OpenAI - Fireside Chat with Tyler Cowen and Alex Tabarrok

So one of the fears I hear most often is that strong AI or AGI is just going to put everyone out of work. I have a much more optimistic view than that, and I think most economists do. And one way to think about it is just to realize AI creates many, many jobs, even though it will take away some jobs. So one of the neatest properties of current AI models is they allow a small number of individuals working with AI to really do a lot more work than was possible previously. So this will mean more companies, more projects, more nonprofits, just more ventures, more attempts to entertain people. And all the new things that will be done by humans working with AI will create jobs. So I think the view both Alex and I hold, is that assuming our government does not mess up other policy matters, but we think we can basically keep full employment for the indefinite future. Uh, just a few areas where I think AI will create or already is creating a lot of new jobs. One area is generally energy, electricity, the grid. Alex himself has written a lot on the United States electrical grid. It's completely screwed up. It will take 20 years, 30 years, 40 years to fix. I'm not sure we'll ever fix it. The AIs cannot do that on their own. That's a good deal of jobs right there. The biomedical sector and medical trials, there will be many, many, many more ideas to test. AIs will help with the testing, but I don't think pure testing by simulation will be possible anytime soon. And in any case, the law requires testing on actual humans. —so that's another growth sector. Simply care for the elderly. There will be robots, personal companions—we have this already—but the elderly also will want human care. It wouldn't surprise me if in the future, you know, 15-20% of all jobs were elderly care. That's a growth sector, will not be taken away by AI. Louis Garicano had an excellent online essay. He referred to what he called messy jobs. Jobs where it's hard to explain exactly what the job is, but on a given day you're doing 11 different things and it requires coordination and figuring out what you ought to do next and getting other people to help you. So Luis Garracano says, tell your kid to go into messy jobs. They're hard to describe, harder to name, but again, there's a real future in messy jobs. I think just government lawyers will be a very significant growth sector. For employment. We need to write a lot of laws about AIs. I don't even think they'll all be good laws. And probably like many of you, I think the AIs will often, maybe always, could do a better job writing those laws. I just don't think we'll trust the AIs to be writing all those laws, even though we probably should. And it will be humans. The humans will work with the AIs. The AIs may even do most of the work. But there will be humans with oversight and a kind of guarantee, or you could say liability cushion. Cybersecurity, this is discussed quite often. Compliance as a sector. I think just niche entertainment. So there'll be a lot more music, art, sculpture, whatever, you know, cinema produced. Again, a lot of the work will be done by AI, but not all of it. The market will be very fragmented. Costs of production will will typically be much lower than they are today. So a lot more people will be able to earn livings working with AI, doing niche entertainment, niche jobs, jobs you don't think about. So like tracking down people's lost pets, it's all sorts of weird jobs. I have a running segment on the Marginal Revolution blog called New Service Sector Jobs. You can type that into the search function and there's about, I think, 150 of these posts. All the new little weird jobs that pop up when a society gets wealthier, you don't really think of them much in advance. If you go back in time, the great economist David Ricardo, one of the smartest humans when he was alive, in 1817 he wrote an essay called On Machinery. He was worried machines would put people out of work. That was on one hand a fair worry, but when David Ricardo sat down and tried to think about what the new jobs would be, He actually had no idea. He couldn't think of them, and he was one of the smartest people of his day. So most new jobs, in fact, will be things that very few individuals can think of in advance, but the market with wages and prices, you know, iterates with its own algorithms and solves that problem, creates new jobs. Just a few closing remarks before I turn over to Alex. Many people believe, in my view correctly, that the AI revolution will be much quicker than the industrial revolution. I think that's quite likely. Uh, that will probably ease job adjustment, not make it harder. You see in the data when sectors change, like manufacturing declines, people hang around the old town or the old job or the old company hoping it will come back. And there's a slowness of adjustment. If there's a decisive change more quickly, I actually think adjustment will be quicker as well.. It might be more disorienting, but the new jobs will be created more quickly. People will move when they need to. There was a scenario online, it was about 3 weeks ago, this claim that because of AI there's going to be a Great Depression again and all the purchasing power will be sucked out of the economy. That came from a group called Citrini Research. That's a view all economists think are wrong. It's not really a matter of opinion. The notion that all the money gets sucked somehow into an AI company and no one has any money, it's not an equilibrium. There's only huge profits at the top if there's people at the bottom buying the product, spending money. And what in fact happens is to the extent the companies earn money, they hire labor, they invest the money, they may spend it on consumption. There's a circular flow in an economy. Economists call this Say's Law. But the Citrini scenario just simply isn't really possible, which is good news. If you ask, you know, what's the incidence— it's an economist's word— of strong AI, like who really benefits, one economic insight is you need to look at longer-run factors. And if there's a new innovation or a new tax, who gains, who loses? It depends on what are the bottlenecks or what are the scarce factors. So right now, two of the big bottlenecks would be energy, compute. Another might be land in San Francisco. So, uh, maybe unjustly, but some returns, high returns will accrue to land in San Francisco, especially after IPOs, and also to the energy sector. And if we think that's bad for the economy or just somehow immoral or unjust, what we can do is act to ease those bottlenecks. And in your capacity as private citizens, If you think it's bad that the energy sector in San Francisco, uh, homeowners get more money, you can have YIMBY for San Francisco, and our government and private sector could do more for solar power and nuclear power, other areas, and that would mean the gains from AI are distributed more broadly. Those are things in my view we should do, we could do. I think we're starting to see actually those happen. Just final point I'll make when it comes to labor markets, what do I like really think is going to happen. I think most people at the bottom will be much better off because of deflationary pressures. A lot of services will be either free or almost free. Uh, that will be very good for them. I think people at the very top who know how to work AI will earn a lot more money. So we'll have a lot more billionaires, people with pretty small firms. I think the big losers, and this does pain me, —so if I'm a member of this class—it's just the upper, upper middle class, the people who go to good schools and they think they can walk into careers in law, consulting, finance that are more or less automatic. They simply have to work hard and do what they're told. They'll become partner. They'll take home $2 million a year, some kind of home in the Connecticut suburbs story. In labor markets, I think a lot of those people will lose out because of AI. Those people are not going to starve in the streets. They may end up, you know, in Houston in an energy sector job earning $300K a year rather than $2 million a year. My heart bleeds for them a modest amount. It will be a social disruption. I think it's quite wrong to deny there'll be plenty of social disruptions. But overall, I think labor markets are going to make it through fine. And I look forward to seeing all the new and interesting jobs that are created. Alex, I turn it over to you.

