What does AI look like when archival concepts and principles inform its development?
In the past, archives have used Artificial Intelligence relying on off-the-shelf tools. This practice has both limited what challenges can be met and made the needs of archives subservient to the larger field of machine learning. It may be a practical thing to do, but many alarming instances of bias have been found in modern machine learning models as applied to archival material. This raises the questions of a) whether off the shelf tools are the best solution for the archival field, b) how archival concepts and principles might influence the development of AI tools intended for records and archives management, and c) how the two fields of AI and archival science can benefit from a partnership. The speakers will discuss the archival design, development, and leveraging of AI to support the ongoing availability and accessibility of trustworthy records. We will first explain the types of AI that are most likely to support archival endeavours, and then illustrate studies on the design of AI tools for the identification of ancient records, for the classification of current records, and for detection of privacy information, as well as presenting several other studies focused on using and developing AI to support archival appraisal, arrangement and description, preservation, and access.
This theme is discussed by Luciana Duranti (University of British Columbia (UBC)-Canada, archival science), Muhammad Abdul-Mageed (UBC-Canada, AI), Luigi Compagnoni (Archivio di Stato di Milano-Italy), Emanuele Frontoni (University of Macerata-Italy, AI), Umi Moktar (Universiti Kebangsaan Malaysia, archival science and AI), and Jim Suderman (I Trust AI Researcher, Canada, archival science). The session will be moderated by Luciana Duranti.