risk classes. Table 2 gives a general overview. With regard to our three experiments one can state: With an increasing probability of getting migration options the pessimistic proxy for the probability of stability decreases. Therefore the lens-like area of pairs for which mutual support is a feasible solution is shrinking - see figure 7. So, in the first experiment we had functioning support relations even among members of classes 1 and 2 and, furthermore, between classes 8 and 9. In the second experiment, based on a lower probability of stability, those classes were too far apart to make mutual support a solution. In the third experiment classes 8 and 9 fall victim to the effect that the further decrease of stability made mutual support among the extreme classes impossible.

Experimental Ethics

315

First experiment: pairings for which 0.903 > a + Second experiment: pairings for which 0.810 > a + Third experiment: pairings for which 0.723 > a +

Fig. 7: The lens-like areas contain those pairings of risks for which in the three experiments the COOP-condition (lemma 2) is fulfilled. To understand the often onion-like structures of the evolved networks one must not limit one's attention to pairings which could support each other in principle. One has to take into consideration problems and consequences of different attractiveness as well. The different degrees of people s attractiveness are characterised by equation (C), from sect. 3.5 above. The solidarity surplus Δ/ depends on py, i.e., the others' probability of becoming needy. Δ; can be maximised by finding a partner belonging to the lowest risk class with which mutual support is possible at all. Figures 8 to 10 show for the foregoing experiments the solidarity surplus which an individual that becomes needy with a probability given by the p\axes gets in an interaction with another individual whose corresponding probability is given by the p2-axes. If no surplus is assigned to a risk pair, then for such a pairing mutual support is not possible. Closer inspection of the figures shows: -

Only towards the extremes do the most attractive partners belong to similar risk classes. The most attractive partners for middle classes belong to classes which are significantly better. The highest surplus can be obtained by classes more to the middle. The higher the probability of stability the higher the maximum surplus that can be obtained by each class.

Consequendy, worse classes try to become neighbours of better classes which are looking for better partners as well. Worse classes are satisfied with neigh-

316

Rainer Hegselmann

Fig. 8: Solidarity surplus, first experiment

Fig. 9: Solidarity surplus, second experiment

Fig. 10: Solidarity surplus, third experiment

bouring better classes, but it is just that which makes those better classes willing to move. Closer statistical analysis of those dynamics has yet to be done. But the simulations done so far support the assumption that under a lot of circumstances wider possibilities of pairing, even when combined with higher maximum surplus for all classes, may result in less stability and lower ratios of satisfied individuals in at least some classes (Hegselmann 1994, pp. 380 f.). Another interesting result is that strong segregation of classes may result in a distribution of payoffs which is more equal than that in the primordial soup (Hegselmann 1994, pp. 383 f.). The analysis of the model is yet to be completed. Questions to be answered regard degrees of clustering and segregation, influence of the size of the migration window, different class distributions, consequences of different decision principles for how to use migration options. At the same time it

Experimental Ethics

317

is obvious that the model itself should be improved and extended. Points that will have to be taken into consideration include: introducing costs for migration, entrance barriers, learning about one's own and others' risk classes or a dynamic of the probability of becoming needy.

5.

A Research Perspective: Modelling Moral Dynamics by Cellular Automata

The model presented above belongs to a certain class of models called cellular automata (CA). From a general point of view basic features of a CA are:10 -

There is a D-dimensional lattice. Time advances in discrete steps. There is a finite number of states. At each site of the lattice we have a cell, which is in one of the possible states. (In other words: finite automata are residing in all sites of the D-dimensional lattice.) The cells change their states according to local rules. Locality is locality both in space and time. So the state of a cell in a next period depends upon the states of neighbouring cells in the last t periods. Usually only the last period affects the future. By definition each cell is part of its neighbourhood. The transition rules usually employed are deterministic·., but «ö/z-deterministic rules are allowed, too. The system is homogeneous in the sense that the set of possible states is the same for each cell and the same transition rule applies to each cell. The updating procedure may consist of applying the transition rule simultaneously. Another updating method may be to select cells at random.

-

-

Originally the CA was introduced by John von Neumann and Stanislaw Ulam at the end of the forties (cf. von Neumann 1966) mainly to give a reductionist model of life and self-reproduction. Usually CA-based models ask for a lot of iterated computations. As computers became more available, many new CA-based models have been developed in science and engineering, and nowadays are used extensively. The applications range from crystal growth, soil erosion, diffusion and fluid dynamics to pattern growth and clashes of galaxies. Moreover, CA can be regarded as parallel processing computers and are therefore interesting for computer scientists. Some steps toward classifications of CA have been made (Wolfram 1986), but the analytical understanding of CA has remained poor. Hence simulations play an 10

A short and easy to read introduction to C A is Casti (1992), vol. 1, ch. 3. See also Toffoli/ Margolus (1987) and Wolfram (1986).

318

Rainer Hegselmann

important role. Because there are natural and easy to handle strategies to visualise low dimensional CA, visualisation has become an important tool. Early examples for CA-based modelling in the social sciences are Sakoda (1949, 1971) and Schelling (1969, 1971). The latter analyses segregation along racial lines in a 2-dimensional world. Nowak, Szamrej and Latane (1990) model the dynamics of opinion formation in a cellular framework. Examples for CA-based modeling with regard to cooperation in social dilemmas are Axelrod (1984, pp. 158 ft), Nowak and May (1992, 1993), Bruch (1993), Hegselmann (1994), Kirchkamp (1994), Messick and Liebrand (1995), and Hegselmann (1996a). In the social sciences those models did not see the light of the day under the name of, nor with reference to, CA; they were baptized "checkerboard models".11 For a host of reasons modelling based on low dimensional CA is a promising approach to analyse moral and social dynamics (cf. Hegselmann 1996a): -

CA and a lot of social dynamics have some basic features in common: Over the periods of time numerous people interact, interactions and information are local, there are overlapping neighbourhoods. Given a social process with those abstract properties it is an obvious and quite natural idea to model it in a CA framework, because then CA cover some basic features of a real world process. The very essence of those models can be described as being multi-agent systems, based on locality and an overlapping interaction structure. Under many circumstances the last two assumptions are much more plausible than the claim that interactions between all actors have the same probability or that all actors have (good or bad) information about all parts of the world. Within the world of CA-based models quantitative explanations and quantitative predictions will often be possible. But one can learn about the "real" world as well. The lessons may be called qualitative understanding, which then allows for qualitative explanations and predictions with

-

11

With regard to the above model one may wonder whether we are leaving the C A framework by allowing for migration. So the question is: how to integrate migration in the CA concept? Basically having cells changing their state is sufficient to include migrations: one makes a distinction between two kinds of states, that of being empty and that of being occupied by an actor. All cells are thus either empty or occupied. The state of being occupied can be realised in different ways. For instance, it can be occupied by an individual of this or that class. An actor's moving to an empty cell in his neighbourhood can then be described as an application of a (complicated) rule by which an occupied centre cell and an empty neighbouring one exchange states. Thus including migration does after all not imply an extension of the C A concept. Migration is a matter of interpreting the states or choosing the right set of states. It is in this sense that models with migration are CA-based models as well. For a short history of CA-based modelling in the social sciences see Hegselmann (1996b).

319

Experimental Ethics

-

regard to real world phenomena. In short: CA-based modelling may provide a qualitative understanding of real world phenomena, where at least for the time being we cannot accomplish more. CA-based models make it quite clear how certain macro-effects are dynamic results of decisions and mechanisms operating on a micro-level only. Order, structure, clustering and segregation can be generated by local, micro-level rules. The basic features of CA (interacting cells, overlapping neighbourhoods, locality, and a discrete time) make it almost impossible not to focus on the evolution of macro structures and the emergence of macro properties. As it seems, there is no simpler modelling framework that would allow for this. Under that perspective, CA-based modelling contributes to a better understanding of micro/macro relations.

What has been said above in favour of CA-based modelling does not mean that there are no restrictions. In Hegselmann (1996a) it is shown that there are even severe dangers in using the CA approach, for example the danger of creating mere artefacts caused by badly designed updating procedures. Other problems are due to restrictions on possible interaction structures, caused by the geometry of neighbourhoods. This said, CA-based modelling provides a promising framework for the analysis of social and moral dynamics, as for example the dynamics of cooperation and solidarity.

References ( 1 9 8 4 ) . Robert Axelrod: The Evolution of Cooperation, New York 1984.

AXELROD BRUCH

CASTI

( 1 9 9 3 ) . Eva Bruch: The Evolution of Cooperation in Neighbourhood Bonn University 1993 (typescript). John L. Casti: Reality Rules: Picturing vols., New York 1992.

( 1 9 9 2 ) .

Structures,

the World in Mathematics,

2

( 1 9 8 6 ) . James W. Friedman: Game Theory with Applications to Economics, second ed., Oxford 1991; first ed. first publ. in 1986.

FRIEDMAN

Rainer Hegselmann: Gleichgewichte spiel, Bremen 1991 (typescript).

HEGSELMANN

( 1 9 9 1 ) .

im iterierten

Solidaritäts-

( 1 9 9 4 ) . Rainer Hegselmann: "Zur Selbstorganisation von Solidarnetzwerken unter Ungleichen: ein Simulationsmodeü", in Homann (1994).

HEGSELMANN

(1996A). Rainer Hegselmann: "Cellular Automata in the Social Sciences: Perspectives, Restrictions and Artefacts", in Hegselmann et al. (1996)

HEGSELMANN

(1996B). Rainer Hegselmann: "Social Dilemmas in Lineland and Flatland", in Liebrand/Messick (1996).

HEGSELMANN

320

Rainer Hegselmann

(1996). Rainer Hegselmann, Ulrich Mueller and Klaus G. Troitzsch (eds.): Modelling and Simulation in the Social Sciences from a Phibsophy of Science Point of View, Dordrecht 1996. H O M A N N (1994). KarlHomann (ed.): Wirtschaftsethische Perspektiven I, Berlin 1994. K I R C H K A M P (1994). Oliver Kirchkamp: Spatial Evolution of Automata in the Prisoners' Dilemma, Bonn University 1994 (typescript). L I E B R A N D / M E S S I C K ( 1 9 9 6 ) . Wim Β . G . Liebrand and David M. Messick (eds.): Frontiers in Social Dilemmas Research, Berlin 1996. M E S S I C K / L I E B R A N D ( 1 9 9 5 ) . David M. Messick and Wim Β. G. Liebrand: "Individual Heuristics and the Dynamics of Cooperation in Large Groups", Psychobgical Review 1 0 2 ( 1 9 9 5 ) . VON N E U M A N N (1966). John von Neumann: Theory ofSelf-Reproducing Automata, edited and completed by Arthur W. Burks, Urbana 1996; Theory first publ. in 1966. N O W A K ET AL. (1990). Andrze) Nowak, Jacek Szamrej and Bibb Latane: "From Private Attitude to Public Opinion: Dynamic Theory of Social Impact", Psychological Review 97 (1990). N O W A K / M A Y (1992). Martin Α. Nowak and Robert M . May: "Evolutionary Games and Spatial Chaos", Nature 359 (1992). N O W A K / M A Y (1993). Martin A. Nowak and Robert M. May: "The Spatial Dilemmas of Evolution", International Journal of Bifurcation and Chaos 3 (1993). SAKODA (1949). James M. Sakoda: Minidoka: An Analysis of Changing Patterns ofSocial Interaction, University of California 1949 (doctoral dissertation) SAKODA (1971). James M. Sakoda: "The Checkerboard Model of Social Interaction", Journal of Mathematical Sociology 1 (1971). S C H E L L I N G (1969). Thomas C. Schelling: "Models of Segregation", American Economic Review 59 (1969). S C H E L L I N G (1971). Thomas C. Schelling: "Dynamic Models of Segregation", Journal ofMathematical Sociobgy, 1 (1971). S I M O N (1955). Herbert A. Simon: "A Behavioral Model of Rational Choice", in Simon (1979); article first publ. in 1955. S I M O N ( 1 9 7 9 ) . Herbert A. Simon: Models of Thought, London 1 9 7 9 . TAYLOR ( 1 9 7 6 ) . Michael Taylor: Anarchy and Cooperation, London 1 9 7 6 . TAYLOR (1987). Michael Taylor: The Possibility of Cooperation, Cambridge, Mass., 1987. T O F F O L I / M A R G O L U S (1987). Tommaso Toffoli and Norman H. Margolus: Cellular Automata Machines, Cambridge, Mass., 1987. W O L F R A M (1986). Stephen Wolfram (ed.): Theory and Applications of Cellular Automata, Singapore 1987. H E G S E L M A N N ET AL.

ULRICH

KRAUSE

Solidarity among Rational Egoists: A Reply to Hegselmann

1.

Introduction

As we all know, it took some time for mankind to find out that the earth is not a disk but a ball. In his paper on experimental ethics', Rainer Hegselmann confronts us with the incredible creed "that the world is a torus" (sect. 3.1). Although I am a mathematician, my main interest in Hegselmanns paper is neither in the mathematical model, nor in its implementation on the computer, but in the fascinating idea to find solidarity even among rational egoists. The paper is not easy to read, because it employs methods which are rather new and not yet well-established within the area of moral philosophy, namely techniques from mathematical game theory and the theory of cellular automata. Moreover, since many problems within these fields are not yet analytically tractable, Hegselmann uses computer simulations in obtaining his conclusions. Such methods have been applied already within various disciplines and have led, in the life sciences (biology, medicine), to a new paradigm, under the heading "artificial life". I consider Hegselmann's paper to be a pioneering piece of work in a new area which could be called "artificial society". Proofs are omitted in the paper. While some proofs are not difficult or are standard, the proof of lemma 2 in section 2 as shown to me by the author is lengthy and complicated (Hegselmann 1991). In the following I will comment on what I consider to be the main points of the paper, avoiding formulas and technical details almost entirely. "Experimental Ethics" consists of the following sections: 1. 2. 3. 4. 5.

Introduction How can solidaric relations be modelled? How can opportunistic choices of partners be modelled? Some experiments A research perspective: modelling moral dynamics by cellular automata

Broadly speaking, it is argued in the paper that solidarity is possible even among hard-boiled egoists, provided that these egoists are sufficiently intelligent, and that some technical requirements are met; the paper then goes

322

Ulrich Krause

on to investigate more closely the structure and the properties of solidarity networks which arise.

2.

The Solidarity Game

Two rational egoists are considered, each requiring help with a certain probability (which later on defines a risk class). In case one individual needs help and the other does not, the latter individual may help the former, bringing about a gain for the former and some cost for the latter. Here the gain is measured as the difference of the utilities of being helped and not being helped (S — D in the paper); similarly, the cost is measured as the difference of the utilities of not helping and helping (Μ — Η in the paper). Gain as well as cost are assumed to be positive. Each of the individuals can choose between the strategies "to help" and "not to help". Cost being positive, a rational egoist will prefer "not to help" if the other individual chooses "to help" and will prefer "not to help" if the other chooses "not to help". Therefore, the two individuals will independently decide "not to help" (which is a dominant strategy). The point, now, is that this happens even if it would be better for both individuals to help each other. This is, applied to the present situation, the famous Prisoners Dilemma (PD) which may be rephrased as: 'individual rationality' need not lead to collective rationality'. If we call an individual's solidarity surplus the difference between expected gain (on receiving help) and expected cost (on giving help), then the PD-situation occurs precisely if for both individuals the solidarity surplus is positive. Thus, the solidarity game among rational egoists becomes a PD just in case solidarity would be beneficial for both individuals. In that situation help comes from game theory, where it has been shown that a PD which is played not only once but is iterated with a positive probability (= discount parameter) allows for cooperation, i.e., solidarity in the present case. (Here Hegselmann follows the road opened by Axelrod 1984 and Taylor 1987.) Admitting future rounds of the game with a certain probability gives room to conditional strategies such as TIT-FOR-TAT (TFT) or TRIGGER (TR) for which cooperation does pay. Broadly speaking, this modelling of solidarity may be viewed as a precise explication of the old do-ut-des-pnnc\p\e in a situation where the future is important. For the coop-condition to be meaningful, the rationality of the individuals seems to have to be of such a kind that utility is cardinal (unique up to positive affine transformations). As cardinal utility it is assumed to be the same for all individuals. What makes individuals different from each other are the different probabilities of their requiring help. Utilities are independent of

Solidarity among Rational Egoists: A Reply to Hegselmann

323

these probabilities in the model under consideration. Dependencies could be meaningful nevertheless. E.g., an individuals utility of helping, when rising with the other individuals' probability of requiring help, would portray some kind of altruism. One may wonder why the only conditional strategies taken into consideration are T F T and TR.

3.

Social Structures of Solidarity

In the next step, in section 3, a whole population of rational egoists is considered. Individuals may differ not only in the probability of requiring help (in fact, there are nine risk classes, each consisting of individuals having the same probability) but also by their position. More precisely, in each period of time an individual occupies a particular cell of a cellular automaton on a torus. In the beginning, each individual is assigned to a cell at random ('primeval soup'). This constitutes neighborhoods where each individual can have at most four neighbors with whom it may or may not start a solidarity game of the kind discussed above. For the next period, using a random mechanism, individuals are given options to move to another cell within certain bounds. An individual may use such an option to get to a more advantageous neighborhood with partners from a lower risk class. (These potential partners, however, will take their own decisions!) For carrying this out, the individuals are equipped with a whole bunch of capabilities. E.g., they are assumed to know their own, and the other individuals', risk class, they know gains and costs as well as the probability of getting an option. (Do they also know whether other individuals do take their options?) Moreover, the individuals are assumed to be intelligent enough to process all the data they are assumed to know. On that base, the individuals proceed by maximizing their solidarity surplus as defined before. Even under the assumptions made, it may be difficult for individuals to find a best position. For this reason it is assumed that individuals are satisfied if a certain minimal level is reached. This introduces a further characteristic of individuals, namely the fraction of the maximum solidarity surplus that makes an individual satisfied. This conception of satisfaction raises the following questions: Why does the fraction not vary among the individuals but is assumed to be uniform , e.g. 50%? How is satisfaction, a kind of bounded rationality, related to rationality as assumed for the solidarity game? (Hegselmann addresses the latter question in footnote 10.) Within this framework, Hegselmann obtains by computer simulation a whole collection of very interesting and, as it seems, stable conclusions.

324

Ulrich Krause

-

If mutually advantageous solidaric relationships are possible at all among unequals who can and must make opportunistic choices of their partners, then support networks will show distinct features of some kind of class segregation. - Extremely good or bad risks will tend to have more difficulties in finding partners with whom they can engage in mutually advantageous and viable relations of solidarity than will members of the middle risk classes. — Middle risk classes profit more from solidarity than extreme risk classes. — Class segregation in general leads to a growth of wealth. In most cases noone will be worse off It is not completely clear to me in what sense Hegselmann speaks of the robustness or stability of his results. It seems that he is speaking from the experience of the computer simulations he has run so far. What does he mean by "extremely stable"? The robustness is a crucial point because Hegselmann's model depends on so many parameters, e.g. payoffs, probabilities of requiring help, ability of type identification, neighborhood structure, probability of getting an option, discount parameter, mobility radius, fraction of satisfaction or minimum level. For this reason, but for other reasons too, it would be valuable if in the future more analytical insights (as already done with the lemmata) could be obtained about the present model.

4.

Final Remarks

The economist and moral philosopher Adam Smith, who might be considered as the father of the modern concept of a rational egoist, is not only the author of the Wealth of Nations, but also of The Theory ofMoral Sentiments (Smith 1759).The latter book opens with a paragraph about "sympathy" the first sentences of which run as follows: "How selfish soever man may be supposed, there are evidently some principles in his nature, which interest him in the fortune of others, and render their happiness necessary to him, though he derives nothing from it, except the pleasure of seeing it. O f this kind is pity or compassion, the emotion which we feel for the misery of others, when we either see it, or are made to conceive it in a very lively manner. That we often derive sorrow from the sorrow of others, is a matter of fact too obvious to require any instances to prove it."

Considerations like these could perhaps be integrated into Hegselmann's model by making one individual's payoff for helping an increasing function of the probability of the other person's requiring help. Another point, also

Solidarity among Rational Egoists: A Reply to Hegselmann

325

related to the concept of a rational egoist, is that the common axiomatic treatment of rationality as developed by economists does not allow for any internal conflict on the side of the individual. A convenient way to model a conflict inside the individual is by introducing several distinct preference orderings according to the diverse aspects under which the world appears to the individual. Rationality then is linked to an individual's ability to merge all these distinct orderings into one single overall preference relation. (See Steedman/Krause 1986, Krause 1991.) This is an aggregation problem that formally bears some relationship to Arrows Possibility Theorem and related results on aggregation in welfare economics. Philosophically, the problem consists, lastly, in deriving a concept of the 'self' which takes internal conflicts of the individual seriously into account. With respect to Hegselmann's model this would amount to a notion of individual rationality which integrates such diverse aspects as, say, payoffs and fractions of satisfaction. In the paper a whole catalogue of assumptions is made about the individual as an information processing unit. The rational egoist as portrayed in economics simply maximizes utility or preferences and has no internal structure, neither with respect to morality nor with respect to information processing. How does an information seeking and processing individual arrive at rationality? An obvious fact about information is that, like money, one never has enough of it. This may be a general reason why individuals need help from each other, a reason, that is, for solidarity. Thus the question of building solidarity upon rationality comes back as the question of building rationality upon solidarity, as is, perhaps, indicated in the quotation from Adam Smith. One might also ask why various individual characteristics appearing in the paper, e.g. minimum level, mobility radius and probability of having an opportunity to migrate, are assumed to be uniform across the individuals.

References ( 1 9 8 4 ) . Robert Axelrod: The Evolution of Cooperation, New York 1984. E L S T E R (1986). Jon Elster (ed.): The Multiple Self, Cambridge 1986. H E G S E L M A N N ( 1 9 9 1 ) . Rainer Hegselmann: Gleichgewichte im iterierten Solidaritätsspiel, Bremen 1991 (Ms.). K R A U S E ( 1 9 9 1 ) . Ulrich Krause: "Eigennutz und ethische Gefühle oder: Wie wird man

AXELROD

SMITH

ein guter Egoist?", Ökonomie und Gesellschaft, Jahrbuch 9 (1991). (1759). Adam Smith: The Theory of Moral Sentiments, vol. I of the Glasgow Edition of the Works and Correspondences of Adam Smith, edited by A. L. Macfie and D. D. Raphael, Oxford 1976; The Theory first publ. in 1759.

326

Ulrich Krause

Ian Steedman and Ulrich Krause: "Goethes Faust, Arrows Possibility Theorem and the Individual Decision Taker", in Elster

STEEDMAN/KRAUSE ( 1 9 8 6 ) . (1986). TAYLOR

(1987). Michael Taylor: The Possibility of Cooperation, Cambridge 1987.

RAINER W E R N E R

TRAPP

The Potentialities and Limits of a Rational Justification of Ethical Norms, or: What Precisely Is Minimal Morality? Abstract: Starting from the insight that, due to certain epistemological peculiarities of normative truth', normative statements cannot claim to be objectively (= Ο) true, the paper systematically works out the idea of basing the O-validity of general moral norms on their O-utility rather than on their O-truth. According to this idea any restriction of choice, in an «-person-conflict of interests S, qualifies as O-valid if it fulfills one of the two following criteria: Either compliance to it by at least a specifiable number k of the η individuals in S would make everybody already in each instance ofS better off than norm-free anarchy (= criterion (I), which establishes two classes of unconditionally O-valid norms each avoiding a corresponding type of trap of prudence), or it would, under certain assumptions of the interacting individuals on the probabilities of the roles taken in their respective lifetime-sequences of situations of type S, increase everybody's utility payoff in the long run (= criterion (II), which establishes three classes of only conditionally O-valid norms). Thus even 'non-veiled' rational egoists refusing to initially concede any rationally unfoundable moral protonorm whatsoever, one that demands some (Harsanyian, Rawlsian,...) impartial standpoint in considering an agreement on mutual restrictions of behaviour, will - so it is argued — have to contract on at least these norms in a fictitious original agreement. The latter's extension defines the system Mmin of minimal morality. Though being far more comprehensive than related approaches to 'morals by agreement' (notably Gauthier's), Mmin will finally be assessed as morally insufficient due to its not containing any compensatory norms. Since some of the latter, according to widespread convictions, are indispensable and since these, at the same time, are not justifiable as O-valid on the basis of whatever brand of veil-free contractarianism, any progamme of founding a satisfactory moral system on mere collective rationality is considered as doomed to fail eventually. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

Introduction Can O-Truth Found O-Validity Claims for TV-Assertions? Utility for Everyone as the Basis for O-Validity Claims What Exactly Is a 'Rational' Justification of Ethical Norms in Terms of Utility? A Contractarian Setting for Our Basic Idea Without Conflict of Interests No Reason to Agree to Restrictive Norms Criterion (I) of O-Validity: Unconditionally O-Valid Norms Some Examples of Unconditionally O- Valid Norms Criterion (II) of O-Validity: Conditionally O-Valid Norms Some Examples of Conditionally O- Valid Norms The Extension of'Minimal Morality': A Short Summary of Our Answer to Our Initial Question 12. The Limits of a Rational Justification of Ethical Norms 13. Justice also Requires Compensatory, Nonwelfarist Principles 14. The Motivation Problem: Why Always Follow Afm[n-Norms? References

328

Rainer Werner Trapp

1.

Introduction

Worldwide we face a variety of rival normative (=TV) systems which, according to the restrictions of behaviour they decree and the reasons they give for this, can be subdivided by different trees of classification. This fact gives rise to the following question which I shall systematically investigate subsequently: Are there any criteria to find out which of the TV-assertions made in an TV-system are objectively valid (= intersubjectively valid; for short: O-valid) from a merely rational point of view? To facilitate the investigation let us reduce each TV-system Mj ( j = l,...,i) to its deontic core, i.e. to those TV-assertions which restrict (via general norms of prohibition or obligation) the sets Als of act alternatives of individuals i G Η in situation-type S to the Mj-permitted choice sets Mj{Als) C A's. The issue of the justifiability of O-validity claims for any kind of further TV-assertions which Mj may contain beyond its deontic core (such as evaluative assertions on motives, persons, characters etc.) will not be broached here.

2.

Can O-Truth Found O-Validity Claims for Ν-Assertions?

Traditionally, claims of O-validity for restrictions of behaviour rest on various ways of 'recognizing the objective trutti as to the deontic status of acts. These ways comprise such heterogeneous approaches as (1) pretended 'privileged access' of founders of religions, prophets, priests or the like to (a) the will of some divine being, or (b) (as in ancient China) to 'the eternal moral laws guiding all (!?) cosmic events'; (2) philosophers' insights into (transcendent, metaphysical, evident, a priori,...) sources of moral truth such as (a) the TV-departments of'Platonic worlds of ideas'; (b) 'eternal hierarchies of material values'; (c) certain formal principles, 'a priori true for all reasonable beings' and allegedly implying adequate moral restrictions for all types of choice situations; (d) 'the TV-presuppositions of assertive speech-acts', said to define a set of'ultimately founded (= letztbegründete) norms of discourse' the denial of any of which would imply a 'pragmatic contradiction'; (e) certain evident 'natural rights' serving - as do the 'norms of discourse' in case (d) - as a point of departure for establishing further norms in some contractarian way, etc. If any pair Mj,Mfr of the TV-systems erected on such an allegedly 'objective' basis is deontically nonequivalent, i.e. if there is at least one possible S and i such that Mj{A*s) φ M^A^), then, already for the formal reason of deontic consistency, not both systems can be O-valid as a whole. Yet even if for all

What Precisely Is Minimal Morality?

