Predictive success, partial truth and skeptical realism

Realists argue that mature theories enjoying predictive success are approximately and partially true, and that the parts of the theory necessary to this success are retained through theory-change and worthy of belief. I examine the paradigmatic case of the novel prediction of a white spot in the shadow of a circular object, drawn from Fresnel’s wave theory of light by Poisson in 1819. It reveals two problems in this defence of realism: predictive success needs theoretical idealizations and fictions on the one hand, and may be obtained by using different parts of the same theory on the other hand. I maintain that these two problems are not limited to the case of the white spot, but common features of predictive success. It shows that the no-miracle argument by itself cannot prove more than a skeptical realism, the claim that we cannot know which parts of theories are true. I conclude by examining if Hacking’s manipulability arguments can be of any help to go beyond this position. Introduction: partial truth and predictive indispensability In its broader understanding, scientific realism is the claim that our scientific theories are true, i.e. that something about them describe correctly mind-independant realities of nature. While its an old philosophical thesis which has known a handful of different versions, most of today’s realists agree that the best argument to defend it has been first proposed by Duhem (Duhem 1914, 37) and reintroduced – on another form – by Putnam (Putnam 1975), Musgrave (Musgrave 1988) and Leplin (Leplin 1997). This argument is the no-miracle argument based on predictive success: it claims that theories which have enjoyed predictive success are mature, i.e. true, because truth is the only explanation of this success which does not make it a miracle. Several problems of this argument have been discussed in the literature, including the issue of the inference from predictive success to truth (Stanford 2000). Here, I would