A woman in glasses reading in a library

28 Oct AI Glasses Are Writing Library Records. Should Genealogists Trust Them?

Posted at 09:00h in African American Genealogy by

A woman in glasses reading in a library

Photo: cottonbro studio on Pexels. Illustrative image.

Picture a cataloguer holding a book. She wears glasses. She looks at the cover and the title page, and a catalogue record starts writing itself.

That’s the prototype the Royal Library of Belgium, KBR, was set to present at the AI4LAM community call on October 20. According to the announcement, the system detects the book, pulls out the bibliographic details, and drafts a MARC21 record while the cataloguer handles the physical item. KBR calls the approach “trust in the loop.” The cataloguer stays involved in reviewing, validating and refining what the machine drafts.

I’m an archivist with an MLIS, and my first reaction was not “that’s magic.” It was “who checks the draft, and against what?”

The mistake I’d make

I’ve been fooled by a tidy record before. A catalogue entry looks authoritative. It has fields, dates, a call number. In my own research I have to remind myself that a clean record is still only somebody’s claim about the item. Rachel Daggs shows up on the brig Tribune manifest as “Rachel Dags.” Spelling alone changes what a database returns. A system that cheerfully normalizes a name can make the person disappear.

What is “trust in the loop”?

The older phrase is “human in the loop,” which can mean a person glances at a screen and clicks accept. KBR’s wording asks for more: staff who actively review and fix the output. That’s the difference between a rubber stamp and a check.

KBR has worked this way before. An earlier KBR tool, built in about ten days on a Microsoft app platform, learned from around 30 photographs of title pages and filled in title and author fields. Staff then checked the generated data afterward. The point isn’t speed alone. It’s that a person decides what stays.

What human verification actually requires

For anyone using AI-made records, including genealogists reading library catalogs, I use three checks. These are my own, not KBR’s procedure:

  1. Check the item. Compare the record to the thing itself, or to an image of it. Names, dates and places must match what’s printed or written, letter for letter.
  2. Check the authority. Is this the right person, place or series? Two people can share a name. A machine can pick the wrong one with total confidence.
  3. Check the trail. Can you see where each fact came from, and who approved it? If the record doesn’t say, treat it as a lead, not a source.

Should genealogists trust them?

Trust the speed. Verify the content. If AI helps a library catalogue a backlog in months instead of decades, we all gain, because more records become findable. But a draft is a draft. Cite the original, not the catalogue entry.

Bring me a record you’re unsure about

Every Thursday at 7:00 PM Pacific I work through reader questions on camera at the Reference Desk. Send one about a record, a catalogue entry, or a document that doesn’t add up. You never have to appear on camera. About six questions get answered live each session. Send yours at the Reference Desk.

Sources

Tags:
, , , ,
Print page