Utilitarianism is the doctrine that the rightness of actions is to be judged by their consequences. What do we mean by ‘actions’ here? Do we mean particular actions or do we mean classes of actions? According to which way we interpret the word ‘actions’ we get two different theories, both of which merit the appellation ‘utilitarian’.
(1) If by ‘actions’ we mean particular individual actions we get the sort of doctrine held by Bentham, Sidgwick, and Moore. According to this doctrine we test individual actions by their consequences, and general rules, like ‘keep promises’, are mere rules of thumb which we use only to avoid the necessity of estimating the probable consequences of our actions at every step. The rightness or wrongness of keeping a promise on a particular occasion depends only on the goodness or badness of the consequences of keeping or of breaking the promise on that particular occasion. Of course part of the consequences of breaking the promise, and a part to which we will normally ascribe decisive importance, will be the weakening of faith in the institution of promising. However, if the goodness of the consequences of breaking the rule is in toto greater than the goodness of the consequences of keeping it, then we must break the rule, irrespective of whether the goodness of the consequences of everybody’s obeying the rule is or is not greater than the consequences of everybody’s breaking it. To put it shortly, rules do not matter, save per accidens as rules of thumb and as de facto social institutions with which the utilitarian has to reckon when estimating consequences. I shall call this doctrine ‘extreme utilitarianism’.
(2) A more modest form of utilitarianism has recently become fashionable. The doctrine is to be found in Toulmin’s book The Place of Reason in Ethics, in Nowell-Smith’s Ethics (though I think Nowell-Smith has qualms), in John Austin’s Lectures on Jurisprudence (Lecture II), and even in J. S. Mill, if Urmson’s interpretation of him is correct (Philosophical Quarterly, Vol. III, pp. 3339, 1953). Part of its charm is that it appears to resolve the dispute in moral philosophy between intuitionists and utilitarians in a way which is very neat. The above philosophers hold, or seem to hold, that moral rules are more than rules of thumb. In general the rightness of an action is not to be tested by evaluating its consequences but only by considering whether or not it falls under a certain rule. Whether the rule is to be considered an acceptable moral rule, is, however, to be  decided by considering the consequences of adopting the rule. Broadly, then, actions are to be tested by rules and rules by consequences. The only cases in which we must test an individual action directly by its consequences are (a) when the action comes under two different rules, one of which enjoins it and one of which forbids it, and (b) when there is no rule whatever that governs the given case. I shall call this doctrine ‘restricted utilitarianism’.
It should be noticed that the distinction I am making cuts across, and is quite different from, the distinction commonly made between hedonistic and ideal utilitarianism. Bentham was an extreme hedonistic utilitarian and Moore an extreme ideal utilitarian, and Toulmin (perhaps) could be classified as a restricted ideal utilitarian. A hedonistic utilitarian holds that the goodness of the consequences of an action is a function only of their pleasurableness and an ideal utilitarian, like Moore, holds that pleasurableness is not even a necessary condition of goodness. Mill seems, if we are to take his remarks about higher and lower pleasures seriously, to be neither a pure hedonistic nor a pure ideal utilitarian. He seems to hold that pleasurableness is a necessary condition for goodness, but that goodness is a function of other qualities of mind as well. Perhaps we can call him a quasi-ideal utilitarian. When we say that a state of mind is good I take it that we are expressing some sort of rational Preference. When we say that it is pleasurable I take it that we are saying that it is enjoyable, and when we say that something is a higher pleasure I take it that we are saying that it is more truly, or more deeply, enjoyable. I am doubtful whether ‘more deeply enjoyable’ does not just mean ‘more enjoyable, even though not more enjoyable on a first look’, and so I am doubtful whether quasi-ideal utilitarianism, and possibly ideal utilitarianism too, would not collapse into hedonistic utilitarianism on a closer scrutiny of the logic of words like ‘preference’, ‘pleasure’, ‘enjoy’, ‘deeply enjoy’, and so on. However, it is beside the point of the present paper to go into these questions. I am here concerned only with the issue between extreme and restricted utilitarianism and am ready to concede that both forms of utilitarianism can be either hedonistic or non-hedonistic.
The issue between extreme and restricted utilitarianism can be illustrated by considering the remark ‘But suppose everyone did the same’. (Cf. A. K. Stout’s article in The Australasian Journal of Philosophy, Vol. 32, pp. r-29.) Stout distinguishes two forms of the universalisation principle, the causal form and the hypothetical form. To say that you ought not to do an action A because it would have bad results if everyone (or many people) did action A may be merely to point out that while the action A would otherwise be the optimific one, nevertheless when you take into account that doing A will probably cause other people to do A too, you can see that A is not, on a broad view, really optimific. If this causal influence could be avoided (as may happen in the case of a secret desert island promise) then we would disregard the universalisation principle. This is the causal  form of the principle. A person who accepted the universalisation principle in its hypothetical form would be one who was. concerned only with what would happen if everyone did the action A: he would be totally unconcerned with the question of whether in fact everyone would do the action A. That is, he might say that it would be wrong not to vote because it would have bad results of everyone took this attitude, and he would be totally unmoved by arguments purporting to show that my refusing to vote has no effect whatever on other people’s propensity to vote. Making use of Stout’s distinction, we can say that an extreme utilitarian would apply the universalisation principle in the causal form, while a restricted utilitarian would apply it in the hypothetical form.
How are we to decide the issue between extreme and restricted utilitarianism? I wish to repudiate at the outset that milk and water approach which describes itself sometimes as ‘investigating what is implicit in the common moral consciousness’ and sometimes as ‘investigating how people ordinarily talk about morality’. We have only to read the newspaper correspondence about capital punishment or about what should be done with Formosa to realise that the common moral consciousness is in part made up of superstitious elements, of morally bad elements, and of logically confused elements. I address myself to good hearted and benevolent people and so I hope that if we rid ourselves of the logical confusion, the superstitious and morally bad elements will largely fall away. For even among good hearted and benevolent people it is possible to find superstitious and morally bad reasons for moral beliefs. These superstitious and morally bad reasons hide behind the protective screen of logical confusion. With people who are not logically confused but who are openly superstitious or morally bad I can of course do nothing. That is, our ultimate pro-attitudes may be different. Nevertheless I propose to rely on my own moral consciousness and to appeal to your moral consciousness and to forget about what people ordinarily say. ‘The obligation to obey a rule’, says Nowell-Smith (Ethics, p. 739), ‘does not, in the opinion of ordinary men’, (my italics), ‘rest on the beneficial consequences of obeying it in a particular case’. What does this prove? Surely it is more than likely that ordinary men are confused here. Philosophers should be able to examine the question more rationally.
For an extreme utilitarian moral rules are rules of thumb. In practice the extreme utilitarian will mostly guide his conduct by appealing to the rules (‘do not lie’, ‘do not break promises’, etc.) of common sense morality. This is not because there is anything sacrosanct in the rules themselves but because he can argue that probably he will most often act in an extreme utilitarian way if he does not think as a utilitarian. For one thing, actions have frequently to be done in a hurry. Imagine a man seeing a person drowning. He jumps in and rescues him. There is no time to reason the  matter out, but usually this will be the course of action which an extreme utilitarian would recommend if he did reason the matter out. If, however, the man drowning had been drowning in a river near Berchtesgaden in 1938, and if he had had the well known black forelock and moustache of Adolf Hitler, an extreme utilitarian would, if he had time, work out the probability of the man’s being the villainous dictator, and if the probability were high enough he would, on extreme utilitarian grounds, leave him to drown. The rescuer, however, has not time. He trusts to his instincts and dives in and rescues the man. And this trusting to instincts and to moral rules can be justified on extreme utilitarian grounds. Furthermore, an extreme utilitarian who knew that the drowning man was Hitler would nevertheless praise the rescuer, not condemn him. For by praising the man he is strengthening a courageous and benevolent disposition of mind, and in general this disposition has great positive utility. (Next time, perhaps, it will be Winston Churchill that the man saves!) We must never forget that an extreme utilitarian may praise actions which he knows to be wrong. Saving Hitler was wrong, but it was a member of a class of actions which are generally right, and the motive to do actions of this class is in general an optimific one. In considering questions of praise and blame it is not the expediency of the praised or blamed action that is at issue, but the expediency of the praise. It can be expedient to praise an inexpedient action and inexpedient to praise an expedient one.
