Patrick Cronin

According to the Classical Pessimistic Induction (CPI), otherwise positive features like predictive and explanatory success actually cast doubt on theories that display them. After all, they negatively correlate with truth. From this historical track record, it is inferred that current best theories are probably not (even approximately) true. The CPI is often wielded against realists about science, but others like Hume (1757), Pareto (1935), and Hájek (2007) extend it to philosophical theories more generally. In this paper, I unveil a priori refutations that defang the CPI in all its guises. To date, treatments of this topic typically assume the following: if the historical proportion of false theories increases, then it has increasingly negative bearing against realism. I show that this assumption is false. I identify three types of problems for the CPI, each attaching to some interpretation or other. The first problem is that the CPI violates a universal adequacy condition on the confirmation relation that restricts the relationship between entailment and disconfirmation. The second problem is that the induction’s target hypotheses preclude the track record from instantiating a list of epistemically significant properties that empirical information is supposed to afford. The third problem is that it is constitutively impossible for the CPI to have random samples; this implies that even the use of randomization procedures cannot help. In sum: my refutation is novel, accessible by pure reflection, and quite general.

Regularity theories of causation assert that causal or nomic notions are to be reduced into “mere” frequencies of particular, non-nomic, co-located qualities and matters of fact. In this essay, I present a critical exploration of Armstrong and Strawson’s explanatory arguments against regularity theories. The shortcomings of these older arguments for nomic realism are identified and a revamped version which is immune to such problems is outlined and defended. I argue that anti-realism suffers substantial disconfirmation due to its comparative inability to unify empirical regularities in the absence of any probabilistic counterweights. I also show that realist theories are much more probable than their anti-realist competitors both individually and in aggregate. This is shown to be the case with even the most humble of observational data. This revamped argument is Bayesian in character; it is immune to the criticisms of Beebee, Everitt, Loewer, and van Fraassen ; and it is empiricist friendly to boot.