Hola, Bonad.
So, good evening everybody and welcome to Ateneu Barcelonès. My name is Andreu Warren and I'm an econ professor at the University of Barcelona and I'm also a member of the economic section here at the Ateneu Barcelonès. And today we have a great conversation ahead of us, and it is a good day to remember that this house, the Ateneu, has been for more than one century one of the places where Barcelona thinks out loud.
And much of that thinking has happened through something we call tertúlies. So for those of you who are not familiar with this term, a tertúlia is a cultural salon, which is kind of an open-ended conversation among people coming from different backgrounds and most importantly among people who don't necessarily agree with each other. Right? So in a tertúlia we argue, we have fun and then we come back the week after with sharper arguments and further discussion. So today this house is host to more than 40 tertúlies every month on topics such as history, poetry, philosophy, technology and of course, last but not least, economics.
Which brings me to our guest today, Professor Tyler Cowen. I learned about Tyler when I was a PhD student at the European University Institute. When I started reading almost every day before starting to work, his blog, joined with Professor Alex Tabarrok, Marginal Revolution. I always very much love the motto of the blog, which is small steps towards a much better world. So this is a very quintessential economist motto: marginal, incremental, positive-sum thinking. And Tyler's style of conversation, both on Marginal Revolution and on the podcast, has the same spirit. He's curious about everything, open to being persuaded and always ready with questions you may not expect.
That actually is exactly the same spirit that we have in the Ateneu's tertúlies and it's one of the reasons why it is so good to have you here today as a celebration of civil and open-minded conversation.
Moreover, this is also quite an interesting moment for the economics profession and for anyone interested in economics.
The last few years have been, I think I can speak in the name of most economists, have been quite disappointing for economists, mainly because on both sides of the Atlantic policy has been drifting away from our general advice.
But there's still a big question, and actually probably the biggest question, is still open and is up for grabs, which is AI. AI is definitely the most important challenge of our time. And while technologists are very good at telling us what the machines can do and what the machines will do, they are less good at telling us how people, how firms, how institutions will adapt and will respond to that technology. So I think economists and economic reasoning in general will have a lot to say, and at least we have a comparative advantage compared to the technologists.
So when I look around myself, for instance, I know that a lot of us are spending a large share of the AI-driven productivity gains working more and being more ambitious, aiming higher. I know that we are not probably a representative sample, but still it's an interesting data point.
So, Professor Tyler Cowen, we are very happy to welcome you to Ateneu Barcelonès and we are very grateful to the Ostrom Institute for organizing this event. Now, it's time to give the floor to Martí Giménez Mauschbach, VP and director of research at the Ostrom Institute. Thank you very much and I look forward to tonight's conversation.
Thanks a lot, Andreu, for the kind words. Thanks a lot for joining us. It's really incredible to see such a packed room tonight and I'm really honored to be introducing Tyler today. So my name is Martí. I am head of research at the Ostrom Institute. So we are a nonpartisan think tank. We are devoted to public policy research, outreach and impact for a freer, more prosperous society. And actually tomorrow we are celebrating our 10th anniversary. And we cannot think really a better way to celebrate our anniversary than with Tyler and also addressing such an important question. On my opinion, it's really the question of the decade, and also hosting it here in a house that has been discussing ideas since the late 1860s. So really, really nice to be here tonight.
So for two years Europe has asked what AI will do to us. But tonight the question is a bit the other way around. So what will we do with it and what will stay stubbornly ours to fix, however clever the machine becomes. So as Andreu was mentioning, we want to understand a bit more the impact on growth, productivity, the way we hire, the way we train, the way we promote, and that is really the modern economy, and our guest has spent his career trying to understand these dimensions.
Tyler Cowen holds the Holbert L. Harris Chair of Economics at George Mason University. He's the chairman and faculty director of the Mercatus Center, one of the most influential centers of economics and policy research in the United States. He sits on the economic advisory council of Entropic, and both Foreign Policy and The Economist have quoted him as one of the most influential economists of our time. He has written around 20 books, I believe. I will not list them now, but I'll just mention three of them: The Great Stagnation, which explained 15 years ago why the rich world had stopped growing the way it used to; Stubborn Attachments, actually one of my favorite books, which argues that we owe faster growth to the generations who come after us; and this spring he also authored The Marginal Revolution, which is a book written to be read with an AI at your side, so quite timely for today's discussion.
That is the formal part, and the informal part is a bit harder to summarize when it comes to Tyler. But we have already introduced Marginal Revolution with Alex Tabarrok. He also co-founded Marginal Revolution University, which has taught economics free of charge to millions of students around the world. He hosts one of the best podcasts ever, Conversations with Tyler, and he also runs Emergent Ventures, which is a philanthropic project built on an idea that a lot of institutions praise but almost none practice, which is: find talented people early, give them money quickly, and get out of the way. So quite an interesting initiative as well. Some of the people who received those grants in their dorm rooms now actually set the agenda in artificial intelligence.
