How Library Science Principles Power Community AI Literacy
Public libraries are dealing with budget cuts, book bans, and the usual existential
questions about whether anyone needs physical collections when they can google
everything. And now AI shows up, and everyone assumes libraries are about to get
steamrolled by it too.
But watch what’s happening at reference desks across the country.
Patrons ask how to determine whether an IRS email is real or a phishing attempt
generated by AI. Small business owners are trying to figure out which AI tool won’t
steal their marketing copy. Retired community members want to use AI to parse
dense municipal documents but do not know where to start. A parent is concerned
about their teenager using AI for homework and whether that’s cheating.
These aren’t edge cases. This is what reference work looks like now.
Much has been written about how libraries can use AI tools internally—chatbots
for reference, algorithms for cataloging, and predictive analytics for collection
development. But there’s another question we need to answer: How do libraries
help communities navigate AI?
And here’s what I realized while building AI governance frameworks: these are
not computer science problems. They’re library science problems—questions about
provenance, access, evaluation, and trust. The exact questions libraries have been
answering for over a century, just dressed up in a new syntax.
The skills to navigate this moment aren’t locked in some Stanford AI lab. They’re
already in the library’s professional toolkit. The field needs to recognize them and
use them.
Why This Is Actually Your Domain
AI governance is framed as a legal or engineering challenge, something that
requires technical expertise libraries supposedly don’t have. But strip away the
jargon and here’s what AI governance actually requires:
Information architecture.
Libraries have spent generations developing classification systems that turn chaos into navigable collections. The AI landscape right now is complete chaos; thousands of models and tools are released every week with no coherent organization. Categorizing these tools, evaluating their utility, and creating functional taxonomies aren’t departures from cataloging. They are the next iteration of cataloging. Moving from the Dewey Decimal System to the Model Registry, from call numbers to capability tags. Same discipline, new substrate. Authority and provenance. Librarians teach patrons to evaluate sources based on authorship, publication venue, and citation trails. Generative AI deliberately obscures this lineage; it presents synthesized information as disembodied truth with no visible source chain. When a patron asks, “Can I trust this AI’s medical advice?” they’re asking the same authority question that libraries have answered about publishers, journals, and websites for decades. Reestablishing provenance in an AI context means understanding the origins of training data, model lineage, corporate ownership structures, and the economic incentives shaping outputs. This isn’t new. It’s the fight for bibliographic integrity in a probabilistic age.
Preservation and the human record. As the web floods with synthetic text, images, and video, the organic “human web” cultural production that has defined the internet era faces extinction through dilution. Model collapse, where AI systems trained on AI-generated content degrade over time, isn’t a technical curiosity. It’s an archival crisis. Libraries need to preserve the human-generated culture before it becomes indistinguishable from the synthetic. Future historians will need access to what people actually created, wrote, and thought before everything became algorithmically mediated—that is, special collections methodology applied to the algorithmic age.
Equitable access and literacy. The digital divide is evolving into an AI access gap. Access to frontier models and sophisticated AI tools is increasingly gated behind subscription tiers. If the library was the great equalizer for books and then for internet access, it has to become the equalizer for AI access. But access without literacy is insufficient. “AI literacy” isn’t a new domain; it’s critical information literacy applied to algorithmic systems. The reference interview now includes teaching people how to ask questions that get helpful answers from AI systems. The evaluation of sources has become the evaluation of model outputs. The skills transfer directly. The mission remains constant. This isn’t about libraries “adding AI” to their portfolio. This is about asserting that information stewardship belongs to libraries, regardless of whether information comes from card catalogs, databases, or neural networks.
What This Could Look Like in Practice
Let me propose three interconnected approaches that translate these principles into actual service. Not massive tech initiatives that require Silicon Valley partnerships but extensions of what libraries already do well. These strategies scale; a small
branch can start with monthly AI office hours and simple documentation templates, whereas an extensivesystem can build comprehensive programs.
