Can AI Unlock the Power of Unlimited, Meaningful ‘Do Overs’ for Beginning Teachers?

Teaching is comprised of a collection of skills that improve with practice. Yet novice teachers are regularly taking charge of classrooms with limited experience—even after completing rigorous teacher preparation programs—forcing them to hone their skills on the job. 

“Learning on the job is very challenging for teachers,” said Julie Jackson Cohen, Charles S. Robb Associate Professor at the UVA School of Education and Human Development. “It often leads to burnout, teacher turnover and even negative student outcomes.”

For years, Cohen has studied how the use of simulator-based practice sessions—paired with feedback from coaches—can boost teachers’ skills. In partnership with Mursion, pre-service teachers practice with a virtual classroom of students voiced by live actors. The good news is the simulations paired with coaching work. The bad news is practice without coaching can be negative, while the actors and the coaches are expensive and difficult to scale. 

“With hundreds of thousands of new teachers entering classrooms each year, we need innovative, scalable methods to help promote effective and efficient skill development,” Cohen said.

Cohen believes generative artificial intelligence might be the answer. Her new project, with new funding from the Bill & Melinda Gates Foundation, aims to build flexible, affordable AI-driven simulations novice teachers can use to practice and receive feedback. 

Unlimited Do-Overs

The heart of this project is creating opportunities for beginning teachers to practice, make mistakes, receive individualized feedback and try again—a cycle that is impossible in a real classroom.

“Teaching is so hard, and we have less and less time to teach people to do it,” Cohen said. “In response to the teacher shortages, we see lots of teachers going straight into classrooms with limited or no practice. That basically means they’re practicing on real students.”

The AI-driven simulator is potentially one answer to Cohen’s question: How can we help teachers feel and actually be prepared so they stay in the profession?

According to Cohen, teachers who enter classrooms unprepared, often create negative beliefs about their students or the schools in which they work. These negative beliefs often lead to teachers leaving the job.

“We are hoping to break that cycle,” Cohen said. “Helping teachers see that there are skills they can improve shifts how they see their students. Instead of thinking it is a student’s fault for struggling to learn a math concept, they see there are things that they can learn to do to help that child be successful. Down the road, this sense of self-efficacy can increase their likelihood of staying in the classroom.”

The simulators Cohen previously used provided opportunities for practice and coaching. But because they used human actors and human coaches, they were limited in the number of times they could practice. Artificial intelligence provides the opportunity for unlimited do-overs.

With Feedback

Cohen and her team aim to train the AI to both generate realistic student responses to what the teachers do in the simulation, as well as to provide automated, customized feedback to teachers to improve the skills needed to teach math effectively.

“Coaching is the heart of this work,” Cohen said. “Over the past decade, we have conducted dozens of studies of simulated practice. We consistently see that practice does not make perfect. Teachers really need information about what they are doing well, how it supports positive student outcomes, and how they can develop skills that might be more challenging.”

When a beginning teacher is practicing in the simulator, they receive detailed and specific information about their strengths and weaknesses. They also have chances to talk through why their instruction might have elicited certain responses from the student avatars. Coaching sessions conclude with the teacher helping to develop a “game plan” for the next simulation session.

This kind of feedback aims to push the beginning teacher to do just a bit more than they could by themselves without generating overwhelm, something Cohen calls the teachers’ zone of proximal development. The coaching aims to provide just enough scaffolding and support for teachers’ iterative skill development.

The two-year grant from the Gates Foundation will support Cohen and her partners: co-investigator Heather Hill, Hazen-Nicoli Professor in Teacher Learning and Practice at Harvard University, Hallie Parten, a UVA research specialist, and colleagues at the Branch Alliance for Educator Diversity.

Together the team will build a prototype, train the AI and build user experiences they will test with teachers and teacher educators across the country, ensuring that the simulator feels realistic and useful.

“We want to be sure we are building something that could be integrated into teacher preparation programs,” Cohen said. “A scalable tool like this could provide hours of practice and targeted feedback, building much-needed skills for pre-service teachers before they take over their first classroom.”