Abstract
With the rapid development of generative artificial intelligence (AI) models, there is a growing need to critically assess their role not only as tools for automating problem-solving, but also as potential mediators in the development of complex cognitive and metacognitive skills. This study empirically investigates the impact of interaction with generative AI on students' ability to transfer algorithmic thinking-a key component of successful computer science education.In a controlled experiment involving 76 students, we conducted a comparative assessment of academic performance across three task types: near transfer, far transfer, and metacognitive tasks. A mixed-methods approach was employed, combining quantitative analysis (success rates, subjective cognitive load, and confidence levels) with qualitative methods (behavioral log analysis and semi- structured interviews). In the experimental group, which used generative AI as an interactive tutor, statistically significant improvements were observed across all task types (p < 0.01), with particularly strong gains in far transfer tasks. Additionally, the group demonstrated higher metacognitive confidence, more meaningful explanations of solution strategies, and a greater tendency toward reflective interaction with the AI.The findings highlight the potential of generative AI not only as a support tool but also as a catalyst for cognitive transformation in learning–provided it is used in a constructive, non-instrumental manner. At the same time, risks related to superficial copying underline the importance of pedagogical guidance and the cultivation of self-regulation in students. This work lays the groundwork for future research on the integration of AI systems into educational trajectories aimed at fostering knowledge transfer and metacognitive development.