AI-Assisted Help-Seeking Trajectories in Programming Education from an SRL-Informed Perspective

This study investigates how university students use generative AI as a help-seeking tool during introductory Python programming tasks. Moving beyond simple metrics like usage frequency, the authors developed an SRL-informed (Self-Regulated Learning) framework to categorize 1,290 student prompts into four functional categories: Conceptual, Implementation, Debugging, and Reflective help. By linking these prompt-level interactions to 17,190 code submissions from 71 students, the researchers mapped the sequential trajectories of how students interact with AI while attempting to solve programming problems.

The analysis reveals that student-AI interaction is rarely a linear, planned process. Instead, many students adopt a reactive approach, frequently turning to AI to troubleshoot errors or generate code snippets only after encountering difficulties. The researchers identified distinct help-seeking trajectories that reflect different levels of regulatory engagement. Notably, while these patterns did not correlate with significant differences in final task scores, they were strongly associated with the efficiency of the problem-solving process. Students who relied on reactive, high-frequency debugging prompts required significantly more code submissions to reach a solution compared to those who engaged in more conceptual or reflective help-seeking.

These results suggest that the educational impact of AI in the classroom is determined by the nature of the student-AI dialogue rather than the mere presence of the tool. The findings highlight a potential risk of 'metacognitive laziness,' where students use AI to bypass the struggle of debugging and planning. For educators, this underscores the need to move beyond simply providing AI access and instead focus on scaffolding students' help-seeking behaviors to ensure that AI acts as a partner in learning rather than a shortcut for task completion.