📄 Research Article
Learning-by-Teaching with ChatGPT: The Effect of a Teachable ChatGPT Agent on Programming Education
Synthesis: Chen et al. (2024) investigate whether ChatGPT can serve as a teachable agent to support learning-by-teaching in programming education. Interacting with ChatGPT as a tutee improved students' knowledge gains and programming ability — especially writing readable, logically sound code — and boosted self-regulated learning and self-efficacy, but had limited impact on error-correction skills because ChatGPT tends to produce correct code, reducing debugging practice.
The Problem with Traditional Teachable Agents
Learning-by-teaching is an effective active-learning strategy, but traditional teachable agents (e.g., rule-based tutoring systems) have limitations — particularly their inability to engage in natural-language dialogue. ChatGPT's conversational ability offers a way to make the teachable agent a natural interlocutor that students must explain concepts to and correct.
Key Findings
- Improved knowledge and programming gains. Interacting with ChatGPT as a teachable agent improved students' knowledge gains and programming abilities, particularly in writing readable and logically sound code.
- Limited error-correction benefit. The approach had limited impact on developing error-correction skills, likely because ChatGPT tends to generate correct code, reducing opportunities for students to practice debugging.
- Self-regulated learning gains. Students' SRL abilities improved, suggesting that teaching ChatGPT fosters higher self-efficacy and better implementation of SRL strategies.
- Role of natural dialogue. Natural-language dialogue supports "socialized learning by teaching," where the conversational relationship with the agent drives engagement and reflection.
Implications
- Conversational teachable agents can extend learning-by-teaching beyond rule-based agents into natural-language domains.
- Design agent fallibility deliberately — a teachable agent that is too competent (always correct) removes the error-correction practice that productive learning-by-teaching depends on.
- Support SRL — the teachable-agent interaction naturally scaffolds self-regulated learning, but may need explicit design to do so consistently.
Connected Concepts
- Learning By Teaching
- Generative AI
- CS Education
- Self Regulated Learning
- Scaffolding
- Active Learning
- Desirable Difficulties
Connected Articles
- Knowloop Confusion To Consolidation 2026 — Teach-back consolidation in a conversational review system
- Curiobot LLM Tutoring Exploratory Learning — LLM tutoring for exploratory learning
- Structured LLM Feedback Programming — Structured LLM feedback in programming
Citation
Chen, A., Wei, Y., Le, H., & Zhang, Y. (2024). Learning-by-Teaching with ChatGPT: The Effect of a Teachable ChatGPT Agent on Programming Education. arXiv:2412.15226 [cs.CY].