This is the conclusion I’ve come to after witnessing the growing popularity of AI tools. AI is everywhere, despite its propensity to produce misleading or incorrect information, and the moral objections of workers who fear losing their livelihoods. Why do patrons choose AI tools while more reliable (and less controversial) options sit on the shelf? Convenience and ease of use. For better and for worse, AI tools eliminate much of the “work” of doing research. Always ready at a moment’s notice, and capable of answering natural language questions, it’s not difficult to see why patrons would prefer querying AI to developing a complex search, navigating an unfamiliar website (or building!), or risking embarrassment by talking to a real person. It’s doubtful whether these conveniences are worth the misinformation, environmental destruction, and questions of intellectual property we’ve endured as a result of generative AI’s rapid growth. Unfortunately, none of these concerns are enough to curb AI tool usage; convenience and the collective advertising power of Silicon Valley make for an effective counter narrative. Instead, I contend that libraries should try to “speak the language of AI”, emulating what makes these tools easy and convenient while discarding the qualities that make them unreliable and harmful.

How could this be accomplished? The simplest solution may be for libraries to have chatbots of their own. The Dr. Martin Luther King library at San José State University has already developed a generative AI chatbot, KingbotGPT, using what is known as a “retrieval augmented generation model” to produce a genAI chatbot resistant to AI “hallucinations”. SJSU Library’s experiment with AI has had promising results, but questions remain about the bot’s reliability. Interestingly, KingbotGPT began its life as a non-generative AI chatbot and was only upgraded to genAI after users (who by that time were familiar with ChatGPT) began treating it as if it already was. Could libraries benefit from taking the opposite tack, creating more primitive chatbots that make using the library more convenient without the same potential for harm? Librarians frequently note that patrons are reluctant to seek out and learn how to use library resources. If that’s the case then those tools must instead be brought to them. A non-generative AI chatbot, accessible on any page, with a UX design based on those of popular genAI tools, would be capable of directing patrons to library resources (physical and digital) and pages that explain their use. Patrons could have their simple, informational questions answered without having to interrupt their research to find a librarian or dig through a website. What’s more, they could do so with minimal effort, using the same methods they are accustomed to from other AI tools.

For those especially queasy about AI, I believe similar benefits could be achieved without relying on a chatbot at all. For example, using the visual language of the AI tool “button” (for example, the Google Gemini star present in the upper right of every Docs file) to create a resource patrons will notice and be willing to click on. Instead of an AI tool, it could open a panel of links to useful resources, perhaps phrased like the sample queries generated by some genAI tools (such as Proquest’s AI Research Assistant). A tool like this could be said to be “speaking the language of AI” without being AI. However, I would discourage this approach. Returning to the example of KingbotGPT, one of the positive results of the project as reported by the researchers was an improvement in their own AI literacy. I would encourage my fellow librarians to consider some good faith engagement with artificial intelligence, at least as a broad concept. Whether the hype dies down or not, these technologies are not going to disappear as if they never existed. It’s our responsibility as librarians to understand artificial intelligence, at least to the point that we can confidently advise our patrons on when it is and isn’t appropriate to rely on it.

Furthermore, it is our responsibility as information professionals to understand that advertising and hype have collapsed a number of different technologies into the broad category of “AI”. ChatGPT is not Deepmind, and neither of them is AI Dungeon. If librarians aren’t willing to learn the difference, we risk not only rejecting useful innovations out of hand, but also alienating our patrons with our ignorance. If we lose the trust of our patrons, we leave them prey to the misinformation and manipulation of AI hucksters. In this new AI craze I see echoes of the early excitement over cryptocurrencies and NFTs, and I’m troubled when I remember the stories of scams and financial ruin that emerged after that enthusiasm faded. Library patrons have already learned to speak the language of AI, and so have the people who want to sell them something. If librarians want a voice in that conversation, then we have no choice but to become fluent. If I had made the effort to speak the same language as the rest of the team during my internship all those years ago, it’s possible that my system would have been put into practice and my work would not have been wasted. I’d like to think I won’t make the same mistake again.