1. The Registry: Reference Work, One Layer Deeper
The most immediate chaos patrons face is tool proliferation. They don’t know whether to use ChatGPT, Claude, Perplexity, or a specialized legal AI for their needs. They don’t know which tools train on their inputs, which respect privacy, or which produce outputs that are reliably accurate.This is a reference problem.What if libraries treated AI tools the way they treat reference materials: as resources that require vetting, contextualization, and active recommendations based on patron needs? Just as you wouldn’t shelve a medical text without evaluating the publisher’scredibility, you wouldn’t recommend an AI tool without examining its privacy policy, data retention practices, and documented failure modes. A Registry would be an actively curated,
patron-facing directory of “library-verified” AI tools, organized by functional category—creative writing,coding support, civic information navigation, health information triage—and annotated with standardized information
What It Looks Like in Practice
Each tool listing answers key questions: Does this system train on your inputs? Is it free, freemium, or paywalled? What are its known limitations and hallucination patterns? What’s the company’s track record on privacy? What tasks is this tool appropriate for, and what tasks should it not be used for? When a small business owner asks which AI tool could help with marketing copy, the librarian doesn’t just say, “Try ChatGPT.” They consult the Registry, assess technical comfort level and privacy concerns, and recommend a specific tool with context: “For your use case, short-form marketing content that you’ll edit before publication, I recommend Tool X because it doesn’t retain your prompts for training, offers a free tier sufficient for your volume, and hasmshown reliable performance for business writing.mHowever, you should avoid using it for anything containing customer data, and here’s why . . .”
That’s reference work. It just operates one layer deeper into the information stack.
Starting Small
You don’t need to evaluate every AI tool on the market. Start with the five most commonly asked about in your community. Create a simple one-page guide for each. Collaborate with other library systems to share the evaluation work so one library tests the privacy policies, another examines accessibility features, and a third documents standard failure modes. Build the Registry as a cooperative project. The Registry positions libraries as the trusted intermediary in an ecosystem designed to eliminate intermediaries. Not building AI tools but contextualizing, evaluating, and routing. Exactly what libraries have always done.
2. Community AI Literacy Labs
Forget the standalone “Introduction to AI” class that three people attend. Instead, embed AI literacy into the services you’re already running.
Job seeker programs: Your workforce development Your workforce development programming already teaches résumé writing and interview skills. Now add a session on how to use AI tools to draft a résumé without losing authentic voice, how to prepare for interviews where employers use AI screening, and how to evaluate which tasks to delegate to AI and which require human judgment.Small business support: If you run small business workshops, add a component on AI tools for entrepreneurs. Not theoretical, practical. Which free tools can help with bookkeeping? How do you use AI for customer service triage while protecting customer data? When should you absolutely not automate?
Civic engagement: Run a workshop series teaching residents how to use AI to parse dense municipal documents, research zoning regulations, and draft public comment letters. The populations most at risk of being left behind by the AI access gap—older adults, non-native English speakers, and people without reliable internet—are precisely the people libraries already serve.
Teen programs: Your teen advisory board wants to use AI for homework. Don’t just say yes or no; teach critical use. When is AI a research assistant versus a shortcut that undermines learning? How do you cite AI-generated content? How do you fact-check AI outputs? Turn the controversy into a literacy opportunity.
Integration Approach
Don’t isolate AI as an exotic technical offering. Normalize it. By integrating these tools into standard library services, libraries model a relationship with AI that is empowered rather than passive. You’re demonstrating that AI is something you use critically and strategically, not something that happens to you. The library becomes the space where the community learns to live alongside algorithmic systems without being consumed by them.
3. Internal Integration
Practicing What We Teach
AI isn’t a separate department. It’s the infrastructure that will eventually touch every library function. What if libraries moved from blanket prohibition to structured experimentation? Create “sandbox” environments where staff can test AI tools for administrative tasks, with clear guidelines on data privacy and output verification.
Staff Applications
A teen services librarian could use an AI assistant to generate initial program marketing copy, then edit for voice and accuracy, reducing administrative burden and freeing time for direct patron interaction. A cataloging team could experiment with AI-assisted subject heading generation for backlog materials, treating the AI output as a first draft requiring professional verification. Grant writers could use AI to draft initial proposals and then refine with institutional knowledge and relationship context. These aren’t wholesale replacements. They’re workflow enhancements that let professional expertise focus on tasks that require human judgment.
Policy Framework
Establish clear internal guidelines: Which tasks are appropriate for AI assistance? What data can never be entered into external AI systems (patron information, personnel records, confidential correspondence)? What does the verification process look like before any AI-generated content goes public? Document your experimentation. What works? What fails spectacularly? What unexpected problems emerge? Share these findings with other library systems. Build the professional knowledge base together.
