Algorithmic epistemic authority and the degradation of organizational knowledge: how generative AI breaks the provenance structure of knowledge management systems
TL;DR
The observation–knowledge nexus and AEA are introduced as KM constructs distinct from automation bias and prior algorithmic authority, and translates them into governance interventions.
Abstract
This study aims to theorize how organizational knowledge management systems are transformed when generative artificial intelligence (AI) is integrated as a knowledge-producing agent. It introduces algorithmic epistemic authority (AEA), a structural condition of knowledge systems, not individual users, that specifies how generative AI breaks the experiential provenance of knowledge and develops three organizational propositions, falsifiable in principle, each with specified disconfirmation conditions and graded tractability. A conceptual study in knowledge management’s (KM) own theory-building tradition and a Type IV contribution in Gregor’s (2006) taxonomy, integrating four constitutive KM-relevant literatures through a four-step chain of reasoning. The framework predicts that generative AI breaks the observation–knowledge nexus structurally rather than incrementally, producing three organizational dynamics: provenance integrity degradation (P1), SECI bypass and tacit knowledge stagnation (P2) and credentialed-interface knowledge coupling (P3). Each proposition requires longitudinal empirical testing. The framework’s claims are specified for base generative large language models and hold with diminishing intensity, as architectural variants (retrieval-augmented, multimodal and Web-grounded) partially restore the observational link. KM systems require architectural separation of AI-generated and human-authored content, interface-level provenance marking, friction for high-stakes consumption, audit infrastructure, domain-expert validation gateways and experiential development pathways. Miscalibrated trust in AI-generated knowledge threatens institutional credibility in credentialed contexts. This study introduces the observation–knowledge nexus and AEA as KM constructs distinct from automation bias and prior algorithmic authority, and translates them into governance interventions.
