Featured
Northeastern education professor Tomás Galguera designed a project in which students worked in small groups using ChatGPT and found it had a positive effect.
OAKLAND — Artificial intelligence can help create more collaborative learning environments and help students build reasoning skills, Northeastern University researchers say.
Working with fourth and fifth graders at the Children’s School at Northeastern’s Oakland campus, education professor Tomás Galguera designed a project this fall in which students worked in small groups debating whether controlled burns are good for forest health.
The activity — called constructive controversy — is common in classrooms for students of this age range, as it seems to foster more complex learning as well as cooperation. But these students had an additional debate partner: ChatGPT.
Two students took one side of a controversial issue and another two took the other side. Each side did research, presented its position and answered clarifying questions. Then students switched sides and added more evidence to the original side’s positions.
Galguera, whose department on the Oakland campus works directly with the Children’s School and its teachers on action research, wanted to know how AI would fit into the mix.
“The geeky side of me was curious to see if ChatGPT had the chops for the whole thing,” he said. “Not simply asking clarifying questions, but then switching, providing the other side and getting engaged in the whole process.”
Working with Lindsay Schaeffer, lead teacher for the fourth and fifth grade combined class, he wanted to observe how fourth and fifth graders would feel about working with AI and how their feelings changed over time. Galguera also wanted to find out how using ChatGPT influenced the way students work with each other and what influence AI had on their reasoning skills.
Galguera and Schaeffer will write a paper documenting the project and its outcomes once they have analyzed their observations and data. But results so far show that students who were nervous at first about working with AI very quickly adapted to it. Significantly, Schaeffer said, it was also clear that including ChatGPT in the process had a positive impact on the way students worked together.
“There was this layer taken out, a layer of what could sometimes lead to misunderstanding or frustration with their peers,” she said. “They were practicing the skills of listening and synthesizing.”
During the project, students conversed with each other and ChatGBT, asking the AI app questions, which asked students clarifying questions in return, Schaeffer said. By the third round their language got more clear and precise.
“They got clearer,” she said. “They got more detailed. They brought in language that the clarifying question might have asked. So it really did matter.”
It was also clear, she said, that students were more receptive to feedback and questions when ChatGPT asked.
Ordinarily when students hear presentations from one another, there’s a possibility of interpersonal dynamics getting in the way, she said. Students might get defensive when questioned by their peers. But when an AI debate partner asks questions or offers information, students hear it differently.
Galguera noted that it helps to be very polite when asking the AI app a question. Using “please” and “thank you” seems to lead to better results, he said.
One student experienced ChatGPT’s preference for politeness firsthand. While doing research, the student was getting tired of the way the conversation was going and the AI app picked up on it.
“It came back at him with, ‘it sounds like we’re ready to move on,’” Schaeffer recalled. “The student said, ‘Whoa, I’m going to stop talking for a minute and let my partner talk.’”
Another student asked ChatGBT questions that none of the other students in the class had asked and was rewarded with information about tree and plant seeds that require periodic fire in order to germinate. That information became an argument in favor of prescribed burns.
“When you provide the same resources for a whole class, it’s a more homogeneous experience for everyone,” Galguera said. “This one was very individualized.”
Some kids in the class were nervous and initially thought there would be robots in the classroom.
“We have some families who don’t allow any screen time,” Schaeffer said. “I think that can translate to fear or distrust. There’s a healthy dose of that in my opinion, but it’s more nuanced.”
Students knew that AI can be biased and even wrong, Schaeffer said. And even though including ChatGPT into their group work became normal, students retained their skepticism.
One day in class Galguera asked ChatGPT to generate an image of a fifth grade class doing a debate. The image that came up showed a class of white students, which led the racially diverse class at the Children’s School to burst into laughter. When Galguera corrected the prompt to ask for an image of a typical fifth grade class in Oakland, California, the result was more reflective of the actual class.
“It’s really important that they saw the bias in the picture, that a classroom full of white children doesn’t represent us,” Schaeffer said. “But more importantly they understood why AI created a class like that and what kind of data is out there.”
What comes next for Galguera and Schaeffer is to consider how working with an AI partner might boost literacy skills. Written transcripts of interactions with AI may help students with drafting their own written work, for example.
One thing is for sure: these fourth and fifth graders have accepted ChatGPT in their classroom.
“Any fear or nervousness they had around it has turned into critical questioning,” Schaeffer said. “They’ve had this foundational experience that normalizes its use.”