Elementor: an Embodied Chemistry Learning Game Using Mixed Reality and Generative Artificial Intelligence

Mixed Reality (MR) game-based learning has been widely adopted in education, enhancing knowledge retention and enriching the overall experience. While embodied learning can be supported by MR environments that provide natural sense-making and trial-and-error in situated contexts, they often lack adaptive and personalized support. This gap can be addressed by Artificial Intelligence (AI), which enables dynamic and contextualized game scenarios. We present Elementor, an MR serious game that integrates chemistry learning and a fantasy game design scenario powered by AI. For fantasy character design, we employ LoRA model training to create anthropomorphic representations of different chemical elements. For dialogue, we employ large language models (LLMs) to enable real-time generation of context-aware interactions. We hope to facilitate engaging and effective chemistry learning. The system was evaluated through an exploratory study combining quantitative questionnaires and qualitative semi-structured interviews (N=28). Results highlight the potential of AI-powered, contextualized MR serious games for improving both motivation and learning outcomes. Our contributions cover three aspects: (1) we propose a novel AI-human codesign workflow for the development of serious MR games; (2) we explore anthropomorphic storytelling with embodied interactions in educational games; and (3) we provide design insights for future MR serious games.