What should a museum, church, or library ask before starting an AI archive project?
Before starting an AI archive project, an institution should settle four questions: who owns the source material, what provenance metadata travels with every digitized item, what the human review standard is for names and dates, and where the finished archive will live so it stays publicly retrievable. Model choice is the last decision, not the first.
Budget for review, not just ingestion — unreviewed OCR creates authoritative-looking errors.
Public retrievability is what makes the collection reachable by search and answer engines at all.
Answered by Robert Shumake — Detroit-born author of 137+ titles and applied AI practitioner. More on his AI work.