David Deming is the Danoff Dean of Harvard College and an economist. He writes a Substack called Forked Lightning and hosts a podcast called The Context Window.
In this week’s conversation, Yascha Mounk and David Deming discuss why AI may force universities into a new era focused on community and virtue rather than knowledge or expertise, whether the business model of American higher education is facing existential or merely serious pressure, and how professors and students should rethink learning and assignments in an age when AI can do much of the traditional academic work.
This transcript has been condensed and lightly edited for clarity.
Yascha Mounk: Tell me a little bit about what you mean by universities entering the third era of higher education because of artificial intelligence. How is AI going to transform Harvard University, and how’s it going to transform the higher education sector more broadly?
David Deming: Well, first of all, it’s a prediction—nobody knows the future. What I’m doing, Yascha, is looking at the trends in front of us and looking at how, in the past, technology and progress have changed what universities really do. I wouldn’t say “only do,” in the sense that we’ve always done a lot of things, but what’s really scarce, what you can only get at a university, and therefore what is kind of the thing that we should really lean into. My thesis was that in the first era, universities were keepers of scarce knowledge.
This is something I said in a convocation speech I gave to Harvard freshmen: the founding gift of this university was by John Harvard, our namesake, and it was a gift of four hundred books. That doesn’t seem like a lot, but it was about two-thirds the size of a Manuel College library in Cambridge, which was a very established elite British university at the time. So this was really a founding gift that put Harvard on the map. You can think of lots of examples in our history where, for example, when there was a great fire in the 1700s that destroyed a lot of books in the library, the Harvard Corporation—the governing body of the university—told students, “Go home. Without our books, what are we?” Of course that changed over time. There were more libraries. It used to be that Harvard was the location of the only printing press in the colonies, in President Henry Dunster’s house. That changed over time too, as knowledge proliferated.
We became organized around expertise rather than knowledge. Knowledge is just words written on a page. The expert curates knowledge and helps people understand how to interpret it. That, to me, is the conception of the modern university. The great universities like Harvard have organized themselves around a collection of expertise in different disciplines, and that has produced huge benefits for society, and I think will continue to. It’s just that with AI now, expertise will become relatively more commodified. That doesn’t mean we don’t have value—it just means that you can actually access the knowledge of Yascha Mounk or David Deming through ChatGPT, or Claude, which has basically downloaded all of human history and synthesized it in a way that gives people access to relatively personalized expertise.
So the question is: what can we still do as a university? I think the answer, as I said in my speech, is that we are still a learning community. There are lots of skills and habits of mind and virtues that you can really only learn in community with other people. You cannot learn them from behind a computer screen. 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. Learning virtues like justice, integrity, and fairness is best done in community with others. I think that’s what we should double down on. Not that experts aren’t going to be important, but that this is a thing that only we can do now, in the age of AI.
Mounk: Let me disaggregate what artificial intelligence might do to universities across three dimensions. The first is that one of the key reasons why universities exist is to undertake research. We needed the kind of resources that universities had, and frankly, the strange marriage of convenience between teaching and research, in order to sustain the salaries of people who were able to do this kind of research. It now looks as though artificial intelligence is going to be increasingly capable of doing a lot of this research.
That is a positive thing, because we actually care about that research. It matters whether you have better or less good economic research in order to set countries on the path towards development, or in order to better manage a potential financial crisis, or whatever it is. It would be very strange to give up on the benefits of AI because we somehow only want humans to do this work, if we think that better medical research can save human lives. But it might raise questions about whether universities are needed in the same kind of way.
The second point is about the financial basis of the university. Is going to university still going to give the same financial benefits to people’s earnings over their lives? Is there still going to be demand for highly cognitively capable human beings in the economy? And if that’s not the case, then who’s going to pay all of the money you need from your MBA students, your LLM students, and your BA students, in order to subsist?
Then the third kind of challenge is that a lot of assignments I would have traditionally done when I was an undergrad, or that I would have traditionally assigned when I was a teaching fellow—as it’s called, not a teaching assistant, at Harvard—when I was doing my PhD there, can now very easily be done by AI. So how do we actually make sure that people are still learning? I’m sure there are other ways to cut the space, but that’s how I think about it. Let’s start with the first one and work our way down. How should we think about this in the context of research, and whether we still need universities as the employers and facilitators of this kind of research?
Deming: This wasn’t a topic of my speech, because it was to undergraduates and was focused on undergraduate life, but I absolutely think that universities should and will remain centers for research. I think that’s for two reasons.
One is a lot of the research that is being funded—let’s just take the case of AI labs, for example. A lot of the research being funded by frontier AI labs is valuable research, but the provenance is unclear. These companies have a financial interest in the research looking a certain way, so you can’t always trust what’s coming out. Not so much that anything is wrong in it—it’s just that there’s a question of whether OpenAI or Anthropic or Google DeepMind or any other lab would publish results that affected their company’s bottom line. You can’t fully understand how to interpret those results, because they’re selected on things that wouldn’t harm the company, for example. That’s just the nature of it. I think also much of that research is very applied—it’s like, how are people using AI to do X? But there’s much more basic, fundamental research that universities do, because the purpose of the university is to produce knowledge, and not for downstream commercial implications or any other implications. There’s basic work that leads to downstream discoveries that really best happens at a place like a university. I think one of the great human inventions is the university, and I very much hope that it will remain a center for research.
