An improved essentiality requirement for deployment realism

Psillos’ “deployment” realism is committed only to the theory components which are essential in deriving novel predictions. H is essential when (1) A novel prediction NP follows from H, together with the rest of the theory RT and supplementary assumptions A, but not from RT+A alone; (2) there is no available alternative hypothesis H* which is (a) compatible with RT and A, (b) non-ad hoc, and (c) potentially explanatory, such that (H*+RT+A)→NP. Lyons (2006) argued that this definition doesn’t work because: (i) it is too vague to be applicable to any historical case: e.g., in (2) it is unclear in which sense and when H* should not be available, what “potentially explanatory” means, etc.; (ii) too many hypotheses (including false ones) would qualify as essential, because typical real life competitors are not compatible with RT and A, and they are ad hoc in Psillos’ sense. Hence, he suggested that deployment realists abandon the essentiality requirement altogether, crediting all the components which were actually employed in deriving novel predictions. But this would be a hara-kiri move, since history is full with hypotheses which were actually employed in deriving novel predictions and subsequently found to be false. In fact, I argue, the essentiality condition is necessary, since it stems from Occam’s principle that we should assume only what is strictly necessary to explain a phenomenon. Therefore, instead of discarding conditions (1) and (2) we can replace (2) with a better working condition: (2’) H cannot be weakened to H’ such that H→H’ and (H’+RT+A)→NP. If NP is risky and H fulfills (1) and (2’), most probably H is true, and any alternative H* is false. In fact, the rate of false hypotheses entailing novel risky consequences is so small that hitting one of them (without using those consequences) would be a miraculous coincidence. Instead, all true (and fecund) hypotheses have true (novel) consequences. Although true hypotheses are fewer than false ones, we don’t find them by chance, but through reliable methods. Alai (2014) § 7 shows how (2’) rules out false components. However, checking whether H fulfills (2’) is not a merely logical task: at any given time some logically possible weakenings of H may be considered physically impossible given certain background presuppositions. E.g., the belief that waves must propagate in a material medium prevents weakening the idea of aether to that of field. Therefore what is essential or inessential (Vickers, 2016) cannot be distinguished prospectively, as hoped by Votsis (2011) and Peters (2014). This is why we cannot foretell the future development of theories. Yet, the (partial) truth of H can be acknowledged independently of its being preserved today; instead if and when H is subsequently refuted, it also appears that H was inessential; this is seen retrospectively, but independently of its refutation. Hence, pace Stanford, the selective realist defense against Laudan’s meta-modus tollens is not circular.