Having said last night that I disagreed with the account of freewill given by Waddington there is some obligation on me to offer some sort of alternative account. I think that Waddington was quite right to stress the importance of the freedom of the will as one of the unique characteristics of human beings. But traditionally to the question ‘are human beings unique, and if so in what ways?’ there have been at least two different answers. One of them is that only human beings are independent, autonomous, and free, but the other is that only human beings use language. I want to talk about these two answers. Are they in competition or do they complement each other? Is autonomy essential to a language user? Is the mastery of language essential to an autonomous agent?
Recently the uniqueness of man as a language user has been much emphasized by the linguists of the school of Chomsky. It has been challenged from two sides. There are those who deny that the natural ability to use language is peculiar to human beings among animal species, and they point to the striking success of the Chimpanzee Washoe in mimicking American sign language. On the other hand, the protagonists of artificial intelligence claim to be able to program computers to use and understand natural languages. There are now in existence, not too far from here, systems which are claimed by their designers to be capable of understanding some bits of normal English, and elsewhere there are systems which are claimed to be capable of answering questions, executing commands, and entering into dialogues. Now if these claims are correct then human beings must agree that not only animals but also non-living artifacts can share the ability to speak a language. Not even the warmest admirers of non-human animals and of computers commonly regard them as fully autonomous agents. Animals may be lively, cunning, or independent but we don't credit them with moral responsibility. And on the other hand the most successful of the artificial intelligence systems run on principles which are wholly deterministic. It appears therefore either that autonomy is unnecessary for language use, or that the appearance which animals and computers give of using language is an illusory one.
Since the time of Hobbes, philosophers have long considered, either with favour or with contempt, the suggestion that in determinism is not essential for autonomy. A decision on this age-long debate wouldn't by itself settle whether autonomy and the use of language necessarily go together. But it may well throw light on the relationship between the two if we take the traditional discussion of the relation between autonomy and indeterminism a step further.
At first sight there seems to be a clear conflict between determinism and freedom. You will remember that because of this apparent conflict Dr Johnson rejected determinism: ‘We know our will is free and there's an end on it’, he said. If we add his hidden premise, we have the simple argument: freedom and determinism are incompatible; we know we are free; therefore determinism is false.
Now philosophers who have rejected the incompatibility between determinism and freedom have commonly made a distinction between various senses of freedom. They have admitted that there are senses in which freedom is incompatible with determinism but they have denied that we know in those senses that we are free. The sense in which we know we are free, they say, is one in which freedom leaves room for determinism.
An early distinction of this kind which was long influential was the distinction between liberty of indifference and liberty of spontaneity. This began as a theological distinction. It was used by people like Jonathan Edwards, who were committed to determinism by their admiration for Calvin. It was later used also by people like Hume who were committed to determinism by admiration for Newton. The concept of liberty of spontaneity is one which approaches freewill through the notion of desire or wanting. It sees the exercise of freewill essentially as the execution of one's wants. To act freely is to act because one wants to. This was the notion I was using last night. It doesn't of course mean, as Waddington seemed to suggest, that to enjoy freedom of spontaneity is to be able to do everything that one wants, but it is to be able in some cases to put one's wants into execution. Now the notion of liberty of indifference does not start from ‘wanting’, it starts from ‘power’ or ‘ability’. It sees freewill as essentially a capacity for alternative action. To act freely on this view is to act in possession of the power to act otherwise. It is only liberty of indifference that presents an obvious contrast with determinism. The contradictory of spontaneity is not determinism but compulsion.
The two types of liberty appear to be distinct and in theory separable. And once the distinction has been drawn it seems easy enough for the defender of determinism to reply to Dr Johnson's simple argument. The freedom which is incompatible with determinism, he can say, is liberty of indifference. The freedom which we know we have is liberty of spontaneity. Whether we enjoy liberty of indifference can't be just a matter of bluff commonsense experience. How could experience show us that there is no sufficient antecedent condition for our actions: that our actions are uncaused? Whether there are such causal conditions depends on the nature of the totality of physical laws which govern life in our Universe. How could the introspection even of Dr Johnson be sufficient to establish the nature of those? How could one feel within oneself the lack of a law correlating one's present action with one's previous history and environment? In another pronouncement Johnson said: ‘All theory is against the freedom of the will; all experience is for it’. But what all theory is against is liberty of indifference; what experience is for is liberty of spontaneity.