Lack of time is not the only reason why an extreme utilitarian may, on extreme utilitarian principles, trust to rules of common sense morality. He knows that in particular cases where his own interests are involved his calculations are likely to be biased in his own favour. Suppose that he is unhappily married and is deciding whether to get divorced. He will in all probability greatly exaggerate his own unhappiness (and possibly his wife’s) and greatly underestimate the harm done to his children by the break up of the family. He will probably also underestimate the likely harm done by the weakening of the general faith in marriage vows. So probably he will come to the correct extreme utilitarian conclusion if he does not in this instance think as an extreme utilitarian but trusts to common sense morality.
There are many more and subtle points that could be made in connection with the relation between extreme utilitarianism and the morality of common sense. All those that I have just made and many more will be found in Book IV Chapters 3-5of Sidgwick’s Methods of Ethics. I think that this book is the best book ever written on ethics, and that these chapters are the best chapters of the book. As they occur so near the end of a very long book they are unduly neglected. I refer the reader, then, to Sidgwick for the classical exposition of the relation between (extreme) utilitarianism and the morality of common sense. One further point raised by Sidgwick in this connection is whether an (extreme) utilitarian ought on (extreme) utilitarian principles to propagate (extreme) utilitarianism among the  public. As most people are not very philosophical and not good at empirical calculations, it is probable that they will most often act in an extreme utilitarian way if they do not try to think as extreme utilitarians. We have seen how easy it would be to misapply the extreme utilitarian criterion in the case of divorce. Sidgwick seems to think it quite probable that an extreme utilitarian should not propagate his doctrine too widely. However, the great danger to humanity comes nowadays on the plane of public morality—not private morality. There is a greater danger to humanity from the hydrogen bomb than from an increase of the divorce rate, regrettable though that might be, and there seems no doubt that extreme utilitarianism makes for good sense in international relations. When France walked out of the United Nations because she did not wish Morocco discussed, she said that she was within her rights because Morocco and Algiers are part of her metropolitan territory and nothing to do with U.N.. This was clearly a legalistic if not superstitious argument. We should not be concerned with the so-called ‘rights’ of France or any other country but with whether the cause of humanity would best be served by discussing Morocco in U.N. (I am not saying that the answer to this is ‘Yes’. There are good grounds for supposing that more harm than good would come by such a discussion.) I myself have no hesitation in saying that on extreme utilitarian principles we ought to propagate extreme utilitarianism as widely as possible. But Sidgwick had respectable reasons for suspecting the opposite.
The extreme utilitarian, then, regards moral rules as rules of thumb and as sociological facts that have to be taken into account when deciding what to do, just as facts of any other sort have to be taken into account. But in themselves they do not justify any action.
The restricted utilitarian regards moral rules as more than rules of thumb for short-circuiting calculations of consequences. Generally, he argues, consequences are not relevant at all when we are deciding what to do in a particular case. In general, they are relevant only to deciding what rules are good reasons for acting in a certain way in particular cases. This doctrine is possibly a good account of how the modern unreflective twentieth century Englishman often thinks about morality, but surely it is monstrous as an account of how it is most rational to think about morality. Suppose that there is a rule R and that in 99% of cases the best possible results are obtained by acting in accordance with R. Then clearly R is a useful rule of thumb; if we have not time or are not impartial enough to assess the consequences of an action it is an extremely good bet that the thing to do is to act in accordance with R. But is it not monstrous to suppose that if we have worked out the consequences and if we have perfect faith in the impartiality of our calculations, and if we know that in this instance to break R will have better results than to keep it, we should nevertheless obey the rule? Is it not to erect R into a sort of idol if we keep it when breaking it will prevent, say, some avoidable misery? Is not this a form of superstitious rule-worship (easily explicable psychologically) and not the rational thought of a philosopher?