If I had to compress Tyler's view of the world into two sentences, the first one would probably be that the problems that matter are rarely the ones in the headlines. And the second one that most things are getting better slowly for reasons nobody planned, and that's why we should pay such close attention to the few that are not. Actually, working on AI, he has been saying something I'd say quite unfashionable for the last few years. He expects it to add about half a percentage point to annual GDP growth. So not 20, not 10, but half a point. And he expects the reason to be mostly us. So ourselves, our regulators, our policymakers, our committees, our inertia, our habits. He calls that view optimistic.
He has around half an hour to convince us on that, and then you and I will try to catch him out on that one. So tomorrow evening the Ostrom Institute will present him with our prize for intellectual leadership and we're very honored about that. But tonight he is simply our guest. Please welcome on stage Professor Tyler Cowen. Thank you.
I have some good news and some bad news for you and they both relate to AI. So, we're in Barcelona. Probably many of you live here or have lived here. There's going to be a lot more tourism in the world. That's the good news and the bad news. So, I think of this trip. Now, I've been to Barcelona five times in my life. That means I like it very much. But I don't really know it well. And my last visit was 20 years ago. My wife has only been two times. So this time before coming, I open up my ChatGPT and I say, 'I'm Tyler Cowen. I'm going to Barcelona. What concerts do I want to see?' And it tells me that Atòmic is playing. So my wife and I, we went to see Atòmic. It says Friday night jazz festival. BJ is playing. My wife and I have tickets for a jazz festival.
We're in the hotel, very nice part of town. And I ask it, 'Where should we go eat tapas?' And it says, 'Oh, a 10 to 12 minute walk is Paco Meralgo from your hotel. That's the best place.' It was very good. Excellent seafood tapas. So the trip is much better because of artificial intelligence that I can ask it anything about Barcelona and it will do better than I could do on my own even though I've been here five times. So this is good news for me. For all of you it's sort of good news and bad news. Like I'm here, I'll spend money, but there's also arguably too many tourists in Barcelona.
You need to tax us. In fact, right, if there's too many of us, tax us, but you're not taxing us, so it's too crowded. But one side effect of AI, it's not talked about much, but there's going to be a lot more tourism in the world and especially from the United States. All those people who work in Anthropic, OpenAI, they are rich. They're having what we call IPO. You know, the companies are going public. They will be rich in liquid. When I was young, like wealthy Americans, they would like visit Los Angeles. If they were radical, they would fly to Hawaii. Now, we come to Europe. You go to southern France, Michelin restaurant, everyone's an American. It's sort of crazy. I don't myself like it. It's sort of bad for me. Like I want Europe for myself, but I don't have it.
So AI is going to make this much worse, but it does mean a lot more wealth, a lot more opportunity, and a lot more ideas spreading across the globe. So my wife and I today were in the museum, the Catalunya — phenomenal museum — and there's many things there you can only see in Catalonia. You don't see it in New York or London. That's why the museum is so good for us. So we see many new things we don't know. You can take a photo. You ask GPT or Claude, 'What's going on in this artwork? Tell me the history. How did it get here?' Whatever. And again it tells you. So more tourism, more flow of knowledge, and again I would say mostly a good thing, but it's just one small way that AI will change our world. It's quite indirect. People don't talk about it much, but over time it will be a pretty big effect.
And in my view, if you stack up all the different effects of AI, 20, 25 years from now, our whole world will be entirely different.
We have a phrase in San Francisco. I don't live there. I'm not one of them. Like, I feel more at home in Barcelona than in San Francisco. But the word they use is 'AI pilled.' They say, 'Are you AI pilled?' If you're AI pilled, it means you realize the world's going to be very different. If you don't realize that, you're not AI pilled.
Just another small example of how the world will be different. Again, I think all the small examples add up to a lot. But take something like religion. I think AI will change religion a good deal. Right now, if you have a question, say about the Bible — well, not now, but say five years ago — you would go to your priest, your rabbi, your reverend, depending on your religion, your imam for the Quran, and ask, and they would know. Today, you don't have to ask them. The entity in our world that knows the most about the Bible, the Quran, the whatever, it's Claude and ChatGPT.
So the religious leader is no longer where you have to go to ask questions about your religion. The AI intermediates your religion. I don't think we know yet how religions will change. They might become lonelier, or maybe the priests and rabbis will lose status, or people will connect in other ways. Some people, you see this already, especially in San Francisco, they start treating the AIs like they're gods or maybe oracles, because after all, they know more than anyone else about anything else. So our whole perception of a relationship to God, to a church, what it means for us to be human, all that will change in ways that I think are very difficult to predict.
