The Widening Gap: The Benefits and Harms of Generative AI for Novice Programmers
A study of 105 novice programmers found that generative AI tools like ChatGPT create a widening performance gap: low-performing students benefit less and may be harmed, while high-performing students benefit more. The research suggests that current AI coding assistants may not be universally beneficial for beginners and could exacerbate existing skill disparities.
Background
- This paper examines how generative AI tools (like GitHub Copilot or ChatGPT) affect people learning to code, focusing on the gap between novices and experienced programmers.
- Key finding: AI assists experienced programmers more than beginners, *widening* the skill gap rather than closing it — novices struggle to evaluate, debug, or meaningfully adapt AI-generated code.
- This challenges the popular narrative that AI "democratizes" programming; the authors argue it may create a new kind of dependency, where novices learn to accept AI output without building core comprehension.
- The paper contributes to an ongoing debate in computing education and the tech industry about whether AI-assisted coding tools are net beneficial for learners, and how curricula should adapt.