The AI Race is Here

The biggest problem with AI is simply that evil humans might use it for destructive purposes, and that is somewhat underrated. Now, I don't doubt there will be evil humans who will create some evil AIs to do their evil bidding, and that is a serious issue. And the better humans, which in my view includes the United States, needs to have the best or at least stronger AIs to combat that.

I've known economist Tyler Cowen for almost 25 years now. We met back in the early aughts after both of us started economics blogs. He is both the most interesting and the most singular person I know. I can't do better than that to explain the wild, wacky, and wonderful experience of knowing Tyler Cowen. I was with Tyler at a conference right after GPT released 3.0, the model that really showed the world what was coming with AI.. We talked about almost nothing else all weekend, and I've been following him on it ever since. So I'm thrilled that he's agreed to join us today to talk about how we think about talent, human capital, and competition in an AI world. Tyler, welcome to the show.

Megan, happy to be here.

All right, Tyler, I have to, I have to start with a Tyler-esque question. Uh, I know you get this a lot, but right now, is AI over or underrated?

Both. Everyone either overrates it or underrates it or both. Hardly anyone is rating it properly. So you have the people who work on AI, and there's a selection mechanism where, well, if you're doing that, maybe you think it's the most important thing ever, and you've seen so much progress in the last 4 years, all those expectations are validated. And then there's the newbies who have attached their status expectations to basically a non-AI world, and they're starting to worry that perhaps all of those will be dashed, and they hold it at bay by saying it's either a big fraud or maybe it will kill us all or whatever else. But they are underrating AI, and the truth, as is often the case, lies somewhere in between.

Relative to which margin do you think AI is most underrated, and relative to which margin is it overrated?

I think the biggest problem with AI is simply that evil humans might use it for destructive purposes, and that is somewhat underrated. They're ultimately glorified cash registers, so they haven't been through a Darwinian evolutionary process where they fight against other species. They're goal-directed. If anything, they're too sycophantic. Now, I don't doubt there will be evil humans who will create some evil AIs to do their evil bidding. And that is a serious issue. And the better humans, which in my view includes the United States, needs to have the best or at least stronger AIs to combat that.

So, I mean, this gets to, I think, where you think AI is going, right? If you think that it is self-accelerating, either because it's going to become a superintelligence and it's just going to start, you know, you're going to reach takeoff and it's just going to turn into some weird machine mind that mere humans cannot possibly understand. Or I think you can have a lesser theory where it just, it enhances your productivity so much that there's no second place for getting to a certain threshold, that once you've gotten there, you're probably going to just stay as the market leader for at least a long, long time. Do you believe in that scenario? And if so, which of those two scenarios do you believe in?

Maybe neither. We've seen a lot of leapfrogging in the AI sector already. So people had written off Google, and now there's people who will insist, well, Gemini is the best. Whether or not you agree, it's not a crazy claim. And certainly for images, it's the best. For looking for data from governmental public sector databases, it's probably the best. It's not the best for everything. So I don't believe in permanent advantage. Some people say, well, Elon is the most determined and he'll control the solar power up there in space because he has the rockets. Or maybe Meta has the biggest stash of cash longer term and they can control the complete stack. All these theories, like, okay, well, some of them will be right, but the fact that there are so many theories says to me it's actually been remarkably competitive, will stay that way for quite a while, probably for the foreseeable future. You have these incredibly strong, smart, sophisticated competitors. And I don't see that stopping anytime soon.

Is the obsession with takeoff and with winning and with this massive investment we're doing in it, is that driven by an inappropriate analogy from the internet where the internet creates a lot of winner-take-all markets, right? And all of these guys, the way they invest, they think about being the winner in the winner-take-all market. They think about being the next Facebook, the next Google. Is that a wrong way to think about AI? And if so, why?

We don't know yet. So some social networks, everyone wants to be part of the same network because their friends are. That can be winner-take-all. That doesn't really apply here. Now, in the case of, say, Google Search, not Google the company, but Google Search, well, it becomes a matter of data and having more data. And that can be self-compounding over time. But it's also interesting. You look at the discussion a year and a half ago, people were saying, oh, Google is cooked. Like AI search is better than Google search, so they're done. Now that hasn't turned out to be true. The share price of Alphabet has done extremely well. And furthermore, Gemini is now a leading AI model. So they're very much not cooked. It's a competitive market, but there's more competition in search now than there ever has been. I use Google for search hardly ever. If I want to go to say, Kennedy Center website. Well, I'll use Google for that 'cause I know where I want to go. But to look for something, I'll just never do it.

Yeah.

I use some AI model, which again could be Gemini, but probably isn't.

I do occasionally search. The funny thing that I use it for is I use it to find old columns. It's still actually like the usually the fastest way to do that.

I agree. I do that. I did that today.

You put Megan McArdle and whatever. Kitchen equipment. Kitchen equipment into Google. Occasionally I'll be looking for something quite specific and I'll use like a quite specific search term and Gemini, the little bit at the top of Google, will hallucinate some completely fake story about me and the kitchen equipment or, but yeah, I don't use it either, right? I use, I actually now have, I have 3 little GPT bots that I made. One of them is my researcher, one of them is my personal stylist, and one of them is my fact checker. And those, usually it's the researcher that I go to or the fact checker that I go to to do most of my search.

And for competition, Chinese open source already is potent competition. It's in essence free. You don't pay things to the AI company per se. And even if people don't use it, don't trust China, so to speak, it's there. UAE announced an open source model. Haven't looked at it yet. It's supposed to be quite good. So There's going to be plenty of competition.

I want to talk about the AI race. I was going to talk about this later, but like, let's just go right now. China versus the US on AI. Who's winning right now?

