Abstract
Objective Based on social cognitive theory and conservation of resources theory, this study examines how perceived AI capability (PAC) relates to learning engagement (LE) in smart education environments. It focuses on the mediating role of academic self-efficacy (ASE) and the moderating role of teacher-student relationship (TSR). Methods A three-wave longitudinal design was implemented over one academic semester. Participants were undergraduate students from a Beijing university (final N = 468) who consistently used an integrated intelligent learning system. PAC and TSR were measured at Time 1, ASE at Time 2, and LE at Time 3. Data were analyzed using structural equation modeling and bias-corrected bootstrapping (5,000 resamples). Results PAC at Time 1 was positively associated with LE at Time 3 ( β = 0.21, p < 0.001). ASE significantly mediated this relationship (indirect β = 0.19, 95% CI [0.13, 0.25]), accounting for 48% of the total effect. TSR moderated the PAC → ASE link (β_interaction = 0.12, p = 0.003). The conditional indirect effect was stronger under high TSR (b = 0.28, 95% CI [0.18, 0.38]) than under low TSR (b = 0.15, 95% CI [0.07, 0.23]), and the difference was significant (Δb = 0.13, 95% CI [0.04, 0.22]). Conclusion Perceived AI capability is temporally associated with learning engagement, both directly and indirectly through academic self-efficacy. Positive teacher-student relationships strengthen this indirect pathway. Theoretically, this study moves beyond technology-centric views by treating PAC as a learner’s interpreted signal of a non-human agent’s competence—distinct from perceived usefulness or system quality—and by showing that AI’s psychological effects depend on human relational resources. Practically, institutions should support teacher-student rapport and self-efficacy alongside AI adoption. Intelligent systems need explainable, process-oriented feedback, and instructors should actively help students turn AI capability cues into lasting learning engagement.
Showing the abstract — retrieve the full paper via the Exa API.