Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
What Markets Tell Us About AI Timelines (with Basil Halperin)
5 Plays1 second ago
Basil Halperin is an assistant professor of economics at the University of Virginia. He joins the podcast to discuss what economic indicators reveal about AI timelines. We explore why interest rates might rise if markets expect transformative AI, the gap between strong AI benchmarks and limited economic effects, and bottlenecks to AI-driven growth. We also cover market efficiency, automated AI research, and how financial markets may signal progress.
- Basil's essay on "Transformative AI, existential risk, and real interest rates": https://basilhalperin.com/papers/agi_emh.pdf
- Read more about Basil's work here: https://basilhalperin.com/
CHAPTERS:
(00:00) Episode Preview
(00:49) Introduction and Background
(05:19) Efficient Market Hypothesis Explained
(10:34) Markets and Low Probability Events
(16:09) Information Diffusion on Wall Street
(24:34) Stock Prices vs Interest Rates
(28:47) New Goods Counter-Argument
(40:41) Why Focus on Interest Rates
(45:00) AI Secrecy and Market Efficiency
(50:52) Short Timeline Disagreements
(55:13) Wealth Concentration Effects
(01:01:55) Alternative Economic Indicators
(01:12:47) Benchmarks vs Economic Impact
(01:25:17) Open Research Questions
SOCIAL LINKS:
Website: https://future-of-life-institute-podcast.aipodcast.ing
Twitter (FLI): https://x.com/FLI_org
Twitter (Gus): https://x.com/gusdocker
LinkedIn: https://www.linkedin.com/company/future-of-life-institute/
YouTube: https://www.youtube.com/channel/UC-rCCy3FQ-GItDimSR9lhzw/
Apple Podcasts: https://geo.itunes.apple.com/us/podcast/id1170991978
Spotify: https://open.spotify.com/show/2Op1WO3gwVwCrYHg4eoGyP
PRODUCED BY:
Recommended
Transcript
00:00:00
Speaker
So within macro, I think the big question is, will AI lead to a speed up in economic growth or will it get bottlenecked by certain sectors or areas?
00:00:11
Speaker
The effect of a aligned AI and unaligned AI goes in the same direction on interest rates, so unlike equities or other asset prices. It's hard to get away from the idea that there will be skyrocketing inequality in a truly transformative AI scenario.
00:00:26
Speaker
But skyrocketing inequality might still be consistent with everyone being better off. Coordination with other people is not something that AIs as they exist today in helpful, harmless chatbots can can really help with.
00:00:39
Speaker
It's very plausible that benchmarks are these narrowly defined tasks that don't really capture the breadth of what a worker does every day. Welcome to the Future of Life Institute podcast. My name is Gus Docker and I'm here with Basil Halperin.
00:00:55
Speaker
Basil, welcome to the podcast. Thanks, Gus, for inviting me on. Excited to be here. All right. Could you give a little background on yourself to start with? Yeah, so I just joined the University of Virginia as an assistant professor in the economics department after finishing a postdoc at Stanford. My background is that I did a PhD at MIT.
00:01:14
Speaker
In past lives, I worked as a data scientist at Uber and maybe relevant to today's conversation, my first job out of college was at a quant hedge fund ah researching and trading inflation-linked bonds.
00:01:25
Speaker
Interesting. Perfect. Right. So the so the the theme for today's the conversation is actually the intersection of economics and ai and specifically what we can learn about ai risk and AI timelines from economic indicators, we could say.
00:01:41
Speaker
You have this fantastic essay on AI timelines and the efficient market hypothesis and interest rates, what we can learn from interest rates and when we're trying to predict when we might get advanced AI.
00:01:55
Speaker
So I'll link that paper in the show notes, but could you sketch out the basic idea here or the basic conundrum? Yeah. So this this paper, which was joint with or is joint with Trevor Chow and ah Zach Maslish,
00:02:09
Speaker
ah the The argument in one sentence is that if markets were expecting transformative to be coming in the next, say, 30 years, either aligned ai going to rapidly accelerate economic growth or unaligned AI that was going to lead to existential risk of the kinds that I know your listeners are very familiar with, either of those possibilities would result in high long-term real interest rates.
00:02:32
Speaker
And looking at markets, looking at real interest rates today, we don't see particularly high uh real interest rates so i break that down maybe starting with what what are real interest rates yeah that that would be useful i think yeah so you know when when you open up the wall street journal the financial times interest rates are not usually if things are going well the sort of thing on the front page right you see stock prices uh interest rates though are this very important price in the economy they affect a lot of other prices
00:03:05
Speaker
So if you do see interest rates in the newspaper, you'll see a nominal interest rate. So when the US government borrows money, it typically issues these nominal loans that pay back in ah dollar terms.
00:03:18
Speaker
So if it issues a 10 year bond at, I think the current rate, something like 1.6% last I checked, that means that in 10 years, it has to pay back the amount of the loan plus 1.6% interest on that loan in dollar terms.
00:03:34
Speaker
Real interest rates are different from those nominal interest rates in that they adjust for inflation that occurred over that period. So they're sort of the the real thing, so to speak. um So ae why look at real interest rates to think about transformative AI?
00:03:50
Speaker
Well, interest rates clear the market. the The way economists think about real interest rates is that they clear the market. a in the supply and demand for saving and borrowing.
00:04:02
Speaker
So if I want to borrow, then I need to take out a loan. The real price, the real cost to me of that loan is the real interest rate.
00:04:13
Speaker
ah Meanwhile, what is the lender get in return for lending to me? They get the real interest rate. So if I really want to borrow, if everyone in the economy really wants to borrow, that would push up interest rates in order for the markets to clear.
00:04:27
Speaker
How does that come back to AI? Well, if we all expect to be super rich next year, you know if I'm going to earn a lot more money next year than i am this year, there's much less reason for me to save today.
00:04:38
Speaker
That lower supply of savings, so to speak, speaking a bit loosely, aye ah pushes up interest rates. Similarly, if we all expect to be dead next year, no reason to save today, that would push up interest interest rates.
00:04:50
Speaker
That's the argument in abstract level can get into more concrete things like, you know, do we see people on AI Twitter talking about taking out long large long term loans, talking about not investing in the 401k? We can get into that. That's the high level abstract argument.
00:05:05
Speaker
Yeah. And this rests on the market for interest or the yeah the market for for loans and the the market that sets the price of of interest. the the market that sets interest rates being efficient in general.
00:05:19
Speaker
So maybe we could explain the efficient market hypothesis and what it is that we are... So if we are saying that the market is wrong, what does that actually mean? Yeah. So the efficient market hypothesis, which is sort of what we're leaning on or or hinting at in our argument, is this idea that Financial markets reflect all available information.
00:05:39
Speaker
So a the price of general motor stock reflects all available information in the world, so to speak, about future profits. aye That's because stocks reflect expectation about future profits because those are paid out as dividends to shareholders of equities.
00:05:57
Speaker
Similarly, with interest rates, I just explained how interest rates reflect the supply and demand for savings. Therefore, we would hope that if markets are financially informationally efficient, they will reflect ah correctly the market participant beliefs about future consumption savings decisions, about future economic growth.
00:06:19
Speaker
ah The idea for markets being efficient is sort of essentially supply and demand, essentially just no arbitrage that if you knew with certainty that General Motors was going to have really high profits next year and that was not reflected in stock prices, you'd immediately want to go out, buy a bunch of GM stock, hold on to that, earn the dividends next year.
00:06:44
Speaker
That's harder if you're unsure. That's harder if you're You sort of don't have the capital to invest in GM um enough to move prices up to the correct level.
00:06:55
Speaker
But that's the sort of basic idea underlying this. ah Markets are good information aggregators, particularly forward-looking financial markets. Yeah, the basic idea here is something like if you think you know, if you think you have a piece of information that the market is not incorporated, you then have an incentive to use that information to try to earn money. And because that because that incentive is is is quite powerful and because there are so many people looking to um to to earn money and to to price assets correctly, the the the assets tend to ah reflect ah all of the information. Also because people now have an incentive to seek out new information.
00:07:34
Speaker
um So, yeah, I think just a quick point here. Why is it that when my uncle says, hears about ChatGPT and invests in Apple or Nvidia or some kind of tech stock and he ah he beats the market, why isn't that a refutation of the efficient market hypothesis?
00:07:55
Speaker
Yeah. ah So I think even if you're someone who thinks I have information about one asset to buy to beat the market, you're you're still a believer in sort of market efficiency for 99.9% of assets, 99.9% of the time. ah so So even if you have some insight,
00:08:14
Speaker
ah it's a good benchmark to trust markets to get things approximately right, sort of as this sort of outside view perspective.
00:08:26
Speaker
There are a number of reasons why someone might be able to beat the market. If you just look at historical data, number one, there might be selection bias. No one talks about all the money they lost on bad cryptocurrencies in late They only talk about all the money they made on Bitcoin.
00:08:41
Speaker
Another reason is that
00:08:44
Speaker
You can earn excess returns in financial markets by taking excess risk. So for example, investing in bonds has historically a very low return, whereas stocks have a higher average return.
00:08:57
Speaker
Does that mean that there's a market failure here or sort of market inefficiency that you can always just invest in stocks and get like a 7% annual real return historically, something like that versus like a 2% roughly historical annual real return on bonds?
00:09:11
Speaker
No, that's not a ah missing arbitrage there. It's that stocks are riskier. If you invest in stocks, sometimes stocks go down a lot and you would like to avoid those big drawdowns ex ante beforehand investing.
00:09:26
Speaker
Investors need compensation in order to bear that risk. Maybe, for example, you're most likely to be unemployed right in the middle of a recession. And those are also periods when stock markets are down.
00:09:39
Speaker
this correlation between aye when you want money the most, when your marginal utility of consumption is highest in econ speak, that correlation with stock market returns means that stocks are risky and and is is one way that people can beat the market.
00:09:58
Speaker
there There also are generally people who have information that hasn't been incorporated into financial markets yet, and they are ah getting compensated for bringing that information mark to the market.
00:10:11
Speaker
So sometimes I think it might be better to talk about market efficiency in terms of like, is the market for information perfectly competitive? Are people getting compensated for going out and doing the costly work of acquiring information, processing information, and bringing that information and incorporating into prices?
00:10:30
Speaker
Yeah, yeah. and I guess one one question here is to ask whether markets are actually good at at pricing in, so to speak, the possibility of either extreme growth from AI or existential risk from AI.
00:10:46
Speaker
These could be very low probability events. These could these events could be far out. And what do we know about how markets price in such information or such possibilities?
00:10:59
Speaker
Yeah. So I think there are two lenses i would think about this from. So one is that a lot of important asset prices are... incorporating expectations or require incorporating expectations about things very far in the future.
00:11:13
Speaker
So the average duration of the stock market, i don't I don't have the number off the top my head, unfortunately, but it's certainly greater than 10 years, maybe maybe even more than 20. That is to say that like the average cashflow, roughly speaking, of stocks is like at least 10 or 20 years out in the future.
00:11:30
Speaker
So markets market participants have to be doing far more, uh far far future or you know maybe not far future by the standards of the f fli podcast but far future by the standards of like contemporary media discourse uh uh forecasting so that that's one thing a second thing though to say is that yeah there is a lot of evidence that markets are worse when it comes things further than in the future and that's because this no arbitrage that i emphasized as important for
00:12:03
Speaker
financial market efficiency is harder for things that take a long time to pay off. So if you've paid attention to prediction markets around elections, this is something that you'll have seen where four years before US presidential election, there'll be a lot of crazy odds on people.
00:12:20
Speaker
i who are never going to win the election. yeah And the reason those odds can persist is because you would have to hold on to a short position against some crazy person who's never going to win the presidency for four years.
00:12:37
Speaker
And there's a high opportunity cost of holding on to that trade because you could be doing other things with your money over that time. Mm-hmm. or because there's random fluctuations in the market over that time. And ah at some point you could get blown out.
00:12:51
Speaker
um In general, limits to arbitrage, this is the technical term for ah things that can prevent arbitrage errors from correcting market mispricings. Limits to arbitrage.
00:13:04
Speaker
There's a good amount of theoretical and empirical evidence that this is more severe with ah arbitrages that take a longer time to pay off. Mm-hmm. Mm-hmm. I mean, So you mentioned your your background in finance, right? I know some people who have profited tremendously from from the COVID pandemic, from Nvidia, from predicting the Trump tariffs and so on. And and you probably know many more of these people than I do.
