Artificial Intelligence in Mathematics Education: ChatGPT’s Capabilities and Limitations in Problem Posing

Abstract

This study investigates the ability of ChatGPT to generate mathematical problems creatively, focusing on fluency, flexibility, and originality. The findings indicate that ChatGPT can produce a vast number of problems, demonstrating a high level of fluency. Additionally, it tends to prioritize the most original problem among its generated outputs. In terms of flexibility, the model successfully formulates problems across various mathematical content areas, showing adaptability in different contexts. Despite its fluency, ChatGPT occasionally repeats content or diverges from the given instructions in later stages, highlighting the importance of expert supervision. When used in educational settings, AI-generated problems should be reviewed to ensure their relevance and accuracy. While ChatGPT has potential as a supportive tool in mathematics education, it should not function independently in instructional environments. Another observed limitation is the inconsistency in how ChatGPT categorizes mathematical problems, which may indicate gaps in its conceptual understanding. Furthermore, its inability to generate visual representations of geometric concepts could contribute to misclassifications or errors. These findings suggest that while ChatGPT can enhance creative problem posing in mathematics, its effectiveness depends on careful monitoring and structured implementation within the teaching and learning process.