Development and implementation of an AI-based handwriting recognition tool for math education: improving students’ attitudes and computational thinking

Abstract

Handwriting recognition for mathematical expressions remains challenging in educational settings, especially when solutions combine Korean text and symbolic notation. This study reports the design and classroom implementation of an AI-based handwriting annotation tool that uses the Mathpix OCR API to convert handwritten solutions into LaTeX and allows students to correct errors. An eight-session program integrated mathematics for AI (functions, loss functions, gradients, and neural networks) with data annotation and human-in-the-loop debugging. Eighty-three 12th-grade students at a high school in Incheon, Korea, completed pre- and post-tests of mathematics attitude and computational thinking. Paired-samples analyses showed small-to-medium gains in mathematics confidence and interest and significant improvements in algorithmic thinking, creativity, cooperativity, and critical thinking; lower raw problem-solving difficulty scores indicated fewer self-reported difficulties. Open-ended responses suggested that students viewed AI as mathematically grounded, recognized stronger links between school mathematics and real-world technologies, and gained confidence despite challenges with OCR errors and LaTeX editing. Although the single-group, short-term design in one school limits causal inference and generalizability, the findings suggest that human-in-the-loop handwriting tools can support mathematics engagement and computational thinking.