Abstract This study investigates the potential of using ChatGPT as a teachable agent to support students' learning through teaching, specifically in programming education. While learning by teaching (LBT) is an effective pedagogical strategy, traditional teachable agents often struggle with facilitating dynamic, dialogue‐based interactions. Our research explored whether ChatGPT, with its advanced conversational capabilities, can effectively support this process. Findings indicate that teaching ChatGPT improved students' knowledge gains and programming abilities, particularly in writing readable and logically sound code. However, its impact on error‐correction skills was limited, likely due to ChatGPT's tendency to generate correct code, thus reducing debugging opportunities. Notably, students' self‐regulated learning (SRL) abilities improved, suggesting that the act of teaching ChatGPT enhances learners' self‐efficacy and SRL strategy implementation. The study discusses how engaging in instructional dialogues with an artificial intelligence (AI) can contribute to the LBT process and explores ChatGPT's specific role in supporting students' SRL. Overall, this research highlights ChatGPT's potential as a teachable agent, offering insights for future research on AI‐driven pedagogical tools and their broader implications for education. Practitioner notes What is already known about this topic? Learning by teaching (LBT) is an effective instructional method that fosters learners' active learning, often within a social context. Previous studies have explored the effect of teachable agents as virtual students, but most agents lack support for dynamic, dialogue‐based interactions in the LBT process. Currently, ChatGPT shows potential to facilitate flexible teaching interactions and allows learners to adapt their teaching strategies. However, the effectiveness of teachable ChatGPT agents remains largely unknown. What this paper adds? This study showed that learning with a teachable ChatGPT agent improved learners' knowledge gains, programming abilities and SRL abilities. This study also found that teaching a teachable ChatGPT agent did not significantly benefit learners' error‐correction abilities. Implications for practice and/or policy Educators can encourage learners to teach or explain concepts to ChatGPT agents through dialogue to promote deeper cognitive processes. When designing ChatGPT‐based teachable agents in programming education, errors should be carefully designed to ensure that students engage with meaningful and representative programming challenges. Future research should explore how SRL strategies can be better integrated into the interactions between students and ChatGPT‐based teachable agents.
Chen et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: