Artificial Intelligence in Higher Education: A Critical Synthesis of Student Learning, Assessment Transformation, and Academic Integrity
TL;DR
The review concludes that artificial intelligence should be treated as a socio-technical component of curriculum and assessment rather than a stand-alone productivity tool, because its educational legitimacy depends on whether institutions can preserve epistemic agency, valid judgement of learning, equitable access, and accountable human responsibility.
Abstract
Artificial intelligence has moved rapidly from specialised analytics and tutoring applications to widely accessible generative systems capable of producing text, code, images, explanations, and feedback. In higher education, this shift has created a closely coupled set of pedagogical opportunities and integrity risks. This critical narrative review examines how artificial intelligence is reshaping student learning, assessment, and academic integrity, with emphasis on the conditions under which educational value is strengthened or weakened. Literature published from 1 January 2019 to 30 May 2026 was identified through accessible scholarly indexes, citation searching, authoritative institutional sources, and verification against DOI and journal records. The evidence indicates that artificial intelligence can expand access to explanation, formative feedback, language support, ideation, and practice, and controlled studies increasingly report benefits for selected learning outcomes. Yet effects are heterogeneous, often short term, and highly sensitive to task design, student expertise, prompting skill, feedback literacy, and the degree of human oversight. Gains in efficiency or performance do not necessarily demonstrate durable understanding, metacognition, or independent capability. Assessment is therefore the pivotal institutional problem: generative systems can assist feedback and evaluation while simultaneously weakening the validity of unsupervised products as evidence of individual achievement. Automated detection is not a dependable solution because accuracy varies by detector, text type, language background, and model evolution, creating risks of false accusation and unequal treatment. The most defensible response is not unrestricted adoption or blanket prohibition, but an aligned model combining explicit AI literacy, process-rich and dialogic assessment, proportionate disclosure rules, human judgement, data governance, and fair procedures for investigating suspected misuse. The review concludes that artificial intelligence should be treated as a socio-technical component of curriculum and assessment rather than a stand-alone productivity tool. Its educational legitimacy depends on whether institutions can preserve epistemic agency, valid judgement of learning, equitable access, and accountable human responsibility.