Professional Documents
Culture Documents
Harsanyi Morality and The Theory of Rational Behaviour
Harsanyi Morality and The Theory of Rational Behaviour
HARE
Whichever way we put it, whether in terms of what I am prepared to
prescribe or permit universally (and therefore also for when I am the
victim) or in terms of how to be fair as between the interests of all the
affected parties, I conclude that the acts I have listed will come out wrong
on the act-utilitarian calculation, because of the harms done to the in-
terests of those who are cheated, or the non-fulfilment of prescriptions to
which, we may -assume, they attach high importance. If we add to this
move the preceding one which rules out fantastic cases, and are clear about
the distinction between judgements about the character of the agent,
judgements about the moral rationality of the action, and judgements
about its moral rightness as shown by the outcome, I think that this
form of utilitarianism can answer the objections I have mentioned. Much
more needs to be said; the present paper is only a beginning, and is not
very origina1.
25
I publish it only to give some indication of the way in
which ethical theory can help with normative moral questions, and to
try to get the discussion of utilitarianism centred round credible forms
of it, rather than forms which we all know will not do.
15 Among many others from whose ideas I have learnt, I should like in particular to mention
Dr Lynda Sharp (Mrs Lynda Paine), in whose thesis 'Forms and Criticisms of Utilitarian-
ism' (deposited in the Bodleian Library at Oxford) some of the above topics are discussed in
greater detail.
2 Morality and the theory of
rational behaviour
JOHN C. HARSANYI
1 Historical background
The ethical theory I am going to describe in this paper is based on three
different time-honoured intellectual traditions in moral philosophy. It also
makes essential use of a great intellectual accomplishment of much more
recent origin, namely, the modern Bayesian theory of rational behaviour
under risk and uncertainty.
One of the three moral traditions I am indebted to goes back to Adam
Smith, who equated the moral point of view with that of an impartial but
sympathetic spectator (or observer).1 In any social situation, each partici-
pant will tend to look at the various issues from his own self-centred, often
emotionally biassed, and possibly quite one-sided, partisan point of view.
In contrast, if anybody wants to assess the situation from a moral point of
view in terms of some standard of justice and equity, this will essentially
amount to looking at it from the standpoint of an impartial but humane
and sympathetic observer. It may be interesting to note that modern
psychological studies on the development of moral ideas in children have
come up with a very similar model of monil value judgements.
2
Another intellectual tradition I have benefited from is Kant's. Kant
claimed that moral rules can be distinguished from other behavioural rules
by certain formal criteria and, in particular, by the criterion of universality
(which may also be described as a criterion of reciprocity).3 For example, if
I really believe that other people should repay me any money they have
borrowed from me, then I must admit that I am under a similar moral
obligation to repay any money I have borrowed from other persons under
comparable circumstances. Thus, in ethical content, Kant's principle of
universality says much the same thing as the golden rule of the Bible:
'Treat other people in the same way as you want to be treated by them.'
Among contemporary authors, the Oxford moral philosopher Hare has
Reprinted from Social Research, Winter 1977, vol. 44, no. 4.
1 Adam Smith 1976.
2 See, for example Piaget 1962.
3 Immanuel Kant 1785.
39
40 JOHN C. HARSANYI
advocated a moral theory based specifically on the Kantian universality
requirement (which he calls the 'universalisation' requirement).4
My greatest intellectual debt, however, goes to the utilitarian tradition
of Bentham, John Stuart Mill, Sidgwick, and Edgeworth, which made
maximisation of social utility the basic criterion of morality - social utility
being defined either as the sum, or the arithmetic mean, of the utility levels
of all individuals in the society.5 (What these classical utilitarians called
'social utility' is often called the 'social welfare function' in modern
welfare economics. But in many cases the term 'social welfare function' is
now used in a less specific sense, without any utilitarian connotations.)
Though many details of the classical utilitarian position may be un-
acceptable to us today, we must not forget what basic political and
moral principles they were fighting for. Basically, both in politics and in
ethics, they fought for reason against mere tradition, dogmatism, and
vested interests. In politics, they conceived the revolutionary idea of
judging .existing social institutions by an impartial rational test, that of
social utility, and did not hesitate to announce it in clear and unmistakable
terms if they felt that many of these institutions had definitely failed to pass
this test. Likewise, in ethics, they proposed to subject all accepted moral
rules to tests of rationality and social utility.
Their main opponents in moral philosophy were the intuitionists, who
claimed that we can discover the basic moral rules by direct intuition,
which, of course, made any rational evaluation of such moral rules both
impossible and unnecessary. Apparently, these intuitionist philosophers
were not particularly troubled by the well-known empirical fact that
people's 'moral intuitions' seem to be highly dependent on accidents of
their own upbringing and, more fundamentally, on the accident of being
raised in one particular society rather than another. Though there were
many notable exceptions, most people raised in a warlike, society or a
slave-holding society or a caste society always claimed to have the clear
'moral intuition' that the social practices of their society had full moral
justification. It was this uncritical acceptance of existing social practices
that the utilitarians fought against by their insistence on subjecting all
moral beliefs to a rational test.
In our own time, these crude forms of obscurantism in ethics have
largely disappeared. But it is still true, it seems to me, that the updated
version of classical utilitarianism is the only ethical theory which consis-
tently abides by the principle that moral issues must be decided by rational
tests and that moral behaviour itself is a special form of rational be-
haviour. I think it can be easily shown that all nonutilitarian theories of
4 Hare 1952.
5 Bentham 1948; John Stuart Mill 1962; Sidgwick 1962; Edgeworth 1881.
Morality and the theory ofrati",""al behaviour 41
morality, including John Rawls's very influential'theory6 and several
others, at one point or another involve some highly irrational moral
choices, representing major departures from a rational pursuit of common
human and humane intereslS; which, in my view, is the very essence of
morality.
Yet, notwithstanding its very considerable intellectual accomplish-
ments, classical utilitarianism was open to some major objections. The
most important step toward resolving most of these objections was taken
by Keynes's friend, the Oxford economist Harrod, who was the first to
point out the advantages of rule utilitarianism over act utilitarianism.
7
(But
he did not actually use this terminology. The terms 'act utilitarianism' and
'rule utilitarianism' were introduced only by Brandt.
8
) Act utilitarianism is
the view that each individual act must be judged directly in terms of the
utilitarian criterion. Thus a morally right act is one that, in the situation
the actor is actually in, will maximise social utility. In contrast, rule
utilitarianism is the view that the utilitarian criterion must be applied, in
the first instance, not to in,dividual acts but rather to the basic general rules
governing these acts. Thus a morally right act is one that conforms to the
correct moral rule applicable to this sort of situation, whereas a correct
moral rule is that particular behavioural rule that would maximise social
utility if it were followed by'everybody in all social situations of this
particular type.
