Beyond Bayesian Accuracy: Skill, Abduction, and the Free Energy Principle in Normative Rationality

Abstract This paper challenges traditional accuracy-centric accounts of rationality by synthesising the Free Energy Principle (FEP) with Charles Peirce’s pragmatist epistemology. Whereas the FEP frames cognition as a biological imperative to minimise surprise through predictive models, we argue that its normative force emerges when integrated with Peircean abduction and skill-based metrics. By reinterpreting rationality through skill scores—Peirce’s 1884 method for evaluating rare-event predictions—we demonstrate that survival-driven inference prioritises context-sensitive skill over abstract accuracy. The FEP’s variational free-energy minimisation aligns with abduction’s dynamic conjecture-making, revealing rationality as a pragmatic negotiation between organismic survival and environmental complexity. Critically, we show that Bayesian accuracy measures (e.g., Kullback–Leibler divergence) fail to capture the adequacy conditions for skillful forecasting, whereas Peirce’s skill score satisfies constraints such as error weighting and directionality. This fusion of FEP and pragmatism advances a naturalistic-normative framework in which rationality is grounded in adaptive, enactive inference rather than idealised coherence, bridging computational neuroscience and theoretical biology with philosophical accounts of inquiry.