Causal Necessity

The Principle of Determinism

We are discussing the argument for a causal theory of perception. One premise of the argument is that every event has a cause. To assess that claim, we need to discuss what a cause is. That has nothing specifically to do with perception, and so the ideas are ones that have come primarily out of the philosophy of science.

There are two main parts to the idea of a cause (of an effect).

1. Whenever the cause occurs, the effect inevitably does too.

Usually, one presumes that the cause occurs before the effect, but that turns out to be unimportant for a general analysis. As a matter of fact, we do sometimes think of a cause as being simultaneous with its effect. (The explosion caused the noise.)

2. This one is hard to state, and indeed, though it is a primary part of our intuitive notion of cause, it is not even clear that it is coherent. It is that the cause causes the effect, the association is not accidental, and somehow some process initiated by the cause gives rise to the effect.

Part 2is intended to rule out accidental associations. In 2003 study of an AIDS vaccine, it didn't work at all on whites, but it appeared to work at least somewhat on nonwhites. That was widely reported in the media and promptly denounced by many AIDS experts. They said that there is absolutely no reason to think that skin color is related to anything that has to do with the efficacy of a vaccine, but that if you massage data with enough different categories, you always find some accidental (random) associations.

Ayer introduces Jean Nicod's (a physicist and philosopher of science) analysis of causation. He suggests formulating the principle of determinism as follows:

Every event e of kind E is a case of an event of some other kind, every instance of which is a case of an instance of E (179).

The "event of some other kind"; is the cause that the principle says every event must have. Thus, for Nicod, a cause of an event e of kind E is an event c of some other kind C such that every event of kind C is a case of an instance of E.

An event may cause many other events, and an event may have many causes, and a kind of event may be caused by some particular other kind in some cases but not others. An event, on Nicod's theory, does not cause another event except insofar as the cause and effect are of appropriate kind, and so the notion of cause doesn't just involve events, it involves kinds, a much more dubious idea. Of course, the same event is of many different kinds, and so it would be important to know more about kinds:

Raising your hand cannot stop a war, but vetoing a UN Security Council Resolution, one would hope, just might stop a war.

Without some restrictions on kinds, this definition just doesn't work: pick some random event that occurred 100000000000 years ago, and say that there is a kind that applies to that event alone. Then every event that occurred since then is a case of an event of that kind, and so that event is a cause of everything that has occurred since then. That seems wrong, and Nicod would eliminate such counterexamples by not allowing single events characterized by their occurrence to be kinds. So consider some kind of event that has occurred, it is a case of every other kind of event that has occurred, since we haven't placed any restrictions on when they can occur. This torpedoes the definition. What to do?

There are other problems: a cause can be of the same kind as its effect, something can be a cause even when it doesn't always work: the posting of a poster can cause me to enter a room (if I want to read that poster), but the posting of a poster certainly doesn't always cause someone to enter a room.

Obvious attempts to fix the definition don't work: requiring the cause to never occur after the effect and requiring it to occur in the neighborhood of the effect still leave too many effects for any given cause.

There is, of course, an obvious kind of solution to the problem: one needs to require that the cause is inevitably or necessarily a case of the effect or that is impossible for the cause to occur without the effect, or the like.

The Animistic Idea of Necessary Connexion

We have introduced an intuitively thin notion of causation: C causes E if whenever things like C occur, things like E occur.

We noted that there is no time direction in this notion, and it can even be that C causes E and E causes C, when they always occur together, on this notion.

We saw that the basic idea has substantial difficulties (not raised by Ayer), in that, for example, things of a kind C that always occur are causes, by this criterion, of everything else.

What seems to be missing is the idea that the cause is in some sense or other responsible for the effect. The most obvious way to picture that is that the cause has some active tendency (Stout) to result in the effect, and the notion of active tendency is one we understand in terms of human tendencies (that is, intentions) to bring about effects. This is a very vivid and appealing idea: I know that certain things have happened because I caused them (for example, I caused—in the precise sense Ayer derives from Collingwood—most of you to hand in papers last week) and that my actions played a real role in the result, and were not just accidentally associated with it---it is no coincidence that I made the papers due last week and that you handed them in last week, one cause of your doing so was my requirement.

