From Text to Text Game: A Novel RAG Approach to Gamifying Anthropological Literature and Build Thick Games

Michael Peter Hoffmann, Jan Fillies, Silvio Peikert, Adrian Paschke

Abstract: This study introduces a novel approach to gamifying anthropological literature using Large Language Models (LLM), specifically GPT-3.5, to create text-based games. Traditional methods of gamifying specialized literature often require costly game design and programming expertise. The method proposed by the authors employs Retrieval Augmented Generation (RAG) to transform anthropological classics into interactive games, potentially expanding the audience for anthropological knowledge. To evaluate this prototype, the researchers developed a corpus of 50 university-level exam questions with human-annotated gold standard answers. Together with an expert in social anthropology, they compared RAG-generated responses to these questions against both the gold standard and non-RAG approaches using a self-designed metric called ATGE (Anthropological-Text-Game Evaluation), which assesses the general quality, ethnographic depth, and honesty of the answer. Results indicate that the RAG-based system