Artificial Intelligence and the Reconfiguration of Knowledge: Evidence for a Structural Shift in Epistemic Systems

Abstract

Artificial intelligence (AI) is becoming a constitutive element of contemporary scientific practice, yet its epistemic implications remain insufficiently theorised. Rather than simply enhancing productivity and efficiency, this study argues that AI reconfigures the underlying structure of knowledge. Using a structured and curated dataset of 600 scientific texts across five disciplines—philosophy, physics, economics, cardiology, and archaeology (2010–2026)—we develop a multidimensional framework capturing conceptual depth (CDI), thematic novelty (TNI), knowledge integration (KII), and epistemic efficiency (EE). AI involvement (AI-share) is operationalised as an inferred and approximate proportion of AI-assisted content, estimated from linguistic and structural features. Across all domains, increasing AI involvement is associated with declining conceptual depth and rising thematic novelty. Knowledge integration remains stable or increases, while epistemic efficiency shows modest gains. These patterns do not indicate cumulative improvement in knowledge quality but instead reveal a redistribution of epistemic value across competing dimensions. We interpret this as the emergence of a recombinatory epistemic regime, in which AI generates variation and human agents perform selection and synthesis. In this configuration, epistemic value arises from the balance between depth, novelty, and integration rather than from depth alone. The central implication is that AI does not merely change how science is conducted—it transforms what counts as knowledge, requiring a reconceptualisation of epistemic evaluation in AI-integrated research environments.

Keywords