Invited Talk: Shared open metadata as critical AI infrastructure

The explosive growth of Generative AI in recent years, attracting billions in investment dollars, is being applied across many areas of human endeavor. This development opens new opportunities and capacities but also raises concerns about Generative AI's tendency to hallucinate statements and facts in an authoritative manner. As a corrective to this tendency, one promising approach is to provide Large Language Models (LLMs) with graph-based knowledge bases in hybrid AI systems that ground LLM responses with factual statements about the world. With libraries, museums, archives, and other cultural heritage organizations increasingly representing their collections as graphs using ontologies such as BIBFRAME, we have an opportunity to serve as authoritative sources by offering shared knowledge bases about our cultures and world. Blue Core endeavors to provide such a shared, open knowledge base using the properties of open RDF-based metadata from contributing institutions, along with AI services to augment and support consumption by LLMs, thereby becoming a critical component of our communities' AI infrastructure.