In recent years philosophers have produced arguments designed to prove that not all human behavior can be predicted or otherwise known in advance, and these arguments have been taken to be relevant to the problem of freedom of the will as well as to the question whether there can be genuine behavioral sciences. Specifically, it is argued that in certain circumstances it is logically impossible that one should come to know decisions, and actions for whose occurrence decisions are necessary conditions, in advance of the occurrence of such decisions. This has been interpreted as a refutation of determinism.
Two antipredictive arguments will be presented separately, and later their import when taken together will be discussed. The first concerns the scientific defectiveness of predictions that influence the predicted event, and the second concerns the logical impossibility of a person's knowing now what he will decide only at some future time.
Influence of Predictions
It is a familiar fact that some prophecies and predictions are self-fulfilling in the sense that the prediction itself produces the predicted event—for example, when all the stock market tip sheets predict that stock x will drop sharply in the next few weeks. We also know, for similar reasons, that some predictions are self-defeating. For example, Jones predicts that he will, as usual, take the easy way out of a difficulty, but then, to prove to himself that he can do better, he does just the opposite. This prediction affected his deliberation and caused him to make a decision opposite to the one he had predicted. Now, the argument that follows does not maintain that a person's predictions of his own future decisions are necessarily or always self-defeating; instead, it maintains that it is logically impossible that by considering causes a person should come to know that his final prediction of what he will decide is not self-defeating, and it maintains that the attempt to achieve such knowledge involves an infinite regress. In other words, this antipredictive argument purports to prove that predictions of one's own future decisions on the basis of antecedent causal conditions cannot possibly be scientifically complete.
It is necessary to state some assumptions and restrictions required by the argument. The first assumption is that decisions are events and hence are the sorts of things that can be caused; many philosophers would reject this assumption. Second, the argument concerns only causal knowledge of future decisions, by which is meant predictions derived with scientific adequacy from what one knows to be all the relevant antecedent causes of the decision, as distinct from predictions not known to be based on all the relevant causes and which consequently yield only a likelihood of the decision's occurrence. Finally, the argument aims to prove only that it is logically impossible for a person to have causal knowledge of his own decision in advance of making such a decision.
Let us assume, then, that some set of circumstances C is causally sufficient for a person S to make decision D and that S has unlimited knowledge of past circumstances and relevant causal laws. Can S come to know that C is sufficient for D ? S may come to make a prediction P that past circumstances C are sufficient for D. We have supposed that as a matter of fact C is causally sufficient for D, but S nevertheless cannot know that this is so unless he also knows that there are no contrary causes. That is, before S can know that C is sufficient for D he must also know that there is no other circumstance which, together with C, is sufficient for not-D. One such probable cause of not-D is the prediction itself. Therefore, S cannot know that C is sufficient for D unless he knows that it is false that
(1) C plus P are causally sufficient for not-D.
S has been allowed unlimited knowledge of past circumstances and relevant causal laws, hence S can know that (1) is false, that is, he can know that making the prediction will not cause him to make a different decision. It does not follow, however, that S now can know that C is sufficient for D, for the same problem recurs: S 's knowledge that (1) is false, which we will call P 1, is a new datum and is itself a possible cause of not-D. Therefore, S cannot know that C is sufficient for D unless he knows that it is false that
(2) C plus P 1 are causally sufficient for not-D.
And S 's knowledge that (2) is false, or this knowledge plus his feelings or attitudes toward (2), constitute a further possible contrary cause, P 2. Thus, an infinite regress arises, within which the agent's prediction on the basis of some evidence C or his revision of the prediction or his final thoughts about the prediction are relevant data in addition to the data upon which the prediction was based. S 's calculating of causes cannot possibly "catch up" with the number of possible causes that must be examined if the prediction is to be scientifically complete, for the final results obtained cannot themselves also be part of the basis of one's prediction.
