When the thesis passes but the mind doesn’t: Generative artificial intelligence, proxy regimes and epistemic agency in universities

Abstract

Generative artificial intelligence (GenAI) tools increase the speed and polish of academic writing, but they also destabilise higher education’s reliance on written artefacts as proxies for understanding. This raises a risk that students can submit fluent text while bypassing the effortful discomfort through which information is usually turned into durable knowledge and judgment. This article synthesises how GenAI-enabled cognitive bypass is described in recent scholarship and identifies conditions under which GenAI use shifts from a cognitive extender to a cognitive substitution. This conceptual article bridges existing theories across disciplines to broaden our understanding of GenAI’s role in students’ cognitive development. Firstly, epistemic hollowing occurs when students are seduced by the fluency of large language models’ (LLMs) outputs, leading to weakened verification processes. Secondly, the uncritical use of GenAI may erode desirable difficulties, and retrieval practices may substitute for key cognitive operations. Thirdly, universities’ audit culture and product-oriented assessment regimes make cognitive bypass rational. Cognitive bypass is framed as a mismatch between tool affordances, learner goals and institutional proxy systems. The implications point towards teaching, assessment and supervision designs that validate processes, require defence and verification and preserve epistemic agency in GenAI-saturated universities. Contribution: The study integrates fragmented GenAI in higher education debates into a three-lens explanatory model of cognitive bypass and translates it into process-oriented implications for teaching, assessment and postgraduate supervision.