The point may be made more clearly if we consider Mill’s comparison of moral rules to the tables in the nautical almanack. (Utilitarianism, Everyman edition, pp. 22-23). This comparison of Mill’s is adduced by Urmson as evidence that Mill was a restricted utilitarian, but I do not think that it will bear this interpretation at all. (Though I quite agree with Urmson that many other things said by Mill are in harmony with restricted rather than extreme utilitarianism. Probably Mill had never thought very much about the distinction and was arguing for utilitarianism, restricted or extreme, against other and quite nonutilitarian forms of moral argument.) Mill says: ‘Nobody argues that the art of navigation is not founded on astronomy, because sailors cannot wait to calculate the Nautical Almanack. Being rational creatures, they go out upon the sea of life with their minds made up on the common questions of right and wrong, as well as on many of the far more difficult questions of wise and foolish. . . . Whatever we adopt as the fundamental principle of morality, we require subordinate principles to apply it by’. Notice that this is, as it stands, only an argument for subordinate principles as rules of thumb. The example of the nautical almanack is misleading because the information given in the almanack is in all cases the same as the information one would get if one made a long and laborious calculation from the original astronomical data on which the almanack is founded. Suppose, however, that astronomy were different. Suppose that the behaviour of the sun, moon and planets was very nearly as it is now, but that on rare occasions there were peculiar irregularities and discontinuities, so that the almanack gave us rules of the form `in 99% of cases where the observations are such and such you can deduce that your position is so and so’. Furthermore, let us suppose that there were methods which enabled us, by direct and laborious calculation from the original astronomical data, not using the rough and ready tables of the almanack, to get our correct position in 100% of cases. Seafarers might use the almanack because they never had time for the long calculations and they were content with a 99% chance of success in calculating their positions. Would it not be absurd, however, if they did make the direct calculation, and finding that it disagreed with the almanack calculation, nevertheless they ignored it and stuck to the almanack conclusion? Of course the case would be altered if there were a high enough probability of making slips in the direct calculation: then we might stick to the almanack result, liable to error though we knew it to be, simply because the direct calculation would be open to error for a different reason, the fallibility of the computer. This would be analogous to the case of the extreme utilitarian who abides by the conventional rule against the dictates of his utilitarian calculations simply because he thinks that his calculations are probably affected by personal bias. But if the navigator were sure of his direct calculations  would he not be foolish to abide by his almanack? I conclude, then, that if we change our suppositions about astronomy and the almanack (to which there are no exceptions) to bring the case into line with that of morality (to whose rules there are exceptions), Mill’s example loses its appearance of supporting the restricted form of utilitarianism. Let me say once more that I am not here concerned with how ordinary men think about morality but with how they ought to think. We could quite well imagine a race of sailors who acquired a superstitious reverence for their almanack, even though it was only right in 99% of cases, and who indignantly threw overboard any man who mentioned the possibility of a direct calculation. But would this behaviour of the sailors be rational?
Let us consider a much discussed sort of case in which the extreme utilitarian might go against the conventional moral rule. I have promised to a friend, dying on a desert island from which I am subsequently rescued, that I will see that his fortune (over which I have control) is given to a jockey club. However, when I am rescued I decide that it would be better to give the money to a hospital, which can do more good with it. It may be argued that I am wrong to give the money to the hospital. But why?(a) The hospital can do more good with the money than the jockey club can. (b) The present case is unlike most cases of promising in that no one except me knows about the promise. In breaking the promise I am doing so with complete secrecy and am doing nothing to weaken the general faith in promises. That is, a factor, which would normally keep the extreme utilitarian from promise breaking even in otherwise unoptimific cases, does not at present operate. (c) There is no doubt a slight weakening in my own character as an habitual promise keeper, and moreover psychological tensions will be set up in me every time I am asked what the man made me promise him to do. For clearly I shall have to say that he made me promise to give the money to the hospital, and, since I am an habitual truth teller, this will go very much against the grain with me. Indeed I am pretty sure that in practice I myself would keep the promise. But we are not discussing what my moral habits would probably make me do; we are discussing what I ought to do. Moreover, we must not forget that even if it would be most rational of me to give the money to the hospital it would also be most rational of you to punish or condemn me if you did, most improbably, find out the truth (e.g. by finding a note washed ashore in a bottle). Furthermore, I would agree that though it was most rational of me to give the money to the hospital it would be most rational of you to condemn me for it. We revert again to Sidgwick’s distinction between the utility of the action and the utility of the praise of it.
