Using LLMs to Adapt Serious Games with Educators in the Loop

Recently, the advancement of generative AI has brought about the opportunity to adapt and personalize learning material to individual students with little effort. This paper explores the application of large language models, such as ChatGPT, to help educators adapt educational serious games at runtime. To incorporate adaptation into serious games in a systematic way, we employ the MAPE-K loop framework. A key focus is the inclusion of educators in the adaptation process, who ensure that AI-driven changes align with educational goals. We thus propose an architecture that integrates player/learner data, game logic, and AI-generated adaptations, monitored and approved by educators via a dedicated browser-based dashboard, in a human-in-the-loop fashion. We show how we integrated this architecture into Untitled Bee Game, an existing educational serious game for eco-sustainability.