The (Artificial) Intelligence of University Education

Most of the conversation about AI in universities is defensive. Students can generate an essay in seconds, so faculty redesign assignments, buy detection software and write policies. Those are reasonable responses to a real problem. They are also far too small.

The essay was never the point. It was a proxy for something harder to see: a student wrestling with ideas, testing them against other minds and arriving somewhere new. AI has broken the proxy. It has also given us the means to build, at scale, the thing the proxy stood for. We can be clever about this. This is our chance.

A seminar across the centuries

Imagine an economics student preparing a question on inflation. Instead of reading five summaries, she convenes a seminar. Adam Smith, David Ricardo, John Maynard Keynes and Friedrich Hayek each take a seat, as agents built from their own writings. She asks how they explain the last three years of prices. They disagree, as they did in life. Smith reaches for the division of labor. Keynes asks about demand. Hayek warns about what any central planner can know. She presses each one, asks them to answer one another and notices where their arguments run out.

Now imagine the same room in every field. A philosophy student puts a question about artificial minds to Aristotle, Descartes and Ada Lovelace. A history student asks Ibn Khaldun and Edward Gibbon why empires decline. A biology student brings Darwin and Mendel together over a modern genome. A literature student asks Virginia Woolf and James Baldwin to read the same contemporary novel.

These are not costumes or chatbots with famous names. Each agent is an agentic representation grounded in a defined body of work, curated by faculty and the library, citing the passages its arguments rest on. It is clearly labeled as an interpretation, not the person. Its value is not that it speaks for Keynes. Its value is that it lets a student argue with Keynes’s ideas at full strength, in real time, and see exactly where those ideas come from.

Discourse, not regurgitation

The difference between this and today’s AI tools is the direction of the work. A student who asks a chatbot for an answer receives a summary and stops thinking. A student who moderates a disagreement between great minds has to think harder than before. She chooses the question, judges the arguments, catches the contradictions and decides what she believes. The machine does not replace her learning. It extends the room in which her learning happens.

Done well, these seminars do what the best university teaching has always done, and what only a few students have ever had: sustained, demanding conversation with the strongest thinkers in a field. The tutorial system at a handful of universities offered something like it to a small number of students. Agentic discourse can offer it to every student, in every subject, at any hour.

Synthesis across the whole university

Universities organize knowledge into departments, and departments rarely talk to one another. Yet the most important ideas often sit between them. Evolution and economics. Ethics and engineering. Poetry and physics.

When every department builds its own agents and corpora, the university becomes a network of ideas that can be connected on purpose. A seminar can bring an economist, a biologist and a philosopher to the same question about scarcity. A student in architecture can test a design against the ideas of urban historians and public health scholars in the same afternoon. Faculty from different departments can design seminars together, and students can follow an idea across the whole of the liberal arts.

The same network can connect universities to one another. A small college can bring its students into conversation with corpora curated by a research university on another continent. Departments that once competed for attention can build shared spaces for exploration.

Exploring what is and what is not

Every subject contains its settled knowledge and an enormous space of what might be: counterfactuals, thought experiments, unanswered questions, ideas never tested. Universities have always taught the first and gestured toward the second.

AI lets students explore that second space directly. What if the printing press had arrived two centuries later? How does a classical physicist respond to an experimental result that contradicts her theory? What does a policy look like when the strongest objections from each tradition are answered in turn? Students can build these scenarios, run them, argue with them and, most importantly, learn to tell the difference between what is known, what is plausible and what is invented.

That last skill matters more than any other. A generation that grows up with fluent machines must learn to ask, constantly: what is the evidence, and where does it come from? A seminar where every claim is grounded in cited sources, and every invention is labeled as one, teaches that habit by practice.

Sorcerers of their domain

The goal is a graduate who commands the ideas of a field the way a master commands a craft: able to summon the right thinkers to a problem, combine their insights, see where they break and add something of her own. Call it sorcery of domain knowledge. It is not memorization, and it is not dependence on a machine. It is fluency, judgment and imagination, built through thousands of hours of real discourse.

Such a graduate is valuable in any profession, precisely because AI is everywhere. Anyone can generate text. Very few can direct a debate among ideas toward a decision worth making.

How to build it

Rearchitecting a university around discourse needs more than software. It needs a design.

• The library as the source of truth. Every agent rests on a curated corpus. Librarians and faculty decide what each agent may draw on, how sources are cited and how disputed texts are handled.
• Faculty as designers of discourse. Professors spend less time delivering content and more time designing seminars, choosing the voices in the room, setting the questions and coaching students through the argument.
• Assessment of thinking. Students are assessed on how they moderate, question and synthesize: transcripts they annotate, syntheses they defend orally, positions they change and explain. The record of a seminar becomes evidence of learning.
• Sovereign infrastructure. Student conversations, unpublished research and licensed texts are sensitive. A university can run this capability on infrastructure it governs, so its data and its students’ work stay inside its walls.
• Clear labels everywhere. Every agent states what it is, what it is built from and where its knowledge ends. No agent claims to be a person.

Guardrails

The design must take its risks seriously. Representing historical thinkers means choosing interpretations, and those choices carry bias. The canon has excluded many voices, and a seminar that only seats the familiar names repeats that exclusion; every department should widen the room deliberately. Students can lean on agents instead of thinking, so the format must demand their judgment at every step. Access must be equal, so the richest experience does not belong only to the richest institutions. And people must stay at the center: students learning from faculty, from one another and from the friction of real disagreement.

This is our chance

Universities have changed shape before, after the printing press, after the research laboratory, after the internet. Each time, the institutions that thrived were the ones that used the new technology to extend learning rather than defend old formats against it.

AI is that moment again. We can use it to produce more summaries and more detection software. Or we can use it to give every student a seat in the greatest seminar ever convened, across every subject the university can imagine.

We can be clever. This is our chance.