Many such issues are discussed by A. K. Stout in the article to which I have already referred. I do not wish to go over the same ground again, especially as I think that Stout’s arguments support my own point of view. It will be useful, however, to consider one other example that he gives. Suppose that during hot weather there is an edict that no water must be  used for watering gardens. I have a garden and I reason that most people are sure to obey the edict, and that as the amount of water that I use will be by itself negligible no harm will be done if I use the water secretly. So I do use the water, thus producing some lovely flowers which give happiness to various people. Still, you may say, though the action was perhaps optimific, it was unfair and wrong.
There are several matters to consider. Certainly my action should be condemned. We revert once more to Sidgwick’s distinction. A right action may be rationally condemned. Furthermore, this sort of offence is normally found out. If I have a wonderful garden when everybody else’s is dry and brown there is only one explanation. So if I water my garden I am weakening my respect for law and order, and as this leads to bad results an extreme utilitarian would agree that I was wrong to water the garden. Suppose now that the case is altered and that I can keep the thing secret: there is a secluded part of the garden where I grow flowers which I give away anonymously to a home for old ladies. Are you still so sure that I did the wrong thing by watering my garden? However, this is still a weaker case than that of the hospital and the jockey club. There will be tensions set up within myself: my secret knowledge that I have broken the rule will make it hard for me to exhort others to keep the rule. These psychological ill effects in myself may be not inconsiderable: directly and indirectly they may lead to harm which is at least of the same order as the happiness that the old ladies get from the flowers. You can see that on an extreme utilitarian view there are two sides to the question.
So far I have been considering the duty of an extreme utilitarian in a predominantly non-utilitarian society. The case is altered if we consider the extreme utilitarian who lives in a society every member, or most members, of which can be expected to reason as he does. Should he water his flowers now? (Granting, what is doubtful, that in the case already considered he would have been right to water his flowers.) Clearly not. A simple argument, employing the game-theoretical concept of a mixed strategy, suggests that each extreme utilitarian should give himself a very small probability (say by tossing dice) of watering his garden. Suppose that there are m potential garden waterers and that f(n) is the damage done by exactly n people watering their gardens. Now if each of them gives himself a probability p of watering his garden it is easy to calculate, in terms of p, the probabilities p1, p2, . . . pm of 1, 2, . . . m persons respectively watering their gardens. Let a be the benefit to each gardener of watering his garden. Then if V is the total probable benefit to the community of gardeners we have
V = p1(a - f(1) + p2(2a - f(2))+ ...+ pm(ma - f(m))
Assuming that numerical values can be given to a and to values of the function f (n) we calculate the value of P for which dV/dp = 0. This gives the value of p which maximises the total probable benefit. In practical cases of course numerical values of f(n) and a cannot be determined, but a good approximation can usually be got by taking p as equal to zero. However the mathematical analysis is of theoretical interest for the discussion of utilitarianism. Too many writers mistakenly suppose that the only two relevant alternatives are that no one does something and that everyone does it. 
I now pass on to a type of case which may be thought to be the trump card of restricted utilitarianism. Consider the rule of the road. It may be said that since all that matters is that everyone should do the same it is indifferent which rule we have, ‘go on the left hand side’ or ‘go on the right hand side’. Hence the only reason for going on the left hand side in British countries is that this is the rule. Here the rule does seem to be a reason, in itself, for acting in a certain way. I wish to argue against this. The rule in itself is not a reason for our actions. We would be perfectly justified in going on the right hand side if (a)we knew that the rule was to go on the left hand side, and (b) we were in a country peopled by super-anarchists who always on principle did the opposite of what they were told. This shows that the rule does not give us a reason for acting so much as an indication of the probable actions of others, which helps us to find out what would be our own most rational course of action. If we are in a country not peopled by anarchists, but by nonanarchist extreme Utilitarians, we expect, other things being equal, that they will keep rules laid down for them. Knowledge of the rule enables us to predict their behaviour and to harmonise our own actions with theirs. The rule ‘keep to the left hand side’, then, is not a logical reason for action but an anthropological datum for planning actions.
