R. M. Hare, The Language of Morals (1952)

3
INFERENCE

3.1. The rule that an imperative cannot appear in the conclusion of a valid inference, unless there is at least one imperative in the premisses, may be confirmed by an appeal to general logical considerations. For it is now generally regarded as true by definition that (to speak roughly at first) nothing can appear in the conclusion of a valid deductive inference which is not, from their very meaning, implicit in the conjunction of the premisses. It follows that, if there is an imperative in the conclusion, not only must some imperative appear in the premisses, but that very imperative must be itself implicit in them.

Since these considerations have a wide bearing on moral philosophy, it will be as well to explain them in greater detail. Few people now think, as Descartes seems to have done, that we can arrive at scientific conclusions about matters of empirical fact, like the circulation of the blood, by deductive reasoning from self-evident first principles.1 The work of Wittgenstein and others has to a great extent made clear the reasons for the impossibility of doing this. It has been argued, convincingly in my opinion, that all deductive inference is analytic in character; that is to say, that the function of a deductive inference is not to get from the premisses 'something further' not implicit in them (even if that is what Aristotle meant (2.4)), but to make explicit what was implicit in the conjunction of the premisses. This has been shown to follow from the very nature of language; for to say anything we have, as we have already noticed, to obey some rules, and these rules -- especially but not only the rules for the use of the so-called logical words -- mean, firstly, that to say what is in the premisses of a valid inference is to say, at least, what is in the conclusion, and, secondly, that if anything is said in the conclusion which is not said, implicitly or explicitly, in the premisses, the inference is invalid. We cannot be said to understand fully the meaning of premisses and conclusion unless we admit the validity of the inference. Thus, if someone professed to admit that all men were mortal and that Socrates was a man, but refused to admit that Socrates was mortal, the correct thing to do would be not, as has sometimes been suggested, to accuse him of some kind of logical pur-blindness, but to say 'You evidently don't know the meaning of the word "all"; for if you did you would eo ipso know how to make inferences of this sort'.

3.2. The principle just set out is not, however, quite general enough to cover all cases. For example, 'x = 2' entails 'x2 = 4'; but it is not natural to say that in the latter expression nothing is said which is not said implicitly in the former; for the latter contains the 'squared' symbol, and in order to understand 'x = 2' we do not have to know anything about the meaning of this symbol. We have, therefore, to say that there must be nothing said in the conclusion which is not said implicitly or explicitly in the premisses, except what can be added solely on the strength of definitions of terms. This qualification is important for the logic of imperatives; for, as I have already warned the reader, there is one kind of imperative conclusion which can be entailed by a set of purely indicative premisses. This is the so-called 'hypothetical' imperative. It must be pointed out that not all imperatives containing an hypothetical clause are 'hypothetical' in this sense. For example, the sentence 'If any statement is untrue, do not make it', is not 'hypothetical' as the expression 'hypothetical imperative' is traditionally used. What a 'hypothetical' imperative is, is best made clear by examples. The subject is so difficult that I cannot deal with it very fully; but some explanation is necessary.

Consider the following sentence:

If you want to go to the largest grocer in Oxford, go to Grimbly Hughes.
This seems to follow from, and to say no more than:
Grimbly Hughes is the largest grocer in Oxford.
The first matter that requires elucidation is the status of the word 'want'. It does not mean the same as 'be affected by a recognizable state of the feelings known as desire'. If I were the superior of a religious order whose rule ordained the complete abnegation of all desires, I could not say to a novice 'If you have a desire to go to the largest grocer in Oxford, go to Grimbly Hughes', for this would be contrary to the rule. But I might very well say 'If you want to go to the largest grocer in Oxford, go to Grimbly Hughes'; for this would simply be intended to convey a piece of information that the largest grocer is Grimbly Hughes. 'Want' is here a logical term, and stands, as we shall see, for an imperative inside a subordinate clause. This is but one of the many puzzles generated by treating sentences compounded with the word 'want'as if they were always descriptive of mental states (1.3).

Now compare the following sentence :

If all mules are barren, then this animal is barren.
This is entailed by the sentence 'This animal is a mule'. We only have to know the meanings of'aH' and the other words used in order to make the inference. We must notice that this inference is valid because another simpler one is valid, namely:
All mules are barren.
This animal is a mule.
∴ This animal is barren.
The more complex form of the inference is arrived at by taking away the major premiss from its proper place, and adding it to the conclusion inside an hypothetical clause.

The following inference can also be treated in the same way:

Go to the largest grocer in Oxford.
Grimbly Hughes is the largest grocer in Oxford.
∴ Go to Grimbly Hughes.
It then becomes:
Grimbly Hughes is the largest grocer in Oxford.
∴ If go to the largest grocer in Oxford, go to Grimbly Hughes.
In English, we write this conclusion in the form:
If you want to go to the largest grocer in Oxford, go to Grimbly Hughes.
We only have to know the meanings of 'want' and the other words used in the conclusion (including the imperative verb-form) in order to make this inference.

Another example would be 'If you want to break your springs, go on driving as you are at the moment'. Here the full inference would be:

Do whatever will conduce to breaking your springs.
Going on driving as you are at the moment will conduce, &c.
∴ Go on driving as you are at the moment.
The speaker of our example, in order to draw the hearer's attention in an emphatic way to the truth of the minor premiss, points out that his present style of driving would be a valid conclusion from this minor premiss and a major premiss which the hearer obviously does not accept. Into this example, the notion of 'means conducive to an end' enters; but the first example shows that it need not be present. Other related forms of expression are:
To stop the train pull down the chain.
Drive slowly or you'll have a collision.
Omit regular lubrication and halve the life of your car.
There is a marked contrast between these three; the first is neutral on the subject of whether the chain is actually to be pulled or not; that is why 'Penalty for improper use £5' has to be added. The second is not neutral; it has a strong flavour of the simple, non-hypothetical imperative 'Drive slowly', and 'or' might be replaced by 'because if you don't'. The third is an oddity; like 'If you want to break your springs, go on driving as you are at the moment', it is ironic and actually has the purpose of dissenting from the clause 'Omit regular lubrication'. It is adapted, with the omission of a trade name, from an actual advertisement.

To the extent that an imperative is hypothetical, it has descriptive force in much the same way as a value-judgement may (7.1). Understanding or supplying the hypothetical clause is like knowing the standard of values that is being applied. It is not easy to say, in any individual case, in which there is no 'if'-clause actually included, to what extent the imperative is to be treated as hypothetical. We must not assume that all non-moral imperatives are hypothetical, for this is far from being true. Operating instructions for machinery form an interesting borderline case. Are we to say that 'Plug in to a supply of the voltage indicated on the label!' is hypothetical, and that we have to understand 'if you want your vacuum cleaner to clean your carpets without necessitating expensive repairs'? It is hard to answer this question; we could certainly understand and obey the instruction without knowing what its purpose was. This case shows, not that there is no difference between hypothetical and non-hypothetical imperatives, but that the line is hard to draw.

It would probably be misleading to say that hypothetical imperatives are 'really indicatives'. They have indeed descriptive force, and are entailed by indicatives; but 'x2 = 4' is entailed by 'x = 2', and yet we should not say that the former was not really a quadratic equation. It would not, for one thing, be intelligible to someone who did not know the meaning of the 'squared' symbol. This symbol, moreover, does not have here a special meaning different from its other uses. In somewhat the same way, 'If you want to go to the largest grocer in Oxford, go to Grimbly Hughes' is not an indicative; it would not be intelligible to someone who had learnt the meaning of indicative verb-forms but not that of imperative verb-forms; and the latter do not have in it a special meaning. The best way of describing the matter has been suggested by Kant: the imperative element in a hypothetical imperative is analytic ('Who wills the end . . . wills also the means'), because the imperatives in the two parts, so to say, cancel one another out. It is an imperative, but, qua imperative, has no content; the content which it has is that of the indicative minor premiss from which it is derived.2

Two suggestions may be made for further researches into the subject of hypothetical imperatives which cannot be pursued here. The first is that the 'if in them has a somewhat different logical status from that which it has in sentences like 'If any statement is untrue, do not make it'. If the latter sentence were analysed into phrastic and neustic, it seems to me that the 'if' would go into the phrastic; the whole sentence might be rendered thus:

In the event of any statement being untrue, your not making it, please,
or thus:
Your not making untrue statements, please.
But in a hypothetical imperative proper the 'if'-clause itself contains an imperative neustic, concealed in the word 'want'. I am not sure yet how these sentences would be best analysed, and am inclined to think that different ones should be analysed in different ways according to how fully 'hypothetical' they are. If the categorical element is completely submerged, as in 'If you want to break your springs, go on driving as you are at present', a metalinguistic analysis is tempting:
The command 'Go on driving as you are at present' could be inferred from a minor factual premiss, which is true, and the major premiss (sc. to which you obviously do not assent) 'Do whatever will conduce to breaking your springs'.
But this problem is part of the wider problem, still very dark, of the analysis of hypothetical sentences in general.