Why This Matters
If librarians haven’t used these tools themselves,they can’t effectively teach patrons to use them. You can’t curate a Registry of AI tools you’ve never tested. You can’t run a workshop on strategic questioning if you’ve never written a prompt. Internal integration isn’t just about efficiency; it’s about building the professional competency to serve as the community’s AI literacy experts. When staff encounter the limitations, biases, and failure modes of these systems firsthand, they become better educators. They move from abstract warnings about “AI risks” to concrete, practical
guidance: “I tried using AI to write that policy summary, and it completely missed the nuance in Section 3; here’s what to watch for.”
What This Means for the Field
For Library Workers
This shift to AI represents a reframing of professional identity, not an event of obsolescence. Routine information retrieval may be increasingly automated, but the higher-order skills—contextualization, ethical judgment, and empathetic service—are
becoming premium assets. Librarians aren’t being replaced. They’re being promoted.
The “human in the loop” that AI governance frameworks require isn’t a generic moderator. It’s a trained information professional. It’s librarians.
For Library Advocacy
When city councils ask, “Why do we need libraries when we have Google, Wikipedia, and now ChatGPT?” the answer becomes specific: because you need a civic institution that ensures algorithmic systems are accessible to people experiencing poverty, comprehensible to older adults, and vetted for the public good. You need a safe harbor that’s not extracting data, not optimizing for engagement, and not making editorial decisions based on shareholder returns. Libraries aren’t supplemental in the AI age.
They’re the civic infrastructure that makes algorithmic governance compatible with democratic values.
For the Democratic Mission
If information becomes entirely synthetic, manipulated by private interests, and unverifiable, the informed citizenry required for democratic governance
collapses. The library must become the verification layer for democracy, the institution where algorithmic outputs are stress-tested, where provenance is traceable, and where access doesn’t depend on purchasing power. This isn’t hyperbole. It’s the
logical extension of the library’s founding purpose.
What We Still Don’t Know
The field lacks shared standards for evaluating AI tools. There’s no equivalent to MARC records or controlled vocabularies for the AI landscape. Most vendors won’t disclose training data sources or data retention policies in plain language. We haven’t agreed on how to archive AI-augmented community memory or what future researchers will need to understand our AI-saturated moment. These aren’t problems any single library can
solve. They require collective professional work, precisely the kind of standardization and methodology development libraries have always done. The American Library Association, the Public Library Association, and state library associations need to
prioritize building shared frameworks, as they did with cataloging standards, intellectual freedom guidelines, and privacy policies. We’re early in this transition. The answers aren’t all clear yet. But the questions are unmistakably library questions.
Claiming the Role
Libraries can’t wait for Silicon Valley to invite them to the governance table. They won’t. Tech companies will move fast and break things, and libraries will be left to repair the informational wreckage. But libraries have something tech platforms
don’t: community trust that tech platforms have squandered, a physical footprint that digital-first services lack, and a methodological framework that computer science doesn’t teach. What libraries need now is the institutional will to recognize that the governance of knowledge, artificial or otherwise, is, and always has been, the
work of the library.
Start small. Build the Registry with five tools. Run one AI literacy workshop integrated into existing programming. Let staff experiment with one administrative use case under clear guidelines. Document what you learn. Share it with other systems. Build the professional knowledge base cooperatively. You don’t need a dedicated AI department or a six-figure tech budget. You need to recognize that the work you’re already doing—evaluation, contextualization, access, literacy—is precisely the work
this moment requires. This isn’t about asking permission to be relevant. It’s about recognizing that the work libraries already do is exactly what communities need to navigate AI systems responsibly, critically, and equitably. The skills are there. The mission is clear. Themoment is now. PL
- FURTHER READING
Project Information Literacy, Provocation Series – Essential reading on how information literacy pedagogy must evolve toward critical engagement
with algorithmic systems. - NIST AI Risk Management Framework (NISTAI RMF 1.0) – Voluntary guidance that maps directly to library values, particularly regarding trustworthiness and accountability.
- Algorithms of Oppression by Safiya Noble (NYU Press, 2018) – Critical examination of bias in search algorithms with direct implications for how libraries should approach AI tool evaluation.
- Free professional development resources –Anthropic Academy offers several free courses relevant to library contexts, including “AI Fluency: Framework & Foundations” (general staff development), “Teaching AI Fluency” (for developing patron programming), and “AI Fluency for Educators” (adaptable to library instruction). Although built around Claude, the frameworks transfer to understanding AI systems broadly.Available at anthropic.com/learn.
