When AI Expands Human Potential: Problem-Centered Epistemic Learning, Premise Revision, and Accountable Knowledge Building

Abstract

Generative artificial intelligence lowers the marginal cost of producing explanations, reframings,comparisons, hypotheses, and other outputs that can function conversationally like knowledgebefore they have been adequately evaluated. This paper calls them knowledge-like candidates andasks how a person can carry a consequential problem from lived experience through AI-mediatedinquiry without confusing candidate production with learning or validation. In the languageof the Learning Sciences and Educational Psychology, the paper proposes a Problem-CenteredEpistemic Learning Architecture. Its unit of analysis is a problem trajectory: lived experienceactivates a task-relative premise structure; language externalizes the problem; GenAI expandsthe representational space; knowledge-like candidates are generated; a conversational branchpoint can lead either toward recursive confirmation or toward challenge, reflective regulation,and distinct validation routes; accountable action then produces consequences that can reviselater premises. The framework separates two outcomes that are often conflated: the warrant ofa candidate artifact and the learner’s own understanding, regulation, and transfer. It furtherproposes that when candidate generation becomes inexpensive, educational scarcity shifts towardproblem formulation, validation, accountable application, access to communities of criticism,and earned relational trust. Existing research is introduced after the architecture as convergentevidence and boundary conditions rather than as its genealogy. The resulting research programmeasks how citizens can transform experiential resources into increasingly warranted knowledgeand responsible action at the lowest feasible cost consistent with domain-sensitive safety.