Russell's Critique of Inference to the Best Explanation: A Reassessment — Epoche B2
Russell Used It: The Problems of Philosophy and Inference to the Best Explanation Correcting the question It is sometimes said that Bertrand Russell criticised inference to the best explanation [1] , and that his target was a naive metaphysical realism rather than the epistemic credentials of the inference itself. The second half of that claim is a repair to the first, and both should go. Russell did not criticise inference to the best explanation in The Problems of Philosophy (1912 [2] ). He used it. Chapter II of that book contains one of the clearest early instances of the pattern in English-language philosophy, offered without embarrassment as the reason to believe in physical objects. This essay traces the process of Russell's argument step by step, shows where the explanatory inference occurs, and then locates the genuine Russellian constraint on such inferences, which lies later in the book and is a different constraint from the one usually attributed to him. The constraint is this: the principle that licenses any inference beyond experience cannot itself be established by experience. That is not an objection to inference to the best explanation. It is a demand about what its defenders must supply, and it is exactly the demand at issue in the contemporary dispute between Bas van Fraassen and the scientific realists [3] . Step one: from things to sense-data Russell opens by asking whether there is any knowledge so certain that no reasonable person could doubt it, and finds that ordinary claims about the table in front of him do not qualify. Its colour varies with the light and the observer's position; its apparent shape varies with the angle; its hardness is a matter of what is felt. What is immediately given, he says, is not the table but sense-data : patches of colour, shapes, textures, sounds. The physical table, if there is one, is what causes these. The question of Chapter II follows immediately: is there any reason to believe that anything exists besides the sense-data? Step two: the inference Russell actually makes Russell grants that there is no logical impossibility in supposing that nothing exists except the data — that life is a long dream in which we create everything that appears. He does not attempt to refute this by argument, and says plainly that it cannot be disproved. What he offers instead is a comparison of hypotheses. His illustration is a cat. If the cat exists continuously whether or not anyone is watching, then its movement from one part of the room to another, and its being hungrier at the second meal than it was after the first, are readily understood. If the cat exists only when seen, then it comes into being in a new position for no reason, and its appetite has grown during an interval in which it did not exist. Russell's verdict is that the second story is not impossible, merely far less satisfactory: the hypothesis of enduring physical objects simplifies and systematises our experience, and the alternative leaves a mass of coincidences unexplained. This is inference to the best explanation. Two hypotheses are consistent with the data; one accounts for regularities the other must leave brute; the more explanatory one is accepted. Russell adds the observation that belief in an external world is instinctive and that we should retain instinctive beliefs that generate no difficulties, which supplies a further ground but does not replace the explanatory one. He is explicit that this is not a proof, and he is comfortable with that. The tempting misreading is easy to see how one might reach. Russell's sense-datum epistemology looks sceptical, and his refusal to claim a proof looks like a concession. But refusing to claim proof is a feature of inference to the best explanation, not a criticism of it. Peter Lipton's formulation in Inference to the Best Explanation (2004) makes the point exactly: what the inference identifies in the first instance is the loveliest explanation, the one that would, if true, give the most understanding, and the substantive claim is that loveliness is a guide to likeliness. Russell picks the loveliest hypothesis and treats it as a guide to truth. Step three: where the real constraint appears Chapter VI turns to induction, and here Russell does impose a limit — on all ampliative inference, his own included. He gives the case of the chicken fed every morning of its life, whose expectations are refuted on the day its neck is wrung. Uniform past experience gives no logical guarantee about the future. Russell's conclusion is not that we should stop inferring; it is a conclusion about the status of the principle we infer by. At the close of that chapter he argues that the inductive principle cannot be proved by experience, because any such proof would use the principle: it must be accepted, if at all, on the ground of its intrinsic evidence. Chapter VII generalises the point — induction is one of several principles that license inference beyond experience and are known, if known at all, independently of the experience they license. Apply this to Chapter II and Russell's position becomes clear and unified. The inference to physical objects is legitimate, but the rule that makes it legitimate — that explanatory power is a mark of truth — is not something the inference can supply for itself. Russell's demand is that such rules be independently underwritten. Step four: the same demand, a century later That demand is the live question. Hilary Putnam's no-miracles argument holds that the success of science would be a miracle if our theories were not approximately true [4] , and so infers realism. Van Fraassen's reply in Laws and Symmetry (1989) is that this uses inference to the best explanation to defend inference to the best explanation, and adds the argument from the bad lot: the rule tells us to take the best of the hypotheses we have actually formulated, and nothing assures us that the true hypothesis is among them, so at most the rule iden