Research Summary: AI literacy over tool design: a mixed-methods study of scaffolded versus unrestricted generative AI in programming education

A recent pilot study explored two approaches to integrating generative AI in master's-level programming education: a scaffolded AI study coach and unrestricted AI tool use. The research aimed to assess the impact of these methods on student learning, with the scaffolded approach emphasizing guided hints and reflection over direct code generation. This study suggests that institutional responses to AI in education are currently focused on control, either through restriction or structured integration.

Why it matters

The increasing integration of generative AI across various educational and operational domains necessitates a strategic understanding of effective implementation methods. This research contributes to the debate on whether to restrict AI use or guide its application, which is critical for developing future talent and maintaining educational integrity. Understanding these dynamics is key to informing policies that support innovation while mitigating potential risks associated with AI misuse or over-reliance.

Key insights

• A seven-week mixed-methods pilot study involved 33 master's-level data analytics students.
• Students were randomly assigned to either a scaffolded AI Study Coach or unrestricted use of AI tools.
• The scaffolded AI Study Coach provided stepwise hints, avoided generating code, limited hints per session, and required post-session reflection.
• The study design aligns with scaffolding theory, indicating an intentional pedagogical approach to AI integration.
• The primary institutional response to generative AI in programming education is currently control, through either restriction or controlled versions.
• The study took place within notebook-based laboratory sessions.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.16784