The Inferentialist Critique and the Reliability Challenge to Scientific Realism — Epoche C1
The claim at issue, and where its support actually lies Bas van Fraassen's central thesis is a claim about what a scientist is entitled to believe when she accepts a theory: on his view she is entitled to believe only that the theory is empirically adequate — that everything it says about observable things and events is true — and not that its claims about electrons, fields or genes are true. This is constructive empiricism, stated in The Scientific Image in 1980, and its usual presentation makes it turn on the line between the observable and the unobservable. The original version of this essay argued that a deeper critique lies underneath, aimed not at what we can see but at the reliability of the inferences by which scientists get from what they see to what they posit. That thesis is right, and it can be given a sharper defence than it received — beginning with a bibliographical correction, because the inferentialist critique is not in The Scientific Image in any developed form. It is in Laws and Symmetry , published nine years later, where van Fraassen devotes two chapters to the rule of inference the realist relies on and produces two distinct arguments against it. Anyone who reads only the 1980 book will find the observable/unobservable distinction and conclude, as most commentators have, that this is the whole of the position. What constructive empiricism actually asserts Three of its terms need pinning down, because the argument later depends on their exact content. Empirical adequacy is not agreement with the evidence collected so far. A theory is empirically adequate, for van Fraassen, if what it says about all observable things and events — past, present and future, observed or not — is true. This is an enormously strong claim, and noticing how strong it is turns out to be the most effective weapon against the position. Observable means observable by beings like us: something is observable if there are circumstances in which, were it present to us under those circumstances, we would perceive it unaided. The moons of Jupiter qualify, because an astronaut who travelled there would see them with the naked eye; the telescope merely brings a genuine observation about. A particle whose passage leaves a track in a cloud chamber does not qualify, because there are no circumstances in which a human being could see it. The distinction is thus anthropocentric by design — it concerns the limits of a certain kind of organism — and van Fraassen accepts that it is vague, replying that a vague predicate remains usable so long as it has clear cases and clear counter-cases. Acceptance is separated from belief. To accept a theory is to believe it empirically adequate and, in addition, to commit oneself to using its vocabulary, extending it, and confronting future problems with its resources. The second component is pragmatic, involves no further belief, and explains why the constructive empiricist need not treat working scientists as making a mistake. This is why the realist manoeuvre the original text dismissed — arguing that empirical adequacy is "a step towards truth" — really does miss its target: van Fraassen is not denying that theories may be true, only denying that accepting them requires believing they are. The rule under scrutiny Realism's usual bridge from the observable to the unobservable is inference to the best explanation, or IBE: from a body of data and a set of candidate hypotheses, infer the truth of whichever hypothesis would, if true, best explain the data. Peter Lipton's reconstruction distinguishes two questions the rule must answer — which hypothesis is likeliest , meaning best supported, and which is loveliest , meaning would provide the most understanding if true — and the substantive commitment of IBE is that loveliness is a guide to likeliness. Van Fraassen's two arguments both attack that commitment, and they attack it in quite different ways. The argument from the best of a bad lot. IBE ranks the hypotheses we have actually formulated and selects the top of that list. But the rule's conclusion is that this hypothesis is true, and that conclusion requires more than that it beat its rivals: it requires that the true hypothesis be somewhere in the set we considered. We have no independent reason to believe this. Human beings have generated candidate explanations for the composition of matter for two and a half millennia, and for most of that period the truth was not among the candidates on the table; each generation nonetheless took the best of what it had. So IBE, as a rule for arriving at truth, presupposes what van Fraassen calls a privilege — that our capacity for generating hypotheses is biased towards the true ones — and the realist owes an argument for that privilege which does not itself use IBE. The argument from probabilistic incoherence. The second argument is technical and is the one that gives the essay's word "reliability" a precise meaning. Suppose IBE is understood as a genuine rule of inference, one that awards the best explanation a bonus in credence over and above what the ordinary updating rule would award. The ordinary rule is conditionalisation: on learning $E$, one's new degree of belief in $H$ becomes $P(H \mid E)$. The Dutch book theorem, due originally to Ramsey and de Finetti, states that if an agent's degrees of belief violate the axioms of probability, then there exists a set of bets, each of which the agent regards as fair by his own lights, which together guarantee him a loss whatever happens. Van Fraassen's extension applies the same construction diachronically to rules for changing belief: an agent who announces in advance that he will update by conditionalisation plus an explanatory bonus can be offered a sequence of bets, all fair by his own announced standards, on which he loses in every possible outcome. The upshot is not that explanation is unimportant but that it cannot be a source of extra probability, since a rule that adds any i