Adaptive AI-generated feedback in higher education: exploring associations with cognitive flexibility and academic psychological safety among engineering students
Abstract
Background Adaptive AI-generated feedback (AIF) is increasingly integrated into higher education; however, its cognitive and psychological implications remain insufficiently understood, particularly in engineering education. Grounded in feedback intervention theory and emotion regulation theory, this study developed and tested an integrative model examining the perceived relationships between AIF, cognitive flexibility (CF), and academic psychological safety (APS), including mediating and moderating patterns. Methods A cross-sectional survey was conducted among 471 engineering students. Data were analyzed using partial least squares structural equation modeling. The measurement model demonstrated satisfactory reliability and validity, with factor loadings above 0.70, AVE values exceeding 0.50, composite reliability values ranging from 0.908 to 0.936, Cronbach’s alpha values ranging from 0.887 to 0.921, and HTMT values below 0.85. The structural model showed moderate explanatory power for CF (R2 = 0.443) and APS (R2 = 0.626), with adequate predictive relevance and no multicollinearity concerns. Results AIF was significantly associated with CF (β = 0.580, p < 0.001) and APS (β = 0.352, p < 0.001). CF was significantly associated with APS and showed a significant indirect association between AIF and APS (indirect effect β = 0.257, p < 0.001). Significant interaction effects indicated that ESAF was associated with a stronger relationship between AIF and CF, whereas MSAF was associated with a stronger relationship between CF and APS. Conclusion Perceived AIF was associated with APS directly and indirectly through CF, while emotional and mindfulness-supportive features were associated with a greater strength in these relationships. Given the study’s cross-sectional design, the findings indicate associations rather than causal effects. This study advances the current understanding of cognitive and affective patterns in AI-supported engineering education and highlights the importance of integrating adaptive, emotional, and reflective features into AI feedback systems.