Today I met with Lili Daie at her request to share feedback on the AI Metadata Assistant. In a previous post here, I was less than enthusiastic about some aspects of this tool. Thanks to Lili for listening and for a good conversation. I wanted to point out something that should have been obvious to me that wasn’t, and that is, that the AI Metadata Assistant’s tendency to sometimes suggest an LCSH that is a non-preferred term is actually a deliberate design decision. Lili pointed out the assumption that such cases would or should get flipped to the correct term as part of a daily preferred term correction job in Alma, assuming that libraries have this turned on.

Again, I should have understood this, but didn’t, so I wanted to clarify this as a follow up to my previous comments.

As she and I discussed, there is still a ways to go before the AI Metadata Assistant can be relied upon for LCSHs that are more complex, i.e., involve correctly constructing subject strings with subdivisions. The tool is improving but it is not always reliable here. (These are my words.)

Steve

Steve Oberg
Assistant Professor and Team Leader for Acquisitions and Discovery Services
Campus Copyright Liaison
Library | Wheaton College (IL)
+1 (630) 752-5852