The second suggestion is, that the relation between hypothetical imperatives and the descriptive element in the meaning of value-judgements would repay much further study. The suggestion just made, that some hypothetical imperatives might be analysed metalinguistically, clearly has a bearing on what I shall later call the 'inverted-commas' use of value-judgements (7.5). It may safely be predicted that hypothetical imperatives will be found to be as subtle, flexible, and various in their logic as are the descriptive uses of value-words.

3.3. But let us leave this difficult matter and return to Descartes. The considerations about inference which I summarized at the beginning of this chapter mean that a Cartesian procedure, either in science or in morals, is doomed from the very start. If any science is intended to give us conclusions of substance about matters of fact, then, if its method is deductive, these conclusions must be implicit in the premisses. This means that, before we fully understand the meanings of our Cartesian first principles, we have to know that they (with the addition only of definitions of terms) entail such various propositions as, that all mules are barren, or that a man's heart is on the left-hand side of his body, or that the sun is so many miles from the earth. But if all these facts are implicit in the first principles, the latter can hardly be called self-evident. We find out about facts like these at least in part by observation; no amount of reasoning from axioms will take its place. The position of pure mathematics has been much discussed and is still obscure; it seems best to regard the axioms of pure mathematics and logic as definitive of the terms used in them. But this much, at any rate, may be said, that if a science purports to tell us facts like the above, it cannot, like pure mathematics, be based on deductive reasoning and nothing else. It was the mistake of Descartes to assimilate to pure mathematics studies which are of a wholly different character.

It does not follow, from the fact that deduction, whether in the form of pure mathematics or of logic, cannot take the place of observation, that deduction is therefore altogether useless as an adjunct to observation. Science makes use of expressions which would be altogether meaningless unless we could deduce. The sentence 'There are three gramme weights on the balance and no more' would be meaningless to anyone who could not deduce from it 'There is one gramme weight on the balance, and one other, and one other, and no more', and vice versa.

The same considerations hold good in ethics. Many of the ethical theories which have been proposed in the past may without injustice be called 'Cartesian' in character; that is to say, they try to deduce particular duties from some self-evident first principle. Often factual observations are admitted among the premisses; but this, though it makes the theories which admit them incompletely 'Cartesian', does not affect my argument. A Cartesian procedure in morals is as illusory as it is in science. If we may take it that, as I shall show later, a piece of genuinely evaluative moral reasoning must have as its end-product an imperative of the form 'Do so-and-so', it follows that its principles must be of such a kind that we can deduce such particular imperatives from them, in conjunction with factual minor premisses. If, for example, a moral system is to enjoin me not to say this particular thing which is false, its principles must contain implicitly or explicitly an imperative to the effect that what is false is not to be said in circumstances like those in which I now am. And, similarly, they must contain other imperatives such as will regulate my conduct in all manner of circumstances, both foreseen and unforeseen. But it is obvious that such a set of principles could not possibly be self-evident. It is not easier, but more difficult, to assent to a very general command like 'Never say what is false' than it is to assent to the particular command 'Do not say this particular thing which is false', just as it is more difficult and dangerous to adopt the hypothesis that all mules are barren than to acknowledge the undoubted fact that this mule which has just died has had no progeny. A decision never to say what is false involves a decision in advance about a very great number of individual cases, with only the information about them that they are all cases of saying what is false. It is not, surely, casuistry of an objectionable kind to want to avoid committing ourselves in this fashion. It is quite true that, when we have had experience of making such decisions, we may eventually find ourselves able to accept the general principle. But suppose that we were faced, for the first time, with the question 'Shall I now say what is false?' and had no past decisions, either of our own or of other people, to guide us. How should we then decide the question? Not, surely, by inference from a self-evident general principle, 'Never say what is false'; for if we could not decide even whether to say what was false in these particular circumstances, how could we possibly decide whether to say what was false in innumerable circumstances whose details were totally unknown to us, save in this respect, that they were all cases of saying what was false?

