Tyler Cowen on how AI will reorder economies, schools, and spirituality

0:00The AI is already smarter than I am. If there were some kind of like economics Olympics and I was up against 03, 03 would beat me. That's not hypothetical. That's today. What's interesting is how little it matters. Did you have a particular oh shit moment when you were using these large language models where you saw something that you hadn't expected? Real depth. So I use 03 a lot to ask it about classical music. I'll say, well, I'm going to listen to Sibelius's first symphony. What should I listen for in the symphony? It's remarkably deep. You could almost say empathetic. It just understands the stuff better, I think, than any human source you would find.

0:34A number of people are going out today saying we're on the verge of a great job displacement. I do not expect huge job displacement. There will be an enormous number of new jobs incorporating AI into the routines of established institutions and also building out new institutions. Those new efforts will be a little slow. And in the meantime, a lot of employers will be somewhat skittish about hiring more people. He said, by 2030, AI could start to fundamentally challenge what it means to be human and cause an identity crisis. I think it will change religion. Some people will start to regard the AIs as oracles.

1:09In some cases, even gods were in for some pretty big shock.

1:18I am so excited about my guest today, Tyler Cowen. Is he an economist or is he a polymath? I think he is both. He shapes conversations on economics, culture, technology, and everything interesting. I have been reading his blog at Marginal Revolution for, I don't dare admit. So wonderful to have Tyler here. Tyler, we're going to talk about AI and the economy and beyond. I am ready. And we have met in real life, just so our listeners know. Indeed, indeed. So let's get started. I'm just back from a trip to Silicon Valley, And I heard a lot of people there talking about the prospect of AI driving 10, 20, 25 % of economic growth.

2:01There would be this bounty that AI would bring us. And I don't really agree with that. I think you don't agree with it either. But what are those projections getting wrong? They're ignoring the all-critical role of human imperfections in systems and institutions. So the more powerful AI becomes, and I'm not a pessimist on that, the more it bumps against people who don't want to adopt it or institutional systems where it does not get incorporated into the workflows. It also hits higher energy prices, constraints from data centers, possible regulatory constraints. The general economic point is that as one thing in your economy gets better, the remaining imperfections become all the more important.

2:45So they're not thinking about this like economists. They're used to working in one sector of the economy where, yes, a lot has changed very rapidly. But when you look at health care, education, government, the nonprofit sector, as you work in these, as I have, you will see the rate of improvement is going to be pretty slow. I certainly feel that. In my first book, I talked about the idea of the exponential gap, which is the technologies can go exponentially, but human institutions are linear. They drag us down, as you say. people have different incentives and there are places where there's just a lot of people that you have to shift and move around.

3:23But I'm curious about whether this time might be different. And let me posit something to you and we can explore this. You know, we know, for example, that we had really meager economic growth really across the world until England in the 1620s. And then in England, we cracked a sustained 1 % growth rate for decade after decade. And it had taken plagues, essentially, to give us better than zero growth. The one exception, perhaps, were the Dutch when they figured out trade and moved in the 16th century to a trading economy. But we have seen the economic regime fundamentally shift. I mean, at least once, perhaps twice, right?

4:07Well, perhaps The first time was when we moved from into cities. And the second time, perhaps, is that period in England in the 1620s. Why might we not have the conditions to do that again now? You can look at particular self-contained systems. One of the first we had was chess. AI in chess is virtually godlike. AI in programming is getting there. It will take a little while, but it becomes a kind of self-contained system. Maybe parts of graphic design. But all that put together is well below 10 % of the economy. And once you have to interact with more complex processes, we need to invent whole new ways of doing things.

4:45We may need an entire generation of institutions and firms to die off and have new ones born again that are built around AI at its center. And don't get me wrong, all of this, I think, is going to happen. But it's much more like a 20-year time horizon. And in the meantime, there are all sorts of tailwinds against economic growth. Well, we're aging, right? We have low fertility rates. Our governments are doing many stupid things, especially in my country. All that tends to lower the growth rate. You know, we've had the internet now for quite a while and maybe measured growth rates are somewhat higher because of it.

