Libraries Shaping AI Together
When a group of Pottsboro, TX residents sat down at the library and asked ChatGPT, “Where can I send a fax in Pottsboro?” we hoped it would suggest the library. Instead, it told them to drive twelve miles to a mail store in another town.
The question was part of a library workshop where residents tested what AI tools know, and don’t know, about our town. We expected that it might not know everything locals know. We did not expect it to send people to other towns to find services and activities available here. Being a tourist destination, it would be helpful for responses to include local restaurants, hotels, and fishing guides.
That moment was on my mind when I went to Washington, DC, for the AI Civics convening hosted by Data & Society. AI Civics treats AI as part of civic life, something communities have a right to shape, not just a tool to learn. Our cohort is a mix of big city library systems, small town libraries like ours, and the National Digital Inclusion Alliance. Each of us is running our own pilot.
Different Libraries, Same Questions
It turns out a library serving a few hundred people and one serving tens of thousands have a lot in common. We keep running into the same questions. Who gets represented in AI? Who gets left out? Who gets a say? And there is no customer service number you can call to fix what ChatGPT says about your town.
What We’re Doing in Pottsboro
Our pilot asks three simple questions. What does AI know about our rural community? What do our neighbors want it to know, and not know? And how do we make that happen?
We started by trying to stump it. Residents asked ChatGPT local questions and graded the answers like report cards. It did fine on broad history. It got tripped up on the small stuff: shops that had closed, longtime businesses it never mentioned, the wrong address for a local store.
Next came “AI we want” MadLibs. People filled in what bugs them about the tech they already use, what they would want AI to help Pottsboro with, what it should never do, and who should decide what gets saved or deleted. One person commented “It annoys me when it tells me how I should get my car fixed. I just want the facts to make my own decision.”
Then we talked about three ways to shape AI: feed better local information to the big platforms, build our own tools, or help set the rules.
Libraries have Valuable Community Information
Libraries are not powerless in this. We hold exactly what AI companies want for their training data: local history, community knowledge, and decades of carefully organized information. That gives us a real say in what gets shared, on what terms, and who benefits.
What’s Next
In DC, we looked for places where our projects overlap, worked on each other’s drafts, and talked through next steps. The goal is a complete draft toolkit by early next year that any library, big or small, can use with its own community.
The gathering showed me library staff across the country are tackling new and difficult problems for their communities. I want to do the same for Pottsboro and help people here thrive. I left feeling hopeful.
If your library is asking the same questions, you can follow the work at datasociety.net/ai-civics.
Tags: ai, AIandpubliclibraries
