Multi-Agent AI System Helps Construct Diverse Art-Historical Narratives

Concerns have been raised that generative AI might lead to standardized interpretations of art. This research explores whether large language model (LLM)-based interactions can instead foster diverse art-historical narratives. The paper introduces "Sanyu Studio," a multi-agent dialogue system where each of 321 Sanyu oil paintings is represented as an agent with mechanisms for facts, interpretation, organization, and memory filtering. A seven-day workshop involving eight art university participants demonstrated the system's capabilities. The study found that user prompts, how evidence was organized, and individual cognitive tendencies significantly shaped the creation of varied yet coherent digital narratives about Sanyu's work. The findings suggest that in situations where historical evidence is scarce, AI systems like Sanyu Studio can enhance human agency. They offer public audiences an interactive and accessible entry point into the complex process of art-historical interpretation, promoting a plurality of perspectives rather than a single, standardized view.