Artificial intelligence tools are widely used by Chinese medical students, yet systematic evidence on usage patterns, critical literacy gaps, and influencing factors remains limited, particularly from large-scale multi-institution studies. This study investigates whether a hypothesised structural “artificial intelligence adoption paradox”—high perceived benefits coexisting with weak information verification competence and unclear academic integrity norms—exists among Chinese medical students. A cross-sectional survey was conducted among 3,194 medical students, with 90.4% from Shanxi and supplementary samples from three additional institutions in Beijing and Hunan. A purpose-built twenty-five-item questionnaire assessed artificial intelligence usage, critical evaluation, academic writing, and ethical attitudes. Internal consistency was confirmed for multi-item domains (Cronbach’s alpha range: 0.71–0.84). Descriptive statistics, Spearman correlations, logistic regression, and K-means cluster analysis were applied. Artificial intelligence adoption was pervasive, with 49.8% of students using artificial intelligence often or always in daily life and 84.5% perceiving efficiency gains. However, in cross-sectional comparison across year groups, confidence in information verification was 47.2% among Year 1 students compared to 35.0% among Year 5 students, and ethical reflection showed a similar pattern in this cross-sectional sample (Year 1: 69.8% vs. Year 4: 48.8%; Year 5: 51.0%). These cross-sectional differences may reflect cohort effects rather than within-person change. Daily artificial intelligence use correlated with perceived efficiency but not with discernment confidence. Males demonstrated higher discernment confidence than females, a difference that warrants further investigation given potential confounding with academic year distribution. Cluster analysis identified three exploratory user profiles: low-intensity adopters, intensive all-domain users, and study-selective users (mean silhouette coefficient = 0.31, indicating moderate cluster separation). A 21.5-percentage-point gap between verification intent and confidence revealed a “knowing-doing gap.” Only 37.0% of students considered direct use of artificial intelligence-generated content as plagiarism, indicating substantial normative uncertainty. This multi-institution cross-sectional study of 3,194 undergraduate Chinese medical students provides empirical support for the hypothesised artificial intelligence adoption paradox: high artificial intelligence use coexists with limited discernment confidence and declining ethical reflection across year groups in this cross-sectional sample. However, the cross-sectional design precludes causal inference, and the observed patterns may reflect cohort differences rather than within-person change. The three identified profiles suggest potential avenues for differentiated pedagogical strategies. Medical curricula should integrate critical artificial intelligence literacy, verification skills, and clear integrity norms from the earliest academic year.
Zhang et al. (Wed,) studied this question.