Van Fraassen's critique of inference to the best explanation

Citation: Okasha, S. (2000) Van Fraassen's critique of inference to the best explanation. Studies in History and Philosophy of Science Part A, 31(4), 691-710 (RSS)
Internet Archive Scholar (search for fulltext): Van Fraassen's critique of inference to the best explanation
Tagged: Philosophy (RSS)

Summary

Okasha introduces the method of Inference to Best Explanation (IBE) and a overview of the critiques that van Frassen has for this method. IBE uses induction to determine the most likely explanation or the explanation that provides the most understanding. IBE is useful in that it can distinguish between multiple hypothesis and can be used to make vertical inferences, which previous modes like Hempel's DN model, cannot accomplish. An example of a vertical inference is the hypothesis that the red shift of one star will occur a certain way because a similar red star also behaved in that way. Sherlock's process of deducing explanations of an event (Moriarty committed the crime), is actually IBE in disguise. It is entirely possible that another person committed the crime, but Sherlock can infer based on past behavior, that the likely and loveliest explanation is Moriarty was the perpetrator.

Critiques of IBE have focused on the general avoidance of inductive methods in favor of deductive and calculative methods. Van Frassen, known for pragmatism, worries that this method does little more than force us to choose the best bad answer of a lot of explanations, working under the assumption that IBE's inclusivity does not allow us to remove 'bad' explanations from the pool of potential explanations, making it likely that we will choose that poor explanation. Van Frassen instead, places IBE in the role of justification, essentially as a way of testing other modes of explanation and not an acceptable mode on its own.

According to Van Frassen, IBE is also poorly suited to statistical methods, particularly Bayesian. Okasha reminds us that, contrary to Van Frassen's claim that IBE can't play nicely with statistics, is the fact that IBE is closely related to hypothetico-deductive methods, which Bayesian statisticians regularly reconstruct using probability studies. HD methods hypothesize an explanation, then use probability to deduce the likelihood that the original hypothesis was correct. Okasha states this very fundamental similarity makes IBE particularly suited for Bayesian statistics, and is not a limiting factor.