Some things to consider when deciding whether to start building with "AI" in libraries and archives

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Notes from a panel talk on whether libraries and archives should start building with "AI." Rather than weighing what the tools can do, it works through six areas of concern: bias in training data, intellectual property, verifiability, the impact on creative work, environmental sustainability, and security and privacy. It closes each one with a concrete tactic for practitioners. Informed by Science and Technology Studies and by Dan McQuillan's Resisting AI, the underlying argument is that the ascendancy of AI is not inevitable: these are decisions that people will make, and what the cultural heritage practitioners decide to do will shape how these technologies get deployed. The PDF is supplemented by a web archive of the original page where the notes were published.