Generative AI as Epistemic Arbiter in Collaborative Problem-Solving: A Study with Fifth-Grade Students

A quasi-experimental study lasting six weeks involved 201 fifth-grade students to investigate the effects of generative AI (GAI) on collaborative problem-solving (CPS). Documented in arXiv preprint 2608.05171, the research contrasted two scenarios: one with GAI assistance and a control group without it. Chi-square analysis indicated notable differences in CPS behavior distributions between the two groups. Students in the GAI-supported group showed increased social behaviors, especially in engagement and conflict management, while exhibiting fewer cognitive behaviors like task planning and solution reasoning. Lag sequential analysis revealed that the control group transitioned from listening to planning, whereas the GAI group moved from task planning to conflict management and then to solution reasoning. Thematic analysis of AI interaction logs indicated that students viewed the AI as a mediator, using it to settle disagreements and direct reasoning. This study underscores GAI's potential to transform collaborative interactions, prioritizing social processes over cognitive ones, suggesting that its integration into educational frameworks could influence student problem-solving approaches, with ramifications for curriculum development and AI tool design.