I think the other thing is—you mentioned this strange combination of research and teaching. I do think there’s a strong complementarity between those two. You can’t really understand a concept well until you can explain it to someone who knows less than you. I think the reason why top talent in all fields, including in computer science—we’ve seen some access to labs, but still a lot of people will come back and are coming back—the reason people want to be in universities is because teaching keeps you fresh, it keeps you engaged with the next generation. Many students here at Harvard are much more AI-native than me, and if I want to keep up with the latest developments, it’s actually helpful to be in the classroom in front of them and see how they react to things. So I do think there’s an intrinsic complementarity between teaching and research, and I personally would go long on the research function of the university in the age of AI. I don’t think that’ll go away. I think if anything, we have challenges like, can we get enough compute to do frontier research? I think there’s some kinds of research that may best be done in industry, but there’s also certain kinds of research that is best done in universities. I don’t actually see that changing. Who knows—but I don’t see it changing in the age of AI.
Mounk: Let’s speak to the second part. How concerned are you that the whole economic foundation of these universities may start to crumble?
One direction from which that could be the case is that university degrees just don’t come to be valued in the same way in the marketplace, in part because people who are able to demonstrate high cognitive skill may no longer be valued in the same way in the marketplace. The second point is that a lot of your money comes from students whose parents are able to pay for an undergraduate education, and if some of them lose their jobs, they may be less able to pay the tuition fees of their children. In something like a business school, a lot of the tuition comes from the fact that you go off to McKinsey or BCG for a couple of years, and then they pay for you to go to business school, with an expectation that you go back to the firm. Perhaps those firms no longer feel that investment is worthwhile in the same way, if they’re increasingly relying on AI tools. And finally, there may just be an expectation that the degree is not going to help you find a job, so why would you take out a loan if it’s not going to give you a job? Finally, if all of that is the case, and a lot of public finances come from payroll taxes and income taxes—and Harvard gets a lot of its money from the federal government, as we learned during some of the attacks from the administration—then that’s yet another source of financial instability. So is the period in which American universities were uniquely rich, in historical terms and in international terms, potentially coming to an end?
Deming: That’s a pretty heavy question, and I don’t really know the answer to it. But here’s what I feel reasonably confident about. I think there are serious threats to the business model of the university, but I’m not sure that those are all related to AI. I think a lot of them are related to demographics. In American universities, the cohort size of high school graduates is starting to shrink, and that’ll put pressure on the number of universities. I think that’s different from putting pressure on the basic model of whether you go to university at all. I think many universities, especially small private colleges in the Northeast where we have more population decline, will go out of business, and you’ll see more consolidation.
We hope you’re enjoying the podcast! If you’re a paying subscriber, you can set up the premium feed on your favorite podcast app at writing.yaschamounk.com/listen. This will give you ad-free access to the full conversation, plus all full episodes and bonus episodes we have in the works! If you aren’t, you can set up the free, limited version of the feed—or, better still, support the podcast by becoming a subscriber today!
If you have any questions or issues setting up the full podcast feed on a third-party app, please email leonora.barclay@persuasion.community
I think also on the research side, some of the research happening in science, engineering, and computer science requires tremendous scale, so it’s hard for many universities to sustain that. You’ll probably see some consolidation—a little bit of the rich getting richer in the university space—and a lot of the pressures facing universities will go down the selectivity distribution, so to speak. I don’t think that’ll really hurt the public universities; if anything, those might go slightly stronger. It’ll be small private colleges that are kind of barely hanging on anyway.
I don’t think that means people won’t go to college anymore, because actually, if you look at the share of, let’s say, 18-to-25-year-olds enrolled in a traditional four-year college over time, it’s actually slightly increased over this entire period. We have this meme that college is over, but it’s not really matching people’s decisions. All of the fluctuation you see in college enrollment over the last decade is in two-year colleges, and that’s because enrollment in community colleges is highly counter-cyclical. When the job market’s really good, people stay out of college, and they go back when there’s a recession. We’ve had historically low unemployment. So I think it’s a little bit of a vibe about the end of college more than the reality. But I think you’re right that there are serious pressures on what we do, and we’re going to have to adapt—you see a lot of belt-tightening of various types, and I think that’s probably a good thing.
So I would say there are pressures on the business of university education, but I wouldn’t say they’re existential pressures. I’d say they’re serious, but kind of normal pressures that certain industries come under, and hopefully it’ll lead to some innovation—there are things we can do better, a lot of things we can do better. Especially undergraduate education—you asked about master’s degree programs, and I think those are under a lot more pressure, because they provide a product that’s much more connected to the labor market, and you can see companies doing things to replace that. But what we do in undergraduate education is really intellectual and moral formation, and there’s just not a lot of places to send an eighteen-year-old that can do that kind of formation. Another way to say it: we have a bundle of things we do, and you can find elements outside of college that do one or two of those things, but it’s hard to find something that does all of them.
So my modest prediction is that universities will survive. We will change, we’ll adapt, and we’ll probably lose some among us. But I’d go long on the university—I’m a believer in it, in what it does. This is not because I think there are no problems; there are actually a lot of problems. But I don’t think college is over.
Mounk: Let’s speak about this piece of intellectual and moral formation, because I think you’re right that that is the core proposition of universities. Of course, the ability of universities to engage in that intellectual and moral formation has been very contested over the last few years. This is partially to do with ideological factors—a sense that a lot of people have that universities are ideological monoliths, and that therefore you send your children off to campus and they’re going to come back with very different views and ideas from the ones they grew up with, which can of course also be a good thing. The whole point of a university certainly is to broaden horizons and open minds, and if nobody changed their mind about the world in the process, that would be a very worrying thing. That was true twenty or forty years ago as well, when most Americans trusted higher education. It may have changed in nature now that most Americans no longer seem to trust higher education, according to Pew and other kinds of polls.
At the same time, there are these questions around intellectual formation—a lot of how intellectual formation happened, and what that intellectual formation consisted of, may shift in the age of AI. So how did you address how universities should change what they do in response to this, in your speech? And then let’s go into more detail about some of those points.
Deming: First of all, on the issue of ideological balance, I think this is a serious problem for colleges all across the country to address. It’s a mix of perception and reality. I think it’s possible, and it is true, that faculty—not just at Harvard, but at almost every university—lean left. However, that does not necessarily mean that politics invades the classroom. I’d distinguish between those two things. The reality is: even if most faculty vote Democrat, there are relatively few faculty who bring their politics overtly into the classroom. That is somewhat rare, and it is usually not a good thing for student learning. When I teach, I never talk about my politics, and I take great pride in having students guess about what I actually think, because my point is to help them think about different issues. I always tell them, even if you’re an advocate, you need to understand the other side in order to make arguments that persuade.
So I think it’s possible to get a good education that teaches you how to think critically and to be ideologically open-minded, even if you’re being taught by mostly Republicans or mostly Democrats. I’d distinguish between those two things—I think they’re both issues, but the issue of people bringing their politics into a classroom where it’s not appropriate. There are some classes where it is, but if you’re teaching organic chemistry and making comments about a certain administration, then that’s just not really relevant. I think that’s pretty rare, actually, on college campuses, but even if it’s rare, it should be even rarer.
Mounk: A couple of words on this point. One is that I don’t think this is exclusively a problem of Harvard University or Johns Hopkins, where I teach. I don’t think it’s exclusively a problem of the United States—I think you have very similar problems in other countries in the Western Hemisphere, in Europe, and elsewhere. I actually think this is going to be a very fundamental problem for universities, but it’s going to be very hard to fix, for reasons that go a little deeper than people generally tend to think.
Fifty years ago, knowing the level of somebody’s education didn’t provide a very strong predictor of who they voted for—if anything, it probably meant they were slightly more likely to vote for Republicans than for Democrats in the United States. But there was always also a kind of hyper-educated elite that leaned a little to the left. The average educated person in America was an upper-middle-class person, and therefore more likely to vote for Republicans, who were generally the party of lower taxes and so on. But there was always the idea that the Harvard professor is a kind of somewhat left-leaning radical, and so it balanced out in a certain kind of way.
We now just live in a period in which, for whatever deep reasons—which I think nobody has fully understood and described rightly—there’s just a huge educational polarization in every developed democracy I can think of, such that the more educated you are, the more you are on the left. As a result, most professors, and most people in the pool from which professors are recruited, are going to be very left-leaning. A lot of the students who tend to come from highly educated households are also going to be left-leaning. So you end up with these institutions that are meant to represent society in some kind of way, to be in conversation with society, but they’re just way to the left of where society is.
So even before we get to all kinds of man-made, homemade problems—which I think also exist, where professors absolutely will not hire somebody known for conservative political points of view—even before you get to that, these structural reasons are driving a huge wedge between universities and the putative social elites that they’re educating and the rest of the population.
Deming: You’re saying the deck is stacked against progress in this area. I think that’s definitely true—everything you say is true about this inherent difficulty in having any kind of balance on the faculty. Here’s what I’d say too: existing alongside that is the reality that we survey our students every year and ask them questions about a series of things we’d call intellectual vitality or civil discourse. We ask them questions like, do you feel comfortable having a conversation with somebody you disagree with? Do you feel like the classroom is a space for you to express your political views, even if they’re in the minority? We’ve seen serious, positive progress on those questions over the last two to three to four years here at Harvard, even though our faculty hasn’t changed its ideology at all, to a first approximation, over that period.
I think they’re both issues. I’m just saying it’s possible to make progress on whether people feel like they can actually get an education that’s ideologically not even balanced, just neutral. A lot of the time, politics doesn’t invade everything—a lot of the time you just want to learn about things. I think that’s possible even under the structural conditions you describe, which are going to take a lot longer to fix. We still need to fix them, but they’re not going to be fixed overnight.
Mounk: Let me speak to the other point you made. I agree with you that certainly my median colleague at Johns Hopkins, and certainly the median professor who taught me when I was a grad student at Harvard, was left-leaning, but certainly did not impose their views on people in the classroom. I’m somebody who has this podcast and writes a lot publicly, so my students don’t have a huge amount of trouble figuring out where I stand politically. I think that makes it incumbent upon me in the classroom to emphasize and model that I will judge them by the quality of the argument, not on whether they say something I happen to agree with. I always explicitly tell them I would much rather they make an interesting argument I disagree with than parrot my views back to me poorly—I have no interest in that. I think students do come to trust me on that over time, but it’s very common that students tell me, thank you, I appreciate that about your class—in such-and-such class, I did not feel like that at all, and we all know you just have to echo back the views of the professor. Again, I don’t think that’s the typical case, but it’s something that deeply marks the students who are aware of those kinds of situations.