Let me speak for a few moments about both economics and politics. And here what we're getting back to what is clearly good news. So if you look at most Western economies, we have aging populations, low birth rates, even in the United States. Typical family now is 1.5 children, right? That's higher than most countries. It's higher than most of Europe. Now it's higher than most of Latin America. Maybe you think in Spain you get all the Latino immigrants. It's for me a great idea. The birth rate in Argentina was just measured at about 1.0. Mexico 1.1. This is astonishing to me. I remember a time when in Mexico seven children in a family was common. And now the average is a little above one. So how many Latino migrants will come to Spain? I'm not sure. But it's not as many there as we used to expect.
So the same countries, United States, most of Europe, Mexico, Argentina, we have big budget deficits. We pay older people a lot in terms of pensions. And our main political parties, they don't want to cut the pensions. Older people vote more often. They vote to keep the pension high. You look at France, you have a radical left-wing party, radical right-wing party. They compete against each other to say we're not going to cut the pensions. So there's slow growth, not enough children, high pensions, budget deficit, common problem around the world. Also in Spain, what are we going to do? I will tell you there's plan A and there's plan B. Do you want to hear plan A? Plan A is that AI will help us grow more quickly and we will manage to pay our bills. That's plan A. It's not that bad a plan. Might not work, but like it's a plan. Do you know what plan B is? There is no plan B.
So, it's plan A or uh-oh. Plan B you could call uh-oh. A lot of countries become like Greece. Like right now, real wages in Greece are one-third of what they were before the Euro crisis still. And the country went through great pain. So, you can't just keep on raising taxes because you slow growth and you don't solve your budget problem. Cutting benefits politically seems impossible and you'd have to cut them a lot. So, we're left with plan A. Plan A had better work for us.
So, I'm on record as predicting for the United States that AI will raise our economic growth rate by half a percentage point. So, let's say normally we would grow 2.2%, 2%. I think now we'll grow something between 2.5 and 3%. Maybe a little more. I'm not sure. Half a percentage point. It sounds very small, right? But half a percentage point difference compounded on top of each other over 20, 30 years, it makes all the difference. So if the United States can grow half a percentage point more, that means our debt-to-GDP ratio, our indebtedness, it converges to a stable situation. We don't have to raise taxes. We don't have to cut our pensions. Country keeps on growing. It's normal. It's pretty good if we can grow half a percentage point more.
So I think with AI we can make American businesses more productive and do that. But again that's a speculation. It's not proof. The latest growth report from America in fact was quite good. That's one nice piece of evidence but it's only one piece of evidence and we will see. But again I stress the point. There is no plan B. So for Spain, I know much less about Spain than my own country. But I believe in Spain because of AI you can grow more rapidly. I don't know how much. Maybe it's a smaller impact than the United States. It's possible and you don't really have a large AI sector of your own. But I'll just say that's your plan A. So root for your plan A because you like us have no plan B. So this is drama in the world. And as I said, I'm pretty optimistic.
I have some friends, friends in San Francisco, again, they think because of AI, like the American economy, two years from now, we'll grow at 20%. To me, that's kind of crazy. I don't think that's possible. I think there's too many bottlenecks. Energy is scarce. Government regulation might intervene. You have a lot of institutions, our government, our universities, our nonprofits, a lot of our big companies, they're very bureaucratic. They're not really going to enthusiastically adopt AI that much in the next two years. So, I think 20% growth for our economy, your economy, that's just impossible. I don't expect it at all. Most economists I know don't expect it. But there are people in San Francisco who call themselves AI pilled and they expect it. We'll see.
There's — I get emails from them. They're like, 'Tyler, do you want to make a bet? You want to bet $10,000?' I don't trust them to pay me back. They think like in five years there won't be money and it will be like a Marxist utopia. There'll be, you know, tapas dripping from the trees for free everywhere. I think that's crazy, too. But I do think we can be more productive. And I suspect many of you here, you work now with Claude or GPT. You are personally more productive.
But what we have not done is make our institutions more productive. So people are freelancing with AI. You have a job. You need to write a memo. You say, 'Claude, write that memo.' Claude writes the memo. You're done. It took you five minutes instead of three hours. That's great, but it doesn't add up to that much in the aggregate. What we need is for the whole institution, all the processes, the budgeting, the finance, the HR department that hires people, the information flows, keeping and storing the data, all of that to be integrated with AI. And that will happen pretty slowly. I think pretty slowly, but still I think we will get higher growth.
One thing we are seeing is the rise of new companies that have very few humans and they're using AI. I call them AI-native companies. Often they're run by teenagers. Very often they're run by immigrants. They're not run by the establishment. They're just smart people. They taught themselves AI. No one at Harvard taught them. They never went to Harvard. And there'll be like three people, five people. And they have maybe some software product they sell to other businesses. And their finance, their hiring, their recordkeeping, their legal department is called Claude. It's like one lawyer, Claude, man, woman, who knows? It's Claude, right? That's their lawyer. They save a lot of money.