5:21But even that you can debate. There are people who'd say, well, productivity still hasn't gone up from the internet. It just takes much longer than people think electricity took 40 to 50 years. I do not think this will be that slow. For some obvious reasons, it's much quicker. It's just on tap, on your phone, on your desktop, wherever. But slow it will be. Yeah, I mean, I'm very sympathetic to that, right? So electricity takes 40 years from the Pearl Street generating station to really being noticeable in U.S. productivity statistics. I think the PC is a bit quicker than that. It's a 20, 25 year time frame.

5:59So we could imagine AI being faster. At the risk of us agreeing too much about this, I'm going to push hard on a couple of points and maybe argue as to why it might be different on this occasion. So you talked a lot about the institutional component to this, right? The fact that the institutions adapt very, very slowly. And, you know, we saw it with something as simple as the typewriter. It took 20, 25 years for the typewriter to make a difference. But I wonder, could it be the case that a generation of managers who grew up in the shadow of Clayton Christensen's book, The Innovator's Dilemma, for whom the word disruption has been part of their career, capital markets that move more freely, the scale of the demand pool from consumers for this, the fact that governments may be doing many stupid things, but also seem to be resigned in some cases and excited in other cases to build out the infrastructure as quickly as we can.

7:00Does that provide a tailwind that might see this moving quicker than that 20-year period? Well, I don't think our managers and consultants and capital markets are also great. They're probably the best in the world in the United States, and there's still plenty of mediocrity there. When was the last time the U.S. built a new nuclear power plant? What are the prospects for that in Britain? In both countries, it's looking up, right? That's good. Probably at the moment, if we start one today, it takes us 10 years if we're lucky. That's just a simple way of showing how much time things take. That's a self-imposed burden.

7:36I would call it an unforced error. South Korea, China can do it much more quickly, and it still seems safe. But I just don't think we're going to completely get our act together all at once. I see no signs of that happening. And if you look at the Trump administration, all these tariffs, you know, they may or may not stick. It depends on the courts. But they slow down trade in the materials to help this happen more quickly. Right. And I think that that's a really important point that is sometimes lost in Silicon Valley, which is that we're embedded in such a wide system where we've made assumptions.

8:07But in reality, the globalization's contribution to economic growth has been so enormous over the last 70 years, it probably dwarfs that of technological innovation. It's an institutional innovation that brought with it lots of technology innovations like supply chains and containerization and so on. And so deglobalization is a really big deal, right? It's a really, really big and distracting deal for companies and governments to have to deal with. So I'm talking your book again. That's how persuasive you are. I'm going to push again. Let's see if we can get this going faster. What about the notion that what we've actually seen in the last year or so has been an unlocking of desire to make those types of investments that the US and the UK found so hard, like nuclear?

8:59So the U.S. electricity system hasn't really grown since the days of Bill Clinton being in power. But now we've got a demand signal, and that demand signal is driving innovation on the one hand from, you know, SMRs and solar and batteries and even fusion. And on the other hand, it is pushing the states and federal government to say, we need to build this, we're going to take regulations out. Could that create a storm that drives things a bit quicker? Well, in this regard, our two countries have very different problems. We have extreme federalism. To build things out, the grid, you need approval from so many state and local governments, from the feds, from various environmental review boards.

9:44Again, it just takes years and years. Cleaning up that process takes years and years. I've heard it estimated that to get rid of a single federal regulation can take two to three years. Again, I'm all for doing that, and I think we're going to do a lot of it, but it will be slow going. Now in the UK, you have far less federalism. You have a bigger fiscal crisis. You have a talent crisis. For some set of reasons I can't explain, you just don't seem to be able to get your act together to do a bunch of things. I don't know why. I mean, I don't know what holds us back. I mean, there tends to be a little bit of a sense of malaise around the UK, peppered by real dynamism in sectors.

