Beyond prompt literacy: Towards a critical epistemological framework for artificial intelligence literacy in higher education
TL;DR
This article advances a reconceptualised AI literacy centred on epistemic discernment, a theoretically grounded, practically enacted capacity for source interrogation, lateral verification, semantic depth analysis and algorithmic auditing, with direct implications for curriculum design, assessment and institutional governance.
Abstract
Artificial intelligence (AI) literacy has emerged as a growing concern in higher education. However, dominant frameworks remain narrowly focused on functional competencies such as operating and prompting AI tools. This utilitarian orientation risks conflating technological fluency with epistemic capacity, producing graduates who can interact with AI systems but cannot critically interrogate their outputs. The substantive gap is not that evaluation is absent from these frameworks, but that it remains epistemologically under-specified. This article advances a theoretical critique of prevailing AI literacy models, arguing that they are epistemologically insufficient for the demands of scholarly and professional life. Drawing on Connell and Keane’s Knowledge-Fitting Theory of Plausibility, virtue epistemology, critical information literacy, and Wineburg and McGrew’s lateral reading research, it locates the core problem in the plausibility trap: large language models generate outputs calibrated for word-coherence rather than accuracy, producing text that performs epistemic authority without warranting it. The article advances a reconceptualised AI literacy centred on epistemic discernment, a theoretically grounded, practically enacted capacity for source interrogation, lateral verification, semantic depth analysis and algorithmic auditing. This reconceptualisation carries particular urgency in African higher education contexts, where the structural biases embedded in globally deployed AI systems collide with institutional commitments to epistemic transformation. Contribution: The article contributes two original conceptual instruments to the AI literacy literature: the plausibility trap and the stochastic alibi and translates its theoretical framework into four mutually constitutive dimensions of an epistemologically adequate AI literacy, with direct implications for curriculum design, assessment and institutional governance.