Explaining the Process of University Students’ Epistemic Trust in AI-Generated Content: A Grounded Theory Study from the Perspective of Philosophy of Education

Abstract

Purpose: This study aimed to explain the process through which university students develop, evaluate, recalibrate, and regulate epistemic trust in AI-generated content from the perspective of philosophy of education.

Methods and Materials: This qualitative study was conducted using a grounded theory approach among 24 university students studying at universities and higher education institutions in Tehran, Iran. Participants were initially selected through purposive sampling and subsequently through theoretical sampling until theoretical saturation was achieved. Data were collected through individual semi-structured, in-depth interviews focusing on participants’ experiences of using, evaluating, trusting, questioning, and verifying AI-generated content in academic contexts. Interviews were transcribed verbatim and analyzed concurrently with data collection using initial coding, focused coding, axial categorization, constant comparison, memo writing, theoretical sampling, and selective integration. Trustworthiness was enhanced through prolonged engagement with the data, peer examination, participant validation, negative-case analysis, reflexive memo writing, and maintenance of an audit trail.

Findings: The analysis generated six main categories: pragmatic entry into AI-mediated knowing, heuristic attribution of epistemic credibility, epistemic friction and disruption of trust, verification and calibration of trust, educational and social regulation of epistemic trust, and consequences for epistemic agency. These categories were integrated around the core category of “calibrating epistemic delegation under conditions of algorithmic opacity.” The findings indicated a dynamic process progressing from pragmatic initiation and heuristic credibility attribution to provisional reliance, epistemic friction, active verification and calibration, and conditional integration. Contextual and intervening factors, including prior knowledge, task stakes, disciplinary norms, AI literacy, information literacy, teacher guidance, and previous negative experiences, shaped this process. Two longer-term trajectories emerged: reflective epistemic augmentation and epistemic dependency.

Conclusion: Epistemic trust in AI-generated content is a conditional, dynamic, and revisable process in which educationally responsible use depends on maintaining learner responsibility, critical verification, evidential judgment, and appropriate limits on epistemic delegation.