Generative AI in education: A Human–AI pedagogical agency framework for learning, cognition, and ethics

Abstract

Generative artificial intelligence (GenAI) has entered education faster than the theoretical and methodological frameworks used to evaluate it. This critical integrative review asks a more consequential question than whether GenAI ‘works’: under what pedagogical conditions can it augment learning without displacing learner agency, epistemic responsibility, and higher-order cognition? A purposive corpus of 56 distinct sources was synthesized, spanning foundational theories of experiential, sociocultural, situated, distributed, social-cognitive, cognitive-load, self-regulated, and technology-mediated learning; established technology-acceptance models; artificial-intelligence-in-education and AI-literacy frameworks; policy guidance; and empirical, systematic-review, and meta-analytic evidence published through 2026. The synthesis indicates that GenAI is most defensibly conceptualized as a fallible cognitive artifact embedded in a distributed learning system rather than as an autonomous epistemic authority. Positive effects on achievement, motivation, engagement, creativity, and higher-order thinking are increasingly supported, but they are conditional on task design, scaffolding, interaction quality, verification practices, AI literacy, and sustained human oversight. Evidence for metacognitive improvement remains weaker, while unstructured use can promote cognitive offloading, overconfidence, dependency, and integrity risks. To integrate these findings, this article proposes the Human–AI Pedagogical Agency Framework (HAPAF), composed of five interacting layers: epistemic positioning, learner agency, interaction design, verification, and governance. The framework reframes effective GenAI integration as an agency-preserving pedagogical design problem rather than a tool-adoption problem.