Integrating AI in Clinical Education: Evaluating General Practice Residents’ Proficiency in Distinguishing AI-Generated Hallucinations and Its Impacting Factors

OBJECTIVE To evaluate the ability of general practice residents to detect AI-generated hallucinations and assess the influencing factors.METHODS This multi-center study involved 142 general practice residents, all of whom were undergoing standardized general practice training and volunteered to participate. The study evaluated AI’s accuracy and consistency, along with the residents’ response time, accuracy, sensitivity(d’), and standard tendencies (β). Binary regression analysis was used to explore factors affecting the residents' ability to identify AI-generated errors.RESULTS 137 participants ultimately included had an mean (SD) age 25.93 ± 2.10, with 46.72% male, 81.75% undergraduates, and 45.26% from Jiangsu. Regarding AI, 52.55% were unfamiliar with it, 35.04% had never used it. ChatGPT demonstrated 80.8% overall accuracy, including 57% in professional practice. 87 AI-generated hallucinations were identified, primarily in the level of application and evaluation. The mean (SD) accuracy was 55% ±4.3%, and the mean (SD) sensitivity (d') was 0.39 ± 0.33. The median response bias (β) was 0.74 (0.31). Regression analysis revealed that shorter response times (OR = 0.92, P = 0.02), higher self-assessed AI understanding (OR = 0.16, P = 0.04), and frequent AI use (OR = 10.43, P = 0.01) were associated with stricter error detection criteria.CONCLUSIONS The study concluded that residents struggled to identify AI errors, particularly in clinical cases, emphasizing the importance of improving AI literacy and critical thinking for effective integration into medical education.