This is a slightly amended discussion of a methodological argument for scientific realism due to Feyerabend, focusing on a difficulty that arose from considering Einstein’s study of Brownian motion and the transition from the phenomenological to kinetic theories of gases.
Suppose we have a theory T that predicts A1 but that actually cricumstances A2 obtain, and further that A1 and A2 are indistinguishable experimentally. This could be because observing the difference is beyond current experimental possibilities or because a demarcation is impossible in principle. In the first instance, note that the response that instrumentation can be expected to improve and therefore provide a means of theory choice is unsatisfactory: prefiguring Galison’s work on instruments as a third pillar needed to understand science (along with theory and experiment), Feyerabend commented in his paper Realism and Instrumentalism that
… the development of instruments of measurement is guided by the ideas and interests of scientists, and it is very unlikely that it will automatically lead to the discovery of all the shortcomings of the theories we possess.
This is especially so when we consider the criterion of empirical adequacy: if we require only that the observable consequences of a theory are true then there is no reason to suspect that T is false; after all, it predicts A1 and that is what we find. T is thus empirically adequate and there is no apparent reason to seek an alternative or to develop more powerful instruments to that end.
Herein lies the force of Feyerabend’s argument. By construction, T is an empirically adequate theory: its consequences are confirmed by experiment and it thus functions as an instrument of prediction. Nevertheless, T is false. Without any additional methodological principles, however, there is no reason to suppose that T could be false; and, more importantly, no reason to try to show that it is.
Since the realist employs the concept of truth, the situation is markedly different. The principle of testability enjoins upon the scientist the necessity of testing “relentlessly … whatever theory he possesses”, as Feyerabend put it. Although this applies equally to the instrumentalist, the fact that T is empirically adequate provides no motivation to continue testing. For the realist, on the other hand, T represents a description of reality (or some restricted domain thereof) with a degree of verisimilitude; it therefore requires testing beyond the satisfaction of empirical adequacy and, if necessary, via the development of other theories – even ones inconsistent with T. Indeed, that is the point: if we introduce new theories and
… succeed in elaborating one of these theories in such detail that it can be compared with T as regards simplicity and effectiveness, if this theory is confirmed where T was confirmed, if it solves some cases which belonged to the class of unsolved problems of T …, if it makes predictions which are not made by T, and if these predictions are confirmed as well…
… then we take the new theory to be more truthlike than T and accordingly reject T as refuted, “despite the fact that no direct refuting instance has yet been found for T.” This, of course, is what occurred with the replacement of the phenomenological theory of gases by the kinetic, so it is far from being an academic exercise.
What we have is thus a methodological justification of realism. Although this was Feyerabend in sophisticated falsificationist form, the difficulties for the falsificationist are the same as for the instrumentalist. What the latter lacks is a justification for developing theories in their strongest forms when the current ones satisfy empirical adequacy. Although one direction of objection open to us is to assert that instances following the above model are few and far between, it is not easy to see how we would go about demonstrating such. This, then, is the claimed methodological weakness of instrumentalism.