Human–GenAI Interaction in Higher Education: Trust as a Central Mechanism Shaping Learning Enhancement and Trade-offs

Abstract While generative artificial intelligence (GenAI) is ubiquitous in higher education, there is only sparse empirical evidence to inform us of the learning impact of human–GenAI interaction. This study aims to fill this void, exploring how students' engagement with GenAI can impact their learning outcomes by studying trust as a mediating variable and taking into account task-oriented learning trade-offs. The proposed framework was validated using survey data from higher education students, and applying structural equation modeling. The findings indicate that the interaction with GenAI in collaboration with the findings of trust is the primary driver in learning outcomes. The positive effect of trust on learning enhancement was complemented by the increase in perceived learning trade-offs, revealing the dual nature of trust in AI-mediated learning. Mediation analysis results show that the relationship between interaction and learning outcomes is partially mediated by trust. The results of the effect sizes show that the interaction effect and the task-oriented use are relatively weak factors in the model, while the trust effect is relatively high. Multigroup analysis shows that Academic level and Gender are key factors in the structural differences, while Usage level is not. This implies that quality of interactions and the formation of trust is more important than the level of usage. The research further contributes to the development of the interaction-centered approach to GenAI in higher education and presents empirical evidence for creating meaningful human–AI interactions with good learning outcomes while avoiding potential adverse effects.