AI literacy, information cocoons, and creativity in healthcare postgraduate education: A latent profile and mediation analysis.

OBJECTIVE: This study aimed to identify latent subtypes of AI literacy among full-time healthcare postgraduate in China, including those specializing in clinical medicine, nursing, basic medicine, and related medical fields, and to examine the mediating role of information cocoons in the relationship between AI literacy types and creativity. METHODS: From April to August 2025, 278 full-time healthcare postgraduates in China were surveyed using scales for AI literacy, information cocoons, and creativity. Data were analyzed using Mplus 8.3 for Latent Profile Analysis and Mediation Analysis, and SPSS 27.0 for univariate and multivariate analyses. RESULTS: (1) Four AI literacy subtypes were identified: Lowest Level-Least Skilled (5.40%), Lower Medium Level-Least Ethical (24.46%), Upper Medium Level-Weaker Attitude (37.41%), and Highest Level-Balanced (32.73%). (2) Logistic regression analysis indicated that gender and grade were significant predictors of AI literacy classification (X2 = 1.48, p < 0.001; X2 = 76.75, p < 0.001). (3) Person-centered analysis (with the Lowest Level-Least Skilled group as the reference group) revealed that the Upper Medium Level-Weaker Attitude (β = -0.513, p < 0.001) and Highest Level-Balanced (β = -0.470, p < 0.001) groups had significant negative effects on information cocoons, while the Lower Medium Level-Least Ethical group showed a positive effect (β = 0.121, p = 0.035). Information cocoons maintained a significant negative effect on creativity (β = -0.380, p < 0.001). Further effect decomposition demonstrated significant mediating effects of information cocoons in the relationships between the Lower Medium Level-Least Ethical, Upper Medium Level-Weaker Attitude, and Highest Level-Balanced groups and creativity, with bootstrap confidence intervals of (-0.251, -0.009), (0.277, 0.671), and (0.248, 0.669), accounting for -17.35%, 36.73%, and 20.13%of the total effect, respectively. CONCLUSION: This study reveals heterogeneity in AI literacy among Chinese full-time healthcare postgraduates and demonstrates that information cocoons partially mediate the relationship between all AI literacy subtypes and creativity. These findings provide important insights for enhancing graduate students' creativity in the context of Generative AI.