The US is clearly ahead. China, you could say, has the best open-source models. But I don't view it as a race of us against them, which is usually the way people think about these things. The long-run equilibrium could be there's a number of lesser powers including, say, terror groups that have bad evil AIs, and that the US and China work together to stamp those out or keep them limited. So our interests and theirs in some ways are quite opposed, say control over the South Pacific. But in a lot of ways, well, we want a lot of global trade. We want the rest of the world to be pretty orderly. We're the two winners of the current global order, and we share this kind of small-c conservatism in many ways. So China, to some extent, will end up on our side with AI. And it could be we work together with them in this super selective, never entirely friendly way to keep order in the world. I think that's more likely than it all just being a race of us against them.

What about Europe? So Europe has basically missed the tech boom. And I've been thinking about the fact that 10 or 15, 20 years ago, you and I were blogging and I—

23 years ago is the number you're looking for, Megan. More for you.

But over the years, we would have arguments with some of our favorite lefties like Matt Yglesias and Ezra Klein. Fun little club. And they would say, well, we should just be more like Europe. We should make those trade-offs. We should have more government spending. We should have more leisure. We shouldn't, you know, try— we shouldn't have this US kind of laissez-faire, winner-take-all economy. And that was an argument that I felt like I could win. 15 years ago, but I felt like I had to work for it. And now I don't feel like I have to work for it. Everyone is looking at Europe and saying, oh, oh yeah, this is really challenging, right? They completely missed the tech boom because of regulatory stuff, because their capital markets aren't deep enough. And now they're falling farther and farther behind the US, and it just looks like a much less pleasant trade-off, especially as you see their fiscal problems coming. Can Europe turn this around? Are they going to be competitive in AI, or are they just too far gone?

There's a few different questions in what you say. I think Europe is currently somewhat underrated, though it's not underrated in AI. Mistral is not currently on a track to make profit and be sustainable. The French government or the EU may in some way try to bail it out of the commercial mess. But at some point, you know, to be a good company and have first-rate products, you need customers who demand first-rate products. So it's not as simple as getting a governmental guarantee. Europe is in a quiet panic because they're going to be dependent on American AI products in a world where at least President Trump and possibly his successors don't treat them the way they would like to be treated, or even the way a lot of Americans would want them to behave. So that's where the panic comes from. But that said, the position of Europe is greatly underrated. If you look at human capital, social capital, political capital, Europe's one of the best places in the world. If you wrote down what are the best 20 cities in the world, by virtually any standard, more than half of them would be in Europe. You could sneak in Sydney, some places in the US, Canada, depends how you rate the cold.

China?

No, not one is in China. It's not close.

Why?

Often it's air pollution. Sometimes it's surveillance. Sometimes it's not sufficiently cosmopolitan. Free speech issues depends how you value those, but they're not going to help any. And just Chinese cities are not that pleasant. They were built quickly. They have amazing infrastructure, great subway lines. Parts of Shanghai are pleasant, but I've been to quite a few second and third-tier Chinese cities and they're not attractive. They're all these—

They're all these giant— residential blocks that are supposed to be self-contained and then there's nothing happening in between them. That's right. That's what it felt like to me. I mean, I was there a long time ago. It was 2010, but—

So we saw Japan and South Korea basically catch up to Europe in some ways, though with way tougher work hours. The question is, will China catch up to Europe? Not is Europe doomed?

My worry about Europe, it's a great place to vacation and I think it's a great—

Great to live.

No, no. I have never lived there, so I cannot— I've lived in London and doesn't count.

It's a different year.

I wouldn't— I also wouldn't say that, that next to Waterloo Station in the winter was a great place to live. Uh, London has many charms, but the Waterloo Station area was in, in, in a dreary March was not one of them. Um, but it— can they actually get out of their own way, right? You had the Draghi— Mario Draghi Italian economist, also served as Italian prime minister, also served as the president of the EU Central Bank, released a competitiveness report last year in which he said Europe needs to figure out how to get out of its own way. In fact, even though they've got this massive market, it's not united. They are— there are, in fact, a lot of internal trade barriers, much more than in the United States. And so even though they have a higher population than America and the idea was the United States of Europe, they haven't gotten there. What they do have is the huge overweening federal bureaucracy. It's kind of the worst of both worlds in a lot of ways, and that they need to fix that and really work on competitiveness. And I see people talking about it, and I talk to people about whether this is actually going to happen, because there was, in the immediate aftermath of this report, there was an outcry, you know, an upswelling of interest in doing something about this. But I haven't heard what concrete things Europe is doing to fix the capital markets problems, the internal trade barriers problems. And I think also the cultural problems of penalizing failure so heavily that it's hard to get stuff off the ground. And the people who do get stuff off the ground tend to come to the United States where it's easier both to get capital, but also I think to fail gracefully.

Well, that's why it's at, say, 70% of US material living standards. But tell me what's wrong with this argument. New technologies basically diffuse, right?

Sure.

Europe has iPhones. They're the same as our iPhones. They're just as good. So they'll remain behind the United States. But as we invent new and better things, which we're now doing at a very rapid clip, often with their aid, I might add, especially in the biosciences, that they'll pick up on those things with a lag and they'll stay at 70% of where we're at. So if you're bullish on America in terms of the delta, you should be bullish on Europe. And they'll be fine.

And are you bullish on America?

Yes, absolutely. So we'll do well. They'll stay at something like 70% of our living standard, and they'll do well too. What's wrong? Why is that wrong? And they're not going to fix all the long list. The Draghi report's great. They'll do like 17% of it. That'll put them up by a smidgen. But for the most part, they will adopt the new ideas that other people come up with.

