Towards AI-assisted Board Game-based Learning: Assessing LLMs in Game Personalisation
DOI:
https://doi.org/10.34190/ecgbl.18.1.3055Keywords:
Artificial Intelligence, instructional design, board games, personalization, moddingAbstract
Board games are used in different educational settings to promote acquisition of disciplinary content, soft skills, foster engagement towards learning content, and sustain motivation. However, designing and conducting effective educational activities with board games requires instructional design skills, knowledge of games, as well as the ability to align the player's internal goals with the learning objectives. Board game-based learning (bGBL) design includes choosing appropriate games and personalising them to better fit with the educational setting and the students' individual needs. This complexity, coupled with a general lack of teacher familiarity with games and game culture, is likely a reason for the relatively low use of board games in formal instructional settings such as schools. Artificial Intelligence (AI) has long been used in education, but the recent diffusion of user-friendly tools for large language models (LLM) opens a new range of possibilities to assist teachers and educators in instructional design: This includes the design and implementation of bGBL. In a preliminary study, we explored the ability of a well-known chatbot, ChatGPT, to select board games and suggest modifications to better align with the classroom context and personal student needs. However, this study was limited to a sample learning unit and was based on a posteriori evaluation by experts. In this contribution, we develop a new testing protocol to assess the reliability, effectiveness, and context sensitivity of several LLMs to adapt given board games to different classroom scenarios. The methodology features blind comparison of AI and human experts. The results suggest that general-purpose AI tools such as Copilot, Claude, and ChatGPT can provide quality and context-sensitive board game modification suggestions, to the point of slightly overperforming human experts in aligning personalisation suggestions to instructional goals. This study represents a first foray in the use of general Artificial Intelligence to assist the modding, or personalisation, of board game-based learning activities and, despite its limitations as a pilot experiment, might pave the way for successful integration of assistive technologies in game-based learning and facilitating its integration in the school curriculum.