Application of ChatGPT for automated problem reframing across academic domains

Abstract

This paper explores the potential of large language models, specifically ChatGPT, to reframe problems from probability theory and statistics, making them accessible to students across diverse academic fields including biology, economics, law, and engineering. The aim of this study is to enhance interdisciplinary learning by rendering complex concepts more accessible, relevant, and engaging. We conducted a pilot study using ChatGPT to adapt problems across 17 disciplines, evaluated through expert review. Our results demonstrate the significant potential of ChatGPT in reshaping problems for diverse settings, preserving theoretical meaning in 77.1% of cases, and requiring no or only minor revisions in 74% of cases. An evaluation performed by 23 domain experts revealed that in 73.6% of cases the reframed problem was considered to add educational value compared to a corresponding abstract problem and to represent a real-world scenario in 57.0% of cases. Furthermore, a survey involving 44 Computer Science students revealed a diverse range of preferences between original and reframed problems, underscoring the importance of considering student preferences and learning styles in the design of educational content. The study offers insights into the practicality and efficacy of employing large language models, like ChatGPT, to enhance interdisciplinary education and foster greater student engagement and understanding.