Behavioral mechanisms and learning outcomes of University Students' GAI-assisted learning in human-AI collaboration

With the rapid advancement of Generative Artificial Intelligence (GAI) technologies, their integration into higher education is becoming increasingly widespread. This transformation is not only reshaping students' learning approaches but also redefining the collaborative dynamics between humans and AI. Based on the triadic framework of Exploration-Exploitation-Adaptation, this empirical study (207 valid questionnaires from Chinese university students, analyzed via structural equation modeling) investigates the behavioral mechanisms and pathways influencing learning outcomes among university students engaged in GAI-assisted learning. It examines how role adaptation, self-efficacy, task-technology fit, and institutional support affect learners' exploration and exploitation behaviors, and how these behaviors in turn impact learning effect. The findings reveal that role adaptation and self-efficacy are the primary internal drivers of GAI-related learning behaviors, while institutional support and task-technology fit serve as essential external enablers. Both exploration and exploitation behaviors significantly enhance learning outcomes, with exploitation showing a more pronounced effect. The model demonstrates good fit and significant path relationships among variables. While the results are consistent with the proposed adaptation-exploration/exploitation-effectiveness pathway, they only reflect correlational evidence and do not establish a causal mechanism. Theoretically, this study enriches insights into human-AI collaboration in higher education. Practically, it offers guidance for the optimization of intelligent educational systems and the design of behavior-guided strategies.