329

pairs Mj,Mk there were no act forbidden by the one and permitted by the other system (so that all these systems were at least arrangeable as a sequence of subsystems of each other or were even all equivalent) I would not concede O-validity to any of them. For all the arguments given for the afore-named sources of moral O-truth fail to convince a mildly sceptical rational being (at least me) of the justification of this pretension. This also holds for the relatively most promising Kant-inspired 'universalizability approaches falling under (2c). But lack of space prevents further discussion here. Before examining what can be taken as an argument for the O-validity of

at least certain kinds of norms, if O-truth cannot, let me point to what I believe to be the root of the predicament in claiming O-truth for any kind of TV-assertion.1 (1) Suppose e.g. that O-truth is claimed for the singular statement that person a ought to do F in S (for short: 0(F$a); let us leave aside here the dispute over how to formalize conditional obligations). How can this claim be checked by rational testers? The reference-semantics for deontic logic, presently the best place to look for truth conditions for obligations, in substance all agree on this: "0(F$a)" is true in the real world WO iff "F$a" is true in every possible world W which,

viewed from WQ, is (1) accessible, (2) technically implementable, and (3) morally perfect. Thus, in order to find out whether "O(Fsa)" is O-true in Wo each tester has (a) to select the subset of all possible worlds W which stand in the mentioned triple relation to Wo and then (b) to check whether "F^a" denotes a fact in each of them. But these perfect worlds W are not given the way the real world WQ is 'given (in a mild, not naive-realistic sense of this term) to the testers' senses when they check whether the empirical (=F) statement "F$a" denotes a fact in Wo. Rather, they may be quite diverse constructions in the minds of different testers. What appears, with respect to a choice situation S, to be morally perfect to a tester χ is contingent upon x's personal history of value formation. This naturally explains why for any pair of Ν-systems there will be some F and S such that adherent χ of Mj intuits that "Fsa" denotes a 'fact' in 'his' perfect worlds WX, whereas adherent j/ of M^ intuits the very contrary for 'his' perfect worlds WY. We do not have to recur to such different 'deontic observers' as a European liberal and a Shiite mullah to therefore find disagreement over the truth value of "O(Fsa)" for many F and 5, where at the same time, due to sufficient uniformity of the cerebral end product of sensory perception of different observers in the same stimulus situation, they perfectly

1

These matters are discussed in detail in Trapp (1988), pp. 105-67, and Trapp (1990b).

330

Rainer Werner Trapp

agree on the truth value of each of the corresponding ^-statements "/^-λ"2. Moreover, contrary to what the Ε-case mostly permits, it is principally impossible in the Ν-case to neutrally decide who is right and who committed a value perceptional' error. The lesson from this, which mutatis mutandis also holds for other kinds of TV-statements, is obvious: Though truth-values for TV-statements allow us to deductively argue in moral matters, this does in no way help ethical Ocognitivism. For epistemology reveals this truth to be a merely subjective one as long as the perfect worlds of different 'moral observers' happen to differ. A corollary of this is obvious: Doubtless one can test, as is often done, general TV-principles Hjy by modus tollens in perfect logical analogy to the testing of general Ε-hypotheses Ηβ. Both in the E - and TV-case testers can, however deeply embedded in theory and paradigm Η χ and Ηβ may be, make all the necessary theoretical TV- or ^-assumptions and then let their TV- or ^"-perception decide over the truth of particular implications deducible from this. Yet there is no doubt, either, that, as far as establishing the provisional 0-truth or definite O-falsity of HE and Hpj is concerned, there is no epiitemological analogy between E- and TV-falsification. For only in the Ε-case can one in a theory-neutral way specify the criteria for a 'normal, exchangeable observer' capable of deciding (as the only or at least final link of a chain of theory neutral measurement instruments in some experimentum crucis) about the provisional O-truth of rival hypotheses Hp and Hg. In all corresponding Ν-cases of testing rivals Ηχ and H^ (where the observer' like in most nonscientific every day Ε-cases is the only measurement instrument' involved) it makes no sense to likewise speak of a theory-neutral, normal and exchangeable observer. Exchanging him might possibly reverse the very result of the 'experiment'. The results of 'tests' of general TV-principles are therefore significant only to the particular testers themselves. If χ registers a positive outcome for, say H f j and a negative one for its rival H ^ , this does in no way provide a basis for justifiably claiming provisional O-truth for Hjy and O-falsity for H ^ , as it normally would in an analogous case of rival (only universally quantified) ^-hypotheses decidable by sensory perception.

2

Contrary claims of extreme epistemological relativists, who insist on an analogous observer relativity in sense perception, rest on gross misinterpretation and overestimation of evidence stemming from gestalt, perceptional and social psychology.

What Precisely Is Minimal Morality?

3.

331

Utility for Everyone as the Basis for O-Validity Claims

But what can be done if the O- validity of any moral norm cannot be justified by its O-truth? Should we simply assume that all moral norms are on a par with norms of cuisine or aesthetics which are true only for certain subjects? Such egalitarianism would be blind to the fact that some moral restrictions are, in a sense to be specified later, undeniably useful to everybody (= O-useful), some are O-usejul under certain empirical assumptions, others under certain additional normative conditions and still others not O-useful at all. It seems reasonable, then, to base different grades of O-validity on corresponding grades of O-utility. The gist of this idea of justifying norms by their utility rests on a wellknown tradition. It even dates back to an ancient source usually not quoted in this context; to my knowledge it was Epicurus who first clearly expressed this view in his Kyriai Doxai. In KD xxxi he writes: "What, with regard to (human) nature, is just is an agreement based on the utility (σύμβολον του συμφέροντος) of mutually not harming each other and not getting harmed" (my translation). KD xxxin reads: "Justice (δικαιοσύνη) is nothing existing in itself but [...] a contract (συνθήκη) not to harm (others) and not to be harmed (by them)." In KD xxxn, and particularly xxxvi-xxxvin he expands this Veil-free' contractarianism, repeatedly paraphrasing that legal norms have to be justified by their "utility for the social interactions in which individuals are engaged". Better known than this initial programmatic hint is the series of its subsequent realizations elaborated in quite different ways by de Mariana, Suärez, Hobbes, Locke, Hume, Spinoza, Rousseau and, more recently, Buchanan, Nozick, M. Taylor, Friedman, Axelrod, Schotter, Gauthier, Sugden and others. Their undertakings of rationally (a) justifying and/or (b) explaining social institutions like conventions, law and morality, and different versions of the state as an agency for enforcing the latter also stimulated a series of German publications by Hegselmann, Hoerster, Kliemt, Raub, and Voss. In hereafter tracing out in some detail the TV-system that, I think, can be grounded on the common basic idea of all these authors ('Morality can be essentially grounded on utility'), I firstly confine myself to the normative aspect (a) and, secondly, renounce any underpinning of utility arguments by a basis of rationally unjustified natural rights (as did some of these authors, including Nozick). I shall attempt to demonstrate that this TV-system is much more comprehensive than so far thought. For it can be shown to contain considerably more (kinds of) moral norms - each justifiable in a merely rational way without

332

Rainer Werner Trapp

recourse to any moral prerequisites like an 'impartial, neutral standpoint' to be taken by everybody initially - than any of the authors mentioned (to my knowledge) argued that it did. Nonetheless this comprehensive N-system will finally turn out to be morally unsatisfactory, whence I shall conclude — not at all with philosophical delight - that mere rationality provides too meagre a basis for establishing a sufficiently comprehensive system of moral restrictions.

4.

What Exactly Is a 'Rational'Justification of Ethical Norms in Terms of Utility?

The examination to be undertaken will make use of a concept of individual rationality according to which rational individuals as such always maximize (expected) effective utility. What precisely is effective utility? Let S be any strategic or nonstrategic choice situation affecting the individuals in H$ = {1,..., n}. Let, for each i £ H$, the particular utility function ut over the possible outcomes x,y,... in S be that one which i would manifest if i assumed all j ( j φ /') to be indifferent between all these outcomes. (Thus the different goodness of these outcomes for others cannot codetermine i's particular preferences for them.) Let us then assume interpersonal unit comparability and model i's effective utility function u*, which also includes the degrees of is external preferences for the well-being of others in H$, by a graded consideration of all particular utility functions involved, i.e. by the weighted average u) = α'j · «ι + . . . + a) • Ui + ... + a'n • un η with ^ a'j = 1 and a'j > 0. j=1 (If also negative external preferences (i.e. a'j < 0) occur, the a'j can, of course, no longer be taken as weights. The weak order by ' > ' over ( a j , . . . ,a'n) may be due to i's decreasing affection or sympathy for the j Ε H$ or to whatever criteria.) The subclass Es C Hs ofpure egoists, evidently, subsumes all i in S fixing on α' = 1, thus basing their choices on u* = ut only. Analogously, diverse kinds of naturally inclined welfarists are characterizable by special weights αγ,..., oc'n. Naturally inclinedpure altruists e.g. choose a\ = 0 (distributing weights in some way or other over Hs\{i}); naturally inclined utilitarians prefer a\ = ... = aln — -η etc. Rational individuals in general will then be those who, as we said above, seek to maximize their (expected) effective utility u* in each S. This, of course,

What Precisely Is Minimal Morality?

333

implies for all possible (a'j,..., a'n) that in cases where a) > 0, «,·(*) > Ui(y) and Uj(x) > Uj{y) for all j φ i hold, i will by choosing χ maximize ut too. Only pure egoists always seek to maximize «,·. This clearly is a more general concept of individual rationality than the one usually applied. It permits us to consider a rational decision maker as a utility maximizer without presupposing him to be an egoist and, at the same time, without already imputing to him any self-restrictions stemming from some (however defined) moral point of view. On the basis of this concept of individual utility maximization this sections questions can now be answered somewhat more precisely than by the crude statements of section 3: A rational justification of a general ethical restriction Ε consists in demonstrating the following: If agreed upon among the individuals 1,...,« and generally (or at least sufficiently) complied with Ε would, up to the border ofPareto optimality, enhance every is effective utility in comparison to what i would receive by alternatively acting in an intelligent, unrestricted wayi i.e. by 'remaining in anarchy as for the situations regulated by Ε. For the limit case of pure egoists an analogous enhancement of all particular utilities involved has to be proved in order to thus rationally justify Ε. These ^-produced real utility gains for everybody (i.e. not only fictitious gains like those obtained from behind a veil of ignorance or from any other already moral point of view) and, consequently, the O-validity of Ε can be of two different degrees: Criterion-(I)-noims (see sect. 7 below) guarantee such gains for each role taker in every single instance of an interaction situation S in which Ε is sufficiently obeyed. Criterion-(II)-norms have less beneficial effects. They produce only long-term expected utility gains for each i provided i makes sufficiently favourable empirical probability assumptions about his particular roles in the series of future instances of S. These assumptions (which are not needed in justifying criterion-(I)-noxms) must be such that the following holds: The expected overall utility losses which i would have to suffer by also obeying Ε in all those instances of S where his role in 5 is unprofitable (so that a general permission of unrestricted maximizing of u* here would serve him better), must be more than outweighed by i's expected overall gains obtained in all those instances of S where general compliance with Ε makes i profit in that it protects him from losses which his role in S would have to suffer in a state of anarchy. The subsequent formal characterization of the properties of the different types of conflictive situations to which criterion-(I)- and criterion-(II)-norms, respectively, apply plus corresponding examples of such situations will make these hitherto rather abstract ideas much clearer. Note that, before we take the notorious dilemmata of rational egoism as a runway for getting off the ground of moral relativism the first subgroup of

334

Rainer Werner Trapp

criterion-(I)-norms, there is a still more principal predicament of rational egoism. For, as Gauthier showed in a valuable impossibility theorem in (1974), rational egoism in general is not even logically possible. One can conceive games in which some plausible conditions of egoistic utility maximizing in strategic contexts turn out to be jointly incompatiblel

5.

A Contractarian Setting for Our Basic Idea

Imagine that all traces of any N-system were suddenly wiped out of our memories. (Inspire your fantasy by looking at some of the results of recent history.) Instead there is only one generally accepted norm, the nonrestrictive principle of anarchy (= A): "Each i e H$ is, in each type of situation S, permitted to do whatever from his A's individual i thinks is best for him." Let us carry the fiction further still and assume (a) that all human beings are ideally rational (consulting decision and noncooperative game theory in their striving for «^-maximization), (b) that some timid characters, somewhat uneasy about the prospects in a state of nature where life threatens to become 'poor, nasty, brutish, and short', have organized a gigantic (say, home computer) all-mankind-conference aiming at a contract on at least those minimal moral (= Mmin) restrictions that, benefitting everybody, would find consensus and thus become O-valid. What kinds of norms would a rational participant i 6 H$ then agree to? Note in particular that i cannot, without petitio principii, be required to ground his assent to Mmin- norms on any already moral point of view. Thus any justification of a norm that requires a Rawlsian or similar starting point can, from the beginning, not count as merely rational. Two things are then clear from the start: (1) For all types S of situation affecting only i himself i will have no reason to agree to any moral norm N(S). If i believes N(S) to be useful like many nonmoral restrictions ("Don't eat five eggs in a row"), he will choose (or forbear) the act(s) prescribed (or prohibited) by N(S) quite by himself. If he thinks otherwise, he will simply disregard Ν(S) without harming anybody else. In either case there is no reason to include N(S) into Mm\n. So why agree to the prohibition of suicide, certain diets, clothings, autoerotic pleasures, dancing on Sundays, working on the Sabbath, eating in daylight during Ramadan, believing in more or less than just one god etc., if really nobody else is affected by such acts even in the long run? (2) Likewise any moral restriction will be superfluous (except the one to be named hereafter) for all S in which n(n> 2) individuals are affected in a nonconflictive way. (a) If S is nonstrategic (i.e. involves only one agent and η — 1

What Precisely Is Minimal Morality?

335

passive recipients of (dis) utility) and if the utility vector of one act (or those of some equally good acts) strongly dominate(s) all others, the agent will choose it (one of them) anyway, (b) In strategic situations with exactly one strongly dominant outcome everybody in Hs will likewise have a self-interest in doing his part to produce this outcome. Thus noncooperative individual rationality here is enough to guarantee Pareto-optimality of outcomes. The exception mentioned would be a first O-valid norm N\ concerning exclusively cases S* where the Pareto-optimal outcome x* only weakly dominates all the others. N\ is the following mild, 'semimoraT demand: "Everybody in S* who would neither gain nor lose by producing or (in strategic cases) coproducing x* ought nonetheless (co) produce x*." Each i would, even as a pure egoist, have reason to agree to N\: S* being possible, i might with p> 0 get into S* taking, with the probability distribution q(r\),..., q{rn) any of the η roles r,· obtainable. Then by a general compliance with N\, as N\ requires, i would gain expected (effective) utility: Being among those who would not gain by x*, i would never harm himself by following N\ in favour of others. Being in a role that benefits from x*, compliance with N\ by others would make i profit. For non-egoists it would pay still more, in terms of effective utility, to agree to N\. What could i gain by agreeing to any further norms prescribing behaviour in nonconflictive situations, e.g. those restricting the pleasures, of whatever kind, consenting adults may share with each other? Note that taking hard drugs or, in the era of AIDS, e.g. certain sexclub activities, cannot count as nonconflictive as, in the long run, other people may be harmed. One might also agree to restrict certain genuinely nonconflictive acts, if committed by children, under the influence of drugs, or (by sober, adult bachelors not in the terminal stage of cancer) in the "Oxford Club for the Cultivation of Russian Roulette" etc. Yet all these restrictions would, rather than qualify as moral norms be paternalistic corrections of those individuals' utility functions based on doubts as to their balanced consideration of short- and long-term consequences. General agreement to a class of further norms would, of course, be rational for strategic situations where compliance with N\ (if necessary at all) leaves more than one weakly dominant optimum to aim at or where more than one strongly dominant optimum, under lack of communication, produces a coordination problem. But such pure coordination norms, void of even minimal moral content, do not belong into Mm[ n , but into a self-enforcing separate pure coordination agreement.

336 6.

Rainer Werner Trapp

Without Conflict of Interests No Reason to Agree to Restrictive Norms

Thus we are justified in saying: All moralproblems to be solved by norms ofMmin presuppose some conflict ofinterests. Let a strategic or nonstrategic «-person-choice-situation S (with η > 2) be conflictive if there is disagreement about the optimal outcome, in that there is no outcome x, such that for all y and /: u*(x) > u*{y). This permits both agreement and disagreement over the ranking of suboptimal outcomes. To keep each conflict of interests as sharp (and thus the need for devices solving it as great) as possible I shall, from now on, argue on the basis of the most pessimistic possibility that all rational individuals are always pure egoists or, said differently, that everybody always defines his effective utilities by only his particular utilities. Limiting our attention to this unpleasant special case does not at all weaken the force of our arguments concerning the O-validity of certain norms. On the contrary: Rather than being accepted by only egoistic rational individuals, our basic argument that any norm whose general compliance benefits everybody should meet consent in this fictitious Mm\ncontract will be accepted by less egoistic individuals as well. Instead of rejecting norms of Mmm nonegoists will, the more altruism they manifest, the more readily agree to further moral norms which (unfortunately) cannot be proved to be O-valid any more, due to their demanding utility transfers to others which even in the long run remain unilateral. Since those further norms do not pay for everybody, their justification cannot be given on a purely rational basis in the sense of section 4. Rather, their acceptance by 'overall payers' requires the additional arational motivational component of sufficient sympathy, solidarity, or affection for those that are the respective norm's overall beneficiaries. An elementary fact motivates our confinement to the counterfactual assumption that everybody in each conflictive situation is an egoist·. If in whatever conflict S all i were naturally inclined utilitarians there would not remain any problems of Mmin S. The proof is obvious: All S without any conflict of interests pose no problem for Mmin. Consider, then, any S that involves some conflict of interests in particular utilities. Take any of the outcome vectors u(xj) = (u\(xj),..., u„(xj)) ( j = 1 , . . . , m). If each i were a naturally inclined utilitarian we would for i = 1,..., η equally get u

1 n K j ) = -n Σ »*(*/)• k=\ x

The effective utilities within each vector would be the same. Then, however, all vectors could (at least partially) be ordered by strong Pareto-dominance. Thus,

What Precisely Is Minimal Morality?

337

if there were no (nonmoral) coordination problem, S would have a trivial e f f i cient, strong equilibrium solution. What was a moral problem on the basis of u* = Ui has vanished for naturally inclined utilitarians. A similar result holds for the general case of nonegoistic effective utilities (for which the proof is all but trivial): Decreasing dispersions (relative to some dispersion measure) of the (a'j,..., a^) and thus of the (u\,..., «*) imply a slimming down of the conflict of interests; this in turn reduces the range of effective utilities for which in strategic situations inefficient noncoop-solutions suggest certain coop-norms. The need of A/m;n-agreements here gradually evaporates. As an illustration of this take the following l-ego'ist-prisoners'-dilemma (— PD2) Si: Si

C

D

C

8,18

2,20

D

10,2

3,3

The Λ/min-norm indicating how to avoid by pure strategies the inefficient noncoop-solution (D, D) — (3,3) would, of course, decree cooperation in S]. (Allowing mixed strategies too would open some more possibilities.) If the a' now decrease, first the PD2-structure and then the conflict of interests dissolve, rendering at the latter point all moral devices for efficiently solving the conflict of interests superfluous: First, for α] < and a.2 < § the dominance ofD vanishes for both players. Still a little less egoism, namely α} < I and a^ < even makes C dominant for both, thus yielding a unique strong equilibrium noncoop-so\\it\on (C,C) = (> 9|,> 15^) which strongly Pareto-dominates all other cells. The conflict of interests has dissolved in an efficient way, before any Μmin-coop-norm can become useful. If (all other payoffs left unchanged) the payoffs (8,18) for (C, C) were lowered (within the range permitted by a PD-structure) up to the point where they become smaller than those of some jointly randomized strategies of mutual exploitation, so that

and

ux(C,C) < />-10+(l-/>)-2< 10 1 „ n y F> ' F > for some p {0 < p < 1), u2(C,C) uz^y e : Ay) and n k (a 2 ) 3z G Hs : uz^y G H s : Ny) > uz^y G Hs" : Ay), and (β)

^3N'(N'

φ Ν) such that:

(ßi) Vz G Hns : uz^y ( fe) 3z G Hns : uz^y

G Hks : Ny) > uz^y G Hks : Ny) > uz^y

6 Hks : Ny) and G Hks : Ny)

340 (lb)

Rainer Werner Trapp

Ν decrees a unique coop-compromise-solution in S if there is more than one efficient, strong equilibrium in S, which is stable under reductionk

{Criterion (II) will be presented below, in section 9.) "Ny", "Ay", ... in criterion (I) stand for "y follows N", "y follows the nonrestrictive norm of anarchy A", ... Since any Ν fulfilling criterion (I) meets general agreement for solving conflicts of the specified type without any further conditions, criterion (I) defines the subclass of Af m ; n -norms which I call unconditionally O-valid (= U O-valid). They comprise, put somewhat less precisely, all those norms which show the highest grade of O-utility: For they would, in each single instance of a situation-type S, if only followed by at least k of η players, either (la)

(lb)

produce some efficient 'shifi to the north-east' of Rn, i.e. (αϊ) make each player obtain at least as much as he would in anarchy, and (02) make at least one player better off than in anarchy, and (β) achieve this in an efficient way, or would efficiently avoid outcome-uncertainty in shaky situations with more than one possible, efficient noncoop-solution.

A few comments on criterion (I) and some examples of norms illustrating it might be helpful: (1) What is meant by "S-payoffs in anarchy" is obviously relative to the specific S-solution-vax'xznts of noncoop-game theory. Though I prefer the solutions in Harsanyi (1977), I leave it open here which noncoop-S-solutions define anarchy. They all agree, on at least this much: applying criterion (I) to them would generate a weighty class of U O-valid norms. (2) One might argue for replacing ">" in (αϊ) by ">" and dropping (012) from (la). As χ might at least once be in a gainer's role in S, however, the weaker requirement made in (a) suffices to make a rational individual χ agree to N . (3) Considering the fertile distinction between foul-dealers and free-riders made by Pettit in (1986) I dub potential foul-dealer-situations (= FD-S) all those which require k — n in to make cooperation (at least weakly) dominate anarchy. In other words, in a FD-S already one defector ('the foul-dealer) would make at least one of the η — 1 cooperators lose compared to his noncooppayoff. Obviously, a lone defector in any PD 2 would be a foul-dealer. 4

An equilibrium-vector (u\,...,u„) is stable under reduction iff any (n — k) -Lateral defection (1 < k < η — 1) does not pay for the defectors. So any equilibrium («1, «2) is per se stable under reduction. For an exact recursive definition see Gauthier (1974), p. 452. (See also example 5 n in sect. 10 below.)

What Precisely Is Minimal Morality?

341

Potentialfree-rider-situations (= FR-S), on the other hand, permit (a) and (β) to become true already for k < n. So a FR-S allows a specifiable maximum of defectors, whose free-riding would not fiilly annihilate anybody's benefit from sufficient cooperation (though the free-riders' extra gains will typically reduce the cooperators' payoffs relative to full cooperation as the subsequent example S'5 of a public-good-game will illustrate). This leads to question (B) "Why always follow TV?" even if it were uncontroversial that question (A) " What norms are justifiable on mere rational grounds?" is satisfactorily answered by criterion (I) and the subsequent criterion (II). If by unilateral defection one would in both FD-S and FR-S gain even more than by cooperation and if in each FD-S a lone foul-dealer would make at least one cooperator even worse off in S than he would be in anarchy, why then always follow U O-valid norms? Why not only follow them if this pays in view of the strategies that are presumed (or even known) to have in fact been chosen by the others? In the particular case of PD"-games (n> 2), where by definition of PD" noncooperation individually even dominates cooperation no matter how the η — 1 others choose to act, this runs up to the paradox that, from the standpoint of individual rationality, one never ought to follow a norm which, being justifiable by collective rationality, one clearly ought to agree to in our fictitious M^m-contract. So why ever follow the respective PD"-norms one has agreed to? Let us defer this notorious question here — it will be taken up at the end of this paper - and insist, contrary to the radical critics of any merely rational justification of morality, who do not carefully distinguish between questions (A) and (B), that even if (B) were unanswerable this would not invalidate the fact that pointing to A/ m j n is definitely a satisfactory answer to (A). (4) Criterion (I) and criterion (II) below refer to the general case of asymmetrical games in which each player may take a different role rt (defined by different numbers and/or types of strategies in his alternative set and/or pay] off asymmetry even if A's — A s for all i,j in 5). In the special cases where all these factors are identical the games are symmetrical. The norms applying to the latter are less complex than those for the general case. Instead of decreeing to each r, a specific strategy (or strategy mixture) from his A's, they simply decree the same strategy a to all rt. (5) Norms aiming at theoretical coop-optima will often decree certain jointly mixed strategies. This, however, will in many (though not all) real life instances of such games be impracticable. Therefore, it is useful to reduce complexity by also pointing to coop- norms for pure strategies only. (6) Another distinction is more important still: If there is only one efficient coo/>-solution Pareto-dominating each noncoop-solution in 5 (as in any

342

Rainer Werner Trapp

PD"(n > 2), if mixed strategies are excluded or, in most public-good games) then the respective M m ; n -norm results directly 'from the logic of S\ Yet this is not so in the games to which part (lb) of criterion (I) refers. For here already in pure strategies more than one efficient Af m i n -norm might be agreed on because there are n(n > 2) efficient, strong (and thus pure) strategy equilibria stable under reduction, each of which is preferred by a different player. Other games, falling under criterion (la), may admit several possible Mmmnorms since their noncoop-solutions, though being unique, are inefficient relative to a continuum of possible pure and/or jointly mixed coop-solutions. (Some examples below will illustrate all this.) Generally, in such cases every norm which, relative to the status quo point (= SQ) of anarchy, decrees some Pareto-front-point of the convex hull around the payoff space in 5 is a potential Af m i n -norm. Which of these norms is agreed upon in our fictitious conference could either be decided by a separate negotiation for each S, or by an advance agreement on some basic bargaining model (= BM) as a general algorithm for putting out coo/>-solutions for such S. There are several 571/-rivals on the catwalk; the most important ones (for Mm\n-purposes), at present, are those by Zeuthen (1930), Nash (1950, 1953), Kalai/Smorodinsky (1975), Kaneko/Nakamura (1979), Gauthier (1978, 1985, 1986), and Perles/Maschler (1981). (Lack of space prevents me from outlining these BMs and their ethical implications5 here.) Guaranteeing, by their very nature, to everybody in S an outcome that is at least as profitable to him as his 5Q-payoff would be and therefore being a mere device for distributing utility gains, each of these BMs is essentially noncompensatory. This trait puts BMs into sharp contrast to those utility aggregation functions which are compensatory in that they also allow us to attribute utility losses to certain recipients, provided these are more than outweighed by the gains of others to such a degree that a situational maximum is reached by the respective distribution. Yet this possibility of their compensatory use does not preclude, of course, that utility aggregation functions may also be applied in choice situations where only utility gains can be distributed in different ways. Thus, rather than on a B M the Afmjn-contractors might also agree on some utility aggregation function as an arbitration scheme - defying the hardship of also interpersonal utility measurement - and restrict its use to only the non5

Highly substantial axiomatic discussions of the formal characterstics of BMs are given in Roth (1979) and Klemisch-Ahlert (1991b). The Kalai/Smorodinsky solution is there proved not to be even weakly efficient for all S if η > 2, the Gauthier solution to require for η > 2 restrictions as for the shape of the hull. Various refinements of basic BMs are collected in Roth (1985) and Binmore/Dasgupta (1987). Trapp (1993) presents a concise survey of the most important BMs including an extensive criticism of their applicability for ethical purposes.

What Precisely Is Minimal Morality?

343

compensatory one needed here. Not contracting on unique A/ m ; n -norms also for these 'many-coop-solution-games - by whatever procedure — would in any case harm everybody. Yet there is one problem specific to these norms. The BM-solutions underlying them depend crucially on the SQ.6 There are, however, various nonequivalent ways of characterizing the latter. One may take a securitylevel-SQ, i.e. the pure or mixed sttziegy-maximin point (= the Shapley SQ = SSQ or SSQmix). In unprofitable games, with no strong equilibrium at all or where the security level and the only strong equilibrium point coincide, this seems a natural basis for negotiating. (Such is e.g. (D,D) in all PD 2 s with a Pareto-front north-east of (C, Q . ) In profitable games with a unique but inefficient strong equilibrium north-east of maximin this equilibrium could itself be the SQ (= ESQ). In other games, notably those with at least two efficient strong equilibria one of which dominates psychologically', one might suggest to construe the SQ on the basis of the threat potential contained in all the possible noncoop- outcomes in S7 (= the Nash-SQ = NSQ). Kaneko/Nakamura, finally, propose the same SQ for all bargaining games S (= 'the worst state for all individuals that we may imagine' = KNSQ) which as such, contrary to all those mentioned, is not defined in terms of any specific S-payoffi. (This will turn out to be ethically less vulnerable.) Depending on which SQ is agreed upon the coop-benefit shares for S can be distributed quite differently among the roles rt in 5. This is less grave, however, for our 'original contract'-bargaining over general norms than it would be for single case bargainings, because these differences will be evened out the more equal are x's chances p{ri) (or relative frequencies in the long run) of taking any of the r/(z = 1,..., n) in 5.

8.

Some Examples of Unconditionally O- Valid Norms

Let us finally illustrate criterion (I) by some examples of norms that satisfy it. Examples (a) to (d) refer to part (la), the other two to part (lb) of criterion (I). All utilities are cardinal. (a) Take situation Si, a special case of a 3 x 4-PD1-extension with the noncoop-outcome (^3,^4) = (1,2):

6 7

A general formal proof of this intuitively obvious fact is given in Klemisch-Ahlert (1991a). See e.g. Luce/Raiffa (1957), p. 110, Harsanyi (1977), pp. 129 f., or Hamburger (1979), pp. 137-40.

344

Rainer Werner Trapp

a2

bx

b2

2,2

bA

3,4

h 1,2

0,12

2,1

2,1

2,2

0,3

10,0

4,0

3,1

1,2

The asymmetric norm N(S2) : Vx(If χ is in S2, then Λ: ought to play ua\ in ri" and "b2 in r2") could be the only pure strategy coop-agreement here. A norm decreeing (a2, b$) would, though satisfying (a) of (la), fail to satisfy (β). If mixed strategies were allowed, too, Ν(S2) would be deficient. Any point on the line segment between (10,0) and (0,12), which also is north-east of (3,4), would be a better coop-so\\xüon than (3,4). All BMs mentioned yield solutions near the midpoint of the Pareto-front. Any such mixed-solution-norm 7V(S2)mix could in one-shot games be complied with by ζ joint randomization yielding this solutions expected utilities. In S2-supergames between the same players one might also agree to alternate between (03, b\) and (a\, £4) in due proportion. (b) Let be any symmetric 2-player-game with options C or D where (C, C) lies north-east of every line connecting the other outcomes, yet is not the noncoop-so\\ix\on. Then, Ν (S3) decreeing C to both, suggests itself as the coo/>-solution. (c) 54 is one of those shaky 1 Chaos- games (as I call them) which offer no better noncoop-solution than maximin (with SSQ = (3,1) or SSQmix = (3,1.7)), since

S4

b\

b2

a\

-2,8

11,1

a2

3,-1

10,2

they only have a mixed strategy equilibrium, which is always weak and thus unstable. Its coop-solution will yield some norm Ν(«S,4)m;x that, relative to the selected SQ-variant, decrees a bargained coop-so\uüon on the Pareto-front part of the line segment between ( - 2 , 8 ) and (10,2). (d) 55 is a variant of a public-good (— PG) game with PD"-structure: Suppose each i (i — 1,..., ri) disposes of a sum st of savings; i is offered a whole spectrum of options: he may invest any part s* (0 < s* < s;) into some PGfund. The money accumulated there would increase by a factor c (1 < c < ή). But, PGs being non-excludable, each i would eventually receive the same share of the sum invested, whether or not i contributed anything. How much should a rational egoist invest?

345

What Precisely Is Minimal Morality?

The noncoop-answet is: Nothing! For any number j (0 < j < τι — 1) and any investment vector of co-contributors i would be better off by a ziTö-contribution (= C,(0)) than by any contribution s* > 0 (= C(s*)), as the following inequation shows, which for all values J,s*-,st-,s* is true if, as assumed, c < n: c-(O) = ( * i + ...+*;·) J

· - + *

η

η So rational egoists, knowing this, would all play Q(0) and end up at the payoff vector (s\,...,sn). This, however, is grossly deficient for many (though not for all) values j,s*,sj,s*. Take the special case S'^ of S5 with equal disposable amount s\ — si — ... = sn = s. Here everyone's maximum contribution s] = s would yield (c-s,...,c-s) instead of anarchy's poor (s,..., s)! Indeed, the norm N(S'^): "In S'^ invest all of j!" would be the only c0ö/>-solution satisfying criterion (la). But even in the general case of S5 all norms decreeing contributions up to = min(ji,... ,s„) would yield a coop-vector of (... ,Sj + 4 i n · (c — 1),...) and thus dominate anarchy's {. ..,·$•,•,...)! Since, indeed, money heaped up in public funds tends, at least in some uses (law enforcement, hospitals, streets, public transport,...), to produce states dominating its private expenditure, this shows that excessive general tax evasion is collectively irrational. Particular values for S5 also permit to specify numbers k(2 < k < n) of cooperators necessary to guarantee Ης -coop-benefits according to criterion (la). In S'^ e.g. for c > j already k full's* = s'-contributors would make the Ηg, payoff vector js-k-c sk-c skc sk-c \ J

( \

η

,···,

η

,

η

+*,···,

η

\-s) /

dominate the anarchy-vector (s,... ,s). So here up to η — k full's* = 0 '-freeriders could be tolerated. (For c = 2 already 6 of 10 full contributors suffice for any s to make N(Ssatisfy (la).) As had already been indicated, the free-riders, however, enhance their own payoff at the expense of the contributors: Instead of the same payoff s • c for everybody, which uniform cooperation in S5 would yield, the k full contributors only receive the smaller payoff ~ each, whereas the n-k full free-riders (zero-contributors) each get ^ + s, and thus make an extra profit in comparison to their payoff in uniform cooperation. Let us now turn to criterion (lb) of UO-validity and have a look at some examples of norms fulfilling it by avoiding the inefficiencies (and sometimes

346

Rainer Werner Trapp

even disasters) possibly arising in anarchy due to the shakiness of the types of strategic situation referred to in (lb). (e) 'Chicken'- or 'Battle of the 5«c«,-structures, as and Sj, respectively, exemplify the dangers of anarchy if there are at least two efficient, strong equilibria (none of which can be everybody's 'darling'): b\

b2

a\

7,3

6,8

42

10,2

0,1

b\

b2

a\

3,4

8,12

*L

10,6

1,1

Any norm N(S(,)m\x and N(Sy)m[x respectively, decreeing some negotiated point on the Pareto-front north-east of the chosen SQ (i.e. any point in between equilibrium-points (10,2) and (6,8) in S uxjx (Vy € H" : Ay), however, is more difficult than similar proofs falling under criterion (I). For it requires χ to make certain empirical assumptions, which were not needed for proving the UO-validity of a norm. This renders criterion (II) more complex than was criterion (I). It reads: (II)

(IIa)

9

Vx, 5, TV [If χ is rational, 5 is a conflictive choice-situation, and TV is a general norm decreeing a certain behaviour to at least one player role r, in S [then χ will agree to TV iff TV fulfils either (IIa) or (IIb) or (lie)]]: (1) 5 is totally conflictive (i.e. a cardinal 2-player-zero sum game) with the noncoop (= maximin)-payoff vector u= (u\, uj); (2) TV decrees the coo/>-solution ν = (v\, vj) φ u;

Hamburger ( 1 9 7 9 ) will be o f particular value for those readers who are interested in socially relevant applications o f game theory presented in a very elementary, introductory way.

348

(IIb)

(lie)

Rainer Werner Trapp

(3) x estimates the distribution of probabilities (or relative frequencies in lx) of taking r, (/' = 1,2) in instances of S (with the same or different opponent(s)) at (px {r\), px {r2) = 1 - / > x ( n ) ) ; (4) χ believes that: ux , (Vy e Hj : Ny) = lj=]px(n) • v, + ^ ( S P H ) > u χ xJx i^y € Hj : Ay) = ΣΪ=ιρ (η) · «;· (1) 5 is a partially conflictive η-player-game (n > 2) whose noncoopsolution is a unique efficient, strong equilibrium u = {u\,..., u„), stable under reduction; (2) Ν decrees the föö^-solution ν = (v\,...,v„) φ u\ (3) χ estimates the distribution ... of taking r,-(/ = 1 , . . . , « ) . . . at (px(ri),...,px(rn))·, (4) χ believes that: ux/ (Vy e : Ny) = Σ"=ιpx{n) • vi + «X(SPH) > (1) S is a n-person //^-choice situation (i.e. nonstrategic) affecting η — 1(« > 2) passive (dis)utility recipients plus one agent y with options Ays = (a\,... ,am)(m > 2) and the payoff-matrix u i( j to be true still after lx. But they can hold jointly incompatible ante-factum-belieft and base their (non)agreement to Ν on them, (b) Situation type S10 (which I name " Aggresso") exemplifies criterion (lib):

•S10

d

-n,d

Λ

3,-15

4,-10

-> a

1,-1

0,0

Player 1 is a potential aggressor (street-robber, rapist, . . . ) who could, unobserved or by surprise successfully attack 2 (= a) or forego this (= -> 0 if every χ assessed p(r\) < Thus the fear to be the victim in more than | of all Si ο-games would produce peace among egoists respecting N(S\q), already without their attributing utility to SPH per se. (c) In the 3-player-game ι, which also exemplifies criterion (lib), cooperation would even pay earlier': C\

a

2

8

b\ 11

11

6

8

5

c2

110

b2 10

b\ 150

A\

25

5

2

-4

a2

7

25 4

b2 -10

11

9

12

5 6

3 8

The noncoop-omcomc of Sn is « = ( 8 , 1 1 , 1 1 ) , the only strong equilibrium stable under reduction. (The other strong equilibrium (12,6,8) is not stable under reduction since if r$ played C2, it would pay for r\ and r2 to jointly switch to {a\, b\) with (25,25)0; u is efficient·, thus single-s hot-nor tYi-cast shifts by cooperation are impossible. The N(S] ι)-decree of (a\ ,b2,c\), however, would pay for almost all values of px{r{) (i = 1,2,3). Even a pessimist χ who believes to never get the best role ?*3 with «3 = 150 must accept N ( S \ \ ) for any «X(SPH) > 0, if he prefers the ^/»-lottery ((1 - p x ( r 2 ) ) , 110;p x (r2), 10) to its noncoop-counterpart ((1 Ρ χ ( / ΐ ) ) & p x i r 2 ) . 11) - which he will for any px{r2) < Only if x fears to almost always be the sucker' r2 will he have to refuse N(S\ 1). For any px(r'3) > 0 the critical norm-acceptance-value would even rise above So among 'normal' egoists N(S\\) would acquire EO-validity. (d) Ad criterion (lie): Let N{S\2) be the classical utilitarian decree of a2 to the only agent in the nonstrategic situation Si 2 . The other two act-alternatives, α ι and -morals mentioned in footnote 3 are most valuable descriptive models explaining how cooperation may in fact arise without any contracting on norms, let alone on enforcing them. Their results are impressive and bear on the answer to be given to question (B) put at the end of section 6 above. Nonetheless, from question (A)'s viewpoint of collective rationality much more cooperation should arise than arose in all these models together. The amount of 'morality' shown to evolve there is firstly confined to only a few games, all falling under criterion (I) (mostly PD2-supergames). It produces much less c0o/>-benefits than could be achieved in a'perfectMmm - world' where all norms of types (I) and (II) are respected. Secondly, for each game S a long evolution is needed for eliminating pig-heads' who stick to unfriendly Sstrategies bred by vain hopes of individual rationality. This also holds for Sugden's meritorious study (1986) which convincingly argues (not only for PD 2 , but for a variety of games) that certain coop-'conventions\ partly even representing an inefficient equilibrium, will spontaneously evolve in supergames of that type. In a nutshell: Waiting for a natural evolution of cooperation from a normative viewpoint would be contenting oneself with too little ofMmin coming too late!10 Finally, keep in mind the following essential property of Mmin: All the general norms it comprises, be they unconditionally or only conditionally O10

Beyond that the socially most important norms, as for example those of property, were in historical fact presumably neither originally won by contract nor by spontaneous evolution ä la Sugden in a world of rational egoists, but were imposed by clan bosses who attained the critical mass of'core collaboration' mainly by arational factors such as bonds of blood and affection.

What Precisely Is Minimal Morality?

353

valid, are of the structure N(S); thus each of them is relative to a certain type S of a strategic interaction situation (represented by the set of players, the set of alternative strategies available to them and the payoff function of the game) or, as in case of criterion (lie), of a nonstrategic choice situation, each involving n(n > 2) individuals in an (at least partially) conflictive way. So none of these norms unconditionally decrees "Everyone ought (not) to do the kind of act λ!"; rather each of them is of the conditional structure: "Everyone ought, if involved in situation type S, taking the agent role r;, to choose the act alternative aj\" So even UO-valid norms are not unconditional, but include as an indispensable component some situation-specifying antecedent. This relativity to a particular mapping from acts or «-tuples of acts onto certain vectors of utility values for each role in S is inherently essential to the very idea of a morality justified exclusively in terms of rationality. Without the information contained therein a potential Mmin-contractor would lack any basis for deciding whether or not his single-instance-N(S)-payoff in each possible role or his expected long-run-payoff represented by a lottery-ticket of possible roles in S will exceed his corresponding payoffs in general anarchy, and therefore, whether or not it is profitable for him to agree to including N(S) into M m i n . With respect to the special case of compensatory norms justifiable according to criterion (lie), this relativity has one obvious implication: Within the theoretical framework of Mmit is impossible to come to a rational decision as for flatly accepting or rejecting any compensatory aggregation function whatsoever as a moral device for justly solving any future conflict of interests. As was illustrated by the above example of a classical utilitarian (lie)-norm-decree the application of a compensatory norm is only justifiable on the basis of a specified situation type plus the additional probability assumptions indicated in (lie).

12.

The Limits of a Rational fustification ofEthical Norms

So much for the potentialities of merely rational morality. Let us now have a short, harsh look at its limits·. Imagine a perfect Mmin-world. Even there most disagreeable states may arise, of which only some examples can be given here: (1) Assume that the bargaining over the 'original' allocation of goods led to a type-(Ib)-contract on a distribution of property-rights pursuant to which lucky j/ obtained property rights over (as only later is detected) some of the worlds most important raw materials, the product basis for the only remedy for cancer, AIDS, ... and similar good things. By rack-price sales,

354

Rainer Werner Trapp

which (leading to north-east shifts and thus to a 'happier and happier' world) would be welcome by M m i n , our monopolist would then acquire the means to buy more and more of the others' land, their labour and services, and thus to make himself a de-facto-dictator. This again would increasingly shift the SQs of further transactions between y and his cocontractors in favour ofy. Would a rational egoist really like to b e / s defacto-subject in this world or would he, from a certain point on, rather hoot away Mm;n? (2) Already Luce/Raiffa in (1957), pp. 128—30, have called attention to grotesquely unfair Nash-income distributions in cases where the players' risksensitivity co-determines their utility functions. An important theorem by Kihlstrom et al. (1981) deepens this criticism by implying that type-(Ib)bargains ä la Nash, Kalai/Smorodinsky and Perles/Maschler ceteris paribus favour a player the more, the more risk-averse his opponent is. KlemischAhlert (1990) generalizes this result to all risk-sensitive bargaining solutions. (3) Already the 'original' allocation of goods may depend on an SQ determined by a 'threat-coalition' of bullies ("If you don't concede to u s . . . , we'll use o u r . . . " ) . Rather than count as a moral solution to the respective puredistribution-problem this appears to be a slap in the face for justice. (4) Suppose the 'peace'-norm N(S\Q) for'AggressO'above had been vetoed by an optimistic rough hoping to fare better with anything goes!' in 5JOgames. Then in Sio anarchy would reign. Yet often a range for type-(Ib)bargaining compromises, profitable for both and thus endorsed by M m m , would remain. Relative to the pressure of the existing SQ e.g. both the victims of kidnappers, of cutthroat-rapists etc. and the respective rascals would profit from an 'in-between-solution' on the Pareto-front ("All right, I'll pay ( d o , . . . ) this but not that, and then you leave me alone!"). To concede this may be prudent·, Mmin complied with here is better than anarchy; but would it lead to a just state? (In many a PD 2 -game it might rather serve justice if both rogues defected and ended up in jail, in SG above if the Red Cross were the heir!) This brings me to a more general criticism: (5) Mmin is purely utility-based. It does not consider to what extent the parties involved deserve coop-utility gains in view of (a) the nature of the SQ and its development, (b) prior sins or merits in multiple respects and further factors 11 . An obvious further deficiency oiM m - m is that it cannot contain any obligations towards 11

I have dealt with this in a different context in Trapp (1988) pp. 3 2 3 - 3 9 .

What Precisely Is Minimal Morality?

355

animals, as already Hobbes observed ("No Covenant with Beasts") or (7) future generations11. (8) Likewise nobody can posthumously benefit from coö/>-arrangements with contemporaries. So one day before his death Emperor Nero would be free (6)

to permit himself the final pleasure o f seeing Rome burn. (Today he might enjoy the beauty o f a nuclear blaze.) (9) W h y should rational egoists contribute to the care o f persons who were born handicapped to such an extent that they are unable to be useful or harmful to them? In an Λ / ^ - w o r l d no norm would protect such persons from being left to starvation. Note that, contrary to what is often said, some compensatory M m ; n -norms could be justified by the reasoning o f type (lie) above. Each egoist might believe that he himself could get sick or old and thus be the beneficiary o f a corresponding social security-a>0/>-norm. Yet he could not even understand being born a cripple with p(ri) > 0 if he knows he is not one. In a word:

A perfect Mmin-world could be perfectly vile.

Certainly, all this is meant to hit only those who claim moral exclusiveness for .A/mjn arguing, as does Gauthier in his version o f Af m i n quite rightly ( 1 9 8 6 , p. 16), that any restrictions beyond Mmm are not rationally justifiable: " O u r theory denies any place to rational constraint, and so to morality, outside the context of mutual benefit. A contractarian account o f morals has no place for duties that are strictly redistributive". By modus tollens I conclude from this that rationality does not by itself give enough groundfor a sufficiently rich moral system M. Either Μ cannot be based on contractarianism or each contractor χ £ Η must accept some impartiality demand in the form o f e.g. precisely that protonorm NQ dispensing with which is o f the utmost importance to Mmm: "Every χ ζ Η (a) assume a fictitious property-reallocation lottery in which A: could [with px = \ (Harsanyi) or with unknown px (Rawls)] draw one of the η bundles o f properties to be found in Η (and so could end up handicapped at birth), and (b) decide over his acceptance or rejection of further Λί-norms before he knows which lot he will draw." Still other interpretations o f impartiality are possible. Like NQ they all amount to very demanding moral norms that cannot be justified as O-valid by universal benefits. So χ would have to arationally accept one o f them as an irreducible deontic bottom pillar. (Remember the result o f section 2 that NQ 12

Cf. Leviathan, part I, ch. 14. In his substantial review of Gauthier (1986), Hegselmann points out that Gauthier's denial of (7) on pp. 2 9 8 - 3 0 0 rests on an insufficient argument. Cf. Hegselmann (1989), p. 157.

356

Rainer Werner Trapp

cannot simply be declared as evidently 0-true\) All those rational egoists who would not profit from any further Af-norms of unilateral care orprotection will, of course, reject NQ . Thus NQ itself and all those further norms which are generally agreed to only on the basis of the general acceptance of NQ will not be O-valid.

13.

Justice also Requires Compensatory, Nonwelfarist Principles

Most of such criticism of Mm;n could be met by accepting classical utilitarianism or maximin as a formal principle for winning concrete M-norms for given types S of conflicts of interests. Yet, though compensatory, both are also welfarist. All brands of welfarism, however, in many S lead to counterintuitive results, as a host of well-known criticism has revealed. This is mainly due to their also neglecting nonutility information concerning desert, motives etc. Beyond that, widespread moral intuitions suggest that just utility aggregation be sensitive to both utility changes and prior levels and also regard the dispersion of deserved levels. Nonetheless the deontic axioms of suchlike utilitarianism incorporating justice', which I presented in Trapp (1988, 1990a), will certainly not meet everyone's, let alone egoists', acceptance. As all pretensions to O-truth for whatever Ν-assertions had to be discarded, and as many compensatory S-norms put out by such an aggregation function are not usefiil to everyone even in the long run, such a moral system cannot be considered O-valid in any sense. So we have to face the sad result: Notwithstanding objectivist affirmations to the contrary, any approach in theoretical ethics has to choose between justifiable claims to O-validity for an insufficient morality of cooperation and only subjective validity for a richer compensatory system considering also nonwelfarist information in moral decision making.

14.

The Motivation Problem: Why Always Follow Mmin-Norms?

All norms N(S) of Mmin are general. They decree to each χ what χ ought to do in S. It cannot be argued - as is sometimes done - that pointing to N(S) is senseless even in criterion-(I)-czses without also delivering reasons to egoists why they shouldfollow N(S) if they know no more than that they are in S. This misses the point, for N(S) is grounded on insights of collective, not of individual rationality. Its very justification that everybody's compliance would make everybody better off than in anarchy remains true even if foul-dealers may force some N(S)-followers below their baseline in anarchy. So the value of an an-

What Precisely Is Minimal Morality?

357

swer to question (A) posed in section 6 above, does not depend on whether or not question (B) can also be answered. Nonetheless the efficient coop-stztts denoted by N(S) can only become real if (B) is answered as well. So why follow N(S) ? In some conditions of some S the answer is easy, for here compliance with N(S) also pays individually. In 'Chicken' e.g. it pays if χ believes the other player will not comply; in «-person 'lumpy'-public-goodgames with a k-of-n-threshold it pays i f * assumes exactly k — 1 others to cooperate, etc. Yet the answer to (B) is difficult if such conditions do not obtain. (In PD"(n > 2) they never obtain!) In a state of nature, N(S) will then have no appeal to egoists. Hegselmann et al. in (1986) point out for PD2-games that both players' question "Shall I follow Af(PD 2 ), i.e. cooperate, or not?" leads to a metagame of PD 2 -structure as well. As noncompliance with 7V(PD2) is dominant the inefficient state of nature will not be bettered by mere agreement to 7V(PD2). What answer to (B) exhibits a way out of these traps? There seem to be only two, a sofi way and the old hard one. The first recommends to develop effective coop-preferences and is exemplified by Gauthier s device of 'constrained maximization. In a similar approach Hegselmann et al. (1986) show (only for PD 2 ) that developing a disposition for conditional PD2-cooperation (guaranteed by 'assurance game'-metapreferences) would avoid the inefficiency trap without running the risk of being exploited by defectors. An alternative suggestion similar to the one above, true not only for 2 PD -games, could lead on from our initial broad conception of rationality and recommend that each χ develop a disposition to have a cooperationguaranteeing critical mass of initial trust that each co-player i considers in his effective utility u* also χ 's interest, and subsequently make the values af in his own 'altruism vector' ( a f , . . . , a*,..., a*) in instance Sj of S depend on his ante-factum belief (which would become more and more justified by growing experience) as for the degree a'x to which each i will in Sj consider him in (a j , . . . , a)·,..., a'n). Of course, this mutual tit-for-tat-Yike dependence of each two players' altruism cannot be based on mutual knowledge in Sj about the factual values in these vectors, since gaining the latter would conspicuously imply an epistemic circle. Just as obviously as this suggestion demands a certain confidence in strangers, the above mentioned suggestion by Hegselmann et al. requires each player to know whether he faces a conditional (or even unconditional) cooperator or a defector. Without this confidence or information one has to model the situation along the probabilistic lines worked out in Gauthier (1986) hereby avoiding some technical deficiencies pointed to in Hegselmann (1989).

358

Rainer Werner Trapp

All this is no safe way to cooperation. Thus, I conclude, rational egoists should not shy at the costs of the hard way and hasten to add a 'penal-law'contract P(Mm\n) to its 'civil law'-predecessor Mmm. Here for each N(S) of Af m , n a penal twin norm NP(S) "For all χ: If χ is in S and violates N(S) then the enforcement agency ΕA ought to penalize χ b y . . . " should be subsequently agreed to. A public-good norm decreeing to what degree contractors have to contribute to the costs for ΕA could be established by criterion (la) of Mmin. As one single ΕA could enforce all norms of Mmm, ΕA would, due to decreasing 'unit cost' for enforcing additional norms, pay the more the more comprehensive Mmin was. Punishment p would ideally have to be strong enough to deter potential defectors by at least nullifying expected benefits from defection. Take PD 2 situations as an example (with the usual orders over outcomes Ti > Rt > Pi > Si for i — 1,2). Let qi be is assessment of the subjective probability of being caught in defection and punished with some penalty p. Some calculation reveals that general agreement on some such penalty p would suffice here to make cooperation dominant for every χ e H, if x's penalty-utility loss of Uj(p) in role η of PD 2 were such that

Ui(p) > max

(Tj-Rj

VΦ

,

Pi-Si H

(Bayesians could already be deterred by less.) In short: Particularly those individuals whose egoism prevents them from accepting more restrictions than those of Mm\n should not put too much trust in their developmental potential towards altruism but rather follow Kant's dictum that 'man is an animal in need of a lord'. The latter does not have to be as terrifying as Hobbes required him to be. How strong should ΕA (as a body possibly consisting of rational egoists, too) be made? If i(i = 1 , . . . , n) has mi(S) opportunities to defect in situations of type S, EA should, to justify its very existence, deter more than one contractor from one defection in one situation-type but, for reasons of pecuniary and, to an even greater extent, other obvious costs, deter less than all contractors from all defections in all situation-types. (In several millenia mankind has not yet found the right point in between these two extremes.) Thus the hard way cannot guarantee full compliance, either. There simply is no safe way to a perfect world., let alone one that really deserves to be called perfect tout court. Hoping to find arguments that completely close the gap between the results of individual and collective rationality or even between 'want' and a

W h a t Precisely Is Minimal Morality?

359

more demanding ought' will, I fear, forever remain vain philosophical illusion. Lack of sympathy and affection cannot be made up for by bare reason.

References ( 1 9 8 7 ) . Ken BinmoreandPartha Dasgupta (eds.): TheEconomics of Bargaining, Oxford 1987. E P I C U R U S (KD). Epicurus: Kyriai Doxai; references follow the standard numbering, as given, for instance, in Opere, edited by G. Arighetti, Turin I960. G A U T H I E R (1974). David Gauthier: "The Impossibility of Rational Egoism", The Journal of Philosophy 71 (1974). G A U T H I E R (1978). David Gauthier: "Social Choice and Distributive Justice", Philosophia 7 (1978). G A U T H I E R ( 1 9 8 5 ) . David Gauthier: "Bargaining and Justice", in Paul et al. ( 1 9 8 5 ) . G A U T H I E R ( 1 9 8 6 ) . David Gauthier: Morals by Agreement, Oxford 1 9 8 6 . H A M B U R G E R (1979). Henry Hamburger: Games as Models of Social Phenomena,New York 1979. H A R S A N Y I (1977). John C. Harsanyi: Rational Behavior and Bargaining Equilibrium in Games and Social Situations, Cambridge 1977. H E G S E L M A N N (1989). Rainer Hegselmann: "Rational Egoism, Mutual Advantage and Morality: A Review-Discussion of D. Gauthier's 'Morals by Agreement'", Erkenntnis 31 (1989). H E G S E L M A N N ET AL. (1986). Rainer Hegselmann, Werner Raub and Thomas Voss: "Zur Entstehung der Moral aus natürlichen Neigungen: Eine spieltheoretische Spekulation", Analyse und Kritik 8 (1986). H O B B E S (1651). Thomas Hobbes: Leviathan, Harmondsworth 1968; first publ. in 1651. H O L Z (1990). Harald Holz (ed.): Die Goldene Regel der Kritik, Bern 1990. K A L A I / S M O R O D I N S K Y (1975). Ehud Kalai and Meir Smorodinsky: "Other Solutions to Nash's Bargaining Problem", Econometrica 43 (1975). K A N E K O / N A K A M U R A (1979). Mamoru Kaneko and Kenjiro Nakamura: "The Nash Social Welfare Function", Econometrica 47 (1979). K I H L S T R O M ET AL. (1981). Richard Kihlstrom, Alvin Roth and D. Schmeidler: "Risk Aversion and Solutions to Nash's Bargaining Problem", in Moeschlin et al. (1981). K L E M I S C H - A H L E R T (1990). Marlies Klemisch-Ahlert: "Distributive Justice of Bargaining and Risk Sensitivity"; disc, paper no. 9001, Dept. of Economics, Univ. of Osnabrück 1990. K L E M I S C H - A H L E R T (1991A). Marlies Klemisch-Ahlert: "Independence of the Status Quo Journal of Economics 53 (1991). K L E M I S C H - A H L E R T (1991B). Marlies Klemisch-Ahlert: Verhandlungslösungen als soziale Entscheidungsmechanismen, Frankfurt/Main 1991. BINMORE/DASGUPTA

360

Rainer Werner Trapp

(1957). R. Duncan Luce and Howard Raiffa: Games and Decisions, New York 1957. M O E S C H L I N / P A L L A S C H K E ( 1 9 8 1 ) . Otto Moeschlin and Diethard Pallaschke (eds.): Game Theory and Mathematical Economics, Amsterdam 1981. N A S H (1950). John F. Nash: "The Bargaining Problem", Econometrica 18 (1950). N A S H (1953). John F. Nash: "The Two-Person Cooperative Games", Econometrica 21 (1953). N I D A - R Ü M E L I N ( 1 9 9 3 ) . Julian Nida-Rümelin (ed.): Praktische Rationalität, Braunschweig 1993. PAUL ET AL. (1985). Ellen Frankel Paul jeffrey Paul and Fred D. Miller, Jr. (eds.): Ethics and Economics, Oxford 1985. P E R L E S / M A S C H L E R ( 1 9 8 1 ) . M. A. Perles and Michael Maschler: "The Super-Additive Solution for the Nash Bargaining Game", International Journal of Game TheLUCE/RAIFFA

ory 10 ( 1 9 8 1 ) .

(1986). Philip Pettit: "Free Riding and Foul DealingJournal of Phibsophy 83 (1986). R O T H (1979). Alvin Roth: Axiomatic Models of Bargaining, Berlin 1979. R O T H (1985). Alvin Roth (ed.): Game-Theoretic Models of Bargaining, Cambridge 1985. S C H Ä F E R / T R A P P (1989). Andreas Schäfer and Rainer Werner Trapp: "Distributional Equality in Non-Classical Utilitarianism: A Proof of Lerner's Theorem for Utilitarianism Incorporating Justice", Theory and Decision 26 (1989). S U G D E N (1986). Robert Sugden: The Economics of Rights, Co-Operation and Welfare, Oxford 1986. T R A P P (1988). Rainer Werner Trapp: "Nicht-klassischer" Utilitarismus: Eine Theorie der Gerechtigkeit, Frankfurt/Main 1988. T R A P P (1990A). Rainer Werner Trapp: "Utilitarianism Incorporating Justice: A Decentralised Model of Ethical Decision Making", Erkenntnis 32 (1990). T R A P P (1990B). Rainer Werner Trapp: "Ein grundsätzliches Argument gegen jeglichen Wertungskognitivismus", in Holz (1990). T R A P P (1994). Rainer Werner Trapp: "Die Verwendbarkeit von Verhandlungsmodellen im Rahmen der Ethik", in Nida-Rümelin (1993). Z E U T H E N (1930). Frederik Zeuthen: Problems of Monopoly and Economic Welfare, London 1930. PETTIT

ANTHONY SIMON LADEN

A Hobbesian Choice: Reply to Trapp

1. If we are to believe Rainer Trapp's conclusion, then those of us in the business of reasoning about morality are faced with a Hobson's choice: Since minimal morality is all that can be "rationally" justified, we are left with no choice but to search for ways to make it more palatable. Trapp's paper tries to do just that: search out and articulate the potentialities and limits of minimal morality. In this reply I hope to suggest a way out of this conclusion, and thus the inevitability of the project Trapp has undertaken, a project he himself claims holds no "philosophical delight". Trapp himself is no minimal moralist, and yet he makes a number of substantive moral assumptions which play right into their hands. Showing this will involve more than purely rational calculation, but also more than the mere articulation of subjective preferences. Trapp's paper can be usefully seen as having two main parts. The first, which includes sections 1-5 and some scattered remarks towards the end, argues for the claim that we are faced with a Hobson's choice, while the second, which might be thought of as the 'meat' of the paper, argues for a particular view of what that choice looks like. Since I think the most interesting philosophical questions surrounding minimal morality take place prior to numbercrunching, I will take a vegetarian's approach and focus my remarks on what I am calling the first part of the paper. 2. Briefly stated, my worry is that Trapp, in setting up such a stark contrast between mere moral preferences and a rather narrow conception of what could count as an objective justification for ethical norms, pre-determines his Hobson's choice conclusions by dressing up what are rather partisan assumptions as something neutral and unavoidable. I want to look closely at the reasoning that leads him to the conclusion in section 13 that " [notwithstanding objectivist affirmations to the contrary, any approach in theoretical ethics has to choose between justifiable claims to O-validity for an insufficient 'morality of cooperation' and only subjective validity for a richer compensatory system considering also nonwelfarist information in moral decision making."

362

Anthony Simon Laden

Trapp's starting point is a familiar one for both the moral skepticism of minimal morality and the less skeptical forms of liberalism; like many, he begins with the fact of pluralism, or as he puts it, the existence of rival ethical norms. But exactly what problem the fact of pluralism poses for philosophy is not so clear as Trapp would have us believe. For him, it is a sign of the insufficiency of the grounds these theories claim for their objectivity. It is, thus, ultimately a reason to reject the bulk of the history of moral and religious thought in a couple of paragraphs. The problem, he claims, is that since all these theories rest their claims to objective validity on the supposed objective truth of their central claims about "the deontic status of acts" (sect. 2), their claims to validity are undermined by the very fact of pluralism. Thus, for instance, the Kantian and the Utilitarian are bound to see the world sufficiently differently that they will deny the 'truth' of the rival view. Having undermined objective truth as a grounds of moral objectivity, Trapp starts off section 3 by asking whether we are left with a view which likens moral norms to norms of taste. We are saved from such an unappetizing conclusion by the fact that some norms are "undeniably usefiil to everybody". These quickly get labelled "objectively useful". Just as quickly, "useful" is defined in terms of increased utility. Then, in section 4, rationality is defined in terms of utility maximization. With this move, the only thing standing in the way of Trapp's Hobson's choice conclusion are the reasonably uncontroversial mechanics of game theory. 3. If there is a way off this path, then, we must have already passed it. In order to begin to suggest where some promising exits might lie, I want to look at the way Trapp dresses his theory in contractarian garb, because it demonstrates rather nicely the way in which partisan assumptions are smuggled in under the neutral banners of 'rationality' and 'the lack of ethical assumptions'. In section 5, Trapp asks that we imagine a world where the only normative system any of us followed was the unrestricted pursuit of our own interest (this might, of course, include pursuing greater utility for others, insofar as their utility contributed to our own as a result of sympathy). He then asks what ideally rational people in such a situation would agree to. He maintains that they could not be required to ground their assent to any proposed system of norms on "any already moral point of view", as for instance, he takes Rawls's contractors to do. Notice, however, the moral position hidden here. Like all bargaining game versions of contract theory, this one is highly dependent on the choice of status quo point, and so that choice can not be a neutral matter. Notice also that two status quo points are at issue here, depending on whether we focus on the game

A Hobbesian Choice: Reply to Trapp

363

as a bargain about expected utilities (in which case initial utility levels determine the status quo) or about action-governing norms (in which case the parties prior allegiance to normative systems describe the relevant starting point). As far as utility levels go, no starting point is any more morally neutral than any other. Both equal starting points, and natural' starting points (ä la Gauthier) represent moral choices, and thus require some justification. Similarly, taking universal unrestricted pursuit of self-interest as the status quo point in a bargaining game is no less of a moral decision than taking the idea of parties as representing free and equal members of a system of social cooperation over a complete life (ä la Rawls) as the organizing idea. That a choice is less moral does not make it less of a moral choice. The important question to ask is what justifies, that is morally justifies, the choices made. In the case of Rawls s theory, the contract is meant to work out appropriate principles of justice to guide a society conceived by its members to be a system of cooperation among free and equal persons over a complete life, and this fact justifies his description of the parties and their motivations. If we are to apply similar reasoning to Trapp's account, then minimal morality will be justified in determining appropriate norms to guide consistently selfish people in their selfish pursuits. And while that is not a wholly useless project (it clearly has a role to play in neoclassical microeconomics, for instance), it should strike us as hardly the only rational project for moral philosophers to be engaged in. The difference, that is, is not between rationality sullied with moral assumptions (Rawls) and rationality in its pure form (Trapp). Rather, it is between two different views of the sort of society for which we are generating norms. In Trapp's world, people are merely rational; in Rawls's they are reasonable as well. 4. A further aspect of Trapp's method of setting up the problem is the limited role that argument amongst people plays in his view of agreement about moral norms. Let's return to his rejection of the objective validity of any moral system other than minimal morality. Trapp starts by summarizing findings about the reference semantics for deontic logic, which he claims provide the best place to look for the "truth conditions" for obligations (sect. 2). Much of the work of identifying these truth conditions involves picking out the set of possible worlds which are 'morally perfect'. Of course, adherents to incompatible moral systems are likely to disagree on this set, and thus there is no common mutually acceptable set of truth conditions to decide between rival norms. Unlike in the analogous empirical case, Trapp claims, "it is principally impossible in the Ν-case to neutrally decide who is right and who committed a value perceptional' error" (sect. 2). The "obvious" lesson he draws from this difference is that the truth-values of normative statements are "merely subject-

364

Anthony Simon Laden

ive" as long as "the perfect worlds o f different 'moral observers' happen to differ." But picture the scene being described here. Two people disagree about the morality o f a certain action, say keeping a harmful truth from a mutual friend. In an effort to determine who is right and thus what they ought to do, each, among other things, describes a set o f morally perfect possible worlds. These sets turn out to be different. At this, each throws up her hands at the irreducible subjectivity of moral judgements and goes her own way. Now that strikes me as a funny picture for a philosopher to rely on. After all, if anybody is in the habit o f arguing for the truth o f beliefs about moral norms, it is moral philosophers. We rarely just state our opinion and walk away if others don't agree. If we did, both conferences and books on the subject would be a lot shorter. Now, it may be that our arguments don't resolve everything (and sometimes they may even resolve nothing), but it seems rather important that we do attempt to resolve our differences by argument, by seeking common underlying reasons and so forth. Once again, we have quietly been put under the sway o f a disputable moral assumption. Having rejected as inappropriate to morality the sort o f objectivity through truth that science pursues, Trapp immediately concludes that the only standard left for moral objectivity is universal love-at-first-sight agreement. But such a ground for objectivity is anything but morally neutral; it is, in fact, a conception o f objectivity that only a minimal moralist could endorse. When it is paired with the notion that the only good grounds for affirmation o f a given norm is its role in increasing your utility, it is hard to imagine anyone else even accepting it. 5. There are other ways to reject objectivity through truth without rejecting moral objectivity. I will briefly sketch out my preferred way o f approaching the problem just to show that once again, Trapp's choice is neither neutral nor inevitable. That done, I will conclude with some remarks about what sort o f choice minimal morality turns out to be. The route from human diversity to utility maximization that Trapp follows is an old one. It relies on a denial of the complexity o f human existence which is seen to lie at the root of the diversity. That is, Trapp, like many utilitarians and others before him, tries to get around the problems o f human difference by relegating all our differences to a superficial surface layer which merely covers up our simple, uniform core. And since underneath those differences, this line o f argument goes, we are essentially rational choosers hellbent on maximizing our utility bundles, moral theory need not be plagued by human diversity, because it can ignore it. T h e problem is, as Trapp points out

A Hobbesian Choice: Reply to Trapp

365

in his paper, that it is impossible to get a sufficiently strong, and thus palatable, moral theory out of the assumptions about people from which minimal morality starts. I want to suggest a different strategy. Instead of treating human diversity as just a problem to be gotten around, perhaps we should also look to it as containing the roots of a solution as well. If we recognize the fact that we don't go through life merely counting up utility bundles and probability distributions, then we will have a lot more material out of which to construct a moral theory. We have projects, plans and commitments; we form and nurture relationships; we deliberate and reason with one another; we have histories which shape us and histories that we shape. It is because of all this that in spite of the fact that we are different from one another, and even differ in the ways in which we are different, we can also share things, from bread to houses to actions to conceptual schemes to, perhaps, systems of norms, in a way that makes them all of ours collectively, and not just each of ours individually. Why not, then, take our capacity, even our propensity, to share, to form plural subjects ("we"s), to seek out consensus and understanding as basic moral facts in the same way that the minimal moralist takes what he sees to be our natural selfishness and quest for ever-greater levels of utility as foundational? If anything, such 'moral' capacities strike me as much more commonplace than those urged on us by the minimal moralist. After all, why do we help our friends or even have friends in the first place? Why do we vote in elections where there is roughly no chance of our vote determining the outcome? Why do we offer directions to strangers who ask for them when we could easily get away with ignoring them or offering simpler lies? The typical response to these questions, that such behavior can be reduced to longer-range utility maximizing calculation via our subscription to the norms of minimal morality merely begs the question. The only reason to give the more tortuous reductionist explanations of minimal morality over the simpler explanations a richer moral theory might offer is because you have already accepted the premise that utility maximization counts as the only rational motivation for action. And, as I hope I have suggested, that is not a position that can be taken neutrally, without moral argument. While minimal morality traces its lineage proudly back to Hobbes, I think one can find traces of the alternative approach that I would advocate in Rousseau, Kant and Hegel, and very clearly in the writings of such contemporary figures as John Rawls, Jürgen Habermas, Amartya Sen and Charles Taylor to name a few. But if that is in fact the case, and we do not follow Trapp in so quickly rejecting the history of moral and political philosophy, then there is a reasonable hope that the project of finding 'merely rational

366

Anthony Simon Laden

justifications' for moral norms is not a Hobsons choice at all, but only a Hobbesian one, and thus one which can be rejected, or at least argued against, on rational as well as moral grounds.

CHRISTOPH FEHIGE, RICHARD M .

HARE, WOLFGANG LENZEN, JEFF

M C M A H A N , P E T E R S I N G E R , T H O M A S S P I T Z L E Y , AND U L L A W E S S E L S

Symposium on Possible Preferences

Christoph Fehige and Ulla Wessels: Introduction to Possible Preferences, p. 367 Peter Singer: Possible Preferences, p. 383 Richard M. Hare: Preferences of Possible People, p. 399 Wolfgang Lenzen: Who Counts?, p. 406 Ulla Wessels: Procreation, p. 429 Jeff McMahan: Preferences, Death, and the Ethics of Killing, p. 471 Thomas Spitzley: McMahan on Psychological Continuity and the Value of Future Goods, p. 503 Christoph Fehige: A Pareto Principle for Possible People, p. 508

C H R I S T O P H F E H I G E AND U L L A W E S S E L S

Introduction to Possible Preferences*

Sometimes our actions make a difference not just to the frustration or satisfaction of preferences that exist (have existed, or will exist), but to the very question which preferences will exist. The corresponding choices — choices that affect the number or the identity of preferences that will ever exist - we can call different preferences choices, as opposed to same preferences choices.1 This symposium is about the morality of different preferences choices. Different preferences choices are important. Clearly, which and how many preferences we have has a lot to do with how we lead our lives, with our character and our welfare. (Compare the person who loves money, and strives to get and keep it, with someone who does not.) Famous sages have pointed this out for millenia, and have given advice on how to shape our wishes.

* We are grateful to Christopher Abbey, Wolfgang Lenzen, Jeff McMahan, and Peter Singer for their helpful comments; and to the Deutsche Forschungsgemeinschaft (DFG) for supporting the research project "Was zählt?"; work on this introduction was part of the project. 1 The terminology is inspired by Derek Parfit's distinction of different versus same people choices, see p. 377 below.

368

Christoph Fehige and Ulla Wessels

But, more dramatically, should lives be led at all·. And for how long? Choices in matters of life and death are among the most important cases of different preferences choices (no life, no preference) and have thus come to dominate the debate. They include the issues of procreation, abortion, population policy, and killing. They are also the central topics of the present symposium and hence of this introduction.

Extra Preferences Suppose that we could bring into existence an extra preference — should we? (Should we, that is, all else being equal? The ceteris paribus clauses will be tacitly assumed most of the time.) If the new preference would be frustrated, it is fairly clear that the answer is no, since preference frustration is bad. But what if it would be fulfilled? What is the value of a satisfied extra preference? There are at least three elementary options: (1) A satisfied extra preference is good. (2) A satisfied extra preference is good if, and only if, its bearer exists anyway. (3) A satisfied extra preference is neither good nor bad; its value is neutral. For brevity's sake, the list has been kept incomplete. Modified versions of claim (2) will be dealt with later. Among the claims that we have ignored completely are that a satisfied extra preference is (pro tanto) bad, and that the (pro tanto) value of a satisfied extra preference could differ from case to case.2 Equipped with options (l)-(3), we can move on to life and death.

Lives: The Beginning May we, or ought we to, create a new person? And may we, or ought we to, have abortions? If at all, then under what conditions? Let us pretend for the moment that preferences are all that counts. (More on other goods below, in an extra section.) Somebody's life is then, for moral purposes, a bundle of preferences. And apart from any preferences the fetus might have before an abortion, the ethics of procreation and abortion boils down to the question of whether certain sets of preferences ought to exist. To see it boil down in some more detail, consider the following list of positions:

2

See Fehige, sect. 1, for some comments on these two options.

Introduction to Possible Preferences

369

The Rabbits*: The Midwives*:

An individual with satisfied preferences ought to exist. Given an individual, she ought to have satisfied preferences. The Mind Readers*: Given an individual with preferences, they ought to be satisfied. Understanding the list, we suggest, is one way to understand the key issues; here are some explanations. Ceteris paribus. Remember that the whole list is peppered with tacit ceteris paribus clauses. The obligations are claimed to hold just as long as they are not outweighed by other considerations. If, for instance, satisfaction could be achieved only at the cost of frustrating either the individual in question or somebody else (say her mother), then the respective obligations to provide the satisfaction might falter; this is something that Rabbits*, Midwives*, and Mind Readers* would all agree to. Down the list. Every claim entails the claims from further down the list, whereas no claim entails one from further up the list. With each step down the list, the obligation to procure satisfaction is tied to an extra condition. Losing the asterisk. Consider members of every group who deny the stronger claims (the claims listed above their own claims, that is). If we call members for whom this holds by their old names minus the asterisk, we get the following list: The Rabbits: The Midwives:

An individual with satisfied preferences ought to exist. Given an individual, she ought to have satisfied preferences; but the Rabbits* are wrong. The Mind Readers: Given an individual with preferences, they ought to be satisfied; but the Midwives* and the Rabbits* are wrong. Procreation and abortion. Now, to see what all this means for procreation and abortion, suppose that Mary is a possible individual who, if she were conceived and born, would lead a life with many more and stronger satisfied preferences than frustrated ones. (Talk of the "possible individual Mary" raises well-known metaphysical problems; but no moral harm ensues if we help ourselves to this expression.) The respective moral judgements will then be these:

370

Christoph Fehige and Ulla Wessels Given that Mary is conceived (and has no preference whose satisfaction would require her survival), refraining from aborting her

Given that Mary exists, satisfying her preferences

obligatory

obligatory

obligatory

The Midwives say:

not obligatory

obligatory

obligatory

The Mind Readers say:

not obligatory

not obligatory

obligatory

Conceiving Mary The Rabbits say:

The call for life weakens. The table shows that, as we move down our list, the call for life weakens: Rabbits prescribe conception and forbid abortion; Midwives don't prescribe conception but still forbid abortion; Mind Readers neither prescribe conception nor forbid (as long as the fetus lacks the relevant preferences) abortion. Mnemonic labels. With some good will, our choice of names for the three positions can now be seen to serve mnemonic purposes. The rabbit is an animal notoriously given to procreation; midwives care for the step from pregnancy to birth; and mind readers look for a mind: no preference, no obligation. Warning. The table shows what follows from the Rabbits', the Midwives', and the Mind Readers' positions under certain standard assumptions about identity and preferences. One source of complications is that some theories of identity or of preferences would, if coupled with the Midwives' or the Mind Readers' position, force us to revise the table (see pp. 373 f. below). The symmetry between contraception and abortion. In one respect, Rabbits and Mind Readers, though at opposite ends of the list, are closer to each other than to the Midwives. Rabbits and Mind Readers agree that contraception and abortion are, morally speaking, in the same boat - the question is just: which boat? Midwives, however, deny this symmetry. The link. We can now link the three positions to the general claims (l)-(3), from p. 368, about the value of satisfied extra preferences. The connections are pretty straightforward. The Rabbits and claim (1): When supporters of (1) look at a life, they give positive weight to every satisfied preference; thus, if there are more and

Introduction to Possible Preferences

371

stronger satisfied preferences than frustrated ones, the life has a positive value and ought to be created. And that is what the Rabbits say. The Midwives and claim (2): When supporters of (2) look at a life, they perform almost the same calculation as the Rabbits. The only difference is that the satisfied extra preferences are not deemed to have positive value as long as the individual does not exist. Therefore, they cannot generate a reason to create her. But since they do get positive weight as soon as she is around, they generate obligations to keep her alive (not to abort her, for instance) and to equip her with preferences that are satisfied. And that is what the Midwives say. The Mind Readers and claim (3): Since according to (3) satisfied extra preferences are of no positive value, there is nothing in a life, or in any part of it, that would have made it bad if the life, or the part of it, had not been lived. Therefore, non-conception is okay, and so is - as long as the fetus has no preferences whose satisfaction would require her survival - abortion. And that is what the Mind Readers say. We have seen, in outline, how the value of satisfied extra preferences determines the moral status of procreation and abortion.

Lives: The End Let us now turn to the end of a life: how bad (or good) is death? The theoretical situation looks quite similar to that for birth, which was presented in the previous section. (In fact, abortion, touched on above, is an issue where the two topics, beginning and end, intersect.) The individual whose death-at-point-of-time-f is at issue will usually have preferences before t, including perhaps some that her death-at-f would frustrate (or fulfil). 3 But, death or no death, these preferences are there anyway, and so their moral impact on the badness (or goodness) of death has nothing to do with different preferences choices; they form, if we compare death-at-f to a survival option, what we could call a same preferences segment of that comparison. Let us pretend, for todays purposes, that the role of the same preferences segment for the badness (or goodness) of death is uncontroversial. We can then turn our attention to the different preferences segment (more precisely, the extra preferences segment): how good or bad is it that death-at-f avoids the existence of the post-f-preferences that the preferrer would have had in the 3

Cf. McMahan, sects. 2 and 4-6.

372

Christoph Fehige and Ulla Wessels

survival scenario? The answer will be informed by our choice between claims (1)—(3), presented at the beginning of this introduction. Let us see how. For an individual to die she has to exist; so where death is at issue the difference between claims (1) and (2) will vanish. (Unless (2) is modified; more on that below.) For defenders of (1) or (2), the preferences that, in the survival option, would come into existence and would be satisfied have positive value and argue against death, just the way we saw them argue against abortion. For defenders of claim (3), however, they don't; for them, the badness (or goodness) of death is determined only by the same preferences segment (see above) and by the negative value of the frustrated extra preferences that survival would bring with it. That, then, was very roughly how the value of satisfied extra preferences will determine the moral status of death and killing.

The Big Three Where have we got so far? Riding roughshod over some major qualifications and variations, here is a synopsis of what each of the three camps says about extra preferences, about the beginning of life, and about its end. The Rabbits say that a satisfied extra preference has positive value. Hence, roughly speaking, creating lives that contain satisfied extra preferences is obligatory, and sexual abstinence, contraception, and abortion are quite frequently wrong; one major reason why killing is wrong is that it prevents satisfied extra preferences from coming into existence. The Midwives say that a satisfied preference has positive value if, and only if, the preferrer exists. Hence, we are under no obligation to create anybody, but must make sure that those who exist get satisfied extra preferences; sexual abstinence and contraception are okay, but abortions and other killings are not, since they deprive the victim of satisfied extra preferences. The Mind Readers say that a satisfied extra preference is neither good nor bad, but neutral. Hence, there is no need to create anybody, and in many contexts we ought even to refrain from it; sexual abstinence and contraception are okay, and so is abortion (unless the fetus has preferences whose satisfaction requires its survival); death is bad in so far as it frustrates the preferences that precede it, but not because it withholds satisfied extra preferences from its victim.

Introduction to Possible Preferences

373

Claim (2) and the Midwives: Variations So much for the large picture. We can now proceed to some of the complications. Quite a few of them concern the Midwives and, which amounts to roughly the same thing, the supporters of claim (2). Identity. Claim (2) presupposes a concept of identity: providing satisfied extra preferences is obligatory given that an entity identical to the recipient of the preferences exists. It follows that the moral substance of claim (2) will vary with the underlying concept of identity. To illustrate this, let us return to the case of Mary and look at the egg and the sperm that, under suitable conditions, would develop into the adult Mary. Now, suppose that according to your concept of identity even the unfertilized egg, or the sperm, or the set {unfertilized egg, sperm} is Mary. If that were so, then even claim (2) would prescribe conception, viz. as a way to provide Mary - who already exists, as an egg or sperm or set - with satisfied preferences. Far towards the other end of the spectrum, imagine a theory that has personal identity over time constituted by mental representation over time: say by somebody's beliefs as to what will happen to him, and by his memories as to what has happened to him. In that case, claim (2) would not prescribe conception and might tolerate even late abortion, or even infanticide. Everything would depend, firstly, on when you think a fetus or baby begins to understand what is meant by a, or its, future; secondly, on how far you think people's memories reach back: to the womb, to the cradle, or just to kindergarten? Elsewhere in the spectrum (especially in between the options outlined in the previous two paragraphs), there are other concepts of identity each of which amounts to another modification of claim (2)'s moral upshot. Thus, concepts of identity generate many of the "cut-off points" that have been suggested in the debate on abortion and infanticide: points up to which killing a fetus, or even a baby, is held to be okay, and after which to be forbidden. 4 Earlier, we presented a table showing what Rabbits, Midwives, and Mind Readers say about conceiving Mary and about aborting her (see p. 370). To get a complete set of entries for the Midwives, we had to assume something about identity, and we have assumed that, as is sometimes held, neither the 4

We shall meet some more of these suggestions shortly. See also Glover (1977), ch. 9, Sumner (1981), § 16, Singer (1993), ch. 6, (1995), ch. 5, the section on "New Reproductive Technology and the Abortion Debate". As to people's identity, see e.g. Parfit (1984), part three, and, with special emphasis on zygotes etc., van Inwagen (1990), sects. 14 and 17-9, and Wessels, sect. 2.4.

374

Christoph Fehige and Ulla Wessels

unfertilized egg nor the sperm nor the set {unfertilized egg, sperm} are Mary, but that the fertilized egg is. In this respect, however, as the foregoing remarks will have made clear, there is considerable leeway within midwifism. Contrary to what the order of our presentation might suggest, the right way for a Midwife to go about these issues is not: to first endorse claim (2) and then analyse her concept of identity to find out what moral claim it is that she has just endorsed. Not to know what one is saying is hardly a benchmark of rationality. For a Midwife, the choice of an identity theory becomes a moral choice. It will thus require a moral argument, not a linguistic one. 5 Identity or identity plus? Now, it doesn't really matter whether you manoeuvre your specifications into claim (2) via identity or more directly. Similar specifications-are sometimes introduced on different routes by different authors. Suppose, for example, you want to say that, if a and b are stages in the history or pre-history of the same physical organism, providing b with satisfied preferences is obligatory only if a has been conscious. (Consciousness is what some people take to be the right cut-off point.) There are two ways to express this. Either you subscribe to claim (2) as it is and explain to us that, according to your concept of identity, an earlier and a later item cannot be the same person unless the earlier one has been conscious.6 Or you take a different concept of identity - a purely physical one, for instance - and write the consciousness requirement into claim (2) straightaway: a satisfied extra preference is good (the modified version would run) if, and only if, its bearer exists and has been conscious anyway.7 Not every condition that has been poured into claim (2) by somebody either via or over and above identity - can be mentioned here.8 It is sometimes held that a fetus acquires a right to life when it quickens.9 More specifically, only by kicking one's mother, we were once told by a sociologist, does one join the human community (a club with funny rules?). Others suggest as a criterion the arrival of the soul. As to the time of arrival, however, opinions 5

6

7

8 9

The same point has been made, in similar contexts, about words like "person", "human being", "life", and "the beginning of life"; their meaning will not by itself settle any moral issues. See e.g. Warnock (1987), pp. 1-6, and Hare (1993), sects. 7.9, 8.7, 10.3, 11.1 f. A position that McMahan, sect. 2.2, seems to sympathize with; similarly Haslett (1996), p. 174. For sentience (rather than consciousness), this is Sumner's strategy in (1981), § 16. (For the difference, or alleged difference, between consciousness and sentience, see ibid., p. 142, and Leist 1990, pp. 145 f.) Leists position in (1990), sect. V.3.a, is similar, but, given Leists notion of a preference, could be couched even in the Mind Readers' format; see p. 376 below. A considerable number of such conditions are discussed in the sources listed in note 4 above. Grisez (1970), pp. 374-97, traces some of the influence this criterion has had.

Introduction to Possible Preferences

375

are divided. Some think the soul is already in the sperm; others hold that it arrives at conception; or when the fetus has the shape of a human being; or at birth; or twelve days after birth; or never; or — not implausibly — that only God knows. 10 Satisfied extra preferences are sometimes thought to be good if, and only if, their recipient exists anyway and would receive them in the natural or normal course of events (if nobody interfered, or something along these lines).11 Sometimes particular psychological connections are required to hold between the temporal stage that receives the satisfied extra preferences and the temporal stage that exists anyway.12 The existence of so many relatives makes it reasonable to use terms like "Midwife", "Midwife*", or "claim (2)" for their entire families; let now each of them cover not just precisely the position for which it has been introduced earlier, but also modified versions similar to those we have just met. In its most general (though somewhat cumbersome) form, the question behind the variations is this: apart from a preference of as that b should have satisfied preferences, which conditions, be they matters of identity or not, suffice for the existence of an object a to generate a reason to provide an object b with satisfied preferences?13 Many answers have been proposed; the differences between them are, literally, vital.