These businesses are starting to be very successful, just like three or four people, high revenue, low costs. So slowly these businesses with few employees and a lot of revenue, if we use AI for kind of everything, they will transform the American economy. I don't know how much of this you have in Spain, but I'm seeing it at different places around the world. A new thing. I call the people who do this the AI maniacs. They are the billionaires of the future because their companies will have such low costs and will be so productive. It's a new mode of running a company that the world has never seen before. And it is possible because of AI.
So that half a percentage point of growth, a lot of that will come from these companies that are small with respect to number of employees but pretty big in terms of revenue. Now some people worry if these new companies have so few workers, like where will the jobs be, right? So that's a very good question. I'm an optimist on jobs, but I think jobs will change a lot.
So in the near future, every job, almost every job, you will have to work with AI. Even if you're like a plumber, a gardener, you'll have to know something about AI. Certainly, if you work in an office, you'll have to understand how to work with AI. It will be like the internet today. How many jobs do you need to work with the internet? Close to every job. Even if you're a basketball player, you know, you make a good play, you want to put a clip on Twitter. That's working with the internet. So you have more fans, sell more tickets, whatever. So pretty much every job now, you have to work with the internet. We're coming quickly to a point where every job you will have to work with AI.
Most jobs won't disappear because in the job you do things in the physical world, like I'm here. This is me. This is Tyler. You did not invite ChatGPT. But I will tell you the remarks of ChatGPT, they're not bad. Like his talk, her talk, whatever, it's not worse than mine. But there's something about you want something in the real physical world that ChatGPT and Claude cannot do. So there'll be human jobs for a very long time. They will involve doing things face to face, persuading people. A lot of it will be working to improve the AIs.
Just two sectors where job growth will be extreme. The first is biomedicine. So the AIs already, they have so many new ideas for drugs, medical devices, treatments. Probably most of those ideas are bad. A few will be very, very good. We need to test them. To test them, you need humans. You need humans in the lab. You need humans to take the drugs. You need humans to record the effects of the drugs and so on. So the biomedical sector will grow to be enormous.
I've made this prediction which I will make here. I would say if you're below, I don't know, 50 years old, 45 years old, I think you will live to be 120. We never expected that before. So I think with AI we will generate an enormous number of ideas for cures. We'll find all the cure ideas that there are pretty quickly. It'll take us a long time for the humans to sift through them and see which ones are good. But we will do it, not in time for me. I'm 64. Sorry. Like what's the cutoff point? I don't know. Like but don't buy a motorcycle, you know? Be careful. Exercise.
Spain already, you have the longest life expectancy in the world. I think that's great. So, a lot of you are going to make it to 120. America is like, we die too early. You know, that's bad. I feel bad about that. But, congratulations. That long life expectancy, it's going to be worth a lot more than you had thought. So, walk around town, keep on doing the olive oil thing, whatever. Don't buy a motorcycle. Like you're a blessed people.
So so many Spaniards will reach something like 120. I don't think you can live forever, like you'll die of old age or maybe your brain just stops working. But I think most of the things that kill you, like cancer, Parkinson's, whatever, we will fix in the next 30 to 40 years and they won't kill you. That to me is a remarkable future. It's something AI can do for us. Now, it does mean these pension problems can get worse. I hope you'll be more productive and those extra years will be productive years and you won't be able to just retire at 60 and be retired for 60 years until 120. Like that's not going to work. We're going to need to revisit that sooner or later.
But I think this future will be very different, and in a way where the biggest beneficiaries are the people who will live the longest. And that is exactly where I am now in Spain. Right? You're number one for life expectancy. Pretty awesome. So you can be richer, have higher GDP growth, and live to 120 if you're below, I don't know, again, I'm just guessing, 50 years of age. Your kids or grandkids, they'll grow up in a world so different from the world you grew up in. I hope they like it, but again, predicting the future is hard, but with more intelligence, we just can do a lot more things.
I sometimes like to say that if you do not think we can handle more intelligence, we're kind of in trouble anyway. You know, like they call it artificial intelligence because it's smart. So North America, Europe, the whole world, we're getting more smarts. This is a big challenge for us. It's actually not easy to handle more smarts. But again, if we're not good at that, what is it we're going to be good at? So I'm still pretty optimistic.
I think for very poor parts of the world like most of Africa, you might see some of the largest gains. There's many parts of Africa it's hard to visit a good doctor, get medical diagnosis. It's not hard in Spain or America, it might cost some money, but there's doctors all over. But if you're in rural Kenya and you're sick, you just speak to AI your symptoms right now, you can get medical diagnosis as good as from, you know, the best doctors in the world. Or let's say you need legal advice or you want advice on how to run your business or you need help with organizing some records or teaching your children something. The places that right now have talent problems, education problems, in relative terms, they should gain the most. And that for instance would include Africa, the poorest parts of India.
