Human Learning Through AI-Mediated Conversation: The Human Learning Conversation Framework for Academic Writing and Assessment

Generative AI unsettles the conventional inference from a polished academic assignment to the understanding of the student who submits it. This article proposes the Human Learning Conversation Framework (HLCF), a provisional pedagogical and assessment synthesis for examining students’ intellectual participation in AI-mediated academic writing. Grounded in writing-process theory, knowledge transformation, cognitive engagement, sociocultural mediation, self-explanation, retrieval, feedback literacy, and self-regulation, HLCF distinguishes observable conversational acts from inferred cognitive engagement and independently demonstrated learning. It organises ten descriptive dimensions into four conceptual domains and relates them to three complementary sources of evidence: contextualised interaction episodes, consequential changes in composition, and independent reconstruction or application. The article positions HLCF relative to recent AI-assessment approaches, presents an illustrative scoring protocol and worked academic example, and develops an argument-based account of assessment validity. It identifies limitations involving manufactured interaction traces, unequal access, language, privacy, task differences, and workload. HLCF is a proposed model, not a validated psychometric instrument; the paper therefore specifies the empirical work required to evaluate its interpretive consistency, relation to independent performance, and possible pedagogical effects.