The ethical implications of introducing FRT into archival metadata description and catalogue search for photographic and audiovisual collections

This dissertation focuses on the ethical implications of introducing facial recognition technology (FRT) into archival metadata description and catalogue search for photographic and audiovisual collections. Relevant debates are informed by and relate to biometric recognition machine learning (ML) technologies more broadly, and discussion of these also features.

The dissertation aims to explore how theory, combined with direct conversation with archivists, can progress ethical debates on this subject. It does so by analysing existing literature on the topic, as well as consulting working archivists via two focus group sessions involving five participants. In doing so, it synthesises theoretical writings with how these may manifest in practice, using the personal and professional views of archivists to kickstart an ethical archival debate that I hope will be picked up and continued by others.

My findings demonstrate where current overlaps exist between theory and practice, for example that archives must remain considered and gradual with their adoption of new technologies, as well as where the two disagree with each other. For example, the dissertation explores clashes between a theoretical desire to open up collections further through the use of technology and practical restrictions around staffing and capacity, as well as the threats this could pose to privacy and consent processes if handled incorrectly.

It begins to bridge a gap between theoretical and practical research approaches and demonstrates a model for future enquiry into the ethical question of what place FRT should have in archival metadata creation and searching.

Key words: 

Machine learning (ML), artificial intelligence (AI), facial recognition technology (FRT), ethics, colonial legacies, decolonisation, algorithmic bias, photographic archives, audiovisual archives, privacy, access, hybrid intelligence, information retrieval, biometric information, explainable AI, dual-use technology.