The theory which I have been presenting so far might be called compatibilism, that is to say the theory that freedom and determinism are compatible with each other. It is sometimes known as soft determinism. Now the difficulty with compatibilist theory as put forward by a writer such as Hume is that it involves a naive conception of mental causation. The determinism which is put forward is a psychological determinism. It is observed that the fact that we can do what we want doesn't mean that we can want what we want. If ah our wants are determined, the theory goes, then it may be true that we can do whatever we want and yet all our actions can still be determined. Now the theory hasn't lacked defenders in this form in the present century. But I think that at the present time most philosophers would regard it as incorrect to think of wants, or desires, or decisions, or the will, as being some sort of mental event which causally determines an action. To say that somebody did an act because he wanted to is not to postulate a mental event as the cause of the action operating through some mental mechanism which we don't yet perfectly understand. There are of course some wants which are mental events: pangs of hunger, stirrings of lust or sudden impulses to pick a flower. Nonetheless the wanting which makes an act voluntary, the wanting which makes the difference between voluntariness and non-voluntariness, is not a mental event. Take a simple example. The sentence that I have just uttered I uttered voluntarily. I chose each word of it, but there wasn't any mental event of desiring to utter the word, no act of willing, between each word and the next, to choose the next word. If some voluntary events, like the choice of words, don't demand a specific event to cause them why should any?
Now certainly when one says that somebody brought about a certain result because he wanted to, one is saying something about the causation of the result. But what one is saying is not that the act was caused by a certain mental event, but rather that the agent was in a certain state in relation to causes when he did it. I think that it is a statement essentially about the absence of certain causes or certain circumstances. This idea is a very old one; it goes back to Aristotle. Aristotle thought that to say that an act was voluntary was to say that it wasn't done under constraint or by mistake. I think that Aristotle's list of the conditions which must be absent is certainly too short and I must confess I don't know how to complete it. But one can criticise Aristotle on this point without agreeing with the proponents of liberty of indifference that in order to be voluntary, an act must be totally uncaused.
The proponents of liberty of spontaneity are right to say that a person's doing x because he wants to is compatible with the causal pre-determination of the event which is his doing x. But this isn't because his wanting to do x is itself a cause which is causally determined. It is rather because the types of causal determinism that are ruled out by saying that a man acted voluntarily don't necessarily exhaust the types of determinism that there are. Pre-eminent among the types of causal determinism that are ruled out are psychological determinisms—by which I mean determinisms in virtue of laws which can only be stated by the use of mental predicates. This includes economic and sociological determinism. But physiological determinism is a rather different thing. It seems to me to be neither entailed nor excluded by the existence of liberty of spontaneity. It seems to me that there is no incompatibility between explanation by neuro-physiological states and explanations in terms of wants and intentions. And this is so even if the laws of neuro-physiology should turn out in the end to be fundamentally deterministic—this may or may not be likely.
There is one objection, an important objection, which is always made to this soft determinist theory. It is made by John Lucas in his book The Freedom of the Will. It was made recently by Professor Anscombe in her inaugural lecture at the University of Cambridge, My actions are mostly physical movements, the objection goes, and if these physical movements are physically pre-determined by processes which I do not control, then my freedom is illusory. So the truth of physical determinism is indispensable if we are to make anything of the claim to freedom. Now I think that this protest can be attacked on purely logical grounds. The protester's argument appeals to the principle—sometimes known as Leibniz's law—that if x is identical with y then whatever is true of x is also true of y. My actions are identical with certain physical movements—the argument goes—these physical movements are determined, therefore my actions are determined. But it's well-known that Leibniz's law cannot be relied on in modal contexts—that is to say, in contexts where the predicates in question involve the notions of necessity and possibility. To take a simple example, the number of member nations in the Common Market is identical with the number six and the number six is necessarily smaller than the number seven, but the number of members of the Common Market is not necessarily smaller than seven. It is only contingently so at this time. But now, the predicate ‘determined’ like the predicate ‘necessarily smaller than’ is a modal predicate. That is, to say that an event was determined is to say that there was no possibility of things turning out otherwise.
This reply to the incompatibilist is suggested by recent work of the American philosopher Donald Davidson. Davidson's own solution to the question of the relationship between freedom and determinism is to say that while being caused is a relationship which links events in themselves, being determined is a property which attaches to events under certain descriptions. For an event to be determined is for it to fall under a description such that there exists a law from which with certain antecedent conditions it could be deduced that an event of that description would occur. For Davidson, events are part of the furniture of the world in a way in which laws are not. According to his theory, laws are essentially linguistic.
