“But Lankes argued this model doesn’t so far fit what’s happening in the library community. The resistance isn’t coming from people who haven’t engaged with the technology. It’s coming disproportionately from people who have. When he conducted focus groups with librarians in the “never AI” camp, he found people who could explain large language models, discuss retrieval-augmented generation, and articulate technically why they considered the tools unreliable. They’ve concluded that a library’s adoption of AI would send the wrong signal about what a library fundamentally is.”

This comes from janefriedman.com/ai-and… and I am not sure if this is particularly true.

In any case, I know this sounds funny coming from me, as I have been advocating that librarians should really understand the technology rather than just parrot buzzwords or just play with tools but I think even if you understand to some level how say Transformer decoder models work, understand self attention, different variants of RAG etc.. It’s still may not be sufficient to know if a tool is fit for purpose without actually trying (and given this is the “never ai” camp, they haven’t tested).

You need both theortical understanding AND experience trying the tools.

Take RAG, understanding basic naive RAG takes 60 seconds, and in today’s environment, this hardly makes you “knowledgable about AI”. At best knowing how RAG works means you know it won’t be a 100% guarantee of source faithfulness. But there are many techniques to reduce the risks, and without trying how would you know what level it is at? A tool that is 99.9% reliable is a lot different from one that is 80%. Theory alone can’t tell you which it is.

In short, theory alone isn’t enough - this knowledge needs to be paired with hands-on experimentation and testing. Run actual searches. Compare results. Break things in controlled ways to understand their failure modes. Only then can you bridge the gap between knowing why a system should work and seeing how it actually behaves with real queries and real collections.

Without theory, you’re reduced to blind trial-and-error with no ability to make connections. Without practice, you’re working from assumptions that may not survive contact with messy reality. Both matters.

The “never AI camp”, I am pretty sure are doing motivated reading anyway and they seem to forget a tool doesn’t need to be 100% perfect to be useful.