I will discuss the moral implications of these two versions of utilitarian
theory in section 9. As I will argue, only rule utilitarianism can explain
why a society will be better off if people's behaviour is constrained by a
network of moral rights and moral obligations which, barring extreme
emergencies, must not be violated on grounds of mere social-expediency
considerations. Prior to the emergence of rule-utilitarian theory, utili-
tarians could not convincingly. defend themselves against the accusation
that they were advocating a super-Machiavellistic morality, which permit-
ted infringement of all individual rights and all institutional obligations in
the name of some narrowly defined social utility.
Virtually all the moral content of the ethical theory I' am going to
propose will come from these three intellectual traditions: Adam Smith's,
Kant's, and that of the utilitarian school. Yet it would not have been
possible to put all these pieces together into an intellectually satisfactory
theory of morality before the emergence, and without an extensive use, of
the modern theory of rational behaviour and, in particular, the modern
6 Rawls 1971. For a detailed critique of Rawls' theory, see Harsanyi 1975a. For a discussion
of some other nonutilitarian theories, see Harsanyi 1975c.
7 Harrod 1936.
8 Brandt 1959, pp. 369, 380.
42 JOHN C. HARSANYI
theory behaviour under risk and uncertainty, usually described
as BayesIan decIsIon theory. The Bayesian concept of rationality is a very
crucial ingredient of my theory.
2 Ethics as a branch of the general theory of rational
behaviour
I propose to argue that the emergence of modern decision theory has made
ethics into an organic part of the general theory of rational behaviour. The
concept of rational behaviour (practical rationality) is important in philo-
sophy both in its own right and because of its close connection with
theoretical rationality. It plays a very important role also in the empirical
social sciences, mainly in economics but also in political science and in
sociology (at least in the more analytically-oriented versions of these two
What is more important for our present purposes, the concept
of behaviour is the very foundation of the normative disciplines of
decIsIOn theory, of game theory, and (as I will argue) of ethics.
The concept of rational behaviour arises from the empirical fact that
hUI?an behaviour is to a large extent goal-directed behaviour. Basically,
ratIOnal behaviour is simply behaviour consistently pursuing some well
defined goals, and pursuing them according to some well defined set of
preferences or priorities.
We all know that, as a matter of empirical fact, even if human behaviour
is usually goal-directed, it is seldom sufficiently consistent in its goals and
in the priorities it assigns to its various goals, to approach the ideal of full
rationality. Nevertheless, in many fields of human endeavour - for ex-
in most areas of economic life, in many areas of politics (including
International politics), and in some other areas of social interaction -
does show sufficiently high degrees of rationality as to
gIve a. surpnSIng amount of explanatory and predictive power to some
analytIcal models postulating full rationality. (Of course, it is very possible
that we could further increase the explanatory and predictive power of our
theories if we paid closer attention to the actual limits of human rationality
and information-processing ability, in accordance with Simon's theory of
limited rationality.9)
Moreover, whether people actually do act rationally or not, they are
often in i.ncreasing the rationality of their behaviour; and they
are also mterested In the conceptual problem of what it would actually
mean to act fully rationally in various situations. It is the task of the
normative disciplines of decision theory, game theory, and ethics to help
9 See, for example, Simon 1960.
Morality and the theory of rational behaviour
43
people to act more rationally and to give them a better understanding of
what rationality really is.
For reasons I will describe presently, I propose to consider these three
disciplines as parts of the s.ame general theory of rational behaviour. Thus
one part of this general theorylO will be:
(1) The theory of individual rational behaviour, which itself comprises
the theories of rational behaviour
(IA) Under certainty,
(IB) Under risk (where all probabilities are known objective probabili-
ties), and
(IC) Under uncertainty (where some or all probabilities are unknown,
and may be even undefined as objective probabilities).
(IA), (IB), and (Ie) together are often called utility theory while (IB)
and (1 C) together are called decision theory.
The two other branches of the general theory of rational behaviour both
deal with rational behaviour in a social setting. They are:
(2) Game theory, which is a theory of rational interaction between two
or more individuals, each of them rationally pursuing his own objectives
against the other individual(s) who rationally pursue(s) his (or their) own
objectives. Any individual's objectives may be selfish or unselfish, as
determined by his own utility function. (A nontrivial game situation can
arise just as easily among altruists as it can among egoists - as long as these
altruists are pursuing partly or wholly divergent altruistic goals.)
(3) Ethics, which is a theory of rational behaviour in the service of the
common interests of society as a whole.
I think it is useful to regard (1), (2), and (3) as branches of the same basic
discipline, for the following reasons:
(i) All three normative disciplines use essentially the same method. Each
starts out by defining rational behaviour in its own field either by some set
of axioms or by a constructive decision model. In either case, this initial
definition may be called the primary definition of rationality in this par-
ticular field. Then, from this primary definition, each derives a secondary
definition of rationality, which is usually much more convenient than the
primary definition in itself would be in its axiomatic or constructive form,
both for practical applications and for further philosophical analysis. For
example, in case (lA) the secondary definition of rationality is utility
maximisation - which is for many purposes a much more convenient
characterisation of rational behaviour under certainty than is its primary
definition in terms of the usual axioms (the complete preordering require-
ment and the continuity axiom).
10 The remaining part of this section will be somewhat technical, but it can be omitted
without loss of continuity.
44 JOHN e. HARSANYI
In cases (1B) and (1 e), the secondary definition of rationality is ex-
pected-utility maximisation (with objective probability weights in case
(1B) and with subjective probability weights in case (Ie)).
In the game-theoretical case (2), the secondary definition is provided by
various game-the0retical solution concepts.
Finally, in the case of ethics (case (3)), as we will see, the secondary
definition of rationality (or of morality) is in terms of maximising the
average utility level of all individuals in the society.
This common method that these normative disciplines use represents a
unique combination of philosophical analysis and of mathematical
reasoning. In each case, a movement from the primary definition of
rationality to its secondary definition is a straightforward mathematical
problem. But discovery of an appropriate primary definition is always
essentially a philosophical - that is, a conceptual - problem (with the
possible exception of case (1A), where the philosophical dimension of the
problem seems to be less important). People familiar with research work in
these areas know the special difficulties that arise from this unusual
interdependence of philosophical and mathematical problems. These are
definitely not areas for people who prefer their mathematics without any
admixture of philosophy, or who prefer their philosophy without any
admixture of mathematics.
(ii) The axioms used by decision theory, game theory, and ethics are
mathematically very closely related. In all three disciplines they are based
on such mathematical properties as efficiency, symmetry, avoidance of
dominated strategies, continuity, utility maximisation, invariance with
respect to order-preserving linear utility transformations, etc.
(iii) Yet the most important link among the three disciplines lies in the
fact that recent work has made it increasingly practicable to reduce some
basic problems of game theory and of ethics partly or wholly to decision-
theoretical problems.
II
3 The equiprobability model for moral value judgements
After the two introductory sections, I now propose to describe my theory
of morality. The basis of this theory is a model for moral value judgements.
Any moral value judgement is a judgement of preference, but it is a
judgement of preference of a very special kind. Suppose somebody tells us:
'I much prefer our capitalist system over any socialist system because
11 In game theory, one step in this direction has been a use of probability models for analysing
games with incomplete information (Harsanyi 1967-8). More recently, a decision-
theoretical approach to defining a solution for noncooperative games has been proposed
(Harsanyi 1975b), On uses of decision theory in ethics, see Harsanyi 1977.