00:13:33
Speaker
So it seems that there are pockets of ah people with special knowledge, people who are extra super smart, say, ah that know something that that others don't? Could it be the case that there are insiders and that know something about AI progress that others don't or that the broader market is not incorporating?
00:13:54
Speaker
It absolutely could be the case. i I am still wary of extrapolating too much from anecdotes because, again, no one hears about the anecdotes where my first investment, funnily enough, when I was like 14 for my bar mitzvah, my dad gave me a few hundred dollars play money to ah invest in the stock market so I could learn how to invest.
00:14:16
Speaker
ay I being 14 years old had no idea ah what to invest in. And so I just go to him I'm like, what should I do with this money? He's like, well, I read this article in the newspaper that said TSMC a good investment. This was 2008 or 2007.
00:14:29
Speaker
oh two thousand seven And so I put the few hundred dollars in TSMC and it like went up a little bit at first. And then the 2008 financial crisis happens, stock market plunges.
00:14:41
Speaker
ae i hold on to it for a few months expecting a recovery. He keeps telling me to hold on. I'm like, no, I got to sell out. So I looked at this a few years ago. I sold out the week that the market bottomed.
00:14:53
Speaker
ah So that's that's that's the average investor for you, me at age 15 or whatever. um And of course, if I'd held on to TSMC today, would probably be doing a lot better. um ae that so So I think that really is an important point that anecdotes are hard to extrapolate from.
00:15:08
Speaker
yeah That said, a second point is that, again, i think the right way to think about What we should expect in markets, in financial markets, is that you should be compensated for doing real work.
00:15:20
Speaker
So if you spend 24 hours a day ah reading ai Twitter, spending time on LessWrong, etc., etc., reading papers on Archive, maybe most plausibly, really doing the hard work,
00:15:32
Speaker
of trying to understand what's going on in the AI industry, then you do deserve compensation for that. And we should expect that you will achieve some alpha reflecting the opportunity cost of your time.
00:15:43
Speaker
That's ah one important point. The third point just is There can be instances where like someone has to go and collect the information to trade and have that reflect in the financial markets.
00:15:55
Speaker
So there can be instances where you happen to be someone who collects some alpha, ah some some excess returns ah because you had the information first.
00:16:05
Speaker
That is possible. Yeah, yeah. Do you have a good sense of whether ideas about either explosive growth or existential risk from AI has spread into kind of mainstream Wall Street institutions?
00:16:18
Speaker
Is it the case that ah the the highly informed people there with all of the compute and all of the data and so on, have they heard the arguments and rejected them or have they perhaps not heard the arguments?
00:16:31
Speaker
So the filtering process, the the the process of information diffusion is happening, happening pretty quickly, still ongoing, is my read. I haven't seen like a definitive survey.
00:16:42
Speaker
So some things I can say are like, we we wrote the initial version of this essay published in January 2023. three So two months or six weeks after ChatGPT was released.
00:16:53
Speaker
Since that time, NVIDIA has gone up like a thousand percent. Microsoft has doubled or tripled or something. ah Real interest rates kind of interestingly the third year horizon, at least, have gone up a percentage point. so ah and And maybe I should say,
00:17:08
Speaker
when When I first started thinking about this issue and debating with very bullish friends in San Francisco whether transformative AI was coming soon, at that point, real interest rates, this is like the depths of COVID 2021, 2020, real interest rates in the US s and the UK and the developed world were we're negative.
00:17:27
Speaker
So if if the US government issued a 30 year bond, bond ah for $100 after 30 years, it would only have to pay off like 98 or $99. It's like a negative 1% interest rate, negative 1.5%.
00:17:42
Speaker
Over the succeeding four years, real interest rates rose by ah two to three percentage points, which is kind of ah large move. That's probably not because of AI.
00:17:54
Speaker
It's probably because central banks around the world have been raising interest rates to fight inflation. That's sort of a separate issue. We can get to short run macro inflation if we want. I love that stuff. Maybe less relevant for your listeners.
00:18:05
Speaker
But definitely kind of interesting that interest rates rose over this time and have continued to rise further since we published this post. So that that's some context. Yeah. And additional context to say, or a reason to bring that up is that I think this market space perspective was particularly useful a few years ago prior to ChatGPT because there was such a smaller fraction of the world thinking about these issues, such a smaller ah amount of information processing by by humans ah trying to argue
00:18:40
Speaker
ah how long until Transformative AI develops. And so having this look at interest rates was was particularly useful then. Still useful today, but even more so then. yeah Anyway, coming back to your question about is this information diffusing through markets or pi ah through the financial industry?
00:18:56
Speaker
Yeah. so ae Certainly two or three years ago, ah there were very few people thinking about it. Today, there are more. So famously, Leopold Ashenbrenner, who was actually sort of a conversation with him inspired the argument with him, debate with him inspired this whole essay, has since launched a hedge fund, situational awareness, had an essay last summer that made a big splash.
00:19:23
Speaker
And just in the Wall Street Journal this week has reported that he his his fund ah with Carl Schulman made a ton of money trading on these ideas. hi Leopold, in his interview with Dworkesh Patel, ah directly cited my work with Zach and Trevor on interest rates saying, yes, he expects interest rates to rise eventually as bond market traders wake up and he has like $1.5 billion dollars in his fund is reported.
00:19:50
Speaker
So $1.5 billion ah and in situational awareness alone is maybe a niche part of the market. Maybe that'll grow. i guess 47% return that was reported is is already growing his fund.
00:20:02
Speaker
But then the industry more broadly, there's certainly ah like Goldman Sachs or other investment bank reports, bare minimum thinking about how quickly is the data center industry growing?
00:20:12
Speaker
ae thinking about what could the ai the impact of AI be on long-run growth. The latest numbers I've seen are that if you look across investment banks, consultancies on average, 10-year growth forecasts are more or less unchanged.
00:20:27
Speaker
But there are individual forecasters in the financial industry who are much more bullish and much more aligned with the man on the street, the woman on the street in San Francisco, so to speak.
00:20:39
Speaker
I mean, how would we even know? You you you discussed an increase in interest rates since 2020, but this is this has probably very little to do with AI. if If we saw interest rates increase, how would we know why they were increasing?
00:20:55
Speaker
Yeah. So I don't think there's a great definitive way, but there are some sort of consistency checks we can look at. So, I mean, one thing you can do is sort of build a full model of interest rates, i trusting that you're able to forecast the path of interest rates well, you understand the determinants of interest rates well.
00:21:16
Speaker
I'm skeptical that macro financial models have that much predictive power in general. Hence why i want to look at markets and look for giant changes interest rates, which is what this transformative AI perspective would predict.
00:21:28
Speaker
um That said, you could still look at other prices, other things in the economy to try to understand i is the change interest rates caused by ai expectations.
00:21:40
Speaker
So for example, the effect of transformative AI on stock prices is plausibly harder to interpret for various reasons we can get into, it's It's very plausible that certain stocks would benefit strongly from expectations of transformative AI. ah For example, NVIDIA TSMC.
00:22:01
Speaker
aye And indeed, we have seen those go up a bunch over the last few years. So that's kind of interesting. I should say that's in the case of aligned transformative AI. Unaligned transformative AI would obviously wipe out not just Taiwan, the United States, and the entire world and send those stock prices to zero at some point.
00:22:19
Speaker
um so So that only helps us corroborate an increase in interest rates due to transformative AI. are Sorry, aligned transformative AI. Yeah. um You can also look at these surveys of financial market participants, of bank analysts, and see if their expectations are changing.
00:22:37
Speaker
That said, and actually this is a very important point that think a lot of people misunderstand, financial market prices do not reflect consensus views, consensus expectations necessarily.
00:22:48
Speaker
They're not meant to track the average level of expectations, like a forecasting aggregator like Metaculous more or less tracks average expectations of participants. Financial market prices and all prices reflect the ah financial market prices reflect the marginal unit of capital, so to speak. They reflect the views of the marginal trader.
00:23:09
Speaker
The marginal trader being the person who ae is just at the indifference point between buying or selling this asset So it's their beliefs that matter.
00:23:21
Speaker
So if there's someone who has very strong beliefs about AI, there will be the person who's disagreeing with others and is trading the asset. put put loosely.
00:23:32
Speaker
um And there are lots of good reasons to believe that that marginal trader is more informed than the average person. Because if you have particularly strong beliefs, then you must have or it's plausible that you have ah better reasons for those beliefs.
00:23:47
Speaker
The people who have spent 10,000 hours reading the bio anchors report or whatever, Maybe they are more willing to make bets in a certain direction. And maybe they also have access to more capital. I think, unfortunately, there are many amateur traders that have strong beliefs and not not a lot ah suits to kind of back up those beliefs. But maybe those people don't have a lot of capital and so they can't really move the market a lot.
00:24:12
Speaker
Totally, totally. um So now I've lost track of where we were on the question, but oh, yes, I was explaining that we could look at surveys of financial market participants to see if their beliefs have changed.
00:24:24
Speaker
But that said, surveys are not definitive because the average belief does not determine the market price, the marginal, but the belief of the marginal trader does. Yeah, yeah. Maybe you can talk a bit about why it's not straightforward to interpret stock prices or equity prices.
00:24:41
Speaker
It's not it's not just that we can we can look at ah yeah why is it that stock prices aren't a perfect indicator of of AI timelines. Yeah, so there are a couple reasons. And I'll say that stock prices are not necessarily uninformative. It's just you might need to make additional assumptions to interpret them.
00:25:00
Speaker
yeah So one thing already discussed is the unaligned versus aligned distinction, where aligned advanced AI would plausibly raise profits of companies a lot, push up stock prices, whereas a unaligned AI would push profits down by exterminating humanity.
00:25:18
Speaker
ah That's one issue. The second issue is that you can only invest in publicly traded companies. For example, OpenAI is not publicly traded. Of course, Microsoft has a 49% share, I believe.
00:25:31
Speaker
ah so So if you want to ah look at stock prices to interpret the effects of AI, maybe you'd show up in Microsoft, but other companies, maybe this is less the case.
00:25:42
Speaker
A related issue is that It's not obvious that advanced AI would indeed, in in the Align case, lead to higher profits. So OpenAI, at least historically, has had this $100 billion dollar profit cap promising that any profits above $100 billion would be rebated to humanity or something like that.
00:26:02
Speaker
and And so advanced AI might not even lead to a higher valuation of OpenAI. Again, historically, that seems to be in the process of being changed.
00:26:15
Speaker
Or there's been talk of this windfall clause that if i <unk> just like opening eyes, hundred billion dollar profit cap, if AI really leads to some massive windfall, then companies could commit to give that windfall to humanity or something.
00:26:30
Speaker
Or coming back to Leopold Ashton Prenner's work for perhaps ah there's been talk of nationalization of companies and then these companies wouldn't earn profits. Final reason why stocks are hard to interpret potentially, ah that's sort of the most economically interesting, is that higher growth rates for the economy, higher expected growth rates for the economy, that is, do not necessarily lead to higher stock prices.
00:26:57
Speaker
The reason for this is kind of subtle. So stock prices reflect the present discounted value of future dividends, the present discounted value of future profits. That's the way that that's the most successful framework for thinking about equity prices.
00:27:12
Speaker
And where you when you say discounted, what do you mean? Yeah. so If you have a company that exists today and tomorrow, it's going to exist, the the stock price is going to reflect the value of any dividends it pays out to you today and the value of any dividends it pays out to you tomorrow.
00:27:29
Speaker
But not just the sum of those two, You discount the value of the profits it pays out to you tomorrow by the interest rate that you could earn in the meantime by putting your money in the bank, earning some interest rate.
00:27:43
Speaker
yeah So exactly to your point, stocks reflecting the present discounted value of future profits. means that although Transformative AI could push up future profits, as the whole thesis of our blog posts and paper argue, it will also raise interest rates.
00:28:00
Speaker
And so it depends on will future profits go up by more than future interest rates go up. That in turn depends on this very important parameter in economics, the elasticity of intertemporal substitution, which reflects how people trade off consumption today versus consumption tomorrow.
00:28:15
Speaker
I can go into more of that. ah It depends on whether this elasticity is above or below one. And the literature is not settled on that. Famously, or sort of famously, macroeconomists think it's below one.
