Asking, Playing, Learning: Investigating Large Language Model-Based Scaffolding in Digital Game-Based Learning for Elementary Artificial Intelligence Education.

This article investigates the integration of large language model (LLM)-based scaffolding within digital game-based learning (DGBL) to enhance elementary students' artificial intelligence (AI) literacy education. Using a quasi-experimental design with fifth-grade students, the study compared an experimental group receiving dynamic, personalized feedback via an LLM-powered avatar ("Mini-Xiaowei") against a control group with traditional static scaffolding. Results showed that the LLM-based scaffolding significantly improved learning achievement and reduced cognitive load, while fostering more active and iterative self-regulated learning behaviors; however, no significant difference was found in flow experience between groups. Additionally, students experiencing higher cognitive load engaged more frequently with the LLM scaffolding, indicating its role in providing targeted support. The study offers empirical evidence supporting the use of generative AI to create adaptive, interactive educational tools tailored to young learners' developmental needs, while also discussing practical implementation considerations and ethical safeguards.