Sure of themselves, not yet beholden: Motivation, trust, and generative AI reliance in higher education

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

As generative artificial intelligence (AI) becomes embedded in higher education, a troubling pattern has surfaced. Some students increasingly delegate thinking, judgment, and task completion to AI, and such reliance can quietly erode the very capacities-independent reasoning, learner autonomy, and critical engagement-that universities exist to cultivate. A common assumption is that reliance simply tracks perceived competence, with students who feel more sure of themselves in working with AI inevitably coming to rely on it. The present evidence suggests otherwise. Drawing on survey data from 518 Chinese university students and combining Partial Least Squares Structural Equation Modeling (PLS-SEM) with fuzzy-set Qualitative Comparative Analysis (fsQCA), we find that AI-induced self-efficacy bears no direct association with reliance; it is related to reliance only in conjunction with trust, the belief standing closest to delegation. Escape and instrumental motivation were positively associated with self-efficacy, whereas social motivation was positively associated with trust. Instrumental motivation was negatively associated with trust after self-efficacy was held constant, a pattern consistent with more functional scrutiny among task-focused users. Contrary to expectation, the configurational analysis did not identify multiple pathways to high reliance: it recovered a single configuration combining social motivation and trust, and no stable configuration without trust. Two approaches based on different analytical assumptions thus converge: self-efficacy and trust are not interchangeable pro-AI dispositions. Universities should therefore move beyond promoting efficient AI use and instead calibrate students' trust and protect learner agency, since making students feel more capable with AI appears unlikely to change how much they delegate.