A new project at the University of Miami School of Education and Human Development seeks to empower students by transforming the role of artificial intelligence in middle school math classrooms—not simply as a technological tool but as a “teachable agent” that students will instruct, guide and learn.
The project, called MAGICAL-Math: Multimodal Artificial Generative Intelligence Championing Advanced Learning Sciences for Math Education, is led by Wanli Xing, professor at the school and the John S. and James L. Knight Foundation Chair in Artificial Intelligence at the University’s Frost Institute for Data Science and Computing. The study partners with the University of Florida, University of Utah and Carnegie Learning Inc.
The project is funded by a $3.75 million grant from the U.S. Department of Education and a $375,000 grant from Carnegie Learning, a provider of K-12 education technology, curriculum and professional learning solutions.
Xing’s team will use generative AI to reverse the role of students. Focusing on middle schoolers, the students will become active participants in learning by teaching AI agents mathematical concepts. The approach is designed to boost achievement, deepen engagement and support students from economically disadvantaged backgrounds.
“MAGICAL-Math will not replace teachers,” Xing said. “Teachers will lay the foundation, and we will be collaborators with them.”
The project will develop a low-cost, multimodal and ethical AI-enhanced system that can be integrated into digital learning platforms already used by schools.
Xing will lead a team that plans to create a software development kit that would allow the technology to be customized and deployed across educational settings.
Laura Kohn-Wood, dean of the School of Education and Human Development, said Wanli Xing’s work “explores how technology, particularly AI, can transform education by creating innovative, personalized learning environments that accommodate student variability across the educational spectrum.”
“This commitment is central to the school’s TransformED vision, which reflects both education’s transformational power and our capacity to improve educational outcomes and advance the greater good,” she noted.
Students will interact with an AI-powered “learning buddy” through multiple forms of representation, including language, visuals and AI-generated feedback.
“We have found that teaching math to middle schoolers can be quite challenging,” said Xing. “We find that students passively receive information and focus on procedural learning activities, which means that students do not own their learning process and do not develop deep conceptual understanding.”
MAGICAL-Math offers an alternative. The premise is simple but powerful. When students have to teach a concept, they often understand it more deeply themselves, said Xing. MAGICAL-Math builds on that “learning-by-teaching strategy” by asking students to explain, correct and guide AI agents as they work through mathematical concepts. Beyond texts and traditional lectures, the AI model offers videos, audio and other visual elements to engage the students.
The study, which will start later this year, will take place in public middle schools serving grades 6 through 8 in Duval, Seminole and Broward counties. Talks are underway with Miami-Dade County school educators, said Xing.
The project is focused on schools that serve large populations of minority and low-income students, with more than half of students in the participating districts qualifying for free or reduced-price meals.
In the first phase, researchers will work with 10 middle school math teachers from varied backgrounds to co-design the AI teachable agent. They also will gather feedback from 100 teachers through an online professional development program about their experiences using generative AI tools in the classroom. The team will then conduct usability and feasibility studies.
The three-year randomized controlled trial is expected to involve 46 teachers and 3,450 students. Teachers in the treatment group will use MAGICAL-Math weekly with students, while control classrooms will continue using traditional math teaching methods without the added support of the MAGICAL-Math agent.
The MAGICAL-Math system will include three major pieces: a generative AI-powered teachable agent that encourages students to learn by teaching; an AI-powered mentor that assesses learning and provides multimodal feedback, including AI-generated videos; and a responsibility module designed to monitor fairness, accuracy and safety in student-AI interactions.
Teachers can interact at any point during this process to ensure the AI agent is not providing wrong or misleading information, said Xing.
To evaluate the tool, researchers will use several measures of student learning and engagement, including Florida Assessment of Student Thinking scores, student interest in mathematics and weekly interaction logs.
Beyond the classroom trial, the project is expected to produce freely accessible digital resources for teachers and students, peer-reviewed publications, conference presentations, outreach materials for education stakeholders and a final dataset, said Xing.