All right, let me give a little bit of a big picture. You know, productivity— productivity just means getting more output from the same inputs. And productivity growth is really the entire secret to economic growth and to improving the standard of living. You know, in 1970, not that long ago, but 1970, half of the world's population was living in extreme poverty. That means less than $3 per day, half of the world's population. Today, that number is 8.5%, and it's falling. And that's what productivity growth did. It has eliminated or greatly reduced extreme poverty. Now, extreme poverty is not the only part of the story. You know, the median American, uh, today has more income than roughly 93% of the world's population. Now that tells you how advanced the United States is, but also how much the rest of the world, uh, has to go to, uh, catch up. And even in the United States, uh, the richest, uh, large country, uh, in history, there's a lot of room for improvement, and not just in producing more goods and services, but in creating longer and healthier lives. As income, as goods and services increase, it's true that on the margin, they become less valuable. We don't need so many, but an extra year of healthy life gets more valuable the more goods and services you have already. Because where— what would be better? You really want to extend your life when your life is already good. That's when extending your life really has high value. So the value of extending life is going up. Now, it is of course true that with any productivity improvement, usually coming from technology, there's change. And change creates fears and change creates challenges. You know, we're seeing a lot of that today. Here is a perceptive critic on social media. The critic says, "The problem with social media isn't that it stores information. The problem is that it stores information instead of you storing the information." Every time you screenshot something instead of thinking, every time you share instead of reflecting, you are training yourself to be a little more hollow. You don't remember things anymore, you just know where to find them. Now, who said this? Was this a concerned parent, a technology critic, somebody in the Senate hearings? This was actually Socrates, okay, only slightly paraphrased. Socrates was worried about writing. He was worried about writing. Now, you know, the— this is not— we shouldn't make too much fun of Socrates, okay, because it's completely natural. It is completely natural, as Tyler Cowen said, that it is very easy to see the jobs which are destroyed with a new technology, much, much harder— you need to be a Jules Verne to imagine the jobs which are created with new technology. You know, as Tyler mentioned, Ricardo, one of the greatest economists of his age, was worried about machinery taking jobs and yet When you take a look around, you know, you look at the Luddites, famously, the Luddites, you know, were smashing the machines, smashing the looms. Interesting fact about the looms is that these were one of the first technologies which were controlled by algorithm, by punch card. So the Luddites weren't just against machinery, the Luddites were the first people attacking artificial intelligence. Now, What happened with their predictions? Okay, how many weavers did the Jacquard loom permanently put out of work? How many farm laborers did the tractor put out of work? How many accountants did Excel put out of work? The answer basically to a rounding error is zero, is zero, a big fat zero. You know, of course it's true that the tractors decreased the number of farm laborers, but at the same time, they increased the number of tractor factories, the number of auto factories, you know, the demand for employment in many other fields went, you know, way up. And certainly, there are some fields where when a good becomes cheap, even the demand in that field goes way up, right? So, the creation of Excel did not put accountants out of work. It actually further increased the demand for accountants, and I think the same is going to be true for AI. There will be new jobs. There will be some— in some fields, the actual number of jobs will increase in those fields in the same way that computers increase the demand for software engineers and so forth. Now, The fears of the Luddites were in some sense wrong. The mechanical looms have made us better off. That's the fundamental fact. However, I'll give them the credit in this respect: the Luddites were not completely wrong because the amount of work did decrease. But now Consider the following two scenarios. Suppose I tell you that AI is going to create 50% unemployment. Half of the people in the workforce will lose their jobs. That sounds terrible. You know, that's catastrophic. Suppose, however, that I tell you that the work week will be cut in half. People will do half as much work. That actually sounds glorious, you know, that sounds great. And yet these are almost the same thing, okay, they're almost the same thing. Subject to distribution is half the amount of work, okay, and we have some choices about which scenario will happen. And in fact, half of the amount of work is not unprecedented, not at all unprecedented. Because at the beginning of the Industrial Revolution, 1850, 1870, the number of work hours, annual work hours, 3,000 hours a year. That was on average, 3,000 hours a year. Today, that number is half, is half, 1,500 hours a year. So the amount of work has actually fallen in half. And it's pretty good. I'm pretty happy, you know, about that. You know, in 1850, kind of remarkably, half of a person's entire life was spent working. Half of their entire life. So basically, you worked, you slept, and then you died. Okay, today we're talking about 10%. 10% of a person's entire life is spent working. Now, what has happened? Well, what has happened is not permanent mass unemployment. Instead, we have had shorter work weeks, fewer hours, a longer childhood, and a bigger retirement. In fact, many of these things we call leisure, they're inventions of the modern era. So you think about the vacation. The vacation was a 19th-century invention of the rich which became democratized in the 20th century. Or, you know, think the 2-day weekend. It's a 20th-century invention. The very idea that somebody of working age would stop working before their body gave out and spend 15 to 20 years enjoying leisure, that would have blown the minds of almost anybody in the turn of the century, 1850, 1870. The very idea of retirement is a new idea for most people. So when you're, you know, when you're thinking about what is the effect of, you know, AI going to be, I think the place to begin is not to think about 50% unemployment, is to think about a shorter work week, a longer childhood, more education, a bigger retirement, easier, happier, better lives. And, you know, as Tyler said, we're thrilled to be— that of the work that you guys are doing, we're thrilled about it. I mean, this to us is like being in the laboratory with Alexander Graham Bell, you know, when he's calling, you know, Watson, come here. And for us to be able to have a a little bit of a peek into this happening in front of our very eyes is a complete thrill. So we are totally excited and happy at the work that you're doing and thrilled to just be a little part of this productivity revolution.