When one attempts to predict a supernova, it is true that in this case, too, the final prediction arrived at is necessarily excluded from the data upon which the prediction is based. However, although it is logically possible that predictions or thoughts about predictions can produce or impede a supernova, it is not scientifically possible that they do so. Therefore, the infinite regress argument is no obstacle to knowledge of, for example, scientific laws or stellar events but concerns only particular events that can be produced or prevented by human agency. And it is clearly applicable to attempted predictions of one's own decisions because we know that speculations and predictions about what one is likely to decide are always among the conditions most likely to be determinative of what one will in fact decide.
counterarguments favoring determinism
The view that this first antipredictive argument casts doubt on determinism may be challenged in a number of ways:
(a ) The argument presents no obstacle to the existence of a complete causal explanation of one's own past decisions.
(b ) There is no logical obstacle to a person's predicting a future decision of someone other than himself, although such prediction does confront a methodological difficulty. That is, suppose that A predicts a future decision of B 's and resolves not to tell B the prediction. Then it appears that A must also predict something about himself; namely, that he will not later decide to revoke his past decision and tell B, after all—and this, according to the infinite regress argument, A cannot possibly do. One complication here is the question whether the regress argument precludes A 's predicting that he will make no decisions at all during a certain future period; if the regress argument does not preclude this, then A can predict that he will not change his mind and tell the original prediction to B. But in any case the solution seems to lie in having A make his prediction of B 's decision from a dungeon or a distant planet or in such a way that he has no time to communicate with B in advance of B 's making his decision; that is, perhaps it is sufficient that it be physically (although not logically) impossible that A should ruin the impeccable scientific basis of his prediction by telling B.
(c ) The regress argument shows no peculiarity of human or even of sentient beings. For it is easy to imagine a simple machine, for which no one would dream of claiming free will or moral responsibility, the behavior of which could not possibly be predicted in circumstances similar to those previously described. We need only suppose that the machine can do two things, x and y, that a prediction of either of these things, punched into a card, can be inserted in the machine, and that we announce our predictions of what the machine will do by inserting appropriately punched cards into the machine. The machine is built to do x when fed the prediction "machine will do y " and to do y when fed the prediction "machine will do x." The situation in which a prediction of a person's decision is defective is fully as artificial as this, and in each situation the prediction is defective for the same reason. In each case, given the causal hypothesis, one can in principle make a scientifically impeccable prediction of what will occur only if neither the person nor the machine is allowed to be influenced by the prediction. Meaning "y " when one inserts the card saying "machine will do x " into the machine is equivalent to telling a person he will decide not-D when one knows that telling him this will cause him to decide D.
It can be argued that the first antipredictive argument shows only that given the causal hypothesis, it is still possible to make predictions competently and incompetently and that one of countless ways in which one can make predictions incompetently is to allow one's prediction to disturb the system that one is trying to predict. However, although it may be the case that the self-defeating prophecy and the self-fulfilling prophecy are equally explicable and, in general, equally avoidable phenomena, it appears that the special situation in which the self-defeating prophecy is unavoidable is important to us—namely, the situation in which we attempt to predict our own decisions. The regress argument also poses a methodological problem for social scientists who wish to circulate predictions of human behavior, but it does not show that there is any event that in principle cannot be predicted.
Logical Impossibility of Self-Prediction
The second antipredictive argument appears to follow from the analytic truth that one cannot know now what, by hypothesis, one will not know until some later time. Thus, one form of this argument (see Karl Popper, "Postscript: After Twenty Years") maintains that exact historical prophecy is incompatible with the fact of advancing knowledge. That is, it is impossible to predict the future decisions and actions of people because these future decisions and actions will be formed and done on the basis of knowledge that, by hypothesis, no one now possesses.
Another form of the argument maintains that it is logically impossible for a person to know what he will decide to do before he actually makes his decision (see Stuart Hampshire, Thought and Action ; Carl Ginet, "Can the Will Be Caused?"; and D. F. Pears, Freedom and the Will ). It is claimed that if a person knows or thinks he knows what he will try to do tomorrow, then either he has already decided what he will try to do or he believes that what he will try to do is not up to him. In neither of these two cases can he decide what he will try to do, for in each case there is nothing for him to decide. Decision is making up one's mind about what one will try to do or about what one will acquiesce in; therefore, to say that one will decide tomorrow appears to entail that there is something one will know then and which, by hypothesis, one does not know now.
