From AI Assistance to AI Dependence: Predicting Students’ Reliance on Generative AI for Academic Work

Abstract

This study examined the factors that predict university students' dependence on generative AI for academic tasks. Following a pilot run, we surveyed around 220 university students in Punjab, Pakistan, using validated multi-item Likert scales in a cross-sectional design. We operationalized AI dependence by tracking how often students used the tools, how much they leaned on them for writing, assignments, research, and problem-solving, and how far they felt incapable of doing university work without them. The independent variables under study were AI self-efficacy, academic workload, perceived usefulness of AI, time pressure, digital literacy, and academic confidence. Exploratory factor analysis matched the anticipated factor structure, while tests of validity and reliability confirmed acceptable internal consistency and construct validity across every scale. We ran correlation analysis alongside hierarchical multiple regression to pinpoint significant predictors of AI dependence, followed by cluster analysis to group students into distinct profiles. Perceived usefulness of AI, time pressure, and AI self-efficacy proved to be the strongest positive predictors of dependence, while academic confidence and digital literacy predicted it negatively, with the overall model accounting for a substantial share of the variance. The cluster analysis yielded three clear student groups—independent users, strategic users, and highly dependent users—separated by notable differences in AI self-efficacy, time pressure, and academic confidence. What these results indicate is that AI dependence grows out of an intersection of psychological states, immediate circumstances, and practical skills, pointing to how universities might better encourage balanced, self-regulated generative AI use.