Coding with ChatGPT: Empirical Evidence of Cognitive Offloading in Computer Science Education

Generative Artificial Intelligence (AI) tools such as ChatGPT, Mistral, and Copilot are reshaping the educational landscape, particularly in programming and computer science education.Their capacity to generate, debug, and optimize code provides immediate performance benefits, yet their influence on long-term cognitive development remains uncertain.This article merges empirical evidence from a quasi-experimental study conducted with 151 first-year computer science students and critical reflection on the pedagogical foundations that have guided education for over half a century.Results indicate significant short-term performance gains in AI-assisted tasks (20-40% improvement) but weak correlations with unaided problem-solving performance (r 0.15, p > 0.05), suggesting limited long-term learning transfer.Drawing on Bloom's taxonomy and cognitive offloading theory, this paper explores how reliance on AI may alter students' metacognitive processes, diminish their ability to think algorithmically, and challenge the traditional models of knowledge acquisition.Beyond empirical findings, it argues for a new pedagogy of AI -one that balances digital assistance with critical reflection, autonomy, and ethical literacy.Generative AI should be viewed not as a cognitive shortcut but as a partner in co-creation, requiring structured integration to prevent the emergence of "artificial learners" with diminished cognitive depth.