Artificial Intelligence Literacy and Critical Thinking in Graduate Education: Auditing AI Generated Conclusions as a Pedagogical Practice

Abstract

Large language models entered the academic work of graduate students faster than the teaching strategies needed to use them critically. The aim of this study was to design and demonstrate a replicable pedagogical practice, the structured audit of AI generated conclusions, in which students classify each statement produced by an assistant according to the evidence supporting it, the missing information, and the evidence contradicting it, using the neutrosophic triplet as a conceptual scaffold. The research adopted a qualitative approach with a design research strategy and an instrumental case study based on public data: the 1973 Berkeley graduate admissions, the canonical example of the aggregation paradox. Findings show that the assistant spontaneous conclusion is correct in the aggregate and wrong in its structure, that the reversal of the pattern by department holds in four of the six departments analysed, and that the triplet separates judgements which the true or false pair collapses. The practice operationalises critical thinking into an assessable artefact.