Exploring pre-service teachers’ conflict resolution strategies and decision-making patterns in AI-based multi-agent environments

Conflict resolution is essential for teachers. Pre-service teachers (PSTs) often struggle to connect coursework with classroom practice, especially in multi-role conflicts. Artificial intelligence (AI) simulations may offer a low-risk environment. This study examined PSTs’ conflict-resolution practices in an AI-based multi-agent setting where virtual participants were available within the same conflict scenario. We developed a chatbot through which 20 PSTs interacted with an aggressor, an accomplice, a victim, parents, and a tutor in a peer-conflict scenario. Using dialogue logs, we applied inductive qualitative content analysis and lag sequential analysis to examine behavioural transitions. The analysis identified two complementary findings. First, PSTs’ strategies varied by student-agent role. When addressing aggressors, they tended to use guiding reflection, emotional support, surface threat, and role endowment; when addressing accomplices, they more often used positive encouragement, guiding reflection, home-school collaboration, and emotional support; and when addressing victims, they relied on home-school collaboration, positive relationship guidance, and emotional support. Second, PSTs’ engagement with AI guidance followed four pathways: relying on AI-generated suggestions, iterative inquiry with AI, doubting AI-generated suggestions, and bypassing AI-generated suggestions. These findings suggest that PSTs selectively negotiated role-specific cues and AI guidance, making visible situated professional judgement in AI-supported conflict-resolution practice.