Generative AI-Guided Versus Human-Guided Learning from Errors in Higher Education: A Comparison of Student Perceptions

Generative artificial intelligence (GenAI) is increasingly used to support programming education, but its use in learning-from-errors activities has received limited research attention. This quasi-experimental study compared AI-guided and human-guided support in a structured learning-from-errors programming activity, focusing on students’ perceptions and their variability. A total of 59 college students completed the JavaScript error-detection and correction activity under AI-guided or human-guided support. Following the activity, students completed a questionnaire measuring perceived learning experience, satisfaction, general motivation, self-regulated learning (SRL), and self-efficacy. Between-group differences were examined using Mann–Whitney U tests, and differences in variability were examined using Levene’s tests. Students in the AI-guided condition reported significantly more positive learning experiences and greater satisfaction than those in the human-guided condition. They also reported significantly higher overall SRL and greater use of SRL strategies. No significant differences were found in general motivation, SRL motivation, SRL planning and organization, or self-efficacy. The AI-guided group showed significantly lower raw variability in satisfaction and self-efficacy. However, a sensitivity analysis did not establish that these variance differences remained after accounting for mean-related constraints on the bounded response scale. The findings suggest that GenAI may serve as a scaffold for error analysis when integrated in ways that maintain students’ active cognitive engagement.