Speaker: Dr Deblina Bhattacharjee, University of Bath
Abstract: As artificial intelligence becomes embedded in how government records are processed and made accessible, a central question emerges: how do we ensure that the pursuit of openness does not come at the cost of trustworthiness? This talk examines the tension between AI-driven access and the principles of provenance, sensitivity, and safety that underpin archival practice.
Drawing on recent developments across institutions including the UK National Archives, NARA in the United States, among others, the presentation surveys how AI is already being applied- from semantic search and automated metadata generation to sensitivity screening and document transcription. It then turns to the risks these applications introduce: hallucinated descriptions entering the archival record, re-identification of anonymised individuals, and the erosion of provenance when AI-generated content becomes indistinguishable from the original.
The talk also draws on my own work at the intersection of AI and cultural heritage, where I develop AI methods for analysing imagery such as artworks and illustrated archives. This experience has demonstrated a recurring lesson: the value of any AI output depends entirely on how well the underlying data is prepared, how transparently the process is documented, and whether domain experts remain central to interpretation. These principles, developed in the context of visual collections, apply directly to the challenges facing government records.
Rather than proposing that AI should or should not be used, the presentation argues that trustworthy use depends on design choices made before deployment: investing in data readiness, maintaining human oversight, labelling AI contributions explicitly, and building governance frameworks that keep pace with the technology. The talk is intended to open a broader discussion among participants about what responsible AI practice looks like across different institutional contexts.
Bio: Dr. Deblina Bhattacharjee is an Assistant Professor in Computer Science at the University of Bath, where she leads research at the intersection of AI and cultural heritage. Her work applies AI techniques, including segmentation, depth analysis, and style transfer, to imagery such as artworks and illustrated archives. She has created widely used datasets for analysing visual art and has organised international workshops bringing together computer scientists and cultural heritage practitioners. Her current research extends to both imagery and documents, using AI to inform questions of provenance and safety. She actively collaborates across disciplines to support the accessibility of cultural collections.