Collaborating in the age of GenAI: a mixed-methods study of model preferences
Аннотация
As generative AI (GenAI) tools flood university classrooms, educators lack evidence about how collaboration should be organized when teams co-create while using multiple AI systems. There is limited research comparatively examining alternative orchestration configurations through which university teams use multiple GenAI tools during extended project work. This exploratory mixed-methods study investigated 97 students’ preferences and experiences with five collaboration models in AI-supported collaborative project-based learning (CPjBL). Data from closed- and open-ended questionnaire responses and in-class mini-focus-group discussions were analyzed using nonparametric tests and inductive content analysis. A Friedman test showed differences across models with small shared consensus, indicating substantial heterogeneity in what “works” for students. Adjusted pairwise comparisons showed that Model 1 (teammates working together throughout the entire project work) received higher perceived-quality ratings than Models 2 and 3, although the very small omnibus effect indicated substantial heterogeneity and no universally preferred model. Descriptive statistics showed a complementary profile: students preferred Model 1 on perceived productivity, comfort, dedication, quality, flexibility, close collaboration, and personal fit; Model 5 on speed, equal contribution, and role clarity. Qualitative analyses identified integration and coherence as the central problem when merging AI-supported, individually produced works. Coping strategies included final joint review and integration as well as setting early standards and guidelines. Findings move the field beyond “one best model”, offering actionable design principles for AI-supported CPjBL.