BEWA: A Bayesian Epistemology-Weighted Artificial Intelligence Framework for Scientific Inference

Abstract

The proliferation of scientific literature and the accelerating complexity of epistemic discourse have outpaced the evaluative capacities of both human scholars and conventional artificial intelligence systems. In response, we propose Bayesian Epistemology with Weighted Authority (BEWA), a computational architecture for truth-oriented knowledge modelling. BEWA formalises belief as a probabilistic relation over structured claims, indexed to authors, contexts, and replication his- tory, and updated via evidence-driven Bayesian mechanisms. Integrating canonical authorial identification, dynamic belief networks, replication-weighted citation metrics, and epistemic decay protocols, the system constructs an evolving belief state that prioritises truth utility while resisting social and citation-based distortions. By anchoring every propositional unit in structured metadata and linking updates to semantic replication and contradiction analysis, BEWA enables automated, principled reasoning across a corpus of scientific knowledge. This work advances the theoretical foundations and practical frameworks necessary for autonomous epistemic agents to assess, revise, and propagate beliefs in dynamic scientific environments.