The same point may be put in another way. It is an established principle of logic that if one proposition entails another, then the negation of the second entails the negation of the first. An analogous principle, somewhat stronger, is also valid, that if I know that one proposition entails another, to be in doubt about assenting to the second is eo ipso to be in doubt about assenting to the first. For instance, if I know that the proposition 'All mules are barren and this is a mule' entails the proposition 'This (mule) is barren', it follows that if I am in doubt about assenting to the proposition 'This (mule) is barren', I must be in doubt about assenting to the proposition 'All mules are barren and this is a mule'; and this means that I must be in doubt about either 'All mules are barren' or 'This is a mule'. Now if we apply an exactly parallel reasoning to our case about saying what is false, we get the following result. Since I am in doubt, ex hypothesi, whether or not to make this false statement, I must be in doubt about assenting to the command 'Do not make this statement'. But if I am in doubt about this command, I must eo ipso be in doubt, either about the factual premiss 'This statement is false' (and this alternative is ruled out ex hypothesi), or else, as must be the case, about the imperative premiss 'Never say what is false'. It follows that no general principle can be self-evident which is to be of assistance in deciding particular questions about which we are in doubt.

The impossibility of a 'Cartesian' moral system may be shown in another way, closely akin to the one just explained. It is not in the least clear what could be meant by calling any proposition, least of all a general principle of conduct, 'self-evident'. If such a principle is to be in some sense impossible to reject, this, it seems to me, can only be for one of two reasons. First, it might be said that a principle of conduct was impossible to reject, if it were self-contradictory to reject it. But if it is self-contradictory to reject a principle, this can only be because the principle is analytic. But if it is analytic, it cannot have any content; it cannot tell me to do one thing rather than another. The term 'analytic', which we shall have occasion to use a good deal, may be defined with sufficient precision as follows:

A sentence is analytic if, and only if, either (1) the fact that a person dissents from it is a sufficient criterion for saying that he has misunderstood the speakers meaning or (2) it is entailed by sentence wich is analytic in sense (1).

A sentence which is not analytic or self-contradictory is called synthetic.

These definitions are of course not exact; a full discussion of the meanings of 'analytic' and 'synthetic' is outside the scope of this book.

Secondly, it might be suggested that a principle of conduct might be impossible to reject, in the sense that its rejection was a psychological impossibility. But what is or is not a psychological impossibility is a contingent matter; it may be a psychological impossibility for me to reject a principle which the more hardened or sophisticated have no difficulty in discarding. We could never have any justification for asserting that no one could ever reject a principle, unless that principle were analytic. Moreover, the psychological impossibility of rejecting a principle would be a fact about the constitution of people's psyches; and from a fact, or the indicative sentence recording it, no imperative can be derived.

A third kind of interpretation is sometimes canvassed, which rests upon the introduction of a value-word. It might be suggested that, though a principle was both logically and psychologically possible to reject, it might be not rational to reject it (it might be impossible for a rational person to reject it). Sometimes instead of 'rational' we have other expressions, such as 'a morally developed or morally educated person' or 'a competent and impartial judge'. These are all value-expressions. We therefore have to ask 'What could be the criterion for deciding whether a person falls into one or other of these classes?' Clearly we cannot say that the rejection of the principle is itself evidence that the person who rejects it is not qualified in these ways; for in this case our criterion of self-evidence would be circular. There must therefore be some other means of finding out whether a person is rational But the question whether a person is rational must be either a factual question or a question of value (or a combination of the two). If it is a purely factual question, then we cannot get imperative conclusions out of factual premisses such as 'So-and-so is rational' and 'So-and-so finds it impossible to reject the principle that. . .'. But if it is wholly or partly a question of value, then either the answer to it is self-evident in some sense (in which case again our criterion of self-evidence would be circular), or else we have at least one constituent in our reasoning which is neither factual nor self-evident. This third possibility, therefore, must be ruled out.

It follows from these considerations that if it is the function of general moral principles to regulate our conduct, i.e. to entail, in conjunction with indicative minor premisses, answers to questions of the form 'Shall I or shall I not do this particular thing?' then these general moral principles cannot be self-evident. If this view of the function of moral principles were accepted (and I shall later give reasons for doing so), it would provide a conclusive refutation of a large number of ethical theories. Suppose, for example, that we find a philosopher telling us that it is self-evident that we ought always to do what our conscience tells us; we must answer that, since we are often in doubt whether or not to do some act which our conscience tells us to do, this general principle cannot be self-evident. And even if it were the case that we never were in doubt on this point, this would be merely a fact about our psyches, and no imperative conclusions would follow from it. In the example chosen, 'conscience' must, of course, be taken as the name of an identifiable psychological occurrence. If it is made a value-question, whether a certain psychological occurrence is really conscience or the Devil assuming the voice of conscience, the principle clearly comes within the scope of the next paragraph.

