Why I am not a Bayesian

Abstract

Confirmation theory must provide a critical instrument for understanding non-deductive scientific reasoning and explaining methodological practices such as the preference for simplicity and the requirement for a variety of evidence. Current probabilistic accounts, particularly subjective Bayesianism, fail to meet these requirements. Bayesianism reduces confirmation to personal degrees of belief, yet it cannot adequately explicate why “de-Occamized” or ad hoc theories are disdained, nor can it provide a robust rationale for the value of evidence variety beyond the mere elimination of competitors. A significant limitation is the “problem of old evidence”: because previously known data already has a probability of unity, the Bayesian kinematic model implies that such data cannot increase the posterior probability of a newly introduced theory. Efforts to resolve this via counterfactual degrees of belief result in logical incoherence or require implausible historical reconstructions of an agent’s mental state. Ultimately, scientific confirmation relies on objective, structural relations between theory and evidence rather than subjective probability distributions. These structural features remain largely uncaptured by the Bayesian framework, suggesting that confirmation theory requires an alternative foundation that prioritizes the logical and content-based connections central to scientific argument. – AI-generated abstract.