Ahmad, MakMa, Kwan-LiuFirat, Elif E.Laramee, Robert S.Andersen, Nicklas Sindelv2024-05-212024-05-212024978-3-03868-257-8https://doi.org/10.2312/eved.20241055https://diglib.eg.org/handle/10.2312/eved20241055Large language models (LLMs) offer new possibilities for enhancing data visualization education, but the impacts on student experiences remain underexplored. Leveraging tenets of behaviorism, constructivism and experiential learning theories, our mixed-methods study examines LLM integration strategies. We conducted two experiments with different groups of students. The first experiment involved 95 Masters of Business Analytics students who created data narratives based on the Titanic dataset either with or without LLM assistance. The second experiment involved 30 Masters of Information and Data Science students who suggested effective visual encodings for different scenarios with or without LLM assistance in a Viz of the Day activity. We collected quantitative data from surveys and project scores and qualitative data from open-ended responses. Our results show that LLMs can enhance students' ability to create clear, accurate, and effective data stories and visualizations, but they can also pose challenges, such as requiring careful prompt crafting, producing inconsistent or inaccurate outputs, and potentially reducing students' creativity and critical thinking. We discuss how our findings suggest a nuanced balance between LLM guidance and human creativity in data storytelling education and practice, and provide specific directions for future research on LLMs and data visualization.Attribution 4.0 International LicenseCCS Concepts: Human-centered computing → Empirical studies in visualization; Empirical studies in HCIHuman centered computing → Empirical studies in visualizationEmpirical studies in HCI10.2312/eved.202410559 pages