As a middle school math teacher, Alfons Prince is used to having a lot on his plate.
“You’re one teacher in a classroom of 25 kids,” says Prince, who teaches seventh and eighth grade at Center City Public Charter School in Washington, D.C. “When a student gets stuck on a problem, it’s hard for me to stop everything and give them one-on-one attention.”
Balancing the individual needs of students with the demands of the whole classroom is an age-old teacher's dilemma. But computer science researchers with the Learning Engineering Virtual Institute believe AI technology can be part of the solution. “If ChatGPT and tools like it can provide more immediate, personalized help to students, that can free up teachers to focus where they’re needed most,” says Zachary Levonian, a senior machine learning engineer with LEVI.
Tutoring is one of the most effective supports for students, but high personnel costs make it tough to scale. Virtual math tutors offer an alternative, but existing systems tend to be inflexible and narrowly focused. If a student gets a problem wrong, the system offers a preset sequence of hints to get them back on track.
The problem, Levonian says, is that kids can get tripped up in all sorts of ways, and the system only has one way to answer.
Levonian and the LEVI team wanted to build a more versatile tool. One that could tune its response to the precise kind of error a student makes. They looked to large language models (LLMs), which excel at interpreting broad prompts and generating relevant responses. “We thought LLMs like ChatGPT could be a one-size-fits-all solution," Levonian says. "It gives teachers the ability to customize high-quality feedback for a wide range of problems and student responses."
In June, they released a ChatGPT-powered prototype called the Math Hint Generator Chatbot. The demo tool draws material from vetted internet sources and a digital textbook. It is capable of writing sequences of targeted hints for a multitude of common errors.
The initiative is supported by the Walton Family Foundation, part of our ongoing effort to look for new ideas in education and support innovators with creative solutions.
Prince was one of the tool’s beta testers. He considers himself to be fairly tech savvy. But Prince has been cautious about introducing technology into his classroom. “I want to make sure my students know how to do the math,” he says. “I fear some of that could be lost if they have a tech shortcut.”
Still, Prince was intrigued by the prospect of having a tool that could “help kids get unstuck without having to wait for the teacher.” When he got access to the demo, he entered real math problems and incorrect responses by his students to see how the chatbot performed. “It was very good at the technical math,” Prince says. “I gave it a problem about finding the volume of a cone. It used the formula correctly and in a way that could help build the understanding of the student using it.”
Although Prince was impressed with the quality of most hints, he says the tool still needs a teacher’s careful eye to vet the output. “I could definitely see using it to prepare for potential questions that kids may have,” Prince says. “Even though I have a scripted curriculum, kids are unpredictable. It would save me a lot of time to have the bot generate hints for anticipated questions ahead of time. I know a lot of teachers could use that.”
We’re really excited about the prospect of hint generation to save teachers time while increasing the personalization and responsiveness of lessons for kids.”
Although the early demo is a work in progress, Levonian says the pilot is about creating opportunities for people to brainstorm about the ways these tools can make their work more efficient. In the meantime, he agrees that the best use for the current tool is in aiding teachers with lesson preparation. Other practical applications could be using the chatbot to create hint sequences for online classwork, or simply sparking ideas for how teachers can more proactively coach kids on how to avoid common mistakes.
Looking ahead, the AI LEVI Lab team plans to integrate an updated version of the tool with Rori, a virtual math tutor they’re using in Africa. Ultimately, they hope to make it available directly to kids so they can access help at any time, at school or at home.
“Before we can do that, we need to step forward to create language that is grade-appropriate and considers cultural context,” Levonian says. “It would need to know how to respond effectively to the kinds of situations human tutors deal with every day.”
Future generations of the Math Hint Generator Chatbot could be tailored for different school systems, curricula and student populations.
“I’m really optimistic about this work,” Levonian says. “We hope it can help educators understand the opportunities of using tools like ChatGPT. Right now, teachers spend a lot of time writing lessons. That means they don’t have time for other important things. We’re really excited about the prospect of hint generation to save teachers time while increasing the personalization and responsiveness of lessons for kids.”