Is there something that you, in your position of authority, can do to address those kinds of situations, understanding that on the other end of that there’s a very strong value of academic freedom? You have to balance two different values here. On the one hand, I think this is an actual abuse of institutional power—the university gives me the right to grade my students, and I will defend the right to do that as I see fit, because that’s a core part of my academic competence, my academic freedom. But if I do that in order to induce my students to parrot political views they don’t believe, on pain of getting a B instead of an A, I think I’m absolutely abusing my authority. So is there something the institution can do to ensure that happens more rarely than it now does?
Deming: Great question. I think the answer is yes, but for all the reasons you suggested—which I can expound on—it’s important to do it in a way that is consistent with our values of academic freedom, and also to take a slightly patient, longer view, because many of the short-term solutions that are out there, I think, don’t work that well.
If I were to go to faculty and say, “you have to stop doing this”—we have an important principle in the university, which is that faculty are more or less sovereign in the classroom, within certain boundaries. I think that’s actually really key to progress. A lot of innovation in research and teaching comes from the autonomy we give faculty who are on the frontier of their disciplines—we give them wide leeway to teach in the way that makes sense to them. So I think that’s a really important principle, which means you have to be willing to accept that, as an administrator, you don’t have a heavy hand with the way faculty teach. As a faculty member myself, I think that’s really important. For people listening to this who are faculty or administrators—once you’re a member of the faculty and you have tenure, you’re an employee for life. We’re all partners in the firm, and we take turns in roles like this. But I think of myself primarily as a faculty member, and I ask myself, would I want someone to do that to me? The answer is usually no.
Having said that, I think you can have a lot of influence on the margins through a persuasion campaign, through hearts and minds, through modeling the way you want our community to engage with each other. It’s not a heavy-handed “hey, you’re doing something wrong, do it better”—it’s more like talking with students. I do a lot of talking with students about these values—again, we’d use terms like intellectual vitality, civil discourse. It really is a skill-building exercise. You are coming here to be challenged. My colleague Tarek Masoud loves to say—and I think he’s right—think about college like a mental gym. You don’t go lift the easy weights; you lift the hardest weights you can, because you’re trying to make yourself uncomfortable and stretch your capabilities. That is what a great university education is supposed to do.
The more I not only say that but model it—I hold these monthly, on-the-record open conversations with students, where the Harvard Crimson comes and asks me questions, and then students ask me totally different questions, and everything is unscripted. I’ve had students come and challenge me on all kinds of things—divestment from Israel, about the Trump presidency, about whether Harvard’s going to do this or that. Sometimes I can’t answer the questions, or I don’t feel comfortable answering them, or I don’t give a great answer, but I try. I do a lot of stuff like that to show people that this is a developmental environment and you can challenge authority, and that’s totally normal, and we’re all playing our parts in this.
I also share with faculty as colleagues what some things that have worked are, and build communities around people who care about these things. We have a group of faculty called the Council on Academic Freedom here at Harvard that advocates for related changes and for us to think about them. I think all of those things, as part of an ecosystem, are good. I can play a role in saying, I’m good with being challenged, I’m good with the idea that I’m not always right. That’s the level at which I typically engage—partly for practical reasons, because I don’t want to be telling faculty what to do in the classroom, but also because I actually think that’s more effective in the long run, and it points the way forward by modeling behavior. When I teach in the classroom, I try to be this way too.
Mounk: To speak to another tension between what the faculty might wish and what might allow the university to actually live up to what it claims to do—a lot of the function of intellectual, and particularly moral, formation has been given up by faculty and delegated to a whole set of staff. Many of the staff are great at their jobs, but I do think that when you’re an incoming student at a top university in the United States today, a lot of the norm-setting and expectation-setting is not done by faculty members—it’s done by administrators at every level of the university. I think there are also some problems about the ideological composition of those administrators—statistics suggest they’re actually much more homogeneously left-leaning than faculty, and in particular more likely to be on the illiberal left rather than the more liberal left.
But there’s also really a problem here relating back to the beginning of the conversation with what the sales pitch is. The idea is that you come here, and we have these exceptional faculty members, and they’re going to engage with you as a human being and help to form and shape you. In reality, what’s happening is that your main point of contact, certainly outside of a lecture and outside the classroom, is an army of administrators—of which there are now, in some places, more than students. Now, I’m sure many faculty members might agree with this point in the abstract. But if you then come to them and say, well, this actually means you now have to run this orientation and be present at this event and so on, they’re going to say, “screw you.” So is there a way in which the university actually can, and has to, reform itself to live up to that promise in a different way?
Deming: You put your finger on it exactly—there are two sides to looking at this. One is that the growth of administration has been faster than the growth of faculty and of students, and there are too many administrators. The other way of looking at it is: why does that exist? I do think part of the reason it exists is because faculty would rather delegate these tasks to somebody else. I count myself in this—faculty want to do research and want to teach. We don’t really want to do a lot of the other things, and maybe we aren’t even particularly good at it either.
