Leveraging Large Language Models to Preserve Indigenous Games: A Case Study of a Chatbot for the Kenyan Game Bano

This paper looks at how large language models can be used to help preserve indigenous cultural knowledge, using the Kenyan board game Bano as a case study. The work is grounded in Cultural-Historical Activity Theory (CHAT), which helps unpack the tension between embodied cultural practice, like actually playing a traditional game, and the more static, text-based nature of digital representation. The core of the study is Simba, an LLM-based chatbot built to capture and share knowledge about Bano's gameplay mechanics, cultural context, and social meaning. The chatbot was evaluated with a group of 15 stakeholders, and the results were combined into a Cultural Preservation Effectiveness Index (CPEI), which came out to 87.6 out of 100. It scored especially well on knowledge accuracy (9.06/10) and on providing cultural context (9.2/10). Beyond the numbers, the paper digs into the contradictions that come up when you try to preserve something embodied and experiential through a digital, text-based medium. It doesn't treat the chatbot as a stand-in for real cultural transmission, but as a complementary tool that works alongside traditional practices. The paper is upfront about its limitations too, particularly the constraints around direct community engagement, and it lays out a framework for thinking about technology as a cultural mediator rather than a replacement for lived tradition.