Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (AI) can support university learning, yet psychological reliance on AI-generated feedback may limit independent mastery experiences. Under an integrated I-PACE–social cognitive framework, this three-wave observational study examined whether self-efficacy for learning and performance statistically accounted for the association between AI dependence and subsequent academic engagement, and whether physical activity moderated the first-stage association. The analysis was based on 1,068 quality-screened students matched across baseline (T1), Week 4 (T2), and Week 8 (T3) questionnaires. AI dependence and physical activity were measured at T1, self-efficacy at T1/T2, and academic engagement at T1/T3. Standardized regression-based PROCESS Models 4 and 7 adjusted for baseline self-efficacy and engagement, demographics, and AI-use intensity. Higher T1 AI dependence was associated with lower T2 self-efficacy ( β = −0.137), while T2 self-efficacy was positively associated with T3 engagement ( β = 0.114). The first-stage interaction was small ( β = 0.073; ΔR 2 = 0.005), with an index of moderated mediation of 0.008 (5,000-resample 95% bootstrap CI [0.002, 0.017]). Inverse-probability-of-completion weighting produced similar core estimates; longitudinal measurement-invariance and latent autoregressive SEM results are additionally reported as sensitivity evidence. The results identify a short-term, temporally ordered indirect association involving academic self-efficacy, not an established causal mechanism. Educators may preserve students’ independent mastery opportunities when designing AI-assisted coursework.
No takes yet. Share an insight, caveat, or question.
Tan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: