Artificial intelligence and the transformation of education: a systematic narrative review towards adaptive epistemic...

Artificial intelligence is increasingly embedded in educational processes, yet much of the literature continues to examine individual tools rather than the relationships amongst pedagogy, curriculum, assessment, knowledge validation, institutional authority, and governance. This systematic narrative review examines these relationships whilst explicitly separating evidence-based patterns from conceptual interpretation. The review followed a PRISMA-informed protocol covering database selection, Boolean search strategy, eligibility criteria, staged screening, type-appropriate quality appraisal, structured extraction, citation auditing, and thematic synthesis. Scopus, Web of Science Core Collection, ERIC, and ScienceDirect formed the reproducible database core, whilst Google Scholar and backward and forward citation tracking were used as supplementary discovery mechanisms. After duplicate removal, title and abstract screening, full-text assessment, and quality appraisal, 124 studies were retained for synthesis. The review addressed four questions concerning how artificial intelligence is conceptualised in contemporary educational research; how it is associated with changes in pedagogy, curriculum, learning analytics, assessment, and governance; what recurring relationships connect adaptive feedback, curriculum sequencing, human-AI roles, epistemic validation, and institutional accountability; and what ethical, epistemological, and governance challenges accompany increasing human-AI participation. The synthesis indicates an uneven movement from isolated tool adoption towards more adaptive and data-informed arrangements, accompanied by substantial debate over technological determinism, platform power, teacher agency, language bias, data colonialism, and epistemic justice. Adaptive epistemic ecosystems and post-linear pedagogy are therefore advanced as synthesis-derived, integrative conceptual constructs rather than predetermined coding categories or universal empirical claims. Their contribution lies in specifying testable relationships amongst human-AI cognition, learning-data infrastructures, pedagogical adaptation, epistemic validation, governance, and contextual boundary conditions whilst preserving the distinction between observed patterns, interpretive synthesis, and future-oriented propositions.