Answers are grounded in this paper and cite their sources.
Large language models (LLMs) are increasingly used as conversational partners in collaborative inquiry, yet little is known about how different LLM roles are associated with preservice teachers’ learning and collaborative inquiry. This study compared four conditions: Supportive LLM that provided perspectives and information, Challenging LLM that raised counterarguments and questions, General LLM, and No-LLM. It examined differences in AI literacy, cognitive load, and discussion dynamics across these conditions. A quasi-experimental study was conducted with 210 preservice teachers who participated in online discussions on controversial issues related to AI in education. Data included self-reported AI literacy and cognitive load measures and discussion logs. Analyses included paired-samples t-tests, ANOVA, ANCOVA, content analysis, and Ordered Network Analysis. Results showed that adjusted posttest AI literacy was significantly higher in the Supportive condition than in the General LLM condition. In the General condition, AI ethics showed a significant within-condition decrease and was lower than in the Supportive condition. The General and Challenging conditions were associated with higher mental load and mental effort, whereas the Supportive condition was associated with lower levels of both. The Challenging condition showed more frequent patterns of idea integration, interpersonal exchange, and discussion facilitation. The Supportive condition showed broader exploration but fewer connections among ideas. The General condition showed repeated questioning and weaker peer idea connections. We hypothesize that the educational value of LLMs depends not only on their presence but also on how their roles are orchestrated across stages of collaborative inquiry.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: