Plain English with Derek Thompson
What’s the Point of College?
Derek Thompson talks with Harvard College Dean David Deming about how AI is forcing a rethink of what college is even for, and why Deming's controversial "AI encouragement" policy is really an attempt to build guardrails around a technology students are already using constantly.
Key takeaways
- Deming frames American higher education in three eras: an early era built around scarce access to books and information (Harvard's founding gift was 400 books), a second era built around specialized faculty expertise as knowledge proliferated, and now a possible third era where AI commodifies expertise itself, pushing colleges back toward their original purpose as communities of learners rather than repositories of information.
- His "AI encouragement" email wasn't a blanket endorsement of AI use—it was a "barbell strategy": either enforce a real ban in controlled settings, or if a ban can't be enforced, acknowledge that students will use AI anyway and build structured guardrails around it rather than punishing honest students who avoid it while others cheat undetected.
- Concretely, Deming plans to have students write initial drafts by hand or on locked-down devices in class, then use AI in supervised stages for feedback and revision, with oral presentations required so he can verify the ideas are genuinely theirs; an approach modeled partly on dissertation defenses, which he argues remain a powerful anti-cheating tool because students who used AI without understanding the material simply can't defend it under questioning.
- He rejects "AI as MOOC 2.0" comparisons: MOOCs failed because they only replicated lectures, missing the personalized tutoring and communal accountability that actual learning requires; AI can offer personalization but still can't replace the social "commitment device" of learning alongside other people who hold you accountable.
- Campus loneliness is real and rising (Atlantic reporting cited shows time spent partying dropped from 32% to 21% of seniors (1994 to 2026) while reported isolation has more than doubled) but Deming says Harvard is testing fixes like phone-free social spaces, more faculty presence in dining halls, and student-led (not administrator-led) events, and he's cautiously optimistic this year's freshman class is more socially engaged.
- In a separate paper using Norwegian administrative data on roughly 9,000 workers, Deming found meetings consume about one hour per average workday (two hours for office workers) and cost firms roughly one-seventh of total labor spending; yet companies with more meetings tend to pay higher wages, and workers who spend more time in meetings see faster wage growth, suggesting meetings are a necessary coordination cost of specialization rather than pure waste.
Notable moments
- Deming's origin story: Harvard's founding gift was 400 books, worth two-thirds of an elite Cambridge college's library, making the university literally a repository of scarce knowledge.
- On students feeling trapped by AI: "It's like social media... I know it's bad for me, but everyone else is doing it and I feel trapped."
- Deming's plan for his AI-history class: blue-book rough drafts in controlled conditions, then supervised AI-assisted revisions and oral check-ins to confirm authorship.
- The strongest counterargument he's heard: that learning should happen "for the love of the game," and colleges should just be so engaging students don't want to cheat; something he says he wishes were true but doesn't trust as sufficient in practice.
- His summary of the meetings paper's finding: "meetings are the broccoli of work, widely disliked, but probably good for us anyway."
- Thompson's aside on journalism training as an analogy for AI-avoidant colleges: preparing a journalist while banning video and podcasting is preparing them "for a labor force that only existed 15 years ago."
- Deming's closing advice, drawn from his own theater background: take acting and improv classes, because presentation and persuasion skills may become more valuable, not less, as AI absorbs more expert knowledge work.
Time saved: 56 minutes.