Differentiation under pressure: psychological pathways and latent profiles of GenAI dependence among graduate...

As artificial intelligence becomes increasingly embedded in academic research, graduate students' dependence on GenAI has emerged as a significant concern. This study aims to examine the psychological pathways linking time pressure with graduate students' GenAI dependence, identify heterogeneous GenAI dependence profiles, and explore the demographic, cognitive, and creativity differences associated with these profiles.

Drawing on survey data from 1,157 graduate students, this study employed PROCESS-based mediation analysis, latent profile analysis, multinomial logistic regression, and analysis of variance to examine the psychological pathways, user profiles, demographic correlates, and creativity differences associated with GenAI dependence.

(1) Time pressure was positively associated with GenAI dependence through two parallel indirect pathways: research anxiety (indirect effect = 0.162, accounting for 41.22% of the total effect) and performance expectancy (indirect effect = 0.108, accounting for 27.48%). The anxiety-related indirect pathway showed a larger association than the performance-expectancy pathway. (2) Four distinct types of GenAI dependence formation were identified: the All-low type (5.77%), the Balanced type (46.42%), the Dual-high/High-dependence type (39.90%), and the High-expectation/Low-anxiety type (7.91%). (3) Gender, academic year, and critical thinking were associated with profile membership. The High-expectation/Low-anxiety and Dual-high/High-dependence profiles showed similarly high creativity levels, both exceeding the Balanced and All-low profiles.

The findings suggest that graduate students' GenAI dependence is associated with both anxiety-related and performance-expectancy pathways under time pressure. The study identifies four distinct GenAI dependence profiles and highlights the role of critical thinking in differentiating these profiles. The High-expectation/Low-anxiety profile may represent a relatively adaptive form of GenAI engagement, although its creativity level was not significantly higher than that of the Dual-high/High-dependence profile.