(Originally published in De Philosophia, Vol. XV, No. 1, Spring/Summer 1999)
Causal determinism is the thesis that all events are causally necessitated by prior events,
so that the future is not open to more than one possibility. It seems to be equivalent to the
thesis that the future is in principle completely predictable (even if in practice it
might never actually be possible to predict with complete accuracy). In fact, one sometimes sees
definitions of determinism that include the notion of predictability.1
I will refer to this concept
of predictability as "absolute predictability," meaning that a completely accurate prediction
of any future event could in principle be given, as in the famous example of Laplace’s demon.
At first, the equivalence between determinism and absolute predictability might appear
unproblematic. If determinism is true, then there is a description of the state of the world at any
given time t1 and a finite set of laws (L1, L2, ..., Ln), both of which appear to be knowable in
principle, and which combined are sufficient
to allow for the prediction of the state of the world at any future time t2. But when we consider
such a notion of predictability, certain paradoxes arise, as can be seen in the following thought
experiment:
Suppose that causal determinism is true and that there exists a foolproof means by
which you can predict the future with absolute accuracy. For instance, suppose you have at your
disposal a super computer that contains the complete data of a given state of the world, and
the complete laws of physics. You can put questions to the computer (e.g., regarding the
outcome of a coin flip), and of course its answer is always correct. But now suppose you ask the
computer about your own future actions. You might ask, for example, whether at time t2 you will
drop the coin you have in your hand. The computer makes its calculation and you are informed
that you will not drop the coin at time t2. Now here’s the interesting question: Is there anything
preventing you from in fact letting go of the coin at time t2? Are there any laws of physics that,
under the given circumstances, would make it impossible for you to do so? It certainly doesn’t seem
so. Under ordinary
circumstances (and there is no reason in this thought experiment to assume anything other than
ordinary circumstances), one can let go of a coin at will. But if you could in fact drop the coin at t2
then you would have defied the prediction. And since it was assumed that the computer could
predict with absolute accuracy, we are left with a contradiction. So either you cannot defy the
prediction (something would in fact have to prevent you from releasing the coin) or there cannot
be such a thing as an absolutely accurate prediction, at least under the above stipulated
conditions.
There are actually at least three possible responses to this dilemma. One response is to
claim that the above thought experiment demonstrates the impossibility of determinism (and by
implication shows the existence of freedom). The reasoning behind this response is as follows.
The notion of determinism implies absolute predictability. That is, if determinism were true, then
the above prediction would in principle be possible. But absolute predictability can conflict with
our ability to perform certain ordinary actions, as in the above thought experiment. And we have
no reason to suppose that we cannot perform such ordinary actions. Therefore determinism must
be false.
A second response is to claim that if determinism is true and we found ourselves in the
situation described by the thought experiment, then we would in fact be unable to drop the coin
at the appointed time. According to proponents of this view, the first response merely begs the
question. For if in fact determinism is true, then it would be the case that at time t2 we would be
unable to perform that particular action. The fact that we don’t know of any conceivable reason
why we could not perform that action at t2 merely shows that we don’t know all of the things that
go into determining our actions. After all, in actuality we are not in a position to be able to predict
the future with absolute accuracy. If, on the other hand, there really were such a computer as the
one described in the thought experiment, then it would know everything that goes into
determining our actions, and would therefore know why we would be unable to let go of the coin
at t2.
But is such a thing possible? Consider what it would be like to be in the situation of this
thought experiment. Would you feel yourself unable to act on your desires? Is it possible to feel
one’s own lack of freedom? It has been claimed that such a state of affairs is in fact possible.2
According to the argument, we have no idea what it would be like to be in such a situation only
because we have never experienced such a state of affairs, never having had our actions
predicted for us with absolute accuracy.
I think that this argument fails to convince, however, and for a very simple reason: We
have much better grounds for believing in our ability to act however we please than we have for
believing in determinism. Given a choice between the first and second responses, then, the first
certainly seems preferable.
But there is, as already mentioned, a third possible response to the dilemma imposed by
our thought experiment. Notice that both the first and second responses accept the idea that
determinism implies absolute predictability. The first response uses this fact to deny the reality of
determinism by means of denying the possibility of absolute predictability (at least under the
circumstances of the thought experiment). The second response claims that there could in fact
be absolute predictability, and that in such a case we could find ourselves in situations where
our lack of freedom would be apparent to us. A third response is to claim that determinism does
not in fact imply absolute predictability, and thus avoid the paradox of the prediction altogether.
For if, even given determinism, the situation described by the thought experiment could never
arise, then there would be no need to deny either the possibility of determinism or the
impossibility of our ever being aware of our own lack of freedom.
To see why absolute predictability is impossible, even if determinism is true, we will first
consider a second argument that is intimately connected with our thought experiment. Suppose
you attempt to predict what you will do at t2, and that your prediction is completed (i.e., you are
aware of the result of the calculation) at t1. Prior to t1, when you were calculating what you would
end up doing at t2, you did not yet know what the prediction would be. But at and after t1, you
would have this new datum, and that could certainly have an effect on your actions at t2. Thus
your action at t2 could be partially caused by something (namely, the information contained in the
prediction) which the calculations could not have taken into account. Therefore, the prediction is
necessarily incomplete, and might be incorrect.
The situation in our original thought experiment is analogous. The computer’s prediction
cannot take into account what effect the prediction itself might have on you. Therefore, the
prediction is not based on everything that can go into causing your actions at the predicted time,
and so might be incorrect. And this explains why you can defy the prediction.
This second argument has also been used as an argument against determinism, for
similar reasons as the first thought experiment. If one maintains that determinism implies
absolute predictability, and concludes from the above argument that absolute predictability
cannot take place, then one has shown determinism to be impossible. But there is no reason to
maintain that determinism must imply predictability, as my own exposition of the argument (which
assumes determinism) demonstrates.
Likewise, the argument has also been denied by some who wish to maintain both
determinism and that determinism implies absolute predictability. It has been claimed that there
are ways of arriving at a prediction that avoid any possible problems arising from effects of the
prediction itself. First, in any given case, the prediction information will not necessarily alter the
original prediction. It might simply have no effect on the outcome. Second, in those cases where
the prediction does have an effect on the predicted event, it can nevertheless be self-fulfilling,
and if one arrived at such a prediction, then one’s forecast would be correct. And there is nothing
in the above argument to rule out such a possibility. Third, and most importantly, consider the
fact that, given determinism, the state of the world is completely determined by prior states,
including those obtaining before t1. Since that is undeniable, it seems possible that one could
arrive at a prediction of the state of the world at t2 based on the conditions prior to one’s own act
of prediction, and thus not have to concern oneself with the effects of the state of the world at t1.
The events at t1 cannot alter the fact that the situation at t2 is completely determined by events
occurring before t1.
Let's consider each of these rejoinders in turn. The first simply does not apply to the
original thought experiment, for in the original scenario we considered those situations where
one’s intent is to defy the prediction. Thus, the original thought experiment concerns only cases
in which the prediction information does have an effect on the outcome.
Let us therefore consider whether one can arrive at a correct prediction by taking into
account what effects the initial prediction will have and adjusting for these (and perhaps that way
arriving at a self-fulfilling prophecy). Suppose you program the computer to proceed as follows.
First, it predicts what you will do if you are not informed of the prediction. There is nothing
preventing the computer from successfully doing this, given the conditions of the thought
experiment. Suppose it predicts you will hold on to the coin. The computer can then perform a
hypothetical prediction based on (a) the state of the world at t1 and on (b) you receiving the
information at t1 that you will not drop the coin at t2. The computer will then have successfully
predicted whether, upon being informed of the original prediction, you will still go ahead and hold
on to the coin, or whether you will decide to defy the prediction. If it predicts you will hold on to
the coin, no problem: it informs you of this, you confirm the prediction, and all is well. This is one
case in which the prediction might be considered self-fulfilling. But note that again this can only
take place in those cases where you do not decide to defy the prediction.
So what if instead the computer predicts that, upon being informed that you will not let go
of the coin, you decide to defy the prediction? The computer has in fact correctly predicted what
you will do, but once again it cannot inform you of that prediction. For suppose that it tells you of
this new, adjusted prediction: that you will in fact drop the coin. Then there would have to be yet
another prediction as to what you will do based on this new information, and the computer would
be back at the starting point. Or, to make things simpler, since we are concerned only with those
cases in which your intent is to defy the prediction, and the computer will always predict this, it is
clear that the computer cannot inform you of its final prediction, for then that would become the
actual prediction you are given, which you will defy. The second rejoinder therefore also fails to
get around the dilemma.
We still have to explain, however, why it isn’t possible for someone to predict one’s own
actions at t2 by working from a prior state of the world that has nothing to do with the prediction
itself. This was the third rejoinder to the above argument, and it appears much more formidable,
for it certainly seems that such a prediction ought to be possible.
In order to properly deal with this objection, let’s consider a third and final argument
related to our thought experiment. In order for the computer to arrive at an absolute prediction, it
must have a complete description of the state of the world at some particular time t0. But can the
computer, or anything else, ever contain a complete description of a given state of the world? I
believe it cannot. For in order to have complete knowledge of the state of the world at t0, the
computer would have to have complete knowledge of its own state of knowledge at t0, since its
own state is part of the total state of the world. But how can it do this without a problematic
infinite regress? Suppose the entire state of the world at t0 consists of the complex fact
(f1, f2, ..., fn) plus the state of the computer. Then in order to have a complete description of
the world, the computer would have to know its own state in the act of knowing (f1, f2, ..., fn),
which we'll represent as (fn+1). So what it would have to know is not just (f1, f2, ..., fn) but
(f1, f2, ..., fn, fn+1). But then it would also have to know its own state in knowing (fn+1),
and so on, ad infinitum.
Nor does it help to consider the possibility that the computer might have knowledge of some
state of the world prior to the time when the computer has this knowledge, for then
it would still have to be explained how such knowledge came about. Either the state of the world
is somehow available at the time it takes place, or it must be inferred by working backwards from
a later state. And in order for the latter to be possible, the later state would itself have to be
completely known.
Since nothing could have complete knowledge of the state of the world at a given time t0,
it is impossible for anything to ever predict the future with absolute accuracyeven in cases
where the prediction does not in some way interact with the predicted event. But, as is evident
from the fact that my expositions of the above situations assumed determinism, none of the
above rules out the possibility that determinism is true. Therefore, determinism does not require
absolute predictability.
Notes
1 For example, Karl Popper, The Open Universe, (London: Hutchinson & Co., 1956, 1982),
pp. 1-2, defines 'scientific' determinism as "the doctrine that the structure of the world is such
that any event can be rationally predicted, with any desired degree of precision, if we are given a
sufficiently precise description of past events, together with all the laws of nature" (italics in
the original).^
2 See for instance David Wiggins, "Freedom, Knowledge, Belief and Causality," in Knowledge
and Necessity, Royal Institute of Philosophy Lectures,1968-9 (Cambridge, 1970).^