Conversational agents have been used to support student learning for some time, but the emergence of Large Language Models (LLMs) poses a novel opportunity to enhance their capabilities in collaborative settings. LLM-powered agents can provide timely interventions in collaborative conversations when a teacher is unable to assist the students. However, the use of LLMs in such tools raises many ethical questions and concerns, especially for use with young, impressionable populations. In this work, we present the human-centered design and evaluation of an LLM-based agent aimed to facilitate small group collaboration in middle- and high-school classrooms. Fifty-eight groups of dyads and triads (145 participants), aged 12-17, collaborated in a jigsaw activity and were assigned to be assisted by our agent or not. The results showed decreased self-reported ratings of social loafing and increased use of language related to respectful collaboration in interactions with the agent compared to those without.
https://dl.acm.org/doi/10.1145/3706598.3713349
The ACM CHI Conference on Human Factors in Computing Systems (https://chi2025.acm.org/)