Epistemological Pluralism and the Evolution of Scientific Theory — Epoche C2
What Duhem claimed, and what Quine added Pierre Duhem's claim in La théorie physique, son objet et sa structure (1906) is narrower than the slogan it became: the physicist can never submit a single hypothesis to the test of experiment, but only a whole group of hypotheses together, so that when the prediction fails the experiment tells us that at least one member of the group is wrong and never which. Two features of his statement are routinely dropped. The first is that Duhem meant physics. He built the argument out of the structure of physical theory — the instruments themselves are read through theory, so that no observation report is innocent of the group being tested — and he did not extend it indiscriminately to every science. The second is that Duhem did not think the resulting freedom was unconstrained. Faced with a choice of which member of the group to abandon, the physicist is guided by what he calls bon sens , good sense: not an algorithm, not a rule of logic, but a trained judgement that can be argued about, that can be got wrong, and that may take a generation to become compelling to a whole discipline. What W. V. O. Quine added in 'Two Dogmas of Empiricism' (1951) is a different and much stronger thesis, and running the two together under one hyphenated name obscures the step. Quine's holism is unrestricted: the unit of empirical significance is the whole of science; any statement can be held true come what may, if we make drastic enough adjustments elsewhere in the system; and no statement is immune from revision, the law of excluded middle included. Duhem's physicist could not test one hypothesis at a time. Quine's scientist can save any hypothesis whatever. The second claim licenses conclusions the first does not, and this essay's earlier form, like most short treatments, attributed the second to both men. Holism is not underdetermination, and the gap has to be argued The passage from either thesis to the conclusion that evidence never decides between theories is the weakest joint in the standard story, and it is worth seeing exactly where it fails, because everything the essay wants to say about pluralism depends on knowing how much room there really is. Two claims must be distinguished. One is that at any given moment a scientist confronted with a failed prediction has a choice of what to revise. That is Duhem's, it is true, and it is transient: the choices made get tested in turn as the auxiliary hypotheses are put to work elsewhere. The other is that for any theory there exists a rival with exactly the same empirical consequences, so that no possible evidence could separate them. That is the global thesis, and it is the one that would make theory choice evidentially idle. Larry Laudan and Jarrett Leplin, in 'Empirical Equivalence and Underdetermination' (1991), showed that even granting the global thesis the inference fails, and their argument has two independent limbs. First, empirical equivalence is not a fixed relation between theories, because the class of a theory's observational consequences is not fixed: consequences are derived only with auxiliary hypotheses, and the auxiliaries change as the rest of science changes. Two theories empirically equivalent relative to the auxiliaries of 1900 may be distinguishable relative to those of 1930. Equivalence is therefore time-indexed, and a demonstration of it at one moment carries no entitlement about the next. Second, and independently, evidence can support a hypothesis it does not entail and which does not entail it, by supporting a more general theory from which the hypothesis follows. Two theories with identical empirical consequences may accordingly stand in quite different evidential relations to the same body of data, because one of them is entailed by a well-supported general theory and the other is not. Empirical equivalence, in short, does not deliver evidential equivalence, and the underdetermination that would make theory choice a matter of taste does not follow from the holism that is actually true. That result narrows the target but does not remove it. What remains, and what the rest of this essay is about, is the fact that the constraints on theory choice are plural and cannot be aggregated by a rule — which is a much more interesting claim than the one it replaces. Where the choice really is non-algorithmic: Kuhn's five values The best-grounded statement of that remaining claim is Thomas Kuhn's 'Objectivity, Value Judgment, and Theory Choice' (1977), and it repays being stated exactly rather than gestured at, because it is routinely mistaken for the relativism its author was writing against. Kuhn lists five characteristics of a good scientific theory on which, he says, practitioners across periods and traditions substantially agree: accuracy, in that the theory's consequences agree with experiment; internal and external consistency, with itself and with accepted theories in adjacent domains; scope, in that its consequences extend beyond the observations it was built to cover; simplicity, in that it brings order to phenomena that would otherwise be isolated; and fruitfulness, in that it discloses new phenomena and new relations. Kuhn's thesis is that these five function as values and not as rules, and he gives two reasons which are separable and both necessary. The first is imprecision: each criterion is indeterminate enough that two competent people can apply it to the same pair of theories and disagree about which wins. Simplicity is his own example — the Copernican system was not appreciably simpler than the Ptolemaic to compute with, and was no more accurate in predicting planetary positions, yet it was simpler in a different and defensible sense, since it fixed the order and relative distances of the planets and made the confinement of Mercury and Venus to the neighbourhood of the sun a consequence rather than a stipulation. The second reason is conflict: the criteria pull against each other, so t