00:28:28
Speaker
Financial economists think it's above one. The estimates in our paper suggest below one. ah But ah it's it's a hard parameter to estimate. This is actually kind of maybe an interesting objection to your thesis. Yeah, the argument or the objection goes something like this. So if we have...
00:28:47
Speaker
advanced AI, we might expect to have products and services that are of much higher quality, say in five years than we than we have today. And so you would expect people to to save money, thereby driving interest rates lower um because, you say, you can you can buy an amazing virtual reality headset in five years, or you can buy medicine that can extend your lifespan in five years and so on.
00:29:12
Speaker
And so this might drive saving um and thereby lower interest rates. and is this Is this incorporated into your argument or how would you think about this?
00:29:23
Speaker
Yeah, so basically i think this is one of the two or three best arguments against the whole thesis. But I i still, in my best guess, think it's it's not powerful enough to outweigh all the other factors.
00:29:34
Speaker
Yeah. So do you mind mind restating perhaps the best version of of the argument that I just tried to give? Yeah, totally. and And I think this will give a good opportunity to get more technical on the the argument that we're making.
00:29:46
Speaker
yeah So the precise reason why ah Higher future growth, traditionally, we think of that as leading to higher interest rate today is that if we're going to be rich in the future, the marginal utility of consumption in the future is lower.
00:30:03
Speaker
Marginal utility of consumption, meaning that a dollar in the future is worth less to me than it is today because I have diminishing marginal utility, where diminishing marginal utility means that i going from earning an income of $100 $1,000 bigger gain from a basic necessities. go from $1,000,000 month $1,900, that doesn't that much because if i only have a hundred dollars a month going into a thousand dollars a month means that i can get more basic necessities
00:30:33
Speaker
like go from a million dollars a month to a million and nine hundred dollars a month that doesn't but doesn't do that much to me so If money is less valuable in the richer future because of doing diminishing margin utility, there's this, again, the argument that there's less reason to save in the future. I'd rather have that dollar today.
00:30:51
Speaker
So that this counter argument about new goods in the future, i think of Phil Trammell as having prominently argued for this. It's really good points, a really understudied point in general in economics.
00:31:01
Speaker
ah He has ongoing work Chad Jones, I think, to flesh out these thoughts. And I think it's a really interesting conceptual idea. ah The argument is exactly as you said, Gus, that like if we're going to have these amazing goods in the future that don't exist today, then potentially,
00:31:19
Speaker
that dollar in the future still is worth more than it is today, even if I'm going to be richer in the future, because there's not that much I can do with the dollar today. The example that Phil gives is that if you're member of Tinker's Khan's Golden Hoard, if you had ah an extra dollar, like what are you going to do buy another horse or something?
00:31:37
Speaker
ah Versus today, if you have an extra dollar, there's like all this cool stuff you can buy, so something like that. um So there's there's not this diminishing margin utility because of these new goods.
00:31:50
Speaker
And that seems very plausible. The reason why I don't think it overturns the argument is a couple of things or two things. One is that if you look historically, looking at economic growth, and this is what we do in the paper, you do just see this strong positive relationship between higher growth and higher interest rates.
00:32:09
Speaker
we We have some nice data that I think is sort of a contribution to normal macroeconomic literature away from AI on this relationship between r and G. real interest rates and growth, showing that historically across 60 different countries, a number of decades, ah higher growth and higher interest rates just really are pretty correlated.
00:32:29
Speaker
um So that's that's one thing that historically, the invention of new goods has not outweighed the traditional diminishing margin utility mechanism. The second is, it's plausible that AI will be different, that AI will lead to all these new goods that, again, all your listeners are very familiar with.
00:32:47
Speaker
life extension, et cetera, amazing things I'd love to have access to. That's all true. That does give a motivation to save for the future, depressing interest rates. At the same time, we'll still be super rich because we'll have transformative AI i leading to rapid economic growth.
00:33:03
Speaker
And so i I will be rich enough to hopefully afford life extension. So on. even Even if even if you're not saving, you mean you will still be rich enough to yeah to yeah basically afford all of the goods that are available in this potentially amazing future.
00:33:20
Speaker
Yeah, but that's that's what I think is really most plausible. um But all that said, I think this is like one of the two or three best arguments against the whole thesis, but overall not convinced.
00:33:30
Speaker
Yeah, does AI change anything here? So you mentioned that this this is not a phenomena we've seen in the past, but maybe AI is different in that um you would expect these these great products and services sooner just because the rate of innovation could be higher.
00:33:47
Speaker
So maybe if you if you don't have to wait 30 years, you have to wait three years in order to to to enjoy better goods and services, you are more inclined to actually save the money or it is more rational perhaps for you to save the money?
00:34:01
Speaker
So what I would say is that it would be great if there was more historical work looking at how much of growth came from new varieties of goods versus more of horses in the golden horde.
00:34:14
Speaker
Like someone should just do that decomposition. And to my knowledge, there's not any definitive work on that or really much work on that at all. no Then you could think about How could the future be different affected by the channels you describe? Like maybe AI in particular ah is is sort of biased towards new varieties rather than more of the same.
00:34:34
Speaker
And again, i think that's extremely plausible. Is it enough to overcome the fact that in this transformative AI world that we're considering, we're having 30% overall income growth?
00:34:46
Speaker
I don't know. 30% annual growth is a lot Yeah. How much is that actually? as you know Maybe you could put that and in perspective for our listeners pretty because you know maybe the difference between 3% and 30% doesn't sound incredible, but it it really is. So maybe you could you could say something about how extreme 30% yearly growth growth rate in the economy might be.
00:35:08
Speaker
Yeah, great question. So so to i expand on that, the the transformative AI aligned scenario we're considering is 30% growth. yeah The reason for that is that's roughly a 10x increase in GDP growth compared to what we see today, which is about 3%, as you say.
00:35:22
Speaker
That number comes from the existing literature, Tom Davidson's work. I think you've had Tom on the show. yeah i it We are looking historically prior to the Industrial Revolution, something like 0.3% GDP growth was what we saw.
00:35:37
Speaker
So there was an order of magnitude increase in GDP growth from before to after the Industrial Revolution, maybe similarly around the agricultural revolution. And so maybe, ah you know, in ai the AI world, we we love talking about OOMS, orders magnitude, maybe there'll be another order of magnitude increase in growth around transformative AI. Yeah.
