ABSTRACT
The hypothetico-deductive model has three weaknesses. It neglects the context of discovery; it is too strict, discounting all relevant evidence that is compatible with the hypothesis but not entailed by it; and it is overpermissive, counting some irrelevant data as relevant. In the last chapter, I argued that Inference to the Best Explanation does better in the first two respects. In this chapter, I argue that it also does better in the third. The problem of over-permissiveness arises because there are consequences of a hypothetical system that do not support the hypothesis, and there are several related ways to generate such consequences. One is by strengthening the premise set. An observed consequence of a hypothesis about childbed fever may support that hypothesis, but it will not support the conjunctive hypothesis consisting of the fever hypothesis and some unrelated hypothesis, even though the conjunction will of course also entail the observation (Goodman 1983: 67–8). Similar problems arise with the indiscriminate use of auxiliary statements. At the limit, any hypothesis is supported by any datum consistent with it, if we use a conditional auxiliary whose antecedent is the hypothesis and whose consequent is the irrelevant datum. This places the hypothetico-deductivist in a bind: to meet the objections in the last chapter that some supporting evidence is not entailed by the hypothesis, he will have to be extremely permissive about the sorts of auxiliaries he allows, but what he then gains in coverage he pays for by also admitting irrelevant evidence. A second way to generate irrelevant consequences is to weaken the conclusion. For any consequence of a hypothesis that supports the hypothesis, there are also innumerable disjunctions of that consequence and the description of an irrelevant observation, and the truth of the disjunction can then be established by the irrelevant observation. ‘All sodium burns yellow’ entails that either this piece of sodium will burn yellow or there is a pen on my table, but the observation of the pen is irrelevant to the hypothesis. Third, as Goodman has shown, by using factitious predicates we can construct hypotheses that are not lawlike, which 92is to say that they are not supported by the instances they entail (1983: 72–5). A grue emerald observed today does not support the hypothesis that all emeralds are grue, where something is grue just in case it is either observed before midnight tonight and green or not so observed and blue. Finally, we may generate apparently irrelevant consequences through contraposition: this is Hempel’s raven paradox. The hypothesis that all Fs are G entails that this F is G, a consequence that seems to support the hypothesis. Since, however, this hypothesis is logically equivalent to ‘All non-Gs are non-F’, it also entails that this non-G is non-F, a consequence that does not seem to support the original hypothesis. ‘All ravens are black’ is logically equivalent to ‘All non-black things are non-ravens’, so the raven hypothesis entails that this white object (my shoe) is not a raven, but observations of my shoe seem irrelevant to the hypothesis (Hempel 1965: 14–15).
