The Bright Side to Generative AI in Schools

While alarm bells are ringing the warnings of potential dangers posed by generative artificial intelligence, University of Virginia education professor Michael Kennedy wonders about the possible impacts.

Yes, there is a potential for academic fraud. But could AI be a good thing, if leveraged properly with appropriate guardrails alongside existing best instructional practices.

“As researchers, we have the ability to test products like ChatGPT in controlled environments in ways that can help answer if and how they can be used for good,” Kennedy said. 

Using $2.5 million in new funding from the U.S. Department of Education’s Office of Special Education Programs, Kennedy and a team of researchers from the UVA School of Education and Human Development are researching whether generative AI could lighten teacher workloads and improve student vocabulary by providing immediate and personalized feedback to students.

AI could be useful as early as third and fourth grades, he said. That’s when vocabulary demands on students increase to include what educators call “academic vocabulary” with words that take on different meanings depending on context such as “revolution, solution, and power.”

“These upper-elementary years are a huge pivot point for students,” Kennedy said. “While some students are still decoding words as they read, they are being introduced to a suite of new vocabulary words that are critical for subjects like science, social studies, language arts and history.”

Kennedy’s research shows using rich images, engaging students with real-world examples, and providing multiple opportunities for students to respond to prompts during instruction is key for students to develop their academic vocabulary. That includes using clear language and offering feedback on students’ mastery and use of the words during discussions and hands-on learning activities.

That is where generative AI could help. Using the grant money, Kennedy and his team will explore if iterative, text-based AI in self-contained multimedia modules could provide feedback to students during academic vocabulary lessons.

That kind of feedback is difficult for classroom teachers working with as many as two dozen students at a time.

“Feedback really isn’t one moment in time,” Kennedy explained. “It is a process with many points of engagement. And teachers are incredibly limited when it comes to time, which makes providing this kind of high-level instruction difficult.”

In the project, students will develop their own vocabulary multimedia modules, requiring them to devise their own definitions, find examples, and write about how the terms are relevant to their lives. The AI program will provide feedback on the accuracy of the modules and offer examples or suggestions to keep them on course.

“High-level feedback is beneficial to all students but can be especially beneficial to students who tend to quietly plow through lessons,” Kennedy said. “Those students are at risk of making a small error that can snowball into a series of mistakes that do not get corrected in real time, which evidence suggests is really important.”

Step one for Kennedy was bringing together a team with expertise in education technology, the science of reading, and culturally responsive practices. The team includes Jennie Chiu, associate professor; LaRon Scott, associate professor and associate dean for diversity, equity and inclusion; Colby Hall, assistant professor; Rachel Kunemund, research assistant professor; and Olivia Coleman, research project manager. 

To create a tool for classroom use, they are starting by identifying non-negotiable guardrails.

“Tools like ChatGPT have access to the entire internet,” Kennedy said. “That won’t work for this application. Generative AI also spits out incorrect answers sometimes and may use language or examples that are not appropriate.”

With the team of experts writing their own answers, Kennedy hopes they can train the AI to be more helpful. And if they succeed, it could go a long way to supporting both students and teachers – especially new teachers. 

A functional AI tool could relieve some of what Kennedy calls “the extremely high cognitive load carried by new teachers.” That leads to a feeling of being overwhelmed. Repeated, day-to-day feelings of being overwhelmed could lead to teachers leaving the profession, he said.

“If you think of cognitive load as expenses coming out of a bank account, first-year teachers are drawing on that account at a very high rate,” Kennedy said. “One of the most effective ways to slow the rate of withdrawal is experience. But of course, that only comes with time.”

The team also hopes to build the AI model to help translate what the students generate in their vocabulary modules into the home language of English language learners so family members can better participate in what their student is learning.

“Creating a way to bridge the gap between school and home for students living in homes where English is not the primary language can be huge for many families and a benefit to students,” Kennedy said.

Kennedy wonders if generative AI could support students developing their vocabulary skills while easing early career teachers’ overwhelming cognitive load. With five years to develop and begin testing their ChatGPT model, he and the team are on their way to uncovering the answers.

“The core of this project is still explicit instruction and students being engaged in opportunities to grapple with the meanings of words,” Kennedy said. “The new technology provides an interesting opportunity to push the boundaries of what is currently possible for teachers and students.”