Introducing M-Powering Teachers: Math Coaching with an AI Assist

Time and again, research has found that coaching teachers is impactful for student outcomes and for improving the quality of instruction. Unlike many other professional development options, it’s fully customizable, meeting teachers where they are and addressing immediate needs in each classroom. Coaching provides support and skill development, empowering teachers to improve their instructional practices, boost student engagement, and create a positive learning environment.

Yet coaching is labor intensive—MQI coaches can spend two to three hours preparing for a single debrief session with their math teachers. Coaches are strapped for time and pulled in many directions, limiting their ability to provide teachers with regular feedback.

With limited human resources and time in the day, how can schools leverage the benefits of coaching? Machine learning technology has offered one possible solution that aims to increase the efficiency of coaching in order to increase its reach: AI-assisted coaching.

Introducing M-Powering Teachers (the M stands for Machine)

The M-Powering Teachers (MPT) platform, a machine learning-based feedback tool designed by a group of scholars with expertise in education, linguistics, and computer science, uses natural language processing to analyze verbal classroom interactions and provide formative feedback to math teachers. 

By feeding transcripts into the system using a training data set coded by humans, the technology can be trained to recognize the same kinds of events in future transcripts. 

When given a transcript of a class recording, it can chart talk distribution (how often the teacher speaks compared to students), mathematical vocabulary use, and whether teachers are taking up students’ ideas, asking focusing questions, and getting students to share mathematical reasoning, instructional practices that have been linked to improved student outcomes in math. The platform also makes it easy to revisit key moments in the transcript—moments where important instructional interactions were happening. 

The platform’s analysis helps ground conversations in evidence, and teachers are seeing the value firsthand. In a recent pilot study, one 7th grade math teacher noted a success to celebrate: of the 55 questions he asked in his lesson, 11 were focusing questions, designed to give space to student mathematical reasoning. The transcript also illuminated places to grow. Multiple teachers were surprised to learn they had spoken 80% or more of the words that appeared, and resolved to provide more space for student talk. 

Leveling up Math Coaching with MPT 

As new AI tools proliferate in education, some might worry. Can an AI tool take the place of humans? 

As district partner Emily Hare, Director of Mathematics for Guilford County Schools, shares, that is not the platform’s aim. Hare, who has been implementing M-Powering Teachers in her district in North Carolina since September, said, “As we are losing funds from COVID relief, we are having to think differently about how we provide coaching. While I’m skeptical about a lot of the AI and machine learning out there right now… [I am drawn to] this version, in how I can make a coach’s job easier and take things off their plate so they can get to the meat of what’s important—having conversations with teachers.”

Hare compared the role of the technology to that of graphing calculators in the classroom: taking over busy work so that teachers and students can dive into math and its meaning.

“How powerful would it be,” Hare asked, “if I free up my coaches’ time where they don’t have to spend an hour and a half in every classroom? We can set up a video that’s transcribed for you, so you don’t have to sit through and watch an hour and a half (which actually takes three to five hours), analyzing and taking notes about what you want to talk about with your teachers. What if that part’s done, and I can skip straight to: here are a couple of moments that are worth pursuing; let me choose some that are going to be a good learning moment?”

In other words: math coaches are not being replaced by a Chat-GPT bot who texts advice to teachers. (In fact, MPT does not use generative AI like Chat-GPT at all in its work. All transcript analyses are done using the project’s own natural language processing models on an encrypted server, ensuring the privacy and anonymity of all classroom and teacher data.)

Rather, the technology is enabling coaches to focus their work on impactful moments and to base their conversations in the actual lesson, rather than their memory and notes, while reducing coaching prep time from hours down to as low as fifteen minutes.

As the research team has described it, M-Powering Teachers Coaching is traditional math coaching, but “with an AI assist.”

The platform relies on coaches to have an impact. An initial research study found that teachers given access to a similar platform without human coaching cycles to complement it underutilized the platform and didn’t change many of the practices targeted by the platform and linked to better student learning. 

Hearing from a middle school math teacher in Guilford County Schools, this finding becomes less of a surprise. She said, “[Talking to a coach about the feedback] really rejuvenated me after some hard days of teaching to hear somebody say… at this specific moment, you nailed this…I was recognized as the leader of my classroom. I always left feeling super encouraged and positive.” 

The researchers hope that integrating the data-driven M-Powering Teachers tool with human instructional coaching can maximize efficiency and reach while holding onto the personal nature of what makes coaching so effective.

Watch the webinar below to see the M-Powering Teachers tool in action; hear from researchers, teachers and coaches; and learn more about how the tool works.

The Future of M-Powering Teachers: Get involved

This technology is new and evolving. So far, M-Powering Teachers is still in pilot mode, and the tool continues to develop in exciting new directions. 

In the future, the researchers hope to expand its features so it can evaluate if student talk aligns with lesson objectives, gauge if student small groups are being used productively, support bilingual classrooms, track teacher growth over time, and much more. 

But even at this stage, there are multiple ways to get involved. The project is recruiting new district partners to help test two coaching models. The study is open to grades 4-8 mathematics teachers, who would record lessons and receive automated feedback and coaching. The MPT app is also free and available to the public, although it is still under development. 

Reach out to Research Project Manager Hannah Rosenstein (hrosenst@umd.edu) to learn more.