Remarks on Inference to the Best Explanation
This session introduced Inference to the Best Explanation as a potential response to Van Fraassen’s arguments for Constructive Empiricism. But that risks blurring two distinct ideas. It may be used:
1: To serve as model of, and shed light on, inference within science. The kind of inferences scientists make may helpfully be thought of as applications of inference to the best explanation (IBE).
2: To shed light on the status of science as a whole, from outside it, to accord either with what Scientific Realists or their opponents think.
Peter Lipton’s book, for example, is mainly concerned with the former project and only addresses the latter modestly. (He does not think that IBE is very good itself at taking on Constructive Empiricism.) His main aim is to use inference to the best explanation (IBE) to shed light on the kind of inference deployed in science.
Using IBE to shed light on scientific inference
Rather than using inductive inferences to arrive at a theoretical view of the world and then, distinctly and subsequently, using explanation to shed light on particular phenomena, he proposes that the initial inferences can best be understood as being based on explanation. Science works by looking for explanations. The best potential explanation is likely to be true.
Still, if explanation is to shed light on our inductive / scientific practices, the notion of best explanation (or best potential explanation) had better not depend too closely on what seems inductively compelling. So best explanation had better not be most likely explanation at the risk of circularity. Instead, Lipton emphasises the potential initial gap between inductive inference and explanation by distinguishing the likeliest and the loveliest explanation.
The likeliest explanation is the explanation that is ‘most warranted’. (An example of a likely but unlovely explanation is that opium puts people to sleep through its dormitive powers. Probably true but unhelpful.)
The loveliest explanation is the explanation which provides the most understanding, or is the most explanatory. If IBE relies on the latter rather than the former, it can provide a substantial, rather than a merely trivial, analysis of inductive inference. It can come as a surprise that a lovely explanation lies at the core of inductive inference.
But the cost of emphasising the distinction – in order to preserve the substance of the analysis – is that it prompts the following worry, which Lipton calls Voltaire’s Objection. Why assume that loveliness and truth are correlated? Why assume that we live in the loveliest world?
Lipton’s response to this emphasises, inter alia, the fact that to be a potentially lovely explanation requires already meeting a number of constraints on adequacy. (He does not have a knock down reply to the worry. His aim is to suggest that his account is no worse off than any other when it comes to Humean worries about induction in general and he takes Voltaire’s Objection just to amount to that, in the end.) But the problem of this kind of reply, this direction of travel, is that it undermines the contrast which gave the analysis substance. So the question is then whether the two ‘vectors’ can be balanced: to have enough distinction to shed some light on inductive practices whilst maintaining enough connection to avoid Voltaire’s objection. After all if the analysis were to work there would have to be a connection and a gap: the paradox of analysis, again.
Using IBE to justify scientific realism
The classic thought is that Hillary Putnam’s comment that: ‘Realism is the only philosophy that does not make the success of science a miracle’ (1975: 73) If success means something like its continued observational and predictive successes (ie its ‘empirical adequacy in Van Fraassen’s phrase), the only account of this which is not miraculous is that the entities etc it postulates and from which it derives its predictions are real and roughly behave as science says.
Of course, Van Fraassen can contest this claim by saying he does not deny the unobservable realm. That might make science miraculous. He is agnostic not atheistic.
Still the realist can frame an IBE argument:
Science is successful (empirically adequate?).
The only / best / loveliest explanation for the success of science is scientific realism (ie the approximate truth of science)
So scientific realism is true.
The problem with this, though, is that it infers from an observable property (the ‘success’ of science) to an unobservable (approximate truth of science). So it would be irrational for Van Fraassen not to apply the basic Constructive Empiricist line: we should be agnostic about unobservables: in this case the underlying truth of science (rather than just its observable claims).
(Recall the implicit argument from underdetermination. Van Fraassen can say that whilst scientific realism is one explanation of the success of science, there is an infinite number of other explanations because there is an infinite number of other theoretical claims about what is unobservable which still accord with the observable success of science. So we can never know that scientific realism is true.)
McMullin adds a further thought to try to justify his claims for scientific realism:
How is the inference from best explanation to the likelihood of the corresponding hypothesis to be supported? Earlier, we listed the three so-called ‘theories’ of truth: correspondence, coherence, and pragmatic effectiveness. It has frequently been noted that the latter two are criteria of truth and are much less plausible as means of definining the nature of truth. Whereas correspondence quite obviously cannot be a criterion; it can at best only have something to do with the nature of truth. [McMullin **: 66]
The latter two are ‘generic criteria’. They are ways of broadly telling when it is appropriate to judge something to be true. There seems to be no operationalised way to check a belief corresponds to a fact, by contrast. It says, instead, something about the nature of truth. But he goes on to say that the generic criteria for truth also closely resemble the kinds of tests scientists use to select theories (on the basis of their explanatory coherence and predictive / pragmatic accuracy.) So
The criteria that scientists would regard as appropriate to the assessment of a theory resemble, therefore, to a striking extent the criteria traditionally thought to be symptoms of truth generally. The values that constitute a theory as ‘best explanation’ are... Similar to those that would qualify a statement as ‘true’. [ibid: 67]
But McMullin’s emphasis on the coherence and pragmatic aspects do not play down his realist credentials: ‘The coherence criteria, in particular, are realist in their implications’ [ibid: 67] So they indicate that the third aspect of truth – correspondence – has also been met.
But note now we are back on familiar territory. McMullin is using the ‘generic criteria’ to argue that the correspondence notion is met. That is, he is arguing from observable properties (coherence and pragmatic effectiveness) to something unobservable (correspondence with a fact). But that is the very argument Van Fraassen will reject and so it cannot be used against him.