Now this Davidsonian answer to the type of position represented by Lucas and Anscombe does not seem to me to be conclusive. Laws may perhaps be linguistic, I don't know, but natural possibilities and impossibilities are surely not linguistic. To take an instance of natural impossibility, ‘I cannot fly’. Now this is not a property which belongs to me only under some descriptions and not under others. However you describe me, whatever description you use of me, it is true that I cannot fly.
Now the necessity and possibility which is commonly formalised in logic—in the area of logic known as ‘modal logic’ which deals with these questions—is essentially logical necessity—the necessity which attaches to logical truths. Many philosophers have believed that logical necessity is ultimately a kind of linguistic necessity. If they are right then the failure of Leibniz's law in modal contexts may be ultimately a linguistic phenomenon from which no lesson can be drawn about non-linguistic necessities and possibilities. I don't myself believe that logical necessity is essentially a matter of language, and quite apart from this there are good reasons for thinking that the ‘cans’ and the ‘musts’ of natural necessity and possibility are not adequately represented by any of the standard modal logics. I shall list some of these reasons in a moment. But first I want to draw attention to the significance of these logical considerations for our overall purpose.
The reason for going into elementary matters of formal logic is this. The expression of deterministic theory demands the use of the notion of natural necessity and possibility; the reporting of free actions demands the use of the notion of rational or mind-guided possibility—and whatever the relation this latter type of possibility may have to the former we clearly need a formal logic for natural necessity and possibility. If we have not yet a logic adequate to formalize these notions then it seems that we are not yet in a position to know how to set about constructing a computer programme to simulate the activity of an autonomous agent.
The basis of contemporary developments in formal logic over the last century is the first order predicate calculus—the part of logic which deals with predicates of objects and quantifiers (the symbols which correspond to the words ‘all’ and ‘some’ in ordinary English). This first order predicate calculus is a system of considerable power whose properties are by now very well known. John Lucas will be talking about some of them tomorrow. Among the things which are known about it is that it is inadequate for the description of human mental states and activities. Human mental states, like belief and desire, have a certain logical property known as ‘intentionality’: they are intentional. When we say that beliefs and desires are intentional in this logical sense, we mean that there are various principles of reasoning which are valid in the first order predicate calculus, and which will lead from true premises to false conclusions if they are applied to the thoughts and wants of human beings. The best known of such principles is once again Leibniz's law—the law that if x is the same as y then whatever is true of x is true of y. This law fails in intentional contexts no less than in modal contexts. Suppose that a detective knows that Mr Hyde is a murderer and suppose that Dr Jekyll is identical with Mr Hyde; still it does not follow that the detective knows that Dr Jekyll is a murderer. So Leibniz's law in such cases does not hold.
Since the failure of Leibniz's law occurs both in modal contexts and in intentional contexts it might be thought that modal logics would be adequate for the description of human activities and capacities. But a number of arguments make it clear that standard modal logics do not suffice to formalise even the simpler statements about human abilities and capacities. In standard modal logics, the following three laws which I am about to enunciate appear either as axioms or as theorems:
- If p then possibly p.
- It is possible that either p or q if and only if either it is possible that p or it is possible that q.
- If it is possible that it is possible that p then it is possible that p.
I think you will agree that these principles sound harmless enough. If you have any objection to them it will be on the grounds of tediousness rather than of falsehood. But if modal logic of this kind is to be able to formalise statements about human abilities and skills then we have to be able to translate a sentence like ‘I can do such and such’ into the form ‘it is possible that I am doing such and such,’ and vice versa. So that ‘I can whistle the Emperor Concerto from beginning to end’ has to be ‘it is possible that I am… etc’ Now once we allow this transformation it's obvious that these three laws can very easily be falsified.
Take the first one. I can't spell the word ‘seize’. I can never remember whether or not it is an exception to the rule about ‘i’ before ‘e’ yet since I usually toss a coin to decide how to order the vowels, from time to time I spell it correctly. So on any particular occasion it is true that I am spelling ‘seize’ correctly, and yet it is not true that I can spell ‘seize’ correctly. So ‘if p then possibly p’ is falsified.
