The No Miracles Argument

If we accept the ampliative methods, in particular the Inference to the Best Explanation, then not only we can hope that our selection of theories among the infinitely many logically possible ones is reliable, but we have a meta-argument to the effect that best theories are in fact true or partly true, and scientific method is truth-conducive, hence scientific realism is right: as noted by Putnam (1975a, 73), this is the only explanation of the stunning predictive and applicative success of science which does not make it a miracle (or if you will, a miraculously lucky coincidence).

It has been objected that the NMA is circular or question-begging, for it pre­supposes the reliability of abductive reasoning, which is rejected by antirealists (Laudan 1981: 45-46; Fine 1984: 85-86: see Ladyman 2002: 218-219). In this sense, the choice between realism and antirealism is a stance (van Fraassen 2002; Psillos 2011a, b). But as argued above, realism seems to be better grounded and more consistent with our epistemic practices at large.

4.1 Novel Predictions

The NMA presupposes that truth is the only (non-miraculous) explanation of success. But certain forms of success can be explained otherwise: success in accounting for known phenomena might be simply due to the theorist’s puzzle-solving ability (ingenuity, imagination and patience). This explanation is not available for the prediction of novel phenomena. However, if the predicted phe­nomena are similar to already known phenomena, the theorist might have induc­tively extrapolated them from the previously known ones, and then built the theory with an eye to accommodate them. For instance, it would be hardly surprising that a theory T based, among other data, on certain chemical and physical properties of carbon-12 and carbon-13, predicted analogous properties for carbon-14.

Therefore the NMA properly applies only to the prediction of novel, surprising (heterogeneous from all that was known before) and bold (a priori improbable) phenomena (Musgrave 1988; Psillos 1999). Further examples of predictions ful­filling these constraints are Fresnel’s bright spot and dark spot prediction; the new chemical elements predicted by Mendeleev; Einstein’s predictions of the retard of clocks in motion and the bending of light; the background radiation predicted by the Big Bang theory; etc.

This means that novel predictions confirm more than mere a posteriori accom­modation; but many objected that confirmation is a logical relation between a hypothesis and its empirical consequences; therefore prediction should have no advantage over accommodation. The ensuing debates (see Alai 2014a: §§ 1 -3) have shown that in order to count for the NMA a new phenomenon NP needs not be historically novel (i.e., not known by the theorists, or not used by them in building the theory): even if NP was used, it is enough that it was not used essentially. That is, it is enough that the theory T was plausible independently of NP, i.e., that it was already the best possible account of a body of old phenomena OP not including NP.

4.2 Objections to the NMA

Since success is essentially the truth of the (novel) prediction ‘NP’, it has been objected that it has a trivial explanation, which entails nothing about T: ‘NP’ is true simply because things are as it states (White 2003; Rees 2012: 302). Yet, we have a real puzzle: how could the theorist find, among the numberless possible theories compatible with the old phenomena OP, one predicting the heterogeneous and improbable new phenomenon NP, without essentially using it? This is the real explanandum in the NMA (White 2003; Laudan 1984a: 92; Alai 2014a: 299, c: 50). It has also been noted that the explanans cannot be that the theorist found a true theory: for the true theories entailing NP are just a tiny subset of those entailing NP, so we would explain a puzzling fact by an even more puzzling one. On a careful analysis, however, the actual explanans is that scientific method is truth-conducive (because the uniformity and causal structure of nature makes ampliative inferences reliable), hence the theorist found a true theory (Alai 2014a: § 1, c: § 5).

Problems with the NMA were raised also by Lipton (1991), Howson (2000) and Magnus and Callender (2004), for whom the impossibility to know the relevant base rate prevents to claim that successful theories are probably true. Paul Hoyningen-Huene (2011) argued that the version of the NMA based on novelty is falsified by what he called “transient underdetermination”.

Some antirealists claim that the success of T can be explained differently: for instance, because T is empirically adequate, i.e., compatible with all the phenomena (van Fraassen 1980, 12); or because all phenomena are as if T were true (Fine 1984; Leplin 1987). But this does not explain why T predicts NP, for T may be compatible with all phenomena without predicting any of them, in particular without predicting NP. For instance, a theory with no empirical content would be empirically adequate. For Stanford (2000, 272 ff.) T is successful because it makes predictions very similar to those of the true theory; but this would be like saying that T predicts NP because it predicts a number of things including NP. So, it would not be an explanation, but just a repetition of the explanandum. Other variants of the empirical adequacy explanation, like Lyons' modest surrealism (2002, 78), and Fine's instrumental reliability explanation (1986, 1991), have the same flaw. In general, no “as if” account really explains: “the hypothesis that it is raining explains why the streets are wet—but ‘The phenomena are as if it were raining' does not” (Musgrave 2006-2007).

Besides, none of these expalanantia explains how the theorist found a theory that predicts NP without essentially using it. Realists explain that T was found thanks to the scientific method, which is fecund and truth-conducive; but antirealists cannot answer in a similar way. In fact, there are just two methods to find theories predicting new phenomena: one is by finding true and fecund theories (the realist method); the other is by inductive extrapolation from known phenomena (Laudan 1984a, § 5; Rees 2012, 302; Wright 2013, chs. 3, 4; Alai 2014d, 125-126), but this is impossible if the new phenomena are radically heterogeneous from the old ones (Alai 2014c: § 6). According to Fine (1984) and Hacking (1983: 64 ) the success of T is just the obtaining of the phenomena predicted by T, hence it is already explained by T itself.

Stanford suggests that even a radically false theory which saves the phenomena is likely to make novel predictions, thanks to “the systematic relationship among phenomena within the same domain of inquiry” (2000: 281). However, if “sys­tematic relationship” means that the phenomena are homogeneous, there is no novelty in that field; if instead it means that they are connected by unobservable underlying mechanisms, then it would be a miracle if T predicted NP without getting those mechanisms right.

Van Fraassen's (1980, 40) “natural selection” explanation was that our theories are successful, simply because the unsuccessful ones were dropped. However, this explains just (1) why we have only successful theories, not (2) what makes these theories successful nor (3) how they were found. When asked “why birds can fly?” one cannot just answer (1) “because those which did not fly were wiped out by natural selection”. It should also be explained that (2) birds fly thanks to wings, feathers, hollow bones, etc. (Kitcher 1993; Alai 2014c, § 7).

5