So that puts— something I've been thinking about a lot is how heavily the onus now lies on the United States and China, right?

Possibly India.

Possibly India. Uh, I was, I was just in India actually last fall, and I was kind of surprised and heartened to see things felt like things were moving in the right direction.

Great street energy there. Where were you, if I may ask?

Uh, God, where was I? I was in Mumbai. I was in Varanasi. Um, you didn't tell me about any of this. This is terrible. Yeah. Mr. Cowen. I've been a little busy launching a podcast, Tyler. Give me time. I was in Nagpur, and it just felt like, A, the street energy was good, but it also felt like it was really interesting talking to government officials because the attitude towards— now, you may or may not like the nationalism and so forth because that wasn't what I was talking to them about mostly. What I was talking about was economic policy, and it was really interesting talking about how they had actually gotten a handle on inflation. Also how that hadn't actually improved their bond yield as much as you want it to yet. Someday perhaps, but they're still paying, their interest rates are still quite high because they've had inflation forever. And it does feel energetic and it does feel like maybe the policymakers at the top are really focused on the same kinds of problems that Ezra Klein and the abundance people are focused on in the US.

And many things they can start from scratch on or near scratch, right? That's a drawback and an advantage.

Like what?

Well, say you're doing a healthcare system, right? The US healthcare system, whatever you think about it, it's very hard to change. We would agree, Ezra would agree. The Indian healthcare system, first, there's a lot of action in their states. Some of their states have hundreds of millions of people, so there's competition across the states. A lot of policy is done at the state level. And there's a lot of areas like cardiac care where they have these super specialist surgeons who do things a new way. They do the same operation again and again and again, and that's all they do. And it's higher quality than much of the US and much cheaper, and it just seems to work. And there's a lot of parts of their healthcare system where I think they can and indeed will do things that way.

Yeah, I actually wrote a column about this when I came back about the fact that I think on both left and right, there's a real yearning for the kind of I don't know, call it 1930 to 1960 period in the US where so much is happening. We're building the Hoover Dam and our economy's growing at, you know, 7% a year. And all we're doing these incredibly scale projects at both the government and in the private sector. And that dynamism has slowed a bit, especially on the government side, right?

Not in the last 5 years. I think we're back at it. mRNA vaccines, immunology treatments against cancer, AI.

Sure.

Anti-malaria vaccine.

We're not building as much infrastructure, and I think people really—

We don't need it, though. Yeah.

But this was actually my point is that, you know, you go to India and you see people getting super excited about getting a toilet, even though often these— this was a big campaign by Modi. And often the toilets, I realized, someone told me, were not actually always hooked up to plumbing. Just merely having something indoor that you could use was a big improvement over going and finding a bush. And that excitement, right? When you get your first toilet, that's an amazing life change. When you get your third bathroom and you've got beautiful marble tile on the floor, like, it's really nice. I'm renovating a bathroom right now. It's nice. I'm very excited about my new bathtub. But I'm not nearly as excited as my grandfather was about having pipes that just— you could turn a tap and make water come.

Are you not excited about living to age 97? I'm not sure you will, but younger people probably will.

I think that's great. No, no, I—

That's very exciting, especially if the time comes when you're otherwise about to die. That's the real excitement. Oh, I can stay alive.

No, I think—

That beats the tile floors.

Absolutely. This is true. I'm excited about MRI. MRNA vaccines. I'm excited about what Moderna is doing on cancer. I don't know if it'll work, but most things don't work, but I'm excited about all of those prospects. But on the other hand, if it's just a really slow incremental thing, it's often hard to see. And I actually think that this is one reason that people are so down on healthcare spending, right? The premium gets taken out of your paycheck every month. Your pills get— you have to pay for And you don't see the stuff that's just incremental improvements, right? So the premiums go up because we got a new cancer treatment and that cancer treatment costs $500,000 or $700,000 or something. But you don't have cancer right now. Now, the person who has cancer and has this new treatment is super excited about it. And you should be super excited that if you had a 1% chance of getting that cancer, you now have— and it was— it used to be a 1% chance of dying from that cancer and it's now a 0.5 or a 0.2 or a 0% chance of dying from that cancer, you should be very excited about that. But that's not how humans think. And I think this is actually a problem for our political economy.

But it's going to happen anyway. That's the good news, right?

Yes. But I wish I had some way to convince people that, you know, there are a lot of marginal improvements happening and they're not as showy as getting electricity or seeing the Hoover Dam. But that doesn't mean they're not great.

But some are pretty showy. GLP-1s. Yeah, it's— they haven't quite fixed obesity, but a major weapon against obesity, diabetes, and it's visible.

Yeah.

And tens of millions of Americans love them. More to come, right?

So I want to go back to the question of AI. Talk to me about human in the loop, because this is something that I think a lot of people are underestimating, is that even if the AI is doing everything, you need a human being who you can blame Who is responsible? It's not even you can blame. Who is responsible for that?

They fall illegally, right?

Yeah, legally.

Fiduciary responsibilities.

But you saw that white hat hacker who managed to twiddle with ClaudeBot?

Yes.

And have you tried ClaudeBot, by the way?

I tried it. It was not— you mean Claude Code?

No, ClaudeBot.

No, I haven't. I'm afraid of it.

So for listeners, the idea of ClaudeBot is that you run ClaudeBot on your laptop and it takes control of your laptop and your life. And it can do all sorts of interesting agentic things and also perhaps send everything in your bank account to Africa in order to help people who need bed nets for malaria.

It's on purpose that I don't have online banking, you know. Really? You know, when I saw Battlestar Galactica way back when, I'm like, I'm not going to get online banking because Captain Adama saved humanity that way.

But does that really— does that really save you?

