The archival description backlog, which kept vast amounts of historical and institutional material effectively invisible
Current factItem-level description was never affordable at the rate collections arrived; the profession's own 'more product, less process' minimal-processing standard was an explicit acknowledgment that most material would only ever get coarse description, producing a 'hidden collections' problem of accessioned but functionally unfindable material.
AI-drafted description -- LLMs and multimodal models generating first-pass catalogue records and finding-aid text from digitised or transcribed material, verified by a trained archivist rather than written from scratch -- shifts labour from drafting to verifying, letting a single archivist move much more previously-undescribed material to a finer grain of description.
A single archivist making a large, previously undescribed backlog collection genuinely findable at a much finer grain of description than minimal processing ever allowed, because AI drafts the description and the archivist verifies it rather than writing every entry from scratch.
ArchivingCuratingA small institution with one archivist clearing years of accessioned-but-undescribed material without needing to hire additional cataloguing staff.
AuditingDocumenting
- Federated multi-LLM archival description research (arXiv 2504.05711)
- 'Making Hidden Collections Visible' human-AI collaboration case study (Library Resources & Technical Services)
- AI-assisted archival metadata generation platform (Springer 2025)
Low risk on direction, since AI-assisted description drafting is a durable capability improvement. Moderate risk on magnitude, because as the tooling becomes standard at well-resourced institutions the advantage shifts from being able to use the tool to being able to verify its output reliably.
This is a collapse in description-drafting throughput, not in appraisal judgment. Every AI-drafted description still needs a trained archivist's verification before it becomes the permanent record, because a confidently wrong description actively re-hides material rather than merely failing to help. The collapse reaches further for typescript-era material than for difficult historical handwriting.