This presentation examines an AI prototype developed in collaboration with Amazon Web Services to support the cataloguing needs of the Cross-Cultural Dance Resources collections at Arizona State University. These collections span more than seventy years of rare ethnographic documentation and provide a vital record of movement-based traditions from many parts of the world. The project explores how automated video analysis and culturally informed metadata design can improve access to dance materials that are often compressed into a single undifferentiated category within cataloguing systems. By integrating Laban movement analysis frameworks, the work investigates how AI can enhance discovery while resisting reductive classification. The results indicate that intelligent chunking and targeted machine learning can reduce processing costs and expand technical capacity for institutions responsible for sizable non-textual heritage collections. At the same time, the project uses these technical outcomes to open a broader critical conversation about the values that shape automated systems. It considers how AI models interpret cultural material, how archival labor shifts when automation becomes a routine part of technical services, and how librarians may influence the ethical direction of these tools. Through this case study, the project proposes ways in which art librarians, as custodians of cultural memory, can guide AI toward practices that respect traditional knowledge systems and contribute to sustainable stewardship of embodied heritage.
Shan Chuah (Tue,) studied this question.