The Efficacy of Using Generative AI to Render Illustrative Diagrams to Accompany Mathematics Word Problems

• In Event: AERA Roundtable Session Friday 11:45 am - Gold 4
  In Roundtable Session: AI-Enhanced Problem Solving and Instructional Design Frameworks

Abstract

Recent advances in generative AI allow new possibilities for auto-generating visuals to accompany educational tasks, like mathematics word problems. In mathematics education, visuals can be beneficial or harmful to learning, depending on the extent to which they support students in making relevant connections to their everyday experiences while avoiding the addition of seductive information to problem-solving. Here, we report a study in which three different GenAI approaches were used to create visuals to accompany mathematics word problems, with one approach being both novel and based on the most recent advances in image generation. We evaluate the approaches with theory-based metrics that measure image quality to examine their affordances and constraints. We give implications for the design of future multi-agent systems.

Authors

• Candace A. Walkington, Southern Methodist University

• Theodora Beauchamp, Southern Methodist University

• Jiabao Wen, Southern Methodist University

• Jonathan Isaac, Southern Methodist University

• Robert DiBiano, Ailectric