Problematic reliance on generative AI among university students: an I-PACE-informed PLS-SEM and fsQCA study

Generative AI has become increasingly embedded in university students' academic learning, offering immediate feedback, content generation, and problem-solving support. Alongside these benefits, repeated reliance has raised concerns about problematic dependence on AI-mediated academic support. Drawing on an I-PACE-informed framework, this cross-sectional study examined psychological and academic factors associated with learning-related problematic reliance on generative AI, defined as a non-clinical, self-reported pattern rather than a clinical diagnosis. Data were obtained from a non-probability online sample of 242 university students who had used generative AI for learning-related purposes. Partial least squares structural equation modeling (PLS-SEM) was used to examine associations among positive affect, generative-AI use self-efficacy, need for cognition, perceived usefulness, cognitive absorption, AI-based mood regulation, academic stress, and problematic reliance on generative AI. Fuzzy-set qualitative comparative analysis (fsQCA) was further used to identify configurations associated with high problematic-reliance membership and its set-theoretic complement. The PLS-SEM results showed that perceived usefulness, cognitive absorption, and academic stress were positively associated with problematic reliance on generative AI, whereas positive affect and AI-based mood regulation did not show significant direct associations. The fsQCA results indicated that no single condition was necessary for high problematic-reliance membership; instead, both outcome sets were associated with multiple configurations. The findings support a cautious distinction between problematic reliance, ordinary generative AI use, and technology acceptance, and they suggest potential implications for AI literacy and academic support in higher education.