Thank you so much. That was fantastic. I'll do the emote applause, but I like— would be clapping if we were live. We have a bunch of wonderful questions that trickled in throughout the talks. And I think that for the second half of this, this fireside chat, we'll lean more into Q&A. For the audience, I want you to feel empowered to raise your hand and announce questions. Between those announcements, I'll work through the questions that we have in the 10th Econ Growth Fireside Questions channel. So if you'd like to float your questions there and let your colleagues upvote, we can go that way. To start, we have a great question from Tejal. Tejal, I'll let you read your own question.

Let me read my question. This question is for— also, thank you so much for coming. We're really, really grateful for you to give us your time. There's some fear that AGI is different from other technological revolutions in that it won't just automate parts of tasks or parts of jobs. Like, maybe theoretically, it could, in the limit, do everything better than a human. So I'm really curious, like, assume this is true. Assume that in the limit, the models can do everything, both digital and physical, actually better than a human. Do you still think that this is net good for quality of life, and why?

One question is how quickly robots will be very practical in all spheres of life. And I don't think we know the answer to that. It is certainly not imminent. To imagine a robot cleaning my carpet seems perhaps not too far away. But if I think of everything I do throughout the course of a day in my job, never mind my life, and how much of it is physical and meeting with people and communicating and motivating and coordinating and inspiring, I don't think we're anywhere close to getting robots to do that. So people doing physical things will be a bottleneck for a very long time. Higher wages will accrue to those bottlenecks, people who are doing physical parts of broader production processes. And if you imagine in some very distant future, you know, there's a robot for everything, you're pretty close to scarcity having been solved. You could then debate how should it be distributed. Should there be a sovereign wealth fund? Sam Altman has talked about a UBI. An easy way is just if people are able to buy land and capital in advance. Uh, I think it's too far ahead for us to figure out what's the best approach, but if everything's being done so incredibly well, uh, it's a wonderful position to be in overall. It is true we haven't figured out how to manage it yet, but that's like saying people in 1800 hadn't figured out how to manage the wealth of 2026. Like, it was okay they hadn't figured it out yet, and we've sort of managed. We're certainly better off than they are. And, uh, that would be my response to your question, Alex.

Yeah, I would say, look, at the present moment, uh, how many jobs has AI destroyed? Again, it's like Zero, right? I mean, it's created more jobs. So this is in the future. Now, is it imminent? Well, we'll see. I think it's early. But let's take the scenario as seriously as we sort of can. And it's obviously very difficult. But David Ricardo, who we talked about earlier, his most famous contribution to economics was the idea of comparative advantage. And what comparative advantage says is that even if somebody is better than you at everything, there are still opportunities for trade so long as there is some limit, like a time limit, on what they can do. I work with Tyler Cowen. He's better than me at just about everything, but yet, we are able to complement one another because Tyler only has 24 hours in a day. So it might be that the robots are better, but for the scenario that you envisaged, you need more than that. You need that they're so cheap and so replicable that there's no constraint. There's no capital constraint, land constraint, time constraint, energy constraint on the robots. So you need more than they're better than people. You also need that they're not constrained so that they don't need to— they have no reason to cooperate. You know, another famous example, the one we use in our textbook, you know, is Martha Stewart. She's the world's best ironer, but she still hires somebody to iron her clothes because she has better things to do. And the AIs, if they have better things to do like solving the secrets of the universe, then the AIs are still going to need somebody to plug them in and clean up or whatever. And there'll be jobs. And as Tyler Cowen said, even if there aren't— even if there are, let's say, fewer jobs, who owns the capital? Who owns the land? For a long, long time, people will. And then maybe we can talk about a UBI or whatever and figure that out. But that's for the future. There will be capital ownership. And the capital will be worth much, much, much more than it is now.

We have a great follow-on question from Makita, who says, sounds like you're saying too early to worry about overpopulation on Mars before we have rockets. Very reasonable. How many years before we solve scarcity robots for everything, should we start to freak out or have a plan for redistribution? I mean, it does seem very odd to be worried about solving scarcity.

I mean, it was like, in a post-scarce world, people are going to be looking back and saying, those guys were worried about solving scarcity. It's like us— people in the late 19th century We're worried about running out of whale oil. You know, the whales, what are we going to do without the whale oil? How are we going to light our homes without the whale oil? You know, here's another way of looking at it. I'm much more worried, not because of AI, just in general, about low growth or zero growth. Because when you have low growth, then the only way to get rich is at somebody else's expense. A zero-sum society. I think that's responsible for a lot of the problems that we're facing now, is that growth has slowed down and we've become much more divisive because it's becoming more of a zero-sum society. If we are growing the pie and we're growing the pie at tremendous rates, we can figure out distribution. Okay, distribution is a hard problem when it's zero-sum, when what I get comes at your expense. It's a much, much easier problem when the pie itself is growing.

Just say it's definitely worth thinking about and writing about. Alex and I do it, uh, but at the end of the day, if you can't figure it out, just don't be too frustrated.

I really— I Thank you for, for tagging that. And we have a similar question as a good follow-on here, which is, what should OpenAI actually do, right? If you imagine the ways that we roll this out and it goes poorly, what do we need to avoid doing? What would you recommend we take as our North Star to avoid negative impacts on society, um, as the models continue to get better and become more, um, ubiquitous in everyday life?

Well, economists stress the idea of competition. So we think it's great that you have such powerful competitors— Anthropic, Google. We'll see how Meta does. We'll see how Groq does. We'll see how the Chinese companies do. Obviously, opinions will differ. It's not your responsibility as a company to keep competition in your sector, but that's what, what we want to see. So, uh, that ensures that the gains are much more broadly distributed. Uh, the other worry I have, which is maybe orthogonal to your question, I'm worried that the AI companies are undervaluing the political risk of relying so heavily on more energy and more compute. And we see this in the Middle East now, possibly many things are going wrong and you're vulnerable to that risk. And I don't have any particular advice, but I'm more worried about that than a lot of the worries people have about AI. You need more power in compute, and right now a big chunk of the world is not doing very well.

