Adaptation in AI-Generated Mathematics Feedback: Empirical Validation of AI’s Feedback Patterns

• In Event: Considering the Interplay between Generative AI and the Mathematics Classroom

Abstract

This study examines patterns in AI-generated mathematics feedback, revealing how ChatGPT-4 demonstrates sophisticated responsiveness to characteristics of student work, yet generates feedback that reflects different instructional priorities and interpretations than those typically made by expert teachers. We identified distinct patterns across empirically validated student work categories through comprehensive qualitative coding and hierarchical cluster analysis of 180 responses to student work. Analysis revealed that AI prioritizes correctness over conceptual exploration and lacks the diagnostic interpretive capabilities of expert teachers. AI increased diplomatic correction strategies (57% versus 11%) for incorrect solutions while maintaining a universal evaluation approach. Findings suggest AI can complement human mathematical discourse, but its adaptations are surface level, highlighting the implications for integrating AI feedback tools in mathematics classrooms.

Authors

• Amanda Reinsburrow, Drexel University

• Cindy Yovanov, Drexel University

• Jason Silverman, Drexel University