AI to review government records: new work to unlock historically significant digital records

Abstract

Government professionals need to work closely with archival institutions to make sure that important born-digital records are identified and preserved in the short term, before being transferred for long-term archiving and access.Knowledge and information (KIM) professionals in government and archivists face similar challenges.In the UK, government records (in digital and paper form) must by law be transferred to archival institutions after 20 years.In the USA, the National Archives and Records Administration issued a memorandum in December 2022, stating that "after June 30, 2024, NARA will no longer accept transfers of permanent or temporary records in analog formats and will accept records only in an electronic format with appropriate metadata." 1 These masses of born-digital government records must be reviewed to select historically significant documents for preservation, and delete ephemeral information.2 This process of appraisal cannot be done manually.Given the volume and complexity of digital information, and the rate at which digital information is being created, using computational appraisals such as Artificial Intelligence is a no longer a choice, but a necessity.Without automation, the risk is that the number of unstructured and unmanaged records keeps growing (these records, and their exponential rise, are often referred to as "digital heaps").The lack of discoverability and accessibility can lead to a lack of transparency and accountability, and to legal liability in case of freedom of information (FOI) requests 3 or public inquiries.In the long term, the inability to find, select and transfer important records has an impact on the historical archive available to researchers and other users in archival institutions.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 10 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.