For an example of the falsification of the second law, consider the case of my two friends, Tweedledum and Tweedledee. They are identical twins, they look totally unlike anybody else that I know, so that when I meet one or other of them I can always tell that it is either Tweedledum or Tweedledee but I can never tell either that it is Tweedledum or that it is Tweedledee and so that provides a counter instance to the second one.
Finally, law number three. It is hard to translate this one at all into an ordinary ‘can’ sentence because it is not clear what is meant by ‘I can can do x’. The only plausible sense seems to be ‘I can acquire the ability to do x’. If it is so interpreted then this law is quite false. Suppose that I am filling in a questionnaire for a job and it says ‘Can you speak Russian?’ I argue that I have the ability to acquire the ability to speak Russian and if it is possible that it is possible that p then it is possible that p so I answer ‘yes’ to the question.
For these reasons and for others which it would take too long to detail it seems to me clear that modal logic of the classical kind is inadequate to deal with the capacities of human beings, I should say that I have not chosen these laws just randomly as things to find fault with. They are the three axioms of a particular well-known system.
Now let us return to the incompatibilist argument favoured by Mr Lucas, that if our actions are identical with physical movements which are determined, then our actions are themselves determined. Whether this argument is valid or not depends on whether the predicates which are involved in the statement of determinism and freedom are predicates to which Leibniz's law applies. If the argument I have just gone through is correct then we can't argue from the analogy of modal logic that the law doesn't apply, because the ‘cans’ and ‘musts’ of determinism and freedom are not those which are formalised by modal logics. However, there are independent reasons for thinking that these ‘cans’ and ‘musts’ are not subject to Leibniz's law as they must be if the incompatibilist argument is to work.
Here is a formal argument which looks all right until you begin to think about it. ‘I can do x, doing x is doing y, therefore I can do y’. This argument does work in some cases, namely if it is some sort of logical truth that doing x is doing y, if it is true by definition. But it doesn't work if all that is meant by this second premise is that a particular instance of doing x is the same as a particular instance of doing y. Let me explain what I mean. Given a dart I can usually hit the dart board. Now on a particular occasion I may hit the dart board by hitting the centre of the board, but it by no means follows that I am capable of hitting the centre of the board. I am not. Any particular exercise of a power or a skill will have other descriptions besides the one which occurs in the specification of the power, and the possession of the power which is specified in no way involves the possession of the power to perform acts according to those other descriptions. The example I have given concerns a human skill, but similar considerations apply to natural possibilities and necessities. I conclude, therefore, that the argument to say that freedom and determinism are incompatible, which is based on arguments of this form, is invalid.
To argue, as I have been doing, for the compatibility of freedom and determinism is in effect to say that liberty of spontaneity does not involve liberty of indifference, at least if liberty of indifference is interpreted in any way which entails indeterminism. But it may well be that the two kinds of liberty are connected in a subtle way. Aristotle thought that it was only when an agent was of a kind to be able both to do x and not to do x that you could attribute a want or desire to do x as opposed to a mere tendency to do x. This may be true, but even if it is it doesn't follow, I think, that when somebody performs a certain action this can't be explained by saying he wanted to do it unless on that occasion it was in his power not to do it. To take a very old example, somebody may stay in a room because he wants to, even though unknown to himself the door has been locked the whole time. But the notion of power which is involved in the definition of freedom of indifference doesn't seem to me on reflection to be one which is incompatible with predictability, much less with determinism. On particular occasions it may well be true that I can do x even if it is predictable that I will not do x. There is nothing contradictory in saying that I can, but will not, do x any more than there is anything contradictory in saying that I could have done x but did not do so. Thirty seconds ago I did not throw the chalk into the third row of the audience. But I could have done, and my not having done so did not take away the power which I then had. The presupposition which I have so far accepted uncriticised from Hume, namely the presupposition that liberty of indifference is incompatible with determinism, is a presupposition which won't, once looked at, bear very much weight.
At last I want to return to the topic of artificial intelligence. At the beginning of this paper I argued that it seemed at first sight an objection to artificial language-using systems that they were deterministic systems. If my argument this evening has been right, then it seems that this is not a valid objection. Even in the case of human language-users, indeterminism doesn't seem to be established. A fortiori there is no reason to demand that if a computer is to be regarded as possessing the ability to use language it should be exempt from determinism. But since we can't be certain that liberty of indifference can be equated with indeterminism it remains an open question whether liberty of indifference of some kind is necessary for the ability to use language. In saying this I think I am allying myself with Waddington in thinking that the questions about Heisenberg indeterminacy are really irrelevant to questions about the freedom of the human will.
Though I think that their being deterministic is no reason why computer systems should not be genuinely said to use language, I think that alleged language-using computers, such as those using the programmes I know about, cannot genuinely be said to be using language. Because even if liberty of indifference is not necessary for a language-user, liberty of spontaneity—the ability to do things because one wants to—does seem to me to be essential. That is to say, only agents who have the power to select their own goals, and to act in pursuit of those goals, can be said to use language. Because it is essential, in order to use language, to be able to confer meaning on symbols; and meaning can only be conferred by agents who are capable of having common purposes and a shared form of life. Of course, the output of a computer which is programmed to produce English sentences is a meaningful output, but the meaning is conferred on it not by the computer itself, but by the living human beings who use the English language.
Liberty of spontaneity is something which artificial intelligences lack. They can perhaps be said to have goals, and perhaps even self-selected goals. If a computer is programmed to parse English sentences when a particular sentence is typed into it then it may perhaps be said to select the parsing of that sentence as a goal, and the choice of that goal governs a large number of complicated sub-routines. But it seems clear that computers don't have the long-term self-selected goals which constitute a form of life as the necessary background to a use of language. To see this, consider that if there were a number of computers isolated from human contact, which had an output isomorphic to the alleged language-using computers this would not be sufficient to make their output into a use of language. If we are tempted to think otherwise, it's simply because their output in present circumstances is one which is interpreted as a language and connects with our human life and behaviour. But for something to be a language it is not enough that it should have a certain complicated mathematical structure. It is essential that that structure should be interpreted and interpretation—the conveying of meaning on symbols—is so far the prerogative of human beings.
Kenny has given us an extremely dense paper with a great many precisely formulated arguments. There is only time to make a very few short comments. He has dealt with two distinct problems, between which he makes connections, and—I am not quite certain I entirely follow his connections. First, he dealt with the problem of determinism and freewill, and then he came to the question of the use of language, by computers for instance. Now on the question of determinism and freewill, he first of all made a distinction between the liberty of indifference, which says ‘I did something but I could have done something else’; and, as he says, this is really incompatible with a belief in determinism. Then he contrasted this with the liberty of spontaneity which says ‘I did this because I wanted to,’ and he then pursued the arguments—is this, or is this not, compatible with determinism? At first he gave some arguments tending to show that it might well be compatible but then he came to the gist of his thesis, a point which was his main argument against allowing that even the liberty of spontaneity is compatible with determinism. This argument is that my actions are mostly physical movements, and if these movements are physically pre-determined by processes which I do not control, then my freedom is illusory; and therefore physical determinism is incompatible with liberty of spontaneity.
He then went on to ways in which this conclusion might possibly be escaped. He described a system of what he called ‘modal logic’ which rejects what he spoke of as Leibniz's law, that ‘if x is identical with y whatever is true of x is also true of y’. He expounded this possible way of escape, but really, I think, came to the conclusion that it could not work—that it did not really give you an escape from the argument. My criticism of this part of his argument was that it all boiled down, at least in the way it was expounded here, to translating these arguments into terms involving words such as ‘can’ and ‘must’ and words of that kind. Now these are notoriously ambiguous words—‘I can do something’ Kenny interpreted to mean it is possible that I am doing it’, but it could also be it is not impossible that I shall do it’; it could also be I am permitted to do it’; and it could be I know how to do it regularly’. And this brings me to what I think is one of my major criticisms: that he tended to try to refute arguments by what I should call once-off examples. For instance, he refers to the argument ‘I can do x, doing x is the same as doing y and therefore I can do y’. Then later he took as an example, playing darts; and if he threw a dart and he did actually once hit the Bull's Eye, then he could certainly say ‘I can hit the board’. But suppose someone came to me and said ‘I'll take you on at a game of darts for a double scotch against a half-pint and I'm just a beginner—I can hit the board’, and if he did hit the Bull's Eye first shot—well, I possibly would begin to wonder. And if he went on regularly hitting the Bull's Eye, I shouldn't accept those odds again the next time. Kenny stated at one point that there was something called the ‘first order predicate calculus’, and this included words such as ‘some’ and ‘all’ but it seemed to me noticeable that in the arguments he expounded the words ‘some’ and ‘all’ were not included. He doesn't say that ‘I can sometimes do x’ or ‘I can always do x’; he just says ‘I can do x’. And then he gives one example of doing x, when doing x is not the same as doing y. Now obviously I will have to study these arguments a lot more carefully than I have yet had time to—but it seemed to me that Kenny was rejecting arguments because he formulated them without any specification of whether a relation holds once, sometimes or always, and then he could give one example when it didn't hold. So I am still rather sceptical about his escape from the argument that liberty of spontaneity is incompatible with determinism. He argued that you could escape from it but I think this escape was based on a failure to specify in his arguments whether he was dealing with a regular occurrence or an irregular occurrence. I personally think you may be able to escape from it, but in quite a different way.
