A Bayesian account of the virtue of unification

Abstract

Theoretical unification consists in a theory’s ability to render disparate phenomena informationally relevant to one another. Within a Bayesian framework, this capacity serves as a distinct epistemic virtue that directly enhances a theory’s evidential support. By defining a measure of informational relevance based on the probability calculus, it is demonstrated that the support provided to a hypothesis by a joint body of evidence equals the sum of the support from its individual components plus a term representing the degree of unification achieved. This account implies that the scientific preference for unified theories need not be stipulated as an a priori bias for simplicity; rather, it emerges as a formal consequence of Bayesian updating when a hypothesis reduces the probabilistic independence of observed phenomena. Case studies in Copernican astronomy and Newtonian mechanics illustrate how specific theoretical structures make previously unrelated data points—such as planetary periods, retrograde motions, and the stability of apsides—constrain and inform one another. Unification thus functions as an empirical virtue rather than a merely pragmatic or aesthetic one, providing a rigorous basis for the superior confirmation of theories that integrate previously independent domains. – AI-generated abstract.