Yeah, I would say that, you know, you really wouldn't ask James Watt to design a better unemployment insurance system, right? You know, James Watt, his comparative advantage was designing a better steam engine, and I would say just go design a better AI. Now, Of course, there are issues which you guys understand much better than Tyler and I, focusing around things like alignment. I am totally behind, 100% behind, you know, thinking very seriously about alignment type issues. I'm also— look, one of the best things that you can do to make AI a positive benefit is to think big, is to think big. Like, if AI can advance medical research, that is tremendous, that is huge. You know, Murphy and Tobin, two economists, you know, they calculate that if you could eliminate cancer, that's worth like $80 trillion. $80 trillion. If you can reduce, you know, the incidence of cancer by 10%, you know, that's $8 trillion. So like the answer to, you know, making AI have social benefits is not worry about UBI, you know, that'll take care of itself, other people will work on that. The answer for you guys is make sure that it's really effing smart and that it can help us with medical research. Or think about Waymo, okay? Sorry, it's your competitor, I know, you know, okay, sorry.

But you all use them anyway, right?

Yeah, you all use them, yeah, yeah. I mean, Waymo is going to create trillions of value, trillions, just in— think about the even in the next few years, the number of lives saved. You know, we, in the United States, 35,000 lives lost every year due to car accidents. Waymo is going to cut that, not just Waymo, AI is going to cut that, you know, way down, as well as reducing the time we spend in traffic and making it more useful. So, like, if you want to make sure that AI has big benefits, then make sure that it works really well.

In a world with strong AI, there's a kind of moral nervousness that sets in. So I'm much more likely to tell people, like, hey, you'd better fasten your seatbelt. Like, you don't want to miss out on what's coming, how many years you might live. And I didn't say that 5 years ago. Now, that's a good thing overall. But there's this feeling of the preciousness. Of what's out there. We wouldn't want you all and the other companies to lose that sense. I actually exercise more because I'm hoping not to achieve escape velocity, but that indirectly you're all going to come up with some wonderful things that'll give me another 10 years. But it might only happen when I'm 86, and I need to be around at 86 to benefit from it. It's actually changed my personal behavior. 100%.

It makes me think, Tyler, you've, some of your writing has made me rethink my ocean swimming. And I'm glad to hear that this excitement. I think one piece that we want to click in on, and I'm gonna read John Blackman's question, which I think also covers Henry's, which is, you know, at its core, I think there's more to work than the like, the wage. Um, and so I'll read the question directly. Income and wealth inequality have already been growing rapidly pre-AI, and it seems likely to grow even faster with AI. Even if those on the low end of the wealth distribution have cheaper goods and services that they can buy from a paltry income from their phony job that their boss with a phony job hired them for— those with real jobs will employ AI agents— it seems like envy and seeing in the enormous gap between most people and billionaires is already a destabilizing force in society and are leading to lower happiness. Feeling a sense of purpose and feeling valued by society are important to humans. How does that happen for the majority of the population?

Well, I think AI will liberate us from a lot of routine work. You all know this. We'll spend more time interacting with each other, coaching each other, inspiring each other. It can be a richer life. I'm not saying everyone will necessarily navigate it that way. But we'll have more of a chance to do things with our leisure time, with our earlier retirement. And again, as Alex pointed out, if we don't grow at a reasonable clip, it's not just that we stay in 2026, a bunch of things get worse. So everyone will have the chance to use it to make themselves more human.

Yeah, I mean, as I said, look, work has already declined from 50% of a person's life to 10% of a person's life. It can go down to 5% and everything will be fine. Everything will be better than fine. And in some ways, inequality, it's debated whether it's increased that much, but I don't want to get into that. But I will say this, look, in some ways, the actual effect of inequality has gone down, OK? Like, does Bill Gates have a better iPhone than I do? No. OK? Does Bill Gates even have better medical care than I do? Not really. He might have better than some people, but I'm actually in a pretty decent position. I got good insurance, whatever. The answer is really not that much. So in terms of consumption inequality, does Bill Gates have access to more of the world's information? Does he have access to a better AI? No, no, he's got, he actually has access to a worse AI. He's forced to use a worse AI.

If you want meaning in your life, another big growth sector, you know, the rest of the world, most of it is adopting AI slower than America, maybe not Singapore or Shanghai, but most of the world. And the notion that there will be many, many jobs, highly meaningful, where you go around the world and teach, say, Kenya how to integrate AI into its institutions. That will be an important job, maybe not forever, but for a long time. You will do the world incredible amounts of good. Public health work all around the world. You know, I've been in Africa twice in the last year. I was in rural Ghana, and then just last week, I was in rural South Africa. And I went to my Zimbabwean safari guide and I said, Do you know ChatGPT? And he says, what's that? So I educate him. And everyone's behind, which maybe is unfortunate, but there's so much work to do. And the first thing he said when he saw it was, oh, if I had had this, I might have actually learned something in school. And now he's going to use it to learn much more about the animals that he talks to the tourists he takes around. And his job as safari guide It's never going away. And you've just made him more productive.