10:22So, you know, the technology startup sector keeps rolling up its sleeves and saying, let's go again. But it is dragging behind it a country that has been somewhat stripped of its muscle and its capacities over the last 15 years in particular. And we have a particular problem with our administrative state. That somewhat gets in the way of the kind of growth that, again, some of the people coming out of Silicon Valley will talk about. But we do also see this sort of wide variety of estimates of what AI would do from some of the investment banks suggesting, you know, a couple of points to GDP and Darren Asimovlu, the Nobel laureate, being much, much more tempered in his estimates.

11:03Where do you end up and why do you end up there? The very best estimate we have is that which is embedded in market prices. And there are plenty of market prices that indirectly give signals about what will happen. I would say if you look at currencies, equity prices, real interest rates, levels of volatility, really anywhere you care to look, the world looks pretty normal, at least for our two economies. There's some weird uncertainties surrounding Trump and tariffs. It's not related to AI. But people in the market who know these AI arguments, they think there may be an effect. NVIDIA obviously is a pricey stock, but they don't see the world being radically overturned anytime soon.

11:44And I think they're correct. If you think they're wrong, you're very able to go out there and bet against them, whatever your view may be. So far, there is capital. People who have real money at stake are essentially agreeing with the position the two of us have set out. I mean, my number is sort of a two and a half-ish percent, so a kind of growth of about half a percent on top of wherever we might be. But there are so many other confounding factors like trade war and political instability that it's hard to imagine. But even that is quite significant if you think about the scale of today's modern economy.

12:19What that means is a doubling by 2050 from today. We're agreeing too much again. And the exact number I've been giving people is an extra half percentage point a year surrounded by high variance. And again, over time, that is a big, big deal. And I'm fairly convinced it will be. But in any given year, you will notice all sorts of anecdotes like, oh, all my friends are using GPT as a therapist, which may not have major economic impact, but it is a sign things are changing. And then in some funny way, over a 20-year period, that adds up to a very different world. And that's what I think we're in for.

12:53We are going to get into that point because you wrote a great essay with my friend Avital Balwit from Anthropic, and I want to explore that. I just want to ask one more thought on the economic question, which is a number of people are going out today saying we're on the verge of a great job displacement. We've heard Joshua Benjo talk about it. Dario Amadai, who's a CEO of Anthropic, has been very vocal. A chief economist at LinkedIn had an op-ed in the New York Times saying particularly entry jobs in white-collar spaces are coming under pressure directly here and now. What do you think huge job displacement means in this context?

13:34What would, for you, feel like huge job displacement? I do not expect huge job displacement. There will be an enormous number of new jobs incorporating AI into the routines of established institutions and also building out new institutions. Now, I do think what will happen is those new efforts will be a little slow to get underway. And in the meantime, a lot of employers will be somewhat skittish about hiring more people because they'll say, well, ideally, I like to keep workers for five years. Maybe this hire will be obsolete because of AI. right. So for now, I'll wait. Reluctance to hire new people will outpace the creation of new jobs through AI.

14:13I'm not sure it will be noticeable in the aggregate numbers. I think it will be noticeable in job market vibe. I don't think many people will be fired. A lot of jobs will change. A lot of people will be replaced by others who have learned how to use AI. But I don't think we'll see a much higher rate of unemployment, at least assuming we do not make other stupid mistakes when it comes to macro policy. Right. But let's unpick that at the moment, because, of course, there is a high rate of employment at the moment in the U.S. sort of historically. So you've got this, you know, you have this tight labor market.

14:46And as an economist, you might look at an aggregate level of unemployment and say it's not really moving in a direction that is nerve wracking. But I suppose one question, though, is in the shift or the gradual destruction or the slowdown in hiring of jobs, does that have particular geographic ramifications? And is the creation of the new job something that takes a little bit too long to evolve, right? That effectively the economy, of course, can absorb and create all sorts of new jobs. But that's dependent on entrepreneurialism. It's dependent on businesses changing the way they hire. So I see these two little gaps emerging, one of which is that there may be places in the U.S.

15:30and, of course, the U.K. and other countries where the slowdown in hiring or whatever limited layoffs there are are quite concentrated. New jobs are created elsewhere, and they take a little bit of time, and that creates a sort of uncomfortable gap. I agree with that. I think parts of each of our countries where you have a lot of young, recent college graduates entering the market for the first time, that's where the negative impact will be the strongest. I have a decent sense where that is in America. you would know better about the UK. But that's exactly where I think it will show up. So where are those hotspots in the US?

16:03Well, there's plenty of people. Say you graduate from Berkeley, UC Berkeley, which is an excellent school. A lot of those people want to stay in the Bay Area. And I'm not sure they'll all be able to get the jobs they're looking for. And that will for them be a problem. Again, I'm unsure how much it will show up in the aggregates, but that's a place there's an outflux every year. They want to hang around. They're not, for the most part, looking to move like from Berkeley to Nashville. And it may be tough. The fact that we're hiring more carpenters, more gardeners, more salespeople, whatever, because we have more wealth, again, could lag behind those negative impacts.

16:38And I expect they will. Are we even measuring the right thing with GDP as we move into a world with AI and there's a greater substitutability of human labor for machine labor? I mean, there has been a long-standing debate amongst economists that, you know, in a digital world replete with intangibles and services and computers that get better every year but stay the same price, that GDP isn't quite capturing the purpose of the economy. But there seems to be such institutional inertia behind it. It's the measure we use. Are we even talking about the right thing when we talk about economic growth in GDP terms, or are we just talking about the convenient thing?

17:17I think GDP is an underrated metric. If you put aside the Gulf oil and gas monarchies, where people want to live is where GDP is high, and the intangibles are pretty closely correlated. Like take the 10 countries you'd most want to be born into. They're all high GDP countries, right? Now you could say, well, Singapore is a bit rich and they're a bit less happy. Maybe, you know, Finland, GDP is a little lower. They're a bit more happy. I'd agree with those points, but I don't think the differences are that large. So you don't think that in a world where we may see increasing share of national income moving towards capital, right, building on this trend that we've seen in the U.S.

17:57and the West over the last 50 years where effectively national income has about 10 points of it has moved from labor to capital. Does that trend continue in an AI world? And if it does, does it create issues? There are a lot of different issues in what you raise. The previous trend, it depends on whether you count Mark Zuckerberg as labor or capital. If you count as labor, but just a very well-paid laborer, there hasn't been much of a change. But I would say it's not clear that AI will concentrate more wealth in the hands of capital. The AI could easily become commodified. We have a lot of open source.

18:33There's potential competition from China. Those companies right now are not making money. There's a number of papers that show those who benefit from the AI service itself are often lower achievers, who now can write a good essay or write a good letter. I'm not sure that's how it will remain as time passes. I'm not sure it goes in the other direction. Sometimes I think the real winners are people who have bought high-quality land, because AI may substitute for capital and or labor, but it really cannot take the place of land. But I would just have a high degree of uncertainty as to where the incidents will fall.

19:07We should worry about this, but I would say not leap to conclusions. Yeah, well, that's back to the future if we're back to land. Actually, on your point about, you know, AI being a skill equalizer, a lot of the early studies that came out looking at large language models from people like Eric Brynjolfsson and Nao and Zhang and others did show that. They showed that, you know, people in the third and fourth quartile got bumped up to median performance. But quite interestingly, more recent studies have shown that if you look at the very, very high performers, the absolutely virtuoso in their field, they somehow are able to use the tool as a tool and achieve greater heights.

19:48And I think I see this when you start to see the instructions for the large language models that, you know, the CLAWs and the chat GPTs get leaked and they've got a prompt and they have this system prompt for these things and they're 10 ,000 words long. They're incredibly sophisticated. And I look at my prompts, which are two lines long, and I think there's no way I could get to that degree of sophistication. And that is about perhaps superstar talent being able to take advantage of the technology. And we see this in other digital markets, you know, so I can make music now as anyone can, but we have a greater degree of concentration amongst the superstars, and perhaps we did 50 years ago.