So there’s a trade-off there: if you delegate spheres of university life that are important to students’ experience to administrators, you delegate that control, and you get the benefit of spending more time on sabbatical writing books, or whatever you’re doing. I’ve experienced that myself—so I’m guessing it’s not a criticism, as you know, but I think what I would say is that I would almost always welcome more faculty engagement with the work I’m trying to do here at the college. I’d say that some faculty are deeply engaged in it, and many are deeply disengaged. I’d say to my colleagues that anybody who has a criticism should—if you accept your role as a tenured faculty member as a partner of the firm, you should not only criticize, you should engage and try to fix the problem, because that’s kind of the only way it gets fixed. If you look around and see a problem, and nobody else is solving it, you’re elected.
I would love a bargain for all universities where we say, okay, we’re actually going to spend more of the marginal dollar on the faculty-student experience—more faculty, smaller classes, whatever it is we need to do to deepen those connections—and less of the marginal dollar on administration, in exchange for greater faculty involvement in student life. I think that would be a good trade for the university, and I would love to make it. I say this as somebody who’s inhabited both roles—I got tenure ten years ago, and I’ve basically been doing some form of, not always a dean role, but directing centers or chairing committees. I do these things not because I like them, actually—a lot of it feels like a tax sometimes, because I love doing research and teaching, that’s what I got into it for. But I also think it’s important: if you’re part of an institution that you believe in, and it has been good to you personally, you serve when you’re asked to serve. I think a lot of faculty feel that way too, but sometimes I wish more did.
Mounk: Some of that is about creating a culture and norm-setting as well. I was an undergrad in Cambridge, England, which at the time still really emphasized one-on-one teaching, in which administrators were barely part of the experience, and you really had this very close contact with faculty members. It’s one of the things that attracted me to academia. When I came through my PhD at your institution, from the very beginning, the culture signaled very clearly: this is not how we get ahead in the world. Yes, of course you want to be a competent teacher—yes, of course you should teach a couple of terms while you’re a grad student, because you need to have on your CV that you’ve done some teaching. But don’t overinvest in teaching. Any term that you can get a fellowship—of which, frankly, there are many available—take the fellowship, so you can focus on research rather than teaching. And if you want to be a tutor in one of the houses, really think twice about that, because you’re going to be dealing with whatever random personal problems the students have, and that’s really going to be a distraction from getting your PhD done.
So I think there’s also a sense in which—we talked a lot, when I was a graduate student, about professionalizing, and the role of the university in professionalizing grad students to become academics. A lot of the core of that professionalization was teaching people that the marginal extra publication in some mid-tier journal is worth a lot more than a slightly higher teaching evaluation, or spending more time in your pedagogical function at the university.
Deming: I do my best to protect graduate students and junior faculty from service requirements that are not going to help them get tenure and make their way in the profession. For people who are just getting started, particularly people on the tenure track, it is important to do your research, establish your scholarly credibility. I guess the question is, once you get tenure, you should really flip that switch, because now you have won the lottery of life. If you have tenure at a great university like Johns Hopkins or Harvard, you have won the lottery of life—you have a forty-year career, God willing, of guaranteed employment. So my view is you have an obligation to give back. Some faculty, not most, but some, still act like junior faculty after they’ve gotten tenure, and are mostly about their own research program, when actually the point is to pay it back by protecting the other people in the same way that you yourself were protected at an earlier time.
Mounk: I guess this is a sense in which tenure is the worst possible system except for all the others. All things considered, I’m a defender of tenure—when it comes to questions around academic freedom and free speech, I very much value the protections of tenure.
But I also think that nowadays, the average graduate student takes six, seven, eight years to go through a program. Most of the time nowadays you then become a postdoc for a number of years. The tenure track has extended longer and longer, for all kinds of reasons. So by the time you actually get tenure, you’ve been in the profession, let’s say, fifteen to twenty years on average. If you have this incentive—that I get through all of this process, and I’m not going to challenge the orthodoxies in my field, and I’m not going to say anything that might be unpopular, because I can’t possibly jeopardize this grand life project of getting tenure—by the time you actually get tenure, you’re not going to change. You’ve been so socialized into one way of doing things that the newfound freedom, in most cases, goes unexercised. I worry that the same is true if, for fifteen to twenty years, you learn that the thing that makes you valued, the thing that is important, the thing that is at the core of your professional identity, is an extra marginal journal publication—flipping that switch, in the way you’re talking about, is going to be very difficult.
Deming: As a description of what actually happens, you may be right, often, most of the time. However, I do think if you get tenure and you keep doing the same thing you were doing before, you’re doing it wrong, because the point of getting it is to then use it to do something you wouldn’t have otherwise—that’s the whole reason to give it. I think it means different things for different people. It could be taking on a risky research agenda, making a bet that could fail, but you still have your guaranteed employment. If it fails, it could mean giving back to your community. But I think, having been given this incredible opportunity, you ought to do something interesting and valuable with it. I had always understood that once I got tenure, I was supposed to contribute in some way, and I’ve tried to do that. I think I’m not the only one—I think most faculty think this way too, but those who don’t should, in my opinion.
Mounk: I think a lot of faculty think that way. Whether a lot of faculty act that way is a different question—I might ask that about myself as well.
Deming: I could say one more thing, Yascha, actually. I know we’re going to zoom out, but I have to say, for somebody listening to this who’s a faculty member—I think you have to understand that the Venn diagram of people who can get tenure at a top research university, and people who can run anything or manage anything, is pretty non-overlapping. You don’t actually have to be that good at management to be effective in a role like this. So I would just say to people, give it a try—not because it’s going to be awesome, but because you’re actually going to give back, and you’ll feel fulfilled by that. That’s my pitch.