I mean, we'll see. But I feel I have an extra layer of protection because to get certain things done, if someone took over my computer, they actually would have relatively little power over me.

If someone took over my computer, they would still need two-factor authentication. They would still need to remember my password, which I can't do half the time. I mean, there would be— there would be a lot of layers between them and getting to my bank account. So if just trying stuff becomes almost free, then what's scarce? What is the scarce thing that then becomes more valuable? Is it taste or judgment or stamina, or is it the willingness to be embarrassed and the willingness to fail?

Well, just trying stuff won't be free. You have to know how to manage the AIs. And in a sense, no one knows that yet. So it's the scarcest thing there is. We're all trying to figure it out.

What makes someone good at at, like, what do you look for when you think about who is good at managing the AIs?

I sometimes compare it to being a very good dog or horse trainer, and this analogy should resonate with you.

I have two bullmastiffs. Yeah. So yes, it does resonate.

The dog wants to cooperate with you. It's been bred that way, but it's somewhat recalcitrant also. And you have to get in a rhythm and a groove where you know what it wants, it knows what you want. It's sort of trained. It becomes a routine. It's quite possible to do, but not everyone is good at it. Being smarter helps you be good at it, but the people who are the best dog trainers are probably not the greatest geniuses of mankind either. So understanding alien intelligences, being flexible on your own, willingness to experiment, taking feedback from how things are going, A lot of factors that are correlated with intelligence, but not so, so, so closely.

It's not just that it is going to increase the returns to just trying stuff because it makes it cheaper to just try stuff, right? It's also going to increase the returns because at the moment, precisely because no one knows how to manage this stuff, we're all learning at the same time. Does this massively advantage people who are like blogger types, I guess is the word, the people who are just like, I don't know, I'm going to let it all hang out there. We're going to see it. We're going to see what happens.

Yes, it does. Like, "Oh, I'll do this," right? And let it rip.

So this brings me to something I've been thinking about a lot. I talked about this with Neal Stephenson on this podcast, that the last 30 years has basically been the revenge of the nerds. I mean, when I was growing up, nerd was not a term of kindness or admiration. It was the people who got picked on and excluded. And then somehow, starting sometime in the 2000s—

Had something to do with those high salaries, right?

Yep. It— being a nerd becomes cool, and it gets not only cooler, it gets more lucrative. And of course, those things are causally related, although I think the relationship is probably inverse. And you and I both benefited, I think, a lot from the revenge of the nerds. Um, but So, but now AI is a better coder than a lot of coders. So does that change? Does that reverse the Revenge of the Nerds? Does it actually mean that there's gonna be more value for people who are good at social engineering, who are good at doing relationships, who are good at just the normal people skills that used to be really valuable in high school before everyone got into being a tech wizard?

Well, that's true, but the nerds still will do fine. So it depends what kind of nerd. If all you know is programming, maybe you should be worried. But nerds on the whole, they tend to be what I call meta-rational. That is, they're good at analysis, good at knowing when they're wrong. Programming, finance, nerdy activities. There's a lot of feedback, right? You have to deal with the feedback. Nerds are ahead of the game in terms of seeing this coming. Learning how to use the new tools. So nerds as a class will do great, though not every single nerd will. But again, any change, you could say that. Now, I think people who are both nerds and are great at networking, they may do best of all because the network really matters.

Why does networking get more valuable?

Here's a story for what you might call an ordinary worker. Everyone sends in a perfect cover letter written by GPT-7 or whatever. And it all looks the same. And who you know gets you opportunity. It's not in every way a good thing, right? But the easier it is to fake things intellectually with AI, the more it will matter what your actual human connections are. An example for startups would be, well, let's say you can now start what might be a serious company with only 5 people, as we talked about before. You still need some capital,. And there'll be a lot of people in that position, but who can vouch for you and go to the VC firm and say, well, I know this person, you know, they would be good at starting a company, give them some capital. And that's another way in which the networking will matter more.

This is a thing that you've talked about that I've been thinking about a lot is just that signaling is about to get really weird.

That's right. And a lot of it will go away because the AI can just talk about you.

Right. And I think a lot of people don't think about how much of their life is signaling. It's just, you know, the obvious example is a bachelor's degree. A bachelor's degree, does it signal that you have now mastered the rudiments of history or English literature or business or whatever you've studied? Maybe. But a lot of what it signals is just, here is a person who put their butt in a seat, —and finished. —and finished, right? They just did it for 4 years, which signals a bunch of stuff about you, from the fact that you probably have parents. Not always. Many— there are people who get themselves out of situations where no one's gone to college and no one's pushing them to. But you probably have parents who are college-educated. If you don't have parents who are college-educated, you probably have an extraordinary amount of grit and, you know, stick-to-itiveness. And that you're, you're telling that to employers. They don't really care whether you learned about Tudor England or about— they care if you're an engineer, but for a lot of the subjects people study, they don't actually care if you have mastered the details of some topic. What they care about is that you are the kind of person who can master the details of some topic long enough to get a degree. And that signal, I think a lot about what's going to happen to education. It seems like it's a potentially existential threat.

There's one scenario. I'm not predicting it, but it's possible. You grow up with an AI. The AI records data on you for years. And when the time comes to look for a job, you send that file in. And you might say, well, who would want to send that file in? But if you don't, it's a negative signal, right? And maybe the equilibrium is you send in some version of that file. It's stripped of personal information, but how good you are at math, how much you know about Tudor England, whatever. The employer can interrogate the thing, and that's your signal.

I worry about late bloomers like myself, right? I was a much more attractive job candidate at 30 than at 15.

