Knowledge, Innovation and Technology in Scientific Research Vol. 6

Artificial intelligence has extended educational automation beyond restricted tutoring and prediction systems into everyday activities of lesson preparation, dialogue, feedback and assessment. Yet the capacity to produce useful instructional outputs does not establish that learners acquire durable understanding. This critical narrative review examines how adaptive, predictive and generative systems redistribute cognitive work and pedagogical judgement across students, teachers and institutions. Relevant peer-reviewed literature and authoritative educational guidance were selected through topic-specific scholarly searches, source verification and thematic appraisal, with particular attention to comparative instructional studies, recent evidence syntheses and ethical scholarship. The most established quantitative evidence concerns structured intelligent tutoring, although effects depend on curricular fit, comparator and assessment. Emerging randomised studies of generative tutoring reveal opposing outcomes: carefully designed scaffolding can improve short-term learning under specific conditions, whereas relatively unguarded assistance may improve performance during use while weakening subsequent unaided achievement. Evidence on automated feedback similarly requires distinctions among comment quality, perceived usefulness, revision uptake and transfer. Generative systems also unsettle the validity of conventional assessment products and alter professional responsibilities for explanation, verification and learner support. Promising uses of AI literacy instruction, human-centred analytics and hybrid teacher–AI orchestration remain constrained by differences in access, local context, data governance and limited longitudinal evaluation. The review proposes an integrated appraisal centred on task design, system mediation, teacher judgement, learner activity and demonstrable independent competence. The educational value of AI is therefore conditional, not technologically predetermined. Future research should compare credible alternatives, study mechanisms and delayed outcomes, and test whether purported benefits persist across diverse schools, disciplines and learner groups.