The argument is very simple. Set up a simple experiment with a subject. Repeat the experiment with conditions as nearly identical as possible. It is of course impossible to have exactly the same conditions.

Instruct the subject to either do or not do some simple action, like raising his arm.

Collect a reasonable number of both raising and not raising the arm. Now argue that, since he showed that it was possible for him to have done or not done the action, retrospectively in any individual case he could have done otherwise.

a careful analysis of the import of the retrospective judgement, "I could have acted otherwise than I did," will, I believe, disclose it to be an empirically meaningless statement.

(Ledger Wood, "The Free Will Controversy," in Mandelbaum, Gramlich, and Anderson, eds. Philosophic Problems (New York, Macmillan, 1957), p. 308

Science professes to draw no conclusions but such as are based on matters of fact, things that have actually happened; but how can any amount of assurance that something actually happened give us the least grain of information as to whether another thing might or might not have happened in its place? Only facts can be proved by other facts. With things that are possibilities and not facts, facts have no concern. If we have no other evidence than the evidence of existing facts, the possibility-question must remain a mystery never to be cleared up.

Williarn James, "The Dilemma of Determinism," in Essays in Pragmatism (New York: Hafner, 1955), p. 42.

I now wish to argue that we can know empirically that a person could have done otherwise.* A person could have done otherwise if he could have done what he did not do. Moreover, if it is true at the present time that a person can now do what he is not now doing, then, later, it will be true that he could have done something at this time which he did not do. This, of course, follows from the fact that "could" is sometimes merely the past indicative of "can." ** What I now want to argue is that we do sometimes know empirically that a person can do at a certain time what he is not then doing, and, consequently, that he could have done at that time what he did not then do. Moreover, we can obtain empirical evidence in such a way that our methods will satisfy the most rigorous standards of scientific procedure.

I shall attempt to show that we can know empirically that a person could have done what he did not do by first considering the more general question of how we ever know what people can do. It is, I suppose, obvious that there is no problem of how we know a person can do something when we see him do it. In this case, the evidence that we have for the hypothesis that a person can do something entails the hypothesis. But all that is entailed by the evidence is that the person can do what we see him do at the time we see him do it. It is at least logically possible that he cannot do it at any other time. Thus, when we project the hypothesis that a person can do something at some time when we do not see him do it, the empirical evidence that we have for the hypothesis will not entail the hypothesis. The problem of our knowledge of what people can do is, therefore, primarily the problem of showing how we know that people can do certain things at those times at which we do not see them do the things in question. The solution to the problem depends upon the recognition of the fact that one fundamental way (there are others) in which we know that a person can do something at some time when we do not see him do it is by seeing him do, it at some other time. However, it is not merely a matter of seeing him do something at some other time that would justify our claim to know that he can do it at the time at which we do not see him do it, but of seeing him do it when certain other epistemic conditions are also satisfied. I shall discuss four such conditions, which seem to me to be the most important. I shall call them the conditions of temporal propinquity, circumstantial variety, agent similarity, and simple frequency.

Temporal propinquity. The amount of time that has elapsed between the time at which we see a person perform an action and the time at which it is claimed that he can perform the action is of considerable importance. For example, if I saw a man perform forty push-ups twenty years ago and have not seen him do it since, that would hardly justify my claim to know that he can do it now. On the other hand, if I saw him do it yesterday, my claim would have much greater merit. The less time that elapses between the time at which we see a person perform an action and the time at which we claim to know that he can perform it, the more justified our claim. This condition requires one qualification Certain actions - for example, running a four-minute mile — require unusual endurance; consequently, if we have just seen a person do such a thing, it is a good guess that, being tired, he cannot do it now. The condition is relevant even in the case of such actions, but we must add the qualification that sufficient time has elapsed between the time at which we saw the person perform the action and the time at which it is claimed that he could have performed the action, so that the effects of fatigue would not prevent or hinder the person from performing it.

