#51 - Tyler Cowen - Markets, Institutions and Why AI Hype is (Probably) Overblown

27:37And that's why we have the problem with CO2 and that's why other solutions are being looked at where we have a better ratio of radiative forcing, as it's called. But when people discuss putting this into the atmosphere, then it's just being discussed as, oh, you want to block sunlight. And it sounds crazy. It sounds like a supervillain thing to just go out and start blocking sunlight. Is there, like, how would we go around solar geoengineering, you think, such that people would be more amenable to it rather than just always see the supervillain version of it immediately? because you're doing something global.

28:15Well, some parties are just going to do it, whether we like it or not. So I don't think it's a matter of convincing people. My main worries are not technical, but geopolitical. Let's say we start doing something or Saudis do something, and then Putin decides he has his own vision for how warm the average temperature should be or whatever, and he does something else. And we get into some kind of pissing war of geoengineering, and I'm not sure that's stable. It's very hard. You can know all the social science you want, but if you really think you can predict what the end of that process looks like, I would say you're fooling yourself.

28:48Yeah, that's a really good point, actually, because presumably Russia would prefer a warmer climate, actually. I mean, in some argument. Yeah, it's not just about the averages. Yeah, melt a bit of permafrost, you get a bit more usable arable land, potentially. Although, again, the actual sort of complexity of the climate shocks that you'll be having, things just become more extreme and Russia is already a climate of extremes. We've got to hope Uncle China twists their arm not to do it, right? It's very complicated. Yeah. Maybe we'll soon all want colder climates because the data centers will prefer it and cost less energy.

29:26Maybe. But I think the point is there's so many variables to forecast. It's just very, very difficult. called, even assuming you have completely mastered all the science. All right. Well, speaking of supervillains, you recently described people who might remain nameless because we try to not politicize this pod, but people who try to force down pharma prices, pharmaceutical prices, as the supervillains of society. Can you explain why? Well, there's a lot of evidence that within health care, pharma is one of the most effective ways of saving lives. And there's a more general point. I use econ terminology.

30:03I just say supply is elastic. You pay more for things, you get more. So whether it's GLP-1 drugs or fixes for HIV AIDS or in the future it will be cancer and dementia, we ought to be paying a lot more for those problems because so many other of our material needs, I wouldn't say they're entirely met, but a lot of us are pretty comfortable. And the things that actually make you suffer are miserable. You getting these conditions, your relatives getting these conditions, it's where we should be spending more money not less so millions of lives are at stake right we've got to do this and it's going to mean spending a lot of money let's not be shy yeah i mean i'm when it comes to trump i i was wondering like oh is he just doing his classic negotiation tactic of ask for the extreme so that they'll compromise and settle on a higher middle than you would have gotten otherwise then i was digging into it more and it just seems like you the ask of because his whole his whole point is that it's it's completely unfair that americans are essentially subsidizing the rest of the world of course it's unfair i love unfair right unfair is how things get done that right exactly it's on it's like what lens do you want to look at yes it's unfair for the small pocket of americans but at the same time yes they're paying a more of a larger pie and if this if well i'd love to hear like you explain why it is that it won't just force the other markets to pay more because he's saying you know that the lowest basically American Americans should pay no more than the uh the lowest paying um other developed nation and that sounds in principle reasonable but can you explain why it's not why it won't work yeah the idealized version is that the others start paying more but it seems like the potential real real version is that they just don't get any um but yeah maybe you can there's some research papers on this question and they've concluded that in basically Americans don't end up really with cheaper drugs.

31:59Other countries pay a bit more, not that much. But basically it's very hard to control the healthcare systems of other countries and get other countries to do your bidding. Typically it backfires and they retaliate against you and you haven't built any political consensus for it. It's a weird sort of thing that Trump, he's like so pro tariff. But in essence, he's insisting that for drugs from abroad, it has to be easier to import them into this country. And he wants price arbitrage. So I think you get the highest returns by allowing for price discrimination. The richest markets pay the most. No, it's not fair, but you get much better final cures and products and pharma and everything else.

32:44So I'm all for price discrimination, even though Americans such as myself, in some cases, are the biggest losers. But we're the ones who want it the most. We have the most money, right? And we get preferential distribution as well. Absolutely. So we get something for it. If you were in power, what is a thing that, what is some, I mean, is there a low-hanging fruit that can be done that will reduce drug prices somewhat for Americans? without... Well, keep in mind, 90 % of prescriptions are generics, right? And for generics, we pay in America below the OECD average. So for most of what you do, you're getting below average prices.

33:22So it's not exactly the ripoff you're told. Now, you could say when you add up all the revenue, on average, you're paying more. But that's because the generics here are so cheap, right, that they sort of disappear from the equation. Maybe the actual negative sentiment that Americans are having is actually less about drug prices and more about the fact that they just don't have a functioning market when it comes to things like you need to go to the ER. You know, as a transplant from the UK, it's insane to me. I had to go get some semi-emergency treatment and I was just like, well, how much is this going to cost?

33:53Oh, we won't tell you. No one will tell you. Yeah. Like, well, what is this? Like, I'm a captive. I have nowhere, I guess it's because I have nowhere else to go. But I mean, I can't even shop around between hospitals because there is no transparency in prices. It just seems that seems like more the absurd thing. And I wonder if that sort of general annoyance that Americans have sort of over spills into drug prices when actually the drug prices isn't the problem. We need to fix that by making it a priority. There were parts of Obamacare that tried to fix it. Trump in his first term for a brief while made a stink about it.

34:26But we've never really seen through a sustained program to make it one of our three or four national priorities and literally just fix the problem so it's like going into mcdonald's and ordering a quarter pounder and we can do it other countries have done it u.s does it for many other things we just have to actually really want to do it what is the um difficulty there is it that they genuinely don't know the prices or what they don't want to tell you right they don't they may not know because they don't know what's the pending reimbursement and so on there's a lot of complications, but to actually just truly literally force them under not quite pain of death, but severe penalties, that there has to be a posted price and it reflects what you actually get.

35:11And I know the quality dimension varies and other complications, but I know it can be done. And again, we've seen it done in other parts of the world. If you go to Latin America where you're just treated by a private doctor, if you have any sense. My wife, once we were in Rio de Janeiro, or we had to visit a doctor. We knew to just go to a private doctor. We show up, we say we're paying cash. We said, what's the price? That was the price. It can be done. We did it. I mean, they're calculating the price after the fact. So I think they can assume that given those few procedures, this would be the price.

35:43If you have a different set of procedures, it would be a different price. Seems somewhat straightforward, given that it happens after the fact. AI should make it much easier. There are real complications to this, I would gladly admit. run it through your row three and it will tell you the price yeah i i had one situation where i had to get an ultrasound and the ultrasound and costing about five thousand dollars and then i looked up what an ultrasound secondhand machine costs which was also around five thousand dollars so if you then get the imagery and feed it into an advanced ai model i think you kind of get break even yeah yeah and just ultrasounds instead of one single use of it that doesn't seem right like there should be some optimization that is being done there we need freer entry we need to fix the problems with third-party payment again it's not simple but if we had the national will to do it we can do much of it not all of it what are your golden rules or advice would you give to people who are listening who want to go out and design a new set of incentives for an industry that they're in?

36:55I tend to give people the same advice no matter what they're doing. That's keep on improving the quality of your peers and find new and better mentors. There aren't that many pieces of advice that are always good advice, but those two come close. And I think in this case also they apply. There's so many complications to new and different incentives, and you've got to rely on the experiences of other people to a considerable extent and you need your network to be great otherwise you're going to fail so i want to discuss um uh ai's impact on the economy that we would be projecting and maybe to start uh i wanted to uh ask you to give an easy to grok understanding of what gdp is what it captures and what it doesn't capture and how consumer surplus place around it?

37:43Well, GDP is the value of final goods and services sold, right? So it's a very rough measure of the size of an economy. And you can divide it by the number of people, which gives you a very rough measure of living standards. So that's GDP. Consumer surplus is the gain you get. So if I pay$7 for a dark chocolate ice cream, as I do, it's worth$14 to me. So 14 minus 7 is 7. My consumer surplus is seven. So they're very different things. Right. And specifically, consumer surplus is not captured within the GDP calculus, right? That's right. But I would say typically they're closely correlated. So the places that produce a lot are also good at making people happier and having more competitive markets and lowering prices.

38:33Not in every sector, not always in health care and scans. But the U.S. economy overall very likely creates more consumer surplus than anywhere else, any large country at least. Would you say that with more and more digital products and services that maybe the ratio of consumer surplus to GDP has the consumer surplus grow? Because I'm thinking of products like Google Maps or Wikipedia that are basically, I don't know, are they just consumer surplus or are they also captured in GDP? I think that's a smaller effect than people think. So Chad Syverson, he's an economist at Chicago. He tried to measure this effect, and he found it was smaller than most people think.

39:14Keep in mind, a lot of the internet is being commercialized still. So GPS is great, but it means the pizza delivery person can deliver more pizzas in a given amount of time, and that will mean higher GDP. It means you might be willing to drive to the theater when otherwise the directions would be intimidating. That boosts GDP. Now, reading Wikipedia probably doesn't boost GDP. But when you really think through what you're doing on the Internet, more and more of it is actually captured in GDP. Though some of it then boosts GDP later, whereas the surplus initially happens and then people figure out how to implement that newly found value into their lives so that they become more productive.

39:56That can be. Like AI right now, a lot of people enjoy it. It's not clear how much it's boosting GDP. I think that's likely to come. But it's mostly a thing for fun or to save time. And that's great. It's consumer surplus, but maybe not much GDP. Right. So that's what I was wondering, whether that AI and kind of cheap LLMs in the future would potentially show up more in these and create high welfare, but not necessarily a higher GDP. I think over time, there'll be more and more of a shift to GDP as agents become things you can actually use, which is not currently the case. There's not really an agent I can use to book my plane ticket, but there will be pretty soon.

40:37And with agents, they'll take over more and more workplace tasks, boost productivity, and GDP will go up quite a bit. But I don't see that, you know, May 2025. Maybe it would be a smidgen in programming. I definitely see it possibly in graphic design. Not many other places. I use it to prepare for podcasts. I give away my podcast for free. That's not GDP, but it's a kind of consumer surplus. And if I sold ads on the podcast, it would mean more productivity in some way. Yeah. So, and you and some other economists like Robin Hansen have talked about that the impact of AI and AI agents may be very significant in GDP, but still like in the one or 2 % range, which is also obviously very significant in terms of GDP growth.

41:23My best prediction, which was highly subjective and speculative, was that it would raise the growth rate per year by half a percentage point. So instead of growing, say, 2 % a year, we might grow 2.5 % a year. So that over time compounds into a big difference, but it takes decades. So you can, A, believe an AI-intense world is very different eventually, but also believe in a given year, most people will be saying ho-hum, and you'll kind of understand why. And that's my view. I think it's already what we're seeing. Yeah, so I did a back-of-the-envelope calculation on like figuring out how many agents we have i'll run you through it quickly so semi-analysis estimates that in 2026 as we mentioned before we'll have about 40 gigawatt of capacity dedicated to ai data centers and roughly probably i'm assuming like 10 gigawatt of that will go for will go towards actively active inference and with 10 gigawatts of inference you can basically run something like 1 ,003 ,000 size agents.

42:29Not obviously today, as you mentioned, they're not yet able to be active agents that are doing very highly productive tasks by themselves. You still need to direct them, but this may change next year. And Jeff Dean, Google's chief scientist, said that he expects agents to work on a junior engineer level in around a year. So if I combine both of these things, then I'm getting to about 1 ,000 ,000 agents that can work in parallel in thinking model style together with the quality of, say, junior engineer levels. So that's about a$100 ,000 payment that we would give a junior engineer in general. So that adds about like$100 billion in productive work in a year.

43:09The binding constraint will be speed of human adoption, not how good the agents are once we get to this point. And I'm not sure what the speed of human adoption will be, but I typically see it in history as having been quite slow. Yeah, so I mean one in the digital realm I would imagine it's a bit faster because in the US we have about 2 million software engineers and probably about the same amount of people that work in administrative and other capacities around them. So there's quite a lot of mental capacity going towards figuring out how to integrate those already, right? and we are not even here talking so it's two million engineers give them another one million agents some people one could imagine say a current open ai engineer would maybe be able to productively use two three agents in parallel but this is not even at that ratio right this is a half an agent per person ratio that we're getting and the world has about 20 to 40 million software engineers so maybe they find some way where like you can capture that productivity gain from it um and anyway so i ran with the numbers for just a second longer just like what happens in five years and the estimate is that the data center capacity keeps doubling about every two years so maybe in five six years after that we get to about 100 gigawatt being dedicated to running inference and then if the agents also start improving in intelligence and skill, then instead of being a junior engineer level, maybe they become senior engineer level in five years beyond that.

44:50And then you would potentially have 100 gigabyte allowing for 10 million agents. If they run at and a senior engineer gets paid$1 million per year at OpenAI and other companies, that adds 10 trillion of potential productive capabilities obviously prior to repricing and software becoming just cheaper and therefore it doesn't show up in actual costs, etc. But I wonder, so at prior pricing, it seems like there is a potential for this use, for this value to come out. But I agree that it's not going to necessarily show up for various reasons. One, because prices are just in one, because people don't yet know how to use them.

45:30Is that what you would point at? Or do you think there are also other reasons why we don't get to that future necessarily? There are two ways in which I might differ from those numbers. First, I don't think we'll be able to expand energy and data centers that quickly, although I would like to. But the second is there's a lot of back and forth between what's an average value and what's a marginal value. I would think about it very literally in what economists call real terms. So let's say AI can do all the software engineering, which sort of will be true soon enough. You then have all this liberated human labor, and you've got to ask, what's its next most valuable thing it can do?

46:06And that's the size of the gain. I'm not sure what that is. I'm not sure it's so wildly impressive. Again, I think it will raise the economy's growth rate, that liberated labor plus other liberated labor. But it's not going to be in the many trillions of dollars by my accounting. But is it only liberated labor, or do we also have? I can see just many more digital goods coming up where, for example, currently education or 10 years ago, education was basically all in person. And in 10 years, I could imagine much more education being directed tutoring one on one via these agents. So isn't that creating a slightly different effect?

46:51I agree with everything you just said, but I think of the marginal value of more digital goods per se is quite small. The marginal value of more human motivation is extremely large. And if we can get the two going together, we will get a huge growth rate effect. But I think that will take a long time. So you could have the world's best tutor for a kid who doesn't want to learn. And I think you're still mostly stuck. The kid might want to learn a bit more because AI is fun, but I don't think it will really solve that problem. Isn't this one of the unique situations, though, where in the past, yes, the economy is always going to be limited by human bottlenecks.

47:26But isn't this the first time that because you have AI agents who will be building stuff for more AI agents and so on, that it could actually get decoupled from those human bottlenecks entirely? Only in self-contained systems. So Chasset clearly has worked, right? Programming, it's starting to work and it will work. Within closed system factories, robotics related things obviously are working, will continue to work. And those are the areas where there'll be big effects. And I think all of those will sum to like a half a percentage point of growth a year, which is a lot. So then going with the calculus, so let's assume that actually the data center capacity keeps growing at that rate.

48:08There are some indicators that show that it may be possible. Oh, it's clearly possible. I just would not predict it, but go on, let's say. Yeah, let's just, let's, in the case that it happens. So you would say, which I would agree, maybe the initial additional AI agents produce something like$1 million in value, but then the next ones become just much lower value because we don't yet know how to reasonably slot them in. Right. And so many of the actual problems of life, this gets back to the earlier discussion of addiction, the problems of human motivation and attitude. And indirectly, AIs might do a great deal there by inventing new medicines, pharma, right, supplies elastic and so on.

48:49But that just takes a long time. And the mere fact that there'll be this cornucopia of genius output, I'm sorry to say most people just don't care that much. I think that's sad, but I feel that's where we're at. We're going to get something like what people call superintelligence and not really know what to do with it. Not for a long time. It's interesting, but you're at the same time. I've heard you talk about fairly strong optimism around longevity improvements. So you believe in like significant improvements in pharma, but you just see them to happen slower, I suppose. Where my take is that looking at what isomorphic labs is doing, they're developing better and better in silico simulations.

49:34So rather than having to do actual experiments in a lab, you can just do it in a simulated lab or at least some of the experiments you start being able to do. where instead of having to have like five times run through the feedback cycle where you have to test, experiment, try, etc., maybe you can bring it down to only three times because you were smarter in every part of this cybernetic loop. It's like a chemical reaction where the slowest stage sets the pace for everything. So, oh, whatever you create, you've got to get through clinical trials, get through the regulators. There's always a quite slow stage in the process somewhere where you don't just do it all by simulation.

50:15And the more you can do by AI, simulation, whatever, the more binding the stupid constraints become. That's true, but I think the total time for development, would you say it can still be halved, let's say? Yeah, I find that plausible within 20 years. That's halved, yeah. And that's great. So I think my grandkids will live to be 97. Okay, great. Yeah, so I'm also more on the halved set than, And I suppose it may sound like it now you imagine that one could imagine the drug drugs just get developed and distributed within months. But no, that's not going to happen without. Yeah. But I would personally be willing to make experiments in the direction of way less in clinical trials and, you know, taking on some costs for the now more extreme benefits.

51:02I just don't think we will. Yeah, certainly. I mean, there were attempts around COVID with human challenge trials, which already kind of shortened the span a little bit. In the UK, they were allowed to do. I think the U.S. still hasn't had any of the experiments with human challenge trials. Is that one of the things that you're considering? But it only shortens it a bit, as you said. And it's not even clear in the UK it shortened it at all. So it's just very tough. When you actually look at progress, it's a slog, even during very exciting times. But I'm optimistic. Again, half percentage point a year is awesome.

51:36It means our government won't go broke. There won't be this very high boost in taxes or extreme cuts in spending. We might avoid a recession or depression, and America will stay a great nation if indeed we get that. So that's a big gain. A whole set of big gains. So the thing that's kind of hard in my mind to square is if I take three things, one is yeah, AI increases productivity significantly for people. Two, we don't get higher GDP growth rates. No, we will get higher rates. Yeah, but not more than like one percentage point. And three, we don't get unemployment. Those seem to not work all three at the same time.

52:20Like one has to give up one of them. Why can't the software engineers be re-employed doing other things?

52:29at the same rate? No, they'd probably be paid less. That's why they became software engineers. Some will retrain and maybe earn more. But unless you're managing a lot of AIs, I predict software engineers will be worse off. They'll be fine. They're not going to starve. They're smart people and good workers. But unless they become masters of the universe, they're losers in this. Yeah, I would have assumed that maybe the point is also around, well, not that many people will actually get the productivity gains from AI because not that many people use it. Even though 400 million monthly active users for chat GPT is quite good.

53:06There'll be more wealth and a lot more projects. And I think like carpenters, gardeners, people who do things like that in a smart way will earn much, much more. So they'll offset the losses of the software engineers and the journalists and the people who write like I do. so it's generally kind of hard to um really confidently project in 10 20 years which jobs will change how i think um but you have thought i imagine about your personal like work and you do multiple things uh quite well so what how do you think is your work as tyler the economics professor how is that going to change with it's changed my life a lot and unlike most people who blah, blah, blah about things.

53:50I really put this into practice. So I travel a lot more. I meet people a lot more. You could say in broad terms, I invest more in charisma. I do a lot more with philanthropic giving. To be clear, it's not my money, but I direct it through Emergent Ventures. And I do more podcasting because I think people are willing to read AI text, but don't want to hear AIs as podcasters. So I've really changed my allocation a lot and for the better. And I've been happy. So let's keep on at that. Most people I know haven't changed anything. And for Tyler, the author, how do you think is that going to affect it?

54:29I'm working now on what I expect will be my last book. And it's a book that will be titled Mentors. And even if the AI could write that book as well or better than I can, I think people want to read a book on mentorship from an actual human who has done it. So that topic, at least I hope, is somewhat immune. But I'm not going to just keep on writing books. I think, I'm not sure the AIs will write the books, but the AI will be a kind of box. And a lot of readers will prefer the box to the book. So I shouldn't be writing more books. But you will still keep writing, I imagine. Of course, on a daily basis, you know, super short term, rapid response to what's happening in the world.

55:09And I don't think that will be obsolete, or at least not for people who are already branded, which would include myself. I'm not sure as a 20-year-old it's what you should start trying to do, though.

55:22I wonder also what you think just in general about UBI and whether it's something that's workable in the world of superintelligence or just far advanced intelligence, or what better system could be used there? Well, since I don't think there'll be mass unemployment, for one thing, I'm less worried. Sam Altman helped fund this recent study, a randomized control trial of UBI. It didn't make people happier. It's perceived as unfair. I just don't think we're going to do it. It doesn't feel or sound American. In these matters, culture is very important. Probably some smaller societies will, maybe that are super high trust.

56:05But I don't see it as the future for large, diverse economies. But how then, in a world where we have superintelligence, presumably that will be the ultimate wealth concentration maximizing? I don't see why. It's going to produce a lot of things on a competitive basis, and a lot of services will be free or near free. It's very complicated, but I wouldn't assume that at all. Let's say your healthcare was mostly free. And productivity and construction would go up by a mere 3x. so homes would be cheaper. People would be more than fine. Healthcare and lodging taken care of. Food's probably cheaper.

56:44So being poor will be much better. Why will healthcare and lodging be taken care of? Well, construction productivity is either stagnated or even gone down in this country for 10 or 15 years. Gone down. There's papers written on this. I don't have the final answers. I'm not sure the authors do. But if superintelligence cannot tell us how to build houses more cheaply, I would say it's not super intelligence. Sure. So people can move to wherever. Some states are going to grant freedom to build. And you'll be able to pipe in entertainment, whatever, with AI, can even without AI. It'll be great, especially for lower income people.

57:21So at the same time, you and Sam Altman have expressed that there will be, in his case, one person, one billion dollar company. In your case, a few people, a billion dollar company is run by a couple of people. So I agree with that. But if not everybody can run a$1 billion company. Sure. And if everybody is now running a much more productive business, then that also doesn't work. Like what are the other people doing? Like there's just the amount of human attention and the population is not elastic. Like there is a fixed limit there at some point that gets hit. So it seems to me that if some people become much, much more productive, then some just don't get to work.

58:11Or do you think that there's some, yeah, do you see this differently? My best guess is the poor are really much better off and the biggest winners. And then you have a lot of people who might have been worth$100 million who will be worth$5 billion because they run this company on their own or with a buddy or brother or whatever. I don't see that that class of people create some huge social problem. But they'll be better off in part because, as you said, like healthcare becomes very cheap. It's like AI therapists are just basically a cent cost or something, right? We'll terraform the desert and investigate new drugs.

58:48We'll need new humans to be in these clinical trials. We'll pay them. We'll need more gardeners. More people will have estates. We'll need more carpenters. We'll need more whatever. more people will be hired to make songs or it's hard to everybody is just everybody is selling on Etsy basically Etsy with individual contributors and handmade things gross massively well a lot of it will be through companies but it could be through an uber like arrangement where oh you know the plumber comes over now the gardener shows up I'm not sure the balance there but I think either can work but I don't I don't see how society because I mean if this is going to happen it's potentially going to happen within five years no no way he thinks it's more 30 oh you think super intelligence is like 30 years away i don't even know what super intelligence means exactly but very very strong ai to do what you're talking about is decades away five years we're still going to have people in state universities complaining that their students are cheating i promise you this it would be quite deflationary that might not go away ever How backward the dialogue would be, yeah.

59:54No, okay. Okay, your timeline's like that. See, mine are like five years. But in part it's because of diffusion, right? Or do you think that the... Diffusion and human imperfection, yeah. Right, not that there... If I understand you correctly, you think that there may well be such an AI agent that is capable of producing what we discussed as like a senior engineer type of work within five years or so? It seems plausible to me. I'm not the AI expert on the technical side, but that's not what I'm denying. Right, right. You're just denying that once we have them, how long will it take for us to adapt to finding useful ways to...

1:00:32Electricity took 40 years, right? And that's, in a way, a no-brainer. And we're now so used to it. It's everywhere. It's in your walls. It's in your smartphone. It's in your car. But that took 40 years. I just think that like we even if very very strong AI is 30 years away we're not that far away from we're already sort of starting to see agents emerge and because they live in the digital realm they are able to essentially that you know they are optimized entirely for that realm in a way than you know any any human mind could could ever be. You just feel like diffusion is easier in the the digital realm in some way, but it does require people finding, yeah, just like some useful task for them as well.

1:01:20I would put it this way. If I'm wrong and the view you're stating is correct, you can make so, so much money investing in markets today that you shouldn't be doing anything else. You shouldn't be wasting your time talking to me. Market prices look very normal along all dimensions for any asset you look at. And if you're right, they shouldn't be. So, like, get to it. But markets have historically been very slow to act on things. That's why you can make so much money now, because the market would be slow to act. I mean, we invested in a semiconductor ETF a few years ago. That's done well. And I would imagine that that's still underpriced as well, which is why I'm holding it still.

1:02:00But that's consistent with my worldview also. Like, real interest rates will go up a lot, if you're correct. and they'll go up within five years when, you know, in your view, it will be more evident. The point around markets actually touches on a different topic. So you've previously said that one of the ways to look at, so some people predict a high degree of existential risk from AI or AGI and that you take more of an agnostic stance towards it because in part it's just very hard to predict how these technologies play out. Do you think that fairly reflects how you think about it? I think I'm more critical of the view than that.

1:02:45I think AI is more likely to save us than destroy us. Any sufficiently distant future, of course, it's hard to predict. There's plenty of other ways we can do ourselves in. AI helps us on all those other fronts. I would agree with that as well, that AI is more likely. I mean, but someone who thinks that one can think simultaneously that it's net good, that the median case is positive, and still that the downside risks are very worth putting a lot of attention to. Those can all simultaneously be true, right? Well, like electricity, we should make it better and safer. And I think we will, of course.

1:03:21But you've also argued with the predictability of how the technology will play out and whether those risks are making sense as people predict them or not. Do you agree with that, Lance? that it's just inherently unpredictable? Again, there's a lot of somewhat loose words in that question. I would just repeat the challenge I've made to what I call them the doomers before. If you really think your argument is correct, take it to peer review and scientific journals and try to publish it and do what the climate change people did, which in fact did persuade nearly everybody because of peer review and the evidence.

1:03:56And what I see people doing is writing these 17-page long blog posts with 39 different points vertically stacked and not a clear model and then expecting everyone to just agree. So I'm not going to base policy on that, nor should I. And no one seems to have any interest in doing this. There is some peer-reviewed articles on peer-reviewed papers on the points around existential risk. Around, but not like actual models where people work through. and try to persuade me it will happen. And I've had people email me dozens of things over the last few years, and I've never seen anything really close to what I'm asking for.

1:04:34Okay, so it's more about peer-reviewed articles that persuade you as well? Well, on any big and important question, yeah. Otherwise, it's like, well, okay, we'll take that into account. Like, the main labs are concerned with safety, and private businesses, as they adopt systems, will be very, very concerned with safety, and there'll be checks and balances. So the argument just has to be made. Is peer review, though, the right information investigation system for this problem? Because we're kind of inherently discussing something that has allegedly, at least in the claim of the doomers, so to say, kind of very little historical precedent and talks about the potential of what would happen in a dynamic.

1:05:20It's talking more about what I would say is it's not predicting if this occurs, then that, then this, then this happens. Any of those individual paths seem unlikely to be accurate, right? And I don't think one can write one of those out. But the statements that I think many are making are more of the type, there are certain attractor states within a complex system, such that it's difficult to predict the specific path, but one can kind of predict better the long-term dynamics of something. I'm not saying we should do only peer review. I'm saying when I see an idea that won't do peer review at all, like the anti-vaxxers, I get very, very suspicious.

1:06:05And something like nuclear weapons, we had intense social scientific investigation of that after they were invented. They were a very new thing. It was highly imperfect what we came up with, but at least it was a start down that path, and we ended up learning a good deal from people analyzing nuclear weapons. And again, I'm just telling people, go do this. Like, you might persuade me. I've changed my mind on most things in my life. What are your best suggestions to try and mitigate? Because, you know, you are someone who hates any idea of incursion on free speech, understandably. but there are genuine risks mounting in the proliferation or essentially the reduction of barriers to very bad actors being able to create very dangerous things through you know whether it's synthetic biology access to chemical weapons bio bio weapons because of llms that are essentially able to disseminate the information giving a how-to cookbook um you know which reduced let's say the both the financial costs, but also the intellectual costs, you know, before you need like 150 IQ to cook up some novel bioweapon.

1:07:17Now, 110 IQ person who can follow some instructions might be able to get that information because of an LLM. And lower tacit knowledge. Yeah, and lower tacit knowledge and so on. Like what solutions do you see that are, that minimize the threat to free speech, but also reduce those very real risks? The national security people I've spoken to tell me they don't feel it has been that hard for quite a while now to come up with very nasty things. And if you think of the sarin gas attack in the Tokyo subway, that's now what, over 30 years ago? My real worry, which I don't know how to fix, is simply that evil terror groups will turn to LLMs for basic management advice.

1:07:57And they'll say things like, don't purge all your members, don't act like a jerk, and 10 other things that are like beyond stupid, but actually will make them far more effective right teach them social tools yeah and i don't know you know we just have to be better too like we'll use the ai on average better than they will so it's an arms race we may not always win but on average they have they have entropy on their side that's the problem like they've always had entropy on their side and we're often our own worst enemies right that's a bigger problem than them destroying us i think right again but if it's amplifying if it's making their job easier than that that differential the offense versus defense differential is going to be even more amplified at higher and higher stakes on an increasingly well in some ways less robust society because most of our institutions are part of shit let's be honest i think it will do more to help the good guys law enforcement will use it there might be too much surveillance for your tastes i worry about that oh i definitely i also worry about that but it's going to hurt the Dead guys.

1:09:01Simultaneously, you know, like this idea of sort of anarcho-tyranny, where like the surveillance are weak, surveillance capitalism will be applied upon, you know, the nice law-abiding people because they're easy to do it against. And then meanwhile, the actual real threats are simultaneously able to get away with using LLMs or whatever information dissemination AI technologies enable them to carry out these very, very dangerous things or make their criminal enterprises more efficient? I think surveillance will win. I don't see how the bad guys stay out of the networks of surveillance. And again, you might think this is bad.

1:09:42I have my own concerns. But that would be my prediction. Big bioterror attacks. Yeah, I mean, that makes sense to me as well as a state that we get to. It puts the risk onto another side, basically, right? because now the question is another dystopia right it's another version where now the surveilling entities may just have too much power uh for some yeah for my taste at least in some ways it will be like a return to early human society where you lived in small villages you didn't have any privacy for tens of thousands of years right more and it wasn't pleasant in various ways. But we're biologically somewhat used to that.

1:10:25And that's one reason why I think it will evolve. Yeah, it's like you don't have any privacy in a small town. But if you live in the big town with people all around you, that's where you have real privacy. And that may have been a temporary stage in human evolution. Yeah, maybe. On the back to the AI risk point, you also expressed that maybe we should be seeing this in markets already as the other great information acquisition tool of humanity. And I wonder, would we actually see it in markets? Like what does it require for markets to really start pricing, including information about risk? Like looking at COVID, for example, it seems like a lot of people and experts were reasonably already assuming the pandemic risk to be very high in January and February, and the markets just lagged massively.

1:11:19But people also know something like COVID. In fact, as we saw, the market ended up going up. So it's not obvious you should have been shorting things, even though there was the short-run panic. I'd put it this way. I talked to a lot of people on Wall Street. I talked to a fair number of people in the national security establishment. And we all talk about the Doomer argument. These are like very, very smart people. The national security people, they're just trained to worry about things, think about risk, big risk. The Wall Street people have massive pecuniary incentives. I've never met one of those people who buys the Duma argument.

1:11:54Not one. What is the Duma argument here, by the way, just to clarify? Well, the leiser would be the extreme example, but the notion that a collective superintelligence will be nothing like us and will decide it doesn't need us and destroy us all. There's weaker versions of it. That's the extreme argument. I suppose, yeah. So that's more the kind of like rogue super intelligence goes out and decides to be against humanity side. But it's not rogue from its point of view. It's just rational. Right. I mean, there are many other risks that I think people are concerned with. Yeah. How AI disseminates.

1:12:28The smart people I know don't believe in it. So what I see sociologically, it's a person's proximity to a certain network of so-called rationalists, San Francisco people, Bay Area communities that predicts whether they believe in the doom or argument. And that to me is a big, big warning sign. It's the same with the anti-vaxxers. You see circles of people socially connected who aren't so keen on peer review, who talk each other into this thing. And it's a big danger sign. Yeah, though I wouldn't have characterized the Duma argument to be that. I mean, if you do, then I would agree that many people don't see that as that likely.

1:13:08But I would have included other things like, for example, that over time, more and more of our economy will run on AI services and be mediated by where a lot of the effective decision making will happen by AI agents. Oh, sure. That's true in all scenarios, pretty much. Right. And as that progresses, if we don't have further developed interpretability or understanding, and it's like we lose some understanding of how our systems work because the decision-making gets done by the other systems, one thing that people worry about is that we lose oversight and don't notice how we start optimizing to a state that we didn't want to get ourselves towards as humanity.

1:13:51That's the whole history of civilization, in my opinion. Like we never have been in control. I think there's this big illusion like, oh, you know, we'll be deposed somehow from controlling what happens. It's no different. I mean, the particular details are different. But that doesn't mean it's necessarily a surviving state. Well, let's say we have five or 10x the wealth. I mean, I'm going to leap at that bargain in a heartbeat. And we live to 97. That's great. But what trade off? at what risk rate? You wouldn't leap at that bargain if it's an 80 percenter. You would leap at it if it was a one percenter, right?

1:14:30Well, one percenter of what? Again, there's going to be plenty of problems that pop up that the AIs will help us solve, unless you think, as Eliezer does, they all go on strike and then eventually do away with us. He doesn't think that. He thinks that there will be a, like, basically goals, because they're black boxes, there is the goal convergence mechanism basically is going will be forever unpredictable no read the title of his new book it's this will it'll kill us all for sure is what it says well he doesn't he says if you build it yes if you build a generally intelligent super yeah sure yeah but okay it's a title of a book it's not it's not the endorses it i've talked to him i've read him he definitely endorses it yes yes so you said that this is just kind of more yeah we were already not in control before and etc.

1:15:20It's interesting. So would you take, so currently we have some externalities and we are hoping that better state capacity or philanthropy or something manages to internalize the externalities that we have to climate change, for example. Yeah. Right. So if we just sped up and kept the externalities where they are and just like increased their production rate as well, that seems to be one of the scenarios that I'm worried about. An acceleration of misalignment, basically. Is that something where you think that we will just, well, if we have those better AI agents and we are smarter, then we'll also figure out how to better internalize them kind of naturally every time.

1:15:58No, there's ways it can go badly. Like we already talked about the scenario where energy use goes up along most dimensions and we fry ourselves. It's not the end of the world, but it's pretty bad and ugly. And I don't rule that out. I'm not sure if it's probability. But it certainly wouldn't shock me if it happened. I wouldn't blame the AIs. I know they would contribute marginally in the form of the data centers, but it would have happened anyway. I would certainly not blame the AIs as well. And again, there's some chance they can come up with a fix for us. So do you think the current, given the sort of various tragedy of the commons, scenarios we've talked about, collective action problems, the tendency towards dopamine capitalism and increasing rates of addiction?

1:16:48I wouldn't say increasing rates. Like for some things, a lot of things, addiction's going down. Sure, okay. Do you not see the current state of our economic model if left completely under status quo conditions and just with a ton of essentially power pumped into it, which is what these new AI agents are going to be. They're just going to accelerate everything across all domains from the really good stuff, healthcare improvements, to the depletion of the oceans, because now the Chinese deep sea trawlers can use AI to even more efficiently map out the location of the tuner and sweep out everything.

1:17:29Do you think that it's fine to just speed up the current system even more and will be fine? or do you think that actual design changes need to be made to the current economic system? For example, more in the direction of Wealth Plus. Well, you need design changes all the time anyway, with or without AI. So we've been depleting fish for a while. Again, it's not an AI thing per se. We need a better system of property rights, international coordination. I don't know if we'll get one. I haven't studied that topic, but it definitely would be better for us and better for the fish. So, you know, the list of things like that, there would be thousands, and we're not going to get them all right.

1:18:14But if you're asking, like, will we be fine, the best signals I see suggest, yeah, if you're born in the United States, you'll be fine, or the UK. Which of the arguments do you find most convincing, not necessarily from Eliezer, but that you've heard from others as well, given that you've been around the community for a while? Well, I think the biggest worry is simply that we have a nuclear war. Again, this can easily happen without AI and that it gets out of control just maybe by accident even. And the chance of that in a given year is quite small. But if you run through enough years, the cumulative probability becomes much higher.

1:18:52That's by far my biggest worry. I hope the AIs help us distinguish, you know, what's a flock of birds versus an incoming missile. But we know nuclear weapons can destroy most of the world. It's a very simple mechanism, and we've come somewhat close to it a few times. I don't see why that isn't everyone's biggest worry. I mean, it certainly on many people's minds has been over the last years, I would assume, right? With Russia, Ukraine, and the South China Sea. yeah well i don't yeah and also the sort of perceived race to intelligence between super intelligence between china and the u.s that even even if it's not actually happening the concern that the other side might develop you know that's a lot of the argument for people to say we must go as fast as possible because we don't want china to beat us to super intelligence they'll say that in one breath and then at the next breath be like but oh the super intelligence is is a while away and we don't need to worry about the other guy you know about about the other guy and the trouble is is that like that's increasing the risk of for example someone thinking it's a good idea to do a first strike on the others to take out their data center.

1:19:56So yeah, I also am extremely concerned about nuclear war. I worry when people use superintelligence, or it used to be AGI, but the terms shift, as a discrete stage that is somehow reached on a day, and then one place has this insuperable lead. I doubt if there's any clear line, and the clear line can cause you to make a lot of errors, like they will get there first. I think most of the time, maybe all the time, will be in the lead. But there's not this clear benchmark as there was with, say, the Hiroshima bomb. And the gradated nature of the discovery should actually put our minds a bit at ease.

1:20:37And our nukes are still going to work. China could have, quote unquote, super intelligence before we do. It's not going to magically like turn around our nukes and send them back to hit Knoxville or something. Well, I mean, if it's also a super persuader, you can't keep all the humans off the grid. Humans are not that easy to persuade. We're too stupid or something. My experience is that humans are kind of easy to persuade. As a poker player whose job it was to convince people of one story rather than another one. You have a lot of people in nuclear submarines with missiles that work, and you've got to persuade a lot of them.

1:21:14I don't think it will succeed. I think China will be afraid of our nuclear deterrent for a long time, no matter what happens with AI. Well, but that's under, I think that's underestimating the power of a super, like, again, if we're playing this game of talking about a super intelligence, we should also, that is like a 1000 IQ in various physical... I don't think that means anything. You're making it a god. It will be very smart. But for it to actually create a set of systems that would capture all of our ICBMs, it's a long ways off. Yeah, I mean, if it's spots where you can't get in with the AI because they are disconnected from communication systems, that it would have any reasonable way to persuade you, then I'd agree with that with nuclear submarines, etc.

1:22:07but I think there is an open question around how much headroom is there for AI agents to be above human capabilities like something like tic-tac-toe the AI will never beat me in however smart it is and there's very little headroom there right but some other areas like writing or research analysis there seems to be quite a lot of improvement ability and to me it seems the persuasion dimension. I think people come out in different sites here, like how much headroom there is above just standard human persuasion. But I can well picture that agents will be, AI agents will be able to be better persuaders than kind of the best, most charismatic politician.

1:22:52Oh, I don't doubt that, but they're not going to persuade American leaders to surrender to China or vice versa i i mean there's some evidence that you know if you look at some of the internal sabotage that's gone on by politicians within the u.s for example to make the southern border unbelievably porous right i mean i don't know what your views are on that but from what i've read and inferred like there's clearly been some someone within the chain has decided that it would be better if they're actually there's massive well the democratic party decided that right right whether one agrees or not right but probably not the whole party as a whole no sure but that's but something probably a foreign actor convinced them that this was actually a good thing that they were doing this they thought these people were going to vote for them like they might have been right even it's not some insidious foreign force that putin bought some facebook ads or had his bots put out some tweets and the democratic party you know pulls its forces from the border it was a normal political calculation that might have backfired of course explain okay explain the islamification of uk you took in a lot of muslim migrants maybe it was a mistake but you did it well because i mean there are still british politicians like clearly it seems like from a rational perspective they're arguing against their own self-interest by being like no no no it's fine no actually islamophobia as a problem and there's having this incredibly, again, relaxed, overly relaxed, in my opinion, approach to that immigration.

1:24:25It's a normal political mistake. Big mistake, I agree. And Starmer finally speaking up. I'm not sure he'll do much, but at least it's politically savvy because it's what voters want. And you see this in many, many European elections, right? Voters want something done. I have very little of it at the feet of external propagandists. so you it sounds like you're more of a mistake theorist than a conflict theorist or well in this particular instance i think it was normal political mistakes okay and facebook ads did not elect trump and it wasn't putin and all the different claims you hear they're mostly wrong maybe entirely wrong yeah i it seems to me that a kind of multi-pronged approach to get an outcome done in terms of persuasion is quite possible because one, populist changes on basis of distributing some information.

1:25:23People are quite amenable to certain memes and ideas, right? The phrase get back control was, I think, quite helpful for Brexit and the specific phrasing matters. And I think that on one hand, the AIs will be able to figure out those better and more persuasive statements, given the population mix-up and the specific messaging that would convince various individuals, as well as then it seems like in some structures, like the Democratic Party, for example, and the example you were just talking about, it doesn't require even to convince all of the politicians. It actually requires convincing some individuals.

1:26:03So you can do targeted individual manipulation and persuasion and kind of like broad society persuasion to get outcomes done. I think AI will improve our views quite a bit. And at the current moment, I know things will change, but it's more objective than any media source, I would say, by far. Well, provided that the owners don't put their own slant on it. Yeah. I agree. I do agree. But even Grok is more objective than anything on the right or the New York Times. Yeah, the current media has decided to be taking political sides for the most part. But that's history. I mean, history of Britain, media, past centuries, my goodness, or US.

1:26:47We had this 30, 40-year window when it pretended to be somewhat objective, even though it was not then. So this manipulability, it's one of my smallest worries about AI. Interesting. I, yeah, we probably won't be able to dig into it, but is there something you would see that would change your opinion about manipulability? Where, say, if you saw more people that used, there was some bias programmed into an LLM that later gets found out, and that leads to, like, the people that were interacting with it developing this now biased opinion as well. Or is there some instance that would change your opinion about manipulability?

1:27:29Well, most voters have pretty, you could say, gross preferences. They want low inflation, a good job, political stability, opportunity for their children. You could add to that list. If the AIs could talk people into not wanting those things, I'd be very worried. But I really don't think they can. Do you not think that TikTok, I mean, there's a lot of evidence that TikTok has been, And certainly it seems like the CCP have been using TikTok as a tool or have been trying to manipulate the trends on TikTok to have certain topics trend more than others, particularly around, for example, Israel-Palestine.

1:28:08You know, they were pushing one side more than the other. There's evidence of that, which would suggest there's some value in doing it because it's quite a risky proposition for the CCP to do. Well, I would say all media in this country for a long time have made Middle Eastern politics too salient compared to, say, civil war in Sudan or Rwanda in the Congo. It's a longstanding problem. Maybe TikTok made it worse. Could be. But, again, very low on my list of worries. It's not going to talk people out of wanting that list of things I gave you a moment ago. So that and then a few of the other topics that we discussed, it seems that kind of like if I had to make a very short summary of the uses that in fact many of the problems that we describe have been occurring in the past as well.

1:28:58This may be and this is just a continuation of them and our way of dealing with them will be able to still find new solutions to the same problems. Yeah. Even if they are more AI enhanced problems, then kind of like the security dynamic of offensive and defensive is not going to go crazy out of whack and crazy out of balance. Just because you add AI, you're improving both. If we can't handle having more intelligence around, I strongly suspect we're doomed anyway. All right. So I very much believe in that in general and that society as a complex adaptive system will adapt. But there are these points where if one area grows a lot and the others can't adapt in time, like how quick is the adaption speed basically, that that can put things out of whack and hurt the system as a whole.

1:30:03Yeah, sure. I mean, bad things happen in all scenarios with or without AI. Sure. Right, but what I'm taking from it is also that if I had the view that the diffusion is slow and therefore, say, the technology-enabled things are not happening at such speed that the rest of society cannot adapt, the rest of society being governments, the ecosystem, etc., you actually see it as being sufficiently slow such that it allows the other parts to adapt. Except for self-contained systems where diffusion can be pretty quick. Okay, yeah. I suppose if I also saw diffusion as being that slow, I would be on the side.

1:30:44I kind of assumed that there will be ways to do it faster, but we also wanted to get into other topics. Would you describe yourself as a competitive person? I think I'm very obsessive and not that competitive. Were there any moments in your life where you either wish you were more competitive or less competitive? I think it's done well by me not to be so competitive. Now, I might be more competitive than like the true population average, but I would say for a person who is sort of known and out there, I'm considerably less competitive than my peers. And I think it's made me quite happy. I don't wake up in the morning feeling mistreated or like someone else did better or they have more Twitter followers.

1:31:29All that is so foreign to me. I wake up every morning. I'm productive. I'm pretty happy. I get to it. I just don't have those negative thoughts. It's been great. And has that been the case for your entire life? I was born that way, my mother told me, and that seems true. Yeah. Yeah, see, I was the opposite. I was born unbelievably competitive, always looking for essentially like who got a higher mark than me in this exam and really, you know, this sort of external comparison. So it's interesting to hear because it does seem like some people are just truly born that way and others are born. And I think you can learn to become more of the other.

1:32:07through exposure, your peer group, et cetera. But it does seem like an actual nature thing more than, well, at least some big chunk of it is nature rather than nurture. So it sounds like you've probably never had a rival. Never had a rival. And I quit chess young in part because I realized I wasn't competitive enough. I didn't truly hate losing. And that doesn't work. You see like Vichy or Kasparov, Magnus. Man, losing to them, it's literally a torture. Yeah, they hate it. them talk about it it's so foreign to me it's like okay i was very very philosophical about losing and that means you should be doing something else you may see the same you know with different poker players right oh yeah i mean one thing that the almost all top poker players have in common is that we hate losing yeah um or at least don't you feel that's a trap like talk about addictions oh totally it's one you can't get out of easily yeah and i it took me you know i wanted to quit poker professionally at least four years before I ended up doing it because it was so much of my identity was wrapped up in it and you know I fell out of love with the game um have you really done it have you really quit well yes I would say so because I I quit I I hung up my hat officially I even like tweeted out I am retiring and in 2019 um I basically didn't play for four years and then randomly well not randomly but four months ago I went back and played a big tournament I saw that Congratulations.

1:33:31Right. I had this biggest win of my life. So there's something in the stepping away and coming back with fresh eyes and probably not having your whole identity wrapped up and actually doing well. I was able to actually see the beauty of the game for what it was and just have a fun time. You definitely played better than you have. Then I would have expected just by pure deterioration of skills. I think that part mattered. Well, I stopped trying to be all like a game theory optimal person and just played very intuitively. Right. And that played off. I mean, you definitely have truly quit. You've played two or three tournaments a year over the last six years, whereas prior it would have been 100 in person and 1 ,000 online tournaments.

1:34:08So you haven't quit. Congratulations. Do you play poker at all? No, I played like when I was eight with the five-card thing, and that's the last time I touched it. And I don't do any gambling. It's a total mystery to me. I don't want to sit down and play games with people. I have no interest. I like to travel and explore and be curious. Well, I suppose the one thing that I find very interesting about games, and we have a deep games crew in London especially with very, very competitive players, is that it creates this contained universe within which you get to maximally live out competitiveness and zero-sumness even, but in a contained way, such that the externalities are arguably positive because you had fun, et cetera, right?

1:34:59And you actually discover, because you put each other in such situations, it's kind of psychologically interesting to see what you discover in each other. Is that not the type of thing that you're curious about? I'd rather sit down with people and talk through ideas, which I do all the time. Podcasting, in a way, is a version of doing that, right? No, you certainly get to understand and meet people through that. There is something different where you see a different side when you put someone in an extreme situation, which you don't discover when you just sit and chat. I would still, I still spend most of my time talking to people rather than competing against them.

1:35:39And I would choose to be so as well. I like traveling with people. It's a good way to see what they're really like. Yeah, that also actually puts people there. What do you think can be done then to make people have a healthier relationship with competition? like you? I don't know. People are born a certain way. There's a distribution. They shouldn't all try to be like me. So I don't think happiness is the goal. And this also relates to our social media discussion. Social media do depress some people, but they put them in touch with the people they want to be in touch with. Often it's family relations.

1:36:14Family relations do not always make you happier. Like we've all learned this one way or another, but you still want to have the ties. so I don't know I agree that is one of the like under under championed points I don't think people should feel bad about being competitive or quote-unquote too competitive it's a way of being there's some value to that authenticity it gives a lot of external public goods when people create I'm not looking to dampen it really well at the same time a lot of the sort of tragedy of the commons and arms race type dynamics that we see in society uh i know you're sort of more skeptical on the the negative mental health impacts on for example of young women on instagram or whatever but like those are often driven actually by sort of misaligned competitive incentives in part because girls want to be you know they want to be they want to have more followers than their other friends and and so on that that is a manifestation of competition perhaps in a more feminine direction but it's still a manifestation of it and that seems to be making people unhappy because I think the thing with competition is it's very much an external locus right you're not I think competition is healthy when you're competing against a previous version of yourself in some ways that's that can be a driver that in a more positive direction but when you're only competing against outsiders uh especially for sort of an evaluation of self-worth I think that's where you get this sort of decoupling between well-being and competition.

1:37:44And it also is somewhat the driver of these race to the bottom type dynamics. I'm hoping AI can help at least marginally with some of these problems. It'll monitor your social media presence, tell you what you're doing makes you less happy, and then offer to do some of it for you. I'm not convinced that's what we'll get, but it's one of the hopes I hold. Yeah, what design principles would you put into such an AI? Well, you would just tell it what you want it to do. You'd say, hey, AI, I feel bad sometimes after social media. Monitor what I'm doing and give me some suggestions and offer to take it over for me when I don't want to do it anymore and to see how it goes.

1:38:20And then ask it two weeks later, am I feeling better now? That sounds kind of Brian Johnson-y because he's one of his thinking is like, if you had an algorithm that would tell you every way to live your healthiest and most physically healthy but also mentally healthy life, would you take it? But I think the algorithm itself will tell you to limit the use of the algorithm. Like here's three or four problems you could fix and still be authentic. Don't let it control whom you're going to marry and 10 other things, but at least do these three things. But then you're still listening to the algorithm effectively.

1:38:52But that's fine. You'll listen to it once and forget about it. And you'll still be you. No, I agree. Though, yeah, I wonder what happens though So if, say, you implement such an algorithm in one of your social media apps and it tells you, well, you should spend 30 minutes on these topics and maybe another 15 here, and those are the 45 minutes that the people got. But now another company develops an algorithm that just simply optimizes for getting more attention and engagement by the users, and people end up spending three hours there rather than 45 minutes on the positive one. Is this the one that captures more revenue and through that kind of just ends up winning?

1:39:33And do we currently allow for even the long-term social media company to win? I think it's going to have to make you feel good to keep your attention because the notion that you need to be on it to stay current won't be true anymore. You can get your AI digest whenever you want it. Again, that's my hope. I don't have a firm prediction, but it seems to me it's quite possible it could work out that way. Mm-hmm. You mentioned you've done a lot of traveling. I'll probably have to define it, but I often talk about the idea of a win-win type mentality. Certain cultures can have more win-win mentalities where, for example, they're better at coordinating.

1:40:19They generally have... Someone is win-winy if they feel happy for someone else's success, whereas someone is sort of win-losey when, you know, again, they had that more competitive mind spirit. Zero-sum mindset. Yeah, zero-sum mindset, effectively. So in your travels, is there a culture that stands out to you as being more win-winy than the rest? U.S. of A. And I wouldn't put us at the top of every category, to be clear. But I think on that particular, I don't know any other country that comes close. Why is that? I've been to over 100 countries. It's one of the ideas the place was founded on, that people have this right to life, liberty in the pursuit of happiness.

1:41:02And a lot of people here really believe it. And there is this opportunity and wealth and it's mostly worked. It's been great. That's why we happily moved here in part as well. There is no other place where people are okay with just people can be very ambitious, explicitly so and other people are actually in favor also and want someone else to win and then it kind of together creates a growth of the pie rather than a zero like divvying up the existent pie right

1:41:38what's one of your weirdest or if you're even open to the word wooiest beliefs well i don't think it's a weird belief but i think there's a reasonable chance that the uap data we observe are actually alien drone probes. I consider the opposite the weird belief, but whatever, other people would call that weird. So why do you think the evidence is in favor of that? I would say this. The evidence I see in the public sphere is not at all convincing. But when I talk to people with very high security clearances or people who've flown F-16s or done other related things, they tell me things that cannot be accounted for and I don't know what to make of it.

1:42:22I don't want to leap to any particular conclusion but I think there's some chance it would be objects, machines from another planet. If you think about how many habitable planets there are I know the distances are great but it doesn't seem so unlikely a priori either and maybe we're being observed in some way. Did you ever read the paper by Toby Ord and Anders Sandberg, the FHI guys, about dissolving the Fermi paradox? Yeah. Did you buy their arguments? They might be right, but even then, there can be quite a few alien civilizations. And since that paper, there's all the more evidence for a lot of habitable planets.

1:43:06So we don't know the number, the initial calculation, that it has to be a truly enormous number. They showed that was wrong. But they didn't sort of prove there's no life out there. And it seems to me much more... They just thought it was improbable. Yeah. I think it's more likely there's quite a bit of life out there than not, though I'd say we don't know. I mean, they came to... If you're just solving a paradox, it's sufficient to prove that it wouldn't be out of the ordinary to not observe it, right? So that's why they said that given their way of running the calculation, you get to like a 50 % chance in the universe that there is no life outside of us.

1:43:44Right. Which is sufficient. It's not saying that there is certainly none. And... You know, those betting odds, I'm against them. I want to get onto your poker table because I think there's life, you know, on the moons of Saturn and or Jupiter. It may not be interesting life, but I bet 50 % on that, I think. Oh, no, no. It was straight... Fair. It was about life as advanced as us, I think. It's very hard to know. Life advanced enough to be sending... Exactly. Exactly, to be sending out radio signals, because that's what the Fermi paradox is actually remarking on. It's like we would expect there to be radio signals banging around the galaxy, but we don't seem to be picking any up, at least carrying any meaning.

1:44:24I wouldn't assume it has to be radio, but we don't know often when someone is communicating with us. I just think it could go either way, and there's clearly some data we can't explain, and we don't know. I don't think there's any conspiracy. I think the government itself does not know, and we ought to keep an open mind. Do you have any other worried beliefs that people would be surprised? Again, like, what's a weird belief? I've maybe lost track of them. Like, I don't think people should drink any alcohol. Is that a weird belief? It's a very common one, though. But it's a very common one. It's definitely a common one.

1:44:59No, I mean, have you ever had any weird experiences you can't explain? No, never. I don't believe in the supernatural. I haven't gone through my life like believing in UFOs or ESP. I even was never convinced there was more than one Kennedy shooter. I hardly believe in any conspiracy theories. What's one that you think is true? Well, I do think the, you know, Pfizer withheld the vaccine or slowed it down to help Biden's chances against Trump. And that was a conspiracy. That's the only major one I can think of that I believe in for the US. I thought they explicitly said so, even at the time. Not that they withheld it, but that they just only gave out the success after.

1:45:43That's why I believe in it. Now, if you're talking about totalitarian countries, I do think there can be many more conspiracies. But they're not things that people here talk about. But like, was Stalin poisoned? I'm not sure. But I think there's some reasonable chance he was poisoned. That would be a conspiracy. see. But the ones people talk about for here, it's just too hard to keep secrets. They dumped the Kennedy files. We didn't see anything. I don't believe in much in that direction. All right. Well, to finish up, we always end each episode with a series of rapid fire predictions. You are one of the kings of predictions, so this should be in your comfort zone.

1:46:26Try and... Ideally, this will be a system one answer for you. um you're allowed to engage your system yeah you are allowed to engage your system too but yes uh so do you consent to this to this exercise absolutely consent i don't believe in system one system two by the way so you're just gonna get the only system i've got you don't believe in it no i don't believe in it at least not for me but i suspect for not for most people most of that book by conman has not held up well okay all right so you were mentored by Thomas Schelling. Correct. The famous game theorist who came up with the concept of a Schelling point.

1:47:05So the probability that you'll ever attend the yearly Schelling point meetup. I didn't know there was a yearly Schelling point meetup, but I don't see why I should go. So the probability is below 1%. Well, you can actually work through what it is. It's Grand Central Station. Well, where would do you think would it be? No, you kind of answered part of it. January 1st, noon at Grand Central Station. Correct. But it's not even focal anymore. Like it's Penn Station or Moynihan Hall, I guess they call it. No, I'm in those places too much already. Why would I go there? Oh, you're breaking the shelling point.

1:47:40Okay. Probability that AI systems collectively exercise more than 50 % of effective global decision power by 2050. 2050. Oh, that's probability below 1%. Okay, 2080? It's a big world. 2080. It's a little hard to define the question. 15%. Probability that by 2035, there will be at least one three-person or fewer unicorn startup. Oh, quite high. I mean, we had mid-journey with seven or eight people, which is more than three, but we're almost there already. So I'll say 84%. probability there'll be 10 of them by the same date. 76%. Probability that an autonomous AI agent system publishes a peer-reviewed paper with zero human co-authors in a top 20 science journal by 2030.

1:48:402030, you said.

1:48:4775%. Wow. Now, probability that the real inflation adjusted US GDP growth averages above 4 % over 2030 to 2035. Oh, quite low. I mean, you would need first a big depression and then a bounce back, which is probably not what you meant to ask about. But I'll just say very low. Okay. Well, then. Well, but still that it's very low being like 10 % likely. or if you had some natural disaster combined with other problems and you lost like 17 percent of your output like could you bounce back quickly over four years with those rates yes maybe but that's the scenario not like oh we just get so awesome so that's basically like 2022 or or 2021 post-covid right we bounce back exactly really high yeah that could happen but it would be that weird bounce back case but otherwise you expect three percent ish oh 2.7 i don't know okay okay oh wow so four percent is actually that low so then we also had the question of uh above eight percent and above 20 i take it those are tiny tiny yeah epsilon yeah uh probability the the u.s uh unemployment rate tops eight percent at any time between now and 2030 because of rapid AI or robotics to fusion?

1:50:12Tiny. I mean, if we make monetary policy mistakes in conjunction with it, that could just happen without AI pretty easily. But because of AI, below 1%. Probability that the median U.S. life expectancy rebounds to over 80 by 2030. It's so complicated how they calculate life expectancy. I can't give you a good percentage. stage. It seems to me too soon for that to happen. So we're now at, I think, 78.7 or so. It's finally started growing again. Yeah, it could. I don't know. I'd have to really do work with numbers, but I don't think whatever happens, I don't think it will surprise us quickly. It'll just continue the current trend.

1:50:58And if we believed in the, well, given the discussion around pharma improvements and healthcare improvements. Likelihood that median US life expectancy grows to over 90 years of age by 2050? Not by 2050, but again, I think young people today will see that with a pretty high probability, 80%. So you're pointing at 2100 or something. You've got to rule out nuclear war accidents, but I think most people will die of old age if we allow them to avoid true catastrophe. Probability that COVID was a lab leak? 50%. All right. Noncommittal. That Epstein killed himself? Well, Dan Bongino just told us he did.

1:51:39And if anyone would opt for that conspiracy theory. Oh, I didn't see this. So I'll say 93 % that he killed himself. Okay. And finally, the probability that David Grush's claims that the U.S. government has secret UFO retrieval program and possesses multiple spacecraft of non-human origin, as well as corpses of deceased pilots. The probability that that is true. Minuscule. Too small to even calculate. Awesome. Well, thank you so much, Tyler. It's been a pleasure for me. And look forward to seeing you both next time.