As GenAI becomes increasingly embedded in higher education, a central question is who retains epistemic agency. This paper reports on the human-centred design of an undergraduate module in Database Technologies. Drawing on human-centred AI principles (Shneiderman, 2020) and Students-as-Partners scholarship (Cook-Sather et al., 2014), it argues that ethical AI integration requires deliberate epistemic friction. Rather than treating AI simply as a tool for efficiency, the pedagogical approach framed it as a contestable cognitive partner, emphasising students’ capacity to question, constrain, and override its outputs.

The study involved 62 students, 70% of whom reported no prior knowledge of the subject material. The module began with participatory design: students co-created and prioritised requirements for a fictional travel information system using in-class polling tools. The lecturer then demonstrated how to translate these requirements into a database schema through iterative dialogue with ChatGPT. In keeping with the concept of cognitive apprenticeship (Collins et al., 1989), expert reasoning was made explicit and chat dialogues were shared with students as artefacts of the process.

Subsequent tuition, alongside weekly lectures, incorporated online revision quizzes and a custom-designed chatbot featuring learning-from-errors diagnostic modules, query explanation, adaptive feedback, and formative MCQs. The end-of-semester test also used MCQ format; however, students were required to provide written rationales rather than simply select an answer, with additional marks awarded for correct explanations. This design operationalised the self-explanation effect (Chi et al., 1994) and ensured that correctness remained attached to reasoning.

To examine whether AI integration scaffolded or displaced independent competence, this study distinguished between AI-assisted and unassisted self-efficacy in the technical material, drawing on Bandura’s (1997) theory of agentic capability. This distinction extends technology acceptance frameworks (Venkatesh et al., 2003) by shifting attention from adoption and perceived usefulness toward epistemic calibration; that is, the learner’s capacity to judge when AI should be trusted, challenged, or overridden.

Survey data (n = 47) revealed a nuanced pattern. Students widely recognised the utility of AI. 85% agreed that AI tools increased their confidence in tackling the material, 89% reported meaningful learning benefits from the chatbot, and 85% felt comfortable trusting it despite awareness that it might produce incorrect outputs. However, indicators of independent competence were more differentiated. While 53% agreed they could write database queries without AI assistance, only 40% agreed they could independently troubleshoot errors. At the same time, 53% reported confidence in adjudicating disagreements between themselves and AI. Taken together, these findings suggest calibration rather than capitulation: students appear willing to use AI as a support tool while still regarding human judgement as the ultimate arbiter of correctness.

Behavioural data from the revision quizzes (n = 53) reinforce this interpretation. Students were permitted unlimited attempts, and performance improved markedly across attempts. In many cases, initial scores were very low, yet students frequently reached 85–100% after repeated attempts. These trajectories suggest engagement patterns more consistent with mastery-oriented revision than with one-shot answer extraction. If students were primarily using AI to generate answers directly, the expectation would be to see rapid convergence on correct responses with few attempts. Instead, the data show iterative cycles of error, feedback, and correction. This pattern aligns with the “learning from errors” pedagogy embedded in the chatbot and classroom approach, suggesting that AI functioned less as an answer engine and more as a diagnostic scaffold supporting repeated practice, thereby supporting the human-centred aim of maintaining students’ epistemic agency.

References

Bandura (1997). Self-efficacy. Freeman.

Chi (1994). Self-explanations improve understanding. Cogn Sci, 18(3).

Collins (1988). Cognitive apprenticeship. Thinking, 8(1).

Cook-Sather (2014). Engaging students as partners. Wiley.

Shneiderman (2020). Human-centered AI. Int J HCI, 36(6).

Venkatesh (2003). User acceptance of IT. MISQ, 27(3)..