CER: AI Tools to Help Instructors Support Peer Learning in Large Programming Classes

Programmers need to be able to work in teams, and students can learn a great deal from their peers. For this reason, many introductory computer science classes adopt peer instruction, where students first try a programming problem on their own, then discuss strategies with classmates, explain ideas to each other, and debug code together. Done well, these group activities help students learn more deeply, stay engaged, and build collaboration skills. However, they are hard to run effectively in large classes, where a single instructor may need to monitor hundreds of groups at the same time. As a result, many groups never truly work together, and many that do show little learning gain. This project addresses the need to better support instructors who manage collaborative learning in large programming courses. The project develops artificial intelligence-based systems that help instructors notice important group patterns, provide timely support, and lead whole class reflections after group work, while keeping the instructor in charge of every teaching decision. The results can improve learning in large computer science courses, reduce the burden on instructors, and make high-quality collaborative learning practical at scale. The project supports faculty development, reaches local schools through outreach, and prepares students for a computing workforce in which collaboration is essential. The research proceeds in three stages. First, the project will identify which collaboration patterns are most useful to track during group programming work. These patterns may include participation balance, code progress, help seeking behavior, discussion quality, and whether group members are converging on a shared understanding. Second, the team of researchers will study how instructors prefer to respond when the system detects a group that needs support. The team will measure how effective different responses are, including private hints, prompts for peer explanation, class-wide comments about common errors, and resources for collaborative debugging. The researchers will also compare different levels of artificial intelligence assistance, ranging from instructor approval of each suggestion to adaptive support that learns instructor preferences while preserving instructor control. Third, the investigators will study how to create effective post activity debriefs that help students learn from patterns across the whole class. In this stage the researchers will see how instructors choose which issues to discuss, decide which alternative solutions to show, and gauge whether students understood the material. Support tools will be evaluated first in laboratory studies and then through classroom deployments in large programming courses. These evaluations will measure student learning, task completion, concept retention, instructor cognitive load, and the usefulness of generated debriefs. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.