Now, turning to his other argument, about language and the liberty of spontaneity, he claimed at one point that only agents that can select their own goals and act in pursuit of these goals can be said to use language. Here I think it is a question of the word ‘use’. Slaves can't select their own goals but have to do what they are told—and they can certainly use language well enough. I think what Kenny was talking about here is a very important thing, but it is not so much the use of language as the ability to invent a language. At the end of his paper he speaks of a number of computers, isolated from human contact, which have an output isomorphic with the language-using computers, and argues that this would not be sufficient to justify saying they were using language. Well, I think it would be sufficient for us to say they were using language, but it wouldn't be sufficient for us to say they had invented it. If they were using English they would be using English; and if the computers could understand each other, and each do what others, were telling them, they wouldn't have invented the language but they would be using it, in my opinion.
I think I ought to pick up the gauntlet which was thrown down and I want to start by being slightly unfair and making a debating point against Dr Kenny. He has wished on me and Professor Anscombe an argument which if it were valid would support a conclusion which we both maintain. He then points out that the argument is invalid and from this invites you to draw the conclusion that the conclusion that we draw is not true. I want to go a bit further, though: first of all, even if Kenny could make a rather stronger point, I think I should still be rather unworried because it is one of the theses that I am committed to maintaining,—and have maintained—successfully—that no formal system of inference can capture all the inferences which we regard as valid. So that even if there were some informal inference which could not be formalised in S4 or in the predicate calculus, nevertheless it would not follow that this inference was not valid. But I want to go one stage further because I am very unhappy with the sorrows of the policeman in London not knowing whether Dr Jekyll is the same as Mr Hyde, or, to take a more classical and more local example, the worry of George IV in Edinburgh, who wanted to know whether Scott was the author of Waverley. And the argument which Kenny has put forward is that George IV did know, presumably, that the author of Waverley was the author of Waverley, and this is a necessary truth, but although it is true that George IV knew that the author of Waverley was the author of Waverley and it is true that Scott was the author of Waverley, he didn't know that Scott was the author of Waverley. But let me take another example. George IV might have wanted to know whether the author of Waverley was the author of Old Mortality or of Ivanhoe, or of many others, and from the fact that he did not know that Scott was the author of Waverley it would not at all follow that he could not have known that the author of Waverley was the author of Ivanhoe. After all, it might very well have had on the title page of one of these ‘by the author of Waverley’. And this is the case in the argument that I am putting forward. I can't formalise it, because it uses two sorts of modality. Essentially, what I am claiming is that the connection between the description of a piece of behaviour which is something which physicists or physiologists might be able to discuss, whether it is determined or not, and the description of a human action is not that contingent. That is to say, although it may not be in the same modality as that used by the Laplacean physicists who from a given initial description of the universe and knowing the laws of mechanics could have predicted the final positions and momenta of all the Newtonian corpuscles—although it might not be the same as that modality, depending on some physical necessity, still there is some necessity of being able to work out from description of behaviour, not perhaps the exact description of action but at least some descriptions of action. Take for example the hypothetical action of Kenny in throwing a piece of chalk into the third row of the audience. Now, if we had been able to predict that he would have done this, we might not have been able to say with certainty whether this action which was being described to us was a philosophical example or as an existentialist doing an acte gratuit, but at least we could have been sure that this was not an exercise in Buddhist impassivity, and we would have known that with some sort of modal necessity. We should have to have a much more complicated modal logic than that which is offered on the board, with at least two sorts of modal operator, m and m’, and what the laws of that modal logic would be I should hesitate to have to say, but I think I can be reasonably sure that if I have got ‘necessarily p’ according to one modality and ‘necessarily (according to another modality) p is identical with q’ then ‘necessarily (according to one or other of those modalities or some other modality still) q’. Well, this is my half answer to Kenny.
