UW–Madison’s Puntambekar and Karumbaiah examine how AI can help teachers and students succeed

By Laurel White 

For the past 20 years, School of Education faculty member Sadhana Puntambekar has worked with teachers and middle school students across Wisconsin to build technology-enhanced learning environments for science learning and teaching, from solving complex design problems and writing about them to contrasting virtual and in-person lab experiments. Now, Puntambekar and a School of Education colleague are partnering to explore how artificial intelligence could be made more useful for students and teachers.

Puntambekar is a Sears-Bascom Professor of learning sciences in the Department of Educational Psychology and director of the Interactive Learning & Design Lab at the Wisconsin Center for Education Research. For the past several years, Puntambekar and her research team have been examining how well an artificial intelligence tool called PyrEval can help students learn to write better essays. That research, which is in partnership with Penn State University’s Rebbeca Passonneau, is helping push boundaries and redefine what it means to harness the power of AI in the classroom.

Sadhana Puntambekar Headshot
Puntambekar

Students involved in the project begin by analyzing PyrEval’s feedback on a selection of essays, then discussing that feedback in small groups. Puntambekar says this pre-writing reflection shows a way teachers can use AI earlier in the learning process as a formative tool, rather than an evaluative one.

“We’re not just taking PyrEval’s feedback on their essays and giving it to the students,” she explains. “Instead, it acts as a springboard for them to think about what they can revise and improve in their essays. This act of reflection is what we think is very unique to what we’re doing — we’re not just relying on the AI, we’re using it as a way to move the learning forward, to help students understand what makes a good essay.”

Once the students write their essays, the tool can also be used by teachers to evaluate trends in learning or misunderstandings across the class. This insight, Puntambekar says, can provide teachers a helpful tool in tailoring instruction and responding more quickly to gaps in students’ knowledge.

“Teachers don’t always have the time to read the essays and give feedback and base instruction on what they see in the essays,” Puntambekar explains.

Karumbaiah

As the project recently entered its third year, a first-year School of Education faculty member joined the effort. Shamya Karumbaiah, an assistant professor in the Department of Educational Psychology, reached out to Puntambekar to offer her expertise to support the work. Karumbaiah, whose dissertation investigated bias in artificial intelligence systems, is helping evaluate how feedback from large language models like ChatGPT compares to the feedback from PyrEval.

Specifically, Karumbaiah is measuring how often the tools make errors or show bias — something she believes may be unavoidable.

“I don’t think we can mitigate all the bias in these models,” she says. “This is an issue that the field has been reckoning with for years.”

The PyrEval project is just one element of a broad swath of artificial intelligence research from both professors.

Another of Karumbaiah’s ongoing projects is looking into how artificial intelligence tools respond to intersectional identities like race and gender identity. So far, her team has examined real-world data and found AI bias against Black girls shows up only when looking at their combined race and gender identities. The researchers have also found that current automated approaches to mitigate AI bias only made those biases worse for Black girls.

Karumbaiah’s team is also developing a tool to support teachers in understanding the benefits and harms of using an AI tool in their classrooms. And, in partnerships with the School of Education’s Diego Román and philosophy professor Harry Brighouse, she is examining how large language models can be made more equitable for multilingual students and how human values like fairness and privacy can be deployed when building AI tools responsibly.

Meanwhile, Puntambekar continues to work as part of a nationwide, $100 million effort funded by the National Science Foundation to advance artificial intelligence innovation in education. She says the primary focus of that project is to foster collaboration between artificial intelligence tools and students and teachers.

Across all of these efforts, both scholars say they are committed to continuing to build their research upon insights and experiences in real world settings.

“Classroom ecosystems are often much more complicated than what we deal with in the academic world,” Karumbaiah says. “I want to ground my research in meaningful contexts that give me access to teacher and student needs and lived experiences. I am trying to build tools that empower users.”

Moving forward, Puntambekar says it’s important to see artificial intelligence resources as aids that can help teachers and students succeed, building upon what many educators already do very well.

“It’s not to say we have a grand thing now and it’s going to be a solution to all of our problems,” she says. “We’re integrating it in the structures that we know work in the classroom.”

Read the full article at: https://education.wisc.edu/news/uw-madisons-puntambekar-and-karumbaiah-examine-how-ai-can-help-teachers-and-students-succeed/