A Study on Individualized Learning Support for High School Mathematics by an Interactive AI Tutor : Educational Potential and Challenges in the Age of Large Language Models

Abstract

This paper is a literature review to summarize the pedagogical potential and challenges of interactive AI tutors in high school mathematics and to identify issues for the next stage of empirical research. The results were compared and analyzed from the four perspectives of (1) learning objects, (2) support design, (3) forms of practice, and (4) learning outcome measures. As a result, the following research gaps were identified: (1) lack of validation for students with significant academic achievement polarization, (2) lack of Large Language Models (LLM) tutor practice in Japanese and high school mathematics, (3) delay in developing indices to measure deep conceptual understanding and metacognition, and (4) lack of demonstration of cooperative operation models between teachers and AI. In particular, the fact that the average score of Mathematics IA (Math IA) in the common test fluctuates ±20 points from year to year and has an amplitude 3 to 5 times that of other subjects supports the urgent need for AI-based individualized optimization support. Based on the above, we plan to introduce the Generative Pre-trained Transformer 4 (GPT-4)-based "Khanmigo-like AI tutor" into Japanese high school mathematics classes and conduct a practical study to verify its effectiveness in deepening understanding and motivating learning for different achievement levels. We hope that this research will serve as a basic reference for designing mathematics education and generating evidence in the age of LLMs.