You would do fine under the system. The AI would see in advance that you're a late bloomer because you have all these correlates of being a late bloomer. Which— such as? Obsessiveness at an early age, interest in books, family background. Just for a start. Articulateness.

Both my parents nearly flunked out of college. I nearly flunked out of college. I don't know what gene controls nearly flunking out of college, but whatever it is, we all had it. We all went on to do fine. My father got a PhD. I got an MBA. I am gainfully employed to this day.

At some point, the AI will see through all of that. Well, we hope so. And stamp the endorsement on Meghan McArdle. But for the world as a whole, it's much trickier. Because not everyone is like you, right? Expand on that thought.

I mean, not in the self-flattering way, but expand on that thought.

Well, how's the AI going to evaluate John Lennon, for instance, who is obviously an incredible talent, highly intelligent? I think the AI would pick up on that. But what will this guy actually do? The AI will ask. It's like, well, he could be part of the Beatles. Okay, but that's a tiny chance. And after that, it's not entirely clear what he would have done. Well, wasn't he—

he was a bit of a street brawler. So maybe he was a security guard.

But you see the issue, right? Yeah. Vincent van Gogh, right? The AI interviews Vincent. Vincent, where'd that ear go? That's one part of the interview. It's not gonna go very well. Now you could say, but a highly creative fellow. Okay, but.

I mean, but can't you say on the flip side though that the AI will not have our human prejudices about people who cut their ears off and try to nail them But people who look bad on paper but are one in a million chances, like Van Gogh, like maybe John Lennon, they might have trouble in this new regime.

Tyler Cowen.

Would Tyler Cowen have trouble in this new regime?

No, not at all. I would be like you. Like, it would see early in my life I was smart, I was good at chess, I read from a very young age. I would do great.

Well, maybe it would just make you go be a chess grandmaster.

Well, it wouldn't make me do anything. It would tell me at some point to leave chess because the pay was low, especially back then. Rational. So that's fine. And I did that anyway.

I do think that even the conscientiousness signal is quite different between I had to turn in some term papers every term or else I wouldn't have passed my classes, or I spent 5 minutes prompting GPT to write a minimally decent essay a few times a month. That's a much easier signal. The latter is much easier to—

It's amazing how many people drop out even with AI and the ability to cheat. Really? You'd be surprised. Why? There's something about finishing that is simply hard and not getting too caught up in the problems of your family, your friends, your personal life. The AI helps a bit, but I encounter students and I hear stories, which I'm really not supposed to relate, and it's just very clear they're not cheating, which is good. But you're wondering a bit, like, how does this whole thing work? But I think Uh, what do you mean? When it comes to the dropout rate, it will not change much because of large language models, and we don't see it changing much, do we? We've had GPT-4 for a while now.

The answer is I don't know. I do know that all my professor friends complain about how many people are obviously cheating in their class.

At a high rate. I'm just not sure it matters that much because it's finishing that matters, not your grades so much in most areas.

So on the flip side, if you want to boost the value of your degree, you need to be teaching people somehow how to use this AI and not to cheat with it. Of course. And I think about what a university like that would look like. You recently kind of outlined some ideas at— in a talk at the University of Austin, where you were a board member, advisory board, which is not a fiduciary thing.

Nor am I paid or anything like that.

What are the, say, the big 3 things that you think higher education needs to do to restructure itself to be, to make students who are AI native and hopefully AI masters?

Higher ed is not one thing. So if you're talking top 10 or 20 schools, they're doing great. They'll still do great. They could exist only as dating services and be fine. Though, of course, they're more than that.

But the kids aren't dating anymore, Tyler. Well, the schools can bring them together somewhat more perhaps. Maybe AI matchmakers where you don't even, you don't even get to choose the A. It'll be like the old, you know, the, what's the name for the, the yentas, right? Who just, they show up and like, well, this is who you're marrying. But aside, outside of that.

All schools should devote a significant portion of their curriculum to teaching the students how to use AI with the following caveats. First, most of the faculty don't know, so there's no one to teach it. In that sense, they can't do it. Often the students know more than the faculty. When you teach the students how to use AI, you're not pushing them away from learning other things. You're using the AI to say, read Homer or learn calculus or whatever.

Oh, I love reading with AI. Yes. Because you just— I mean, it's not like I'm doing it every page, but just the ability to stop and say, I don't understand this. What is happening here? What am I missing? I do it now with economics papers too, where if I come across something where I'm not understanding what they've done in this paragraph, I just copy the whole paragraph and I say, could you please translate this into layman's terms? It's extremely good at it.

My day is just filled with AI prompts every day. I run my own podcast. I have to prepare for guests. So one pending guest, he's an expert on ancient Egypt and Ptolemaic Egypt. It'll be a good episode, but those are hard topics for me to learn. It's like Chinese dynasties. You can memorize them, but what does it all really mean? I find it really difficult.

I did not realize the extent to which— because now my brain is old and un-plastic— but just how much easier it is to learn Western history because the names are not as immediately legible.

And you can link everything maybe to a place you've been or some other thing you know. Yeah. If you're from Ireland, where your family was from, right? Much easier. So ancient Egyptian history, I've always found very hard. And now I'm reading his books. There's so much I don't know. Many, but not all of his books are written for specialists. So I just keep on asking, asking, asking. I can prepare for that podcast in less than half the time it would've taken me a few years ago.

So you wrote a book with Dan Gross about talent. We've talked about how people should think about preparing themselves for AI, but how do companies think about finding talent in an AI era?

Right before AI, you and I were at a conference.

That's right. When GPT-3.0 came out and it was all anyone talked about that weekend, or at least all anyone, the people I talked to. There were a lot of people at that conference. But so it was right before that. How have you, how is anything in the, that you wrote in Talent with Dan Gross, Has your thinking on any of that changed with the advent of AI?

