Yet Another Digital Surrogate? Computer Vision and the Future of Collections Management Systems

The collection catalog – and today, the museum collection database – has been the foundational tool for art history for almost two centuries. Despite enormous changes over this time in our theoretical approach to the discipline, basic cataloging of objects (who created them, how and when they were made, what concepts they illustrate or formal properties they exhibit, where they have been displayed, and who has seen them) is essential to virtually all of our historical argumentation. Whether we are using these descriptions as evidence in historical argument, or challenging and replacing them with new historical contexts, art cataloging remains central. Data-based approaches to art history, including computer vision, may make description even more central, with its focus on iconography, style and formalism, pose analysis, etc.What could computational art history outputs like trained models or image vectors do for the field if they could be first-class digital objects within museum collections databases, joining the digital surrogates, conservation and provenance records, and scholarly bibliographies already curated there? In this talk I will discuss recent work on a prototype project to integrate simple machine learning mechanisms into collections management for the Carnegie Mellon University photo archives, and what its user interfaces and back-end architecture suggest for future CMS designs. Such a shift could help transform museum computer vision experiments from fun toys into deep research and exploration tools, massively magnifying our research impact in the real world and transforming the experience of both expert and casual users of these systems.This presentation was made for the conference DH Nord 2020: The Measurement of Images - Computational Approaches in the History and Theory of the Arts, November 18th, 2020