Mounk: Well, let me make the counter-pitch. I have a longstanding spiel about deans, and I hope you won’t take it personally, David, which is that I think being a tenured faculty member is one of the great jobs in the world. It’s the closest you can get to being a leisured aristocrat. I don’t think most academics are at leisure—they actually work very hard, and I work very hard. But you have the financial resources and the freedom to think about whatever you think is important and interesting in the world, and that is an immense luxury.
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. Faculty members, they might be left-leaning and progressive, but more than anything else, they’re anti-authoritarian in instinct and outlook. They hate the university president, they hate their dean—some more than others, but that’s the general tendency. And then it’s a managerial role, but you don’t have any power—you can’t actually fire people or hire people. You have influence, of course, and you make certain important decisions, but you have way less power than you would as a regional manager of Trader Joe’s. So I’m very, very happy to be an academic with job protections. I would much rather be a regional manager at Trader Joe’s than be in your role. Is that unfair?
Deming: No, for your preferences, it sounds totally reasonable. To be honest with you, people ask me—when I first took this job, I was fielding questions, we were doing this new resident tutor training, and somebody said, “Was this your dream job? And if so, what did you do to get it?” I said, without thinking, “I left my dream job to take this job.” My dream job was a tenured professor at a place like Harvard. That’s what I—but the reason to do it is not because it’s going to be better than what you’re doing. The reason to do it is if you care. In my case, I took this job a week after we announced that we were suing the Trump administration. So if somebody says to you, again, if you believe in a place and what it does, and it’s been good to you personally, you could be asked to serve, and you serve. Think about it not as something that’s going to be awesome, but as a service responsibility. I am sustained by the impact I’m having, not by the consumption value of doing it every day.
That’s the reason to do it, and to not do it forever, by the way. My view is people should take turns—I don’t think it’s good to have a dean of the college for twenty years. I don’t intend to stay in this job that long, or anywhere close to that long. I think it’s good to have a rotation, and to have people with that service mindset and the mindset of a faculty member. I plan to return to teaching and research. I think that’s important for the governance of the university, to have people who identify as faculty in positions of authority, because then we get the type of university that is how faculty would like it to be governed. Somebody’s got to do it. That’s my counter-pitch—again, it’s not about personal utility. It’s about service.
Mounk: It’s the ideal of ruling and being ruled in turn. I’m very happy in this realm of life to be ruled—I have no aspiration to rule. Let’s get into the heart of some of the questions and debates about AI and learning.
Obviously, part of the core problem is that universities have this role as validators in the world. One of the reasons why a degree from Harvard is valuable is that if you have a high GPA and so on, that’s supposed to indicate something about your intellect, about what you’ve learned, about your ability to jump through a lot of hoops, and that’s valued in the marketplace. The way we traditionally assessed that was, in the humanities and to some degree the social sciences, by setting you a few days to go write an essay, and in the natural sciences, math, and so on, by asking you to do a problem set. Nowadays, Claude and ChatGPT and the like can do that for you very, very well. The deeper problem is that we, in a slightly more idealistic description of what universities were up to, also thought that learning to write well and learning to work through these kinds of problem sets were really valuable intellectual skills. But there’s a real question about the extent to which that remains the case when this machine can now do it for you.
So, before we get to how we need to change assessments and all of those things—what do you think students still have to learn? Do we still have to learn to do math and sciences in the same way? Do we still have to learn to write an essay in our own voice, if, in theory, Claude and ChatGPT could do it for them?
Deming: Basically, yes, they do. But I think that what, in the fullness of time, we think of as mastery of those concepts ought to be at a higher level. What it means to be good at writing will change. But I think the core skill of being able to formulate an argument, express yourself, think critically, interrogate different positions—I can’t imagine that no longer being valuable. In fact, if anything, I think it’ll be more valuable. I just think maybe we’ll have higher standards in terms of how you’re able to embody it in some kind of performance, whether it’s an oral exam or a presentation or whatever. I think we’ll expect people to have done much deeper research on the claims they’re making, because they have access to this tool.
So, again, it’ll be very disruptive, but I think ultimately it will raise our standards in a very helpful way. And I don’t think AI will replace us as experts, especially in spaces that are inherently contested and contestable. You write and think a lot about politics—I think it’s going to be a golden age for politics. That’s because politics, unlike, let’s say, programming, is not a verifiable domain where there’s one widely accepted correct answer. The code compiles or it doesn’t. Politics is not actually a domain where you’re debating issues with a right answer—the answer is an outcome of people’s beliefs, their values, and power structures. I just don’t think AI is a technocratic solution for what is ultimately often a political process.
So I think the people who will benefit are the ones who know how to use the tool to become experts in something, and then, having become experts, will use the skills they’ve gained outside of that expertise to persuade others of their point of view, to bring people along, to help people make meaning in ways that speak to their individual situations. So I think persuasion is likely to become much more important in the age of AI. I’m pretty bullish on the university’s ability to be a unique place where people can do that, but it does require some adaptation. That’s what I’ve been focused on, what I’ve been writing about, what I’ve been thinking about.