Daniel and I wrote a paragraph in that book. We said, we know AI will be important, but we're not going to talk about it because we know whatever we say will change so rapidly. It would require something of its own in the future. And that's held up very well, that paragraph. You need to test people on how well they use AI. Now, for most endeavors that you might want to hire them for or fund them for, So one way to do that, if you can, is just sit down with them and ask them to start prompting and see how they do. It's not always feasible, and you can help them and see how they respond. It's a very simple way of conducting an interview where you learn a lot about the person quickly.

What is your sign of someone who's good? Like, what do you look for when you see someone prompting the AI?

Well, if they're amazing at it to begin with, that's wonderful, but you don't, really need that. Most people are not. The ways you can learn to be a good prompter, you can learn very, very quickly. So you just see how quickly can they pick up the tips you give them. And that's telling you something about how they'll adapt to the AI. It's not perfect, but given where we are now, now over time, the AIs will make the judgments. There's a number of companies taping large numbers of interviews. And over time, they'll track people and see how well they've done 10 years later. We need the 10 years to arrive. But we'll have a lot more data about what in an interview predicts success. And of course, we'll use it. This is in some ways intimidating. It's like the John Lennon question, right? Hey, Vincent, where's the ear, right? Guy could not answer question about ear. Negative signal would be the normal response.

I feel like Van Gogh would've in fact answered the question. It's just that the answer would've creeped out the AI. I don't think, I mean, he was pretty open about it.

I think he was. You have to be when it's the ear.

Indeed. Indeed. So does AI raise or lower the returns to conscientiousness?

I think it raises the returns because the AIs we're working with right now, and this could change quickly, but you don't have to be a genius to work them. You need to pay attention and be alert and flexible, but not genius. So simply the ability to keep at it, I'd say some mix of conscientiousness with a minimum of flexibility. The returns to that have gone up. If you're just super rule-following and have no flexibility, that kind of conscientiousness, probably the returns are down.

Does it raise or lower the returns to intelligence?

Raises, because you can multiply your efforts much more.

But it's also a substitute for your efforts, right?

It's both, but at least in the immediate future, short of intelligent humanoid-like robots, the fact that you can operate in the physical world gives you a huge advantage over any AI. And if you think about, like, human beings in nature, if we fought against different animals, you know, mano a mano, we often lose.

I don't know if apparently if you ask men, they're like, absolutely, I could defeat that bear.

You could beat the badger. But a lot of times you lose. Sometimes you win. But in practice, you win by integrating all of your different facilities and getting weapons, civilization, whatever you need to beat the relevant animals. So when you're competing against the AI, you're never— or not necessarily beating it on pure smarts now, or certainly not in the future. It's the fact that you can integrate so many abilities. So physically, people can run a bit, somewhat fast. They can jump a bit, somewhat. They can throw things somewhat. There's a bunch of somewhats, but we integrate them all together. And in the physical world, we do fine against animals.

You don't think that that gets beaten by specialized robots that do one thing except— or do a few things exceptionally well?

In factories, we've had automation for a long time. People adjust to it. Now, is there some very distant future where there are human-eyed robots that move through space and act like Megan MacArthur, but look like liquid metal or something. I just cannot picture it.

Why wouldn't they want to be as beautiful as me, Tyler Cowen?

They would want to be, but what robot could be, right? I find it hard to speculate about a world that different, but that's far away. And in the meantime, you combine all these, you do it to some degree, like you can get on a plane, you can shake someone's hand, you can appear in a meeting, you can charm someone, whatever. And the AI can't, so.

I'm pleased that you think I can charm people.

This is why it's a magnifier for most intelligent people. But again, if you're not flexible, it's going to be a big problem.

So if you had to rank the channels by which AI might make talent search better, right? We've already talked about how it could get worse, right? Is just that it destroys signaling.

But it destroys signaling by competing with signaling. It outcompetes it. You don't need 4 years of undergraduate college necessarily in some cases if the AI can vouch for you. So you have a cheaper signal would be how I would put it.

But I mean, the cheaper signals themselves can be a problem, right? And I think that we've seen this with recruiting where everyone is just flooded with perfect AI resumes that are completely impossible to distinguish from each other.

But the AIs, I think, will judge talent pretty accurately in the intermediate future. Yes.

Will they also help with just matching, with getting you the people that you want on the job?

Absolutely. And they already do.

But it's also, it's a problem in finding talent, right? It's a problem that you, that is now, I mean, you've seen this in colleges where everyone is applying to every elite college and then the acceptance rates go down. But are the matches better or are they worse now?

I just think it'll be much easier. And then people in other countries already have access to GPT, whatever, Gemini is free. So they can do things with AI and show off their talent. And yeah, I'm very bullish on talent allocation in a world with AI.

What should people be doing if you, if like a normal person, not a Tyler Cowen, but just a normal person who's kind of sitting in their job, they've got a good white-collar job, they're thinking about where to send their kids to school, what should their kids major in, or what should I do to make myself AI ready? Like what do you say to those people? I assume you get this question a lot.

Multiple times a day. There's a lot of scenarios. It depends on the ages in question. If you're 43 years old and have a good white-collar job, I just say learn how to work with AI. No one's going to be sure what's coming for you, but better to learn things now than start from scratch. But I do tell them, look, look, we don't know. You might have to change jobs. There will be plenty of jobs. There'll be a lot more jobs in the energy sector. And the biomedical sector, and you might need to be more flexible than you desire. Academics also, or people who wanted to be academics especially, are at risk. So if you thought you were gonna be a mathematician or say economist, I'm not sure what those jobs will look like in 5 years or 10 years.

