Training mathematics teachers’ error behavior through ChatGPT-based error situations

This study examines the potential of ChatGPT as a training tool for prospective mathematics teachers, focusing on error behavior and instructional strategies for solving word problems. ChatGPT consistently generated incorrect solutions, mirroring common learner error patterns. Based on the responses of prospective teachers<br /> (N = 26) to these errors, three types of guidance emerged: co-constructive, directive, and non-responsive. These types represent varying instructional strategies guiding learners, as identified in previous research. In addition, participants evaluated the realism and usefulness of ChatGPT-based interactions. While many found the tool valuable for practicing guidance techniques and anticipating learners’ misconceptions, they noted limitations at the same time, including a lack of emotional nuance. Overall, the findings emphasize both the opportunities and limitations of using artificial intelligence-based dialogue systems in teacher education.