Morality and the theory of rational behaviour
45
under our capitalist system I happen to be a millionaire a?d have a
satisfying life, whereas under a socialist system I b,e m all probabIl-
ity at best a badly paid minor government offiCIal. T?IS may, be, a, very
reasonable judgement of preference from, hIS own
point of view. But nobody would call it a v,alue JudgeI?ent because It
would be obviously a judgement based pnmanly on self-mterest.
Compare this with a situation where ,would a
preference for the capitalist syste,m as the system WIthout
knowing in advance what particular socIal posltion he would occupy
under either system. To make it more precise, let us assume that he would
choose between the two systems under the assumption that, in either
system, he would have the same probability of occupying any of
available social positions. In this case, we could be sure that hIS chOlce
would be independent of morally irrelevant selfish considerations. There-
fore his choice (or his judgement of preference) between the two systems
would now become a genuine moral value judgement.
Of course it is not really necessary that a person who wants to make a
moral of the relative merits of capitalism and of socialism
should be literally ignorant of the actual social position that he does
occupy or would under each system. But it is, necessary t,hat he
should at least try his best to disregard this morally Irrelevant
information when he is making his moral assessment. OtherWIse hIS
assessment will be a genuine moral value judgement but rather will be
merely a judgement of personal preference.
For short reference, the fictitious assumption of having the same proba-
bility of occupying any possible social will called the equipro-
bability postulate, whereas the entire precedmg declslOn model based on
this assumption will be called the equiprobability model of moral value
judgements. , "
We can better understand the implications of this model If we subject It
to decision-theoretical analysis. Suppose the society we are considering
consists of n individuals, numbered as individual 1, 2, ... ,n, according to
whether they would occupy the 1st (highest), 2nd (second highest), ... ,
nth (lowest) social position under a given social system. Let VI', V2, ,
V denote the utility levels that individuals 1,2, ... ,n would enJoy under
system. The individual who wants to make a judgement
about the relative merits of capitalism and of soclahsm WIll be called
individual i. By the equiprobability postulate, individual i will act such a
way as if he assigned the same probability lin to his occupymg any
particu!ar social position and, therefore, to his utility reaching anyone of
the utility levels VI' V2, , Vn , ."
Now, under the assumed conditions, according to BayeSIan declSlOn
46 JOHN C. HARSANYI
theory, a rational individual will always choose that particular social
system that would maximise his expected utility, that is, the quantity
(1)
representing the arithmetic mean of all individual utility levels in society.
We can express this conclusion also by saying that a rational individual
will always use this mean utility as his social welfare function; or that he
will be a utilitarian, who defines social utility as the mean of individual
utilities (rather than as their sum, as many utilitarians have done).12
Of course, this conclusion makes sense only if we assume that it is
mathematically admissible to add the utilities of different individuals, that
is, if we assume that interpersonal comparisons of utility represent a
meaningful intellectual operation. I will try to show that this is in fact the
case.
In describing this equiprobability model, I have assumed that individual
i, who is making a moral value judgement on the merits of the two
alternative social systems, is one of the n members of the society in
question. But exactly the same reasoning would apply if he were an
interested outsider rather than a member. Indeed, for some purposes it is
often heuristically preferable to restate the model under this alternative
assumption. Yet, once we do this, our model becomes a modern restate-
ment of Adam Smith's theory of an impartially sympathetic observer. His
impartiality requirement corresponds to my equiprobability postulate,
whereas his sympathy requirement corresponds to my assumption that
individual i will make his choice in terms of interpersonal utility compar-
isons based on empathy with various individual members of society (see
section 5).
This equiprobability model of moral value judgements gives us both a
powerful analytical criterion and a very convenient heuristic criterion for
deciding practical moral problems. If we want to decide between two
alternative moral standards A and B, all we have to do is ask ourselves the
question, 'Would I prefer to live in a society conforming to standard A or
in a society conforming to standard B? - assuming I would not know in
advance what my actual social position would be in either society but
rather would have to assume to have an equal chance of ending up in any
one of the possible positions.'
12 For most purposes the two definitions of social utility are mathematically equivalent. This
is always true when n, the number of people in society, can be regarded as a constant. The
two definitions, however, yield different decision criteria in judging alternative population
policies. In this latter case, in my view, the mean-utility criterion gives incomparably
superior results.
Morality and the theory of rational behaviour
47
Admittedly, this criterion - or any conceivable moral criterion - will
still leave each of us with the great moral responsibility, and the often very
difficult intellectual task, of actually choosing between these two alterna-
tive moral standards in terms of this criterion. But by using this criterion
we will know at least what the actual intellectual problem is that we are
trying to solve in choosing between them.
My equiprobability model was first published in 1953, and was ex-
tended in 1955.13 Vickrey had suggested a similar idea,14 but my work was
independent of his. Later John Rawls again independently proposed a very
similar model which he called the 'original position', based on the 'veil of
ignorance'. 15 But while my own model served as a basis for a utilitarian
theory, Rawls derived very non utilitarian conclusions from ?is own. Yet
the difference does not lie in the nature of the two models, whIch are based
on almost identical qualitative assumptions. Rather, the difference lies in
the decision-theoretical analysis applied to the two models. One difference
is that Rawls avoids, any use of numerical probabilities. But the main
difference is that Rawls makes the technical mistake of basing his analysis
on a highly irrational decision rule, the maximin principle, which was
fairly fashionable thirty years ago but which lost its attraction a few years
later when its absurd practical implications were realised.
16
Our model of moral value judgements can also be described as follows.
Each individual has two very different sets of preferences. On the one
hand, he has his personal preferences, which guide his everyday behaviour
and which are expressed in his utility function Ui. Most people's personal
preferences will not be completely selfish: But they will. assig?
weights to their own interests and to the of faml.ly, theIr
friends and other personal associates than they wIll aSSIgn to the mterests
of strangers. On the other hand, each will also
moral preferences which mayor may not have much mfluence on hIS
everyday behaviour but which will guide his thinking in those - possibly
very rare - moments when he forces a special impersonal and impartial
attitude, that is, a moral attitude, upon himself. His moral preferences, un-
like his personal preferences, will by definition always. assign the same
weight to all individuals' interests, including his o,:n. These
erences will be expressed by his social-welfare function W;. TypIcally, dIf-
ferent individuals will have very different utility functions Ui but, as can be
seen from Equation (1) above, in theory they will tend to have identical
social-welfare functions - but only if they agree in their factual assump-
13 Harsanyi 1953 and 1955.
14 Vickrey 1945.
15 Rawls 1957; 1958; and 1971.
16 First by Radner and Marschak 1954, pp. 61-8. See also Harsanyi 1975a.
48
JOHN C. HARSANYI
tions on the nature of the individual utility functions V and on the con-
version between individuals' utilities (as tdecided by inter-
personal utIlIty comparIsons) - which, of course, may not be the case.
By definition, a moral value judgement is always an expression of one's
moral preferences. Any evaluative statement one may make will automati-
cally lose its status of a moral value judgement if it is unduly influenced by
one's personal interests and personal preferences.
4 An axiomatic justification for utilitarian theory
I now propose to present an alternative, this time axiomatic, justification
for utilitarian theory. This axiomatic approach yields a lesser amount of
philosophically interesting information about the nature of morality than
the equiprobability model does, but it has the advantage of being based on
weaker - almost trivial - philosophical assumptions. Instead of
usmg very specific philosophical assumptions about the nature of moral-
ity, it relies merely on Pareto optimality and on the Bayesian rationality
postulates.
We need three axioms:
Axiom 1: Individual rationality. The personal preferences of all n
individuals in society satisfy the Bayesian rationality postulates.1?
Axiom 2: Rationality of moral preferences. The moral preferences of at
least one individual, namely, individual i, satisfy the Bayesian rationality
postulates.
Axiom 3: Pareto optimality. Suppose that at least one individual j (j = 1,
... , n) personally prefers alternative A to alternative B, and that no
individual has an opposite personal preference. Then individual i will
morally prefer alternative A over alternative B.
Axiom 3 is a very weak and hardly objectionable moral postulate.
Axiom 1 is a rather natural rationality requirement. Axiom 2 is an equally
natural rationality requirement: in trying to decide what the common
interests of society are, we should surely follow at least as high standards
?f rationality as we follow (by Axiom 1) in looking after our own personal
mterests.
17 Most social scientists do not realise how weak the rationality postulates
are that Bayesian deCiSion theory needs for establishing the expected-utility maximisation
As and Aum.ann have shown (Anscombe and Aumann 1963), all we
IS the of. conslste.nt preferences. (complete preordering), a continuity
aXiom, the sure-thmg pnnClple. (aVOidance of domli1ated strategies), and the requirement
that .our preferences for should depend only on the possible prizes and on the
speCific events the prize. (The requirement can be replaced by
aXIOms the behaVIOur of numencal probabilities within lotteries. In
the literature, these aXIOms are usually called 'notational conventions'.)
Morality and the theory of rational behaviour 49
Axiom 1 implies that the personal preferences of each individual j (j = 1,
... , n) can be represented by a von Neumann-Morgenstern (=vNM)
utility function V;. Axiom 2 implies that the moral preferences of indi-
vidual i can be represented by a social welfare function W;, which mathe-
matically also has the nature of a vNM utility function. Finally, the three
axioms together imply the following theorem:
Theorem T. The social welfare function Wi of individual i must be of the
mathematical form:
(2)
n
Wi=.L a;U; with a
j
>0 for j=1, ... , n.
1=1
This result
18
can be strengthened by adding a fourth axiom:
Axiom 4: Symmetry. The social-welfare function Wi is a symmetric
function of all individual utilities. (That is, different individuals should be
treated equally.)
Using this axiom, we can conclude that
Equations(2) and (3) together are essentially equivalent to Equation(1).19
I realise that some people may feel uncomfortable with the rather
abstract philosophical arguments I used to justify my equiprobability
model. In contrast, the four axioms of the present section make only very
weak philosophical assumptions. They should appeal to everbody who
believes in Bayesian rationality, in Pareto optimality, and in equal treat-
ment of all individuals. Yet these very weak axioms turn out to be
sufficient to entail a utilitarian theory of morality.
5 Interpersonal utility comparisons
In everyday life we make, or at least attempt to make, interpersonal utility
comparisons all the time. When we have only one nut left at the end of a
trip, we may have to decide which particular member of our family is in
greatest need of a little extra food. Again, we may give a book or a concert
ticket or a free invitation to a wine-tasting fair to one friend rather than to
another in the belief that the former would enjoy it more than the latter
would. I do not think it is the task "f a philosopher or a social scientist to
18 For proof, see Harsanyi 1955.
19 There is, however, the following difference. Equation (1) implies that social utility must be
defined as the mean of individual utilities rather than as their sum. In contrast, Equations
(2) and (3) do not favour either definition of social utility over its alternative.
50 JOHN C. HARSANYI
deny obvious fact that people often feel quite capable of making such
compansons. Rather, his task is to explain how we ever managed to make
such comparisons - as well or as badly as we do make them.
Simple reflection will show that the basic intellectual operation in such
comparisons is imaginative empathy. We imagine ourselves
to be In the shoes of another person, and ask ourselves the question, 'If I
were now really in his position, and had his taste, his education, his social
background, his cultural values, and his psychological make-up, then
what would now be my preferences between various alternatives and how
much satisfaction or dissatisfaction would I derive from any alterna-
tive?' (An 'alternative' here stands for a given bundle of economic com-
modities plus a given position with respect to various noneconomic
variables, such as health, social status, job situation, family situation, etc.)
In other words, any interpersonal utility comparison is based on 'what I
will call the similarity postulate, to be defined as the assumption that, once
proper allowances have been made for the empirically given differences in
taste, education, etc., between me and another person, then it is reasonable
for me that 0.ur basic psychological reactions to any given
wtll.be much the same. Of course, it is only too easy
to mIsapply thIS slmtlanty postulate. For instance, I may fail to make
for differences in our tastes, and may try to judge the
th.at. a devoted eater derives from eating fish in terms of my
own mtense dIslIke for any kmd of sea food. Of course, sensible people will
seldom make such an obvious mistake. But they may sometimes make
much subtler mistakes of the same fundamental type.
In general, if we have enough information about a given person, and
make a real effort to attain an imaginative empathy with him, we can
probably make reasonably good estimates of the utilities and dlisutilities he
would obtain from various alternatives. But if we have little information
about him, our estimates may be quite wrong.
In any case, utilitarian theory does not involve the assumption that
people are very good at making interpersonal utility comparisons. It
Involves only the assumption that, in many cases, people simply have to
make such comparisons in order to make certain moral decisions - how-
ever badly they may make them. If I am trying to decide which member of
is in gre.atest need of food, I may sometimes badly misjudge the
situatIOn. But I SImply have to make some decision. I cannot let all
members of my family go hungry because I have philosophical scruples
about interpersonal comparisons and cannot make up my mind.
interpersonal utility comparisons do pose important
problems. In particular, they pose the problem that they
reqmre us to use what I have called the similarity postulate. Yet this
Morality and the theory of rational behaviour 51
postulate, by its very nature, is not open to any direct empirical test. I may
very well assume that different people will have similar psychological
feelings about any given situation, once differences in their tastes, educa-
tions, etc. have been alloweCl 01". But I can never verify this assumption
by direct observation since I have no direct access to their inner
feelings.
Therefore, the similarity postulate must be classified as a nonempirical a
priori postulate. But, of course, interpersonal utility comparisons are by
no means unique among empirical hypotheses in their dependence on such
nonempirical postulates. In actual fact, whenever we choose among
alternative empirical hypotheses, we are always dependent on some
nonempirical choice criteria. This is so because the empirical facts are
always consistent with infinitely many alternative hypotheses, and the
only way we can choose among them is by using a priori nonempirical
choice criteria, such as simplicity, parsimony, preference for the 'least
arbitrary' hypothesis, etc.
Our similarity postulate is a nonempirical postulate of the same general
type. Its intuitive justification is that, if two individuals show exactly
identical behaviour - or, if they show different behaviour but these dif-
ferences in their observable behaviour have been properly allowed for -
then it will be a completely arbitrary and unwarranted assumption to
postulate some further hidden and unobservable differences in their
psychological feelings.
We use this similarity postulate not only in making interpersonal utility
comparisons but also in assigning other people human feelings and con-
scious experiences at all. From a purely empirical point of view, a world in
which I would be the only person with real conscious experiences while all
other people were mindless robots would be completely indistinguishable
from our actual world where all individuals with human bodies are
conscious human beings. (Indeed, even a world in which I alone would
exist, and all other people as well as the whole physical universe would be
merely my own dream - solipsism - would be empirically indisting-
uishable from the world we actually live in.) When we choose the assump-
tion that we actually live in a world populated by millions of other human
beings, just as real and just as conscious as we are ourselves, then we are
relying on the same similarity postulate. We are essentially saying that,
given the great basic similarity among different human beings, it would be
absurd to postulate fundamental hidden differences between them by
making one person a conscious human being while making the others
mere robots, or by making one person real while making the others mere
dream figures. (Strictly speaking, we cannot exclude the possibility that
somebody who looks human will turn out to be an unfeeling robot; but we
52 JOHN C. HARSANYI
have no scientific or moral justification to treat him like a robot before the
evidence for his being a robot becomes overwhelming.)
There is no logical justification for using the similarity postulate to reject
th.e hypothesis that o t ~ r people are mere robots (or mere dream figures)
yet to resist interpersonal utility comparisons based on the very same
similarity postulate. It is simply illogical to admit that other people do
have feelings and, therefore, do derive some satisfaction from a good meal
in the same way we do; yet to resist the quantitative hypothesis that the
amount of satisfaction they actually obtain from a good dinner - that is,
the personal importance they attach to a good dinner - must be much the
same as it is in our own case, after proper allowances have been made for
differences in our tastes, in the food requirements of our bodies, in our
state of health, etc. A willingness to make interpersonal comparisons is no
more than an admission that other people are just as real as we are, that
they share a common humanity with us, and that they have the same basic
capacity for satisfaction and for dissatisfaction, in spite of the undeniable
individual differences that exist between us in specific detail.
The long-standing opposition by many philosophers and social scien-
tists to interpersonal utility comparisons goes back to the early days of
logical positivism, when the role of nonempirical a priori principles, like
the similarity postulate, in a choice among alternative empirical hypoth-
eses was very poorly understood. We owe an immense intellectual debt to
the logical positivists for their persistent efforts to put philosophy on truly
scientific foundations by combining strict empiricism with the strict
mathematical rigour of modern logic. But there is no denying that many of
their specific philosophical views were badly mistaken, and that they had
little appreciation in their early period for the importance of a priori
principles and, more generally, for the importance of theoretical ideas in
empirical science.
One would think that after so many years the time had come to escape
the narrow confines of a long-obsolete logical-positivist orthodoxy and to
have a fresh look at the problem of interpersonal utility comparisons.
6 The use of von Neumann-Morgenstern utility functions
The utilitarian theory I have proposed makes an essential use of von
Neumann-Morgenstern (= vNM) utility functions. Many critics have
argued that any use of vNM utility functions is inappropriate, because
they merely express people's attitudes toward gambling, and these atti-
tudes have no moral significance.
2o
This objection is based on a rather
common misinterpretation of vNM utility functions. These utility func-
20 See, for example, Rawls 1971, pp. 172, 323.
Morality and the theory of rational behaviour 53
tions do express people's attitudes to risk taking (in gambling, buying
insurance, investing and other similar activities). But they do not merely
register these attitudes; rather, they try to explain them in terms of the
relative importance (relative utility) people attach to possible gains and to
possible losses of money or of other economic or noneconomic assets.
For example, suppose that Mr X is willing to pay $5 for a lottery ticket
that gives him a 111,000 chance of winning $1,000. Then we can explain
his willingness to gamble at such very unfavourable odds as follows. He
must have an unusually high utility for winning $1,000, as compared with
his dis utility for losing $5. In fact, even though the ratio of these two
money amounts is only 1,000 : 5 = 200 : 1, the ratio of the corresponding
utility anddisutility must be at least 1,000 : 1. (If we know Mr X's
personal circumstances, then we will often be able to carry this explana-
tion one step - or several steps - further. For instance, we may know that
his strong desire for winning $1,000 arises from the fact that he needs the
money for a deposit on a badly needed car, or for some other very
important large and indivisible expenditure; while his relative unconcern
about losing $5 is due to the fact that such a loss would not seriously
endanger his ability; to pay for his basic necessities - food, lodging, etc.
21
)
In other words, even though a person's vNM utility function is always
estimated in terms of his behaviour under risk and uncertainty, the real
purpose of this estimation procedure is to obtain cardinal..:utility measures
for the relative personal importance he assigns to various economic (and
noneconomic) alternatives.
No doubt, since social utility is defined in terms of people's vNM utility
functions, our utilitarian theory will tend to assign higher social priorities
to those individual desires for which people are willing to take con-
siderable risks in order to satisfy them. But this is surely as it should be.
Other things being equal, we should give higher social priorities to in-
tensely felt human desires; and one indication that somebody feels strong-
ly about a particular desired objective is his willingness to take sizable risks
to attain it. For example, if a person is known to have risked his life in
order to obtain a university education (e.g., by escaping from a despotic
government which had tried to exclude him from all higher education),
then we can take this as a reasonably sure sign of his attaching very high
personal importance (very high utility) to such an education; and I cannot
see anything wrong with our assigning high social priority to helping him
to such an education on the basis of this kind of evidence.
21 Fundamentally, any explanation of why a given person's vNM utility function has any
specific shape and, in particular, why its convex and concave segments are distributed in
the way they are, will be typically in terms of substitution and complementarity relations
among the commodities consumed by him. Mathematically, indivisible commodities are a
special case of complementarity.
54 loHN C. HARSANYI
7 Preference utilitarianism, hedonism, ideal utilitarianism, and
the question of irrational preferences
The utilitarian theory I have proposed defines social utility in terms of
individual utilities, and defines each person's utility function in terms of
his personal preferences. Thus, in the end, social utility is defined in terms
of people's personal preferences. This approach may be called preference
utilitarianism. It is not the same approach that was used by the nineteenth-
century utilitarians. They were hedonists (hedonistic utilitarians), and
defined both social utility and individual utility functions in terms of
feelings of pleasure and pain. A third approach, called ideal utilitarianism,
was proposed by the Cambridge philosopher Moore, who defined both
social utility and individual utilities in tenus of amounts of 'mental states
of intrinsic worth', such as the mental states involved in philosophy,
science, aesthetic appreciation of works of art, experiences of personal
friendship, etc,22
Both hedonistic and ideal utilitarianism are open to serious objections.
