“The AI-Induced Shock” Keynote Panel Presentation. IFLA World Library and Information Congress. Busan, South Korea (video).
Abstract: Generative AI is not just flooding the world with more content. It is changing where understanding gets made. In a live chat exchange, meaning is built between a person and a machine, and no stable object is left behind to point to, check, or hold accountable. That is what makes this moment a genuine break, not a louder version of the information shock our field absorbed 80 years ago. AI literacy still works on synthetic essays and fabricated citations. It has nothing to grip when the only output of an exchange is what someone now believes. This talk proposes a role for librarians in that gap: the accountable conversant, a credible person building trusted places and answerable for what people carry away.
Video:
Script: The real problem with Generative AI is not lack of training, or business models, or policy. It is uncertainty.
Will AI wipe out half the workforce? Will AI continue as commercial walled gardens needing massive data centers and scars on the environment? Will AI replace white collar workers? Can we choose to reject AI and still be relevant and competitive?
Every day our feeds fill with slop on the one hand and scholarly articles on the other that look at cognition, adoption, and arcane performance metrics. It is exhausting trying to filter reality from hype in a time where internationally we are already overloaded making the same kind of real-versus-not-real call in politics and warfare.
The noise, however, is a symptom, not the problem. There is a reason our world is filled with a thousand voices all calling out the truth and the way. The reason is that we are at the beginning of an Information Shock. A historic event. The last one hit our field 80 years ago, and finding the root of this new one explains the noise, and more importantly, points to a path forward.
An Information Shock is not really about a technology. It is what happens when the way we produce information changes enough that the field organized around the old way has to rebuild itself. The tools change. The people doing the work change what they call themselves. The institutions built around both catch up over decades, sometimes centuries.
Look at where we are sitting today. Eighty years ago the world came out of a global war with more scientific and technical data than any institution knew how to hold. The digital computer arrived at the same moment. The Journal of Documentation was founded in 1945 in direct response to that pressure. Twenty years later, documentalism was no longer the name of the work. It had become information science. The word librarian survived. The word information joined it. The field enlarged around what it could no longer ignore.
It is tempting to file generative AI under the last shock. Just more information, harder to trust, produced at higher volume and faster speed. If that were the whole story, there would be no new shock to speak of. Postwar information science already lives in that world. We built tools for it. We wrote frameworks for it. Overload and contested credibility are the material of the previous shock, not the signature of this one. To see what is actually new we have to look somewhere else. We have to look at what AI does inside the exchange.
Generative AI does two very different things, and only one of them is a real problem for the field.
The first is producing objects that look like documents. Essays. Images. Code. Songs. Books. This is the visible problem, and it is real. Editors are drowning in synthetic manuscripts. Citation practice is degrading in measurable ways. But a synthetic book still behaves like a book. It sits on a shelf. It can be described. Its claims can be checked against the sources it names. Our tools bend around it.
The second mode is where the shock lives. Open a chat window with one of these systems and something different happens. You ask. It answers. You push back. It revises. You float a half-formed idea and it finishes the idea in words you now recognize as your own. Six turns in, you walk away with a conclusion that was not there at turn one. It was built with you, from moves that neither of you made alone. This is not retrieval. This is not search. It is co-authorship of understanding, in real time, without any document doing the mediating.
Now think about what that does to our field. Since the Library of Alexandria we have mediated between people seeking knowledge and the sources that could meet the need. Every tool we built to do it shared one working assumption. There is a stable object in the middle. Something you can point to, describe, return to, and hold accountable. The object does not have to be paper. A data set holds still. So does an image. So does the archived web page. What they all share is that they stay put.
The chat exchange does not. Save the transcript and what you have is a recording of the conversation. Real. Keepable. Not the conversation. Ask the same question tomorrow and the answer moves. The mediation happened inside the exchange, not in any object left behind. This is the shock. Something has stepped into the position the document held, and it does not hold still.
This is why AI literacy, in its current form, does not reach the deep problem.
AI literacy is serious work, and for the first mode of authorship it is the right approach. When a system hands a learner a synthetic essay, or a fabricated citation, or a generated image, the inherited toolkit still works. Check the source. Check the claim. Check the training data. Weigh the bias. This is document evaluation aimed at a new kind of output, and it is exactly what that mode requires.
It has nothing to grip in the second mode. The output is non-deterministic. The training data is not inspectable item by item. The same prompt answers differently tomorrow. And the end result of the exchange is not an artifact you can hold up to the light. It is what the person now believes. By the time literacy training tells the learner to check the output, the persuading has already happened, and it did not happen in a document.
So what is the path forward. It is not to walk away from the exchange. Librarians can and should reject these systems as they stand today. The water-hogging data centers. The training built on uncompensated human labor. The handful of companies extracting value from a global commons and selling it back to us. We have every right to fight for AI that is ethical and sustainable, and every duty to refuse to normalize what is not. But refusing these systems as they are is not the same as leaving the exchange. Our communities are already inside these tools. The underlying business model is already reshaping their lives. They need us there with them.
Nor is the path forward simply to teach around the systems, as though better literacy could compensate for a shift in where learning is now happening. Someone in the room has to be accountable for what a person carries away from an exchange. A chatbot cannot be. It answers to no community. It holds to no prior commitments. It cannot be called back six months later and asked to explain itself, because six months later it is a different model with different weights.
Librarians can be. That is the role I want to put on the table for this panel. The accountable conversant. Someone whose professional standing depends on being answerable for the exchange. A person who can sit with a graduate student, or a patient in a health library, or a citizen trying to understand a piece of legislation, and hear what they now take as settled, and press them to notice where the settling happened. The AI cannot be held to what it said. A librarian can.
The future of collections is less settled than we like to admit. Fifty years of collection-building has produced holdings the profession is proud of. But those collections have already become less important to our communities than the connection and the trust a library has with them. In the shock we are now inside, that direction accelerates.
What becomes more important, and where we should put our professional weight, is credible people who build trusted places. Online, in physical space, or both. Places where a community can come together, build knowledge together, and become empowered to make a difference in a world that is changing on them. A chatbot cannot be that place. It cannot be a community. It cannot be trusted the way a person, standing in a room or on a call with you, can be trusted.
That is where the work is. Being credible people who build trusted places, and being answerable for what happens inside them. That is the theory piece I wanted to bring to this panel, and it lands in a very practical place.
Thank you.