Holmes’s Inference

Of course, Holmes’s inference is not absolutely airtight. It is possible that Straker is innocent and Martians with hypnotic powers over dogs commit- ted the crime. But that only goes to show that this inference is neither valid nor deductive in our sense. It does not show anything wrong with Holmes’s inference. Since his inference is inductive, it is enough for it to be strong.

Inferences to the best explanation are also defeasible. No matter how strong such an inference might be, it can always be overturned by future ex- perience. Holmes might later find traces of a sedative in the dog’s blood or someone else might confess or provide Straker with an alibi. Alternatively, Holmes (or you) might think up some better explanation. Still, unless and un- til such new evidence or hypothesis comes along, we have adequate reason to believe that Straker stole the horse, because that hypothesis provides the best available explanation of the information that we have now. The fact that future evidence or hypotheses always might defeat inferences to the best ex- planation does not show that such inferences are all bad. If it did show this, then science and everyday life would be in trouble, because so much of sci- ence and our commonsense view of the world depends on inferences to the best explanation.

To assess such inferences, we still need some standards for determining which explanation is the best. There is, unfortunately, no simple rule for deciding this, but we can list some factors that go into the evaluation of an explanation.3

First, the hypothesis should really explain the observations. A good expla- nation makes sense out of that which it is intended to explain. In our origi- nal example, the broken lock can be explained by a burglary but not by the hypothesis that a friend came to see you (unless you have strange friends). Moreover, the hypothesis needs to explain all of the relevant observations. The hypothesis of a mistaken police raid might explain the broken lock but not the missing valuables or the lack of any note or police officers when you return home.

The explanation should also be deep. An explanation is not deep but shal- low when the explanation itself needs to be explained. It does not help to ex- plain something that is obscure by citing something just as obscure. Why did the police raid your house? Because they suspected you. That explana- tion is shallow if it immediately leads to another question: Why did they suspect you? Because they had the wrong address. If they did not have the wrong address, then we would wonder why they suspected you. Without an explanation of their suspicions, the police raid hypothesis could not ade- quately explain even the broken lock.

Third, the explanation should be powerful. It is a mark of excellence in an explanation that the same kind of explanation can be used successfully over a wide range of cases. Many broken locks can be explained by burglar- ies. Explanatory range is especially important in science. One of the main reasons why Einstein’s theory of relativity replaced Newtonian physics is that Einstein could explain a wider range of phenomena, including very small particles at very high speeds.