I think especially the poor countries that have good internet connections are best situated. So my wife and I, we were in Kenya a few years ago. And this is rural Kenya. I don't just mean Nairobi. The internet was better than back home in the United States. We have like okay internet. It's not great. It was incredible in Kenya. So that makes me more optimistic about Kenya because connecting with the AI will be easy. Most of Africa right now is not using AI. I was in northern Ghana not too long ago, very rural, like the biggest city in northern Ghana, it just has one traffic light still. And I went around, I was asking people, 'Do you use ChatGPT? Do you use Claude?' And they looked at me like, 'What's that?' And they're not using it yet, but it is spreading rapidly.
And you know, even at home, we use it as our lawyer. We often use it as our doctor. So my wife and I, we have a dog. Dog's name is Spinoza. Yes. Spinoza, you know, the philosopher. And if Spinoza has a problem, we don't bring Spinoza to the vet right away. That's the old world. We ask GPT, like Spinoza has a rash. You take a picture of the rash. You ask, 'Do we need to take Spinoza to the vet?' ChatGPT basically says no. So we save money. Like ChatGPT is vet number one. The plan B is the human vet. Like there's at least a plan B. But you save a lot of money by having a plan A.
Not too long ago, Spinoza ate some raisins. Raisins, they're supposed to be bad for dogs. He's fine, it turns out. But you ask AI and again you get an initial diagnosis. AI said, 'You'd better take him to the vet.' So the vet gives him something. He throws up the raisins. Now he's fine. So we took him to the vet. It told us to take him to the vet. So this is already a kind of revolution. It's not really recorded in statistics. It saved us one trip to the vet with the rash. It mandated another trip to the vet with the raisins. Might have saved Spinoza's life or at least made him healthier. Avoiding that first trip saved us, I don't know, $600 for like nothing. And it's pretty great. So like that's now, and the AIs are going to get better.
Now, that all said, the world is getting very weird. It will disorient us. Our jobs will change a lot. Just close with a remark or two about short-term risk. The biggest short-term risk is just cyber attacks. So with AI, cyber offense has outraced cyber defense. So you probably saw around Europe all these refrigerators in grocery stores. They turn it up to 50 degrees centigrade. Is it some like evil terrorist? Is it Putin wanting to send a message? Is it Iran wanting to send a message? Is it like random? You know, I don't know. I don't know if we know. But of course it's bad. They tried to attack a blood bank I think in France that was like stopped, but it was in the process of happening.
So you need good AI for cyber defense. Right now US government has said to Anthropic, to OpenAI, you cannot give people the best cyber defense. The government just doesn't let those companies distribute it because we're worried if the best cyber defense gets into the hands of China, Iran, North Korea, whoever you're worried about, that they will use that to develop even better cyber offense. I think we need to bite the bullet, distribute the best cyber defense, otherwise we're just going to have hack, hack, hack, hack, hack, one after the other. For the next year, I think there'll be significant hacks every week, maybe almost every day. The North Koreans stole six billion dollars from Binance. That's a very tech-savvy, sophisticated company.
So, some of the short-run AI is going to hurt. People will be afraid of it. It's not always going to feel good. Long-run gains are pretty high. The short-run gains, well, part of that is my tourism and this healthier dog we have at home. So, there's many things AI can do for us. It cannot really give meaning to your life, but for a large number of practical tasks, it's really going to change everything we humans do. Those are my basic remarks. And with that, now I sit down and field some questions. Thank you all for coming and listening.
Thank you, Tyler. Can you hear me? Is it working?
Okay, great. I think this was really insightful. I'm sure there's a lot of questions from the floor, but I'll take the opportunity to start out with a few from my end. So, the first one is around diffusion of innovation, something that you also alluded to, right? So, a few days ago, we lost William Nordhaus, who did a lot of research on how new technologies spread and how also badly we measure them. So for AI to accelerate the current baseline growth rate you just mentioned, it cannot just exist in a few localized tech hubs, you need to spread to the physical real world where there's still a lot of inertia and bottlenecks, as you said, right? So life sciences or healthcare, the sector I'm in, but also manufacturing, government. So how do you see that interface between the LLM and the AI world and the physical world? Do you see the mix between LLMs and biosensors or humanoid robots to be the future in that sense, or how do you do the connection between the high, fast adoption sectors and the more slowly adoption ones?
Humanoid robots, humanoid robots, they're pretty far away. In the United States, AI is spreading pretty rapidly, I think faster than anywhere in Europe. So just about a week ago, I gave a talk. I was speaking to a man there. He runs a roofing company in Missouri. So they make physical stuff. You would think, well, if anyone won't benefit from AI, it's someone running a roofing company. And he said, 'No, we use AI for everything in the factory to make our asphalt, to have more efficient processes.' And he said they're blowing away their competitors because they're using AI more effectively than the other companies they're competing against. So we're just seeing in at least the dynamic competitive parts of the American economy a lot of AI usage.
