Exploring the Impact of Peer Reflection on Agent Participation in a Large Language Model Simulation Environment

Abstract

Peer reflection is a crucial process that enhances learning effectiveness and encourages the review of thoughts and actions. This study introduced peer reflection into a multi-agent discussion system utilizing Large Language Models (LLMs) and quantitatively and qualitatively evaluated its effects based on changes in utterance volume and content. Four agents, each with distinct individual characteristics, engaged in peer reflection after an initial discussion and then participated in a subsequent discussion. The results showed variability in the improvement of both utterance quality and quantity during the latter discussion. These findings suggest that peer reflection has the potential to vitalize discussions and enhance learning effectiveness in educational dialogue environments that employ LLMs.