How to Do Things with an Infinite Regress

Scientific method may be viewed either as an argument justifying a conclusion or as a procedure for finding the right answer to some question. Both conceptions occasion the problem of empirical regresses. According to the former approach, it is hard to say what the point of a regress is. According to the latter, we can solve for the strongest sense of singlemethod performance that could be covered from a regress of procedures. Several types of regresses are solved in this sense. Some of the solutions are shown to have sufficient power to deal with Duhem’s problem. 1 CONFIRMATION AND NATURALISM Here is a familiar way to think about the philosophy of science. Our empirical claims must be justified. Usually, evidence does not and never will entail them, so they must be justified some weaker way. Thus, there must be a relation of partial support or confirmation falling short of full (deductive) support that justifies them. The principal task of the philosophy of science is to explicate the concept of confirmation from practice and historical examples. Any feature of scientific method or procedure that is not addressed to the nature of this relation is extraneous to the philosophy of science per se, although it may be of tangential psychological, sociological, or purely computational interest. Thus virtues such as confirmation, explanation, simplicity, and testing are relevant, but the logic of discovery (procedures for inventing new hypotheses) and computational efficiency are extraneous (e.g., Laudan 1980). After the justifying relation is explicated from historical examples, the obvious question is why it should be that relation rather than another. It is no longer stylish to seek an a priori answer to this question; one responds, instead, with the naturalistic view that if scientific standards are to be justified, that justification must itself be scientific. The next question is how scientific reasoning can justify itself. Somehow, circles are more fashionable than regresses, but without a clear picture of what justification is supposed to accomplish it is hard ∗This paper benefits from helpful suggestions by John Earman, Jim Lennox, and Oliver Sculte.