But we have major sectors like government, nonprofits, education, where AI is either not being used or not being used effectively. There's a lot of freelance use of AI. You're in a company, you're in a government, you say, 'Claude, write this memo for me.' But you wouldn't say the institution is using AI. So, I think like 40% of the American economy is kind of doing nothing or just getting frustrated, but the competitive parts of America are using it with joy and doing great. So, you know, about Spain, it's harder for me to say, but I do know Europe in general is behind the pace and fewer people here are AI pilled.
Great. So, you've also mentioned the concept of or the possibility of, you know, billion-dollar tech companies run by few employees, and I guess also with a lot of AI agents on the side, that I guess is due to reduction of transaction costs like Coase, right? But at the same time within a firm you could also have a reduction in the coordination cost that allows the firms to really expand and grow. How do you see that balance? Do you see we're going to have both sort of a lot of companies becoming mega companies and at the same time a couple billion-dollar companies run by a few employees, or how do you see that coming into play?
I think if you measure by revenue, some companies will grow much larger. I'm not sure they'll grow larger in terms of employment, but the companies that use AI best will just open up significant leads on their competitors in a way that had not been possible, you know, five years ago or really even two years ago. So there'll be an increase in the variance, would be the technical term. I wrote a book about AI 15 years ago. The title is Average Is Over. Companies that use AI on average will do very well and companies that don't will just fall behind and be outcompeted. So we'll see a lot of like clusters or clumping of AI success. But other places will fail at competing. So I think that's the way to think about it is higher variance.
Okay. While you think, and we have a few already, but I'll just take the chance about one more. So of course economic growth is, I'd say, partially driven by talent that's sort of highly skewed IQ on the distribution, whereas I'd say politics is mostly dependent on the median usually. How would you, do you see AI changing that dynamic at all? I guess people can also start using LLMs to ask on policy. Yeah, maybe I'm just curious about your thoughts on how cognition being now less scarce, how that will affect both on the economic side but also on the political side.
It's very hard to predict politics in any country, even your own. But I would say this, if you ask the best AI questions about economics, politics, you get very sensible answers. You don't have to agree with all of them, but it's better than the New York Times, better than Twitter, better than whatever on the internet, better than your favorite podcast host. It's just very good, and it's mostly pretty centrist, pretty moderate. So, I'm not sure how much people now are listening to their AIs, but it should make them more moderate and less radical.
In my country, we just had a big election yesterday. The results were quite moderate. So, people strongly preferred the Democratic Party candidates, but they rejected a lot of the far-left policy positions. So they're saying like, 'Well, Trump for us was too radical. But what the left-wing Democrats want to do, that's also too radical.' I'm not saying this is because of AI. I don't really have evidence for that. You know, part of me hopes AI had a role in that. But so far AI is just making people smarter and wiser and more sane to the extent they listen to it. It doesn't get them angry. It tells them how great they are. You know, there's worse things than that.
Okay, so I think we have a few questions. I'll trust someone to manage the mic.
Yeah, but we'll start over there on the third row.
Right, I'm going to try to ask a very Tyler Cowen question. Would you rather live in a world where the fertility rate is 0.5 and the GDP growth is 5%, or a world where the fertility rate is 2.5 and the GDP growth is 1%? And I ask this because I wonder if the question should be less how do we balance our federal budget and more how do we regain what technology has taken from us and maybe led to this fertility crisis.
I don't know how to increase fertility rates. I know that Hungary has tried. They spent a lot of money and mostly it failed. Which world I would prefer depends on whether or not I'm already born. But I think the reality is we need to do both of those things. We don't need 2.5. That would be fine. But, you know, 2.1 would be acceptable. The United States can worry the least about this because we can always attract very high-quality immigrants.
Some other countries can do this, but a lot of them cannot. Spain is not in a terrible position. You can get pretty high-quality immigrants from Latin America because you share a common language. But if you're say Armenia, I was just visiting Armenia. Three million people, many leave to earn more money. Birth rate is not high enough. Who can they get to go to Armenia? The language is hard to learn. It's not, you know, even the same script as we have. So they have a much bigger fertility problem.
than we do. You know, you can blame technology for it, but I think the real issue is people have decided having children is not that fun. I hope we can persuade them otherwise, but I'm not the ideal ambassador in that cause. I would just put it that way.
You just recently interviewed Luis Garo in your podcast, and he has a take that Europe can be the smart second mover, that it can just focus on the application of AI and basically forget about making foundational models. Do you agree? Do you think this is basically Europe just being dependent on US and China open sourcing whatever they want?
Well, right now I don't think Europe has a choice. There's not a European company really in the running for having a good foundation model, and there's many obstacles. You would need a more integrated capital market. It would have to be easier to fire people. You would need more of a startup culture that encourages the young and deemphasizes seniority. You would need to enjoy your leisure time less. That's hard in Spain and Barcelona in particular. You'd need different forms of deregulation, and you would have to play a lot of catch-up. I suspect all that together, while I'd be happy about it, is impossible. So, I think you will be more dependent on foreign models. I think you will be more dependent on the United States.
I'm not sure Spain will like this because of your western location in Europe. In military terms, you've depended much less on the United States for a long time than Poland, Germany, United Kingdom. Those countries have been dependent on the US a long time. They don't always like it, but in fact, they're fine. And Spain will become more like them. You'll be more dependent on America. I apologize. It will feel bad. But I don't think you have to end up worse off. There might be foreign policy stances you take. You'll have to modify or give them up. And it's not just a Trump thing. Trump is a symptom. The next set of presidents, they will use AI as a kind of tool to twist your arm. So yeah, you won't be that happy, but I think, you know, Germany, for instance, has lived under this since World War II, or Japan has for that matter. They depend on the US for protection. So you'll be more like that.
Hi Tyler, I would like to ask you about AI and fertility rates. You often mention in interviews that AI will reshape the job market, and especially the white collar upper middle class jobs, and you say that parents often, they're anxious, they ask you how to prepare their children for this age when AI will reshape the job market. Do you think that if this fear about the future of work spreads, will it affect negatively the fertility rates? That's one question. Second question, obviously the US is at the spearhead of technological innovation, but I think it still is a spearhead when it comes to political innovation as well. In this decade, we had two political experiments. One was Operation Warp Speed. The second one was DOGE. While the former was a success, the latter was rather a failure. But what do you think are the lessons from DOGE for Europe in the context of AI and regulatory barriers that surround it?
There's several different questions in there. One reason Warp Speed worked well, like everyone was against COVID. So that was not controversial, and we just pushed on it, and it worked, with help from the United Kingdom. DOGE, most Americans don't really want to cut government. Whether or not you agree with that position, even Republicans, they might say, but they don't really want to. So you try to cut government, it's unpopular. At best, it's divided, maybe 50/50, but usually it's unpopular. So you try to do something unpopular. You rely on this one guy, Elon Musk, who's super busy, like running five different companies at once, and he is himself, let's say, unusual. And if it doesn't go well, maybe that's not a big surprise.
Now, will people not have children because of AI? I don't know. We don't understand fertility rates very well. I'd be surprised if AI discouraged people. Like sometimes you hear people, like they've heard, oh, the climate change, I can't have kids. Like I've heard that. I don't think in the aggregate it has really mattered, but you hear, I think we'll hear it with AI. Some of my San Francisco friends, they're like, 'Oh, the robot nanny. The kid will be so easy. Everyone will have three kids.' I don't believe that either, but I really don't know. I'm not good at predicting fertility rates, and there's a lot of social consequences of AI that I don't think anyone can predict. So, it could as easily go up as down. The nice thing is we'll have more wealth that, to the extent subsidies matter for children, will be able to afford them better. I hope that helps us. I'm not convinced it will. Again, I'm familiar with Hungary spending 5% of GDP. They get a good two or three years increase in the birth rate, and then it goes back to where it was. Didn't really work, or maybe worked just a little. I wish I knew. It's one of our biggest problems, maybe our biggest problem, and I don't know anyone with a good answer.
There's been talk about the possible end of work due to AI, like the robots will take all the jobs, or at least there will be job market quick displacements. So what's your take on universal basic income? And also there's a criticism of technofeudalism that all the wealth of AI is being concentrated in a few hands.
Two different questions. First, any major tech innovation, most of the gains go to the users. It's true for language, for fire, for the printing press, for penicillin, and AI I think will be the same. So if you look at the market valuations for Anthropic, OpenAI, they're pretty high. Like Anthropic might hit two trillion. That number sounds crazy, but it's not that high. In fact, not compared to the overall aggregate benefits AI will bring. So you'll have a lot more billionaires, but the general citizenry, including in Africa, all over the world, will be much better off. So I don't think we'll get, you know, technofeudalism. And furthermore, the government may step in and just regulate AI more, or take shares in the AI companies, and do things to change the structure of what is happening.
And in terms of jobs, I think humanoid robots are far off. There's no company making humanoid robots that right now is worth a lot of money. Do I think someday it can happen? I do. But it's not pending. What is working for us is intelligence, not humanoid robots. I think factories will be mostly some kind of robot, mostly not humanoid, just automation, which we've had for a long time. So the United States used to be well over 35% manufacturing jobs, and we went down to about seven, eight percent. That's a huge change. There were problems, but US is still a rich country, an orderly country, kind of normal American, whatever. So, we're going to go from like 7% now to 2% or 1%. Again, there will be some losers, but it will be a pretty small shift compared to what we've already done. So, I don't think it will be massively disruptive.
When humanoid robots do come, whenever that point may be, I actually think governments will ban them. The people won't want it. Again, hard to predict, but I think it will creep people out to just be dealing in a world where these humanlike things walk around and are also very smart. But it's far enough off. I just don't really have a prediction one way or the other. But it may be like the Blade Runner world. You only let them work on other planets, like on the moon, and you ban them from Earth. That would not surprise me at all. Whether you think that's a good idea or not, that's kind of what I expect.
So, I have a question, and it's like the two branches of AI in the future. The first one is like these corporations like OpenAI or Anthropic, and they have like this big infrastructure, and they sell us like this service. I don't know, I pay $20 or 100. And the other one is that we have like better computers, more RAM, more storage, more memory, and we have our own local offline AIs. What do you think that will be like? This shift, will it be like more weight on the centralized and local offline AI, or do you think that it will win, this centralized service as infrastructure?
So far the trends we have seen are greater decentralization. So you have numerous companies with very good AI. It's not just two. And then there's Chinese AI. And over time, most of it will get better. Maybe some of it will go away. So I think, I don't know, within 10 years there will be a world where you won't have to know anything about computers or programming, and you'll just talk to some box, maybe just talk to your wall, talk to your refrigerator, and say what you want, and it will happen for you. In technical terms, that is more or less possible now. It's not integrated into real world systems or software very much, but there's not anything stopping it from happening.
So, right now, you know, you have these agents, you have to manage the agents, maybe you check on them. Sometimes they do the wrong thing. You wake up at 3 in the morning, you check on your agents, you set them straight. I think that's a very temporary state of affairs that you have to manage them, and the agents will be able to figure out what it is you want. And again, you can just issue commands and create what you want in information or recordkeeping or entertainment or answering your — it won't even be email anymore. It'll just be a very unusual situation where you will have extremely powerful command over knowledge and information at will. A lot of people, frankly, won't know what to do with that. So I think the people who will just take the initiative to do something with it at all will do very well in life, and some people just won't do much with it. In the same way like the Amish don't do much with driverless cars, they still have horse and buggy. Again, I just think it will be a very different, very strange world. But if you want to, it will be quite easy in a decentralized way to use all this AI to your advantage.
Hi. I do work on the Spanish, in Barcelona, AI tech small bubble. So, yeah, you're right to say it's very small, and it's mostly building products on top of the large foundational models. So my question is basically, do you think there's space for small AI companies to really capture value, or most of the value is going to be captured by these foundational models? And this connects with my second part of the question, which is, Elon Musk recently said that the big bottleneck that we're probably going to be faced with late this year or early next year is that while we are still producing chips at the rate we need for more compute, creating new energy sources and energy infrastructure is not happening at the same rate, and we're probably going to hit a wall soon. I wanted to know a bit of how you think about these two ideas and how this can slow or not the process and maybe change the structure of how we build AI instead of only focusing on these big foundational models, or not.
I'm very optimistic about European AI companies. It's common practice in the United States to look down on European AI, European AI efforts. I think human capital and talent here are very high. And look, you can take the best foundation model, Fable or whatever OpenAI is going to put out very soon, give it to a company, the company has no idea what to do with it. They are just totally baffled, I assure you. And OpenAI and Anthropic, they've been creating these teams of people to go around and tell the companies. Like the teams don't really know what to tell the companies. We haven't figured this out yet. So, I think there's amazing room within every part of Europe to have intermediary companies that bridge what the AI model is and what the businesses actually need, in a way where I think Europe, most parts of Europe, will be super successful at that. And it's happening now. So I'm very optimistic there.
Now on energy, again it depends a lot which part of Europe you mean, but I think Norway and Finland in particular, and possibly the United Arab Emirates, assuming there's a stable peace in the Middle East, they will supply a mix of energy and compute to the rest of Europe, and that will go fine. It will take some number of years, but not take an amazingly long period of time. It does require some buildout and some extra effort, but I don't see anything impossible in that. And there's other European countries that are possible candidates, and other places in the Middle East, the Saudis have plans. Again, who exactly will succeed, you can debate, but there's a lot of money in this. And I think it will happen, and it will create jobs, and there's places that want to do it. It could help Northern Europe and the Nordics quite a bit. Colder weather can be an advantage.
And then there's the question, what happens with nuclear power? That seems to be making a comeback. What happens with solar power in space? Elon is bullish on that. I don't feel I can judge that question, but at least it's something that might happen. So there will be temporary energy bottlenecks, but I don't think there will be very long-term energy bottlenecks, even for Europe, where energy problems tend to be endemic and in a lot of European economies energy is quite costly. But ultimately, this gives you all the more reason to solve those problems. So I'd be cautiously optimistic on that, too.
Hello, thank you for your presentation and your talk. What do you see as the biggest risks and downsides of AI?