Designing Learning by Teaching Agents: The Betty's Brain System

The idea that teaching others is a powerful way to learn is intuitively compelling and supported in the research literature. We have developed computer-based, domain-independent Teachable Agents that students can teach using a visual representation. The students query their agent to monitor their learning and problem solving behavior. This motivates the students to learn more so they can teach their agent to perform better. This paper presents a teachable agent called Betty's Brain that combines learning by teaching with self-regulated learning feedback to promote deep learning and understanding in science domains. A study conducted in a 5th grade science classroom compared three versions of the system: a version where the students were taught by an agent, a baseline learning by teaching version, and a learning by teaching version where students received feedback on self-regulated learning strategies and some domain content. In the other two systems, students received feedback primarily on domain content. Our results indicate that all three groups showed learning gains during a main study where students learnt about river ecosystems, but the two learning by teaching groups performed better than the group that was taught. These differences persisted in the transfer study, but the gap between the baseline learning by teaching and self-regulated learning group decreased. However, there are indications that self-regulated learning feedback better prepared students to learn in new domains, even when they no longer had access to the self-regulation environment.