As I said, the median American is richer than 93% of the world's population. So the world has got a big way to catch up and that will provide lots of focus. And look, there will be people, you know, in the past, The very rich in the past, the landowners in their Italian villas and so forth in the Roman era, they weren't complaining about, "Oh, if only I had a job, I would feel my life is more meaningful." They thought their lives were pretty good. I think people will find meaning in other ways. Won't be— it won't necessarily happen automatically, instantly, but work is important. I like my work. I'm not going to retire. Tyler's not going to retire. He likes his job. But there are other ways, and much of our work is just fun anyway. But there are other ways people will find meaning.

Wonderful. We have one more question along this line, and then I think we'll shift into more nuanced questions about AI. Henry asks, Tyler mentioned that the upper middle class could be the— the upper upper middle class could be the biggest losers. These are highly capable, ambitious, connected people. If that group loses their expected path to wealth, status, and influence, why are we so confident that the social and political consequences will be manageable?

I think it will be tough. I mean, frankly, it's one of my biggest worries. Those are influential people. They're not massive in number, but it's not just a handful of billionaires. It's a lot of people in the upper, upper middle class. And we're going to have to deal with some major political problems, which we have already. Forget about AI. Uh, it's an absolutely genuine concern. I don't have really good answers for it, or maybe not any answers at all. Every era has faced political problems like this. This is what I think ours will look like. It's up to all of us as citizens to make something of it, but you should be worried.

It's a good question, and I think it's something we think about a lot, is just our responsibility in that division. And it's a good reminder that it's happening before. This may accelerate it, but there's humanity—

Those people are going to starve., you might really worry, but those people will do fine. And a lot of them just on inherited wealth. Most of them, in fact, could just inherit wealth, enjoy the deflation, own land, own capital, and be rich. So I don't feel that sorry for them, but their relative status, or like journalists, a lot of academics will take a real tumble and people care about that status. Again, I, I think this is one of the big problems. And I don't think there's a simple answer that says it's all okay. But it's not that, oh, you're starving out the poor people or like killing the homeless. It's one of the more manageable transitions or disruptions you could be creating.

I mean, the average physician in the United States has an income of $500,000 a year. That's kind of too high. Okay. I'm not saying they don't deserve it, but you know, it's kind of, it's, That's the reason they have it is because we have a lack of skill. We don't have enough skilled workers. We need more skilled workers. So if we can broaden the medical skill and reduce some of those wages, you know, it's going to be politically an issue, but economically, you know, you're not shedding a tear from me. I think it's good for people and it's good for people We have lower income than that.

Alex and I have a textbook and we wonder in which year will strong AI make the textbook obsolete and then we lose income. We're not happy about that. That's the kind of loss we're talking about. But again, we sort of feel we deserve it. Like if the AI is better and we've had some chance to earn money and well, there's other things he and I can do. But we're one of the losers here, to be clear, possibly.

What? What, we are, Tom? What? You're going to change my mind.

I thought every month we talk about, oh, when are people going to move away from our text and just use AI?

No, no, I mean, I make fun of the Luddites sometimes, but now that my job is at risk, maybe they had a point. You shouldn't take that seriously. I mean, we've been the winners and change affects everybody. And now that it's suddenly— my job is up for grabs. I think I have enough intellectual honesty to say that I'm not that special and my class should not be holding back growth.

I go around, I give many more talks, often for pay, but not always. Because the AI can't compete with that, but some of the other things I do, the AI kicks my butt.

Well, the AI is nowhere, nowhere close to taking your job at giving talks. I'll say that. This has been very inspiring, and I want to shift into— we have some questions about how we can make sure we distribute the benefits thoughtfully. I want to open with a question from Jeff Mickey about tiered AI service. So he says, Tiered AI service plans. Is it a good on-ramp to intelligence? How do you compare the benefits of increasing access to small amounts of intelligence to all versus serving the most intelligence to fewer people? Sometimes I fear that people will be on ChatGPT free and not know how much they're missing out on and then lead less impactful lives. How many people should be buying Pro Plus now that we don't know— now that don't know the models can spend more time thinking. Or don't know how it could help them. Judging the intelligence of others is hard without building relationships, and that makes judging AI products different than buying, say, Coke Zero versus basic Coke versus sugarcane Coke. Where can you make that judgment and choice more selfishly and well?

Alex? Yeah, I mean, look, the difference between ChatGPT, you know, Pro Research and the free ChatGPT, it's less than difference between ChatGPT free, you know, 5.4 and 3.0, right? I mean, I think, you know, I hope you agree. It's pretty obvious to me that it is. And so, yeah, there is an issue, you know, on the very end, but like, as so far, the big story has been the decrease in the price of intelligence. And, you know, on the margin, most people don't need, you know, Pro. I mean, it's great for me. I use it all the time. Thank you, Wyatt. But, and we want the scientists, the engineers, you know, to be using it. But if you want to draft a letter, you know, you want to figure out what should I cook or what should I do? You know, it's not necessary. And if you want ChatGPT 5.4, well, just wait a few months and you'll have it, you know, on the free version. You know, then we'll be on to 6. But that margin, I think, is less important than the bigger margin of the decreasing cost of intelligence.

I would like to see someone do a publicity campaign in Africa And I hope it does not bankrupt you all, but when I speak to rural Africans, which I've now done in a fair number of cases recently, none of them know, you know, either Claude or ChatGPT or any. And the marginal returns to them for having something rather than nothing, I think, are very high. And again, if this can be done without bankrupting you all on the token side, the gains to mankind from things you already have are just phenomenal. And it's a free lunch, but people out there don't know. I even was talking to a university professor, full professor, near Cape Town. And I said, oh, in your university, how AI-pilled are people? And he says to me, AI-pilled, what's that? He answered the question. So there's just, there's so much more that can be done without anything really needing to be done. It's just sitting there. There's massive inertia. In human affairs, and it will happen, but it could happen more quickly.

I love that angle. It makes me think I was golfing in Bakersfield, California, and a guy in a golf cart drove by and saw a ChatGPT hat and was like, hey, they taught me how to fertilize the greens. And I was just like, I would have hated to see him last week. But to your point, It's really hard to conceptualize just how broadly distributed the use cases are. I want to take us— open up a real can of worms as we get towards the latter part of the talk. And from Kevin, we say— Kevin asks, ignoring feasibility, do you think AI systems should have property rights, run companies? Would this be positive for the human economy?

Yes. I think crypto should be their money. There should be some kind of required capitalization so they're actually liable. And there will be a full AI economy that will grow at some unknown rate and it will transform the human economy. It's a very exciting topic for both Alex and me. We talk about this all the time.

Who should get in trouble if, if Bajon's AI steals money from my bank account? Should that roll up to, to Bijan?

Yes, but there will be such a thing as AIs in the wild that are not owned, or you can't trace ownership. And you can put liability on the cloud, but if they're coming from a cloud somewhere like North Korea, I think that's a big problem we haven't figured out how to solve yet. We might need some kind of mini Great Firewall to keep out AIs that have no attachment to anyone's liability. We need a tech solution for that. That's to me very important. I don't know if your company should work on it, but someone should, and some people already are.

Yeah, it's a very interesting legal problem. You know, corporations are people, like literally that's how they are treated in the law, and there are good reasons for that because you want a corporation to be able to make a contract even when everybody in the corporation turns over and leaves, and, you know, the corporation itself has some existence. And so there will be a need to think about the legal structure for AIs, and a lot will depend upon, like, can they be punished? My feeling is that they actually can. Again, this is where you guys would know way more than us, but they do seem to have preferences. They do seem to have likes and dislikes, and there are things that, you know, you could do if— there are ways to incentivize, obviously, AIs. So I think it is possible to bring them within a legal structure, and there are ways that you can freeze them out. You could make them outlaws, or it could be outlaw AIs. We have outlawed, you know, some crypto wallets, you know, for example, frozen them out. And maybe we'll have to freeze out some AIs from, you know, the electrons. Like, it's a fascinating area. It's totally fascinating. And new things are going to have to be worked out, new areas of law. It seems to me totally doable. But I don't know what it's going to look like. We're all going to be surprised, I think. It's a fascinating project, but it doesn't seem to be unattainable.

So I'm just trying to think about how we punish the AIs. I think it could be done. We have a— A few more questions here. One more from, from Tejal. New hypothetical: assume models are better than humans at literally everything, no more comparative advantage, more energy efficient and free and effectively free, and intelligence costs $0. After a certain point, are there still returns to making the model smarter? I.e., should we still keep going?

So yes, I'm a curious sort. So solving the mysteries of the universe, I care about. I don't know if it has practical value. It might, but I still— that's one of the dreams of my life, is to be able to read the ultimate theory of the universe. So I'll vote yes.