Circumstantial variety. The greater the variety of circumstances under which we have seen the person perform an action, the more justified we are in claiming to know that he can perform it. There is also a qualification needed here. Sometimes, though we have not seen a person perform an action in a very great variety of circumstances, we have seen him perform the action under circumstances very similar to the circumstances he is in when it is claimed that he can perform it. In this case, the greater the similarity of the circumstances, the better the evidence.

Agent similarity. If the condition of the agent changes radically, from the time at which we see him perform an action to the time at which it is claimed that he can perform it, then our evidence that he can perform the action may be greatly weakened. For example, if we see a man lift a two-hundred-pound weight, and he subsequently breaks his arm, our having seen him lift the weight is surely not very good evidence that he can do it now that his arm is broken. Thus, the greater the similarity of the condition of the agent, at the time when we see him perform the action, to the condition of the agent at the time at which we claim that he can perform it, the greater the justification of our claim. To some extent this condition, like the preceding one, may be formulated as a condition of variety rather than as a condition of similarity. That is, if we have seen the agent perform an action at times when his condition has varied greatly, then, even though the condition of the agent at the time at which it is claimed that he can perform the action is quite different from what it was when we saw him perform it, the claim might, nevertheless, be fairly well justified. However, it seems to me that with respect to the circumstances, variety is more important, while, with respect to the condition of the agent, similarity is more important. The reason for this is that great changes in circumstances are often unimportant, while small changes in the condition of the agent may often be crucial.

Simple frequency. Other conditions aside, the more frequently we have seen a person perform an action, the more justified we are in claiming to know that he can perform the action when we do not see him perform it.

These conditions are related in various ways. For example, temporal propinquity tends to produce agent similarity, because generally people change less in a shorter time than in a longer time. Of course, circumstantial variety contributes to simple frequency, and vice versa. Thus, these conditions, which are simple canons of inductive evidence for a certain sort of hypothesis, are inductively interdependent.

Moreover, the importance of the various conditions depends to a considerable extent upon the kind of action involved. With respect to actions that one usually retains the ability to perform for a long time, such as wiggling one's ears, the condition of temporal propinquity is less important, whereas with respect to actions that one quickly loses the ability to perform, such as running a four-minute mile, the condition of temporal propinquity is much more important.

However, if all of these conditions are very well satisfied with respect to any action, we possess sufficient empirical evidence to support the hypothesis that a person can perform the action when we do not see him perform it, and, in the absence of any evidence to the contrary, we are certainly justified in claiming to know that the hypothesis is true. Indeed, these conditions are typical of the usual canons of inductive evidence; if they are satisfied, then, by the usual canons of inductive evidence, our evidence is excellent.

It will not be difficult to imagine an experiment, which we could quite easily carry out, that would enable us to obtain such evidence. To avoid unnecessary complications, we shall concern ourselves with one very simple action, the lifting of an arm. Now let us imagine that we find a subject who is normal in every way, and fabricate an experiment to investigate when our subject can, and when he cannot, perform this very simple action. For example, we might first instruct him to lift his arm whenever we tell him to, and see that he does this. We might then instruct him to lift his arm whenever we tell him not to, and see that he does this. We might then tell him to heed or not to heed our instructions, as he wishes, and see that he sometimes lifts his arm when we tell him to, and sometimes does not, and that he sometimes lifts his arm when we tell him not to, and sometimes does not. We might then run this same experiment under a variety of circumstances, indoors and outdoors, under stress and under relaxed conditions, with a weight attached to his arm and without impediments, etc. Moreover, we might keep careful records of the condition of the subject throughout all our experiments, and, finally, we might vary the condition of the subject by the use of drugs, hypnotism, etc.