Pleasure and Other Goods So far, our presentation has proceeded as if all that counted were preferences. What, one might ask, has happened to happiness in the hedonic sense: to pleasure, or feeling good? Preferences aside, if Mary isn't conceived, or is conceived and aborted, or is raised but killed, then she will have no fun, or less than she could have. Is this irrelevant?14 If we want to say that pleasure counts, we have two basic options. 10

11

12

13

14

For the twelve-day theory, see Batchelor (1901), p. 240. Prominent sources for the "never" theory can be spotted with the help ofLange (1866), esp. vol. 1, part 4, ch. II, vol. 2, part 3, ch. II; for the other claims, with the help of Emmel (1918), Grisez (1970), esp. chs. IV f., and Stockums (1924). Cf. Lenzen, sect. 4, and the critique by Wessels, sect. 2.4, subsections "What Fertilized Eggs Do by Themselves" and "External Intervention, Inner Nature, Biological Constitution". Cf. Tooley (1983), pp. 130-2, McMahan, sect. 3, and the objections from the beginning of Spitzley's critique; Lenzen, too, seems to hold (in sect. 6.1) that such connections make at least a quantitative difference. The point of this general wording becomes clearer if we think of a and b as entities that might be different time stages of the same object, organism, or person. The issues we have touched upon in this section - the interplay of identity, psychological continuity, desires, and the right to life - receive a detailed (and difficult) discussion in sects. 5-1 to 5.3 of Tooley (1983). Cf. Lockwood (1979), pp. 157 f., 164-9, Lenzen, sect. 3.

376

Christoph Fehige and Ulla Wessels

Linkingpreferences to pleasure. One option is to translate pleasure jargon into preference jargon. ^ This has two pleasant effects: we get a unified terminology, and the three positions we have come across so far cover the hedonic terrain as well. Given any translation of pleasure into preferences, we can simply proceed along the routes sketched out above and check what the results of each of the claims (l)-(3), coupled with the translation, are. A lot will then depend on the precise wording of the translation. Should we, for instance, say that it is analytically true of every individual that she wishes to have pleasant consciousness? (In that case, assuming that the fetus is the same individual as the corresponding adult, even Mind Readers would forbid abortion, since it would frustrate an existing preference for pleasure. The case from the second column of the table on p. 370 could simply not arise.) Or, slightly weaker, should we take it as an analytical truth that pleasant consciousness implies the preference for having more pleasant consciousness later on? (Thus Leist 1990, p. 147; in that case still, even Mind Readers would forbid abortion if and after the fetus has had some fun.) Or should we say that the strongest analytical truth in this area says that individuals want to spend pleasantly the time during which they are conscious?16 Not linking preferences to pleasure. If we don't want analytical links, then we have to work with two separate accounts: one for satisfied extra preferences, one for extra pleasure}7 Pretending that by now we have got to grips with extra preferences, we have to decide anew upon the principles that cover the value of extra pleasure. What principles, then? Preferences, if there, should be satisfied; similarly, consciousness, if there, should be pleasant. That much is clear. But what is the value of a pleasant extra moment of consciousness? There are at least three elementary types of options: ( H I ) A pleasant extra moment of consciousness is good. (H 2) A pleasant extra moment of consciousness is good if, and only if, its bearer exists anyway. (H 3) A pleasant extra moment of consciousness is neither good nor bad; its value is neutral. This list should evoke memories,18 and the reader is invited to indulge in 15 16

17 18

Cf. Kant (1785), pp. 415 f., Singer (1993), p. 131, Fehige, sect. 1. This is what Singer (1993), p. 131, seems to have in mind; for this position and the difference it makes, see Wessels, sect. 2.2., subsect. "How the Suicide Argument Survives Various Objections". Cf. e.g. Lockwood (1979), pp. 164 f. If it doesn't, see the section on extra preferences towards the beginning of this introduction.

Introduction to Possible Preferences

377

them, and their reverberations, at her leisure. And if you think there are other good things than preference satisfaction and ήδονή, you can pour them into the same sort of structure.19

Several People So far, our ceteris paribus clauses, mostly tacit, have helped us to concentrate on the fate of just one real or possible person: ought she to be conceived? May she be aborted? Ought she to get a satisfied extra preference? How bad would death be for her? Things become more complicated still if we move on to cases that involve several (possible) parties. The more parties, the more ways that other things can fail to be equal. Considerations concerning the quantity of lives lived might, for instance, conflict with those concerning their quality. Raising the number of people might lower the standard of living - should there rather be ten very happy people or twenty people who are 'half as happy'? And if not twenty, what about thirty? If we employ Derek Parfit s term "different people choices" for choices that make a difference to the number or the identity of people who will ever exist, 20 then the area in question is that of aggregationalproblems in the realm of different people choices. Many of the famous problems from that area, too, are presented and discussed in this symposium: replaceability and the NonIdentity-Problem, the Absurd Conclusion, the Repugnant Conclusion, and the Mere Addition Paradox.21 Aggregational problems of different preferences choices come in in impersonal versions, too: is it better for someone to have ten strong satisfied extra preferences or to have a hundred weak ones? Having nodded in their direction, our introduction will not pursue aggregational issues much further. Though fascinating and important, they are mostly posterior to the ceteris paribus questions which, for precisely that reason, have been treated here in some more detail. We are unlikely to know 19 20 21

Cf. McMahan, sects. 2.3 and 3. Parfit (1984), sect. 120. Replaceability and the Non-Identity Problem: see Parfit (1984), ch. 16, Fehige, sect. 7, Singer, sect. 2, Wessels, sect. 2.3, subsect. " T h e Suicide Revisited"; the Absurd Conclusion: see Parfit (1984), ch. 18, Fehige, sect. 7; the Repugnant Conclusion: see Parfit (1984), ch. 17, Fehige, sect. 6, Hare, sect. 4; the Mere Addition Paradox: see Parfit (1984), ch. 19, Fehige, sect. 6, Singer, sect. 3. T h e cruelty of average utilitarianism is another example, see Parfit (1984), sect. 138.

378

Christoph Fehige and Ulla Wessels

the value of several lives of a certain type as long as we don't even know the value of one them; and we are unlikely to know the value of one life of a certain type as long as we don't even know the value of one part of it. And some of the aggregational puzzles in population ethics, we think, have far more to do with general questions of justice than with different people choices in particular. To frame this as an empirical hypothesis: if a general principle of justice, applied to a certain different people choice, comes up with a judgement that a certain moral thinker finds counter-intuitive, then the chances are that there is a same people choice with, in some sense, the same property. That is to say, applied to that choice, the principle will also come up with a judgement the thinker finds counter-intuitive, and some structural analogy between the sources of counter-intuitiveness will leap to the eye. Take, for example, the so-called Repugnant Conclusion, a claim generally taken to follow from, and to embarrass, total utilitarianism (the doctrine, that is, that bids us to maximize the sum total of welfare). T h e Repugnant Conclusion says that for any number k, however large, there is a number η such that the existence of η people who consider their lives barely worth living is better than the existence of k very happy people. Now, compare this to the following claim from the realm of same people choices: for any number k, however large, there is a number η such that satisfying one tiny preference of each of η persons is more important than saving k of these people from getting roasted in hell for one hundred years each. Call this the Hellish Conclusion. It would be quite a feat to spot somebody who finds the Repugnant Conclusion counter-intuitive but not the Hellish Conclusion. There seems to be, if a problem at all, then a general problem for utilitarianism: an enormous lump of disutility, inflicted on each of no matter how many persons, can always be outweighed by providing each of a sufficiently large number of individuals with one crumb of utility. Suppose our hypothesis that most of the aggregational trouble is of a general type can be confirmed along the lines thus illustrated. Then the right morality for different people choices is likely to be found not so much in aggregational principles tailor-made for, and checked and discussed with a special view to, population ethics; but rather in a combination of, on the one hand, general principles and discussions of justice (making little or no particular reference to different people choices) and, on the other hand, the type o f p r o tanto considerations - about the value, for instance, of a satisfied extra preference or an extra moment of pleasure - that this introduction has focused on. Same people or different people, justice itself is then deeply uniform. All we have to do is let it loose on the right concept of welfare.

Introduction to Possible Preferences

379

The Contributions to this Symposium We can now link the papers that await us to what has been said so far. The most elementary connection is that the symposium includes representatives from each of the three camps. Here is a more specific guide. Peter Singers contribution begins on p. 383 and should be read along with this introduction; he surveys major arguments, options, and problems, quite a few of which he himself has shaped or initiated over the last twenty years. Ground covered includes the relation of preference to happiness, replaceability, the Non-Identity Problem, the Mere Addition Paradox, Hurka's theory of the diminishing marginal value of happy lives, and Heyd's claim that, in essence, what should function as the moral measuring rod for different people choices are the wishes of the procreators rather than those of the offspring. Singer sees the Rabbits' and the Mind Readers' position as the essential candidates today; having compared the drawbacks of each of them, he concludes that, currently and by a narrow margin, the Rabbits have the better case. Richard M. Hare, whose paper begins on p. 399, is a Rabbit, and probably the paradigm Rabbit. He has been defending the position rigorously and repeatedly since at least 1975 (see the reprinted essays 5 f. and 10—12 in his 1993); as far as we know, he is also - appeals to the Bible and to moral intuitions aside - the inventor of the only genuine argument for it. Hare argues that to be moral is to have analogous preferences for analogous situations; that some real-life people want to have been born; and that, hence, morality requires them to have, for analogous situations (in which other people's birth is at issue), analogous preferences (preferences for those peoples birth, that is). Hares paper restates his view and adds further explanations. Taking as his point of departure an argument of Hajdin's, Hare also discusses asynchronic preferences, the relations of preferences to pleasure, the roles all these should play in moral reasoning, and their impact on the morality of possible people in particular. Wolfgang Lenzen, whose paper begins on p. 406, is a Midwife. He presents one specific form of midwifism, based on a theory of the value of life and on a minimal moral theory that tries to combine the maxim "neminem laedere" with a number of utilitarian ideas. Like most versions of midwifism, his theory implies that it is, roughly speaking, not obligatory to conceive, but obligatory not to abort. What underpins the verdict is firstly certain differences between a fertilized egg and an unfertilized egg; secondly the claim that it may be immoral to deprive an individual of future goods even if these are not yet desired. The paper includes a critical discussion of the Mind Readers' and of the Rabbits' position, and replies to some of their objections against midwifism.

380

Christoph Fehige and Ulla Wessels

Ulla Wessels, whose paper begins on p. 429, criticizes the positions of Hare and Lenzen. Challenging Lenzens background theory, she sets out to show that, by itself, undesired future happiness does not argue against death; and that, if it did, it would also argue for the creation of happy individuals. And, she goes on, suppose that we grant Lenzen his background theory: does it really entail that it is when sperm and egg fuse that the right to life begins? Lenzen does not give us the theory of identity that could support that inference. Hares argument, too, says Wessels after a detailed analysis, fails. Under one reading, the method it employs could generate deontic contradictions and is therefore inadequate. Under the other readings, the argument is left with a large hole: in order to show that people with satisfied preferences ought to exist it needs the extra premiss that satisfied preferences ought to exist. Most of Wessels's points are, or imply, fairly general objections to rabbitism and midwifism; thus, to the extent that the objections are valid, her paper is an argument ex negative for mind reading. JeffMcMahan, whose paper begins on p. 471, is a Midwife*. His paper concentrates on death (both pre- and postnatal) and on the role of preferences in explaining why death is usually bad. A discussion of the standard problems and answers in the field gives rise to a new theory that preserves parts of previous approaches, but modifies and supplements them. All Midwives* say that a persons goods-at-point-of-time-ί' speak against her death-at-an-earlier-pointof-time-ί. McMahan's central innovation is the claim that the degree to which they do so is proportional to the psychological connectedness of the personat-t with the person-at-ί'. The innovation is explained and motivated, and is shown to avoid certain counter-intuitive implications of unmodified midwifism. Thomas Spitzley, whose paper begins on p. 503, criticizes McMahans contribution. Firstly, he questions the central building block of McMahan's theory, the notion of psychological continuity: the notion itself, he says, raises conceptual puzzles, and McMahans moral claims involving it are arbitrary, counter-intuitive, or both. Secondly, Spitzley is sceptical about McMahan's use of moral intuitions in general. Thirdly, he doubts whether McMahans theory covers as large a terrain as it claims to: if we want to calculate the badness of a natural death along McMahans lines, then what survival scenario are we supposed to compare it to? Christoph Fehige, whose paper begins on p. 508, is a Mind Reader. He explains and defends what has figured in this introduction as claim (3) - viz., that a satisfied extra preference is of no value. He calls that position antifrustrationism and proceeds to develop a Mind Readers morality based on it. Some

Introduction to Possible Preferences

381

of his reflections on the way (those that concern, as he puts it, the format of Pareto principles for different people choices) are independent of antifrustrationism or mind reading. His paper also discusses the Mere Addition Paradox (and its antifrustrationist solution), the Repugnant Conclusion (plus a very repugnant cousin of it, both rejected by antifrustrationism), and various related problems. *

It has been our aim to facilitate the reader's orientation in the realm of possible preferences and in this symposium. The field itself is complex, and our introduction could not help sharing this fate to at least some extent. Our exposition was centered around two times three claims: firstly, around claims (l)-(3) about the value of satisfied extra preferences; secondly and correspondingly, around the claims of the Rabbits, the Midwives, and the Mind Readers. Keeping these trinities in mind, and understanding everything else as variations on that theme, is perhaps a helpful strategy for finding a way through the jungle of arguments, counter-arguments, methods, intuitions, thought-experiments, and subtle differences.

References Contributions to this symposium have been referred to by authors' names only, without year numerals, and are not listed here. (1901). John Batchelor: The Ainu and their Folk-Lore, London 1901. Karl Emmel: Das Fortleben der antiken Lehren von der Beseelung bei den Kirchenvätern, Leipzig 1918. G L O V E R (1977). Jonathan Glover: Causing Death and Saving Lives, Harmondsworth 1977. G R I S E Z (1970). Germain Grisez: Abortion, New York 1970. H A R E (1993). Richard M. Hare: Essays on Bioethics, Oxford 1993. H A S L E T T ( 1 9 9 6 ) . D. W . Haslett: "On Life, Death, and Abortion", Utilitas 8 ( 1 9 9 6 ) . VAN I N W A G E N (1990). Peter van Inwagen: Material Beings, Ithaca 1990. K A N T (1785). Immanuel Kant: Grundlegung zur Metaphysik der Sitten, Berlin 1911, vol. IV of the Royal Prussian Academy edition of Kant's gesammelte Schriften; Grundlegung first publ. in 1785. L A N G E (1866). Friedrich Albert Lange: Geschichte des Materialismus, second, rev. ed., Frankfurt/Main 1974, 2 vols.; first ed. first publ. in 1866, second ed. in 1873. L E I S T (1990). Anton Leist: Eine Frage des Lebens, Frankfurt/Main 1990. BATCHELOR

EMMEL ( 1 9 1 8 ) .

382

Christoph Fehige and Ulla Wessels

(1979). Michael Lockwood: "Singer on Killing and the Preference for Life", Inquiry 22 (1979). P A R F I T (1984). Derek Parfit: Reasons and Persons, Oxford 1987; first ed. first publ. in 1984. S I N G E R (1993). Peter Singer: Practical Ethics, second ed., Cambridge 1993; first ed. first publ. in 1979. S I N G E R (1995). Peter Singer: Rethinking Life and Death, Oxford 1995. S T O C K U M S (1924). W. Stockums: "Historisch-Kritisches über die Frage: 'Wann entsteht die geistige Seele?'", Phibsophisches Jahrbuch der Görresgesellschafi 37 (1924). S U M N E R (1981). L. W. Sumner: Abortion and Moral Theory, Princeton 1981. T O O L E Y ( 1 9 8 3 ) . Michael Tooley: Abortion and Infanticide, Oxford 1 9 8 3 . W A R N O C K (1987). Mary Warnock: "Do Human Cells Have Rights?", Bioethics 1 (1987). LOCKWOOD

PETER

SINGER

Possible Preferences 1. 2. 3. 4. 5.

My Position in Practical Ethics Criticism of My Position in Practical Ethics Why I Am not a 'Mind Reader' Life as a Journey And in the End ... References

1.

My Position in Practical Ethics

It may be useful to begin by re-stating the position I developed in the original edition of Practical Ethics (Singer 1979, henceforth ΡΕ). I shall then indicate some sources of criticism, and possible ways of handling them. In ΡΕ I approached these issues via the question: why is it wrong to take life? I sought to answer this question both in respect of life that is merely conscious, that is capable of experiencing pleasure or pain, but not self-aware or capable of knowing that it exists over time, and hence not capable of wanting to go on living; and in respect of life that is self-aware and capable of wanting to go on living. I reserved the term "person" for the latter kind of self-aware being, irrespective of species. In considering the wrongness of killing a conscious being that is not a person, I suggested a very simple answer: we value pleasure, killing those who lead pleasant lives eliminates the pleasure they would otherwise experience, therefore such killing is wrong. I then pointed out that stating the argument in this way conceals something that, once noticed, makes the issue anything but simple: "There are two ways of reducing the amount of pleasure in the world: one is to eliminate pleasures from the lives of those leading pleasant lives; the other is to eliminate those leading pleasant lives. The former leaves behind beings who experience less pleasure than they otherwise would have. The latter does not. This means that we cannot move automatically from a preference for a pleasant life rather than an unpleasant one, to a preference for a pleasant life rather than no life at all. For, it might be objected, being killed does not make us worse off; it makes us cease to exist. Once we have ceased to exist, we shall not miss the pleasure we would have experienced." (ΡΕ, p. 85.)

384

Peter Singer

To show that this is not merely splitting hairs, I invited readers to consider the opposite case: a case not of reducing pleasure, but of increasing it: "There are two ways of increasing the amount of pleasure in the world: one is to increase the pleasure of those who now exist; the other is to increase the number of those who will lead pleasant lives. If killing those leading pleasant lives is bad because of the loss of pleasure, then it would seem to be good to increase the number of those leading pleasant lives. We could do this by having more children, provided we could reasonably expect their lives to be pleasant, or by rearing large numbers of animals under conditions that would ensure that their lives would be pleasant. But would it really be good to create more pleasure by creating more pleased beings?" (PE, p. 86.) I then distinguished two possible approaches to these issues. The first, which I called the "total" view, accepts that it is good to increase the amount of pleasure in the world by increasing the number of pleasant lives, and bad to reduce the amount of pleasure in the world by reducing the number of pleasant lives. The second approach, the "prior existence" view is to count only beings who already exist, prior to the decision we are taking, or at least will exist independently of that decision. This view denies that there is value in increasing pleasure by creating additional beings. I believe that the prior existence view is more in harmony with most people's intuitive judgments today. At least in modern secular societies, we do not generally think that couples are under a moral obligation to have children when the children are likely to lead pleasant lives and no one else is adversely affected. But I do not place much weight on this for two reasons. First, I am sceptical about the value of such intuitive judgments as a test of moral theory, since in different times and circumstances people may have quite different intuitions. Second, it is not clear that our intuitions in this area are consistent. In particular, while we commonly hold that there is no moral obligation to have children when they will lead pleasant lives, we also hold that it is wrong to bring into the world a child who will suffer for a few months from an incurable disease, and then die. We would think it wrong for a couple knowingly to conceive such a child; but if the pleasure a possible child will experience is not a reason for bringing it into the world, why is the pain a possible child will experience a reason against bringing it into the world? The prior existence view must either hold that there is nothing wrong with bringing a miserable being into the world, or explain the asymmetry between cases of possible children who are likely to have pleasant lives, and possible children who are likely to have miserable lives. I note in passing that it will not do to say: "The possibly happy child we

Possible Preferences

385

fail to bring into existence does not grieve at the loss of the happy life she might have experienced; whereas the miserable child we bring into existence does experience the misery of her existence." For to say this is to focus on only one alternative in each case. One could just as well say: "The possibly miserable child we fail to bring into existence does not rejoice at having been spared the misery she might have experienced; whereas the happy child we bring into existence does experience the happiness of her existence." In ΡΕ I suggested that proponents of the prior existence view might be forced, here, to a rather unconvincing response: that there is nothing directly wrong in conceiving a child who will be miserable, but once such a child exists, since its life can contain nothing but misery, we should reduce the amount of pain in the world by an act of euthanasia. Since euthanasia is a more harrowing process for the parents and others involved than non-conception, this gives us an indirect reason for not conceiving a child bound to have a miserable existence. This response is weak, because it is possible that once the miserable child exists, it will be beyond our power to carry out the act of euthanasia. Then where does the wrongness lie? This reveals an oddity at the heart of the formulation of the prior existence view: it refers to being who already exist at the time of the decision "or at least will exist independently of that decision". Thus it says nothing about decisions that cause beings to exist. So it is possible that no objection can be made to a decision that causes a being to exist, but from the very moment that decision has been made, and the process that will bring the being into existence is set in train, it is obligatory for us to stop this process if we possibly can. Surely there is something wrong with a morality that can ground no objection to doing A, the sole effect of which is to cause X to come into existence, and then tells us that we must do what we can to end the existence of X at the earliest possible moment. Let us turn now to the total view. One implication of the total view is the replaceability of conscious beings that are not persons. This appears to be a helpful argument for meat-eaters: although meat-eaters are responsible for the death of the animal they eat and for the loss of pleasure experienced by that animal, they are also responsible for the creation of more animals, since if no one ate meat there would be no more animals bred for fattening. The loss meat-eaters inflict on one animal is thus balanced, on the total view, by the benefit they confer on the next. (This assumes, of course, that the animals lead pleasant lives until they are killed; a very doubtful assumption in today's world of closely confined, intensively reared animals; and, at least within the framework in which we are now considering the issue, it also assumes that the animals are not capable of understanding that they have a future.) The argument also has a consequence that many people find much more disturbing: it

386

Peter Singer

makes infants replaceable, up to the time at which they are capable of forming preferences to go on living. This is almost certainly not in the first month or two after birth. Hence the notorious suggestion in PE, which in Germany can still only be discussed behind closed doors, that even an infant whose life prospects are not totally miserable - for example, a haemophiliac - could be killed, if the parents were planning to replace the infant with another who would have fewer obstacles to a completely fulfilling life.

2.

Criticism of My Position in Practical Ethics

The replaceability argument was probably the most controversial, and widely criticised, argument in PE. Unfortunately none of the critics have offered satisfactory alternative solutions to the underlying problems to which replaceability offers one - if not very congenial — answer. Derek Parfit has described another hypothetical situation that amounts to an even stronger case for the replaceability view. (Parfit 1976a; for a restatement see Parfit 1984, p. 376.) He asks us to imagine that two women are each planning to have a child. The first woman is already three months pregnant when her doctor gives her both bad and good news. The bad news is that the fetus she is carrying has a defect that will significantly diminish the future child's quality of life - although not so adversely as to make the child's life utterly miserable, or not worth living at all. The good news is that this defect is easily treatable. All the woman has to do is take a pill that will have no side-effects, and the future child will not have the defect. In this situation, Parfit very plausibly suggests, we would all agree that the woman should take the pill, and that she does wrong if she refuses to take it. The second woman sees her doctor before she is pregnant, when she is about to stop using contraception, and also receives bad and good news. The bad news is that she has a medical condition that has the effect that if she conceives a child within the next three months, that child will have a significant defect - with exactly the same impact on the child's quality of life as the defect described in the previous paragraph. This defect is not treatable, but the good news is that the woman's condition is a temporary one, and if she waits three months before becoming pregnant, her child will not have the defect. Here too, Parfit suggests, we would all agree that the woman should wait before becoming pregnant, and that she does wrong if she does not wait. Suppose that the first woman does not take the pill, and the second woman does not wait before becoming pregnant, and that as a result each has a child with a significant disability. It would seem that they have each done some-

Possible Preferences

387

thing wrong. Are their wrongdoings of equal magnitude? If we assume that it would have been no greater hardship for the second woman to wait three months before becoming pregnant than it would have been for the first woman to take the pill, it would seem that the answer is yes, what each has done is equally wrong. But now consider what this answer implies. The first woman has harmed her child. That child can say to her mother: "You should have taken the pill. If you had done so, I would not now have this disability, and my life would be significantly better." But if the child of the second woman tries to make the same claim, her mother has a devastating response. She can say: "If I had waited three months before becoming pregnant you would never have existed. I would have produced another child, from a different egg and different sperm. Your life, even with your disability, is definitely above the point at which life is so miserable that it ceases to be worth living. You never had a chance of existing without the disability. So I have not harmed you at all." This reply seems a complete defence to the charge of having harmed the child now in existence. If, despite this, we persist in our belief that it was wrong of the woman not to postpone her pregnancy, where does the wrongness lie? It cannot lie in bringing into existence the child to whom she gave birth, for that child has an adequate quality of life. Could it lie in not bringing a possible being into existence — to be precise, in not bringing into existence the child she would have had if she had waited three months? This is one possible answer, but it commits us to the total view, and implies that, other things being equal, it is good to bring into existence children without disabilities. A third possibility is that the wrongdoing lies, not in harming an identifiable child, nor simply in omitting to bring a possible child into existence, but in bringing into existence a child with a less satisfactory quality of life than another child whom one could have brought into existence. In other words, we have failed to bring about the best possible outcome. This last seems the most plausible answer, but it too suggests that at least possible people are replaceable. The question then becomes: at what stage in the process that passes from possible people to actual people does replaceability cease to apply? What characteristic makes the difference? In ΡΕ I tried to solve the problem by drawing a line between beings that are merely conscious, and those who are persons. I suggested that in respect of self-conscious individuals, leading their own lives and wanting to go on living, the replaceability argument holds little appeal, because it clearly thwarts the desires of such beings to go on living. Beings that lack the conception of themselves as living beings with a future, however, cannot be harmed in this 'personal' sense. Killing them does reduce the quantity of happiness in the universe. But this wrong, if it is wrong, can be counter-balanced by bringing

388

Peter Singer

into existence similar beings who will lead equally happy lives. So, I argued, perhaps the capacity to see oneself as existing over time, and to have other non-momentary, future-directed interests, is the characteristic that marks out those beings who cannot be considered replaceable. Another way of putting this point is to say that rational, self-conscious beings are individuals, leading lives of their own, and cannot in any sense be regarded merely as receptacles for containing a certain quantity of happiness. They have, in the words of James Rachels, a life that is biographical, and not merely biological (Rachels 1986, pp. 24-7). Beings who are conscious, but not self-conscious, on the other hand, more nearly approximate the picture of receptacles for experiences of pleasure and pain, because their preferences will be of a more immediate sort. They will not have desires that project their images of their own existence into the future. Their conscious states are not internally linked over time. We can presume that if fish become unconscious, then before the loss of consciousness they would have no expectations or desires for anything that might happen subsequently, and if they regain consciousness, they have no awareness of having previously existed. Therefore if the fish were killed while unconscious and replaced by a similar number of other fish who could be created only because the first group of fish were killed, there would, from the perspective of fishy awareness, be no difference between that and the same fish losing and regaining consciousness. For a non-self-conscious being death is the cessation of experiences, in much the same way that birth is the beginning of experiences. Death cannot be contrary to an interest in continued life, any more than birth could be in accordance with an interest in commencing life. To this extent, with non-self-conscious life, birth and death cancel each other out; whereas with self-conscious beings the fact that once self-conscious one may desire to continue living means that death inflicts a loss for which the birth of another is insufficient compensation. Richard M. Hares test of universalizability (Hare 1981) supports this view. If I imagine myself in turn as a self-conscious being and a conscious but not self-conscious being, it is only in the former case that I could have forwardlooking desires that extend beyond periods of sleep or temporary unconsciousness, for example a desire to complete my studies, a desire to have children, or simply a desire to go on living, in addition to desires for immediate satisfaction or pleasure, or to get out of painful or distressing situations. Hence it is only in the former case that my death involves a greater loss than just a temporary loss of consciousness, and is not adequately compensated for by the creation of a being with similar prospects of pleasurable experiences. In reviewing PE in The New York Review of Books, H. L. A. Hart raised

Possible Preferences

389

an objection to this attempt to solve the problem (Hart 1980). He suggested that for a utilitarian, self-conscious beings must be replaceable in just the same way as non-self-conscious beings are. Whether one is a preference utilitarian or a classical utilitarian will, in Hart's view, make no difference here, because: "Preference Utilitarianism is after all a form of maximizing utilitarianism: it requires that the overall satisfaction of different persons' preferences be maximized just as Classical Utilitarianianism requires overall experienced happiness to be maximized. [...] If preferences, even the desire to live, may be outweighed by the preferences of others, why cannot they be outweighed by new preferences created to take their place?" (Hart 1980, pp. 29 f.)

