ABSTRACT
Scientific hypotheses cannot be deduced from the empirical evidence, but the evidence may support a hypothesis, providing a reason to accept it. One of the central projects in the philosophy of science is to account for this nondemonstrative inductive relation. The justification of induction has been a sore point since the eighteenth century, when David Hume ([1777] 1975) gave a devastating skeptical argument against the possibility of any reason to believe that nondemonstrative reasoning will reliably yield true conclusions (see Induction, Problem of ). Even the more modest goal of giving a principled description of our inductive practices has turned out to be extremely difficult. Scientists may be very good at weighing evidence and making inferences; but nobody is very good at saying how they do it. The nineteenth-century American pragmatist
Charles Sanders Peirce (1931, cf. 5.180-5.212) coined the term abduction for an account, also now known as ‘‘Inference to the Best Explanation,’’ that addresses both the justification and the description of induction (Harman 1965; Lipton 1991, 2001; Day and Kincaid 1994; Barnes 1995). The governing idea is that explanatory considerations are a guide to inference, that the hypothesis that would, if correct, best explain the evidence is the hypothesis
that is most likely to be correct. Many inferences are naturally described in this way (Thagard 1978). Darwin inferred the hypothesis of natural selection because although it was not entailed by his biological evidence, natural selection would provide the best explanation of that evidence. When astronomers infer that a galaxy is receding from the Earth with a specified velocity, they do so because the recession would be the best explanation of the observed redshift of the galaxy’s spectrum. On the justificatory question, the most common
use of abduction has been in the miracle argument for scientific realism. Hilary Putnam (1978, 18-22) and others have argued that one is entitled to believe that empirically highly successful hypotheses are at least approximately true-and hence that the inductive methods scientists use are reliable routes to the truth-on the grounds that the truth of those hypotheses would be the best explanation of their empirical success (see Putnam, Hilary; Realism) because it would be a miracle if a hypothesis that is fundamentally mistaken were found to have so many precisely correct empirical consequences. Such an outcome is logically possible, but the correctness of the hypothesis is a far better explanation of its success. Thus the miracle argument is itself an abduction-an inference to
the best explanation-from the predictive success of a hypothesis to its correctness. Like all attempts to justify induction, the miracle
argument has suffered many objections. The miracle argument is itself an abduction, intended as a justification of abduction. One objection is that this is just the sort of vicious circularity that Hume argued against. Moreover, is the truth of a hypothesis really the best explanation of its empirical successes? The feeling that it would require a miracle for a false hypothesis to do so well is considerably attenuated by focusing on the history of science, which is full of hypotheses that were successful for a time but were eventually replaced as their successes waned (see Instrumentalism). The intuition underlying the miracle argument may also be misleading in another way, since it may rest solely on the belief that most possible hypotheses would be empirically unsuccessful, a belief that is correct but arguably irrelevant, since it may also be the case that most successful hypotheses would be false, which is what counts. Abduction seems considerably more promising
as an answer to the descriptive question. In addition to the psychologically plausible account it gives of many particular scientific inferences, it may avoid weaknesses of other descriptive accounts, such as enumerative induction, hypothetico-deductivism, and Bayesianism. Enumerative inferences run from the premise that observed Fs are G to the conclusion that all Fs are G, a scheme that does not cover the common scientific case where hypotheses appeal to entities and processes not mentioned in the evidence that supports them. Since those unobservables are often introduced precisely because they would explain the evidence, abduction has no difficulty allowing for such ‘‘vertical’’ inferences (see Confirmation theory). If the enumerative approach provides too narrow
an account of induction, hypothetico-deductive models are too broad, and here again abduction does better. According to hypothetico-deductivism, induction runs precisely in the opposite direction from deduction, so that the evidence supports the hypotheses that entail it. Since, however, a deductively valid argument remains so whatever additional premises are inserted, hypothetico-deductivism runs the risk of yielding the absurd result that any observation supports every hypothesis, since any hypothesis is a member of a premise set that entails that observation. Abduction avoids this pitfall, since explanation is a more selective relationship than entailment: Not all valid arguments are good explanations.
