Flipping the Roles: Engaging Middle School Students in Learning-by-Teaching With a Generative AI-Powered Teachable Agent

• In Event: Motivation and Creativity in Technology-Based Learning Environments

Abstract

Teachable agents simulate a less knowledgeable peer for students to teach, promoting learning-by-teaching. Recently, generative AI has offered the opportunity to transform the design and development of teachable agents. We created a generative AI-powered teachable agent to support middle school students’ mathematics learning. We conducted a pilot study to test usability and improve the initial prototype. The interviews with teachers and students revealed issues with our prototype, such as a lack of pedagogical support, diversity in agent characters, and socio-emotional interactions. After addressing them, we conducted a quasi-experimental study, the main study of which involved 212 6th-grade students. The experimental group showed a significant knowledge gain in pre-and post-test, whereas the control group showed a smaller improvement.

Authors

• Wanli Xing, University of Florida

• Yukyeong Song, University of Tennessee

• Chenglu Li, University of Utah

• Zifeng Liu, University of Florida