Harnessing GenAI for Teaching and Learning Math

The Problem

Millions of children around the world lack foundational math skills needed in daily life, problem solving, and many careers. Teachers are often too stretched to deliver the individualized instruction students need or lack the tools to diagnose students’ misunderstandings and tailor their teaching accordingly. Providing individualized support has historically been too expensive for most schools, particularly in low-resource settings where large class sizes, limited internet, and shared devices are barriers. 

The Solution

In collaboration with Microsoft and the education company Eedi Labs, the Computational Policy Lab will support the development of and rigorously evaluate low-cost generative AI-powered tools aimed at transforming math teaching and learning. Prior randomized controlled trials (RCTs) of Eedi’s low-cost, scalable solutions  have demonstrated clear positive effects: in a two-year RCT in the UK, students made approximately 3 additional months of progress in math after only 10-15 minutes of use per week (effect size=0.30 SD at 24 months). 

Building on this success, the new tools will be designed to give teachers real-time insights into individual students’ challenges and offer tailored instructional suggestions. They will also enable students to engage in self-paced, interactive learning to complement their classroom work. The team will test these new tools in low-resource middle school classrooms, first in the UK and then in the Global South, with the potential to reach millions of students over time. 

The Research

The team is exploring two EdTech solutions to help teachers and students:

1. Diagnostic tools to help teachers identify why students are struggling, one of the most persistent and under-addressed problems in education research. The tools will use data from homework and in-class activities to generate suggested guidance on how to help students with their biggest challenge areas. 
2. Five to 10-minute AI-powered video explanations that let students learn at their own pace and interrupt the video to ask questions as they go. 

The research will be evaluating the solutions above using individual- and classroom-level randomization and mixed-methods analysis to generate rigorous causal evidence, with the goal of measuring average treatment effects and variation across student subgroups.  

The Impact

Through this collaboration, the team hopes to explore and validate approaches to making personalized, high-quality math learning accessible to every child, everywhere. 

The research instruments, evaluation data, documentation, and code developed will be made opensource and shared through education and AI research networks. Eventually, the researchers hope to create a public-facing platform that provides access to ready-to-use AI interventions for educators around the world.

This initiative marks the beginning of a new line of work within the Computational Policy Lab focused on harnessing AI to improve teaching and learning.