00:35:55
Speaker
So 30% growth, that's a lot, a lot more than 3% that we see on average today. And even a lot more than like the really fast growth episodes that you might think of in history.
00:36:06
Speaker
So China had this astounding sustained growth episode from the reform and opening up period to around 2010. Things have slowed down a bit since then. So remarkably fast, but their remarkable growth rate was sustained three decades of 10% annual growth, 10% versus 30%, still large gap.
00:36:24
Speaker
so large gap
00:36:27
Speaker
put 30 in perspective uh i don't know you can use moore's law as one benchmark perhaps where moore's law we think of as this astoundingly fast thing uh computing power doubling every year historically so 30 isn't quite as fast as morse law but it's like nearly there more's law is like 40 44 42 percent in old growth something like that historically um so 30 would be like the economy as a whole is growing as fast as the incredible progress, nearly as fast as incredible progress we've seen in computing systems over the last 60 years.
00:37:00
Speaker
Yeah. Yeah. so to say So life would change rapidly and and kind of tremendously under 30% growth Yes. ah So if if we have 2% or 3% growth historically in ah the developed world in the post-war era, then that's like, what, a 36-year doubling time for for incomes? So like once a generation, your income doubles.
00:37:25
Speaker
30% growth means every two, two and a half years, your income's doubling. It's a totally different world. Yeah, yeah, totally. It's actually surprising to me to go back to something you mentioned earlier, that we don't have more research on the question of whether...
00:37:39
Speaker
most growth comes from kind of new inventions or most growth, whether it comes from kind of more production of already existing things. that that's That seems like a massively important and and and interesting and and and kind of deep question that we should we should know more about.
00:37:56
Speaker
I totally agree. um I can speculate that one reason why it's hard is that it's it's hard to think about the introduction of new goods um because it breaks a lot of things both economically and philosophically.
00:38:10
Speaker
So like, would you rather live in the year 1500 without vaccines or today? uh, is it's, it's much harder to make that comparison versus would you rather live today versus 1980 with,
00:38:28
Speaker
approximately the same set of goods or something, hi because you're comparing preferences over different, non non totally overlapping sets of goods. That sort of just breaks a lot of basic ah microeconomic theory.
00:38:43
Speaker
So again, Phil Trammell, Chad Jones, I think are doing some very cool work on this. Hopefully don't inspire others. Yeah. yeah i mean Just thinking about this for the first time, it seems to me like new goods and services are introduced all the time.
00:38:58
Speaker
The kind of set of ah good goods and services that that's available to me right now, right in front of me, is is very different from the one that my, say, dad had access to 30 years ago. it's So isn't there kind kind of a ah Yeah, I mean, what do we do with the fact that this is already happening, that new goods and services are continually being introduced? So ins in some sense, economists must must be thinking about this problem just because it's it's a reality.
00:39:27
Speaker
So I think if you if you talk with Phil, he would say that economists are not thinking hard enough about this issue and sort of slip it under the rug. So one way you can slip it under the rug is by only making local changes, or sorry, ah making local comparisons. Local in the sense of aye comparing you today to you two years ago.
00:39:45
Speaker
Because for over two years in the modern era, at least, two years, there's like not that much change happening. Those are pretty comparable. And like... if If you ah read in the footnotes or whatever of your your favorite econ textbook, you'll you'll see notes that comparing over different decades is a lot slipperier because of new good introduction, how that affects price indices construction, how that affects this adjustment from nominal to real, and nominal GDP versus real GDP.
00:40:18
Speaker
Mm-hmm. So so like this this is a known issue. hi There's just like not a great way around it beyond taking these local changes and sort of extrapolating them, um at at least as far as I'm aware.
00:40:30
Speaker
Yeah, got it. Got it. Okay, so so if we go back to the question of economic indicators for AI timelines, we've talked about interest rates and... Maybe summarize again for us, why is it that interest rates is like the thing to focus on?
00:40:45
Speaker
Why is that a great indicator? Why in particular, is that a number that incorporates a lot of information? Yeah. So one one way of framing this is that Paul Cristiano had this blog post over a decade ago saying three three implications of advanced AI.
00:41:03
Speaker
And the three implications he lists are, number one, growth will speed up. Number two, wages will fall. Number three, humans won't control or sort of ah set the future, thinking about advanced AI.
00:41:16
Speaker
alignment so So plausibly with advanced AI that is superior to human humans at all tasks, wages will be driven down to zero. yeah An issue with looking at wages to understand, to forecast AI capabilities is that wages will not get driven down to zero until we sort of have those capabilities at hand.
00:41:35
Speaker
Interest rates, on the other hand, are forward looking. Financial market prices in general are forward looking. So like the US government issues 30 year bonds regularly. Those incorporate expectations about future savings decisions over the next 30 years.
00:41:51
Speaker
UK government issues 50 year bonds. I think Austria has 100 year nominal bond. Maybe Argentina does too. So this these instruments exist. looking forward a lot. ae and And so that's useful because it's useful for forecasting instead of just contemporaneous yeah ah economic conditions.
00:42:11
Speaker
um Additionally, interest rates are useful, going back to the discussion about stocks, because the effect of aligned AI and unaligned AI goes in the same direction on interest rates, unlike equities or other asset prices.
00:42:25
Speaker
So within the class of economic indicators you can look at that are forward-looking. Interest rates are nice because they both go in the same direction for aligned and unaligned AI. And I at least do really want to take seriously this these risks from unaligned AI.
00:42:38
Speaker
Yeah, yeah. And then third, as you say, in general, financial market prices in particular, even more so than other prices in the economy, i are useful to look at because aye prices financial market prices update quickly are liquid, unlike wages again. For example, those are sort of sticky, only maybe update every year if you're lucky.
00:43:03
Speaker
um So we we have lots of empirical evidence that financial markets are good at behaving in a forward-looking way. There's various sort of amusing historical anecdotes you can point to providing some evidence, uh, I won't say demonstrating, but providing some evidence that financial markets are forward-looking in useful way.
00:43:22
Speaker
One that I like, and I think is, uh, relatively robust is, uh, uh, Armin Alcheon, who, uh, was a great price theorist, economist, worked at RAND in the 1950s, the, the sort of defense think tank.
