Critical thinking dispositions and perceived support-dependence discrepancy in GenAI-assisted learning: evidence from human–AI interaction pathways

Background Generative artificial intelligence (GenAI) can support academic performance while also creating a perceived gap between AI-supported task completion and independently available competence. This study examined how multidimensional critical thinking dispositions are associated with cognitive offloading, cognitive enhancement, and perceived support-dependence discrepancy (PSD) in GenAI-assisted learning. Methods A cross-sectional survey was administered through Wenjuanxing from March 15 to May 20, 2026, using stratified purposive sampling across six Chinese universities. Of 990 submitted questionnaires, 797 eligible and complete responses were retained. The hypothesized pathways were tested using structural equation modeling, with a separate observed-score moderation analysis for academic stress and additional robustness and measurement-model assessments. Results Inquisitiveness was positively associated with cognitive enhancement (β = 0.515, p < 0.001) and cognitive offloading (β = 0.270, p < 0.001). Open-mindedness (β = 0.091, p = 0.039) and systematicity (β = 0.096, p = 0.029) showed small positive associations with cognitive enhancement, whereas analyticity and truth-seeking were not significantly associated with either interaction pathway. Cognitive offloading (β = 0.448, p < 0.001) and cognitive enhancement (β = 0.207, p < 0.001) were positively associated with PSD. The Academic Stress × Cognitive Offloading interaction was not significant (β = 0.048, p = 0.123; delta R-squared = 0.002). Alternative CFA specifications, a common-method-factor analysis, and reliability sensitivity analyses yielded broadly consistent focal conclusions, while the HTMT assessment provided qualified evidence of discriminant validity. Conclusion Critical thinking dispositions showed heterogeneous associations with GenAI interaction pathways, and both offloading and enhancement were associated with perceived support-dependent competence gaps. Because the outcome was measured with a single self-report item and the design was cross-sectional, the findings indicate statistical associations rather than objectively verified overconfidence, calibration bias, or causal effects.