Now, suppose that we instruct our subject to heed or not heed our instructions as he wishes, and insure that the condition of the subject, as well as the circumstances in which he is placed, are those we have found to be most propitious for arm-lifting. Moreover, suppose that we watch him lift his arm, then avert our eyes for a moment, and, subsequently, see him lift his arm again. In this case, the conditions of temporal propinquity, circumstantial variety, agent similarity, and simple frequency would certainly be satisfied.

Consequently, we would then have sufficient empirical evidence to support the hypothesis that the agent could have lifted his arm during that brief period when we did not see him lift his arm, and, consequently, we would be justified in claiming to know that the hypothesis is true.

Furthermore, this claim would be justified whether or not the agent lifted his arm at the time in question, and, indeed, would be justified even if we knew that he did not lift it. In fact, even if we do not avert our eyes but see that he does not move his arm at the time in question, this in no way detracts from the value of our evidence. Under the conditions we have imagined, the fact that our subject does not lift his arm need provide no evidence whatever to support the hypothesis that he cannot lift it.

The latter claim is surely the crux of the matter. To see that it is justified, let us suppose that we know from what the agent tells us that he did not try or make any attempt to lift his arm. The relevance of such knowledge is this. If the agent tried and failed, that would be evidence that he could not perform the action, but there is a great difference between failing when one has tried and mere nonperformance. If we are able to rule out the hypothesis that the agent tried and failed, and if the condition of the agent as well as the circumstances in which he is placed are those we have found to be most favorable for arm-lifting, then the mere fact that he does not lift his arm would not support the hypothesis that he cannot lift it.

An analogy should help to clarify this point. Suppose that a car is tuned and checked so that it is in perfect operating condition and is then placed in circumstances that must favor good performance. If someone tries to start the car, turns the key, sets the choke, etc., and the car fails to start, that is evidence that it cannot start. On the other band, if no attempt is made to start the car, then the mere fact that the car does not start in no way supports the hypothesis that it cannot start.

Therefore, if we know from our experiment: (i) that the condition of the agent and the circumstances in which we have placed him are ideal for arm-lifting, and (ii) that his not lifting his arm provides no evidence that he cannot lift it, then our experimental empirical evidence is sufficient to justify our claim to know that the agent could have lifted his arm at a time when he did not lift it.

Since there is no impossibility, of any sort involved in our imaginary experiment, there being no logical difficulty involved in actually carrying it out, it follows that it is possible to know empirically that a person could have done otherwise. It seems altogether reasonable to suppose that, were the experiment actually carried out, the results would be approximately what we have imagined them to be. Moreover, the uncontrolled but abundant evidence of everyday life also clearly provides us with empirical evidence sufficient to justify our claim to know empirically that a person could have done otherwise. Indeed, the experiment I have asked you to imagine is not necessary for the attainment of such knowledge, but it is sufficient, and that is the point at issue.

At this point, I wish to consider several objections that might be raised against the preceding argument. In the first place, it might be objected that the empirical evidence that we would obtain from our imaginary experiment would not establish categorically that the agent could have done otherwise, but would rather establish hypothetically only that the agent could have done otherwise, if certain conditions had been different .6 However the experiment would establish both that, at certain times, the subject could have done otherwise, if certain conditions had been different, and that, at other times, the subject could have done otherwise, had the conditions been precisely as they were. It is most important not to obliterate this distinction. For example, we might discover that, when our subject is hypnotized and given certain instructions, he can lift his arm only when he is told to. In such a case, if he does not lift his arm, we would only be justified in asserting hypothetically that he could have lifted his arm if we had told him to. Or, suppose that we instruct him, again under hypnotic control, to lift his arm only if he has decided to do so five minutes earlier. In this situation, if the subject does not lift his arm, we would only be justified in asserting hypothetically that he could have lifted his arm, if he had decided to five minutes earlier. However, if we have just seen him lift his arm under the most favorable circumstances, and, without the conditions being altered in any way, he does not lift his arm now, we would be justified in asserting categorically that he could have lifted his arm — and no ifs about it.