Exploring Middle School Students' Interactions and Perceptions of Using Generative AI in STEAM Learning Environments.

Generative artificial intelligence (GenAI) is rapidly reshaping educational practice, yet its role in supporting K-12 STEAM learning remains underexplored. Using an exploratory design, this study aims to implement two GenAI-based STEAM learning environments, exploring middle school students' interaction behaviors and perceptions of using GenAI across two 120 min modules: Mathematics-Engineering and Mathematics-Arts. Data were collected from 12 eighth-grade students in China through student-GenAI interaction logs and follow-up interviews. The results revealed that the students working in groups mainly used GenAI to acquire STEAM knowledge/information, relied on exact copying of instructions, favored direct retrieval/review prompts, and used GenAI mainly to request information. Comparatively few interactions involved review, verification, evaluation, or reflection. The participants perceived various roles of GenAI and expressed high satisfaction and trust; however, they demonstrated limited critical evaluation of GenAI outputs, lacked a sense of engagement, and achieved minimal feelings of accomplishment. Although the interaction counts were relatively comparable across the two modules, satisfaction was slightly higher in the Mathematics-Engineering module. Most students perceived no module difference in trust and GenAI utility. Given the small sample and short-term two-module design, the findings are exploratory and context-specific, limiting their generalizability and precluding conclusions about the comparative effectiveness of GenAI-supported STEAM learning. This study provides preliminary evidence on how students interact with and perceive GenAI in STEAM contexts and highlights the need for context-sensitive integration, explicit support for critical evaluation, faded prompt scaffolding, and deeper human-AI collaboration.