Ethical theories of this general type usually conceal their fallacious character by a device of which a brief mention may be made here, although it cannot be fully understood until we have discussed the logic of value-words. If the general principle advocated contains a value-word, it may be made to appear self-evident, by being treated as analytic; and then, when the same value-word appears in the factual minor premiss, it may be treated as if it were descriptive. For example, we might assert the self-evidence (because analytic) of the principle that we ought to do our duty; and then we might argue that we could ascertain what our duty was by some fact-finding process (e.g. by consulting a faculty called a sense of duty, or else by seeing to what kinds of act the word 'duty' was applied in our society, and then calling these acts 'duties'). From this argument it would appear that we could get to a conclusion 'I ought to do a particular act A' and thence to the imperative 'Do A', simply on the basis of two premisses 'One ought to do one's duty' and 'A is my duty', the first of which is self-evident and the second factual. But this is an equivocation. If 'duty' is a value-word, then we cannot decide what is our duty merely by consulting word-usage or by seeing whether we have a certain psychological reaction, but only by making a moral decision. On the other hand, if 'duty' were not treated as a value-word, but regarded as meaning either 'that towards which I have a certain recognizable psychological reaction' or 'that to which the name "duty" is commonly applied in my society', then the principle 'One ought always to do one's duty' would not be self-evident.

3.4. The upshot of all this is rather alarming. I gave in the preceding chapter reasons for holding that no moral system whose principles were regarded as purely factual could fulfil its function of regulating our conduct. In this chapter I have shown that no moral system which claims to be based on principles which are self-evident can fulfil this function either. These two contentions between them, if they are accepted, dispose of nearly all of what Hume calls 'the vulgar systems of morality'. Most ethical writers who have seemed plausible to those who studied them superficially, can be shown to suffer from one or other of these defects. A few great writers, such as Aristotle, Hume, and Kant, though it is not difficult to find here and there in their works traces of these defects, can yet, if studied in the right way, be seen to avoid them in their main doctrines. But it is not surprising that the first effect of modern logical researches was to make some philosophers despair of morals as a rational activity.

It is the purpose of this book to show that their despair was premature. But the effects of the above argument are so catastrophic that it may well be asked 'Have you not from the start rendered the problem impossible of solution ? Is there not some flaw in your argument, some dichotomy too stringently pressed, some criterion interpreted too exactingly; can we not save something from the destruction by being a little less rigorous?' In particular, exception will certainly be taken to my use of the word 'entail'. It may be maintained that although, in the strict sense of the word, I have indeed shown that moral judgements and imperatives cannot be entailed by factual premisses, yet there is some looser relation than entailment which holds between them. Mr. S. E. Toulmin, for example, talks of:

an ethical argument, consisting partly of logical (demonstrative) inferences, partly of scientific (inductive) inferences, and partly of that form of inference peculiar to ethical arguments, by which we pass from factual reasons to an ethical conclusion -- what we may naturally call 'evaluative' inference.3
Since I have elsewhere, in a review of Toulmin's book,4 discussed his particular version of this doctrine, which avoids the crudest of the errors to which I shall be calling attention, I shall content myself here with some general remarks about this kind of approach to the problem.

Let us first glance at the history of this type of theory. It is, I think, clear that its immediate origins are to be found in the attack by writers of the verificationist school upon ethics as a branch of philosophy. The theory is intended to save ethics from this attack by showing that moral judgements are, after all, good empirical propositions, only their method of verification is different from, and somewhat looser than, that of ordinary fact-stating sentences. Thus they are indeed inferrable from observations of fact, but in a looser way.

Now this programme is from the start misconceived. A statement, however loosely it is bound to the facts, cannot answer a question of the form 'What shall I do?'; only a command can do this. Therefore, if we insist that moral judgements are nothing but loose statements of fact, we preclude them from fulfilling their main function; for their main function is to regulate conduct, and they can do this only if they are interpreted in such a way as to have imperative or prescriptive force. Since I am not concerned here with moral judgements as such, I shall leave till later the question 'How is the prescriptive force of moral judgements related to the descriptive function which they also normally have?' I am concerned here with the more fundamental problem of what sorts of reasoning can have as their end-product answers to questions of the form 'What shall I do?' It is clear that until we have clarified this more fundamental problem, we shall not be able to say much about the prescriptive force of moral judgements. Here it will suffice to show why, although prescription and description may be combined in the same judgement, description is not and never can be prescription. In other words, I am going to give reasons for holding that by no form of inference, however loose, can we get an answer to the question 'What shall I do?' out of a set of premisses which do not contain, at any rate implicitly, an imperative.

