AI Dependency, Academic Self-Efficacy, and Depressive Symptoms Among Medical Students: A Structural Equation Modeling Study in Saudi Arabia

Background: The rapid integration of artificial intelligence (AI) tools into medical education has opened up new possibilities for academic support. However, it has also raised concerns about over-dependence on AI. Students’ use of AI may have implications for students’ academic self-concept and psychological well-being, but the interrelationships among these constructs are not yet sufficiently understood. This study aimed to examine the association between AI dependency and academic self-efficacy and depressive symptoms among medical students in Saudi Arabia. Methods: We conducted a cross-sectional study among 207 medical students recruited from three public universities in Riyadh, Saudi Arabia. Data were collected anonymously using an online questionnaire distributed through QR-coded recruitment announcements. AI dependency was assessed using a 9-item subset of the AI Dependence Scale, academic self-efficacy using the 5-item General Academic Self-Efficacy Scale, and depressive symptoms using the two depression items of the PHQ-4. AI dependency and academic self-efficacy were modeled as latent constructs in a structural equation model. The model examined the direct association between AI dependency and depressive symptoms and the indirect associational pathway through academic self-efficacy. Results: The mean AI dependency score was 2.83 (SD = 0.78), while the mean academic self-efficacy score was 3.87 (SD = 1.07) and the mean depressive symptom score was 2.05 (SD = 1.42). AI dependency was positively associated with academic self-efficacy (β = 0.298, 95% CI 0.157–0.439, p < 0.001). Academic self-efficacy was inversely associated with depressive symptoms (β = −0.212, 95% CI −0.382–0.041, p = 0.016). AI dependency showed a strong positive direct association with depressive symptoms (β = 0.608, 95% CI 0.491–0.725, p < 0.001). The indirect association through academic self-efficacy was negative (β = −0.063, 95% CI −0.124–0.002, p = 0.052), while the total association remained positive and significant (β = 0.545, 95% CI 0.409–0.671, p < 0.001). The model explained 49% of the variance in academic self-efficacy and 37% of the variance in depressive symptoms. Conclusions: Greater AI dependency was associated with higher academic self-efficacy but also with substantially higher depressive symptom levels among medical students. Academic self-efficacy was inversely associated with depressive symptoms, although evidence for an indirect pathway between AI dependency and depressive symptoms was limited. Our findings highlight the importance of distinguishing productive AI use from excessive dependence. This supports the need for responsible AI integration alongside attention to medical students’ psychological well-being.