It is of course true that preference utilitarianism is a form of maximizing utilitarianism in the sense that it directs us to maximize the satisfaction of preferences, but Hart is on weaker ground when he suggests that this must mean that existing preferences can be outweighed by new preferences created to take their place. For while the satisfaction of an existing preference is a good thing, the package deal that involves creating and then satisfying a preference need not be thought of as equivalent to it. Again, universalizability supports this way of conceiving preference utilitarianism. If I put myself in the place of another with an unsatisfied preference, and ask myself if I want that preference satisfied, the answer is (tautologically) yes. If on the other hand I ask myself whether I wish to have a new preference created that can then be satisfied, I will be quite uncertain. If I think of a case in which the satisfaction of the preference will be highly pleasurable, I may say yes. (We are glad that we are hungry if delicious food is on the table before us, and strong sexual desires are fine when we are able to satisfy them.) If, on the other hand, I think of the creation of a preference that is more like a privation, I will say no. (We don't cause ourselves headaches simply in order to be able to take aspirin and thus satisfy our desire to be free of the pain.) This suggests that the creation and satisfaction of a preference is in itself neither good nor bad: our response to the idea of the creation and satisfaction of a preference varies according to whether the experience as a whole will be desirable or undesirable, in terms of other, longstanding preferences we may have, for example for pleasure rather than pain. Exactly how preference utilitarianism ought to evaluate the creation and satisfaction of a preference, as distinct from the satisfaction of an existing preference, is a difficult issue. In my initial response to Hart's criticism (Singer 1980), I suggested that we think of the creation of an unsatisfied preference as putting a debit in a kind of moral ledger that is merely cancelled out by the satisfaction of the preference. (Some will see in this model confirmation of

390

Peter Singer

Marx's scornful remark that Bentham's utilitarianism is a philosophy suitable for a nation of shopkeepers!) The moral ledger model has the advantage of explaining the puzzling asymmetry in our intuitions about bringing into existence a child who will lead a thoroughly miserable existence, and failing to bring into existence a child who, in all probability, will lead a happy life. The debit view of preferences just outlined would explain why we should feel that the former is seriously wrong, but the second is not a wrong of any kind: to bring into existence a child, most of whose preferences we will be unable to satisfy, is to create a debit that we cannot cancel. This is wrong. To create a child whose preferences will be able to be satisfied, is to create a debit that can be cancelled. This is, on this model, ethically neutral. The model can also explain why, in Parfit's example, what the two women do is equally wrong for both quite unnecessarily bring into existence a child who is likely to have a larger negative balance in the ledger than a child they could have brought into existence.

3.

Why I Am not a'Mind Reader'

It is probably because I have, in recent years, referred to this 'moral ledger' model that the editors thought that I would represent, in this exchange, the view that they describe as that of the Mind Reader: the view that all we ought to do is make sure that the preferences that have existed, exist or will exist are satisfied ones (see, in this volume, the "Introduction to Possible Preferences", where the alternative positions - those of the Midwives and the Rabbits - are also presented.) Unfortunately, at least as far as this symposium is concerned, I have spent some time in preparing a revised edition of Practical Ethics (Singer 1993); and in so doing, I have come to see that the 'moral ledger' view has undesirable implications as well as desirable ones: it makes it wrong, other things being equal, to bring into existence a child who will on the whole be very happy, and will be able to satisfy nearly all of her preferences, but will still have some preferences unsatisfied. For if the creation of each preference is a debit that is cancelled only when the desire is satisfied, even the best life will, taken in itself, leave a small debit in the ledger. Since everyone has some unsatisfied desires, the conclusion to be drawn is that it would have been better if none of us had been born. Does this difficulty mean that we must reject the moral ledger model? What other options are there? To accept the conclusion would be to adhere to a form of Pessimism that surpasses even that of Schopenhauer. It is absurd to hold, of even the most fulfilling lives, that because some desires could not

391

Possible Preferences

100

World A

World Β

World C

Fig. 1

be satisfied, it would have been better if they had never been lived. So we cannot accept the conclusion. Can we deny that it follows from the moral ledger model? Not, I think, unless we find a way of modifying that model. Could we attach to it a stipulation that sets a level of preference satisfaction, below complete satisfaction, as a minimum for overcoming the negative entry in the ledger that is opened by the creation of a being with unsatisfied preferences? This might be the level at which we consider a life ceases to be worth living, from the perspective of the person leading that life. Such a 'solution suffers from the disadvantage of being ad hoc. In addition, though, and more seriously, the modification can only be made at the cost of giving up the advantages of the moral ledger view. For once we set a minimum level that overcomes the negative entry in the ledger, it would seem undeniable that to bring into existence a being whose life will exceed this minimum by a significant margin is a good thing ... and thus all the implications of the total view return to haunt us, if in a modified form. There is also another problem with views like the moral ledger view, one that Parfit pointed out many years ago in response to my first foray into this difficult area (Parfit 1976b). This is that if we claim to be indifferent between worlds in which more people are brought into existence, it is easy to show that our preferences can become intransitive. Consider the populations in figure 1. World A has a billion very happy people, who can satisfy nearly all of their preferences. World Β has the same billion very happy people at the same level of happiness, plus another billion whose existence does not in any way detract from happiness of the first billion. This second billion are also happy, although because they live on a different continent, with a much colder climate, they are not able to satisfy quite so many of their preferences, and hence are not quite so happy, as the first billion. There is, at this stage, no contact between the populations in the two continents. In World C, the two continents are

392

Peter Singer

in contact, and those who do not enjoy cold, snowy winters have moved to the warmer continent. This has made it a bit more crowded, and slightly reduced the happiness of those who were living there before, but it has made those remaining on the colder continent happier still, because now they are less crowded. Overall, the shift in population has satisfied more preferences than it has frustrated, and has increased average happiness in the world as a whole. So how does a defender of the moral ledger view, or indeed any Mind Reader, compare the different worlds? It would seem that anyone who holds such a position must be indifferent between A and B, because the additional people in Β lead good lives, and do not detract from the degree of satisfaction of the original billion in Α. Β is, at least, not worse than A. But in comparing Β and C, we are comparing populations of the same size, and there is more preference satisfaction in C, so C must be better than B. Now let us compare A and C. Since the original billion in A are worse off, and since on this view the addition of extra people leading happy lives is a matter of indifference, it seems that C must be worse than A. Therefore the Mind Reader holds that C is better than B, but Β is not worse than A, and A is better than C. Obviously there is a problem here. Other possible solutions have been suggested. Thomas Hurka (1983) has pointed out that when a species is endangered, we place great value on an increase in numbers; but this value declines as the population size grows, and beyond a certain point, we may put no value at all on further increases. Similarly in the case of the human population, he suggests, we might accept what he calls "the variable value view". At low levels of population this principle gives more value to increases in population size than it gives to increases in average levels of well-being; but at higher levels of population the balance changes so that increasing average levels of well-being becomes more important than adding to the size of the population. Finally a point is reached at which further additions to the size of the population are of no value at all. Hurka himself notes that the mere addition paradox remains a serious objection to a variable value view. It can be illustrated by the diagrams above, if we suppose that one billion is the point at which further additions to the size of the population are of no value at all. (If this figure seems too small, merely multiply all the population sizes in the diagram by a factor of 10, or 100, or whatever is required to reach the point at which further additions cease to be of value.) Then the variable value view must say that Β is worse than A; because once we ignore its greater population size, the only thing left on which we can base a comparison between A and Β is the higher average level of welfare in A. But the difference between A and Β is merely that there are additional people

Possible Preferences

393

in Β whose average level of welfare is - through no fault of anyone's — lower than the average level in A. Despite this lower average level of welfare, the people in Β are all happy and able to satisfy most of their preferences. They are glad that they are alive, and their existence has no adverse effect on the welfare of anyone else. How then can it be a bad thing that they exist? Parfit (1984, pp. 401-12) has suggested another reason why we should reject the idea that there is a point beyond which further increases in the size of the population have no value, even if the additional population consists, and will continue to consist, solely of very happy people who can satisfy virtually all of their preferences. This reason is based on our judgment that intense suffering is a bad thing, and the more intense suffering there is, the worse the overall state of affairs is. Let us then make the plausible assumption that in any large population of very happy people there will be a few who, through no fault of anyone else, experience intense suffering. (Perhaps they are infants born with very painful diseases, who live only for a short time in pain and then die.) Now suppose that the world has 5 billion people in it, as it now does (roughly) have, and that of these, 10 000 experience lives of intense suffering. This is, unfortunately, almost certainly an extreme under-estimate of the number of people in the real world who experience lives of intense suffering. Despite the existence of rather more than 10 000 people living lives of intense suffering, most of us think that the continuation of human life on earth is a good thing. This suggests that we believe that the positive value of the lives of4.99999 billion people outweighs the negative value of 10 000 lives of intense suffering. Now imagine that we suddenly discover that there are 10 other inhabited planets in our solar system, each inhabited by 5 billion beings very like us; but on each of these planets, instead of 10 000 people living lives of intense suffering, there are only 10 who have this miserable kind of existence. Is the universe as a whole a better or worse place than we had believed it to be, before we discovered extra-terrestrial life? Most of us would think it a better place. Some might think it no different in value. But could anyone think it worse? Surely not. Yet that is exactly what we would have to believe, if we accept the previously mentioned assumption, that intense suffering is a bad thing, and the more of it there is, the worse it is; and if at the same time we hold that 5 billion is the point at which further increases in population size cease to have any value. For then there would be no additional value in the existence of the extra 50 billion people we have now discovered; but there would be disvalue in the lives of an additional 50 who are experiencing intense suffering. (Once again, nothing depends on whether the point at which further increases in population are of no value is 5 billion, or 10 or 100 times that figure; whenever it is reached, the objection holds.)

394

Peter Singer

In his recent book Genethics, David Heyd (1992) has proposed a more radical solution to these problems. He calls his position "generocentrism". This is, he tells us, "essentially the thesis that choices about whether to bring possible people into existence can and should be guided exclusively by reference to the interests, welfare, ideals, rights, and duties of those making the choice, the generators', the creators or the procreators." This, at least, is how he wants to handle pure cases of such "genesis problems", as he calls them. Pure cases are those in which no third party, other than the generators and their potential offspring, is affected. In the real world, of course, third parties will be affected by the procreative choices made by others. These third parties may be actually existing people, or they may be future people who do not yet exist, but who will exist, irrespective of the procreative decision we are about to make. Thus Heyd makes a sharp distinction between, on the one hand, actual people (whether present or future) and potential people, who may or may not come into existence. His generocentric ethics tell us that we ought to give full consideration to the interests of all actual people; but we are under no obligation to give any consideration to potential people. This is a daring attempt to solve all genesis problems, but it fails. Suppose that for genetic reasons, any child that a couple can have will suffer from a very painful condition for about six months, and then die. Nevertheless, the couple would like to have such a child. They will gain some pleasure from the experience of pregnancy, and from feeling that they are caring for another being. The pleasure they gain will be significantly less than the pain the child suffers, and they know this. Should they go ahead and have the child? It seems that if they accept Heyd's view, their reasoning must go like this. Before they have made any decision to have the child, their interests count but those of the child do not. So they may rightly decide to have the child. The child is duly conceived. Now the child becomes actual, and its interests count. Therefore they should not have the child, for the child's suffering outweighs their interest in having the child. So they do what they ought to do, and decide to have the fetus aborted. But once a decision is made to abort the child, the child ceases to be actual, and becomes merely possible. Therefore the child's interests no longer count, and from that moment, the couple's interests again lead to the conclusion that the pregnancy should be continued; but then . . . and so on. This seems an absurd situation for any ethical position to be in. Can Heyd avoid it? He might say that so long as the possibility of abortion exists, the child is not actual. Then the parents would be right to continue the pregnancy until the child is born, and if euthanasia is not permitted, the child would have to go through its miserable life for the benefit of its parents. This hardly seems the right answer either.

Possible Preferences

395

Heyd must also face a number of other problems. First and most obviously, on his view the human race would not do wrong if we were all collectively to agree that we did not want to have any more children. We would thus bring our species to an end. On a generocentric view, if that was what we wanted, the decision could not be criticised on ethical grounds, even if the prospective future of our species were much better than it really is - that is, even if war, poverty and disease had been banished from the face of the earth. If this conclusion can be swallowed, there is worse to come. Suppose that, instead of not reproducing at all, the present generation embraced an ethic of "apres nous, le deluge"; it decided to embrace a lifestyle of high energy and resource consumption, attending only to short-term pollution problems that might have an adverse impact within a single generation. Let us suppose, too, that the present generation did this, well aware of the fact that there would be future generations who would suffer because of the long term pollution problems and depleted resources that it would inherit. The policy of the present generation seems selfish and clearly wrong. But can a generocentric ethic show it to be wrong? It might seem that Heyd has no problem here. For there will be future generations. Since actual' people can be future as well as present, this means that the policies of the present generation will cause actual people to suffer. But the problem is that, as Parfit has pointed out, our choice of consumption policies affects the way in which people live, and hence the partners they meet and the time at which they have their children. It thus affects the identity of future people. It may be that, say, a hundred and fifty years after the present generation decides to adopt its high consumption policy, there is no-one alive who can truthfully say: "If the earlier generation had been less selfish in its choices, I would now be living a better life, in a less polluted world." For to this one could respond: "If the earlier generation had been less selfish, your parents would never have met, or if they had met, would not have had a child together, or if they had produced a child, would have produced a different child; so you would not have been living at all." So the members of future generations may not be actual, even though we may know that there will be some people alive in the future.

4.

Life as a Journey

Let us try another approach. Shakespeare talks of life as an "uncertain voyage". In another essay (Singer 1987) I have contemplated using this idea as a way of explaining an important difference between the wrongness of killing

396

Peter Singer

people at various stages of their lives (and we can include here, in addition, the putative wrongness of failing to bring a possible being into existence). We can see the lives of self-conscious beings as arduous and uncertain journeys, at different stages, in which various amounts of hope and desire, as well as time and effort have been invested in order to reach particular goals or destinations. Suppose that I am thinking of travelling to Nepal, where I plan to trek to Thyangboche Monastery, at the base of Mt Everest. I have always loved high mountains, and I know that I would enjoy being in the Himalayas for the first time. If during these early days of musing on the possibility of such a trip an insuperable obstacle arises - perhaps the Nepalese Government bans tourism on the grounds that it is an environmental hazard - I will be a little put out, naturally, but my disappointment will be nothing compared to what it would have been if I had already arranged to take the necessary time off work, perhaps bought a non-refundable plane ticket to Kathmandu, or even trekked a long distance towards my destination, before being barred from reaching my goal. Similarly, one can regard a decision not to bring an infant into the world as akin to preventing a journey from getting underway, but this is not in itself seriously wrong, for the voyager has made no plans and set no goals. Gradually, as goals are set, even if tentatively, and a lot is done in order to increase the probability of the goals being reached, the wrongness of bringing the journey to a premature end increases. Towards the end of life, when most things that might have been achieved have either been done, or are now unlikely to be accomplished, the loss of life may again be less of a tragedy than it would have been at an earlier stage of life. The great virtue of this journey model of a life is that it can explain why beings who can conceive of their own future existence and have embarked on their life journey are not replaceable, while at the same time it can account for why it is wrong to bring a miserable being into existence. To do so is to send a being out on a journey that is doomed to disappointment and frustration. The model also offers a natural explanation of why Parfit's two women both do wrong, and to an equal degree: they both quite unnecessarily send out voyagers with fewer prospects of making a successful journey than other voyagers whom they might have placed at the starting line. The women's children can be thought of as replaceable before the journey begins, but this does not require us to hold that there is an obligation to bring more children into existence, let alone to regard people as replaceable once life's journey has properly begun. But the journey model, too, has its problems. Too many questions remain. If lives are like journeys, we can see why it is more important to help those whose journeys are already underway to complete their journeys well, than to

Possible Preferences

397

arrange for more travellers to start. But if we easily can arrange for more travellers to start, and the prospects are good for successful completion of their journeys, is it prima facie desirable that more journeys should be made? And what of lives that are not like treks towards Mt Everest, but more like many strolls to the corner store? Does it matter less if they are interrupted, especially after one such stroll has been completed, and before the next has really been thought about? The journey model is helpful, but only up to a point. Like the moral ledger model, it cannot ultimately solve the problems we are considering.

5.

And in the

End...

So where, in the end, do I stand? Much as I would like to find a convincing defence of the Mind Readers position, I have to acknowledge that up to now I have failed to do so. As I have suggested, there are factors that justify us in giving preference to the desires of existing individuals over the creation of new ones. Nevertheless I must, for the moment, admit that the Rabbits have the best case.

References ( 1 9 7 6 ) . Michael Bayles (ed.): Ethics and Population, Cambridge, Mass., 1976. ( 1 9 9 8 ) . Christoph Fehige and Ulla Wessels: "Introduction to Possible Preferences", this volume. G O R O V I T Z E T AL. ( 1 9 7 6 ) . Samuel Gorovitz, AndrewL. Jameton, Ruth Macklin, John M. O'Connor, Eugene V. Perrin, Beverly Page St. Clair and Susan Sherwin (eds.): Moral Problems in Medicine, second ed., Englewood Cliffs, NJ, 1976. H A R E ( 1 9 8 1 ) . Richard M. Hare: Moral Thinking, Oxford 1981. H A R T ( 1 9 8 0 ) . H . L. A. Hart: "Death and Utility", The New York Review of Books, May 15, 1980. H E Y D ( 1 9 9 2 ) . David Heyd: Genethics: Moral Issues in the Creation of People, Berkeley 1992. H U R K A ( 1 9 8 3 ) . Thomas M. Hurka: "Value and Population Size", Ethics 93 (1983). P A R F I T (1976A). Derek Parfit: "Rights, Interests and Possible People", in Gorovitz et al. (1976). BAYLES

FEHIGE/WESSELS

PARFIT

PARFIT PETTIT

(1976B). Derek Parfit: "On Doing the Best for Our Children", in Bayles (1976). ( 1 9 8 4 ) . Derek Parfit: Reasons and Persons, Oxford 1984. E T AL. ( 1 9 8 7 ) . Philip Pettit, Richard Sylvan and J. Norman (eds.): Metaphysics and Morality, Oxford 1987.

398

Peter Singer

(1986). James Rachels: The End of Life, Oxford 1986. (1979). Peter Singer: Practical Ethics, Cambridge 1979; also referred to as PE. (1980). Peter Singer: "Right to Life", letter in The New York Review of Books, August 14, 1980. (1987). Peter Singer: "Life's Uncertain Voyage", in Pettit et al. (1987). (1993). Peter Singer: Practical Ethics, second ed., Cambridge 1993.

RACHELS SINGER SINGER

SINGER SINGER

RICHARD Μ .

HARE

Preferences of Possible People

1. I must ask forgiveness for starting with a dish that was not on the menu. I want to draw attention to a paper by Mane Hajdin, whom I did not know about until a student of mine found his paper (Hajdin 1990) in a data-base. This paper has altered my treatment of preferences so radically that it might affect my view on the preferences of possible people. I do not think that it actually does; but I need to discuss the question. Hajdin sets out to show that my theory of moral reasoning can be so modified as to make irrelevant two kinds of preferences that have been an embarrassment to me, namely those that Dworkin has called "external" preferences, and those that I have called "now-for-then" preferences. Another name for the first of these is "non-experiential preferences" - that is, preferences for things other than experiences of the preferrer. Another name for the second is "asynchronic preferences" - that is, preferences for what should happen at times other than the time when the preference is had. I shall for the most part use the second pair of expressions and their antonyms. The effect of the elimination of these two kinds of preferences from moral reasoning would be to make my own theory much more like the "happiness-utilitarianism" advocated by Richard Brandt (cf. Brandt 1989). Already in Hare (1981), sect. 5.6, I had said that there could be a form of preference-utilitarianism that was in effect the same as a form of happinessutilitarianism - namely a form that restricted the preferences taken into account to synchronic experiential preferences (i.e. now-for-now and thenfor-then preferences for experiences of the preferrer), and excluded from consideration asynchronic non-experiential preferences. This is because "happiness" can be defined for some purposes as a state in which we get the experiences we prefer to have and not those which we prefer not to have (there are of course other possible definitions, but this one gets fairly close, I think, to Brandt's thought). In Hare (1989) I developed this idea. More recently, in Hare (1995a), I came closer to saying that I accepted this modification of my theory. 2. The way Hajdin gets rid of non-experiential and asynchronic preferences is really very clever, and he puts it extremely clearly in this excellent paper.

400

Richard Μ. Hare

He points out that a move I make in my theory of moral reasoning uses a principle which Allan Gibbard has called the "Conditional Reflection Principle" (cf. Gibbard 1988, p. 58). Gibbard cites my own formulation of the principle: "I cannot know the extent and quality of others' sufferings and, in general, motivations and preferences without having equal motivations with regard to what should happen to me, were I in their places, with their motivations and preferences" (Hare 1981, sect. 5.5). This is the same principle as Wilfrid Hinsch calls "das Prinzip der bedingten situativen Identifikation" (Hinsch 1994). I wish that when I introduced the principle I had given it some less cumbrous and more perspicuous title, but for the moment I will use Gibbards expression. Hajdin points out that the operation of this principle in moral reasoning, as I describe it, ofitself excludes asynchronic non-experiential preferences from the reasoning. This is because what makes the reasoning work is our thought about what it would be like for our preferences to be frustrated or satisfied in our experience at a given time. To use my own example which he cites: what makes me think that I ought not to drive very close to the car in front is the thought that, if I were in the car in front, I should not like a car to run into me from behind, thus frustrating my preference for not being injured in a rear-end collision. Hajdin rightly says that "Merely imagining what it is like to be an occupant of that vehicle while it is travelling smoothly (even with all his preferences) may well leave my driving unaffected" (Hajdin 1990, p. 307). If Hajdin is right that only the thought of the frustration or satisfaction of preferences in my experience can affect my moral thinking (according to my theory), then clearly asynchronic and non-experiential preferences will be excluded from the thinking. This will save me the embarrassment caused to me by Cheops in Gibbard s paper who wanted to have a big funeral; I can think what it would be like for me to have this preference were I Cheops, but not what it would be like for me to have this preference frustrated (for I would be dead by the time the 'maimed rites' took place). I have in the past spoken of death-bed wishes and promises in a way that assumed that such asynchronic non-experiential preferences could have a place in moral reasoning; but, as Hajdin says, I have ways of doing without this move (Hajdin 1990, p. 309, fn. 3, citing Hare 1988c, pp. 235 f.). 3. The question obviously next arises of whether the dismissal of asynchronic non-experiential preferences from moral reasoning ought to make me revise my view on the preferences of merely possible people, about which I agreed to speak in this paper. I think that this would be too hasty a conclu-

Preferences of Possible People

401

sion. But first let me summarize the view as it has stood up to now. It is that in deciding whether or not to bring people into existence (or for that matter other animals - but people for short) we ought to have regard to the preferences which those people would have, if they were brought into existence. Among these would be the preference for existence over non-existence, which we may presume anybody to have who is happy. I have dealt at length in other places with some popular arguments against this view (for example, in Hare 1988a, 1988b and 1975; see also 1995b and 1995c). But for now let me set out the argument for the view, which I still think conclusive. It depends on the thesis of the universalizability of moral statements, which can be formulated for the purpose as follows: one cannot with logical consistency make inconsistent moral statements about two situations which one admits to be qualitatively identical. By "inconsistent moral statements" I mean "statements which would be inconsistent if made about the same situation". Now suppose we ask an existing person to make moral statements about the actions of those who brought him into existence by begetting, conceiving and not aborting him. It is of course possible that he is an unhappy person, and therefore wishes that they had not brought him into existence. Let us obviate this possibility by supposing that the person we are asking is happy. Since I have never wanted to maintain that we have a duty to bring unhappy people into existence, this will not affect the argument. Now there are three things that the person can say about what his parents did: (a) They did right (i.e. what they morally ought); (b) They did wrong (i.e. what they morally ought not); (c) What they did was neither wrong nor obligatory (i.e. it was neither the case that they did what they morally ought, nor the case that they did what they morally ought not). I shall call (c) the "amoralist" position with regard to these acts, and postpone discussion of it for the moment (I deal with it at length in Hare 1989). 4. If (c), the amoralist position, is for the sake of argument excluded, the person we are asking has the choice between (a) and (b); that is, he has either to say that they did what they ought, or that they did what they ought not. Faced with this choice, I think it obvious that the person will choose (a), and say that they did what they ought in begetting, conceiving and not aborting him. But because of universalizability, if he says this he has in consistency to say the same about any situation resembling the present case in its universal properties. Difference in time is not going to make any moral difference be-

402

Richard Μ. Hare

cause, as I have argued in Hare (1981), sect. 6.4, temporally referring expressions, which have an implicit singular reference to a temporal point of origin, cannot occur in moral principles, and therefore any principle which applies to this case must apply to qualitatively identical cases at other times, past or future. So he has to say that if an identical situation were to recur now, the parents concerned ought to beget, conceive and not abort the possible child in question. I think that, if the amoralist position is excluded, this argument is conclusive, and shows that a normal happy person has to acknowledge that in cases just like his, except that the begetting, etc. has not yet taken place, it would be wrong not to beget. It does not show, and I have never claimed that it does show, that all failures to beget and conceive and all abortions are wrong. There can be differences in some, perhaps in many, cases, which would make them right. But of course it has to be shown that these differences are morally relevant, which can in many cases be done. I am only claiming that the preferences of possible people can have weight in our moral reasoning. This amounts to denying a claim that Christoph Fehige makes in his excellent but difficult paper (Fehige 1998). The claim is, as he put it in an earlier version of that paper, that "it's not our job to raise the number of preferrers, it's our job to look after the preferrers who are, have been or will be, around". There are a great many other claims that I have not made, but to which some people wrongly think I am committed. In particular, as I have explained in the articles referred to, my position does not require a vast expansion of the population (how much it should be expanded depends on the choice of the best population policy, which depends in turn, in large part, on factors irrelevant to the present argument). I must also point out that my argument applies equally to the preferences of other future children that parents might have if they do not bring this one to term; they too have to be considered. This has the effect of greatly weakening the force of my view as an argument against abortion. All procreation is choice (even if the choice is not deliberate), and the choice to have a child later rather than now may do something for the interests of the later child at the cost of harming the interests of the child that the present foetus will turn into. And if the right population and family planning policy is being followed, the number of children that ought to be born is born, and the preferences of those children are all the preferences that we can satisfy within the limits of the right policy. 5. I can best use the space remaining to me by reverting to what I called the "amoralist" position (answer (c) in the list I gave). I said that the person we were asking might reply that what his parents did in begetting, conceiv-

Preferences of Possible People

403

ing and not aborting him was neither wrong nor obligatory. This answer is quite consistent with his preferring their bringing him into existence; such a simple preference is not a moral judgement, and is consistent with the moral judgement that they neither ought to, nor ought not to. He might, in short, reply that it was morally indifferent whether they brought him into existence or not, but at the same time express the preference for them having done so. It looks as if this would defeat the appeal to universalizability, because mere preferences are not universalizable. I argued in my reply to Brandt (referred to earlier) that this move can be countered by an adaptation of an argument I use against amoralists in general in Hare (1981), sects. 10.7 ff. The argument is complex, but I will try to summarize it. Suppose that we ask the person whom we are talking to, what he prefers should happen in all the other possible cases, past present and future, that resemble his own case in their universal properties. He could say that he expresses no opinion about them - no moral opinion, that is. He prefers that in his own case he should be brought into existence; but he has no preference as to whether the other similar people in similar situations should be brought into existence or not. But this leaves him exposed to a demand for the reasons why he differentiates between his own case and those of others. By hypothesis, the cases are qualitatively identical. So his reason could only be that his case affects him, but the other cases do not affect him. He has thus become, as regards this type of case, an amoralist egoist concerned only for his own interest, and not for those of others. If he treats this case in such a way, it is hard for him to avoid treating many other cases, whether cases of abortion or of any other harm to peoples interests, in the same way. He therefore is in danger of becoming an amoralist egoist for an unlimited number of kinds of case, and therefore of adopting a general stance of amoralist egoism, whose disadvantages I pointed out in Hare (1981). However, I am not yet satisfied with this way of dealing with the amoralist position (c), but shall have to leave it for the present. 6. Returning now to the Hajdin move of excluding non-experiential asynchronic preferences from moral reasoning, I ask: Does this move harm the argument I have outlined for giving consideration to the preferences of possible people? I think it does not. That this is so is suggested by the close similarity there is, to which I alluded earlier, between Brandt's happiness version of utilitarianism and the restricted preference version that the Hajdin move leaves us with. For it is obvious that possible people, if they become actual, may be happy - that is, they may have their synchronic experiential preferences satisfied and not frustrated. If this happens, we shall, according to the happiness

404

Richard Μ. Hare

version of utilitarianism, have done right ceteris paribus to bring them into existence. So on this version of utilitarianism there is really nothing to stop us saying that their preferences have to be considered in moral reasoning; if the Tightness of actions is determined by their effect on happiness, and the action of bringing these happy people into existence increases happiness (as it does) then it is pro tanto right. The same ought to hold for the restricted version of preference utilitarianism that we are considering, if the two kinds of utilitarianism are so similar. And indeed it does hold. The preferences that we satisfy by bringing into existence people who will have them satisfied and not frustrated are synchronic experiential preferences. The happy people we bring into existence will be happy (on this understanding of "happy") because the experiences they have satisfy, and do not frustrate, the preferences which they have at that time. To use our other terminology, their now-for-now and then-for-then preferences will be to that extent satisfied and not frustrated. External preferences and now-for-then preferences do not need to be mentioned in the reasoning. If we can maximally satisfy these experiential synchronic preferences we shall have maximally satisfied all the preferences that could come into the reasoning, and shall be exemplary utilitarians. I conclude that Hajdins move could help to relieve my theory of some embarrassments, and so dispose of some unfinished business, as I called it in Hare (1981), sect. 5.6, without making me say anything different from what I have said in the past about the preferences of possible people. I remain uneasy, however, because of a hankering I still have to include within the scope of moral reasoning some asynchronic non-experiential preferences, such as the desire that my children should be looked after when I am dead, or that my wife should not commit suttee. So there is still some unfinished business.