I think about the academic problem a lot in part because I can imagine ways to turn a university into a more AI-hardy place, right? If I were gonna start that university, would I start with people who are selected by their ability to master the current path to tenure? Probably not. Are those people going to be interested in doing the things that will be productive in the future? Also probably not. Strong agree. Um, and the governance of universities is crazy. Right? Breaking anyway. And I said this to an academic who came back. I said, you've got this huge problem, which is that you are kind of selling a bundle. You are selling all of these classes that signal various things to employers, but your individual— you could be really good at producing AI-literate students, but employers have no way of seeing that you personally are very good. They need to have a basic faith in the bundle. And I said, and the problem with this is that the administration doesn't have any governance way to make the university better as a whole. And he came back and he said, I don't think you understand. No, we design our classes ourselves. And I was like, yes, that's the problem. Exactly. Like the, the, in fact, it is a problem that this structure that has given a massive autonomy to professors. And I'm not even arguing about whether this was a good or a bad structure 20 years ago. But right now it's a real problem to reorienting the university to do something incredibly different. I know that's been a problem before.

And put declining enrollments on top of that. So you can't just solve the problem by throwing money at it.

How do professors that you talk to about this, how many of them understand that problem? Very—

oh, declining enrollments, they all understand.

No, no, the declining enrollments, but the problem of— Very few. Very few.

People are starting And they're, oh, we're going to do innovative things to blend AI into our classes, which is of course a good idea. And then we're going to have a policy to limit cheating. That's of course a good idea. But there's not a true sense of urgency based on a real understanding of how much things at some point will just have to change. It's easy for people to be in denial when they're tenured. They feel they have guaranteed jobs, which may or may not stay guaranteed, but it feels that way. And then it's a real pain to restructure your research, your teaching, your whatever, to incorporate AI in a major way. Most people don't want to do it. And they go, "Oh, I tried GPT, whatever, it hallucinated for me." Or, "Oh gosh." Or they only regurgitate. Yes.

And I think that this is an incredibly dangerous attitude. Look, I have respect for someone who says, "I don't like this future. It's not what I wanted. I like the old thing and I want to keep that." And I totally respect that. It's not my personal approach to it, but—

I hope we keep parts of the world like that, like we have the Amish today. Absolutely. Yeah.

But on the other hand, we've got a bunch of regulations that assume imperfect compliance, right? We've got speed limits. On highways, very few people drive the speed limit. And it's dangerous to do so often. Yeah. I sometimes drive the speed limit because I'm basically a grandmother in a younger person's clothes, although not that much younger these days. You don't want a system— you can imagine a pretty, pretty easily an AI system that would have perfect ability to detect speeding violations, right? You could just string cameras along highways. They would have a combination of license plate recognition and facial recognition for whoever's in the car. And you have to assume those systems get cheaper and better very quickly, along with drones, right?

We've done that for a while, right? That's hardly new. We can do a lot more of it and better, right?

And, and it could be everywhere. It could be in all public places. But the thing is, you don't, if you do that, you also have to adjust the speed limit. If you do that without adjusting the speed limit, you're gonna get a bunch of different kinds of problems, including a lot more traffic, right?

There may be some future world, not near, but where there's no speed limits, but there's complete surveillance of vehicles. And at the end of the month, you pay a fine based on how well you drove.

But isn't that also true for companies to some extent? Like they've got a bunch of rules, but, and formally, right? The informal rule is do not make a problem for HR. But the former rule is much more restrictive than that. And are we, as compliance and monitoring get much easier and cheaper, are we gonna have to go reteach people how to— are we gonna have to retool the rules? And how do you retool the rules in either the amount of detail that human intuition and give and take does, or do we just have to rethink rules entirely?

We'll have too much surveillance. So what China does, if you jaywalk, I think it's in Shanghai, they don't even send you a ticket. They just deduct it from your account and you get a little notice that the money's gone. I don't think America is going to go that route, but the whole notion of rules and very strict thresholds we'll need to redo. And again, government will lag and there'll be these significant social problems in the intermediate periods. And then how are people going to agree on the new system? Whatever you and I think it may be, everyone's going to have an opinion on matters such as this. But no one's talking about this yet, right? No one's talking about it. It's going to take quite a while to converge on something that becomes a new consensus.

How do we want rules to look? Yes. Right? I mean, this sounds like such a dumb question, right? Like a kindergarten question. Yes, exactly. But it's central. But it is central. And as he said, what I thought was really useful in the post was he said, Don't think of this as like you're like a congressional regulator, right? Think like you were at the Continental Congress. You want to make the rules for making the rules.

And it's game theory all the way down. Yeah. So what, what do you think?

Have you had any useful thoughts on that yet?

As an economist, I'm led in the direction of Pigouvian pricing. I'm not sure that's socially sustainable.

People hate games. I'm going to interrupt you to explain what Pigouvian pricing is, where you want to basically kind of price externalities into— am I saying this right?

Yeah, so imagine a system where all your driving is watched, though the personal information is disposed of. The AI decides how well you did that week. And if you were driving recklessly, you pay a fine in proportion to the gravity of your mistakes. So that encourages people to drive better. You don't have a fixed rule. It requires A, a lot of trust in the AI, B, people accepting Pigouvian pricing. And we know they hate carbon taxes, right? They hate—

They hate having to be— I mean, and this is the thing. On the one hand, there is a possibility that AI could eliminate crime as we understand it, right? Because the ability, in the same way that, you know, one, Derek Thompson recently asked, why did all the serial killers go? And there's a lot of answers to that question, but the most, you know, we don't have so much lead poisoning anymore. We don't have, we haven't sent so many men to war. They're not so psychologically scarred. As the World War II generation was. We probably do a marginally better job policing child abuse, and most serial killers seem to have come from pretty abusive situations. But one answer is just, it's way harder to be a serial killer now than it used to be. And that—

Both cameras and DNA evidence. Yeah.

