This lightning talk will discuss how some preliminary experimentation with Copilot leads to an investigation of the ways it gets things wrong; in this case Ngā Upoko Tukutuku/Māori Subject Headings. I was attempting to get a suitable summary of a recorded interview, Copilot offered to find appropriate subject headings for the summary. It kept finding new and exciting ways to be incorrect. Eventually I did find a way to make it work - but at the cost of more time than purported use of AI would save! That led to some deeper questions of AI and the usage of Māori data.
This falls under Whakairohia te rākau taumatua \ Crafting collaboration and innovation, specifically disproved use cases. I think this will be interesting to the attendees of NDF due the combination of the appearance of Māori data and AI’s inability to correctly use it, being more inclined to give an answer than be correct. This has implications for how libraries and other institutions use AI, especially when creating records using Māori data, and it would be great to connect to other attendees who are attempting similar use cases.
Nik Crombie (Kahugnunu, Rongomaiwahine, Rakaipaaka) is currently a Digital Research Services Coordinator at Te Herenga Waka - Victoria University of Wellington and has worked in libraries for most of her career. She is co-chair of the Local Contexts Library Working Group. Her professional interests include data sovereignty, indigenous metadata, accessibility and digital preservation.