Over-Reliance in the Generative-AI Era: A Suppression-Aware Reanalysis of AI Dependency and Critical Thinking in 1,580 University Students
Year: 2026
Type: article
Abstract: Generative artificial intelligence (GenAI) can accelerate information retrieval, drafting, and problem solving, yet the educational risk is not use itself but uncalibrated dependence that substitutes external output for internal evaluation. This study develops a suppression-aware statistical lens for examining that risk. We reanalyze a real, openly documented 2026 survey of 1,580 university students in Vietnam that measured AI dependency (AID), critical thinking (CT), academic self-efficacy (ASE), academic stress (AST), and performance expectations (PE). Using the published construct-correlation matrix, we reconstruct standardized multiple regression, partial correlation, incremental variance, multicollinearity diagnostics, and all theoretically relevant control subsets. The zero-order association between AID and CT is positive (r=0.356), but after jointly accounting for ASE, AST, and PE it reverses sign (β=-0.092, 95% CI [-0.129,-0.055], p<0.001; partial r=-0.121). The full model explains 62.9% of CT variance, and adding AID last contributes ΔR²=0.0055. Sensitivity analysis shows that the sign reversal emerges only when shared positive variance associated with academic efficacy/expectations is sufficiently separated from dependency-specific variance. The principal contribution is methodological: bivariate correlations can mask a negative unique association between AI dependency and critical thinking. A suppression-aware reporting protocol is proposed for future GenAI-in-education studies and for the design of verification-first, reflection-preserving learning activities.
Authors: Yuxin Tang, Haotian Gu
Institution: Shanghai Jiao Tong University
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: diamond
OpenAlex ID: W7222912407