When Generative AI Enters the One Department Nobody Called

Generative AI Has an Epistemology Problem. Librarians Don't.

I’ve been consulting with a group of higher education librarians recently, and this work has taught me something revealing about the kind of expertise higher education institutions reach for when a new technology arrives.

Institutional responses to generative AI, for example, tend to look towards IT departments, learning consultants or individual faculty. While understandable, this is not particularly productive, because in doing so generative AI typically becomes either an infrastructure problem or a pedagogical wild west. Or both.

In either case, the institutional imperative reflects deeper assumptions about what kinds of knowledge matter, whose expertise actually counts as expertise, and which professional traditions are considered foundational to the university’s core work rather than peripheral to it.

Generative AI continues to tease out these assumptions with unusual clarity. We think of this technology as software, as a tool, as a teaching challenge, or as something faculty should own because they own the curriculum. But this logic quietly bypasses a more fundamental question: what kind of problem is generative AI, really? Not in terms of its technical architecture, but rather in terms of what it does to the conditions under which knowledge is produced, communicated and evaluated in academic life and beyond.

Librarians, I can tell you, have a great deal to say on this matter.

The Shape of the Problem

Generative AI does something philosophically consequential to information. It decouples form from content, fluency from accuracy, and confidence from accountability. It also produces prose that reads as authoritative regardless of whether it is true or meaningful, and it does so without an author, without methodology, without a review process - indeed without any of the institutional mechanisms that academic culture has developed over centuries to make knowledge claims susceptible to evaluation by peers.

The result is not just a new kind of source - and let’s be clear, generative AI output is not a source in the traditional sense of the word. It is, however, a fundamental challenge to the very frameworks students and faculty use to decide what counts as a source, what counts as empirical evidence, and what critical engagement with information actually requires.

Questions that surround the quality and purpose of information are not new. They are among the oldest questions in epistemology - questions about what knowledge is and how it is brought to life. What is new is the scale and speed at which generative AI makes these questions urgent for ordinary academic work. Ancient philosophical questions now have immediate practical consequences.

Librarians have spent their professional lives navigating exactly the terrain that generative AI now makes treacherous: what does it mean to find, evaluate and use information responsibly when the information environment is unreliable, politically shaped, or structurally opaque? They have continuously rebuilt their practice around this question through every major change in how knowledge is accessed - from physical collections to databases, from the open web to algorithmic curation. That accumulated experience is not peripheral to the AI challenge in higher education. It is among the closest things to a direct preparation for it that any professional group currently possesses.

The Limits of Single Ownership

Yet, the structural response most institutions default to - assigning generative AI to whichever department has the most bandwidth or the loudest voice - produces characteristic and predictable problems. When AI sits with IT, the questions that get asked are the ones IT is trained to ask: licensing, security, access control. Legitimate concerns, but definitionally upstream of the harder questions about learning and knowledge. When AI sits with teaching and learning consultants, the focus shifts to pedagogical integration - also legitimate, but insufficient without an epistemological foundation that can anchor what responsible use actually means. When AI sits with individual faculty, innovation accumulates in pockets while institutional learning accumulates nowhere.

What these failure modes share is not incompetence, but instead a structural problem: each department optimizes for the questions it knows how to ask, and remains blind to the questions it doesn’t. The solution is not to find a better single department. It is to build cross-functional ownership: IT managing infrastructure, faculty contributing subject knowledge, learning consultants supporting pedagogical integration, librarians anchoring the work in information competency, among others.

Of course, this is much harder to organize than assigning a single owner. It requires sustained coordination and tolerance for genuine disagreement about what the real problem is. But those disagreements, handled well, are precisely the point.

The Questions Nobody Is Asking

Under pressure, institutions reach for the most visible and legible forms of expertise, and information competency often does not announce itself the way technical or pedagogical expertise does. Librarians tend not to make a fuss. They have cultivated a service orientation that, while perhaps professionally admirable, has made them easy to overlook precisely when their expertise is most needed.

What gets lost in their absence is something we may take for granted, but which is essential to higher education. Institutions can develop AI policies, update academic integrity guidelines and run faculty workshops without ever seriously engaging the epistemological questions that generative AI raises.

The work may look complete. But the harder questions - questions about what it actually means to know something in an AI-saturated information environment, about how students develop genuine critical judgment when fluency and accuracy have been decoupled, and about what information competency even requires when the concept of a source has become philosophically unstable - these questions tend not to get asked unless someone in the room is trained to ask them.

That is what most institutions are currently leaving on the table. Not a resource, exactly, but rather a way of thinking that only becomes possible when people who disagree about what the problem is are forced to sit with the discomfort of each other's questions.

Now go talk to your colleagues in the library.