We often think of physical causation in similar terms. The tension in the bow causes the bow to elongate once the string is released. It plays an active role. All one ordinarily sees is the shape of the bow changing, not the tension, but we believe that the tension plays the causal role.

None of this improves the situation over Nicod's simple association picture. Why?

I'm going to explain by making a quite different point that isn't strictly relevant because I think it helps in a metaphorical sense in understanding this one.

Lewis Carroll wrote an essay called What the Tortoise Said to Achilles, to make the point that in logic one needs not only axioms, but rules.

The tortoise is trying to convince Achilles of something, say q. Achilles is a little slow.

The tortoise says You believe p right? Achilles agrees. You believe If p then q right. Yes. Then q. Achilles scratches his head, I'm not sure that follows.

After a bit of thought the tortoise realizes he has omitted something from the argument:

p

If p then q.

If, if p then q and p, then q.

Achilles assents to all this, but won't conclude q

p

If p then q.

If, if p then q and p, then q.

If if, if p then q and p, then q, and if p then q, and p, then q.

The infinite regress is clear.

We have the same kind of regress for causation:

String is released.

Causal result, bow shoots arrow.

String is released.

Bow was tense.

Causal result, bow shoots arrow.

String is released.

Bow was tense.

Tension without a countervailing force gets released.

Causal result, bow shoots arrow.

String is released.

Bow was tense.

Tension without a countervailing force gets released.

Some mechanism causes the tension to be released.

Causal result, bow shoots arrow.

No principle of active tendency serves to guarantee that the effect happens as a result. One can always insist on a new principle that that active tendency has certain results, and so we may as well give up at the outset---all these bells and whistles add complexity without helping the explanation. At some point, we just have to admit that we take the association to be causal, and so we expect the effect on the basis of the cause.

What remains is to explain away why this idea of some inevitable process or animating tendency seems so central to causation. One can trace the history of what we now take to be scientific causal explanations back to a belief desire model taken from human experience. For example, in Aristotelian physics, objects fall because they want to be at the center of the earth. Indeed, the whole notion of a physical law, which arises in medieval times, and so we have pretty detailed knowledge of the history, arises out of the notion of a moral law, and so does the idea of laws, as imposed by governments. In the case of moral law, the punishment must follow the transgression. That, Ayer claims, citing Kelsen, is the origin of the idea that the effect must follow the cause.

Just because an idea has an origin in motivations we no longer subscribe to doesn't mean it's a bad idea. Lots of ideas come about in all sorts of strange ways. What matters in deciding whether they are worth keeping is whether they play a useful role today. As we have seen, adding the idea of an animating tendency to our expectations of association does not help in explaining those associations: it just adds another case—we expect associations between the animating tendencies and the results. More complexity, no help.

It seems, unreflectively, that we see that certain things are causes of others, that that is an observed phenomenon that requires explanation, whether or not it helps in explaining something else, namely, apparently inevitable associations. As Hume pointed out, we see objects in configurations, never their tendencies. We are constructed to expect that things of types we have often seen will recur. That is, we are constructed to develop habits. One can explain why that is useful, and why it makes it look like we are seeing tendencies.

We thus have an historical and psychological explanation of why it seems to us that causation involves something more than inevitable association, and reason to see that that idea is of no explanatory use. A reasonable understanding of causation must do without it.

We have also seen that mere association, the picture discussed last class, is inadequate. What to do? Next time we discuss another proposal for accounting for the inevitability of an effect following its cause: the idea is that the statement that the effect occurs follows logically (with no other principles) from the statement that the cause occurs.

Ayer uses properties that are often thought of as causal to guarantee that we have, in my terms not his, common reference to material objects (159-161). But it is not at all clear that such properties can do the job if they do the job because they are causal properties, once we take causation to be thin.

-- ShaughanLavine - 28 Feb 2003 - 3 March 2003 - 27 Feb 2007