00:43:37
Speaker
And, uh, The hydrogen bomb, had super bomb, had just been tested for the first time, and it was not known ah publicly what sort of material was used to develop the bomb, just like uranium was used to develop ah the the atomic bomb.
00:43:59
Speaker
ay And so he went and looked at the stock performance of various metal producers and saw that I'm going to get the element incorrect. Something like the the lithium producer had outperformed other ore producing companies in in the period around the the hydrogen bomb test.
00:44:19
Speaker
And he writes this report internal to Rand saying, oh, this is evidence. I think it's ah lithium is a key ingredient in the hydrogen bomb. And famously, his superiors who knew what went into the development of that technology forced him to like burn the draft of the paper.
00:44:36
Speaker
ah So so like that's a very cutesy anecdote. We shouldn't read too much in into cutesy anecdotes versus systematic analyses. um But that's like one vivid example of of why financial markets are good at incorporating forward-looking information.
00:44:50
Speaker
Yeah, yeah. It's actually and an interesting point because, I mean, there was a bunch of secrecy surrounding the development of of nuclear weapons. And you might imagine that there would be secrecy surrounding the development of AGI also. Maybe this is even a nationalized progress,
00:45:07
Speaker
project that's completely locked down. and And how would that affect whether the market would know anything or incorporate anything about AI progress into the the interest rate or into the price of of public companies?
00:45:21
Speaker
Yeah, great question. So there's definitely a world where it's like, 10 people in a basement working on AGI. None of that information leaves the basement. They don't link to anyone.
00:45:33
Speaker
No one really knows or everyone keeps their lips shut. No one secretly trades and make a bunch of money and no financial market prices show up until the AI in a box is unleashed upon the world. yeah i That's logically consistent.
00:45:48
Speaker
If AI occurs sort of more gradually, as i think one should compared to like the 2010s your discussion, Nick Bostrom, super intelligence, your discussion of AI seems much more plausible today than it did 10 or 15 years ago.
00:46:04
Speaker
There's just a lot of public information. The information will get incorporated or the information will get leaked. So I don't know how much people in these labs are are doing trading on the side. if If you know you're you're an ML researcher and anthropic and you're up to that, I'd love to hear from you.
00:46:22
Speaker
um There are other cutesy examples from history. So Suresh Niddu and co-authors have a sort of hilarious example of, they look at 20 different CIA orchestrated coups during the cold war.
00:46:39
Speaker
And they go and look at, it's it's something like United Fruit Company in Costa Rica, I i believe, I could be getting that wrong, ah was like the monopolist fruit producer in Costa Rica a There was some sort of revolution in Costa Rica, forgive my ignorance of Latin American history here, where United Fruit Company was pushed out.
00:47:03
Speaker
CIA, for reasons of fighting communism, goes and tries to orchestrate a coup in Costa Rica. And in this paper, they show that the stock of United Fruit Company and in parallel in these other 20 incidents, that prior to the coup attempt, ah there was excess returns to these companies.
00:47:20
Speaker
And they've had this anecdotal evidence, ah again, sort of incredible, that there were just a lot of leaks from the government initiatives and that and insiders were trading on these expectations.
00:47:32
Speaker
Yeah, so this would mean that we would have to expect AI development to be unrealistically contained and secret for it not to affect affect public markets, even you know especially if we have stories of of of leaks like that that that affect ah prices that that's actually That's surprising to me that that you would see a consistent pattern like that.
00:47:56
Speaker
and is is it Would it just be kind of insiders seeing an opportunity to make to make money and then acting on that? Yeah. yeah or Or government regulators eventually get permission to go into these labs and keep track of what they're doing. It leaks out via that.
00:48:14
Speaker
yeah any Any sort of story like that. ah Yeah. like Obviously, this is all speculation. And again, it's totally logically consistent that the information remains private, either through social pressure legal and legal force, etc. And like it'd be interesting. I haven't done this.
00:48:32
Speaker
If we like went and looked on Polymarketing Calci or something about GPT-5 release dates, how much does it look like there's leaks going on with that? That that would be some interesting evidence. ah But I think the overriding...
00:48:44
Speaker
the more important point plausibly is is the fact that takeoff is slower than the AI in a box in the basement. Yeah. There's the slower takeoff part. There's also a question of how we're developing AI. so So in the era of scaling, you need massive data centers. That is something that gets out almost immediately.
00:49:06
Speaker
And perhaps if we're moving to a paradigm of trying to automate AI research and development, that's something that can perhaps be done more internally and more in secret.
00:49:19
Speaker
And yeah, of course, we're we're we're speculating here, but this is this is something that could push us in direction push in the direction of less public information, I think. That sounds totally right to me.
00:49:30
Speaker
ah Big picture, what i would say is that there there are definitely scenarios you can tell where interest rates will not move before some sort of transform of AI. ay I think it's like the best guess and high probability that they will go up quite a bit beforehand and maybe even already have.
00:49:48
Speaker
And then another thing to say is that if you have ai in the basement where it's developed in secret, no one leaks out about it, perhaps via the story you tell, like if that leads to i nanobots terraforming the Sahara Desert with solar panels and then colonizing the stars and so on, and that is not leading to richer humans for one reason or another or richer vast vast majority of humans say even aye that would not show up interest rates because there wouldn't this consumption smoothing mechanism of people expecting to be rich in the future lowering savings rates today that said like if sam altman's getting super rich in the future then maybe he has enough money to push around markets uh even if he's the only one to move interest rates so
00:50:40
Speaker
the mechanism or The story here really is focused on this particular transformative AI scenario of either unaligned AI, human extinction, or aligned AI leading to consumption growth.
00:50:52
Speaker
Yeah, yeah. ah how How would you settle an argument between you and and a person like Daniel Cocotelo, who's been on on this podcast? And just for listeners tora to remember here, ah Daniel has as very short timelines. He expects us to get to AGI by 2027, perhaps 2028 now.
00:51:11
Speaker
and And this is not a story like developing ai AGI in a basement. This is more like developing AGI in a desert of like... ah robots quickly building up ah the the facilities you need. And this is all happening extremely quickly. So in a sense, is if you have very short timelines,
00:51:33
Speaker
How do you argue? How would you argue with such a personal? How would you settle differences and and find out who's right beyond just waiting and seeing? ah Just because if there are information that can't be incorporated into interest rates, it seems like the mechanism you would you would use for predicting what's going to happen is not really active.