References BRANDT ( 1 9 8 9 ) .

tice 15

Richard B. Brandt: "Fairness to Happiness", Social Theory and Prac-

(1989).

Christoph Fehige and Georg Meggle (eds.): Zum moralischen Denken, Frankfurt/Main 1995. FEHIGE ( 1 9 9 8 ) . Christoph Fehige: "A Pareto Principle for Possible People", this volume. GIBBARD ( 1 9 8 8 ) . Allan Gibbard: "Hare's Analysis of'Ought' and Its Implications", in Seanor/Fotion ( 1 9 8 8 ) . HAJDIN ( 1 9 9 0 ) . Mane Hajdin: "External and Now-for-Then Preferences in Hare's Theory", Dialogue 2 9 ( 1 9 9 0 ) . FEHIGE/MEGGLE ( 1 9 9 5 ) .

Preferences o f Possible People

405

(1975). Richard M. Hare: "Abortion and the Golden Rule", in Hare (1993); article first publ. in 1975. H A R E (1981). Richard M. Hare: Moral Thinking: Its Levels, Method and Point, Oxford 1981. H A R E (1988A). Richard M. Hare: "Possible People", in Hare (1993); article first publ. in 1988. H A R E (1988B). Richard M. Hare: "When does Potentiality Count? A Comment of Lockwood", in Hare (1993); article first publ. in 1988. H A R E (1988C). Richard M. Hare: "Comments on Griffin", in Seanor/Fotion (1988). H A R E (1989). Richard M. Hare: "Brandton Fairness to Happiness", Social Theory and Practice 15 (1989). H A R E (1993). Richard M. Hare: Essays on Bioethics, Oxford 1993. H A R E (1995A). Richard M. Hare: "Welchen Nutzen maximiert der Utilitarist? Replik auf Kusser", in Fehige/Meggle (1995). H A R E (1995B). Richard M. Hare: "Abtreibung, Empfängnisverhütung und Zeugungspflicht: Replik auf Lenzen", in Fehige/Meggle (1995). H A R E (1995C). Richard M. Hare: "Zum moralischen Status potentieller Personen: Replik auf Schöne-Seifert", in Fehige/Meggle (1995). H I N S C H (1995). Wilfried Hinsch: "Präferenzen im moralischen Denken", in Fehige/ Meggle (1995). SEANOR/FOTION (1988). Douglas Seanor and N. Fotion (eds.): Hare and Critics: Essays on Moral Thinking, Oxford 1988. HARE

WOLFGANG

LENZEN

Who Counts? 1. 2. 3. 4. 5. 6.

Introduction Some Moral Theory The Value of Life and the Wrongness of Killing Abortion vs. Contraception Critique of the Rabbits' and the Mind Readers' Positions Objections to the Midwives' Position References

1.

Introduction

According to Christoph Fehige's and Ulla Wessels's classification (from the introduction to this symposium) I am a Midwife, i.e. with respect to the issues of abortion and contraception I defend the common sense view: (MIDW 1)

There is something seriously wrong about abortion, while nothing at all is wrong about contraception or, somewhat more exactly, about all those forms of contraception which aim at preventing an ovum from becoming fertilized.

This position might be formulated in many different ways. One way of expressing it in the terminology of rights would be to say: (MIDW 2)

No ovum has a right to become fertilized, but every embryo in whatever stage of its development — has the prima facie right not to be aborted, i.e. not to be killed.

Another way of expressing this, in terms of duties, would amount to stating: (MIDW 3)

Nobody has the duty to procreate, but every pregnant woman has a prima facie duty to let the embryo develop and live its life.

Still another way of putting this, in terms of possible vs. actual persons, would be to claim: (MIDW 4)

Although human embryos lack some essential characteristics of personhood and hence are only possible persons or future persons, they have the same moral status as actually existing

Who Counts?

407

persons. In contrast, human gametes which never happen to develop into embryos remain merely possible persons and thus are no proper objects of moral considerations. There are two strategies for arguing in favour of this position. The first, constructive strategy consists in showing how the Midwives' evaluation of the morality of abortion and contraception can be derived from a sound ethical theory in conjunction with a plausible account of what is in general wrong about killing. The second, 'destructive' strategy consists in refuting the alternative positions of what Fehige and Wessels call the Rabbits and the Mind Readers by showing that these views either are based on false theoretical assumptions or give rise to intuitively unacceptable conclusions. Before addressing this twofold task, however, let me explain what kind of restriction I have in mind when formulating the Midwives' position by means of a prima facie clause. To say that an embryo has the prima facie right not to be aborted, or to say that the mother in spe has the prima facie duty to let the embryo develop does not mean that abortion must be morally wrong in every case and under all circumstances. As I have argued, in Lenzen (1991), abortion may be justifiable: - in almost every instance of so-called medical indication; - in most instances of so-called eugenic indication; — in many instances of so-called criminal indication; and perhaps also — in some instances of so-called social indication. In other words, I more or less subscribe to the present abortion-legislation in Germany - the so-called Indikationenlösung - while I reject the so-called Fristenregelung which wants to permit any abortion up to the 13th weak of pregnancy. The reasons for this rejection will be given below when I discuss the shortcomings of the Mind Readers' position. First, however, let me sketch the ethical theory upon which my further investigations will be based. This is a blend and refinement of three well-known principles of moral philosophy, namely the Golden Rule, the maxim "Neminem laedere", and the principle of utilitarianism.

2.

Some Moral Theory

2.1. A Minimal Principle of Ethics. To begin with, there are two quite different versions of the Golden Rule. While the positive version requires us to

408

Wolfgang Lenzen

do to others what we want them to do to us, the negative version only forbids us to do to others what we do not want to be done to ourselves. The idea behind the latter maxim is perhaps better expressed by saying that we should never do to others what they do not want to be done to them. Furthermore, in order to avoid problems resulting from unreasonable or capricious wishes, one may add the further qualification that we should never do to others what they reasonably want not to be done to them. We thus obtain the following formulation of the Negative Golden Rule: (GOLDneg)

Action A of an agent X is morally wrong if and only if (for short: iff) there is at least one other individual Y such that Y reasonably wants A not to be done to him.

The notion of someone reasonably wanting something shall be interpreted in such a way that individual Y has a reason for wanting A not to be done ifif A conflicts with his legitimate interests. Although the latter concept of'legitimacy' is not without problems 1 , it shall be presupposed as a primitive concept here. In its traditional version, the related maxim "Neminem laedere" simply says "Do no harm". Accordingly action A of an agent X is to be considered as morally wrong i f f A does harm to at least one other individual Y. The concept of harm is meant to cover not only physical but also mental or psychic injuries. Here it will be understood in a very broad sense so that action A does harm to an individual Y whenever A (or the immediate or future consequences of A) violate the legitimate interests or preferences of Y. This interpretation of "Neminem laedere" basically expresses the same ethical principle as GOLD n e g . In what follows, however, it will be taken only as a necessary but not as a sufficient condition for an action to be morally wrong 2 and it will be referred to as Minimal Principle ofEthics·. (ETH m i n ) Action A of a person X is morally wrong only if A violates the legitimate interests of at least one other individual Y. This weak principle is strong enough to solve some problems in applied ethics. In particular it allows a clear evaluation of the morality of (at least certain 1

2

Within the framework of utilitarian ethics, 'legitimate' interests might be distinguished from 'illegitimate' ones by means of weighing factors such as the 'Verdienstparameter' discussed in chapter II of part 1 of Trapp (1988). In particular, if an agent X faces a moral dilemma such that whatever he decides to do he cannot prevent doing harm to some individual Y, many people would be reluctant to say that X necessarily acts morally wrong.

Who Counts?

409

instances of) suicide and also of (at least certain instances of) euthanasia. For reasons of space, however, these issues cannot be dealt with here.3 On the other hand, E T H m i n is evidently too weak to judge those actions that favour the legitimate interests of some individual Y while violating the legitimate interests of some other individual Z. Here we need the conceptual means to compare the harm or the good done to the one with the harm or the good done to the other individual. This is where utilitarianism enters the scene. 2.2. Utilitarianism. Utilitarian ethics is based on the assumption that for every relevant action A and for every individual Yi (from a certain set {Y\,..., Yn}) we are given a numerical value u(A, Yj) - the utility of A for l·/ which somehow corresponds to the degree to which the individual Yi wants A to happen (or wants A not to happen). With the help of these values the total utility of an action A is defined as the sum of the utilities of A for all individuals involved, i.e. U (A) = Σ» U(Aj).

This principle is much too demanding, however. According to UTIL super every action A which is not optimal, i.e. for which there exists an Ä such that U(A') > U(A), would have to be classified as morally wrong. But most of our everyday activities, such as drinking a cup of coffee, smoking a cigarette, or walking along the river, fail to be optimal; it is easy to construct alternative actions with a higher total utility, e.g., drinking no coffee, smoking no 3

Cf., however, the brief remarks on suicide in section 5.2 below. More substantial discussions of these issues may be found in Lenzen (1991) and (1997).

410

Wolfgang Lenzen

cigarettes and donating the money saved to charity. Nevertheless we would be reluctant to consider these activities as immoral. In order to develop a more realistic utilitarian ethics, one has to take a closer look at the correspondence between the utility of an action A for an individual Y, u(A, Y), and the strength with which Y wants A to be done (or wants A not to be done). Consider two alternative actions A and Β such that, e.g., u(A, Y) — +15 while u(B, Y) = - 3 5 . Clearly, since A has a much higher value for Y than Β, Y prefers A to B, i.e. Y wants A to happen rather than B. Let the latter preference-relation be expressed by "Α >γ Β". Then the following correspondence between preferences and utilities is taken to hold for arbitrary actions A and Β: (REPR)

u(A, Y) > u(B, Y) i f f A >Y B.

It would seem quite natural to suppose that action B, which in our example has a negative value of —35, therefore is negative for Y' in the sense that Β violates the interests of Y. Equally one would want to assume that A, which has the positive value of + 15, must therefore favour F's interests. However, the standard axioms of the theory of utility do not guarantee that this is always the case. To be a little more precise, the condition of an action being 'positive' for Y can be formalized, e.g., by introducing a 'tautological action T, to be interpreted intuitively as Ά is either done or not done'. Then action A is 'positive' for Y iff Y wants A to happen and thus prefers A to T; analogously Β is negative' for Y iff Y wants Β not to happen and therefore prefers Τ to B.^ The crucial condition mentioned above then amounts to requiring that — for every individual Y — the 'neutral' action Τ has a value u( Τ, Y) = 0. Let us call a utility function u which satisfies this requirement normal. Although the standard theory does not entail that every utility-function must be normal, it can be proven"* that for every function u which represents the preferences of an individual Y in the sense of satisfying REPR, there exists a normal function u* which also represents these preferences. Therefore in what follows we may assume that every utility-function considered is in fact normal. Since for every normal utility function u*, for every individual Y and for every action A, a positive value u (A, Y) expresses the degree to which A fa4

5

Alternatively one may postulate that for every action A there exists a contrary action' No η-/I (formally: ->A) and defined to be positive for Y iff A >y ->A. Cf. Lenzen (1980). The existence of normalized utility functions is not only guaranteed by abstract theorems of the mathematical theory of utility - cf. Lenzen (1983) but is also warranted by the following intuitive interpretation. Take uT(A, Y) to represent the (positive or negative) amount of money that Y is willing to pay for making A happen (or not happen).

W h o Counts?

411

vours the interests of Υ, i.e. the degree to which Y wants A to be done 6 , the normal total utility of an action A, U*(A), is positive iff the good done by A to some of the individuals concerned outweighs the harm done to the others.7 Therefore we can state the following utilitarian principle which avoids the problem of supererogation and, at the same time, generalizes the idea of G O L D n e g and thus supplements E T H m i n : (UTIL)

Action A is morally wrong iff its normal total utility, U* (A) — Σκη u*{A, Yj), is negative.

3.

The Value of Life and the Wrongness of Killing

The life of an individual X - whether human or not - has a value that depends on the sum of values of everything that X has ever lived or experienced in his life.8 Since I am here not concerned with the problem of vegetarianism, I will concentrate on human lives only and not try to speculate about the average value of the life of a fish, a chicken, or a pig. Let us begin by considering the completed life of some person X. If X dies, say, at the age of 84, her past life may be split up either into temporal units - 1001 months; 30 535 days; 732 850 hours; etc.— or into a sequence of states and events A\,... ,A„. This sequence might begin with some prenatal events or with the first sucking at the mothers breast. It will contain some typical events as hurting a knee as a baby, celebrating Christmas, writing maths homeworks, lying in the sun, reading newspapers, 6

7

8

Analogously, a negative value characterizes the degree to which A conflicts with the interests of Υ, i.e. the degree to which Y wants A not to be done. Therefore u*{A, Y) = 0 iff Κ does not care whether A happens or not. To be somewhat more precise, one here needs the further premise that the utility functions for the different individuals Yi are 'intersubjectively comparable'. This could be guaranteed by a slight modification of the procedure mentioned in footnote 5. Let u" (Α, Υϊ) represent the amount of money that Yi would be willing to pay for making A happen (or nor happen), if Yi has the same fixed amount of money at his disposal as every other Yj. Nagel (1970, p. 2) has argued that the value of life exceeds the value of all ones experiences: "life is worth living even when the bad elements of experience are plentiful, and the good ones too meager to outweigh the bad ones on their own. The additional positive weight is supplied by experience itself, rather than by any of its contents" (my emphasis). It is hard to see, however, how "experience itself" without any content might be valuable. W h y should life without (positive) content be better than life without consciousness or better than the 'life' of someone in deep coma? More recently, Nagel repeated the somewhat paradoxical claim that being alive would be a value in itself and that therefore death should be considered as a loss even if life is no longer worth living; cf. the final chapter of Nagel (1986). There, however, no real argument is given; the reader is just referred to Wollheim (1984).

412

Wolfgang Lenzen

10

20

30

40

50

60

70

80

Years

Fig. 1

meeting lovers and friends, etc. Typically it will end with illness, pain, and the fear of death. Each of these states or events has a value for X which can in principle be measured by a normal utility function u*. Since most of the things we experience in life are positive for us, the value of X's total life, V(L,X) = will normally be much greater than 0. Moreover, at almost every instance t of her life, not only has the past life of X normally had a positive value V(L then clearly these could override the ceteris paribus judgement that, since ^1001 doesn't want life, death would be all right. This leaves us with the possible future person stages ^1002» ^1003> etc. from the survival scenario. The objection bids us to take into account their happiness as well. To make sure that this happens, let us require that £1001 empathize with them, and that tool's overall preference, in order to have moral authority, must give the same weight to £1001 s own interests as to those of every possible later bi. Thus we demand empathy - a prudential, intrapersonal type thereof. 12 But what precisely will empathy come up with? It will certainly require £1001 to pity every person stage which is unhappy; the unhappier a person stage, the less good its existence.13 But what about the other way round? What about the stages which would be happy? Does the Midwife suggest that we (or person stage £1001) should pity a person stage for not existing if its existence would be happy? Note the deep, deep contradiction that is lying in wait for the Midwife here. Midwives insist on arguing just from the interests of objects that exist (have existed, will exist); it is a central tenet of midwifism - since without it, Midwives would be Rabbits - that the fact that an object would be happy if it existed does not make its existence, compared to its non-existence, a good thing. The central tenet would be violated if we now took, as the objection we're dealing with suggests we do, the interests ofthe possible future stages as a moral reason for their existence\ This is an argument a Midwife cannot use. (To employ one of the Midwifes favourite expressions, person stage £1002 isn't deprived of anything if it never exists.) Note the subtle difference: a Midwife can say (though I don't see why, and though I think the suicide argument speaks against it) that, if the future person stages are happy, it is in the persons interest to have them. But a Midwife can not say, and hence cannot support the previous claim by saying, that the interests of the possiblefuture person stages generate a reason for their existence. And this answers objection (2). Objection (3) to the suicide argument says that it would justify us in bringing up a fetus even if we knew that, for some unavoidable reason or other, his life would be hell. The ex-ante absence of a preference against damnation would, 12

13

A demand that is not unusual - see, e.g., Hare's requirement of prudence, referred to in note 56 below. See also the reply to objection (3).

436

Ulla Wessels

by the reasoning of the suicide argument, show that damnation is not worse than non-existence; therefore, the suicide argument would permit the creation of a life of eternal suffering. This is a reductio ad absurdum. Reply: It would be one indeed, but the suicide argument does not entail the scandalous verdict. In order to count, an individual s preference (or lack of preference) must be rational; it must be compatible with the verdict rationality would come up with if faced with the options at issue. What does this mean for damnation? Unhappiness is something that rationality advises us to avoid. Thus, if our morality is to be guided by rational preferences, we must ascribe to people a preference to avoid suffering. We must ascribe it even in cases in which it has no or little psychological reality - as is the case with a fetus, which has no 'real' preference whatsoever. In that sense, i.e. in the sense that includes the ascribed preferences, the fetus does not lack preferences against eternal suffering, and thus the suicide argument does not apply. 14 But, it might be asked at this point, should we not, just the way we have ascribed preferences against an unhappy future, also ascribe preferences for a

happy (as opposed to no) future?

The answer is twofold. The Midwife just cannot say that the interests of

the individual's possible future person stages should prompt us to do so - see the reply to objection (2). So we're left with the question whether it is simply in the individual's interest to have, even where a corresponding preference is lacking, a happy future as opposed to no future. That no is the right answer is precisely what the suicide argument itself is out to suggest. 15 Objection (4) to the suicide argument says that the step from the moral evaluation of the suicide to that of the fetus went wrong. For we should not, in general, treat an entity that cannot grasp a proposition p as if it had no preference for p. Reply: Well, not ifpure rationality has a different verdict about ρ — see the reply to objection (3). But otherwise, the procedure is correct and alternatives would be hard to justify. If an individual cannot grasp ρ and pure rationality does not recommend p, then it is clearly unwarranted to treat her as if she wanted ρ . ι β 14 15

16

For a similar discussion, cf. Fehige (1998), sect. 1, subsect. "Orexigenesis vs. prophylaxis II". As to rational preferences and happiness (or welfare), cf. also the discussion in sect. 3.3 of this paper. The point becomes blatant if we think of compensation. Suppose there's an individual who has no 'real' preference for ρ (where p is not recommended by pure rationality) and we treat her as if she wanted p; something she does indeed want, although less strongly, is q. In a situation where she can have just one of the two goods, we would be required to give her p.

Procreation

437

Objection (5) to the suicide argument says that, if we accept the suicide argument, and thus the claim that a happy future as such does not speak against death, then, apart from other people's sorrow and the like, very little would speak against somebody's death at all, and that is implausible. Reply: Death would be bad for a person because of her past and present preferences - preferences for her survival or preferences that can only be satisfied if she survives. In the rare cases where such preferences are lacking, I don't see why death should be bad for the person. And since most people have many strong preferences of this type — how strong becomes obvious when we see how much pain and misery are still preferred to death! - , death will usually remain a terrible thing, even if we deny that the future as such, without an orectic anchor in the past or the present, speaks against death. We do not need AntiDeath in order to say that killing the man on the Clapham omnibus would be very bad. 17 Thus, the five objections to the suicide argument do not work, and the argument remains intact. It is possible that somebody rationally wants to commit suicide even though he is fully and vividly aware of the fact that he would be happy if he went on living. He can simply prefer non-existence to a happy existence. In such a case, we ought not to prevent him from committing suicide. Similarly, we have no moral obligation to keep somebody alive who has no preference for his happy survival. A fetus who has no preferences whatsoever resembles, in the relevant respect, the latter case; therefore, we have no moral obligation to keep it alive. Two more remarks on the scope of the argument. Firstly, the argument works no matter which goods a proponent of Anti-Death has in mind when he speaks of a "happy future". It doesn't matter whether what he's talking about are future experiences, future satisfied preferences, or future objective' goods of whatever kind. For any such goods, it is always possible to rationally prefer the end to a sequel full of them. (The same thing applies to our discussion of principles (*) and (**) in the reply to objection (1). Roughly speaking, different notions of happiness won't make a difference.) Secondly, there are ways of weakening Anti-Death that won't help. To say that a happy future makes death bad if this or that condition is fulfilled, is, as can easily be seen, no way out 1 8 (unless, of course, the "if" clause requires ex-

17 18

In other words, we would be required to satisfy a preference she doesn't have, and doesn't by virtue of rationality have to have, at the cost of one she does have. This would be a weird notion of benefitting the individual. Cf. Fehige (1998), sect. 1, subsect. "Orexigenesis vs. Prophylaxis I". Jeff McMahan, e.g., thinks that, preferences aside, an individual's interest in future goods de-

438

Ulla Wessels

ante preferences for survival, in which case the character of the theory is altered radically). Conditions or not, one can, without lack of empathy for one's future person stages, lack any preference for the future in question; therefore, the suicide argument will work, even against the weakened versions. By itself (especially if it hasn't been desired beforehand), a happy future - that is what the argument shows - never makes death bad. 2.3.

Problems with the Conjunction of Anti-Death and Not-Pro-New-Life: The Suicide Revisited and the Wretched Child

My second and third objections, entitled "The Suicide Revisited" and "The Wretched Child", concern the conjunction of Anti-Death and Not-Pro-NewLife - in other words, the alleged moral asymmetry of the case in which the person in question exists anyway and the case in which she does not. 19 The Suicide Revisited Objection no. two is again to some degree an intuitive point, just like many of Lenzen's.20 Consider once more the case of the suicide b from the previous section (2.2): b has no wishes, not even implicit ones, for survival (not even for a happy future), nor for anything that her survival could help to achieve.21 Furthermore, let us assume that there are no friends, relatives or any other people whose interests would be affected by b's death. (In the previous section, I advocated a particular verdict about b\ death or future. Now my objective is different. I want to point out that, whatever the verdict is, it should be the same in her as in a certain other case.) Now look at a segment of happy future: a certain number of days, with a certain number of pleasant experiences, vastly outnumbering the unpleasant ones, and a certain number of satisfied preferences, vastly outnumbering the frustrated ones. Suppose that we can either give that future to b, or create a

19

20 21

pends on the psychological relatedness between her present and future person stages (McMahan 1998, sect. 3). The relevant modification (note the italics to come) of our suicide argument would proceed from a case in which, although aware of the fact that the possible future person stages that are psychologically related with ^ιοοι would enjoy a happy life, person stage ^1001 would have no preference for a happy survival. Technically, of course, arguments against one of the conjuncts - such as the argument that has been presented in section 2.2 above - are also arguments against the conjunction. What I'm now concerned with, however, are the problems which arise from marrying Anti-Death to Not-Pro-New-Life; I will show that the marriage is a mismatch - regardless of any merits or shortcomings each partner might have in its own right. See note 8 above. Note that the suicide differs from Lenzen's comatose (1998, sect. 5.2) who, I take it, was more normal in that, before the accident, he had a preference for survival.

Procreation

439

new individual who will live through precisely that future; suppose also, and again improbably, that there are no morally relevant side-effects either way (for example, no benefits or costs, emotional or other, involved in killing b and in creating a new being) - then what should we do? Give the future to the existing person or create a new one and give it to her? It doesn't matter. Both scenarios contain the same number of happy days, and both scenarios are equally good. 22 Lenzen thinks it does matter - see his claims Anti-Death and Not-Pro-New-Life. He says that we are doing a good thing if we give the happy future to the suicide (who never wanted it, not even implicitly), but not a good thing if we create a person and give it to her. I would of course agree that under normal circumstances providing an existing person with the happy future would be better than creating a person and providing her with it. Normal people want a happy future and that would create a moral asymmetry between the existing person and the extra person (who, by virtue of her non-existence, has no such preference). Wishes of this type get frustrated if we give the future to the extra person, but not if we give it to the existing person; thus we'd better give it to the latter. The case at issue, however, was not normal, and preferences for survival were not there. Their absence, I suggest, is the absence of any reason to favour the existing person. It is precisely by having no preference for a happy future that the previous self has left us with no moral reason to tie the happy future to him rather than to a new bearer. Therefore, both scenarios are equally good. If you share this intuition, you cannot support Anti-Death and Not-Pro-New-Life. One of them will have to go. The Wretched Child Here is my third objection, and the second one against the conjunction of Anti-Death and Not-Pro-New-Life. Clearly, we want morality to forbid sadistic creations - in other words, the creation of persons who will be miserable throughout their life. Lenzen agrees that such creations are wrong. 23 But look at the strange group of claims we get if we add this reasonable requirement, which Lenzen himself endorses, to his Anti-Death and Not-Pro22

The issue I'm raising here is that of "replaceability"; cf., e.g., Parfit (1984), § 122, Singer (1993), pp. 1 6 1 - 5 , and, in this volume, Singer, sects. 1 f., and McMahan, sect. 3.3. It is normally discussed on the basis of more realistic cases: should we kill a disabled fetus if the mother will have another try and is then likely to have a healthy child? Or if a 14-year-old girl is pregnant and can have a less happy child now or an abortion and a happier child later what ought she to do?

23

Cf., e.g., Lenzen (1998), sect. 6.2.

440

Ulla Wessels

New-Life:24 (A) (B)

On the one hand, a bad future is a reason not to bring a person into existence. On the other hand, a good future is not a reason to bring a person into existence.

Whence this asymmetry? The question becomes even more puzzling if we recall: (C)

A good future is a reason not to end a person's existence.

How do these claims go together? If a bad future makes existence a bad thing, then why does a good future - and one that is good enough to make continued existence a good thing! - not make existence a good thing? I find this hard to answer. Not so hard, of course, if you set out to save the three intuitions that conception is usually not obligatory, and the existence of a wretched child is worth avoiding, and abortion is usually wrong. Indeed, these three almost force you to embrace claims (A), (B), and (C). This, however- as an unbiased look at (A), (B), and (C) suggests - , is rather an argument against the former trio. I asked how claims (A), (B), and (C) go together. Note that the answer could not be: "While the creation of an unhappy person leads to the existence of somebody who suffers, and thus harms the person in question, the creation of a happy child does not benefit that child — because the child didn't exist at, and would never exist without the, creation." This would be morals based on a linguistic muddle - a real howler. If creating a happy person couldn't count as benefitting her because at the point of creation there was no "her" to benefit, then creating an unhappy person couldn't count as harming her because at the time of creation there was no "her" to harm. Thus we have to put a ban either on both ways of speaking ("being created can harm a person" and "being created can benefit a person"), or on neither of the two. Tertium would be cheating. If we put a ban on both of them, (A) will have to go, and the creation of misery will be in order.25 Since this route is deeply unattractive, and is not the one Lenzen chooses to take, I will not pursue it here. 24

25

The requirement appears as (A) in the following list, paraphrases of Anti-Death and NotPro-New-Life as (C) and (B) respectively. This becomes particularly clear when we invent situations in which, after the act of procreation, there is no way of putting an end to the misery, i.e. no way of killing the person in question. The causal road to relief is blocked. In such a situation, the two-step construction Lenzen once had in mind (see sect. 2.4 of his 1 9 9 0 and note 1 6 in his 1 9 9 8 ) could not avoid the cruel result that producing misery is all right.