The former presupposes a now completely outdated hedonistic psycho-
logy. It is by no means obvious that all we do we do only in order to attain
pleasure and to avoid pain. It is at least arguable that in many cases we are
more interested in achieving some objective state of affairs than we are
interested in our own subjective feelings of pleasure and pain that may
result from achieving it. It seems that when I give a friend a present my
main purpose is to give him pleasure rather than to give pleasure to myself
(though this may very well be a secondary objective). Even if I want to
accomplish something for myself, it is by no means self-evident that my
main purpose is to produce some feelings of pleasure in myself, and it is not
the actual accomplishment of some objective condition, such as having a
good job, solving a problem, or winning a game, etc. In any case, there IS
no reason whatever why any theory of morality should try to prejudge the
issue whether people are always after pleasure or whether they also have
other objectives.
As to ideal utilitarianism, it is certainly not true as an empirical observa-
tion that people's only purpose in life is to have 'mental states of intrinsic
worth'. But if this is not in fact the case, then it is hard to see how we could
prove that, even though they may not in fact act in this way, this is how
they should act. Moreover, the criteria by which 'mental states of intrinsic
worth' can be distinguished from other kinds of mental states are ex-
tremely unclear. (Moore's own theory that they differ from other mental
states in having some special 'nonnatural qualities' is a very unconvincing
II Moore 1903.
Morality and the theory of rational behaviour 55
old-fashioned metaphysical assumption lacking any kind of supporting
evidence.)
More fundamentally, preference utilitarianism is the only form of utili-
tarianism consistent with the important philosophical principle of
preference autonomy. By this I mean the principle that, in deciding what is
good and what is bad for a given individual, the ultimate criterion can only
be his own wants and his own preferences. To be sure, as I will myself
argue below, a person may irrationally want something which is very 'bad
for him'. But, it seems to me, the only way we can make sense of such a
statement is to interpret it as a claim to the effect that, in some appropriate
sense, his own preferences at some deeper level are inconsistent with what
he is now trying to achieve.
Any sensible ethical theory must make a distinction between rational
wants and irrational wants, or between rational preferences and irrational
preferences. It would be absurd to assert that we have the same moral
obligation to help other people in satisfying their utterly unreasonable
wants as we have to help them in satisfying their very reasonable desires.
Hedonistic utilitarianism and ideal utilitarianism have no difficulty in
maintaining this distinction. They can define rational wants simply as ones
directed toward objects having a real ability to produce pleasure, or a real
ability to produce 'mental states of intrinsic worth'; and they can define
irrational wants as ones directed toward objects lacking this ability. But it
may appear that this distinction is lost as soon as hedonistic and ideal
utilitarianism are replaced by preference utilitarianism.
In actual fact, there is no difficulty in maintaining this distinction even
without an appeal to any other standard than an individual's own per-
sonal preferences. All we have to do is to distinguish between a person's
manifest preferences and his true preferences. His manifest preferences are
his actual preferences as manifested by his observed behaviour, including
preferences possibly based on erroneous factual beliefs, or on careless
logical analysis, or on strong emotions that at the moment greatly hinder
rational choice. In contrast, a person's true preferences are the preferences
he would have if he had all the relevant factual information, always
reasoned with the greatest possible care, and were in a state of mind most
conducive to rational choice. Given this distinction, a person's rational
wants are those consistent with his true preferences and, therefore, consis-
tent with all the relevant factual information and with the best possible
logical analysis of this information, whereas irrational wants are those
that fail this test.
In my opinion, social utility must be defined in terms of people's true
preferences rather than in terms of their manifest preferences. But, while it
is only natural to appeal from a person's irrational preferences to his
56 JOHN C. HARSANYI
underlying 'true' preferences, we must always use his own preferences in
some suitable way as our final criterion in judging what his real interests
are and what is really good for him.
8 Exclusion of antisocial preferences
I have argued that, in defining the concept of social utility, people's
irrational preferences must be replaced by what I have called their true
preferences. But I think we have to go even further than this: some
preferences, which may very well be their 'true' preferences under my
definition, must be altogether excluded from our social-utility function. In
particular, we must exclude all clearly antisocial preferences, such as
sadism, envy, resentment, and malice.
23
According to utilitarian .theory, the fundamental basis of all our moral
commitments to other people is a general goodwill and human sympathy.
But no amount of goodwill to individual X can impose the moral obliga-
tion on me to help him in hurting a third person, individual Y, out of sheer
sadism, ill will, or malice. Utilitarian ethics makes all of us members of the
same moral community. A person displaying ill will toward others does
remain a remember of this community, but not with his whole personality.
That part of his personality that harbours these hostile antisocial feelings
must be excluded from membership, and has no claim for a hearing when
it comes to defining our concept of social utility.24
9 Rule utilitarianism vs. act utilitarianism
Just as in making other moral decisions, in choosing between rule utilitar-
ianism and act utilitarianism the basic question we have to ask is this:
Which version of utilitarianism will maximise social utility? Will society
be better off under one or the other? This test very clearly gives the
advantage to rule utilitarianism.
In an earlier paper25 I proposed the following decision-theoretical model
fOl" studying the moral implications of the two utilitarian theories. The
23 For a contrary view, see Smart 1961, pp. 16-17.
24 The German neo-Kantian utilitarian philosopher Leonard Nelson proposed a distinction
between legitimate and illegitimate personal interests (Nelson 1917-32). He argued that
the only interests we are morally obliged to respect are legitimate interests. Thus, under his
theory, exclusion of antisocial preferences from our concept of social utility is merely a
special case of the general principle of disregarding all illegitimate interests. Unfortunately,
Nelson did not offer any clear formal criterion for defining legitimate and illegitimate
interests. But it seems to me that a really satisfactory theory of legitimate and illegitimate
interests would be a major step forward in utilitarian moral philosophy. Yet discussion of
this problem must be left for another occasion. (The reference to Nelson's work lowe to
Reinhard Selten.)
25 Harsanyi 1977.
Morality and the theory of rational behaviour 57
problem we to consider is that of making moral decisions, that is, the
of the morally right action is in a given situation or
ill a giverNiass of SituatIOns. In actual fact, analytically it is preferable to
redefine this problem as one of choosing a morally right strategy. Here the
term 'strategy' has its usual decision-theoretical and game-theoretical
meaning. Thus a strategy is a mathematical function assigning a specific
action to any possible situation, subject to the requirement that, if the
agent has insufficient information to distinguish one situation from
another, then any strategy of his must assign the same specific action to
both (In technical language, all choice points belonging to the
must the same specific action assigned to them.)
