If the Robots Are Coming, Why Aren't Interest Rates Higher?
Transformative AI, Existential Risk, and Real Interest Rates
The source recording for this episode has no speaker labels, so this transcript runs as continuous dialogue without attributing individual lines.
>> Welcome, everyone, to the Justified Posteriors Podcast, the bi-weekly podcast that updates our priors about the economics of AI and technology. I'm Seth Benzell, maxing out my borrowing limits before the AI spends my money for me, coming to you from Chapman University in sunny Southern California. >> I'm Andrey Fradkin. Recording this podcast so that the AGI is treating me well in our positive AGI future, coming to you from Cambridge, Massachusetts. This podcast is brought to you by the Digital Business Institute at Boston University's question school of business. So Seth, what are we talking about today? What's on deck? >> That's a great question, Andrey.
Today, we're talking about how would you invest your money if you were told that in five years and 10 years, we're going to have the artificial general intelligence revolution, that in the near future, not right now, but in the very near future, we're going to have AI technologies that stupefy the imagination. How do you take advantage of information like that? What would you do if I guaranteed you that that would happen? >> Well, Seth, because I believe in the efficient market hypothesis, there's nothing I could do to take advantage of that situation. >> But rather, it says that you can't beat the market, right? There's some sort of sense in which that you wouldn't, it's unlikely that you would have some sort of special insight into the development of AI that would allow you to trade at an advantage versus other people. >> But yet, still, even with everything about the future being common information, prices evolve over time.
And in fact, this paper that we're reading has a very explicit hypothesis about what the impact of future developments in AGI, or future expectations today about future developments in AGI, what that's going to do to the price of assets today. Are you curious to find out, Andrey? >> Do enlighten me, Seth. What is the claim that this paper is making? >> All right, so today we're reading a paper titled "Transformative AI, Existential Risk and Real Interest Rights," who authored by Trevor Chow and Basil Halperin of Stanford, Basil being a friend of the show, and Jay Zachary Maslisch of Oxford and the Global Priorities Institute, GPI being one of these organizations in Oxford, that sort of a long-termist organization has been thinking about computers in AI and AGI for a long time.
And if I had to kind of summarize the argument of this paper in a sentence, it's that right now, interest rates, real interest rates, so if you save money, what's the amount of payment, additionally, you will get back from saving that money a year from now, or a certain amount of years from now, that that number is very low. And then the second claim of the paper is that is incompatible with AGI or advanced AI arising in the near future. So in other words, anyone out there who's worried AI is going to take your job, AI is going to change the world, AI armies are going to be marching across the continent, you know what, take a sigh of relief.
Basil's got your back. Basil says, "Not happening anytime soon." What do you think of a claim like that, Andrey? Do you want me to give my prior? Yeah, why don't you give me a prior before having this paper, what would you have thought of a claim like, low interest rates today indicate that markets do not believe that artificial general intelligence is imminent. How would you evaluate that? I guess I put a pretty low prior on that claim. Let me explain kind of where I'm coming from on that. I tend to be skeptical of the ability of macroeconomists to predict anything, and especially people in finance as well.
They don't exactly have an amazing track record of forecasting things very well. I think there's kind of a broader kind of like under scoring claim there. You're meant to kind of naturally think that if the market is not predicting it, that it's not likely to happen. But I guess another way to try to evaluate that claim is to also look at a prediction market or something like that, which seems like a more direct way to assess what people think the likelihood of AGI will be in the future. So I think I'm a little skeptical of the macro models, and I'm a little skeptical of using an indirect instrument to test for this market, believe it rather than a prediction market.
What do you think Seth? Well, that's a good question. So I'm looking up right now. What does metallicus think is the chance of AGI? So do you have a specific metallicus AGI prediction in particular? I should be looking at that you think would do a better prediction job? I'm also looking it up right now. It would guess that we could try to use whatever Basel is thinking, and we can get to that because that's not a claim about AGI, that's a claim about economic growth. That's an interesting point there. But if we look at, let's say, when will the first general AI system be devised?
And Metakelus, which is this prediction market, what is the community prediction? It's a bi-modal distribution, Seth. I love it. And there's one mode that's towards the end of this decade, and then there's another mode about 90 years from now. So I think if you were taking this literally, we would say that there's some probability it's imminent, and then there's some probability that it's going to happen quite a bit later. Yeah, no, this is a fascinating graph with the two modes. I've had one mode in three years, and then one mode in 100 years. All right. So maybe let me tell you a little bit about how I evaluate this claim.
Andra, I think you were going to have a disagreement today because I found myself before I read this article, strongly in agreement with the idea that low interest rates indicate that markets do not believe that really revolutionary AGI is imminent. And I'd put that claim at about 90%. Now, I think two areas that we might have diverged on are first that, you're saying that sometimes markets are dumb, right, that maybe markets aren't good at predicting things about the future, right? I feel like that's not engaging directly with Basil's claim. I think Basil's claim is a claim about what markets think, not about what is actually going to happen with AGI.
And then maybe he goes on to argue that, and also markets are pretty smart. The other reason that I'm persuaded by this argument is for reasons that are not super, I guess we'll talk about this more in the evidence side. But I'm in just on its face. If somebody tells you, I'm going to come around with a really new exciting technology that you can build a factory and make output from this new technology that's going to be way better than the factories you could have built in the past. Just mechanically, that's got to increase investment demand. And, so there you go.
So I believe that AI will increase investment demand. That's got to increase interest rates for a certain level of savings. Interestingly, the paper really focuses on the supply of savings side of the ledger. But I think like, even before thinking about the supply of savings, yeah, like if technology explodes, there's going to be a lot of really exciting investment opportunities. And on its face, that should lead to increases in interest rates. So I feel pretty strongly about this claim before I even read the paper at maybe 90%. So are you willing to put a number on your evaluation, Andrey? I guess let me like highlight where I think my prior is coming from.
It's not like, I don't think we can use, supply and demand reasoning to think about interest rates. Certainly, I'm not that much of a macroeconomics nihilist. But I just think that there are lots of kind of frictions in these markets. And generally, things aren't even the strong relationships in macroeconomics that people like to talk about or are actually kind of really hard to disentangle on the data. So, the market believing that AGI is imminent with some fairly high probability might not move interest rate that much. And so that's kind of maybe where I'm coming from there. So as a result, I don't have a higher priority than this.
Let's say 10%. 10% is where my prior is. Oh, wow. Have we ever been we're opposed? Let's see if Basil's arguments can move you a little bit. I think maybe before we get too deep into the evidence, though, why don't we take a minute to explain what does Basil mean by AGI, right? Because, we all have kind of our pet understandings of what it would mean for artificial general intelligence to be achieved. He actually has two really specific definitions here, right? The two scenarios he has in mind are first kind of the eukatastrophe, which is that AI causes GDP to start skyrocketing. So greater than 30% GDP growth rates per year, which would be I don't know, has any country ever achieved a 30% growth rate other than like coming out of like they just got nuked so when the year before.
I don't know. I think that would be unprecedented. I don't think so. And then the second scenario he has in mind is the AI's rise up and kill all humans. In which case, Basil also thinks interest rates would increase in the period of time leading up to that would be the AGI catastrophe instead of the AGI eukatastrophe. So how do you feel about those definitions of AGI, Andrey? Are you happy using those as working definitions? No, I'm not happy about them, of course. How do you never happy? No, no, no, there's some things I'm happy about. But, you're the sake of argument. All right.
Listen, 30% GDP growth, sir. I think that can happen. Yeah, big deal, big deal. I guess what I would say is I could imagine AGI without 30% growth rate easily in a variety. There's a question about which parts of the argument rely on the GDP growth definition versus just humans are going to be really happy in the state or the world. For example, if the AI convinces everyone to go into the matrix and I was one, you go to the matrix and the matrix is essentially paradise. So, everyone will be happy, GDP growth doesn't have to go through the moon. Yeah, I don't know how material it is that is, that's the definition.
That's so interesting. So, okay, so you object to GDP being sky rocketed as a threshold because you can imagine scenarios where welfare is high, but measured GDP doesn't increase because of free goods, free digital stakes in the matrix. I can also imagine a scenario, and maybe we can talk about this a little bit again at the end of our podcast where we're thinking about critiques of Basil's framework. I can imagine a situation where we really do get country of geniuses on a cloud, right, where AGI is going, but that that fails to create 30% GDP growth rates, right? It turns out that we'll have a huge amount of intelligence, but we will be bottlenecked by something in the economy that isn't as elastically supplied.
And depending on where that bottleneck is, I think that could have very different impacts on interest rates. But let's come back to that point later. Yeah, I mean, I think I have to just be clear, of course. Humans still care about land, and land is scarce, and the AGIs don't figure out how to overturn existing regulations. Clearly there will be bottlenecks in the economy, for example. So, the question is like, your load is AGI just a genie? Or are we talking about like, just very intelligent beings who are still constrained in a variety of ways, right? Right. And so, I think the most sympathetic reading of Basil's paper is that this is an evidence against genies.
It's not evidence against extremely powerful genius AIs. Yes. And then what do you think, Seth, about the other side, the existential version of AGI? Yeah. Is that too extreme? Or should we have existential risk without everyone just dying right away? Right. So, he's imagining a future in which that is not useful to move your savings into, right? For his argument to work that catastrophe scenario has to be so sucky that it doesn't make sense to save into it, that's got to be a pretty complete obliteration of the human race. You might imagine, even if there was like only a handful of survivors, it would be incredibly valuable for them to have saved resources into that state of the world.
They literally have to be like zero humans left for there to be no value in saving into that future. So, I don't think his paper is particularly focused on the catastrophic vision of AGI. I think we can talk about in a minute, whether we think if there was commonly anticipated catastrophic AGI, what the impact on interest rates would be. But it is interesting how, in his theoretical models, the absolute worst outcome and the absolute best outcome, point in the same direction in terms of interest rates. Yes. And I guess, we're still not actually getting to the meat of his content, but maybe you anticipate this.
The AGIs, or whatever future state of the world, might just be a state of the world where the dollar is no longer used, or financial instruments are no longer used. And so, we're just not able to save using the standard means of saving, but we might be able to save in other ways. People are going to want to move consumption into temporally, right? That's what I'm saying is that, all right, so I might be able to want to, I might want to move consumption into temporally might be maybe I'll store food, right? Yeah, let's go back to like, olden times, it's very simple, right?
Yeah, very some acorns store some wheat, et cetera. But I might still believe that the financial system is doing because there is an AI running around that's hacking into all the banks, right? And now, the question is like, well, what's the, what's the real interest rate in that world? Well, actually, in a macroeconomic model that there would still be a real interest rate, but it just probably isn't the real interest rate that is commensurate with how we measure it today. This is a really, I think, important point, which is that if you think that people are going to have precautionary savings leading up to the AI catastrophe, you catastrophe.
That's certainly theoretically consistent, but it is not what the agents in Basil's models do, right? So there's kind of like two issues here. The first issue is, do you save into the catastrophe? Basil's claim is, no, you do not save into the catastrophe, either the catastrophe or the eukatastrophe. The second point is if you get your you're just gonna, you're just gonna go through a bucket list, right, Seth? You're just gonna go through the bucket list now before you kick the bucket. Exactly. You're gonna go take your resort vacation in Thailand. You're going to buy everyone around to drinks at the bar because, if that's all there is with my friends, then let's keep dancing, right?
But you're bringing up a second claim, which is precautionary savings might not even show up in the market, right? Because you might not trust the financial system with your precautionary savings. So I guess maybe the argument would be the flip side. So imagine you were someone who was worried about AGI and you normally have precautionary savings to deal with just like an ordinary depression. So like you're normally worried about just a regular great depression. So you keep cash under your mattress, right? That cash under your mattress, in some sense, is not getting the market rate of return. It's not really pinning down the world interest rate.
You can argue that there's some like crowding out effects. However, if you were to dissave the money that's under your pillow, because you were worried about AGI, because that market money was not in the market in the first place, maybe your dissaving would not show up as higher interest rates. Is that kind of where you're going? Perhaps. I think there's a fair point there. I think if it was literally cash under the mattress, I feel like that still saves Basil's story because there's some sense in which your cash by being out of the market is disinflationary, et cetera, et cetera. But if it was literally like a stock of acorns in your yard, that was never interacting with the market.
And so yeah, if you dis saved your acorns, that would not show up in higher global interest rates. I think that's a fair point. Anyway, how about you go through the macro argument? Yeah, how about I make, how about I go through the evidence? How about I lay you down some basil slots here? So let's start with fact number one. So there's an old truism. I think this is from Tyler Cowan. Maybe this is Cowan's third law, which is that all claims about interest rates are incorrect. So get ready for some claims about interest rates. The first claim that we need for Basil's argument to work is for interest rates today to be low.
And in fact, if you do look both historically over very long time series, today's interest rates are relatively low by kind of the standards of the long sweep of history, as well as by more recent US history. So right now I have the real 10 year interest rate in the US, according to Fred. Thank you, the Federal Reserve Bank of St. Louis. According to those folks, those fine folks, the interest rate today is around, really interest rate is around two percentage points. If you that's a little bit elevated over where it was in terms of a nadir, both in the mid 2010s and during the recession due to the pandemic, if you'll recall Andrey, there was this all these statements about the millennial lifestyle subsidy and the zero interest rate economy.
Do you remember all that stuff? Do you remember that set? But maybe for the benefit of our listeners, do you want to explain the real and the real and nominal interest rate? You are rugged in for real versus nominal. Okay. So I'm about to buy a house. When I buy, when you buy a house, the man at the mortgage office will say, you're going to pay a 6% interest rate on this loan for the house. That interest rate is a nominal interest rate in the sense that if you borrow $100,000, you're going to have to owe $106,000 at the end of the year.
But you have to remember that in the background of this, the value of money itself is also changing, right? So there's a sense in which if you have $100,000 today, that's more valuable than $100,000 a year from now, because a year from now, there'll be some inflation and, the dollar that used to buy you a Snickers bar now only buys you half a Snickers bar. So when I say the real interest rate, what I mean is kind of the difference of those two. So when you save money, you get a nominal rate of return. So if I had a 6% interest rate, my $100 would turn into $106 at the end of the year.
And then you can subtract up of that the inflation rate. So let's say there was 3% inflation this year. So $100 today is worth what $103 would buy you a year from now. So you can imagine taking the difference. So if there was a 6% nominal rate and a 3% inflation rate, you get a 3% real interest rate. So taking into account that the dollar is getting weaker over time, what's the return to saving money for a year? Okay, so that's the difference between real and nominal. We want to focus on real interest rates, because in the opinion of macro economists in the long run, the important thing are the real rates, not the nominal rates.
Okay. So real interest rates recently at around 2%, they've come up a little bit since the pandemic and since the mid 2010s, because the Federal Reserve is trying to bring inflation under control. And there's this trade-off, again, way into macro. But for reasons that are beyond the scope of this talk, in the short run, if the government raises interest rates, it tends to lower inflation. In the long run, it works a little bit differently. And so that's why we have slightly elevated interest rates today. Maybe Basel would say we have slightly elevated interest rates today, because people think there's a percentage chance of AGI is going back and forth.
What does that compare to historically? Historically, if you look at the mid 1980s, real interest rates were in the 6, 7, 8% range, much higher. That's coming off of the Volcker deflation when Paul Volcker, who was the Federal Reserve Chairman at the time in the mid and late 80s, began raising interest rates in order to bring inflation under control in the middle of the Reagan administration. So if you look at this curve, and we'll post this in our show notes, of real interest rates since the mid 1980s, you see pretty much a slow decline from 6, 7, 8% in the mid 80s to a nadir of 0% or slightly negative in the mid 2010s, and now back up to 2% today.
So I would say to summarize, broad swaths, real interest rates in the United States are coming down. You can tell stories about the year to year pattern, but the aggregate secular pattern seems to be that there's a lot of money looking for investment opportunities that's just not finding productive investment opportunities. The marginal product of investment on its face seems pretty low. And that's why Andrey, you and I back in the mid 2010s, we got the quote unquote millennial lifestyle subsidy with all of these companies that had very, very long term time horizons because of the low interest rate. We're willing to make these big layouts of free products for millennials in the hopes of hooking us forever because if the interest rate is zero, you can make an investment today in order to write it, make money 100 years from now, and it still counts because 0% interest rate, money 100% 100 years from now, so great.
Any comments on the millennial lifestyle subsidy? Did you get anything good? I remember Uber first rolled out. That was fun. It was very subsidized. Uber and the competition being Uber and Lyft was a very excellent thing for my lifestyle, I have to say. Yeah, I mean, the 0% interest rate phenomenon is, very, very interesting phenomenon. I do want to kind of mention something. I wonder if one potential reason for interest rates going down in the 80s was that the risk of nuclear war diminished over the 80s. So right, by the argument of this paper, if you think the world's going to blow up, interest rates go up.
So maybe what you're watching is Gorbachev is lowering interest rates. Okay, now, I know I'm getting a little ahead of myself, but did the markets predict the end of the Cold War? Well, let's look at this time series interest rates. They do come down from 90 to 95 a little bit. I get from 80 from to 85. Just from 80 to 85. Sure. They got what are you saying? The Cold War ends in like 1990. Yeah, it's hard to see it in the time series, right? This is kind of one of the big challenges, Basel's pacing here, right? Which is that everything causes stuff to happen to the interest rate.
So to make the claim that any one particular thing is driving the interest rate, you're working uphill because like the interest rate is one of these prices that everything in the economy feeds into and feeds off of. Would you say that's fair? Yes. Like all the probabilities, all the different things that can happen are accounted for in the interest rate. Okay, so let me tell you what my initial reactions were when I first read this claim. I think I ran into this paper about a man two years ago now and it led to us hiring a Basel for the Stanford Digital Economy Lab, where I think he's thriving.
My initial reaction was interfrates have been in the secular decline. And in my opinion, it has a lot to do with what must be a global saving one, that there's this rise of a Asian middle class and a middle class across the world due to, the strong global economic growth of our recent round of globalization. And there's this huge amount of savings that's chasing very few investment opportunities. And so yeah, if you wanted to boost interest rates, you need to create a lot, a lot more investment opportunities. And given how deep this glut is, I think you'd need really, really strong in technological innovations to overcome that.
But you know, Bay's Basel here is contemplating a technological innovation that's good enough to increase growth rates by 30% percent. So yeah, if you had a giant AI shock, that would be enough to raise the marginal product of new investment possibly to very high levels. And so you know, I play around with macro simulations like this. And in my macro simulations, if you 10x the rate of automation, that's enough to kind of get interest rates to start going up again. So not getting the sky right rocket to like 30% or more, but to keep them from collapsing back into, zero interest rate negative territory, you need like a big increase in automation that would increase demand for physical capital.
But that level of technological shock really only increases GDP growth rates by a certain amount. So I think all things together, if you had a big enough technological shock, that would be enough to increase growth rates. And then enough to increase interest rates because of the investment demand. And then the only question is like, how far can that permute back in time? If I told you that 10 years from now, there's going to be this amazing investment opportunity that's going to raise interest rates, to what extent can that raise interest rates today in order to answer the question we're going to have to consider different models of how people decide to save.
Seth, I'm a little unclear about your reasoning because you're really focusing on that return in the future. But the way I interpret it is that the financial return in the future doesn't matter. It's just the future is going to be glorious, right? Like the robots are going to handle everything for us, we're going to have all of our material needs fulfilled. In principle, it could be one company develops the AGI. It has a limited demand for capital. And the story doesn't really depend on that much relative to the fact that if I know that the future is going to be glorious, why would I save anything for it?
I think so. I think this is a really important point. I think the paper is laser focused on the issue that you bring up, which is the saving side behavior. What's the decision to save and invest? What I'm pointing out is that the interest rate is pinned down by supply and it's been down by demand, right? So at the same time there's investment demand and the interest rate also has to be the marginal profit capital. So yes, savings, with side of the scissors does the cutting, right? Yes. So yes, so that's what I'm trying to bring in. Okay, so maybe let's talk about these different saving rules that he considers.
He focuses first on representative agents. So a representative agent is kind of the simplified way economists like to think about savings and consumption in the economy when they're being lazy, which is they imagine one guy who owns all of the income in society and he lives forever and has a certain discount rate over the future. And in this case, if we're in the good world where the euchatastrophe is coming, AGI is going to be amazing in the future, then the good AGI being on the way causes growth expectations to increase. AGI is making all of the capital and all of the labor income. So if growth expectations increase, that means their income expectations increase.
And because they're planning on future income, they want to, move some of that consumption to today. So they're going to really dis save today in order to try to balance consumption between today and tomorrow. So that's the good case why AI in the future would cause a representative agent to dis save today, thereby raising interest rates, right? Because if you dis save, you move up the curve in terms of the projects that get funded and you move to higher and higher marginal product projects are not funded. What about the bad case? So suppose we think AGI is going to come around the corner and instead of leading to the euchatastrophe of 30% growth, it's going to lead to the bad catastrophe of the world being destroyed.
Well, in this scenario, bad AI is going to lower the future marginal propensity to consume to zero. The future, sorry, the future marginal value of consumption to zero because you can't eat things in the future because you're dead. The Terminator fucking killed you. And so what do you do? You party today, we've all been the science fiction cliche of, the world's ending tomorrow. So we have a big party tonight. And what is a party? A party is eating up all of your stock at acorns, right? So representative agent case pretty seems pretty straightforward that if growth is going to increase a lot, interest rates have to increase a lot.
Fair enough. Yeah. Oh, okay. Now you might say, but Professor Menzel, it's not the case that all of saving in society is produced by one guy who gets all of the income in society. Rather, savings are made by distributed people or a lot of different settings and a lot of different domains. And in particular, I would emphasize the overlapping generations aspects, right? People don't live forever. They're born and have a life cycle over the course of which, they might borrow to go to college, then they come out of college, and then they save for a house, and then they buy a house, and then they pay that down.
And now they're saving for retirement, and then they retire, and then they spend down their retirement savings. And then if they are the perfect, most God blessed economist, and on the day that they die, they've spent the last penny of their asset. And all and then that I believe that's how you become the saint in economics. So in this case, I would have to say I was a little bit confused by the analysis in the paper. It's very abbreviated, and it's not clear who is making the income, and when in the overlapping generations model that gets sketched. What Basel says is that basically it's just the outcome is the representative agent model plus this additional parameter.
That's a function of some complicated OLG things. Let me tell you instead the way that I think about the OLG scenario. In the overlapping generation scenario, you'd imagine someone who lives for two periods. And then just to keep things simple, we're going to assume logarithmic preferences. That's not super important to our audiences, but it makes the math easier. Let's say this person makes labor income when they're young, and then they're going to consume their capital income when they're old. And then let's assume the saving rate is a constant. What that's going to lead to is the young people, if they're only making the labor income, that tomorrow there's going to be some growth, and then that might raise tomorrow's labor incomes.
Doesn't matter to me. I'm just trying to, I made some labor income when I was young in my 20s, and I want to eat some of that while I'm young, and I want to eat some of that when I'm old. In this setting, a shop to AI is going to raise interest rates immediately if it's automating, raising the marginal product of capital. But to the extent that AI also raises labor share of income. Now, this is kind of like an unrealistic model I'm getting into now, but you can imagine a universe where AI raises labor share of income. And if what households are doing is that they're saving a constant fraction of their labor income, that would increase the supply of savings in the economy.
And so you can get a world where there's actually higher capital accumulation, lower interest rates as a result of this AI shock. As long as the AI shock redistributes in the right way between people who want to consume their capital and people who want to save their capital. Right. So I think what this aspect of the discussion misses, and I think he talks about labor share as being one of the important parameters, is I would have taken the argument one step further, which is to say, AI probably automates, which probably lowers labor share, which probably reduces supply of savings in an overlapping generation's model. So that would be a force that would tend to push again in the direction of anticipated increasing interest rates, or at least our current AI interest increasing interest rates.
In terms of a bad AI shock in the OLG world, I think if thinking about the catastrophe will terminate or the AI's kill all of us, in overlapping generations world, I'm a little bit concerned that if it's too far away, that's not going to show up in interest rates at all. I think if that is in the far enough, if it's more than a generation away, that should have no impact on interest rates in an overlapping generations model. Assuming that there's no agents that live super long. But if there was one in the near future, you would get the basil outcome. So if I'm a young person today, and I think the world is going to get destroyed tomorrow, maybe I don't save anything today.
Maybe I eat everything today. So I would say the heart of the story gives you some flavor to what is the timeline over which the AI has to be imminent to have these effects, right? Because it's got to be in the lifetime of the people who are making the savings decisions for it to bite, right? Whereas there is no quote unquote lifetime of the representation. Yeah, well, it's very in the weed set. I think we can talk about these different sorts of models. I think these toy models do kind of show that if you believe something is happening tomorrow, there will be a really strong response.
But I think where things get funkier is, well, if we know that AGI is coming tomorrow, something must have already happened today to make us believe that, right? Right, we're not the market is never going to be like, Oh, 100% we're getting AGI 10 years from now. There's no uncertainty, and there's no evidence that we have AGI today, right? There's a sense that there's a transition path that is likely to be reflected in a bunch of economic variables as we approach this moment of, AGI, unless you truly think of like, AGI occurs one day and the next day we're in singularity.
Certainly there are people who believe that. Yeah, well, let's come back to that at the end. Most of us, yeah, most of us believe that even if it's things are fast moving, right, they're going to, we're thinking that fast, years rather than decades, that's fast. And so we're not talking about days. And as a result, we don't expect the market to think that there's 100% chance of something 10 years from now for which there isn't already economic evidence for it. To me, it's just kind of like, it's weird to expect it. Now, if we think like, does the market think that there is a 5% chance of AGI 10 years from now, I would have loved to see a number that says, here's how much the real interest rate should be higher based on that probability.
That is not something this paper gives us. There's like a hint at a calibration. There's like a let me find the hint of the calibration. There is right where he discussed the overlapping generations agent benchmark calibration under transformative AI. So he's imagining basically a representative agent model in the steady state. Suppose that there was an agent who had, log preferences, again, that's in the weeds, but that standard has a discount rate of 1% for the future. And what he says, if we anticipated AI increasing the growth rate by 30%, I assume this is in the next year that would raise real interest rates to 31%.
Unclear what the time horizon has to be for us to anticipate that 30% increase, whether it's got to be right now or 10 years from now. Yeah, I think that's maybe the biggest thing lacking in this paper is a close consideration of the time horizons at play. Although maybe that's a good transition into the little bit of the empirical evidence. Andrey, do you want to briefly tell the audience about the cross country analysis he does? Yeah, so this is kind of an interesting thing, right? So far, we've been talking about the real interest rate. And if you have just one real interest rate, then we are in a difficult situation and trying to evaluate empirical claims at just one time series.
But actually, different countries have different real interest rates. And so, to a budding macroeconomist, this seems like a natural place to run their favorite type of regression, their cross-country or should, which is not the most stellar of empirical strategies, but you know what? It's the best they have. It's a start. It's sick and gander. Essentially, the theory needs two numbers, right? We need expectations of the GDP growth and the real interest rate at this time period. Both of those numbers are kind of tricky to obtain, and we think about it. The real interest rate, in order to calculate that, we need to kind of subtract out the expected inflation, because we don't know what inflation will be in the future.
And so, there are kind of two ways of doing that. One is, there are these special types of bonds, treasury, inflation-protected bonds, and other variants of those that are actually real, that have real interest rates embedded in them. So, you can kind of use that. Or you could ask people, kind of, what do you think inflation is going to be? And then similarly, GDP growth rates, you can use some forecasts. You can actually use the realized GDP growth. Obviously, there might be some differences in those two numbers, because of the stellar reputation of macroeconomists and forecasting GDP. There are some deviations in that.
But you try to see whether the real interest rate is explained by the GDP growth forecast when we consider across different countries and over time. And in fact, there is a positive correlation, I think. So, I think that's kind of the evidence. That's what I'd say about that. Yeah, so not super surprising, but it is consistent with the story, right? So, again, the story is, big explosion in anticipated growth should lead to big explosion in interest rates today. And the cross-country evidence seems consistent with the idea that increases in growth rates lead to increases in GDP. R squared isn't particularly good, but the relationship is there both in the cross-section and when controlling for, like country-fixed effects.
He kind of does a difference version. Okay. So, that's the evidence. Andrey, are you ready to move into our scathing criticisms? Go for it. All right. I want to skate because the authors are beloved, and I think there's a lot of good thinking in here. But one paper I do not see cited here, which I would have loved to see a site for is my paper with Eric Green-Lolfson, digital abundance and scarce architects, which considers exactly the scenario that we've been dancing around. So, what if we get this country of geniuses on the cloud, but it doesn't create 30% growth rates, right? Why might that happen?
Well, it might be that there's a scarce bottleneck in society that prevents us from really taking huge advantage of this country of geniuses on the cloud. And now that here's the key point, and then that scarce bottleneck has to not be capital that you can invest in on the margin, right? So, why don't we say that the scarce bottleneck is like literally genius entrepreneurs, right? So, we've got this country of geniuses on the cloud, but we need a special entrepreneur character to come and find the opportunities to deploy this new tool. In such a scenario, those scarce architects are going to get giant shearers of output.
They might get, 50%, 90% share of output, even if output's increasing a lot. But if this AI technology we're developing is a substitute for the kinds of capital that can be invested in, you would get a world where GEP goes up by a lot. It's not going to go up by maybe you can imagine a world even where, it goes up by 30 percentage points. But the bottleneck and the income is not going to capital, and therefore, the marginal product of capital investment is low, and therefore interest rates are low. So, I there's certainly no, there's no economic contradiction in the story I just told.
You might say it's unlikely that in a world where we've got a country of geniuses on a cloud, no one can, ordinary people can't think of productive capital investments. But it's, there's nothing internally inconsistent about that scenario. And I think something can be very smart and still not solve, the substantial share of society's problems that might be driven by kind of just the very nature of human beings, and what they want in commensurate with what other human beings want, for example, like having very nice beachfront property. I'm really even farther than that, right? So like, imagine that output is the C, yes, again, a little in the weeds here.
Imagine that GDP is a function of how good your AI is, plus your capital stock is one input, and then your second input is Elon Musk's, right? If we keep on pouring in more and more AI, make your AI arbitrarily abundant, you can make output arbitrarily abundant. So you could have GDP growth of greater than 30%, you could be solving all of society's problems. But again, the bottleneck is the Elon Musk. Elon Musk makes 100% of the income. And depending on assumptions on his saving rate, you might get very low interest rates in such a scenario, even though we are getting big bad GDP growth.
I guess to me, kind of a greater worry about the AGI world is the AGI's come up with better ways of running society. And then they just decided to wipe everyone to zero, for example, as right. And so they steal our money. I just have a hard time believing that our financial system remains constant through the AGI transition, right? And that really puts us in a bind about how to make money off of this, for example. Well, it's not about making money after us, but yes, it's not really about making money. It's about interchangeable consumption. So here's another thing that's not in Basil's model, which is the, we also hinted at this, which is the Yetkowski devastating for a strike.
If the AI is super nice and sweet to us, right up until the moment decides to kill all humans, we wouldn't expect that to show up in interest rates, right? But I mean, I guess you would say, well, the market doesn't anticipate it. I'm not making any claims that are not about what the market anticipates. Okay, so fair enough. So the market doesn't anticipate a Yetkowski death stating for a strike. But I don't know, maybe the market's not always right. I guess what I like, the way I think about this is there's some truthiness to this argument, right? If the market is certain enough, we should certainly observe changes in the interest rate.
But clearly, we're not there right now. And so maybe the question is, if we believe that there will come a time, and we need to have an opinion on the time, but there will come a time where it'll be apparent that EGI is coming, or that 30% GDP growth is coming, so on and so forth. How do we take advantage of that today? Because I think that's why people are so interested in this paper, really, right? People are really interested in this paper because there is this question that is very commonly discussed, certainly in San Francisco is like, how do you profit those, your insider information from hanging out with all these AI people that AI is a really big deal, you know?
Well, probably buy it. What's at your yum in 2015, right? But how much? Oh, sorry. But I think one implication of this is, at some point, even before AGI happens, if it's pretty clear that's going to happen, the markets will react. So if you can make a bet on future real interest rates at a sufficient time horizon, you stand to make a lot of profit, and you'll be really wealthy relative to everyone else for that short period of time until AGI comes and everyone will be equally wealthy. Yeah, I mean, it's certainly not a Basil's story, but I understand why that's why San Francisco's or San Franciscoites are trying to deal.
So I think we're moving into sort of wrap up austere remote. I just wanted to bring up one more sort of like conceptual critique of what Basil's doing here, which is related to this idea of a money tree. Have you ever seen Andrey, one of these funerary money trees that are made by certain Chinese groups? No, I haven't actually. All right, cool. So I saw a couple examples of this. One example that's beautiful is in the San Francisco Museum of Asian Art, where kind of like in a tomb, so you know, revered ancestor, we're burying you. And with you, we're going to bury a kind of a metal tree replica that has kind of money hanging off of the limbs in the thought that, okay, now in the afterlife, you're going to have this money tree, you're going to be stuck, you're going to have all the money you need.
Or when you think about it, why would if I'm a guy who's about to die, why would I want to move money from my limited physical existence into heaven? Heaven, there's the things don't cost money in heaven, Andrey. Why do I need money? Heaven, right? Well, I mean, heaven is not a universally acknowledged outcome, right? These are presumably not Christians. But to make the analogy to the setting in the paper, Basil asks the question, if heaven is on the way, why would you save money to bring to heaven? Well, one answer might be, well, in heaven's got really good stuff, right? So all of the arguments of this paper are about equalizing the marginal product of consumption.
So the idea here is, well, in the future, we'll have lots of money, so we'll be able to consume a lot. So the marginal product of consumption will be low, will be satisfied, and therefore we should do more consumption today. But maybe we'll invent something on tray so amazing, so beautiful and delicious that in fact, you want to move all of your money into the future for just a slight chance of getting epsilon more of these future wonder products. Maybe one example is you might imagine an immortality drug, which is invented shortly after AGI, which is still expensive in the medium term. If I told you AGI was going to invent an immortality drug, but it's going to be, a million dollars, you'd probably save into the AGI future, right?
Yeah, that's totally right. So imagine it was a ability to upload your brain into the cloud. There's still vast physical costs involved in running billion dollars. And transferring your brain there into running your brain as a in silico process. And maybe eventually they'll figure out how to get everyone uploaded, but they can't really resurrect people who are already dead. And so as long as the other probability of dying, every year, you really care about getting that technology ASAP. And there's actually an interesting argument made by some people that actually, if you think AGI is coming very, very soon, then you should be really, really risk-averse in terms of your physical activities, right?
You're about to get over just get the best life ever. And all you have to do is just not go skydiving or motorcycle. Oh, I love this. Okay, so Basil's follow-up paper is gonna be the rate of skydiving indicates AGI is not sinned. Yes. Yeah, but I mean, I think that's maybe a great place to start transitioning into posterior because it's not just dissaving that you should do if you think AGI is around the corner. There's a lot of things you should do differently. And maybe there should be a paper that considers all of these different behavioral adaptations. Yeah, I mean, that would be really fascinating.
I guess what I'm curious about set is, so, there's this kind of narrow claim about these interest rate arguments and the paper. But I guess, given today's interest rate, how much does that inform your belief about the probability of AGI at a 10 or 20 year horizon? Oh, so now you want to ask me what? Okay, so it seems like to answer that question, I need to take both on the actual percentage chance of AGI and to take on how good the market is at incorporating information about AGI. I would say that having read this paper, my prior that my belief that low interest rates indicate that markets do not believe AGI is imminent, I think that goes from 90% it goes increases to like maybe something like 95%.
And the reason why is because I was approaching this paper from the perspective of the supply side. It seems really clear to me and really obvious to me that technology should increase the demand for investment and a good enough technology should increase the demand for investment by a lot. To the extent that you think that those opportunities are coming in the near future, you should be keeping your powder dry for going investments and stuff that's going to keep your money locked up for a long term in the short term. And I think if you looked at the required rate of return on new capital projects that are going to take 30, 40 years to pay off a dam somewhere or a bridge somewhere, the fact that those interest rates are not at a super high level, those are interest rates that are taking into account.
Supposedly what the cap money could have been doing, alternatively 34 years from now, I think that tells you that the market doesn't think that there's going to be AGI. Now, how do I get from whether the market thinks there's going to be AGI to whether there really is going to be AGI? I think it tells you more than zero. I think it tells you, let me put it this way. I think that the case that we get really dramatically elevated interest rates before AGI as a premonition of AGI, let's say five years before AGI, again, by this 30% increase or instruction in the world criteria, I think 80% chance, 90% chance that you would see at least five years before the AGI, you'd see that showing up in interest rates.
Yeah, so I guess before I read this paper, and I just wasn't really thinking about these issues very much, other than just my general skepticism of macroeconomic arguments and claims about interest rates, I think I almost certainly agree with you that if we're right on the cusp of AGI, we're going to start seeing things with interest rates. I guess thinking about this problem more, it's not clear to me given the uncertainties involved, we would see large movements in the interest rate today, even if there was a 5% chance of AGI priced in. That's kind of where I was hoping that there would be a calibration done to tell me how much that would be.
But I still, I never like to think of the market as one person. I don't think of Mr. Market, I think. You think of Mr. 401K owner who has never even heard about AGI? Yeah, so I'd be kind of participating. Like what share of market participants think AGI is coming at a given time scale, and is that a substantial percentage? I still don't know because they just clearly they exist, but I don't know what share they are, what share of capital they have. I think still my punch is that we'll see evidence of incoming AGI before with other macroeconomic indicators. I find it hard to believe, and I guess I'm a bit of a gradualist in this way, but I find it hard to believe that we're going to have 30% GDP growth if we didn't have already an elevated GDP growth a year before.
I think that's exactly right. It's not like it's going to go from 2% to 30%. There'll be a ramping up. Yeah, and so then I guess if we're going to, let's say we have this conversation 15 years from now, and AGI is considered imminent by a large share of the market, I assume that that's because we've had technological breakthroughs, massive GDP growth, and yes, then we get weird, sort of elevated interest rates that's probably going to happen, right? But at that point, the game And that's why you're not hardly calling it, right? At that point, you're not calling the game anymore. I think that's kind of why the more interesting question, I think for a lot of people, a very practical question is what do we do now if we think that, let's say, there's a 5% chance of AGI, very, very powerful AGI, 15 years from now.
And one strategy could be, well, we know if that's going to have 15 years from now, 10 years from now, we're going to have very high interest rates, and so we should bet on that. It's not obvious to me that that's the strategy that one should take. Did you see what Sam's recent post was about what you should invest in when AGI is coming? He says land and luxuries. And hey, if people start binning up land, that's going to increase the real rate of return to land. And by the no arbitrage condition, if you start increasing the rate of return to one asset, you increase the rate of return to all assets.
Yeah. So I think Sam Altman is implicitly thinking that interest rates are going up. But I guess the value of land is valuable in a lot of intermediate scenarios. The value of land is a good investment as long as we have robust GDP growth. Period. I don't need AGI for that. So robust global GDP growth might, I think that's a pretty reasonable prediction and the median prediction, right? So of course, we can have pandemics, wars and stuff is obviously risk and uncertainty, but like it would be the history suggests that global GDP growth will continue happening and you scarce things will always be scarce.
Scarce land will always be scarce and therefore it's a pretty good investment, right? I just don't think you don't need. Yeah, I for it. But it ramps up everything. It'll be even more scarce in the world. Yeah, even more in more abundant, right? Yeah. All right. So I think Basil, you heard our advice for you what your next paper should be your next paper should actually tell us economists where we should be placing our bets, where we should be making our investments, as well as looking into the rate of skydiving as a potential additional forecasting tool for AGI. Do you have any parting thoughts, Andrey, other than that are beloved and intelligent guests should listeners should like, rate, and subscribe to our podcast.
Stay safe out there, guys. Stay strapped. Well, thanks for tuning in to Justified Posteriors, the podcast that updates ours and your careers about the economics of AI and technology. Beautiful. See you next week, guys.