Reassessing Inference to the Best Explanation in the Philosophy of Scientific Realism — Epoche C1
Two Routes to Scientific Realism Scientific realism is the position that our best scientific theories should be read as approximately true descriptions of a mind-independent world — including the parts of the world no one can observe, such as electrons, fields and viruses — rather than as convenient instruments for predicting what our detectors will register. The classic argument for it proceeds by inference to the best explanation (IBE), a pattern of reasoning named by Gilbert Harman in 1965: when one hypothesis explains the available evidence better than any rival, that explanatory superiority is itself a reason to believe the hypothesis. We use the pattern constantly in ordinary life — the crumbs, the open jar and the guilty dog together make "the dog ate it" the thing to believe, though nobody saw the eating. Applied globally to science, IBE yields what Hilary Putnam called the no-miracles argument: the sustained predictive and technological success of theories that posit electrons would be an inexplicable fluke unless something like electrons actually exists, so realism is the best explanation of success itself. IBE-based realism, however, has well-known vulnerabilities, and they explain why an alternative route became attractive. Larry Laudan's 1981 "confutation of convergent realism" assembled a historical list of theories that were predictively successful in their day yet posited entities we now reject — the caloric fluid of heat theory, the phlogiston of combustion chemistry, the luminiferous ether. If success did not certify those posits, why should it certify ours? And Bas van Fraassen pressed a structural worry: IBE licenses belief in the best explanation we have thought of , which may merely be the best of a bad lot; ranking rivals by explanatory virtues such as simplicity and unification presupposes, contentiously, that those virtues track truth rather than human convenience. Hacking's Interventionist Alternative Against this background, Ian Hacking's Representing and Intervening (1983) proposed to shift the ground of the debate from theorising to experimentation. His entity realism holds that we are entitled to believe in an unobservable entity not when it figures in our best explanation, but when we can routinely use it as a tool to investigate something else. His flagship example is drawn from real laboratory practice: in the Stanford experiments that searched for free quarks, physicists altered the charge on a niobium ball by spraying it with electrons and positrons; and in the PEGGY II source built for scattering experiments, polarised electrons were produced and steered as a matter of engineering routine. Hacking's famous verdict on the sprayed positrons — if you can spray them, they are real — condenses the view: entities graduate from hypothesis to equipment. The position seemed to sidestep both classic objections at once. It does not infer from explanatory goodness, so the bad-lot problem lapses; and because manipulation relies only on low-level causal properties of the entity (its charge, its mass, its response to fields) rather than on any high theory about it, the entity's credentials look robust against the theory changes that power Laudan's pessimistic history — electrons survived the replacement of classical electrodynamics by quantum theory precisely because engineers went on using their charge throughout. This prioritisation of doing over representing also resonates with a strand of Latin American philosophy that locates truth in concrete practical engagement with reality rather than in disembodied theoretical structure; the appeal of Hacking's picture is not parochial. The argument of this essay is that the appeal, though genuine, cannot fund the claim usually built on it — that interventionist realism is a more basic or less problematic foundation than IBE. Examined closely, intervention presupposes explanation at every step, and where intervention is impossible, science does not stop being rational. The two considerations that follow develop each point in turn. Intervention Presupposes Explanation Consider what "using an electron as a tool" concretely requires. To build an electron gun one must know that electrons carry a fixed negative charge, that they are deflected by electric and magnetic fields in a calculable way, that certain materials emit them when heated or illuminated, and that their spin can be aligned by suitable optical pumping. Each of these items of engineering knowledge is a causal hypothesis about an unobservable entity, and each was originally accepted because it best explained laboratory phenomena: Thomson's 1897 cathode-ray measurements were taken to show a universal charged corpuscle because that hypothesis explained the constant charge-to-mass ratio across cathode materials; Millikan's oil-drop experiments were taken to show charge quantisation because discrete jumps in droplet behaviour were best explained by integral multiples of a unit charge. The experimenter who sprays electrons is thus not enjoying a theory-free encounter with reality; she is deploying a compact body of well-confirmed explanations. The success of the intervention — the niobium ball's charge changes exactly as the manual predicts — is evidence, and powerful evidence, but of what? Of the truth of the causal explanation that guided the design. Without an antecedent explanatory hypothesis about the electron's properties, one could neither conceive of using it to alter a charge nor interpret the altered charge as the electron's doing. This point has been pressed in the critical literature on Hacking, notably by David Resnik, who argues that experimental practice supplies no route to entity realism that bypasses inference to the best explanation: the manipulability argument is best reconstructed as a particularly strong instance of IBE — the best explanation of our reliable success in building electron-spraying devices is that electrons exist and have roughly the p