Let me just make one more positive point. He was maintaining that the liberty of spontaneity was not the same as the liberty of indifference, which is of course true, and that the liberty of spontaneity was the only thing which we need be worried about with freewill and is compatible with determinism. But this won't wash. He defines the liberty of spontaneity in terms of some sort of ability to do what one wants. But supposing someone here had been hypnotised, by, say Kenny, into throwing back that piece of chalk and then did it. Now it is clear that he wanted to do it and it is clear that in many senses he was free to do it—he had the liberty of that apparently spontaneous action of throwing back the piece of chalk!—but we would not say that he was really free to do it, and if the matter came to the court of law the fact that he had been hypnotised would be evidence against holding him responsible. Or to take another example—the sad cases of men born with xyy chromosomes; insofar as it is made out that people with this genetic constitution cannot but be rather too tall and rather too violent, they are acquitted of criminal responsibility. We don't hold kleptomaniacs responsible. We don't think that they are free. Although they undoubtedly want to steal, and can do what they want, the whole point is that they are not free not to want to steal and this is the issue of the freedom of the will on which I am taking my stand and which seems to be more fundamental than the many true and important things that Kenny has said,
I don't really want to take more than a minute because I think Kenny ought to have the chance of replying to what has already been said. I think Waddington was very unfair to Kenny. I don't think that he really said at all what Wad suggested but any way I would like to take up Kenny's question whether we could speak of language use without the language-user having selected his own goals. I thought Waddington made a correct point when he said that one couldn't really speak of the proper use of language unless the language was interpreted. But that is a different question from whether the language-user has selected his own goals or not. Now it seems to me that if you say ‘Please move out of the way’, and I pick up my chair and move out of the way, you know that I am at least understanding what you said,—that your language has been interpreted by me. But I might not have selected my own goals; I might have been a robot to which you addressed this remark, and a robot in fact could very well do that kind of thing—or so I am told. And so I want to insist that we must separate, in our discussion of what counts as a real language-user, the question of the self-selection of goals—which is of course a very interesting business—and the question of the interpretation of utterances in the language, which I think is something we can claim to have made a start on already.
I would like to begin by thanking Waddington for his summary of what I said, which I thought was very accurate up to the final point which I would like to thank Longuet-Higgins for mentioning. I do think that the line of argument that I followed gives you an escape from the argument to show that freedom and determinism are incompatible. I do think we can escape from the Lucas argument. Lucas was, of course, quite right that I hadn't proved that the two were compatible. I don't quite know how one can prove the compatibility of two things save by disproving the arguments that are brought up to show they are incompatible, and I await with interest the further arguments to show that they are incompatible.
I agree with Waddington that the ability to do something can be displayed by doing it regularly. But I don't think that it is sufficiently established by a single exercise of a certain skill. Now the examples that I gave, as several speakers pointed out, were on the whole rather out of the way examples, and of course they were just single examples. But the point of my argument was this: that the Anscombe-Lucas argument appeals tacitly to a certain logical law and that logical law doesn't work. Now if a logical law is to work it must work always. It is no good saying ‘Oh, I know in that example that you gave it didn't work but I can think of lots of other examples where it would work—just look at Scott and the author of Waverley’. If something really is a logical law then it always works and a single counter-example is enough to overthrow it.
John Lucas was perfectly correct in what he said about hypnotism—that action under hypnotic suggestion is not free. Hypnotism was one of the things which I had in mind when I said that I thought Aristotle's definition of voluntariness was inadequate. He said that an action was voluntary if it was not done under duress or by mistake. I said I thought his list of the things that must be absent was too short, and I would certainly agree with John that hypnosis is one of the things that has to be absent. About the xyy chromosomes, I am not sure—I would like to hear more about them because I have heard slightly different accounts of the data from the one Lucas gave.
Finally, with regard to Longuet-Higgins’ remarks, I would not wish to claim that in order for somebody to be a language-user he has to select every goal off his own bat—clearly this would be far too strong a claim. It is only that in order to be a language-user you do have to be capable of choosing your own long-term goals. Why do I think this? Because I think that the interpretation, the giving of meaning to language is ultimately a matter of intentions. If one is to mean something by something then one has to intend to produce certain effects, and so on. I am not yet convinced that, clever as they are, Longuet-Higgins’ computers have intentions. But this is a topic we shall return to.