Trust formation and fairness perception in GenAI: the divergent psychological pathways of computer science and business-related students.

BACKGROUND

The rapid proliferation of generative artificial intelligence (GenAI) in higher education raises critical concerns regarding educational equity. However, the psychological mechanisms linking algorithmic characteristics to students' perceptions of fairness remain underexplored. Integrating fairness heuristic theory, privacy calculus theory, and the technology acceptance model, this study examines how GenAI attributes influence perceived educational fairness and explores differences between computer science students and students in selected business-related majors.

METHODS

Survey data were collected from 419 Chinese university students using a purposive sampling method, including 222 computer science students and 197 students in business-related majors. A covariance-based structural equation modeling approach was utilized to analyze the pathways through which GenAI algorithms influence perceived educational fairness, alongside a multi-group analysis to evaluate the structural differences between the disciplinary groups.

RESULTS

Structural equation modeling reveals that algorithmic transparency enhances perceived algorithmic fairness (β = 0.434, p < 0.001), which sequentially builds technological trust (β = 0.582, p < 0.001). This trust positively influences perceived learning impact (β = 0.551, p < 0.001) and perceived educational fairness (β = 0.425, p < 0.001). Data privacy concerns negatively affect technological trust (β = -0.090, p = 0.037). Crucially, multi-group analysis uncovered divergent psychological pathways between disciplinary groups. The positive effect of transparency on perceived algorithmic fairness was stronger among students in business-related majors than among computer science students (β = 0.597 vs. 0.257, p = 0.022), whereas the negative association between data privacy concerns and technological trust was statistically significant only within the computer science group (β = -0.154, p = 0.027). Furthermore, perceived learning impact mediated the trust-fairness relationship only among computer science students (B = 0.158, p < 0.001). Conversely, the mediating effect of perceived algorithmic fairness between transparency and trust was stronger among students in business-related majors than among computer science students (B = 0.329 vs. 0.134, p = 0.017).

CONCLUSION

This study provides a nuanced explanation of how GenAI influences students' perceptions of educational fairness, offering insights for designing inclusive AI-driven educational environments. Furthermore, it highlights that preventing GenAI from exacerbating the digital divide requires differentiated governance strategies tailored to the distinct cognitive pathways of students from different academic majors.