Dependence on Generative Artificial Intelligence Among Medical Students and Its Association With Critical Thinking: A Cross-Sectional Study

Introduction: Generative artificial intelligence (AI) (e.g., ChatGPT and Google Gemini) is exceptionally useful in the educational field for learning new concepts. However, besides its benefits, it also introduces the risk of pathological dependence on AI. Hence, this study aims to evaluate the prevalence of AI dependency and its association with critical thinking. Methods: A cross-sectional study was conducted at a government medical college in Tamil Nadu. The study population comprised undergraduate medical students. Data were collected using a structured digital questionnaire that consisted of validated Generative AI Dependency and Critical Thinking Disposition Assessment Questionnaire (CTDAQ-20) scales. Statistical analysis included non-parametric group comparisons, Spearman's rank correlation, and multiple linear regression. Results: Among the 722 participants, the prevalence of Generative AI usage was 695 (96.3%). Among users (n=695), the median AI dependence score was 2.1 (IQR: 1.6-2.7), with males having higher dependence than females (p=0.019) and final-year students reporting higher dependence than students in other years (p<0.001). The overall median critical thinking score was 71.0 (IQR: 64.0-77.0), with female students scoring higher (72 vs. 69; p<0.001), and first-year students having better critical thinking scores than students in other years (p=0.031). Overall, AI dependence had a significant inverse correlation with overall critical thinking (rho=-0.248, p<0.001), and multiple linear regression confirmed that higher AI dependence (B=-1.487, p<0.001) and increasing age (B=-0.868, p=0.008) were associated with lower critical thinking. Conclusion: Generative AI usage is ubiquitous among medical students. However, greater dependence on AI is independently associated with diminished critical thinking, particularly among final-year and male students. Therefore, it is hypothesized that the medical curriculum should consider incorporating assessment methods that require active learning and clinical reasoning, such as objective structured clinical examinations (OSCEs) and direct observation of procedural skills (DOPS), rather than relying primarily on traditional theory-based examinations, and should promote AI literacy that supports the use of AI as an adjunct rather than a replacement for human cognition.