The theones use different decision rules in solving this
problem. For both theories, a moral decision problem is a
involving maximisation of the same quantity,
utIlIty. But the two theories impose very different mathema-
tical constramts on this maximisation problem. An act-utilitarian moral
agent assumes t?e strategies of all other moral agents (including those
of all utIlItanan agents) are given and that his task is merely to
choose. hiS own strategy so as to maximise social utility when all other
strategies are kept constant. In contrast, a rule-utilitarian moral agent will
regard not only his own strategy but also the strategies of all other
agents as variables to be determined during the maximisa-
tIOn process so as to maximise social utility. For him this maximisation
process is subject to two mathematical constraints: one is that the
to be chosen for all rule-utilitarian agents must be identical
(smce, by definition of rule utilitarianism, all rule-utilitarian agents are
to follow the same general moral rules); the other is that
strategI.es of all nonutilitarian agents must be regarded as given. (On
thIS last pomt both utilitarian theories agree: people known not to believe
in cannot be expected to choose their strategies by
trymg to maxlmIs.e.soc.Ial utility,. They may follow traditional morality, or
some. other nonutIhtanan moraltty, or may simply follow self-interest, etc.
But, m. any case, for the purposes of a utilitarian decision problem, their
strategIes be as being given from outside of the system.)
These differences m the decision rules used by the two utilitarian
theories, and in particular the different ways they define the constraints for
the. utilitarian. problem, have important practical impli-
catIOns. One ImplIcatIOn IS that rule utilitarianism is in a much better
position to organise cooperation and strategy coordination among dif-
ferent people (coordination effect).
For;'example, consider the problem of voting when there is an important
measure on the ballot but when voting involves some minor incon-
58 JOHN C. HARSANYI
venience. Suppose there are 1,000 voters strongly favouring the measure,
but it can be predicted with reasonable certainty that there will also be 800
negative votes. The measure will pass if it obtains a simple majority of all
votes cast. How will the two utilitarian theories handle this problem?
First suppose that all 1,000 voters favouring the measure are act utilita-
rians. Then each of them will take the trouble to vote only if he thinks that
his own vote will be decisive in securing passage of the measure, that is, if
he expects exactly 800 other people favouring the measure to vote (since in
this case his own vote will be needed to provide the 801 votes required for
majority). But of course each voter will know that it is extremely unlikely
that his own vote will be decisive in this sense. Therefore, most act-
utilitarian voters will not bother to vote, and the measure will fail, possibly
with disastrous consequences for their society.
In contrast, if the 1,000 voters favouring the measure are rule utilita-
rians, then all of them will vote (if mixed strategies are This
is so because the rule-utilitarian decision rule will allow them a choice only
between two admissible strategies: one requiring everybody to vote and
the other requiring nobody to vote. Since the former will yield a higher
social utility, the strategy chosen by the rule-utilitarian criterion will be for
everybody to vote. As this example shows, by following the rule-utilitarian
decision rule people can achieve successful spontaneous cooperation in
situations where this could not be done by adherence to the act-utilitarian
decision rule (or at least where this could not be done without explicit
agreement on coordinated action, and perhaps without an expensive
organisation effort).
Though in some situations this coordination effect may be quite impor-
tant, it seems to me that the main advantage of rule utilitarianism over act
utilitarianism really lies in a different direction, namely, in its ability to
take proper account of the implications that alternative systems of possible
moral rules would have for people's expectations and incentives (expecta-
tion and incentive effects).
For example, consider the problem of keeping promises. Traditional
morality says that promises should be kept, with a possible exception of
cases where keeping a promise would impose excessive hardship on the
promise maker (or perhaps on third persons). In contrast, act utilitarian-
ism would make the breaking of a promise morally permissible whenever
this would yield a slightly higher social utility - perhaps because of
unexpected changes in the circumstances - than keeping of the promise
would yield. But this would greatly reduce the social benefits associated
with the making of promises as an institution. It would make it rather
uncertain in most cases whether any given promise would be kept. People
would be less able to form definite expectations about each other's future
Morality and the theory of rational behaviour
59
behaviour and would have a general, feeling of insecurity about the future.
Moreover, this uncertainty would greatly reduce their incentives to engage
in various socially very useful activities on the expectation that promises
given to them would be kept. (For instance, they would become much less
willing to perform useful services for other people for promised future
rewards.)
As compared with act utilitarianism, rule utilitarianism will be much
closer to traditional morality in maintaining that promises should be kept,
subject only to rather rare exceptions. An act utilitarian always asks the
question, 'Would this one act of possible promise breaking increase or
decrease social utility?' In contrast, a rule utilitarian has to ask, 'What
particular moral rule for promise keeping would maximise social utility?'
As a result, an act utilitarian can consider the socially unfavourable
effects of promise breaking only to the extent that these have the nature of
causal consequences of individual acts of promise breaking. No doubt, one
act of promise breaking already will somewhat reduce people's trust in
promises, but normally this effect will be quite small. In contrast, a rule
utilitarian can also consider the causal consequences of a general practice
of repeated promise breaking. But, more importantly, he can also consider
the noncausal logical implications of adopting a moral rule permitting
many easy. exceptions to promise keeping.
More particularly, he will always have to ask the question, 'What would
be the social implications of adopting a moral rule permitting that prom-
ises should be broken under conditions A, B, C, etc. - assuming that all
members of the society would know
26
that promise breaking would be
permitted under these conditions?' Thus he will always have to balance the
possible direct benefits of promise breaking in some specific situations
against the unfavourable expectation and incentive effects that would
arise if people knew in advance that in these situations promises would not
be kept. In other words, rule utilitarianism not only enables us to make a
rational choice among alternative possible general rules for defining
morally desirable behaviour. Rather, it also provides a rational test for
determining the exceptions to be permitted from these rules.
Limitations of space do not allow me to discuss the moral implications
of rule utilitarianism here in any greater detail.
27
It can be shown, however,
that rule utilitarianism comes fairly close to traditional morality in recog-
nising the importance of social institutions which establish a network of
26 In trying to evaluate a possible moral rule from a rule-utilitarian point of view, we must
always assume that everybody would know the content of this moral rule. This is so because
in principle everybody can always find out by direct computation what particular set of
moral (ules (i.e., what particular moral strategy) is optimal in terms of the rule-utilitarian
criterion.
17 But see Harsanyi 1977.
60 JOHN C. HARSANYI
moral rights and of moral obligations among different people in society;
and in maintaining that these rights and obligations must not be infringed
upon on grounds of immediate social utility, with the possible exception of
some very rare and very special cases. (The main social advantages of such
stable rights and stable obligations, once more, lie in their beneficial
expectation and incentive effects.) But of course we cannot expect that the
rule-utilitarian criterion would confirm traditional views on these matters
in all particulars.
10 The utility of free personal choice
As Rawls has rightly pointed out,28 traditional utilitarianism tries to
impose unreasonably strict moral standards on us, because it requires us to
choose every individual action of ours so as to maximise social utility.
Thus, if I feel like reading a book for entertainment, I must always ask
myself whether my time could not be more usefully devoted to looking
after the poor, or to converting some as yet unconverted colleagues to
utilitarianism, or to taking part in some other socially beneficial project,
etc. The only ways I could possibly justify my taking out time for reading
this book would be to argue that reading it would givt: me exceptionally
high direct utility (so as to exceed any social utility I could possibly
produce by alternative activities), or that my reading the book would have
great instrumental utility - for example, by restoring my temporarily
depleted mental and physical energy for future socially very beneficial
activities.
There is obviously something wrong with this moral choice criterion. It
is not hard to see where the problem lies. Any reasonable utilitarian theory
must recognise that people assign a nonnegligible positive utility to free
personal choice, to freedom from unduly burdensome moral standards
trying to regulate even the smallest details of their behaviour. Suppose we
could choose between a society with the highest possible moral standards,
regulating every minute of our waking lives in full detail, and a society
with somewhat more relaxed moral standards, leaving us a reasonable
amount of free choice in planning our everyday activities. It is very
possible (though it is by no means certain) that, by imposing much stricter
standards, the former society would attain higher levels of economic and
cultural achievement than the latter would. Nevertheless, many of us
might very well prefer to live in the latter society - at least if the differences
in the economic and cultural standards between the two societies were not
unduly large.
28 Rawls 1971, p. 117.
Morality and the theory of rational behaviour 61
What this means analytically is that, apart from the social utility W we
assign to the outcome of any given activity, we must also assign some
procedural utility V > 0 to our having a free personal choice among
alternative activities. Suppose we have to choose between two alternative
strategies and ':. likely to yield the outcome utilities W':' and ':',
with W':' > Then classical utilitarianism would select strategy as
the only morally permissible strategy. But, it seems to me, we must
recognise as being an equally permissible strategy, provided that
+ W>:.
11 Conclusion
I have tried to show that there is a unique rational answer to the philo-
sophical question, 'What is morality?' I have argued that, by answering
this question, we obtain a very specific decision rule for choosing between
alternative possible moral codes.
Even if this conclusion is accepted, this will not mean that practical
moral problems from now on will become simply matters of solving some
well-defined mathematical maximisation problems. Solving such prob-
lems will always involve extremely important questions of personal judge-
ment we have to use our own best judgement whenever we lack
completely reliable factual information about some of the relevant
variables. We will often lack reliable i,nformation about other people's
manifest preferences and, even more so, about their true preferences. Our
interpersonal utility comparisons may also be based on insufficient in-
formation, etc.
But the most fundamental source of uncertainty in our moral decisions
will always lie in our uncertainty about the future, including our uncer-
tainty about the future effects of our present policies, both in the short run
and in the long run. It seems to me that careful analysis will almost
invariably show that the most important source of moral and political
disagreements among people of goodwill lies in divergent judgements
about future developments and about the future consequences of alterna-
tive policies.
I have tried to show that an updated version of classical utilitar'ianism is
the only ethical theory consistent with both the modern theory of rational
behaviour and a full commitment to an impartially sympathetic human-
itarian morality.
On the other hand, neither the concept of rationality alone, nor a
commitment to a humanitarian morality alone, could yield a useful ethical
theory. 'Rather, we need a combination of both. As I have argued in
discussing Rawls's theory, even the best intuitive insight into the nature of
62 JOHN C. HARSANYI
morality will yield a highly unsatisfactory ethical theory if these insights
are conjoined with a highly irrational decision rule like the maximin
principle. Conversely, even the most careful analysis of the concept of
rationality cannot show that rationality entails a commitment to a
humanitarian morality.
Kant believed that morality is based on a categorical imperative so that
anybody who is willing to listen to the voice of reason must obey the
commands of morality. But I do not think he was right. All we can prove
by rational arguments is that anybody who wants to serve our common
human interests in a rational manner must obey these commands. In other
words, all we can prove are hypothetical imperatives of the form: 'If you
want to act in a way that an impartially sympathetic observer would
approve of, then do such and such', or: 'If you want your behaviour to
satisfy the axioms ... then do such and such.'29 But I do not think that this
negative conclusion is a real setback for moral philosophy, or has any
important practical implications at all. As a practical matter, all of us have
always known that rational discussion about moral issues is possible only
between people who share some basic moral commitments, such as a
common interest in a truly humanitarian morality.
Let me end with a disclaimer. I think the utilitarian theory I have
described in principle covers all interpersonal aspects of morality. But I do
not think it covers all morality. There are some very important moral
obligations it fails to cover because they are matters of individual morality
and of individual rationality. Perhaps the most important such obligation
is that of intellectual honesty, that is, the duty to seek the truth and to
accept the truth as far as it can be established - regardless of any possible
positive or negative social utility this truth may have. (Telling the truth to
others may be constrained by tact, respect for other people's feelings, or
commitments to secrecy, etc. But admitting the truth to ourselves is not.)
Intellectual honesty requires us to accept even very unpleasant truths
rather than withdraw into a dream world or a fool's paradise based on
self-deception. It also requires us to accept wholeheartedly the truth that
we are not alone in this world but rather share a common human nature
with many millions of others. Acceptance of this particular truth is, of
course, not merely a matter of theoretical rationality; rather, it is also the
intellectual basis of all social morality.
29 Harsanyi 1958.
* The author wants to thank the National Science Foundation for supporting this research
through Grant Soc 77-06394 to the Center for Research in Management Science, Univer-
sity of California, Berkeley.
3 The economic uses of
utilitarianism
1
J. A. MIRRLEES
economists, when evaluating alternative economic policies, are utili-
tanans. At any rate they look at something they call the total utility of the
outcome. This paper is intended to argue in favour of this procedure. 2 It
may be as well first to exemplify it.
An interesting question is how much income ought to be redistributed
from th?se with high wages and salaries to those with low wages. To
It, one set a model in which each individual's utility is a
numencal functIOn of hIS net income, after taxes and subsidies and of the
of labour he supplies. Each individual, supposedly how
hIS Income ?epends on the labour he supplies, decides how much to supply
by what will maximise his utility. All these labour supply
taken determine the output of the economy. A redis-
tnbutIve system, conSIstIng of taxes and subsidies, is feasible provided that
the out?ut of economy is sufficient to provide for public and private
expendItures, pnvate expenditures being determined by private net in-
comes. The object of the exercise is to find which feasible redistributive
system yields the greatest total utility.
This is not the place to defend ,the simplifications of such an economic
analysis, far less to discuss how it might be improved. Even within the
model outlined, assumptions as to the kinds of taxes and subsidies that are
possible have a effect on the results. I shall want to return to
1 A public lectu.re with the same title was given at University College, London, in February
1977. The mam arguments were the same, but it is doubtful whether there are any common
sentences: I am for that invitation and the opportunity it provided to
attempt to artIculate an economIst's defence of utilitarian methods as used in much
economics. I should like to acknowledge discussions on
these WIth J. R. Broome, P. A. Diamond, and A. K. Sen, and their comments on the
first draft of thIS paper. Comments by P. S. Dasgupta and Q. R. D. Skinner were also useful.
2 ?een so many presenting versions of utilitarianism, or defending it
agamst cn.tI.cIsm (many of whIch I have read only cursorily or not at all), that it is hard to
wntmg another. But there of emphasis from the major statement by
VIckrey (1960), and substantIal dIfferences from Harsanyi (1953, 1955, and later
Qoth of.whom these matters from the point of view of economic problems.
T.akmg that pomt of VIew, I found that I wanted to deal with a number of matters not
dIscussed by Hare (1976) and Smart (1973) in their statements.
63