The Use of Prompt Engineering in Creating Mathematical Modelling Activities with Artificial Intelligence Tool ChatGPT

Abstract

Mathematical modelling and modelling activities are important for making sense of mathematical concepts in different extracurricular and daily life contexts. However, teachers may have difficulties in designing these activities in a way to establish meaningful relationships with real life, in accordance with the modeling process and the objectives of the lesson. With a good understanding of how to interact with it, AI tools can be effective tools for teachers in situations where they face difficulties or inadequacies. The aim of the research is to reveal how ChatGPT, an artificial intelligence tool, designs mathematical modelling activities when prompt engineering is applied. ChatGPT has been utilized as a tool for designing mathematical modelling activities at the middle school level. Version 3.5 of ChatGPT has been utilized as the tool for data collection. As part of research, ChatGPT developed mathematical modelling activities using four distinct input formats. Feedback indicated that a gradual increase in performance could be achieved through careful structuring of the inputs. To this end, researchers elaborated on the desired outcome's format, provided examples, and emphasized boundaries as needed.  The activities generated by ChatGPT, addressed and examined in terms of appropriation with modelling activity design principles. The research revealed ChatGPT's excellence in designing activities aligning with reality and modelling generalization principles. However, it was unsatisfactory in creating activities aligning with model building, self-evaluation, and model documentation principles. The results indicate that successful prompts for initialization include defining a structure to follow, assigning roles, and using questions for best results. On the other hand, successful prompts for structuring involve defining clear goals, providing good examples, and emphasizing included and excluded information.