Artificial Intelligence for Active Learning in Education: A Critical Narrative Review of Pedagogical Value, Adoption Resistance, and Ethical Governance

Abstract

Artificial intelligence is being incorporated into education at a pace that exceeds the development of stable pedagogical evidence, institutional capacity, and enforceable governance. Its educational value is often framed through personalisation, rapid feedback, and efficiency, yet these affordances do not necessarily produce active learning and may instead encourage cognitive offloading, dependency, or superficial task completion. This critical narrative review integrates three bodies of scholarship that are commonly treated separately: active-learning theory and evidence, resistance to pedagogical and technological change, and ethical frameworks for artificial intelligence in education. Literature published from 1 January 2010 to 5 June 2026 was examined, with earlier foundational sources included where conceptually necessary. The synthesis indicates that artificial intelligence supports active learning most plausibly when it elicits explanation, prediction, retrieval, critique, revision, and peer dialogue rather than supplying polished answers. Meta-analyses of intelligent tutoring systems and recent experimental studies of generative artificial intelligence suggest beneficial average effects, but confidence is constrained by short interventions, locally developed assessments, weak evidence on retention and transfer, uneven disciplinary coverage, and rapid technological obsolescence. Resistance is not adequately understood as reluctance or deficient digital competence. It may signal concerns about pedagogical legitimacy, professional identity, workload, surveillance, reliability, inequity, and the erosion of human relationships. Effective mitigation therefore requires participatory design, transparent purpose, protected alternatives, assessment alignment, professional learning, and institutional support. Ethical principles are necessary but insufficient unless translated into lifecycle governance covering educational necessity, data minimisation, fairness testing, human oversight, contestability, disclosure, incident response, and periodic withdrawal decisions. An integrated framework is proposed in which pedagogical alignment, learner agency, adoption conditions, and accountable governance are treated as interdependent design requirements. Artificial intelligence can catalyse educational improvement, but only when it is subordinated to defensible learning purposes and when institutions accept responsibility for both pedagogical and social consequences.