Epistemic Generativity: A Structural Definition

This paper introduces a formal definition of epistemic generativity as a structural property of evaluative systems. It distinguishes systems that optimize under a fixed evaluative structure from those capable, in principle, of operating on that structure itself. Contemporary artificial intelligence systems are characterized by optimization relative to externally specified criteria: they select actions or outputs according to given standards but do not revise those standards. Epistemic generativity, by contrast, is defined as the condition under which a system can select among, and revise, its own evaluative criteria in response to evidence. The paper argues that prevailing uses of “generativity” in machine learning and adjacent fields—typically associated with behavioral novelty, compositional capacity, or scale—do not capture this distinction, as they remain consistent with optimization under fixed evaluation. It further clarifies the status of out-of-distribution failure as performance degradation relative to existing criteria, rather than evidence against those criteria, absent endogenous revision. On this account, behavioral sophistication does not entail evaluative authorship: intelligence, understood as effective optimization, is structurally distinct from epistemic agency. The analysis provides a minimal, classificatory criterion for separating optimization from generativity and grounds implications for responsibility and governance. Where evaluative structure remains external to a system’s operation, accountability attaches to those who specify, constrain, and update the standards under which the system functions.