20:29I agree with that too, but that does not negate the fact that poorer people, they now have free excellent medical advice, free excellent legal advice, free excellent therapy. So I suspect, and again, this is a guess, but people at the bottom of the earnings ladder and education ladder, and then the superstars at the top, they will be the ones that benefit. And on net, it's probably egalitarian, but with both effects. Yeah, it's a complicated one. So let's turn to those social effects for the moment. You wrote a great essay in the Free Press with Abital Balwit about AI. And, you know, what's interesting about what our discussion about the economy is that we've ended up in this place saying, you know, it's important and it will take, but it will take time, right?

21:11We were imagining that time in 20 years, 25 years. But the pace and urgency in that essay in the free press was quite different. And I wanted to reconcile those. You know, essentially you said by 2030, AI could start to fundamentally challenge what it means to be human and cause an identity crisis as we lose our position to these entities that could be smarter than us. And I'm just curious about how we reconcile the first part of our conversation with what was quite an impactful essay. Well, the AI is already smarter than I am, even if you just restrict it to economics. If there were some kind of like economics Olympics and I was up against 03, 03 would beat me.

21:52That's not hypothetical. It's not the future tense. That's today. What's interesting is how little it matters. So demand for my services insofar as I wish to supply them, it's up, not down. And that's what will take a long time. But I would make the side point, that's a co-authored essay and we each wrote it. And Abital thinks the pace of change will be much faster than I think it will be. So the essay, you could say, is a weighted average of our two opinions. I'm fine with doing things like that, but she does think it will be faster. I mean, she's in the heartland of Silicon Valley, right? And one of the things that I found fascinating, actually, is the speed of iteration within the Valley.

22:32I haven't seen it as febrile and excited since the 1997-1998 period, but based on much, much more knowledge of how to build companies, how to iterate, how to test things. And the speed of learning there is really singing to San Francisco's ability to agglomerate all of this incredible talent. And so it's interesting to me that the people who are, in a way, best positioned to observe this change happen and are surrounded with the resources that allows them to take advantage of it, feel that challenge of meaning coming towards them faster than perhaps you or I do. That's absolutely true. And I largely agree with their predictions about the pace of AI improvement.

23:16But when it comes to the actual world, I feel they have underinvested in the social sciences and history. And I've had long extended back and forths with these people for now a few years. And I'll stick with my current view. I feel I win those debates. What is the metric or the number, the indicator, the data point that you would have to see to think, oh, I got this wrong? Well, if I saw, for instance, real interest rates make a big leap upwards for some reason other than higher budget deficits, I would think, oh, the market's telling us something. And then I would change my mind. Real interest rates are up a bit.

23:52It's probably mostly because of higher U.S. government borrowing. A modest amount of it is probably from higher CapEx expenditures coming from the AI sector. And that does reflect higher growth prospects for the future. It just does not seem extraordinary in terms of, you know, predicted impact. So unpack that for the non-economists who are listening. Why would a leap in interest rates signal some kind of a change? Well, if there's much higher productivity, whether coming from AI or something else, people want to invest a lot more money in whatever is being productive. So partly they're borrowing money.

24:27The demand for capital is higher. The opportunity cost of capital is higher because it could be used for more energy for AI or for the AI itself. And all that will mean higher real interest rates. And I expect that to some degree. It just seems to me within a quite manageable level of change. Yes. And there are so many other confounding factors, as you say, debt servicing and the ebb and flow of the rest of the economy. So we haven't seen what you would describe as a shift to a new interest rate regime just yet. So that's the indicator that we might want to look out for. But come back to this question of O3 being smarter than you now.

25:05Did you have a particular oh shit moment when you were using these large language models for the first time in the last few years? Was there a moment where you saw something that you hadn't expected? Well, the first GPT-4, that was my first moment. It's like, whoa. Now, much earlier, I had moments in my history with chess, which is extensive. I saw the advancement of AI in chess, and I had many moments much earlier, even before Deep Blue. But, you know, thinking about large language models, GPT-4, O1 Pro, and now O3 have been my biggest moments. And two of those are quite recent. GPT-4 now is a little, you know, two-plus years ago.