Decoding Pictorial Collections Using Faces

This paper explores the potential of Artificial Intelligence (AI) technologies, specifically facial recognition, for enriching metadata in pictorial collections. We briefly discuss ethical and practical concerns, such as establishing metrics and standards for AI evaluation, as well as mitigating bias and privacy risks. Then three key AI methods are discussed: custom-trained face recognition (recognizing a given trained individual), notable individual detection (recognizing any of a set of pre-trained historical figures), and frequently occurring individuals (clustering faces to find those faces that appear most often in a collection). These techniques can be combined with biographical databases and biometrics to provide date approximations. Where faces appear in overlapping manners, these approximations can be spread by inference across large swaths of a collection, creating a network effect on dating and identification metadata. Finally, the paper advocates for the creation of a universal heritage-specific open-source face recognition database that could democratize access to historical data and foster cross-disciplinary research.