However, there is a difficulty here. What is it that one knows as a result of decision and that one cannot know prior to the decision? From the fact that a person has decided to do something, it does not follow that he knows what he will do or try to do in the future. Decision does not give one knowledge of anything that will occur in the future because the mere fact that a person has decided does not ensure that he will not falter, change his mind, or die tomorrow. Hence, it appears to be mistaken to assume that because decision entails ignorance prior to decision, this ignorance is of something which one will know later as a result of decision; what one comes to know when one decides is nothing in addition to the decision itself and not any fact about the future. The reason for this appears to be that "decision" is an intentional concept.
Sometimes a person claims to know what a future decision of his will be, and various explanations of his supposed mistake can be made: (a ) He has already decided, and he confuses with the act of decision itself some future reaffirmation, announcement, or implementation of his decision. (b ) He has tentatively decided and plans at the last moment to reappraise his decision, but he thinks that he knows the result of that reappraisal because of his tentative decision. In this case, if he does not deliberate again at the last moment, then he merely reaffirms what he has already decided, and if he does deliberate again, then it is impossible that he should know in advance the result of his deliberation, even though this new decision agrees with his earlier tentative decision. (c ) He construes a future reaffirmation of a decision already made to be a new decision because its time, place, or context differs from that in which he first decided. (d ) He confuses a guess, likelihood, or probability with knowledge of his future decision.
It has also been claimed (for example, by Richard Taylor, in "Deliberation and Foreknowledge") that if a person knows or thinks he knows what he will do in the future, then it is impossible for him to deliberate about what he will do, for deliberation also presupposes ignorance. "Jones is deliberating whether to do x " entails "Jones does not know whether or not he will do x." But here a distinction must be made between the agent's belief or knowledge that he will do a particular act in the future and the agent's belief or knowledge that this particular act he will do is in some sense not up to him. If a person believes that he will do x, he cannot deliberate whether to do x, even though he believes that he will do x freely, that what he does is up to him. On the other hand, if a person believes that what he will do is not up to him, then he cannot deliberate whether to do x, even though he lacks knowledge or belief about what he will do. Hence, although it has been claimed that both foreknowledge and lack of freedom preclude deliberation and decision, these claims nevertheless require separate argument, and only foreknowledge is relevant to self-prediction and the paradoxes thereof.
It might be thought that the two antipredictive arguments are not truly distinct, and indeed some philosophers have written as though these arguments were but two approaches to the same logical point. But they are distinct, except insofar as they can be put to similar purposes. The first argument applies to all predictions that can causally influence the events predicted, whether these events happen to be decisions, revolutions, or stock market trends. It is thus broader in scope and does not require that the event also be of that special sort which, in certain circumstances, is logically impossible to know in advance. The second argument attacks the very idea of foreknowledge, however obtained, of occurrences that entail prior ignorance and does not, as does the first argument, attack the scientific adequacy of predictions that can influence the predicted events.
Logical Impossibility of Causing Decisions
Many philosophers would maintain that if some set of antecedent conditions is causally sufficient for the occurrence of an event, then it is logically possible that the event be predicted or known prior to its occurrence. From this claim, together with the second antipredictive argument, can be constructed the following argument that attempts to prove that it is logically impossible that decisions have causes (see Ginet, op. cit.): If it is logically possible for a decision to be caused, then it is logically possible for a person to know what his own decision will be before he makes his decision; it is not logically possible for a person to know what his own decision will be before he makes his decision; therefore, it is not logically possible for a decision to be caused.
