Harvard's Dean: Universities No Longer Sell Expertise—They Sell Character and Community
Universities once charged for knowledge and expert status. Now that AI can replace expert functionality itself, what keeps students are the character and collaboration skills only community can teach.
The video won't play here. Listen to the audio instead:
The argument · timestamps estimated from transcript position
Universities shifted from stewarding collections to building community
Deming notes that Harvard's founding gift consisted of 400 books donated by John Harvard—roughly two-thirds of Emmanuel College Cambridge's collection at the time, yet sufficient to establish the university. The first era of universities was the age of scarce knowledge collections; when knowledge itself became abundant, universities reorganized around disciplinary expertise. Now that AI can package expert knowledge into accessible products, the scarce resource universities possess is character and habit—things that can only be learned within a community.
— David DemingYou can't fully trust research from AI labs
Deming argues universities should remain research centers for two reasons. First, research at cutting-edge AI labs is influenced by commercial interests: companies are unlikely to publish results that damage their bottom line, making it hard to fully trust such work's source and filtering. Second, teaching and research are naturally complementary—explaining a concept to someone who knows less is the best test of whether you truly understand it. This complementarity is partly why top talent chooses university positions over full-time lab work.
— David DemingUniversities face a shrinking pipeline, not an AI crisis
Deming acknowledges severe pressure on the university business model but argues the root cause is declining high school graduation rates, not AI. Small private colleges in the Northeast will be eliminated and the sector will consolidate further, but public universities may actually strengthen. Four-year university enrollment for 18-25-year-olds has actually risen slightly in recent years. What is genuinely volatile is two-year community college enrollment—which moves counter-cyclically, falling when jobs are plentiful and rising when the economy weakens. Unemployment has remained consistently low.
— David DemingFaculty lean left by structural sorting, not by conspiracy
Mounk raises a structural issue deeper than ‘campus political correctness’: fifty years ago, education level barely correlated with voting patterns—the educated even leaned slightly Republican. Today, in virtually every developed democracy, education correlates strongly with leftward politics. The more educated, the more left-leaning. This means the faculty population and the hiring pool themselves are structurally left-leaning, separate from whether individual professors bring politics into the classroom—and far harder to correct.
— Yascha MounkTenure doesn't change people; by then it's too late
Deming argues that tenure's purpose is to be ‘truly used’—to take research risks or repay the academic community. He himself spent ten years in administrative service after earning tenure. Mounk counters: getting tenure now averages fifteen to twenty years, during which people are trained that ‘publishing another paper matters more than daring to speak.’ By the time they actually possess that security, they've likely already lost the instinct to use it.
Tests either lock AI out completely or design it in
This semester Deming is pursuing a barbell strategy: on one end, completely block AI by reverting to pen-and-paper blue book exams with no electronic devices. Harvard's Ec10 (Introduction to Economics) has already lowered the weight of term papers and increased midterms from one to three. On the other end, accept that students will use AI when writing long essays. Rather than pretend to enforce a ban that punishes honest students, redesign assignments to incorporate AI into the workflow itself.
— David DemingNear-perfect homework masks hollowed-out learning
Brown University's economics course experienced what looked like a ‘cheating scandal’: the professor hadn't explicitly ruled out AI, and suspiciously polished work arrived, with many scores clustering near perfect. When he hastily administered a no-AI final exam, many high-scoring students performed poorly. Deming calls this not cheating but mass ‘learning shortcuts’—students saw someone else's correct answer and mistook comprehension of it for their own understanding. Their grasp was fundamentally brittle.
— David DemingBlue books can't measure the work of revision and refinement
Mounk agrees with the barbell strategy's direction but identifies a gap: writing ability isn't producing a passable essay in three hours—it's the process of repeated revision, demolition, and rewriting. That's what blue book exams cannot test. His solution: separate them. Some courses retain pure handwritten closed-book exams; others explicitly permit students to freely use AI on the final paper.
— Yascha MounkIn their own words · checked verbatim
Therefore, we should return to that really quite old idea of a university, or a collegium —a medieval learning community of teachers and learners, people learning together in community.
David Deming1:00
I think being a dean is one of the worst jobs in the world. For one, as I’m sure you’ve noticed, structurally everybody hates you.
Yascha Mounk35:00
I said, without thinking, “I left my dream job to take this job.”
David Deming36:30
what I’ll be experimenting with in my own class this spring is what I would call, and others have called, a barbell strategy, which is basically all or nothing.
David Deming41:30
The main thing, I think, is we cannot create a situation where we say on our syllabus, “20-page paper due in two months, bring it back to me when it’s done, I ban AI, and the problem is solved from my end.”
David Deming43:00
I’m always tempted—I never quite say this to students—but I’m always tempted to say to a student, if they say “I’m just bad at writing,” to answer, “well, in that case, you’re just bad at thinking.”
Yascha Mounk45:00
It’s like, this is the worst they’re ever going to be. Whatever you think it can do now, it’s going to be able to do more in a year.
David Deming48:30
So it’s not really cheating, it’s more like compliance—maybe “cheating scandal” isn’t the right term. It’s not really cheating, it’s actually just showing how people are using AI to substitute for their own effort.
David Deming49:30
Figures
| Harvard's founding gift | 400 books | 1:00 |
| Harvard's founding collection as share of Emmanuel College Cambridge's holdings | approximately two-thirds | 1:00 |
| Ec10 midterm exams | increased from 1 to 3 | 42:00 |
| Years since Deming earned tenure | 10 | 33:00 |
| Brown University AI-assisted homework scores | near 100% | 49:00 |
Glossary
- Collegium
- A medieval teaching community where students and teachers lived together to learn collectively, which Deming identifies as universities' original organizational model.
- Barbell strategy
- A polarized approach to AI assessment: either completely prohibit AI through closed-book exams, or openly integrate AI into assignments.
- Blue book exam
- The traditional American university closed-book handwritten exam without electronic devices, often used to prevent AI-assisted writing.
- Tenure track
- The long evaluation and promotion pathway from doctoral degree to earning permanent faculty status.
- Intellectual vitality
- Harvard's metric for assessing whether the campus enables free expression of disagreement and rational debate.
How to listen
University administrators, teachers, and anyone concerned with how AI is reshaping higher education's business model and teaching assessment.
The middle section on tenure and the dean's career choice (around 29-37 minutes) can be skipped.