A qualitative study on how Computer Science Students use LLMs to learn programming and how they perceive it affects their learning

Sammanfattning

Programming education is increasingly being influenced by large language model (LLM) based artificial intelligence tools such as ChatGPT and GitHub Copilot. While these tools provide new opportunities for learning and programming support, their impact on students' learning processes remains uncertain. This study investigates how computer science university students use LLM based AI tools when learning programming and how they perceive the effects of these tools on their understanding and learning process. A qualitative approach was used, consisting of semi-structured interviews and a programming task where participants were encouraged to think aloud while solving the task. Four computer science students participated in the study. The collected data was analyzed using thematic analysis. The results indicate that participants primarily viewed AI as a supportive learning tool resembling a tutor or teaching assistant that provides explanations, guidance, and immediate feedback. Participants described advantages such as flexibility, availability, and fast responses. A central finding was that educational pressures, including stress and workload, influenced how AI tools were used. During lower stress situations, AI was mainly used for explanations and learning support, whereas higher stress levels shifted usage toward assignment completion and code generation. Participants also expressed both positive and negative perceptions regarding AI use, emphasizing that learning outcomes depended largely on how the tools were used rather than the tools themselves. The findings suggest that AI usage in programming education is influenced not only by individual preferences but also by broader educational conditions. While LLM based tools can support learning and understanding, their use may also affect deeper engagement with learning processes depending on context and usage patterns.