Exploring LLM-Based Teachable Agent for Learning by Teaching in Physics Education

Abstract

Active participation and interaction in learning methods have been shown to enhance learning effectiveness, with "learning by teaching" recognized as a particularly effective approach. Recently, the advantages of Large Language Models (LLMs), such as their ability to provide personalized feedback and facilitate natural language interaction, have made them increasingly popular tools for supporting student learning. Based on this background, this study analyzed the impact of LLM-based Teachable Agent (TA) on learners and explored their cognitive and affective effects during the learning process. An experiment was conducted with 24 high school students, examining the differences between traditional individual learning and teaching an agent-based system to study physics concepts. The results indicated that the TA-based learning approach improved learners' conceptual understanding and problem-solving abilities. Additionally, TA-based learning increased learner motivation and engagement, yielding more positive outcomes in user experience evaluations compared to traditional methods. This study demonstrates the potential of LLM-based TA to foster active learner participation and highlights their applicability across diverse technology-driven learning scenarios.