Scientific Explanation

Why penguins can't fly.

As always, we start with the logical positivists.

The logical positivist theory of scientific explanation is called the covering law model of explanation. It has two parts: the inductive-statistical model of explanation (I-S) and the deductive-nomological model of explanation (D-N). The basic model is the D-N model, and the I-S model is, more or less, the D-N model with statistics tacked on.

So, according to the D-N model, a scientific explanation of an event works like this:

You have

No cheating is allowed: the laws actually have to be used, and laws must be truly general laws, not ad hoc rules made specially for the occasion.

There is a serious problem with the theory of D-N explanation: You can not only D-N "explain" the length of the shadow cast by a flagpole in terms of the height of the flagpole, you can also D-N "explain" the height of the flagpole in terms of the length of the shadow.

The problem always illustrated using the flagpole example is called the symmetry problem.

Godfrey-Smith dismissed the D-N model of explanation as wrong and no longer worth considering. He is wrong. What the example shows is that the D-N model is missing something, not that it is wrong so far as it goes. As philosophers like to put it, it may be that the D-N model is a necessary, but not a sufficient condition for something to be a good explanation.

What do we need to add? Godfrey-Smith's favorite extra ingredient is causality. The proposal is that a causal explanation is a D-N explanation in which the circumstances cause the result.

Many philosophers are inclined to reject the idea of causal explanation out of hand, because they don't believe in causation. (That is becoming less true.) Empiricists don't believe in causation, largely for Hume's reasons. I have a different worry about causal explanation. If A causes B, then the occurrence of A explains B. If A always precedes B, then the occurrence of A indicates B, but may not explain it. But that seems to say that what we mean by "A causes B" is neither more nor less than that A is suitable to explain B. I think that the causal model of explanation is correct, but empty. All it says is that explanation has a direction, and we call that direction the direction of causation. The symptom of that is that no one has a decent theory (explanation) of causation. So, the causal theory, though it may be true, isn't terribly useful.

Kitcher and Friedman proposed the unification model of explanation. An explanation unifies the phenomenon being explained with others, it makes it seem expected because it is of a recognized and understood type. Explanations of magic tricks are pretty clearly of this form. Traffic has been explained, rather surprisingly, in that way: traffic is a one-dimensional fluid in which the cars are the molecules. An explanation through unification is a D-N explanation but one in which the laws used and the way in which the explanation is set up is also used in many other explanations. Explanation is reduction of strangeness, seeing that something that can be D-N explained is explained in a familiar and general way. We take the direction of "causality" to just be the direction that leads to a unified understanding. The picture of explanation that I've just suggested, while it was not a part of the official logical positivist theory of explanation, was always there on the sidelines: it was common to comment, for example, that D-N explanations enabled one to explain many things on the basis of a few laws.

How does the unification model handle our main obstacle: the symmetry problem? Kitcher, for example, wasn't so sure that it does, and so he sometimes invoked a mixed model that includes causation. It seems to me that it works in a very straightforward way, and that it works better than the causal theory: We can explain the length of the shadow of a flagpole as part of a perfectly general theory of shadows. That theory is even part of a theory that explains much more: the images produced, not only by opaque objects, but by lenses and mirrors. There is no such theory that goes from shadows to the objects that produced them.

What about my other case, that of a flagpole constructed to cast a shadow of a certain length? In that weird case, explanation does go the other way, but the causal laws of ray optics don't change. What is our explanation in such cases? Whatever it is, it involves my intentions in constructing the flagpole, and it is part of a general theory of how intentions to build things with certain properties result in things with those properties.

Godfrey-Smith says that there are many kinds of explanation. So does van Fraassen. But Godfrey-Smith's point is the Kuhnian one that each paradigm (in the broad sense) brings with it its own notion of explanation. He seems to say that there are no general principles under which all such explanations fall. van Fraassen says something much more surprising: he argues that science isn't in the explanation business at all, that when we give explanations based on science, just as when we design gadgets based on science, we are using science for a purpose outside of science proper.

Godfrey-Smith is certainly right that what counts as an explanation is different in different scientific traditions, but I think that in every case it takes the form of unification of explanations in accordance with as few principles ("laws") as possible. What counts as a law, and what needs explanation differs a lot.

Explanation in Biology—not covering laws, but covering principles.

The last section of the chapter is on the metaphysics of causation. Is cause something over and above regularity?

-- ShaughanLavine - 07 Nov 2005 - 29 Oct 2007 - 31 Oct 2007