Generative artificial intelligence can solve complex math and engineering problems in seconds. How do students actually use these tools when given the chance? Purdue University researchers Aparajita Jaiswal and Gaurav Nanda recently published a study to answer that question.
Jaiswal and Nanda observed 38 undergraduate students in a senior level operations management course taught by Nanda. In this exploratory study, students first solved a linear programming problem manually. They then used generative AI to tackle the exact same task. The researchers wanted to see if students used the AI simply to get an answer, or if they treated the software as a collaborative learning partner.
The results revealed a distinct lack of deep collaboration. The researchers classified student interactions into three categories: passive, active, and constructive engagement. They found limited evidence of interactive engagement. Students did not seem to question the AI's assumptions or progressively develop ideas, even during multiple back-and-forth conversations.
Nanda serves as an assistant professor in the School of Engineering Technology, and his coauthor Jaiswal is a lecturer in the Engineering Projects in Community Service (EPICS) within the College of Engineering.
Nanda noted that the specific assignment structure heavily influenced this outcome.
"The assignment asked the student to use AI to solve the linear programming problem after they had solved it manually," Nanda said. "Therefore the student's intention of using the AI might have been more confirmatory and somewhat exploratory in nature."
Nanda added that the AI provided clear, detailed answers with accurate reasoning on the first try. The software even provided additional insights about sensitivity analysis that the students did not have to compute manually. This accuracy essentially eliminated the need for students to argue or iterate with the program.
Jaiswal noted that this distinction is crucial for the future of STEM education.
"I think this highlights an important distinction between using AI and meaningfully engaging with it," Jaiswal said. "It also raises interesting questions about how the nature of a learning task influences students' interactions with AI."
How can professors stop students from simply copying AI-generated answers? Both researchers agree that educators must rethink how they design their assignments. They recommend designing assignments that require students to critically evaluate, verify, and justify AI-generated responses. Rather than simply accepting AI-generated answers, students should be encouraged to question their accuracy and underlying reasoning.
"One of the main things I would emphasize would be to design the assignments or exercises in a way which asks students to validate and verify the output provided by AI," Nanda said. He highly encourages students to ask the AI for citations and then independently validate those sources for accuracy.
Jaiswal offered a specific example for mathematical optimization problems.
"Instead of asking students to use AI to find the optimal solution, we could ask them to examine the AI's reasoning and explore what would happen if one of the problem's constraints changed," Jaiswal said. "Students would then need to explain whether they agree with the AI's interpretation and justify their reasoning."
This approach forces students to move beyond accepting an easy answer. The researchers want educators to use AI to create new opportunities for students to reason, question, and develop their own understanding of complex technical concepts.
Additional information
