An Instrumentalist Account of Predictive Success — Epoche C1
The argument at issue, and its strongest case Quantum electrodynamics — the quantum theory of the interaction between electrons and light — predicts the strength of the electron's magnetic moment to a precision at which calculation and measurement agree at the level of parts per billion. A realist about science takes such an agreement to be evidence that electrons exist and that the theory describing them is at least approximately true, on the ground that nothing else could account for it. This is the no-miracles argument, given its familiar formulation by Hilary Putnam in 1975: realism is the only philosophy of science that does not make the success of science a miracle. The argument, and whether it can be resisted without denying the success, is the subject of this note. Set out in steps, the argument runs: Mature scientific theories achieve extraordinary predictive success, including successful predictions of phenomena unknown when the theory was framed. The best explanation of that success is that the theories are approximately true and their unobservable entities approximately as described. Therefore we are justified in believing that they are. Notice the form. Step two selects a hypothesis on the ground that it would explain the data better than its rivals, and step three concludes that the hypothesis is true. This pattern is called inference to the best explanation, and identifying it matters for two reasons taken up below: it is the same pattern of inference that scientists use inside theories, and it carries a substantive assumption — that being the best available explanation is evidence of being true — which is precisely what is at issue between realists and their opponents. A theory that was successful and false The original version of this note attempted to break the link between success and truth by appealing to traditional divination practices which, it said, have sometimes yielded surprisingly accurate predictions about weather or harvests. No evidence was offered for that claim, no study was cited, and no measure of accuracy was given, so the example has been removed rather than repaired. It is also unnecessary, because the history of physics supplies a documented case that does the same work far more forcefully, since it involves precisely the kind of success the realist regards as decisive. In 1818 the French Academy of Sciences set a prize competition on the diffraction of light. Augustin Fresnel submitted a treatment based on the hypothesis that light is a wave in an elastic medium — the luminiferous ether — filling all space. Siméon Poisson, examining the entry, derived from Fresnel's equations the apparently absurd consequence that the shadow of a small circular disc should have a bright spot at its exact centre, and offered this as a reductio of the theory. François Arago performed the experiment and the spot was there. This is novel predictive success of the strongest kind available: a consequence nobody had anticipated, derived from the theory, taken by a competent contemporary to be a refutation, and confirmed. If any success licenses belief in the entities a theory posits, this one does. Yet the luminiferous ether does not exist. There is no elastic medium filling space, and the vibrations Fresnel described are not vibrations of anything. The realist's inference, applied in 1818 to the best-supported optics of the day, would have delivered a false conclusion about what the world contains. What the historical record does to the argument One case establishes that the inference is not deductively valid, which nobody disputed. The interesting question is whether it is a good inductive inference, and that is a question about frequencies rather than possibilities. Larry Laudan pressed exactly this point in 1981. He assembled a list of theories that were successful by the realist's own criteria and whose central theoretical terms are now taken to refer to nothing: the crystalline spheres of ancient and mediaeval astronomy, the humoral theory of medicine, effluvial theories of static electricity, the phlogiston theory of chemistry, the caloric theory of heat, the electromagnetic and optical ethers, and theories of spontaneous generation. His argument is an induction over the history of science. If reference failure has been the normal condition of successful theories rather than an aberration, then success is weak evidence of truth, and the no-miracles argument is an inference from a premise that the historical record does not support. Kyle Stanford sharpened the induction in 2006 by asking what the failures have in common. His answer is that in case after case, the scientists of a period could not conceive of an alternative that was equally well confirmed by the evidence then available and that later became the accepted theory. Darwin's contemporaries had no notion of Mendelian inheritance; the difficulty was not that they weighed it and rejected it. If that pattern is the norm, then no present-day assessment of "the best explanation" can be trusted to have canvassed the field, because the historical record shows that such assessments have systematically failed to. Hacking is not an instrumentalist, and the difference is the point The original note recruited Ian Hacking as an instrumentalist philosopher offering an alternative to realism. This is a misdescription, and correcting it is the most important change in this revision, because the correction preserves a stronger version of the note's thesis than the one it argued for. Hacking's position in Representing and Intervening of 1983 is a realism — an entity realism. He believes electrons exist. What he denies is that we should believe the theories about them. His criterion for belief in an entity is not that it figures in a successful explanation but that we can use it as a tool: when the properties of a supposed entity are understood well enough that we can deploy it to interfere with something else, the case for i