While the digital archives of memory institutions have expanded exponentially in recent years, much of this material remains inaccessible. With the advent of mass digitization programs and the ever-increasing flow of incoming born digital material, the range and extent of digital heritage collections has grown enormously over the past two decades. However, due to a constellation of factors – from copyright restrictions and data protection regulation to outdated technical systems – it can often prove surprisingly difficult for researchers and the wider public to access and make use of these collections. Digitally collected does not simply equate to digitally available.

How can AI be used to improve this accessibility? This talk presents the work of KBLab, data lab at the National Library of Sweden (KB), in applying AI models trained at the library as a means of enhancing accessibility to the library’s digital collections. In the first part, we explain how and why KBLab has established itself as a key actor in the training of new Swedish AI models based on the library’s collections. In the second part, we offer a case study showing how these models can be applied to making parts of the collections currently lacking metadata searchable for researchers and users at large. More precisely, we present an example of how multimodal AI can be applied as the basis for an image search system to search and analyze large image collections, based upon the following demo: https://lab.kb.se/bildsok/. We conclude with some broader reflections on the possibilities and challenges of establishing an AI-based infrastructure, including the potential for new organizational configurations to aid further exploration about the role of AI at the library.

KBLab is a national infrastructure for digital research at the National Library of Sweden. Read more about our development projects within AI and data science on our blog.

Love Börjeson is director at KBLab and responsible for R&D at KB, as well as an affiliated researcher in applied language technology at The Stockholm School of Economics Institute for Research.

Chris Haffenden is research coordinator at KBLab and affiliated researcher at the Department of the History of Science and Ideas, Uppsala University.