Justifying theories III: Inference to the best explanation
The basic problem facing both falsificationists, such as Popper, and inductivists is that we appear to need to make ampliative inferences that take us from evidence about a body of data to claims that go far beyond that evidence.
We would like to be justified in thinking that these claims are likely to be true. The problem of induction suggests that we have no such justification; Popper's response is unsatisfactory in part because it declares that we don't need such a justification. Is there another way out?One possibility that has been explored by philosophers of science in recent years is that neither induction nor conjecture is the best way to understand what we are doing when we move from data to theory. Instead, the American philosopher Gil Harman suggested, what we are doing when we construct a theory on the basis of data is that we are trying to find the theory that best explains our data. So this view of the relationship between data and theory is called “inference to the best explanation”; I'll call this suggestion the “ITBE model” for short.
Let's consider Mendel's experiments again. What Mendel noticed was a series of patterns in the results of plant-breeding experiments. For example, if you crossed a white-flowering pea with a red- flowering one, you sometimes got just pink offspring and sometimes you got both red and pink. Furthermore, when there were red and pink flowers in the offspring, the plants that had them came in about equal numbers. What Mendel showed was that if you supposed that red plants were either RW or RR and that white plants were WW, you could explain these results. So he proposed his theory of genes, according to the inference to the best explanation model, as the best explanation of the data.
As a result, the ITBE model must draw on a theory of explanation.We saw earlier that, on the DN view of explanation, if a theory explains the data, then the occurrence of the data could have been predicted (given a description of the initial conditions). This is because the explanandum is a logical consequence of the theory and the specification of the initial conditions. On the DN model, then, a body of data is explained by any theory from whose laws it can be derived, provided that the theory is true. This last proviso was Hempel's “empirical adequacy condition.” Hempel insisted on this condition because any finite body of data can be shown to be the logical consequence of an indefinitely large number of incompatible theories. (And, of course, being finite beings, we always have a finite body of data.) His idea was that you had an explanation only if you had a true theory from which your explanandum could be derived.
But now you can see that we can't use Hempel's account of explanation if we are going to use the ITBE model. For Hempel's empirical adequacy condition means that we have an explanation of something only if the theory is true. But then we couldn't use the ITBE model to give us reason for believing that a theory was true because we'd have to know that the theory was true before we could tell whether it provided any explanation (never mind the best explanation) of the data; thus we'd have to know whether it was true in order to find out whether we had an explanation that gave us a reason to believe it was true! So we had better drop the empirical adequacy condition. Instead, then, of requiring that a candidate explanation relies on a true theory, we can say that a candidate explanation is one that would explain the explanandum if it were true.
Then the ITBE model amounts to this: you have a reason to believe a theory T if you can derive a true explanandum, E, from T's laws (and a specification of initial conditions) and this derivation provides the best available explanation of E.
The major task for the ITBE model is thus to specify how we are to compare explanations in order to decide which of a class of candidate explanations is the best. And the right way to do that is to give some criteria for deciding which of two explanations is better, since if there is a best available explanation, it will just be the explanation that's better than any others that are available.Two criteria for preferring explanations that have been proposed are simplicity and power. Using simplicity as a criterion means that if you have two candidate explanations for a phenomenon, the simpler one provides the better explanation, and (according to the IBTE model) the theory it uses is thus more likely to be true. It's not entirely obvious what it means for one explanation to be simpler than another. But there is an old principle, known as Ockham's Razor (which is named for the fourteenth-century English philosopher William of Ockham), that says you should not multiply entities beyond necessity. What it means, in effect, is that if you can construct a theory without postulating an entity, then you should do so. So we could follow this lead and argue that an explanation that appeals to fewer entities (and is, presumably, therefore less complex) is simpler than an explanation that appeals to more.
As for explanatory power, a theory is more powerful if it explains more phenomena (or more kinds of phenomena) than another. So an explanation E that uses a theory T is preferable to an explanation E' that uses a theory T' if T explains more phenomena (or kinds of phenomena) than T'.
Notice that both Popperians and inductivists will accept this latter claim. For a theory that we know explains a wide range of phenomena has been exposed to a wide range of potential falsifica- tions—which will satisfy the Popperians that it is corroborated—and has a large number of supporting instances—which will please the inductivists.
But the ITBE model does not hold that a theory covering a wide range of phenomena gives a better explanation because it is more likely to be true: rather, it holds that the theory is more likely to be true because it provides a better explanation. This must be so if the ITBE model is to be a competitor to inductivism and fal- sificationism.To see why, consider whether the ITBE model is a real alternative to inductivism. We can argue by reductio. Suppose the ITBE theorist agrees that the reason that an explanation E is better than an explanation E' is that E has greater inductive support than E'. Then, while it might then be true that a good explanation gave you reason to believe the theory that it used, this would only be because the theory already had good inductive support: and then that would be the real reason why the explanation gave you reason to believe the theory. So if the ITBE model is to be a competitor to induc- tivism, it must deny that the reason that E is better than E' is that E has greater inductive support. (A similar argument shows that the ITBE theorist must deny that the reason that an explanation is a better explanation is that it is more highly corroborated.)
This fact draws attention to a first major challenge for the ITBE model. Why should the fact that a theory would provide a simple or a powerful explanation if it were true be reason to believe that it is true? Aren't we at risk of making the assumption we rejected when discussing verificationism in 2.6, namely, that the universe is organized for our epistemic convenience? After all, some very complicated theories—the quantum theory, relativity theory, the DNA theory of inheritance—are now believed to be correct. So why assume that simplicity is a sign of truth? Isn't it an empirical question whether or not the universe is simple? And if so, doesn't the ITBE model just stack the cards in favor of a particular answer to that empirical question?
Similarly, why should the fact that an explanation covers a wide range of phenomena that we have looked at be grounds for thinking it is true? The ITBE model, recall, denies that inductive evidence gives grounds for believing a theory.
So it can't rely on the idea that a powerful theory has lots of confirming instances. And it denies that corroboration gives grounds for believing a theory. So it can't rely on the idea that a powerful theory has survived a wide range of possible disconfirmations. Once more, we can say that there is ample empirical evidence that some powerful theories are false: Newtonian physics is false (that is why it was replaced by relativity and quantum theory). Explanatory power is thus clearly consistent with falsehood. So why should we take it sometimes to be reason for thinking that a theory is true?So the ITBE model has some work to do to explain why a theory's providing good explanations is grounds for thinking it is true. And there's another set of problems for the ITBE model: simplicity and power seem to pull in opposite directions. You can usually make a theory more powerful by making it less simple. For one of the easiest ways of expanding a theory to account for more phenomena is to add to the theoretical entities that it makes use of. (Chemical theories, for example, gained explanatory power as new elements were postulated, producing a chemistry that had greater explanatory power but that was also, at the same time, more complex.) So a second major challenge for the ITBE model is how to decide whether to put more weight on simplicity or on power.
The ITBE model has a certain plausibility. It does seem right to say that one reason for believing that there are genes, which behave as Mendel proposed, is that this hypothesis provides a simple, powerful explanation for a great range of data about biological inheritance. Certainly, as I said when I was introducing Mendel's theory, that's one of the reasons why people came to believe it. And, more generally, scientists often appeal to the simplicity and power of the explanations a theory provides when they are seeking to defend it. But we have seen that there is another possible explanation for this fact, namely, that simple, powerful explanations usually have higher inductive support or greater corroboration. So inference to the best explanation may not be a real alternative to inductivism and falsificationism.
4.13