Reading Comprehension

The justification of inductive inference — the move from observed instances of a regularity to confidence in its continuation — has proved, since Hume first pressed the question with full force in the eighteenth century, one of the more persistent embarrassments of empirical philosophy. The difficulty Hume identified is not the practical one that induction sometimes fails; that it sometimes fails is obvious. The difficulty is that the rule by which we expect the future to resemble the past cannot itself be justified by appeal to the past without circularity, and cannot be justified a priori without making nature into a logical truth.

Two centuries of subsequent philosophy have attempted, in various ways, to find a way around the difficulty. Pragmatist accounts argue that induction is justified instrumentally because we have nothing better and any rational creature would be in the same position; reliabilist accounts argue that induction is justified to the extent that the inductive method actually tracks regularities in the world; Bayesian accounts treat the question as one of updating credences under specified priors. Each has merits within its own framework; none, on the most careful examination, has dissolved the original difficulty in the way its strongest defenders sometimes suggest. The various attempts have, in the main, produced more refined statements of the problem rather than solutions to it.

This is not a counsel of paralysis. Practical induction is unavoidable; no agent could function while withholding all inductive inference, and the question Hume raised has never been a practical recommendation to do so. What the question presses is a question about the kind of warrant induction has — whether it is the rigorous warrant of demonstrative proof, the looser warrant of well-functioning practice, or something in between whose specification has proved harder than the original disputants assumed. To live without an answer is not the same as to live without an inquiry, and the inquiry has yielded, over the centuries, a clearer picture of what an answer would have to do.

What survives, then, is a question whose unsettledness is itself instructive. Some questions in philosophy admit of clean solutions and disappear from the literature once they are solved; the problem of induction is not, on present evidence, one of these. Its continued resistance to dissolution suggests that the warrant of empirical inference is bound up with features of how rational agents must operate in time, with limited information, under constraints whose own justification is not separable from the activity they make possible. To say more would be to claim, on slim evidence, the kind of progress the literature has so far declined to deliver.