This argument is, in the following way, of more apparent relevance to the traditional problem of freedom of the will and in particular to a theory of human agency: Let us suppose that decisions are necessary conditions for the occurrence of certain actions, and let us suppose further that decisions are part of the causes of such actions. If so, then any set of causes sufficient for the occurrence of such an action must include a decision as part of the set, for whatever is sufficient for something to occur must include everything necessary for that thing to occur. But the decision, by the preceding argument, is uncaused, and therefore no set of causes existing prior in time to the decision can be sufficient for the occurrence of the action. The decision can thus be viewed as a partial, uncaused cause of the action, which, together with ordinary causes, is sufficient for the occurrence of the action.
Difficulties of the following sort have been raised against the argument that maintains that it is impossible that decisions be caused: First, it has been doubted that it follows from the causal hypothesis that it is possible for a person to predict his own decisions; for the possibility of predictability in principle need not include the possibility of predictability in all possible circumstances (see A. J. Stenner, "On Predicting Our Future"). As we have seen, it is not obvious that paradoxes arise when we suppose someone to predict decisions of persons other than himself. Second, a premise of this argument maintains that from the hypothesis that decisions are caused, it follows that one could in principle make a scientifically adequate prediction, based on knowledge of antecedent causes, of one's own future decision. But the first of the two antipredictive arguments claims that this does not follow at all, because it is impossible to establish that one's prediction has no contrary influence on the predicted event. That is, the first antipredictive argument, if sound, shows that the causal hypothesis does not entail the apparent absurdity that in principle one could, by considering antecedent conditions and relevant causal laws, come to know one's own decisions in advance.
Canfield, J. "Knowing about Future Decisions." Analysis 22 (6) (1962): 127–129.
Fain, Haskell. "Prediction and Constraint." Mind 67 (1958): 366–378.
Ginet, Carl. "Can the Will Be Caused?" Philosophical Review 71 (1) (1962): 49–55.
Hampshire, Stuart. Thought and Action. London: Chatto and Windus, 1960.
Hampshire, Stuart, and H. L. A. Hart. "Decision, Intention and Certainty." Mind 67 (265) (1958): 1–12.
Herbst, P. "Freedom and Prediction." Mind 66 (1957).
Lucas, J. R. "Freedom and Prediction." Part 1. PAS, Supp. 41 (1967): 163–172.
Lyon, A. "The Prediction Paradox." Mind 68 (1959): 510–517.
MacKay, D. M. "Brain and Will" (BBC Third Program, 1957). Printed in A Modern Introduction to Philosophy, edited by Paul Edwards and Arthur Pap. New York: Free Press, 1965.
MacKay, D. M. "On the Logical Indeterminacy of a Free Choice." Mind 69 (1960).
Moran, Richard. Authority and Estrangement: An Essay on Self-Knowledge. Princeton, NJ: Princeton University Press, 2001.
Moran, Richard. "Self-Knowledge: Discovery, Resolution, and Undoing." European Journal of Philosophy 5 (1997): 141–161.
Murdoch, Iris, Stuart Hampshire, Patrick Gardiner, and D. F. Pears. "Freedom and Knowledge." In Freedom and the Will, edited by D. F. Pears, 80–104. London, 1963.
Perry, David. "Prediction, Explanation, and Freedom." Monist 49 (1965): 234–247.
Popper, Karl. "Postscript: After Twenty Years." In his Logic of Scientific Discovery. London: Hutchinson, and New York: Basic, 1959.
Rabinowicz, Wlodek. "Does Practical Deliberation Crowd Out Self-Prediction?" Erkenntnis 57 (2002): 91–122.
Schlesinger, George. "The Unpredictability of Free Choices." British Journal for the Philosophy of Science 25 (1974): 209–221.
Stenner, A. J. "On Predicting Our Future." Journal of Philosophy 61 (14) (1964): 415–428.
Stocker, Michael. "How to Prevent Self-Prediction." Journal of Philosophy 65 (1968): 475–476.
Taylor, Richard. "Deliberation and Foreknowledge." American Philosophical Quarterly 1 (1) (1964).
Andrew Oldenquist (1967)
Bibliography updated by Benjamin Fiedor (2005)