Bill James, who has a great book on sort of popular— It's called Popular Crime, and it's just about big, famous— Wonderful, underrated book. Holy—

One of the most underrated books underrated books.

I love that book, in part because it's episodic and you can just dip into it whenever you're feeling a little blue. But, you know, he says in the book that at one point he was gonna do— because Bill James is a sportswriter, he's the Moneyball guy. But what he apparently in his spare time was doing was reading, like, every true crime book ever written. And then he wrote a book about the true crime books. Very autistic thing to do, of course. Yes. Wonderfully autistic thing to do. And it's incredibly well-written, and it's super funny. But he said at one point, he decided to sit down, and he was gonna help the cops by figuring out, out of all of these crimes, how were the serial killers caught? And he would have some sort of statistics. He's a statistics guy. And then he stopped the project because the answer was almost always the same. It was just victim got away. And once the victim had gotten away, you had a description of the guy, and you could go looking for him. And the point he makes in this is that it's in fact just incredibly difficult to catch people who are killing strangers because they're— that they have no connection to.

Crisscross strangers on a train, right? Yeah.

Hitchcock understood this. And so the prob— but that has actually changed, right? Now they've caught— they started catching people by where were you parked? Have you— where were you suddenly using your credit cards? There was a guy in Alaska who was flying a plane in, killing people, and then flying it out. And he eventually got caught because he used a credit card, right? And those things have changed. And it's actually— so you could imagine more changes along that lines where it effectively becomes impossible to commit murder and get away with it. And that would be good, I think.

But the same technologies government can use to extract all the surplus from us, right?

And also it could be used to create more crimes. Right. In a way that China has, right? I mean, jaywalking was not really a crime. It was like, maybe, I know in Germany, Mike, economist Mike Munger has a funny story where he's in Germany and he starts to jaywalk and some little German lady starts batting at him with her umbrella screaming, "Die Kinder! Die Kinder!" Basically saying, "Don't do this. You're setting a bad example for the children." But in the United States, it was not a crime. No one was going to book you for jaywalking. I love street jaywalking. Yeah. And but you could make it a crime if you had ubiquitous surveillance already. And so it's not just extracting all the surplus, it's extracting all of the weirdness. That's right. And the little— the things we do and the spandrels to make life slightly more bearable.

One way to put it, a world with strong surveillance AI requires much more trust in government, than we have in many places, including most of the United States. And no one has a good answer to that problem.

And no one's really thinking about that problem as far as I can tell.

Well, China thinks about it a lot. China, sure, but like— And we attack them a lot for what they do. Just China. But we don't turn the mirror on ourselves very much.

Is there a lot of trust in government in China right now? No. In my opinion, no.

Yeah, I don't think so. People may trust the government to do particular things well, which it might do well. But there's not general trust in government.

How does that collide with more and more AI use? Does AI enable China to just keep cracking down more and more?

Or maybe they overstep, right? And the whole structure breaks somehow. So there's always two effects. One's the short run, what changes at the margin? And then there's the how do other parts of the system react to that? And we need to look at both. So we don't know.

All right, we're going to do a few lightning round questions. I'm ready. What are people underrating for the next 5 years in AI? What is the most unexpected thing that's going to happen in the next 5 years that no one's thinking about right now?

I mean, there's enough people thinking about it. No one won't apply. But just the sheer level of frustration of trying to get things to work, I feel, is often underrated by AI partisans. I also think people underrate—

and I've written a couple of columns about this— that people underrate the slowness with which institutions change. Of course.

What is your most absurd belief about AI? Here's an absurd hypothesis I entertain. I certainly don't believe in it at 0.5, but that there's some Lamarckian-like mechanism that turns down the dial. We're not in a simulation, but there is something deeply strange in, say, epigenetics, where when the number of people goes up enough, fertility goes down. I'd give that some small percentage of being true, though it goes against every science I've ever been taught.

If you were appointed AI czar— Yes. —of your university— Yes.

—what would you do? I would just get faculty using it more. Because ideally, I'd like to have a third of all instruction devoted to teaching students how to use AI. But it's clear we don't have the resources or faculty to do that. What's the resource constraint versus the faculty constraint? Anything in a university takes budget, even if it looks free when you write it down on a paper napkin. And there's just not enough money for anything, much less that. So, oh, do we have to get people access to this? Or what if someone has a disability? And my goodness, it goes on, right? So it's a tough one, but to work to a world where having a lot of classes teach AI proficiency And some do because the instructor just decides to, but most do not. I'd want to be in a position where I had the faculty to actually do that.

And last question, and this is for the entire game, Tyler. So think carefully.

This is double jeopardy, right? Yep.

What's something you're optimistic about, and particularly something you're optimistic about that most of your audience would find implausible?

If you ask, what am I optimistic about in the future of AI? I think it's simply that the models will just keep on getting better. They're already much better than people realize. That's a critical fact. Most people are not spending enough money on having the very best model. That may change over time. Should they be spending more? Many more people should be spending more. Yes, absolutely. And they will be. But right now it's easy. You just use the app on your phone. You don't even necessarily, you know, flick on the menu to the best version of it because some versions are cheaper to supply than others. And the result is that people think these things are much worse than they already are. And I'm optimistic about the fact that people are going to learn this and the models are going to keep on getting better. And there's no natural end to that process.

Well, I think we're going to have to leave it there for now. Tyler, thanks for coming on the show.

Megan, my pleasure.