What changes for libraries and librarians with AI?

The shift

Answering a question by finding, retrieving, and synthesizing the relevant sources — the classic reference-desk transaction — goes from scarce (a trained intermediary's time and knowledge of where things live) to abundant and instant. But the same abundance that answers any question in seconds also floods the information environment with plausible, unattributed, confidently-wrong content, which makes the opposite function — someone accountable vouching for what's real and where it came from — scarcer and more valuable, not less.

The axioms

• Finding the right information is hard, so a trained intermediary who knows where things live and how to search is valuable — scarce because search skill and knowledge of the collection took training to acquire.
• Answering a patron's factual or research question requires an expert who can navigate sources — scarce because synthesis across sources used to require a person's hours.
• Organizing a collection so it can be found — cataloguing, classification, metadata — requires skilled human labor per item — scarce because describing and classifying each work took trained judgment and time.
• A library's collection is trustworthy because a professional selected, vetted, and stands behind each item — scarce because vetting sources for authority and accuracy is expensive judgment.
• Teaching people to find and evaluate information (information literacy) is a service the library provides — valuable because the average person can't reliably tell a good source from a bad one.
• The library is a physical place people go to access materials, computers, and staff — scarce because the materials and infrastructure existed in one location.
• The library provides equitable access to information for people who can't otherwise afford it — scarce because paid resources (databases, books, expertise) were gated by cost.
• The library is an institution whose name certifies that what it holds is authentic and where it came from is documented — scarce because provenance and preservation require an accountable custodian over time.

Invalid axioms

1. The reference intermediary who knows how to find and synthesize an answer is the scarce, valuable one. Finding, retrieving, and synthesizing across sources is now free and instant for anyone with a chatbot. The classic reference-desk transaction — "I have a question, help me find the answer" — is the exact shape of work AI does at near-zero cost. The habit-trap: libraries still staff and physically anchor a reference desk as a core service, and library-science training still centers search-and-retrieval skill, as if knowing how to look things up were still the scarce good.
2. Cataloguing and metadata require skilled human labor per item. Generating a catalogue record, subject headings, classification, and descriptive metadata from a work's contents is now cheap and fast to draft. The habit-trap: technical-services departments are still staffed and budgeted as if describing each item were the bottleneck, when the remaining scarce act is checking the machine-generated record against ground truth, not producing it from scratch.
3. Basic research assistance — "help me get started on this topic" — is a professional service worth a person's time. Producing a competent starting map of a topic, a reading list, or a literature overview is now abundant. The habit-trap: research-support services are still framed around the librarian doing the initial legwork, when a patron can get a plausible version of that legwork instantly and now needs help with the thing that came after — verifying it.

Unchanged axioms

1. Someone accountable must vouch for whether a source is real, authoritative, and correctly attributed. A model produces fluent citations, some of which don't exist, and plausible claims, some of which are false — and it can't be answerable for either. A librarian who curates and stands behind a collection, or authenticates a source's provenance, is providing exactly the ground-truth verification and accountability the model can't. This gets more valuable as unattributed AI content floods the environment, not less.
2. Teaching people to verify — to tell a real source from a fabricated one — is scarcer-value work now, not obsolete. The classic information-literacy skill (evaluate authority, spot bias) was already scarce; the flood of plausible synthetic content makes the ability to check a claim against its actual source the scarce competency of the era. This is a rising-value reframing rather than a survival: the demand went up because the pollution went up. Calibration flag: how well ordinary people can self-serve this via better-cited, more reliable models is moving fast — if models become dependably grounded and transparent about sources, the teaching need thins even as the institutional vouching need holds.
3. Preservation, provenance, and authentication of the record require an accountable custodian over time. Establishing that a document is authentic, unaltered, and traceable to its origin — and keeping it that way for decades — is trust-and-accountability work, not synthesis. No amount of abundant text generation substitutes for a named institution certifying "this is the real thing, and here is where it came from." As provenance itself becomes contested, this rises in value.
4. The library as physical civic space rests on physical presence, not information access. The building serves people who need a safe, free, staffed place — for community, for shelter from the elements, for in-person help, for access to hardware and human assistance. That function was never really about the information being only there; it's about the physical and social good of the place, which token-generation doesn't touch.
5. Equitable access is a distribution-and-trust problem, and gating didn't disappear — it moved. The library's role of providing free access to what others pay for still holds, because the best AI tools, databases, and reliable sources are increasingly gated behind subscriptions and cost. The scarcity shifted from "access to books" toward "access to good tools and vetted sources," but the underlying equity function — someone providing free, trusted access to the have-nots — is intact.

New axioms

1. Librarians as guides through AI-polluted information, and teachers of verification, is an emerging role nobody has standardized. When answers are free but a meaningful fraction are confidently wrong or fabricated, the scarce service is helping people verify — and the profession has no settled curriculum, credential, or service model for "verification guide" the way it had one for "reference." It's being improvised library by library.
2. The library's role as a trusted-source and proof-of-provenance institution is rising, but the practices for it don't exist yet. As synthetic content makes "is this real and where did it come from?" a routine problem, an accountable institution that authenticates provenance becomes valuable — but standards, tooling, and staffing for provenance-at-scale (verifying and attributing AI-era material) haven't been built.
3. Equitable access when AI is gated is a new front, not the old one. If the reliable, well-grounded AI tools cost money and the free ones are the polluted ones, the access gap re-forms around quality of tool and trustworthiness of source rather than access to a book. Whether and how libraries provide free access to trustworthy AI — and who pays — is unsettled.
4. Cataloguing and reference staff are being freed up faster than their role is being redefined. The work that justified much of the headcount is now cheap to draft, but the higher-value work (verification, provenance, literacy teaching) isn't yet defined as a job with training and metrics. The transition risks being read as "libraries need fewer people" rather than "the same people should be doing the scarcer, harder-to-fake work."

Where it breaks

Libraries are still anchoring service around the reference desk and search-and-retrieval skill (INVALID axiom 1) at the exact moment the scarce, rising-value work is verification and vouching for what's real (NEW problems 1 and 2) — the desk everyone still staffs answers the question AI already answers for free, while the question that actually needs a human ("is this answer real, and where did it come from?") has no desk, no curriculum, and no owner yet.

A second collision: the profession's freed-up capacity from automated cataloguing and reference (INVALID axioms 1–2) is arriving as a budget-cut argument, while the equity gap is quietly re-forming around access to trustworthy AI rather than access to books (NEW problem 3) — so headcount may be cut on the old rationale precisely as a new, harder access problem appears that would need those same people to solve.

ai-axioms.vercel.app/axioms/education/librarians-ai-axioms