Mounk: Very briefly, in terms of assessment and so on, what do you think that adaptation would look like?
Deming: I want to be humble about this, I don’t have all the answers, we’re all figuring this out as we go along—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. You can have environments where, if you think assessment of mastery is very important for your class, you want to make it fully AI-proof. That could be a return to blue book exams under controlled conditions. If you’re teaching a big lecture class like Ec 10, intro to economics, the Ec 10 professors, Jason Furman and David Laibson here at Harvard, have basically downweighted take-home problem sets and increased the number of midterms from one to three, under controlled conditions. That’s an attempt to basically get AI out of the process—that’s one end of the barbell.
I think the other end of the barbell is: if you either have assignments that are too meaty to fit within a three-hour window, or you actually want to ask something deeper and longer-term of students, you just have to accept that AI is going to be part of the process and design around it. That doesn’t mean, “let AI do the writing or the thinking for you”—absolutely not. It’s more like: understand that students are going to be at the ChatGPT window or the Claude window as part of their process of reasoning through an idea, and try to design around that reality.
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.” Students are sitting there feeling like they have a choice: either behave with integrity, which is the right thing to do, and don’t use AI, or do the wrong thing and use it without telling anybody. Either way, if you do the right thing, you feel bad about it and disadvantaged, and if you do the wrong thing, you’re rewarded for it. So we just have to accept the fact that everybody is using AI for everything, all the time, and design around it. I think about it as trying to build a set of guardrails around education, to acknowledge that this technology exists and try to use it for good rather than ill. That’s going to mean different things in different classes.
Mounk: Speaking to the inherent positionality—to use a word beloved by our colleagues in the humanities—of being a dean, your suggestion that perhaps we could use AI, especially with some of those longer papers, has attracted a lot of controversy. I’m going to defend you on this and then attack you on something totally different.
I was struck that this is very similar to something I’ve been arguing within Johns Hopkins, and something I did in my classes over the last year or so. I think universities have to do two things. The first is to ensure that students still learn how to write. I think there’s no substitute for writing in terms of figuring out your thoughts. 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.” Because a lot of the time when we speak with each other, I can say, “my view is this, my opinion is this,” and jump over the necessary logical steps, which is very easy to do in oral communication. It’s the moment you sit down and try to write down your argument that you realize there’s a gap between this point and that point, and it becomes hard to write because you really have to refashion your argument.
I grew up with a very poor writing education in Germany, where I was a lazy high school student, and I realized if I wrote sufficiently convolutedly that my teachers didn’t really understand what I was saying, they’d give me an A. I was very lucky to have decent writing instruction as an undergrad, which didn’t just make me a better writer—I think it made me a much, much more careful thinker. So I absolutely think we need to preserve that as a core part of instruction.
At the same time, I also think we need to teach students how to use AI tools, which are obviously going to play a huge role in their lives and professional lives. I myself use AI tools for all kinds of things—not to write prose, but for all kinds of other things, in research, creating graphics, and other things. By the way, that might be great—instead of getting fifteen or twenty seminar papers at the end of a term that are of pretty good quality, perhaps students can actually go and do something even more ambitious, even more interesting with those tools. That would be more fun for me as a faculty member too.
So I did exactly what you advocated for, in terms of thinking this should be at the level of the university education—I don’t think it needs to necessarily be the case in each course. Since I think we haven’t yet gotten there in a systematic way, over the course of an education, where you have some writing-intensive courses that are non-AI and other contexts where you’re really using AI in a guaranteed way, I now assign, for the first time in my life, blue book exams, where students have to demonstrate mastery at writing out an argument about some subject over the course of three hours, or two hours, or whatever it is. And I’ve allowed some students, in some contexts, to use AI however they wish for their final seminar papers. I think that’s worked pretty well. So I’m going to defend you on that—how do we make sure there’s still some substantial writing component over the course of an undergraduate education, in two ways? First, how do we make sure there’s actually some context where there’s going to be blue book exams, where you have to write essays? Second, how do we grapple with the fact that a lot of writing skill is not just sitting down in a time-pressured environment and being able to produce a passable essay in that context—it’s going back and revising and rewriting and improving. That’s just much, much harder to do in a blue book exam.
Deming: I completely agree with that. I’ll admit that when I sent this email out to students, if you read the whole text of the email, I was bending over backwards to say, “these are just my thoughts, this is not policy, I don’t know the answer, but here’s how I’m thinking about it.” I probably underestimated how people would just not care about that and see it as the word of a dean, therefore some kind of authority. The reality is, not only can I not tell faculty who teach writing-intensive classes what to do, I don’t want to tell them what to do, because they know better than me. So I wasn’t even really trying to—I was just saying, in the spirit of getting the ball rolling, let’s play the game. That was kind of the spirit of it. But I can see why it would be at least taken out of context in certain ways. I don’t take back the sentiment, but I do think I could see how people would get touchy about it, especially given this moment we’re in.
