If These Walls Could Talk: Critical Play with Large Language Models in Museums
A new paper proposes a framework for using large language models (LLMs) in museums, specifically for creating role-playing chatbots that simulate historical figures or artifacts. The author, Anders Sundnes Løvlie, identifies a core dilemma: LLMs are inherently untrustworthy in their factual accuracy, which is a significant issue for museums. To address this, the paper suggests designing for "critical play," where LLM bots are intentionally unreliable but still engaging and representative of historical narratives, styles, and perspectives. AI
IMPACT Suggests a new approach for integrating LLMs into cultural institutions while acknowledging and designing around their inherent limitations.