Patterns of Learner-AI Interaction and Academic Performance in an Object-Oriented Programming Course

Synthesis: Examines how different forms of learner-AI interaction relate to learning outcomes in object-oriented programming courses. Identifies distinct patterns of GenAI use among students and correlates them with academic performance, finding that certain interaction patterns (seeking explanation rather than code generation) are associated with stronger learning outcomes.

Key Findings

• Survey data were collected from 210 undergraduate students in a first-year object-oriented programming (OOP) course where GenAI use was permitted for coursework but prohibited in assessments.
• Students used GenAI significantly more often for explanation seeking and debugging than for code generation.
• Cluster analysis identified five distinct learner-AI interaction profiles, including a "smart" high-usage pattern characterized by low reliance on code generation and high use for conceptual support and learning gains. Usage patterns were associated with differences in perceived assignment difficulty, self-assessed understanding, trust in AI-generated code, and norm-related attitudes.
• Critically, no significant differences in assessment performance were found across clusters — self-directed GenAI use alone did not lead to measurable learning gains.

Study Design & Method

This full research paper investigates how students integrate GenAI tools when learning OOP and how different patterns of use relate to learning experiences and outcomes. The study surveyed 210 first-year undergraduates about their self-directed GenAI use, academic performance, perceived difficulty, understanding, and trust. Cluster analysis was used to derive learner-AI interaction profiles, which were then compared on the self-report and performance measures.

What this means for practice

• Learners. Use GenAI for explanation seeking and debugging rather than code generation: students reached for those purposes significantly more often, and the profile with the lowest code-generation reliance reported the strongest conceptual support.
• Learners. Do not read heavy tool use as progress — the higher-usage profiles reported more trust in AI-generated code alongside greater perceived assignment difficulty and lower self-assessed understanding.
• Learners. Expect self-directed use alone to move no grades: no significant assessment-performance difference appeared across the five interaction profiles, so treat the tool as support for reasoning rather than a substitute for working problems through.
• Learners. Justify AI-assisted solutions and compare AI-generated code against your own, since the study ties deeper engagement to active explanation seeking, critical evaluation of outputs, and reflection on process rather than to unguided task completion.

Limitations

• All measures were self-reported, with no logs of actual AI interaction, so the five clusters are perceived usage profiles subject to recall bias and social-desirability effects, especially on rule compliance.
• The sample was 210 first-year undergraduates at a single institution in a single OOP course, limiting generalization to other settings and programming paradigms.
• A moderate silhouette coefficient means the clusters overlap, so the profiles capture broad tendencies rather than sharply separable categories.
• The design is observational rather than experimental, so the findings are descriptive associations; engagement depth was not measured and some clusters were small, leaving the study possibly underpowered for small performance differences.

Citation

Marina Lepp (2026). Patterns of Learner-AI Interaction and Academic Performance in an Object-Oriented Programming Course.