What I want to underscore is: when you’re in a role like mine, where you’re responsible for undergraduate education, you have to look beyond the position-taking aspect of it—AI is good or AI is bad. It’s happening, people. We have to adjust to it. It’s not that we’re trying to sneak AI into the classroom where it’s not there—it’s just the reality that students are using it all the time. There are going to be faculty who teach these classes who have better ideas than me, and many of them are already doing a lot of these things. But I think, at a high level, one solution is to get much more deeply than maybe we’re comfortable doing, or used to doing, into the process of a student’s writing. You mentioned a three-hour blue book exam—one way to do this is to say, “okay, I want you to go write your rough draft in a three-hour controlled condition. I’m not even going to grade it, I’m just going to give you feedback on it, and that is the basis of your longer paper.” Then you turn in an outline, a first draft, a second draft, and I ask you, “why did you come up with this change?” In the same way that the more you understand what an AI model can do, the more you can align it, the more you understand a student’s thought process and their writing process, the more you can ensure that it’s actually them and not the AI. So I think it’s probably going to be fine to have writing-intensive courses where people use AI, as long as faculty and the teaching staff are deeply engaged in the process of helping students improve and test their ideas and refine them.
Mounk: As long as we’re not naive about the capabilities of the AI models, right? What I’m really struck by in academia is that there are obviously many academics who are deeply immersed in AI and totally understand its capabilities, and there are many others who, for ideological reasons, or simply because they’re not on the younger side and are slower to adapt, or for whatever other reason, have no sense of what AI can do. I was in a conversation recently where somebody said, “I know how you can deal with teaching this lecture class—you just tell them they can use AI to answer the ICA prompt, but then they have to themselves critique the answer that Claude gave, and that’s something AI can’t do.” And I thought, what do you mean that’s something AI can’t do? Of course AI can do that as well.
Deming: To be fair, it probably couldn’t do it that well two years ago. The technology is just moving so fast.
Mounk: Right, no, that’s fair. But the number of times I’ve heard faculty members smugly explain to me how AI-proof their assignments are, and I’m thinking, I could get Claude to do your assignment in two minutes—it’s quite astonishing.
Deming: That’s another part of the challenge for me, too, that maybe I didn’t fully grok at the time—I’ve been studying AI as a labor economist since before I became dean, and I’ve been pretty involved in that space. Not in a “I work with” or “I have relationships with any of the AI labs” kind of way, but I’ve studied it academically and used it regularly. So I’m pretty up to date on what’s happening in the AI space and have a sense of what the frontier models can do. 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. You just look a little bit down the pipeline here, and we have to get adapted to this like yesterday. There was a major cheating scandal at Brown last year with AI, and there’s no reason that wouldn’t be Hopkins or Harvard. I just feel a sense of urgency about fixing this.
Mounk: Tell us about that, because it was a fascinating scandal—walk my listeners through the nature of it.
Deming: An economics professor, Roberto Serrano—I may forget some of the details—he basically gave an assignment and allowed students to use, or actually I don’t even think he said you could use AI, he just didn’t say anything about it, and he got all these assignments back that were suspiciously well done. People were doing economics at a level much higher than what he was teaching—almost suspiciously so. The upper end of the grade distribution, relative to past years, was just—there were people getting basically 100% on these pretty difficult economic problems. So he decided, okay, now I’m going to give a final exam, but I’m not going to let you use AI. And he showed that the students who scored at the very top—many of them scored very poorly on the final exam. 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.
The important thing to me about that was that students knew in advance the exam was going to be in person and that they wouldn’t be able to use AI on it. So, in some ways, they were cheating their own thought process. Students thought, “I did well on this, I really get this”—in the same way that when you ask a friend, “hey, what’d you get for number seven,” and they show you, you have this instinct that now you understand it because you see the right answer, but because you didn’t struggle through it, you don’t actually understand it, or your understanding is brittle. What that showed me is that we are enabling learning shortcutting en masse in universities, if we are just silent about this. I feel an obligation to stop that from happening, because this is students’ chance to learn how to think and to learn how to write. Every year we delay is another cohort of students who are going to be harmed by this. That’s what really drove me to roll the ball on the court and, maybe, wear the black hat a little bit. I think it’s important to do—I think that’s what somebody in a position like mine needs to be doing.
Mounk: I strongly agree with you that the worst version of this equilibrium is the kind of don’t-ask-don’t-tell approach—I’m going to pretend you’re not able to do this with AI, and you’re going to pretend you did it by hand, and I’m not going to really check what you did. That ends up favoring the least engaged and the least good student. As you’re saying, in many areas now, one mark that something is done by a genuine student—we’re talking about very smart, very talented students here—is that it’s suspiciously good, because the AI models are that good now. If you know that many of your classmates are going to be using AI, and you have to pay the double penalty of, A, having to do the work when you’d rather be getting more sleep or going out and having fun with your friends, and B, getting a worse grade than them, that’s just putting students in an impossible position. We don’t have to be entirely surprised if some of them get tempted to cheat.
There’s one piece that I strongly disagreed with. I don’t know whether that’s something other people have disagreed with you on, so hopefully you’re going to get incoming from a slightly surprising direction here. I think, in general, it makes sense to go to blue book exams where we can. It makes sense to allow AI in contexts where you can do cool projects. For one of the projects I’m thinking of assigning now, I’m just going to have people build a public resource around democracy, obviously using AI tools—build a cool website that, in some way, gives these resources around it. That’s a nice different kind of thing. And then for another exam, we’re going to do a pen-and-paper one. There are going to be some contexts where we still want people to do extended writing projects that don’t use AI.
In the rest of this conversation, Yascha and David discuss AI detection software and the impact of the Trump administration’s attacks on Harvard. This part of the conversation is reserved for paying subscribers…