AI Goes to School for Math-Instruction Study

INSIDE THE MIDDLE SCHOOL classroom, a video camera swivels on its tripod while five microphones capture clues to the age-old question: How do you teach math to kids?

Recordings from this lesson and thousands like it across the country will be fed into an artificial intelligence (AI) program trained to spot instances of student engagement and the teaching practices that elicit it: Maybe a pupil explains her reasoning—“it can’t be divided because it’s a prime number!”—or raises her hand several times.

It’s the first half of a project, funded by a $4.5 million grant from the Gates Foundation/Walton Family Foundation and led by UMD’s Center for Educational Data Science and Innovation (EDSI) to create a massive database arming scholars and ed-tech companies with real-world classroom data to mine for best practices.

Elementary and middle-school math scores have tumbled from pre-COVID levels, widening the gap between high and low achievers as other countries leapfrog the United States in international rankings.

The three-year UMD effort, initially supported by a Grand Challenges Team Project Grant, blends AI with roll-up-your-sleeves rigor. The recordings will be made anonymous with technical wizardry, then turned into transcripts that educators will scour to flag examples of quality instruction. Those annotations will get fed into EDSI’s MPowering Teachers program and other AI systems to support effective instruction.

“There might be teaching aspects we didn’t know of that are really important for learning,” says the study’s lead researcher, education policy Associate Professor and EDSI Director Jing Liu. “The data open up this possibility.”