Artificial Intelligence in Higher Education From Automation to Responsible Augmentation

Abstract

Artificial intelligence (AI) is often presented as a sequence of increasingly capable tools, yet educational outcomes do not rise automatically with model capability. This narrative review argues that the decisive variable is alignment: AI produces durable educational value when its affordances are aligned with pedagogy, assessment, educator agency, and institutional governance. The review traces the field from automation and learning analytics through adaptive support and generative co-production to an emerging model of responsible augmentation. Recent meta-analyses, controlled studies, systematic reviews, policy documents, and selected empirical studies were critically compared across three interdependent domains: resources and learning environments; teaching and learning; and management, assessment, and evaluation. The synthesis explains why apparently contradictory findings coexist. Structured, curriculum-aligned AI tutoring can improve immediate learning, whereas unconstrained answer generation can raise supported performance while weakening independent performance. Likewise, automated assessment can improve feedback speed but becomes educationally unsafe when opaque scores or unreliable detectors are treated as final decisions. Evidence concerning international and multilingual learners further shows that language, culture, affect, and data representation are not peripheral variables. The review proposes a responsible-augmentation framework in which alignment, agency, equity, and accountability operate across all three domains. It concludes with a research agenda prioritising longitudinal transfer, calibrated human–AI collaboration, subgroup validity, transparent governance, and institution-level implementation studies.