What is it about?
Data science students increasingly learn by tackling real coding projects, often with help from AI chatbots. But we know surprisingly little about how they actually use these tools. We built an AI tutor that lives directly inside a Jupyter notebook — the coding environment students use — and gave it access to their code and data so it could offer relevant help. We then watched how 24 students at two universities used it while solving a real Kaggle challenge. We found three distinct styles: some students used it for quick answers, some leaned heavily on it to generate code, and others used it to think more deeply and work independently.
Featured Image
Photo by James Harrison on Unsplash
Why is it important?
As AI chatbots become a standard part of how students learn to code, a key question is whether they build genuine skills or simply hand over answers. Rather than asking only whether AI tutors "work," we show how the same tool produces very different behaviours in different students and how those behaviours shift over a single session — with some students growing more reliant on AI over time while others move toward solving problems on their own. We also release our Jupyter templates and the library used to build the tutor as open-source, so other educators can adopt and study the approach. This offers teachers a practical way to see how their students are really engaging with AI, not just whether they are using it.
Perspectives
What struck me most was how differently students engaged with the same tool. It was a reminder that giving students an AI tutor is not a single intervention — it is really many, depending on how each learner chooses to use it. I hope this work encourages educators to look past whether students use AI and instead ask how, because that seems to be where the real learning happens.
Leon Ciechanowski
Akademia Leona Kozminskiego
Read the Original
This page is a summary of: Interaction Patterns With Data Science AI Tutor In Project-Based Learning, July 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3803400.3809369.
You can read the full text:
Contributors
The following have contributed to this page
