Middle Schoolers Teach AI Agents to Solve Math Problems

Researchers at the University of Miami are launching a project that redefines the role of artificial intelligence in middle school mathematics. Instead of acting as a simple tool for answering questions, the AI is designed as a "teachable agent" that students must instruct, guide, and correct. This approach aims to transform passive learners into active educators of the technology itself.

The initiative, known as MAGICAL-Math, is led by Professor Wanli Xing and partners with institutions including the University of Florida and Carnegie Learning. Funded by a $3.75 million grant from the U.S. Department of Education and additional support from Carnegie Learning, the three-year study will take place in public schools in Florida. The goal is to boost student achievement and deepen conceptual understanding by forcing students to explain mathematical concepts to an AI that can make mistakes.

Flipping the traditional classroom dynamic

Traditional math instruction often relies on students passively receiving information and practicing procedural steps. Professor Xing notes that this method rarely leads to deep conceptual ownership. By asking students to teach the AI, the project leverages the "learning-by-teaching" strategy. To explain a concept clearly enough for a machine to understand, students must break down their knowledge, identify gaps in their own understanding, and articulate the logic behind mathematical processes.

The AI agent is designed to be multimodal, interacting with students through language, visuals, and audio. It is not intended to replace human teachers but to serve as a collaborator. Teachers will lay the foundational knowledge, while the AI provides a low-stakes environment for students to practice explaining and correcting errors. This interaction is intended to make learning more engaging and accessible, particularly for students in economically disadvantaged backgrounds who may lack other tutoring resources.

Targeting underserved communities for impact

The study will focus on middle schools in Duval, Seminole, and Broward counties that serve large populations of minority and low-income students. In these participating districts, more than half of the students qualify for free or reduced-price meals. The project is specifically designed to support these students by providing a personalized, AI-enhanced learning environment that can adapt to individual variability.

In the first phase of the research, the team will work with ten middle school math teachers to co-design the AI agent. They will also gather feedback from one hundred additional teachers regarding their experiences with generative AI tools. This collaborative design process is intended to ensure the technology is practical, ethical, and integrated smoothly into existing digital learning platforms used by schools.

Building a customizable educational toolkit

A key component of the project is the development of a software development kit. This toolkit will allow the technology to be customized and deployed across various educational settings. By creating a low-cost, open system, the researchers hope to make this AI-enhanced learning approach scalable. The system is built to integrate with current digital learning platforms, reducing the need for schools to purchase expensive new hardware or software suites.

According to Laura Kohn-Wood, dean of the School of Education and Human Development, the project aligns with the school's vision of using technology to transform education. The aim is to create innovative learning environments that accommodate the diverse needs of students. By empowering students to take charge of the learning process, the project seeks to improve educational outcomes and advance the greater good, as reported by GN technics/ai (en-US).