Memory Institutions Meet AI

Lessons from Critical Technology Discourse

Abstract

Galleries, libraries, archives, and museums (GLAMs) across the globe are building new datasets to render their collections open, machine-readable, and internet-accessible. The new generation of GLAM datasets have wide reach, offering to turn institutions inside-out so that remote audiences can view, download, share, and remix digital assets. GLAM institutions have treated the associated turn to open data as inherently positive—able to promote cultural understanding and appreciation in ways that promise scale, accessibility, and customization. However, some critics suggest that upsides of technology for GLAM datasets need to be balanced with risks that can arise from their design, development, and integration into artificial intelligence (AI) technologies. In this work we ask: how should GLAMs account for the emergence of AI-driven experiences built upon GLAM datasets? We seek to answer this question by flagging key ethics and governance issues in tandem with supplying some guardrails for navigating them. We examine GLAM datasets from a sociotechnical perspective. Drawing on our experiences as researchers spanning multiple areas (computer science, computer vision, AI ethics, art history, and cybersecurity) and working in different sectors (industry and academia), we identify salient concerns and remediations from critical technology discourse on dataset development for AI systems.