I conclude that in every case if there is a rule R the keeping of which is in general optimific, but such that in a special sort of circumstances the optimific behaviour is to break R, then in these circumstances we should break R. Of course we must consider all the less obvious effects of breaking R, such as reducing people’s faith in the moral order, before coming to the conclusion that to break R is right: in fact we shall rarely come to such a conclusion. Moral rules, on the extreme utilitarian view, are rules of thumb only, but they are not bad rules of thumb. But if we do come to the conclusion that we should break the rule and if we have weighed in the balance our own fallibility and liability to personal bias, what good reason remains for keeping the rule? I can understand ‘it is optimific’ as a reason for action, but why should ‘it is a member of a class of actions which are usually optimific’ or ‘it is a member of a class of actions which as a class are more optimific than any alternative general class’ be a good reason? You might as well say that a person ought to be picked to play for Australia just because all his brothers have been, or that the Australian team should be composed entirely of the Harvey family because this would be better than composing it entirely of any other family. The extreme utilitarian does not appeal to artificial feelings, but only to our feelings of benevolence, and what better feelings can there be to appeal to? Admittedly we can have a pro-attitude to anything, even to rules, but such artificially begotten pro-attitudes smack of superstition. Let us get down to realities, human happiness and misery, and make these the objects of our pro-attitudes and anti-attitudes.
The restricted utilitarian might say he is talking only of morality, not of such things as rules of the road. I am not sure how far this objection, if valid, would affect my argument, but in any case I would reply that as a philosopher I conceive of ethics as the study of how it would be most rational to act. If my opponent wishes to restrict the word ‘morality’ to a narrower use he can have the word. The fundamental question is the question of rationality of action in general. Similarly if the restricted utilitarian were to appeal to ordinary usage and say ‘it might be most rational to leave Hitler to drown but it would surely not be wrong to rescue him’, I should again let him have the words ‘right’ and ‘wrong’ and should stick to ‘rational’ and ‘irrational’. We already saw that it would be rational to praise Hitler’s rescuer, even though it would have been most rational not to have rescued Hitler. In ordinary language, no doubt, ‘right’ and ‘wrong’ have not only the meaning ‘most rational to do’ and ‘not most rational to do’ but also have the meaning ‘praiseworthy’ and ‘not praiseworthy’. Usually to the utility of an action corresponds utility of praise of it, but as we saw, this is not always so. Moral language could thus do with tidying up, for example by reserving ‘right’ for ‘most rational’  and ‘good’ as an epithet of praise for the motive from which the action sprang. It would be more becoming in a philosopher to try to iron out illogicalities in moral language and to make suggestions for its reform than to use it as a court of appeal whereby to perpetuate confusions.
One last defence of restricted utilitarianism might be as follows. ‘Act optimifically’ might be regarded as itself one of the rules of our system (though it would be odd to say that this rule was justified by its optimificality). According to Toulmin (The Place of Reason in Ethics, pp. 146-8) if ‘keep promises’, say, conflicts with another rule we are allowed to argue the case on its merits, as if we were extreme utilitarians. If ‘act optimifically’ is itself one of our rules then there will always be a conflict of rules whenever to keep a rule is not itself optimific. If this is so, restricted utilitarianism collapses into extreme utilitarianism. And no one could read Toulmin’s book or Urmson’s article on Mill without thinking that Toulmin and Urmson are of the opinion that they have thought of a doctrine which does not collapse into extreme utilitarianism, but which is, on the contrary, an improvement on it.
1 The Philosophical Quarterly, Vol. VI (1956), 344-354. Page breaks are indicated in brackets. Based on a paper read to the Victorian Branch of the Australasian Association of Psychology and Philosophy, October, 1955. Since publication of this article Professor Smart has preferred to follow Brandt's terminology of “act-“ and “rule-utilitarianism” instead of “extreme” and “restricted utilitarianism.”