Procreation

441

Lenzen goes in the other direction and is prepared to permit both ways of speaking: "[L]et it here be taken for granted", he writes, "that procreation would [... ] be good for a happy but bad for a wretched child." 26 But if something is good for somebody, then clearly it is ceteris paribus good. Thus Lenzen is committed to the claim that it is ceteris paribus good to bring a happy person into existence. In that case, (B) will have to go. We haven't been shown - and I fail to see how one could show - why we should endorse the puzzlingly asymmetric trio of (A), (B), and (C). 2.4.

Problems with the Inference from Anti-Death and Not-Pro-New-Life to Anti-Abortion and Not-Pro-Fertilization: Identity and Its Relatives

We come to the fourth objection against Lenzen's position. Having questioned in the previous two sections (2.2 and 2.3) Anti-Death itself and Anti-Death in combination with Not-Pro-New-Life, we now ask whether the conjunction of Anti-Death and Not-Pro-New-Life, tenable or not, entails that of Anti-Abortion and Not-Pro-Fertilization. Wolfgang Lenzen thinks that it does. Whether he's right will depend on questions of identity. On the one hand,

it would have to be shown that a fertilized egg is the individual whose future we are talking about; only then can, in Anti-Death, the clause "given an individual who exists" get to work and generate Anti-Abortion, i.e. the obligation to lead the fertilized egg into a happy life and, a fortiori, not to abort it. On the other hand, it would have to be shown of other items that they are not the individual whose future we are talking about: items such as the unfertilized egg, the sperm, and the unfused pair of an unfertilized egg and a sperm. 27 If one of these objects were the individual, then - contrary to Lenzen's theory — that object, too, would fall under the jurisdiction of Anti-Death and would have to be led into a happy future (just like the fertilized egg, see "on the one hand"). Contrary to Lenzen's theory, conception, too, would then be obligatory, as a step on the object's journey to happiness, which, as has just been reported, would itself be obligatory. Only if none of these objects is the individual, 26 27

Lenzen (1998), sect. 6.2. Lenzen himself is fully aware of the fact that his theory has to make a moral difference not only between a fertilized and an unfertilized egg, but also between the pair of sperm and egg before and after they fuse. Cf. Lenzen (1995), p. 231.

442

Ulla Wessels

is Not-Pro-New-Life rather than Anti-Death applicable and can generate Not-Pro-Fertilization, i.e. the permission to refrain from conception. Thus, Anti-Death and Not-Pro-New-Life pass the buck to our theory of identity. The moral asymmetry between non-conception and abortion boils down to the identity-theoretical asymmetry between the unfiised and the fused pair of sperm and egg. The former is not the individual whose happy future we're talking about whereas the latter is. Identity, however, is notorious - a philosophical maelstrom if ever there was one. Criteria for the identity of objects in general and of people in particular are controversial and abound with puzzles. If this is so, and if for Lenzen so much hinges on questions of identity, then why doesn't he discuss them? Where is the theory of identity (and where is the defense of it) that bears out his central identity-theoretical claim?28 Identity is supposed to do a lot of work, but receives little attention. This disproportion is all the more baffling since serious identity-theoretical proposals to be found in the literature clearly do not bear out Lenzen's central claim. -

Some theories deny that identity obeys bivalence. It need not be the case, they say, that either a is identical to b or a is not identical to b. As to the beginning of life, it may then well be that the unfertilized egg is not identical to the subsequent child, and that the mass of, say, 128 adhering cells is, but that the intermediate stages neither are nor are not. 29 Some theories say that identity is a matter of degrees, not of all or nothing. 30 Faced with gradual developments like that of a human life, such a theory would argue that identity emerges, and that the fertilized egg is at best 'slightly more identical' to the child than the pair of sperm and unfertilized egg. Some theories base identity on similarity or shared properties.31 The pair of egg and sperm just before the fusion and the entity just after the fusion

-

-

28

29 30

31

I.e. the claim that neither the unfertilized egg, the sperm nor the unfused pair of unfertilized egg and sperm are the individual whose happy future we're talking about, whereas the fertilized egg is. For this type of view, see, for instance, van Inwagen (1990), chs. 14 and 17-9. See, for instance, Hare (1981), ch. 5.4, and Lyon (1980), pp. 1 7 8 - 8 0 . Cf. also Parfit (1971), (1984), part III, and (1995), esp. part II; strictly speaking, Parfit himself is not after identity, but after 'what matters in survival' - this, however, is precisely Lenzen's topic. T h e Buddhist view seems to be similar, cf. chs. 2 f., esp. pp. 3 3 - 4 0 , of LaFleur (1992). As does Wittgenstein, see (1958), pp. 6 1 - 3 ; cf. also Loux (1978), pp. 1 2 4 - 6 , Borowski (1976), and Lyon (1980).

Procreation

443

share, intuitively speaking, such an 'amount' of properties that they would have to count as the same object by this type of standard. They are close neighbours in time and share most of their physical make-up, of their history, of their causal potential, 32 and of their position in space. — Many theories hold our brain or our mind to be a necessary component o f our identity.33 They would not support Lenzen's position, since the early fetus does not have a brain or a mind, and thus could not be the same individual as, e.g., the seven-year-old child. All these are candidates. The answer to the decisive question — to the question "Who is who?" — is far from obvious. We have to diagnose, at the very least, a monumental lack of arguments at the heart of Lenzen's theory.

Three Moves That Won't Help Although Lenzen himself does not seem to build on them, I would like to mention three moves that could not help support his views. (1) It would be of no use to say that the controversy among identity theories has nothing to do with morals and that, if we are looking for a morally relevant notion of identity (which we are), the choice of fertilization as the crucial point is just obvious. It isn't. This is why even people who are quite happy with Lenzen's general position (Anti-Death and Not-Pro-New-Life) come up with different cut-off points (points, that is, before which killing a fetus or even an infant would, and after which it would not, be all right). Even they cannot agree what identity, in the morally relevant sense, is; some of them think, in contrast to Lenzen, that it requires a brain, or even consciousness.34 To declare that the choice of an identity theory is a moral choice is to say that it requires a moral argument, not that it requires none. (2) What about the fact that there's space between the egg and the sperm before they fuse - doesn't that settle the issue? No. Imagine we could cut adults 32

W e might be tempted to say: "Fusion makes an enormous difference in causal potential since it is necessary on the way to a happy person." If this argument were valid, identity would be renewed whenever a necessary condition were met. For instance, the acquisition o f consciousness, being necessary on the way to a happy person, would also create a new individual. This would ruin Lenzen's theory since the individual who experiences the happy future would then not be the same as the fertilized egg.

33

David Lewis, Anthony Quinton, and Sydney Shoemaker are among those who have proposed psychological criteria o f personal identity; Thomas Nagel and John L. Mackie think that we are essentially our brains. For references, see Johnston ( 1 9 8 7 ) , notes 3 and 18. Cf. also in this volume McMahan, sect. 2.2: "[E]ach o f us began to exist when the brain o f his or her body developed the capacity to support consciousness and mental activity."

34

See, e.g., Leist ( 1 9 9 0 , ch. V), Sumner ( 1 9 8 1 , ch. IV), and M c M a h a n ( 1 9 9 8 ) .

444

Ulla Wessels

in half and put them back together again. Then while Paul is disassembled, we would still say "Look, there's Paul cut in half'. The space between the halves does not impugn his identity. Neither, of course, does the absence of space suffice to establish identity. People touch each other without becoming one; and bacteria live in people, fetuses in mothers, etc., and are still not one with their hosts. (3) Lenzen's claim about identity could not be supported by adducing genetic make-up as a criterion. As a necessary condition for identity, genetic make-up would not show that identity starts with the fusion; at best, that it starts no earlier than with the fusion (but perhaps much later) - a result that is too weak for Lenzen's purposes. Could genetic make-up be a sufficient condition for identity? No. Firstly, the fused pair of egg and sperm can still split up and would thus be identical with two or more children - a reductio ad absurdum?** Secondly, we can think of brainwashings that would change our identity without changing our genetic make-up. And if Τ had been raised with 'my' genetic make-up, but in a different culture and century and class, then I would have become a different person. The connection between genes and identity is too weak to serve as the missing link. What Fertilized Eggs Do by Themselves What is, according to Lenzen, the crucial difference between before and after fertilization? As we have seen, the answer that identity is the difference (i.e. that the individual whose future we're talking about comes into existence at fertilization) is not obvious, to put it mildly. It needs back-up. Lenzen needs to show us properties that constitute identity and coincide with fertilization; or at any rate properties that - identity or not — make a moral difference and coincide with fertilization. I will now look at Lenzen's candidates for such properties. I will ask whether (or how, or at what moral costs) they succeed in singling out fertilization as the crucial threshold. The morally relevant difference between the fertilized and the unfertilized egg, Lenzen tells us, is that the latter "does not by itself develop into a being who lives a life worth living. Its 'life' normally ends with menstruation, and as such has no more value than, say, the 'life' of a finger-nail or the 'life' of an appendix." 36 35

36

Cf. Singer (1995), p. 94. Genetic make-up is nothing that Lenzen himself wants to rely on, cf. his (1995), p. 231. Lenzen (1998), sect. 4.

Procreation

445

Now, in what sense of "by itself' is it true that an ordinary fertilized egg that turns into a happy life does so by itself, while an ordinary unfertilized egg (or an ordinary pair of an unfertilized egg and a sperm 37 ) that turns into a happy life does not do so by itself? I suggest we check out the various readings that come to mind. First reading of "by itself: Let us say that something that happens does so by itself if it happens 'normally'. This interpretation does not support Lenzen's claim that when an ordinary fertilized eggs turns into a happy life it does so by itself. As Lenzen himself confirms, neither the unfertilized egg nor the fertilized one normally makes its way to a happy life: "[0]nly 5 out of 100 fertilized human eggs survive up to delivery, and the corresponding survival-rates for embryos and fetuses at later stages isn't very much higher" 38 . Thus, the odds are both against unfertilized and fertilized eggs and won't yield the qualitative difference. Second reading of "by itself': Let us try another statistical interpretation - one that refers not to the events but to what people do; let us say that something that happens does so by itself if it is true that it would happen if everybody did what is 'normal' for them to do. Under this reading it might be true of some real-life fertilized eggs that turn into happy persons that they do so by themselves. But now there is more bad news for Lenzen. Firstly, the same might hold for some unfertilized eggs — again fertilization has failed to make the difference. Secondly, the whole idea of giving moral authority to the question what people usually do is bizarre. (To the extent that we grant that authority, we exclude the possibility of a widespread vice.) The third and most important point is an application of the second one. It is easy to see that, within Lenzen's theory, the current interpretation of "by itself" would imply that, if most women had abortions, abortions would be all right. Since this is doubtless a verdict Lenzen would reject, this interpretation cannot be the one he has in mind. 37 38

See note 27 above. Lenzen (1998), sect. 6.1; cf. also Singer (1993), ch. 6. The problem I am addressing here should not be confused with the one that Lenzen wants to solve, and can solve, by his expected utility approach (cf. his 1998, sect. 6.2). Once we have established that, in Lenzen's sense, the happy future of the fetus 'counts' whereas that of the unfiised pair of sperm and egg does not, then the expected utility approach would correctly insist that, when calculating the badness of killing a fetus, we weight its possible future happiness with the (somewhat dim) probability that the fetus, if not killed by us, would live to see it. But we are still at the step I have just italicized; expected utility doesn't help, and wasn't invoked by Lenzen to help, with this problem.

446

Ulla Wessels

Third reading of "by itself': Suppose we have a sufficiently clear distinction between acts and omissions. To avoid confusion with the word "act" in the wider sense (in which even omissions can be acts), I will call acts that are not omissions active acts, an expression that is ugly but unambiguous. Let us say that something that happens does so by itself if it would have happend even if nobody had performed any active acts. Again, this reading does not support Lenzen's claim that when an ordinary fertilized egg turns into a happy life it does so by itself. While pregnant, mother must do at least everything that is necessary for her own survival, then she must give birth, then the child has to be brought up. Obviously, all this involves lots of active acts; mother has to eat, for instance, and later either she or somebody else will have to feed her child - and clearly eating and feeding are active acts, not omissions. Fourth reading of "by itself': When I say that something happens by itself (so somebody might try to explain their usage to us), I have in mind a certain list of actions. The way I use these words, something that happens does so "by itself' if it is true of the event that it would happen even if nobody performed an action from the list I have in mind. Now everything will depend on the list. Say that "fertilizing" is on it, but that things like "eating while you're pregnant" or "feeding after birth" are not. Then Lenzen will be all right. On the one hand, a fertilized egg that turns into a happy person would do so by itself. Fertilization is on the list, but fertilization is not necessary (not for the step in question, that is); mother's eating while she's pregnant, feeding after birth, and other such manoeuvres, are necessary for the fertilized egg to acquire a happy life but are not on the list. On the other hand, unfertilized eggs that turn into happy persons never do so by themselves. That would require fertilization, and fertilization is on the list. Fine - but what is the right list? Any list will, in Lenzen's framework, be equivalent to a particular moral creed; it will not be, no matter by how charitable a standard, a general principle from which the particular creed can be derived. Some lists of actions would generate Lenzen's results, lots of other lists would not. To say that we have obligations to refrain from certain ac39

Nothing changes if we replace mother with a machine. Building and running the machine would require active acts as well — in real life, at any rate, and the existence of science-fiction scenarios in which fetuses grow up by themselves is of no help to Lenzen's position. Firstly, science-fiction scenarios only show that it is logically possible for a fetus to grow up by itself; but Lenzen thinks that in many real-life cases they grow up by themselves and that thus it would in many real-life cases be wrong to kill them. Secondly, we can also tell science-fiction stories in which the pair of sperm and unfertilized egg grows into a person by itself; hence, if science-fiction stories counted, fertilization would again fail to make the difference.

Procreation

447

tions (which would prevent fertilized eggs from turning into happy persons), but that we have no obligation to refrain from certain other actions (which would prevent unfertilized eggs from turning into happy persons) is not giving a reason why abortion is wrong and non-conception is not; it's just a paraphrase. Hence the fourth interpretation turns the expression "by itself' into a wild card; insert it into a moral theory that grants authority to what happens (or would happen) by itself and you can get practically any result you like, just by tailoring the relevant list of actions to suit your tastes. The exciting question is where the list comes from. The answer would require a mort general notion of 'by itself, and that, in turn, throws us back on the other readings. They, however, have already failed us. Further readings of "by itself"? What other type of explications of "by itself" could there be? I have no idea. I'm sorry if I've missed the decisive one. But I've done my best, and it would really be Lenzen's job to say a word or two on the concept that is supposed to bear so much weight in his theory. Moreover, I don't see where additional types of candidates could come from. There seem to be just two sources that could provide us with the basic material for an explication of "by itself": on the one hand, statistics and its cousins (standard or normal conditions, standard actions, causality, similarity, etc.); on the other hand, the difference between acts and omissions. The exercises we have been through make it reasonably clear that this material won't get us anywhere near an explication of "by itself" that will make the notion entail the difference that Lenzen assumes it entails. External Intervention, Inner Nature, Biological Constitution Have I been paying a malicious amount of attention to the innocent words "by itself' that appear just once in Lenzen's paper? I don't think so. The words aren't innocent, Lenzen's theory rests on them. True, he gives a few other brief hints that use different words. But it is easy to show that they are in the same boat and would succumb to the same treatment. One example is Byrnes claim (quoted approvingly by Lenzen) that "the possibility that an ovum will become a person depends upon external intervention". I just don't get it: so does, as most parents will have noticed, the possibility that a fertilized egg turns into a person!40 40

C f . our above discussion of reading no. one of " b y itself'. W h a t I'm saying here is true if the word "person" is used to imply that the entity will indeed experience happiness; but the word has to be used in this way if Byrne's quote is to support Lenzen's theory.

448

Ulla Wessels

Of course, if we simply define "intervention" so that fertilization counts as an intervention, whereas the many active acts (not just omissions) necessary for the step from pregnancy to birth (let alone from pregnancy to, say, kindergarten) do not count as interventions, then we obtain the desired result. But this is, in another guise, the list approach we encountered as reading no. four of "by itself" - a big moral issue is placed in the lap of a terminological caprice. A second example is Byrne s claim (again quoted approvingly by Lenzen) that fertilization changes "the inner nature and biological constitution" of the ovum; this is in line with Lenzen's own remark that "an unfertilized ovum qua its nature is not capable of making experiences"41. These references to nature are - what else could they be? - yet another version of the claim that the road a fertilized egg has to travel to kindergarten is in a different statistical or causal class than the road a pair of sperm and unfertilized egg has to travel to kindergarten. But it is not. Certainly, we can point to one active act, viz. bringing about fusion, that is necessary in the one case and no longer necessary in the other. But - as long as statistically the odds are against even the fertilized egg,42 and as long as myriads of other active acts are necessary even in order to pave its way into a happy life43 - why should, of all the active acts, that one catapult an object into a different statistical or causal class? Simply to say that it does ("period!"), whereas the others don't ("period!"), is, to say the least, somewhat arbitrary.44 Hence the Inference Fails Section 2.4 has been devoted to the inference from Anti-Death and Not-ProNew-Life to Anti-Abortion and Not-Pro-Fertilization. The upshot, I think, is clear enough. We can choose any one of various closely related terminologies (be it that of what happens "by itself', or "normally", or that of "nature", or "identity"), and we can stipulate meanings for these terms that will generate a difference between pairs of sperm and egg that have fused and those that have not (differences such as: after fusion the pair has a different "nature" or 41 42 43 44

Lenzen (1998), sect. 4. Cf. Lenzen (1998), sect. 6.1, and above, the discussion of reading no. one of "by itself. Cf. above, the discussion of reading of no. three of "by itself'. Helga Kuhse and Peter Singer tell a story in which the action that brings together egg and sperm in a laboratory is, by ordinary standards, tiny and undramatic (1992, p. 59). This reminder is helpful for our context, too. But it is true anyway that, in vitro or not, many of the acts required to turn a fetus into a happy life have, by ordinary standards, at least the causal calibre of those required for fertilization. I mention "ordinary" standards since no doubt standards could be defined that put fertilization in a different class; but for that sort of manoeuvre, see above, the discussion of reading no three of "by itself'.

Procreation

449

"identity" than before; or after fusion, but not before, the pair will attain a happy life, if at all, then "by itself' or "normally"). But this will only work if the stipulations are tailor-made for the moral results. General explications of the relevant concepts (in terms of statistics, causal distance, acts and omissions, genetic make-up, brain continuity, psychological connectedness, similarity, etc.) will, as we have seen, fail to yield the difference between egg and sperm before and after fusion. If somebody wants to argue for the difference, he cannot therefore, as Lenzen thinks he can, invoke these concepts. No general criterion will do Lenzen the favour of singling out fertilization; he will be reduced to saying that fertilization just is his criterion. There is, Lenzen tells us elsewhere, "eine riesige qualitative Differenz" between the cells before and after they fuse, and this "begründet den gravierenden moralischen Unterschied" between abortion and non-conception.45 But he has simply not told us what makes the tiny physical step, which we're all aware of, a "huge qualitative difference". Of course the step is necessary on the road to personhood or a happy future — but so are a thousand others. Does Lenzen not notice that, as it stands, the claim about the difference just amounts to the claim that abortion and non-conception have different normative status? And that, if it amounts to it, it can hardly support it? Lenzen also says (same place as just quoted) that the huge qualitative difference is usually called "the mystery of life". I give him that much: it sure is a mystery. 2.5.

Wolfgang Lenzen and the Midwives: Conclusion

What, then, should we think of Lenzen's theory? No doubt it is as carefully crafted a version of midwifism as we can wish for - a version that clearly deserves, and rewards, close study. However, close study also reveals at least three fundamental flaws. Firstly, there are good reasons to deny Anti-Death, in other words good reasons to believe that, where an individual has no ex-ante preference against death or its consequences, death cannot be bad for that person. These reasons carry over to the denial of Lenzen's ban on abortion. For all this, see section 2.2. Secondly, even if Anti-Death were tenable, the moral asymmetry claimed by the conjunction of Anti-Death and Not-Pro-New-Life is not. The mere question whether somebody is around anyway makes, other than the question whether he is around and wants this or that to happen no moral difference. Two arguments to that effect were presented in section 2.3. 45

Lenzen (1995), p. 232.

450

Ulla Wessels

Thirdly, even if, contrary to what "firstly" and "secondly" have just told us, Anti-Death and Not-Pro-New-Life were jointly tenable, Lenzen has not shown that the two of them object to abortion but tolerate non-conception see section 2.4. Far too much identity theory is left unexplained and undefended, so that the central identity-theoretical claim - Wolfgang Lenzen's answer to the familiar question "When does a life begin?" - is presented to us as sheer dogma. Plug the identity-theoretical dogma into a moral principle sensitive to identity, and you'll get a moral dogma.

3.

Richard Hare and the Rabbits 3.1.

Hare's Argument

Richard Hare believes that individuals with satisfied preferences ought to exist — and thus that we have obligations to conceive and not to abort them. 46 In the words of the introductions to this paper and this symposium, he is a Rabbit. Hare has put forward what is, as far as I know, the only genuine argument in support of the Rabbits' claim. In a nutshell: to be moral is to have analogous preferences for analogous situations; some real-life people want to have been born; hence morality requires them to have, for analogous situations (in which other peoples birth is at issue) analogous preferences (i.e. preferences for those people's birth). To see what is going on, we need a more detailed version. Consider a situation that we will call the past existential situation S. In S, one of two possible worlds had to be brought about, either the S-birth-world or the 5-non-birthworld. In the ^-birth-world, an individual exists that has far more and stronger satisfied preferences than frustrated ones and that, ex post, rationally wants to have been born. In the S-non-birth-world, the individual does not exist and, therefore, wants nothing. In fact, the S-birth-world has become actual; the individual exists. Let us call her Mary. Situation S is shown in figure 1; in the figure, and henceforth, "> x " stands for "is preferred by χ to". Let us pretend, for the sake of Hare's argument, that the question of Mary's existence or non-existence affects nobody else's interests. Then all we have to take into account is her preference in the matter, her actual ex-post preference, that is, for the 5-birth-world over the S-non-birth-world. 46

Some qualifications are supposed to apply. For instance, if after an abortion another child comes into existence instead of the aborted one, and is happier than the latter would have been, then an abortion may be permissible - cf. Hare (1975), sect. 10.5.

Procreation

451

Fig. 1: The past existential situation S

Hares general theory, universal prescriptivism, says that morality requires us to have the same preferences with respect to all situations which are universal copies of each other. 47 Thus, if Mary has a preference for the birth-world over the non-birth world with respect to situation S, she must, on pain of immorality, have the analogous preference with respect to every situation that is a universal copy of S. The next step is one from Mary's universal preference to a corresponding "ought"-judgement. What ought to be the case is, according to universal prescriptivism, what a universal preferrer wants to be the case. Hence, if Mary has, with respect to S and all its universal copies, a preference for the birth-worlds, we can say that, in S and all its universal copies, the birth-worlds ought to be brought about. So that is the moral judgement about the past existential situation S and all its universal copies. Let us now turn to one such universal copy of S, the present existential situation S*. S* raises a moral problem right now. Somebody is currently wondering whether Mary*, who plays in S* exactly the part Mary played in S, ought to be born. (Throughout this paper, I will give names with asterisks to merely possible situations and the people in them. Names without asterisks refer to actual situations and to actual people.) We can represent S* in the same way as S - see figure 2. We already know that, in S and all its universal copies, the birth-worlds ought to be brought about. We also know that S* is a universal copy of 5. Hence, the birth-world* of situation S* ought to be brought about; Mary* ought to be born. 48 47

48

A universal copy of a situation Κ is a situation that resembles Y in all its universal properties; a universal property, in turn, is a property that can be specified without reference to specific individuals. Cf. Hare (1981), chs. 1.6 and 6. It might be objected to Hare's argument that it is never applicable to the world as it is because, in real life, a new existential situation that raises a moral problem will always differ in

452

Ulla Wessels

Fig. 2: The present existential situation S*

Summing up the argument (with short for "preferred by individual χ universally" and >ς?" for "better than"): With respect to past existential situation S: S-birth-world >Mary 5-non-birth-world (Mary is a universal preferrer) With respect to past existential situation S and all its universal copies: their birth-world >Mary their non-birthworld (universal prescriptivism) With respect to past existential situation S and all its universal copies: their birth-world »ς? their non-birth-world (S* is a universal copy of 5) With respect to present existential situation S*: S*-birth-world* S*-non-birth-world* This is Hares argument for an obligation to procreate. 49 It is ingenious. I will call it the Rabbits' argument. Its clue is the innocence of its premiss (of its first

49

some universal property from any past existential situation; we won't find an existing person whose life would be universally identical to that of the person whose creation we are considering. Hare has foreseen this objection and has added an argument to the effect that it is legitimate to replace the requirement of universal identity by the weaker requirement of relevant similarity. Let us grant that Hare can indeed meet the objection and just limit ourselves to asking whether we would have an obligation to create a new person whose life would be universally identical to that of the existing person who is happy to be alive. Cf. Hare (1975), (1988a), (1988b), and (1998). My presentation deviates from Hare's to some extent, but not significantly. Hare has agreed that my reconstruction captures his argument (personal communication).

Procreation

453

line, that is). If we buy the argument, then whether we ought to bring an individual into existence can be deduced, in the end, from what ex post an actual person rationally wants to have happened. Compared to other questions procreation is sometimes said to confront us with (such as: what are the desires of merely possible people? How bad is it not to exist? And can we harm people who do not exist?), statements on actual rational ex-post preferences are rather unsuspicious. There's hope that, if we could deduce the morality of procreation from such statements alone, we could solve a host of major problems by comparatively unproblematic means. But, I will ask, is Hares method sound? Can, in general, actual rational ex-post preferences play the part Hare wants them to play? In section 3.2, I will show that, as it stands, the answer is no. If we consult actual rational expost preferences the way Hare suggests, the advice we get can be inconsistent. There are two ways out of the inconsistency, which will be discussed in sections 3.3 and 3.4 respectively. Both of them, however, would leave the Rabbits' argument with a large hole. Roughly speaking, in order to yield the conclusion that people with satisfied preferences ought to exist, the argument will need the additional premiss that satisfied preferences ought to exist. 3.2.

A Self-Contradiction: The Vermeer-Novalis Case

The situations S and S* are cases of what we can call different preferences choices — choices which affect the number or identity of preferences.50 Our question is: suppose somebody has a rational ex-post preference with respect to a different preferences choice that concerned himself; can, in general, that preference be a reliable guide for moral judgements on universally identical choices?51 Consider the following thought experiment featuring what we will call Vermeer-Novalis situations (for short, VN situations). Let us start with the past VN situation Τ. In Τ, it was an open question which of two possible worlds would be brought about, either the world in which an individual c turns into an expert on Vermeer and has, ex post, a rational preference of strength 5 for having become an expert on Vermeer (=: the Τ-V-world), or the world in which c turns into an expert on Novalis and has, ex post, a rational preference

50

51

For the term "different preferences choice", cf. the beginning of Fehige/Wessels (1998); it is a modification of Parins term "different people choice" from Reasons and Persons (1984, sect. 120). The objection that Hare doesn't say that in general they can will be dealt with at the end of this subsection.

454

Ulla Wessels

Fig. 3: The past VN situation Τ

Fig. 4: The present VN sitaution T*

of strength 5 for having become an expert on Novalis (=: the ^T-N-world).52 In fact, the T^-V-world became actual; c became an expert on Vermeer. We assume that, before the decision, c had no preferences either way. If we write "„ > x " for "is preferred by χ with strength n \ we can represent the case as usual — see figure 3. Same procedure as before, with situations S and S*. By adding asterisks across the board, we baptize a universal copy of Τ. Say that the copy, Τ*, raises a moral problem right now: somebody is wondering whether c*, who plays exactly the part in T* that c played in Τ, ought to become an expert either on Vermeer or on Novalis - see figure 4. Let us now apply Hare's inference pattern: With respect to past VN situation Τ: Γ-V-world 5 > f Γ-Ν-world 52

W h a t is the strength of a preference? I'm not sure, but let us assume (as Hare's system in general and some of his applications of the Rabbits' argument in particular have to) that some such notion is available.

Procreation

455

(c is a universal preferrer) With respect to past VN situation Τ and all its universal copies: their V-world 5 »