AI-assisted adaptive problem-based learning in a programming course: its effects on students' problem-solving and debugging skills

Abstract

This study examined differences in problem-solving and debugging skills between students who participated in AI-assisted adaptive Problem-Based Learning (PBL) and those who participated in the Case Method in a programming course. A quasi-experimental nonequivalent pretest-posttest control-group design involved 69 students from the Department of Curriculum and Educational Technology, Universitas Negeri Padang. The experimental group (n = 35) followed Arends' PBL with data-informed adaptive grouping and structured use of ChatGPT, Gemini, and Claude during the investigation phase, while the control group (n = 34) followed the Case Method without generative AI. Multivariate analysis of covariance (MANCOVA) tested the two posttest scores while controlling for both pretest scores. The results showed a significant multivariate difference between the instructional conditions, Pillai's Trace = 0.456, F(2, 64) = 26.780, p < 0.001, partial η² = .456. Follow-up tests also showed significant differences in problem solving (partial η² = 0.335) and debugging (partial η² = 0.435). Adjusted means were higher for the experimental group on both skills. The findings suggest that an instructional package integrating PBL, adaptive grouping, and structured AI assistance has the potential to support problem-solving and debugging skills. However, because the three components were implemented simultaneously, their individual effects cannot be separated.