GENERATIVE ARTIFICIAL INTELLIGENCE AND STUDENTS' INDEPENDENT LEARNING IN PROGRAMMING EDUCATION: A NARRATIVE REVIEW OF EMPIRICAL EVIDENCE
Abstract
Background: Generative artificial intelligence (AI) tools are now widely available to students who learn programming, yet it remains unclear whether higher task performance with AI leads to durable, independent learning. Methods: A narrative review of empirical studies published in 2023–2025 was carried out in September–October 2026. Studies were included if generative AI was used in a learning task and if performance with AI, performance without AI or learning processes were measured. Five controlled or experimental studies formed the core evidence; one qualitative study, one classroom deployment and two surveys were used as supplementary evidence. All bibliographic data were verified against publisher records. Results: AI consistently raised performance while it was available, but its effect on unassisted performance ranged from negative to neutral. Negative effects appeared when learners obtained complete solutions, whereas hint-based or explanation-oriented use reduced or removed them. Several studies suggest that AI may widen the gap between learners with high and low prior knowledge and weaken self-monitoring. Only one study measured delayed retention, and only one measured transfer. Programming courses should therefore assess unassisted performance separately and organise AI use in guided stages.
