Scholarly Communication and Open Initiatives Librarian San Diego State University
OpenAI learning models like ChatGPT and GPT-4 have exploded onto the learning and library scenes challenging expectations for what our future work will look like. If Artificial Intelligence (AI) can produce workable programming code such as Python, can it be taught to produce structured metadata in formats like MARC 21 or Dublin Core? If so, it has the potential to help libraries improve discovery of open access resources which often have poor metadata quality due to a limited cataloging workforce.
With this issue in mind San Diego State University Library researchers experimented with open access resources available to OpenAI learning models to see if they could be taught to produce good quality structured metadata. What standards do they do well? When they produce an error, can updates to the query fix the issue for all subsequent queries? Additionally, how well do OpenAI learning models handle applying subject headings?
This poster will share the results and discuss the findings as well as thoughts on the impact learning models might have for the cataloging profession.