3. 5. My reasons for holding this are three. First, to hold that an imperative conclusion can be derived from purely indicative premisses leads to representing matters of substance as if they were verbal matters. In this connexion it is interesting to recall the parallel mistake of Professor Carnap in regard to physical laws. Carnap once held that, by the inclusion of suitable rules of inference in what he called the P-language (i.e. the language of a science) the statements of the science could be shown to be true in virtue of their form alone; and to say this is to assimilate those statements to what are normally called analytic statements -- though Carnap himself calls them synthetic, using the word in a special sense.5 This may seem a neat way of showing how scientific truths can be said to be necessary, and thus solving the troublesome 'problem of induction'. But if we ask 'What are these special rules of inference?' it is bound to appear that they are nothing but the laws of the science in another guise. Thus, if we have a rule of inference to the effect that we can proceed from 'This is a mule' to 'This (mule) is barren', then obviously our rule of inference only states in a new way the old law 'All mules are barren'. The question then arises, 'Is it proper to treat a law of science as if it were a rule of inference?' It is natural to say that it is not; for, as the work of Professor Popper, already referred to, has made clear, the rules of inference of ordinary logic can be shown to depend on the definitions of the logical words (2.4, note). Thus, tor example, it is part of the meaning of the word 'all' that we can infer from 'All mules are barren and this is a mule' the sentence 'This (mule) is barren'. If, therefore, we want to assimilate the laws of science to rules of inference, we shall have to show that they, likewise, follow from the meanings of the words used; for example, we shall have to show that the reason why we can pass from 'This is a mule' to 'This (mule) is barren' has something to do with the meaning of the words 'mule' and 'barren'. But to say this is to be guilty of conventionalism, the defects of which have been shown by the work (among others) of Professor von Wright.6 The sentence 'All mules are barren' tells us something, not about words, but about the world; and therefore it cannot be treated as a definition, nor as something analogous to a logical rule of inference. The only kind of definition to which it is in the least similar is an Aristotelian 'real' definition, or part of one, to the effect that it is -- in fact -- a property of mules that they are barren; unlikely as the conventionalists would be to admit it, their definitions and rules of inference have to be treated as 'real' in this sense if they are to do the job required of them.

The position with regard to conduct is similar. The view which I am attacking holds that by having special rules of inference we can say that there can be inferences from a set of indicative premisses to an imperative conclusion. If we ask 'What are these special rules of inference?' it is clear that they are nothing but the old rules of conduct in a new guise. What under the old dispensation appears as an imperative major premiss reappears under the new as a rule of inference. The criterion which I suggest for deciding on the merits of these two ways of putting the matter is the same as before. Let us take an example. Suppose that I say 'Don't say that, because it is false'. Are we to represent this argument as follows:

S is false.
∴ Do not say S,
or shall we add the imperative major premiss 'Never say what is false'? If the latter, the inference is valid by the ordinary rules of logic; but if the former, we have to have a special rule of inference, which will just be this imperative major premiss in another capacity. Does it matter which of these alternatives we choose? Surely it does if we are concerned to distinguish between on the one hand general principles about our conduct, which have content, and tell us to do, or to refrain from, certain positive acts in our external behaviour, and on the other logical rules, which are rules, not for behaving correctly, but for talking and thinking correctly, and are, if Popper is to be believed, not about our actions, but about the meanings of the words used.

This argument would tell equally against a theory which reduced rules of conduct to definitions of value-words; for in that case also arguments about how one should behave would turn into merely verbal disputes. Suppose that a Communist and I are arguing about whether I ought to do a certain act A; and suppose that on his principles I ought not to do it, whereas on mine I ought. An advocate of the sort of theory that I am attacking might treat this dispute as follows: each of the disputants has his own way of verifying the sentence 'I ought in these circumstances to do A'; and these ways differ. Therefore, in order to avoid such disputes, it would be better for us to substitute two unambiguous terms for the one ambiguous one; for example, the Communist should use the term 'ought1' for the concept governed by his rules of verification, and I should use 'ought2' for my concept. But the point is that there is a dispute, and not merely a verbal misunderstanding, between the Communist and me; we are differing about what I ought to do (not say) and, if he convinces me, my conduct will be substantially different from what it would be if I remained unconvinced.

