Information overload is an old story, and the field already built tools for it. The last Information Shock, at the end of World War II, shifted the scarce resource from information to attention and left the field with information retrieval, literacy training, and decades of practice managing distrust and abundance. If AI only produced more information, or even more information you couldn’t trust, none of that would require a new shock. What’s different is where AI sits. In its conversational mode, it takes the position between a person and what they want to know, the position the document itself used to hold.

Generative AI writes in two different registers. The first produces objects that still behave like documents: essays, articles, images, code that can be shelved, cataloged, and checked against the sources they claim. The second happens inside a chat, where a person and the system build an answer together in real time. That exchange has no stable object behind it. Ask the same question tomorrow and the system may answer differently, since the process generating the answer is stochastic and the training data underneath it keeps shifting. A transcript can be saved, but it records the exchange without being what did the mediating.

AI literacy works well for the first kind of AI authorship, where the output is an artifact you can hold still and interrogate: check the sources, check the claims, check the dates. It has much less to offer in the second mode, where there is no artifact, just an exchange that persuades in real time and looks different for the next person who has it. Controlled studies already show these systems can be more persuasive than a human debater. Literacy assumes a document in front of the learner. When the shock removes the document, teaching people to evaluate a document more carefully doesn’t reach the problem.

It’s already happening, and it doesn’t depend on anything like artificial general intelligence arriving. Roughly half of American adults report using AI chatbots, and a majority say they read the AI-generated summary at the top of a search result rather than the sources beneath it. People ask these systems for medical advice, voting advice, and companionship, and act on the answers without tracing them back to anything. None of that waits on the technology improving. The shock is a fact about how these systems are already used, not a bet on where they’re headed.

Professionals adapt by shifting from managing documents to standing accountable inside the exchange itself. That means building new tools aimed at conversation rather than retrieval, rethinking literacy training so it addresses what a person comes to believe rather than just what a document says, and claiming a professional identity built on being answerable for outcomes, not just on organizing sources. The field has done this before, moving from scribe to librarian to documentalist to information scientist. The task now is to name and build whatever role comes next.

Libraries become the place where the accountable conversant stands: professionals answerable for what a person carries away from an exchange with AI, rather than gatekeepers of documents. As chatbots take over the document’s role in mediation, the librarian’s job shifts from helping people find and evaluate sources toward sitting inside the learning exchange itself, someone with professional standing who can be held to account for whether what a person now believes tracks reality. The collection doesn’t disappear. It becomes the stable ground people and librarians use to check the exchange against, rather than the starting point for it.

While all of commercial generative AI was built with copyrighted  works, much of what trained these systems came from pages, scholarship, and collections that institutions made freely available, on the assumption that openness would widen their readership. The result has massive implications for artist and scholar alike. Commercial AI providers now sell that material back as chatbots and agents, often with the original author stripped out of the answer entirely. Economists have a word for this: enclosure, the privatization of what was a common resource. Museums and libraries face a sharper version of the same problem, since licenses written for human readers didn’t anticipate a reader that ingests an entire collection at once. This is one of the concrete places the shock has already broken something, well before any larger argument about learning or identity.