Evaluating AI’s interpretations of Teachers’ Mathematical Thinking

• In Event: Teaching Practices and the Learning Sciences

Abstract

This study examines the interpretation boundaries between qualitative researchers and AI. We compared how the Large Language Models (LLMs) and human researchers evaluated the presence and strength of four core mathematical themes across five topic groupings of teacher interview responses. We find out that, without human-developed codebooks, AI recognizes patterns in phrasing but needs refinement, while humans interpret data based on the depth of context-dependent understanding.

Authors

• Kun Wang, Rethink Learning Labs

• Rachael Eriksen Brown, Pennsylvania State University - Abington

• Chandra H. Orrill, Rethink Learning Labs

• Cheng Tang, University of Georgia

• Shiyu Wang, University of Georgia

• Yaxuan Yang, University of Georgia

• Allan S. Cohen, University of Georgia