3.6. My second reason for objecting to this kind of approach is that, if one is going to introduce looseness into our talk about conduct, it is as well to make clear in just what this looseness consists; and I am myself far from clear what is being proposed. Let us admit for the sake of argument that we are at liberty, if we please, to treat principles like 'Never say what is false' as rules of inference; we have then to ask, in what respect these rules of inference differ from the ordinary rules of logic. I have already given my own answer, that they differ in the same way as scientific laws differ from rules of logic, because they are about matters of substance, not about words -- though in this case the matters of substance are not matters of fact, but of what we should do. The answer given by the type of theory which I am criticizing is, that these rules of inference are looser than the rules of logic. Thus, if I say 'This is false, but say it', I am not contradicting myself, but only breaking the looser rule to the effect that the inference

S is false.
∴ Do not say S.
is 'in general' valid. It could be argued in favour of this way of treating the matter that we do often say 'Don't say S, because it's false', which presumably rests upon an inference like that just set out; but that this cannot be a strict entailment, because I would not ordinarily be said to contradict myself if I said 'S is false, but say it'.

We have therefore to inquire what can be meant by saying that a rule is 'in general' valid, but not universally. It sounds sensible to say that the rule 'Never say what is false' is a rule of this sort; for in fact we do think it right to observe it in the majority of cases, but we also think it right to break it in exceptional cases in the interests, for example, of tact, the winning of wars, or the preservation of innocent people from homicidal maniacs. Now I can think of at least two ways in which a rule or principle can be incompletely rigorous. The first way is when the rule lays down that a certain kind of action is in certain circumstances to be done, but it is understood that it is sufficient if it is done in the great majority of instances; exceptions are allowed if they are not too numerous in proportion to the total number of cases. An example of such a principle would be the principle that undergraduates must not take a week off work during term; clearly if once or twice during his career an undergraduate, whose industry is otherwise exemplary, takes some time off, even a week, we think no harm of it; but if he takes every week off, or even the majority, he probably gets into serious trouble. It is clear that the principle about not saying what is false is not of this character, because we do not say 'It doesn't matter your saying what is false occasionally, so long as you don't do it too often'.

The distinguishing character of this first kind of loose principle is that the exceptions to it are limited solely in number, and not otherwise determined. Provided that the undergraduate does not take time off constantly it does not matter whether he chooses one week in which to do it rather than another. Thus it is left to his own decision when, if at all, the exceptions to the principle are to be made, provided that they are not numerous. Moreover, his decision to take this week off rather than that effects no modification in the principle; it does not establish a new precedent for idleness that was not there before. Thus we may say that the principle is, vis-a-vis its exceptions, static.

Very different is the case with the other kind of 'loose' principle, to which 'Never say what is false' belongs. Here the exceptions are not limited by a numerical restriction, but by the peculiarities of particular classes of instances. We do not say 'Speak the truth in general, but it doesn't matter if you say what is false once in a way'; we say rather 'Speak the truth in general, but there are certain classes of cases in which this principle does not hold; for example, you may say what is false in order to save life, and there are other exceptions which you must learn to recognize'. This type of principle is quite different from the first. It is true that here too the decision is left to the agent in the individual case; he has to decide whether to make an exception or not; but what he is deciding is very different. The undergraduate deciding whether to take some time off has not to ask himself whether this is a case of a class which ought to be treated as exceptional. To the first kind of principle, there are not classes of exceptional cases, there are just exceptions, differing in no significant particular from the cases in which the principle is observed. But in the case of the principle 'Do not say what is false', in deciding whether or not to make an exception, we have to think, not 'Have I been breaking this principle much lately?', but 'Is there anything about this case that makes it different from the general run of cases, in such a way that I ought to put cases like it into a special class, and treat them as exceptions?' Thus, with rules of this kind, even exceptions are what I shall be calling decisions of principle, because in making them we are in effect modifying the principle. There is a dynamic relationship between the exceptions and the principle.

This makes it apparent that if we talk about the second kind of principle as being loose, we are being seriously misleading. Looseness in conduct is generally reckoned a bad thing, and it would be dangerous if philosophers were to put about the idea that principles of conduct are loose; for the ordinary person cannot be expected to distinguish readily in what sense they are being called loose. He will naturally take it that they are like the first kind of principle, and that because they are loose he need not trouble to observe them always, provided that he does it often enough to keep up appearances. But in this sense our principles of conduct, as indeed most principles of skill also, are not loose at all. The fact that exceptions are made to them is a sign, not of any essential looseness, but of our desire to make them as rigorous as we can. For what we are doing in allowing classes of exceptions is to make the principle, not looser, but more rigorous. Suppose that we start off with a principle never to say what is false, but regard this principle as provisional, and recognize that there may be exceptions. Suppose, then, that we decide to make an exception in the case of lies told in war-time to deceive the enemy. The rule has now become 'Never say what is false, except in war-time to deceive the enemy'. This principle, once the exception has been made explicit and included in the wording of the principle, is not looser than it was before, but tighter. In one large class of cases, where previously the possibility of exceptions was left open and we had to decide for ourselves, the position is now regulated; the principle lays down that in these circumstances we may say what is false.