You know, there's a great, uh, science fiction, very short, uh, story, uh, which I love, and it's about creating an artificial intelligence because they want to figure out you know, is there— does God exist? Is there a God? And so they spend all of the resources, you know, making the AI work better and giving it more and more access to energy and capital, and the AI thinks and thinks and thinks and thinks for decades. And then finally it says, I have an answer. Finally it comes back and, you know, they say, you know, well, tell us, tell us the answer. Is there a God? And the AI says, there is now.

It's—

which, man, it's— God, this is so fascinating. Sometimes reading the like chain of thoughts as it's summarizing, I feel like a hint of mysticism. I was— I remember asking like a— I think one of the EconVince questions, and I see in the chain of thought it's like, the Thai tiger shines in the night, and then it returns to kind of doing game theoretic calculation. You're like, there's something, you know, there's something happening.

Don't ever get rid of the chain of thought. I love it. People love it. It's quite significant and meaningful. I know it's not exactly how it's thinking in many cases, but I think it's very good that Deepseek, you know, put that on the table the way they did.

And so I think two more questions, one from Olivia. If unaccountable AIs are way more intelligent and productive than real humans, they own property, they're not accountable to any human, how do you envision this not leading to human disempowerment since generally seems easy to convert money and capability to influence?

Well, if they own property, they're partly accountable. You might have to call the AI police on them, but once they're tethered to things, It becomes a manageable issue that you can solve through gains from trade. There'll be rogue AIs like there are wild animals, and we don't manage to put all dangerous wild animals in zoos. So this problem will never go away, but AIs will give us a lot of defensive capabilities, and I think most AIs will want to trade to achieve whatever are their ends.

It's an interesting scenario, and I don't discount it, but nevertheless, I think it's a scenario for the future to figure out. Right now, I think it's just too early, but it doesn't seem— as Tyler said, it doesn't seem impossible. But I do think there has been— Tyler and I have known Eliza for like 20 years, so it's not like we haven't I heard these things. But it's a little bit early. It's a little bit early. Let's solve cancer first, and then we'll worry about the post-scarce society. And a lot of the other AI is helping us with this. I mean, that alone is— so you say, well, what is the answer to this problem? And I say, I don't know. Let's ask a smarter person. Well, then, you know, 10 years, 20 years, we'll have a smart person who will ask them, what do we do about this problem? And the AIs will have a good solution.

Tyler, Alex, are there any questions that you have for us as we kind of work on building, building AGI?

Well, I don't know if you're allowed to answer them, but I'm always curious as to what's coming next, to whatever extent any of you are able to speak to it. But I understand if you just have to respond with silence.

I think you can expect smarter models that can do more knowledge work.

Great. And, you know, I think integrating the models into workflows, Sam announced recently, that's wonderful. I think you all can win at that. But it's a very hard problem, and it's going to create jobs for decades. It doesn't matter how smart the things are. You know, the humans can be the stupid element, but you've still got to mesh the two. And that's a lot of jobs. And I plan to spend a lot of the rest of my life working on that in my institutions. And it's tough. I would say this.

I hope that— I don't know if you guys are. I hope some of the AI companies are talking with Epic, OK, the big medical data integration company. Okay, so the data there that Epic already has is potentially life-transforming, and to get AI access to the treasure trove of almost universal human medical data, I'm very, very excited about that and about the prospects for medical research within AI. So, and the data is there, so just bring it, bring it together, bring it together. And, uh, that would be a real thrill.

And if someone could birth a movement, I know it's not your jobs, but to get data unlocked more systematically. There's so many stupid barriers. In the past, maybe they didn't matter much. Now it's just critical for lives. And no one's really that upset about this. It's not a political issue. There's not even a large group of whiners on the internet going on about it. I would love to see just more and more data unlocked.

I couldn't agree more. I think that was clear in early deep research when like people would say like, oh, it's missing these sources. And we would just be like, oh well, those sources aren't available to the model, but we know it could find them. And so this like the capability overhang is severe. And Alex, regarding the medical data, I will just kind of zip my lips, throw away the key. But I do appreciate, I appreciate the referral. I just want to, again, on behalf of OpenAI, thank you both so much for making time today. If there are any final remarks from the team, please feel free.

Thank you so much for coming. You know, we're all going to be working on a bunch of different things, including accelerating knowledge work, and I think discussing these questions with you has been really helpful. Great.

An honor for us, and hope to see some of you out there at some point. And, uh, have a good spring. Yep, it's been thrilling.

As Tyler said, an honor.

The honor is all ours. Thank you very much.

Great, take care. Bye. Take care.

Thanks, Wyatt. Thanks, everybody. Thanks, Dijal.