Researchers have found that giving AI “peers” in virtual reality (VR) a body that can interact with the virtual environment can help students learn programming. Specifically, the researchers found students were more willing to accept these “embodied” AI peers as partners, compared to voice-only AI, helping the students better engage with the learning experience.
“Using AI agents in a VR setting for teaching students programming is a relatively recent development, and this proof-of-concept study was meant to see what kinds of AI agents can help students learn better and work more effectively,” says Qiao Jin, corresponding author of a paper on the work and an assistant professor of computer science at North Carolina State University. “Peer learning is widespread in the programming field, as it helps students engage in the learning process. For this work, we focused on ‘pAIr’ learning, where the programming peer is actually an AI agent. And the results suggest that embodying AI in the VR environment makes a real difference for pAIr learning.”
For this study, the researchers enlisted 18 adults who were interested in programming, 14 of whom had little or no previous experience in programming. The study participants were asked to perform two programming tasks with support from an embodied AI agent. One of the tasks was structured, giving the students a clearly defined goal; the second task was unstructured, allowing the students to create whatever code they were interested in. Students also performed a separate structured task, interacting with a voice-only version of the AI agent.
“We found that both versions of the AI agent supported the students’ ability to learn programming,” Jin says. “But the importance of the embodied agents stood out when we interviewed study participants after they had completed the tasks.
“Students reported feeling more engaged with the pair learning process when interacting with the embodied AI, and that this increased both their confidence and their motivation. Students also reported that the gestures made by the embodied AI agent – such as pointing to sections of code on the screen – also improved their ability to understand instructions from the AI.”
However, the study also revealed some unexpected challenges associated with the embodied AI.
“We found that the embodied AI could also be distracting,” says Jin. “When the AI was responding to a question from students, it was generally not a problem. But some students reported that physical movement by the AI that was unrelated to what the AI was saying could be a distraction – particularly when the AI was being proactive, rather than acting in direct response to the student.
“Ultimately, we feel this study reveals the potential of embodied AI agents for advancing programming education,” Jin says. “And the study provides valuable design insights for creating more intuitive and supportive AI programming partners.”
The paper, “From Tool to Partner: Exploring the Roles of Embodiment on AI Agent in Pair Programming,” will be presented at the IEEE Symposium on Visual Languages and Human-Centric Computing (VL/HCC), being held in October in Raleigh, N.C.
The first two authors on the paper are Xiaoran Yang and Yang Zhan, Ph.D. students at NC State. The paper was co-authored by Noboru Matsuda, an associate professor of computer science at NC State.