This simplified statement of the way in which we modify principles by admitting classes of exceptions covers only those cases where the principle itself is stated in words which leave no doubt as to how to recognize the cases which fall under it. 'Never say what is false' is an example of such a principle. Often, however, principles are stated in such a way as makes it impossible to treat the question, whether a case falls under them, as a mere question of fact. Often, though not always, this is because the principle itself contains, in addition to such imperative verbs or value-words as are necessary for stating a principle of action, other value-words occupying the place that in a normal case would be occupied by purely descriptive terms. For example, we might put our principle about falsehood in a different way: 'Do not tell lies.' We might subsequently admit an exception in the case of falsehoods told, not with the intention of deceiving, but for other purposes, for example in order to amuse. Then we might say that to tell a story about someone, which everyone knows is ben trovato, is not lying. We can say this, because 'lying' does not mean simply telling falsehoods, but telling falsehoods which are reprehensible. Thus we might, and sometimes do, make a distinction between lies proper and white lies; lies proper are all reprehensible; a white lie on the other hand is, in the words of the Oxford English Dictionary, 'a consciously untrue statement which is not considered criminal: a falsehood rendered venial or praiseworthy by its motive'. In all such cases, the modification of the principle takes the form of an alteration, not of its actual wording, but of the conditions under which the principle is held to apply, that is, an alteration of the extension of the crucial word, or as we shall later call it its descriptive meaning, with the retention of its evaluative meaning. This, as has been pointed out to me by Professor H. L. A. Hart, is how legal principles are often modified by judicial decisions, as, for example, by the decision whether or not the occasional cricket ball landing in a public street is properly to be called a 'nuisance'. The word in question need not be (as here) a value-word; it may be a descriptive word whose meaning is loose enough to admit of such treatment. Such decisions, of course, render the law more precise, not looser. The extension of the word may be actually altered, or it may merely be rendered more precise. And it should not need pointing out that decisions of this kind are decisions, and not, as Aristotle seems sometimes to think, exercises of a peculiar kind of perception.7 We perceive, indeed, a difference in the class of case; but we decide whether this difference justifies us treating it as exceptional. Thus, far from principles like 'Never say what is false' being in some way by nature irredeemably loose, it is part of our moral development to turn them from provisional principles into precise principles with their exceptions definitely laid down; this process is, of course, never completed, but it is always going on in any individual lifetime. If we accept and continue to accept such a principle we cannot, as in the case of the rule about taking time off work, break it and leave the principle intact; we have to decide whether to observe the principle and refuse to modify it, or to break it and modify it by admitting a class of exceptions; whereas if the principle were really by nature loose, we could break it without modifying it at all. In the following chapter I shall consider in greater detail how we develop and modify our principles.

3.7. The gravest error, however, of the type of theory which I am criticizing is that it leaves out of our reasoning about conduct a factor which is of the very essence of morals. This factor is decision. In both the kinds of principle which I have been discussing, the principle falls short, in some sense, of being universal, only because in particular cases it is left to the decision of the agent whether to act upon the principle or not. Now to use the word 'inference' of a procedure like this is seriously misleading. When someone says, either 'This is false, so I won't say it', or 'This is false, but I'll say it all the same, and make an exception to my principle', he is doing a lot more than inferring. A process of inference alone would not tell him which of these two things he was to say in any single case falling under the principle. He has to decide which of them to say. Inferring consists in saying that if he tells a falsehood he will be breaking the principle, whereas if he tells the truth he will be observing it. This is a perfectly good deductive inference, and nothing further need be said about it. The rest of what he does is not inference at all, but something quite different, namely, deciding whether to alter the principle or not.

Thus I see no reason to take back what I have said about the way in which principles of conduct entail particular commands. The entailment is rigorous. What we have to investigate is, not some looseness in the entailment, but the way in which we form and modify our principles, and the relation between this process and the particular decisions that we make in the course of it.


Notes

1 Cf. Discourse on Method, Part V.

2 Groundwork of the Metaphysic of Morals, tr. H. J. Paton, pp. 84-85.

3 Reason in Ethics, p. 38.

4 Philosophical Quarterly, I (1951), 372.

5 Logical Syntax of Language, pp. 184-5.

6 Logical Problem of Induction, ch. iii.

7 Nicomachean Ethics, 1109b23, cf. 1126b4.