00:51:52
Speaker
Yeah. So I think there's a ah bunch of points one can make here. One is that I'll note that personally, i find the market-based perspective most useful for being less worried about short timelines worlds.
00:52:07
Speaker
Because as discussed earlier, you might expect markets to be sort of more efficient at these shorter horizons. ae So like that this perspective gives me tangibly, substantially more confidence that AI 2027 is less likely.
00:52:21
Speaker
yeah That's one point to make. Another point to make is perhaps markets are wrong. Perhaps AI 2027 is correct. And indeed, like perhaps this is a good time for me to say, I personally am substantially more bullish on prospects for transformative than markets are.
00:52:36
Speaker
And in fact, I've reallocated my most valuable asset, possibly my most valuable asset, at least for now, my human capital away from traditional macroeconomic issues that I studied for a decade. i was obsessed with monetary policy for um almost a decade before i trying GPT-3 for the first time in summer 2020, freaking out a little bit, and now sort of spending at least half of my time working on economics of AI. aye So to some extent, I won't fight the AI 2027 argument fully. Yeah.
00:53:07
Speaker
yeah um That's the second point. A third point is, again, this market-based perspective can only speak to, i think, i this particular scenario of advanced AI leading to rapid economic growth or human extinction.
00:53:25
Speaker
ah can't lead to It can't speak to if this is all happening in the desert, not affecting the broad-based economy. If there's sort of a separate AI robot economy, yeah this market perspective won't speak to that.
00:53:36
Speaker
And then finally, yeah, we we can we can debate inside views about is AI 2027 likely me or not and set aside the market-based perspective. I think that would be the most productive thing to do on on that, but which is maybe another point to make here, which is Inside views on AI timelines, like Ajay Kotra's bio-Anchors report, like AI 2027, like the work done by many others, aye I think is extremely useful, extremely useful complement to this market space view, which does not, for example, take a stance on i the compute centric worldview, ah which I think the vast majority of
00:54:16
Speaker
recent AI work, AI forecasting work ah is is leaning pretty heavily, if not entirely relying on the idea that scaling up compute is sort of all you need, at least eventually, to have transformative AI.
00:54:29
Speaker
The market-based perspective just saying, will we see rapid economic growth or human extinction? That's all you need. Yeah. Yeah. Makes sense. um Yeah, let's see. Oh, you you mentioned ah perhaps a lot of profits from from AI development will flow to someone like Sam Altman or say and a broader group of investors and leading AI figures.
00:54:54
Speaker
What does it mean if wealth is extremely concentrated? What does that mean? you know Specifically, the question is something like they might save and invest differently from from the average person and so if the if the wealth is much more kind concentrated what does that mean for for interest rates yeah so um what i can say is that the idea that ai could lead to ah rapid economic inequality or even like immiseration of large swaths of humanity while making others really rich uh
00:55:29
Speaker
There are elements of that, which I would put in, I mentioned before, this list of two or three best critiques of this whole framework. ah So it could be up there. where ah like If you talk to people in San Francisco about their savings behavior, I think you do sort of get one of two responses, but it's two very polarized responses. One is, yeah, I'm not i'm not saving for my ah kid's college education, which is a real-life example from former your former guest and my colleague at UVA, Anton Koronek.
00:56:00
Speaker
He was interviewed by NPR and him and his wife talk about putting less into their kids' college savings accounts. So that's one possibility consistent with our story. The other possibility is people talking about wanting to save a lot because of uncertainty about the future.
00:56:17
Speaker
yeah um there's There's more that could be said there. But base ah one one important point is that that that would indeed push down interest rates and is a potential counter-argument. The way I think about that is, wait missline what would what exactly would push down interest rates in this scenario?
00:56:34
Speaker
So if I'm really worried about not having a job in five years because of AI, yeah even if there's 30% GDP growth, if that's all captured by Sam Altman and co, and I'm unemployed, then i don't want to draw down my savings today so that I have those savings in five years when I'm unemployed.
00:56:52
Speaker
yeah And that higher savings rate will push down interest rates, so to speak again. Again, interest interest rates clearing the market and supply and demand for savings. yeah um So I guess number one, historically, again, we don't see that higher economic growth is associated with this higher eucyncratic risk that would lead to higher savings and lower interest rates.
00:57:11
Speaker
That's one point. Again, AI could be different. And there's good stories and mechanisms to think that maybe it could be. So another thing I think about here is that Asset prices in general are sort of like a wealth weighted, risk tolerance weighted average of people in the economy.
00:57:28
Speaker
This came up in the discussion earlier about heterogeneous beliefs and the marginal trader. um But it's also particularly relevant here where i if if Sam Altman in the future is capturing all this wealth and he's just saving it all, that's that's depressing interest rates.
00:57:44
Speaker
um So if from 2025 2030, we have sort of the economy as exists today of unemployed, is super rich, and there's consumption growth within that from to today,
00:58:02
Speaker
that would mean that from twenty thirty to twenty thirty five interest rates are reflected by are reflecting the consumption growth of that one percent and so the ten year interest rate today reflects like the average of the first five-year period and the second five-year period.
00:58:20
Speaker
So the 10-year interest rate today would still reflect ae at least in half the rapid consumption growth in the latter half of the period. Although the first five years might have depressed ah interest rates due to precautionary savings due to fears of unemployment by me and the masses.
00:58:40
Speaker
Yeah. yeah so So you think this is this is plausibly a good counter argument, but how plausible is is the scenario itself? Do you think that a i will lead to kind of extreme wealth inequality?
00:58:54
Speaker
Yeah. So to large extent, I think this is a question of political economy and depends on what the political response is. So i economists for the last decade have spent a lot of time thinking about the China shock in the US where China entered the World Trade Organization. this led to a lot more trade between ah u s and China.
00:59:12
Speaker
And arguably, this led to losses of manufacturing jobs in the US. And standard the The standard econ argument would be more free trade between China and the US is good for the economy as a whole. It might impact some people negatively, but those people can be made better off by, for example, taxing consumers who get cheap goods from China, transferring ah some of that tax income to those who lose their jobs or helping them retrain to new work.
00:59:41
Speaker
there's there was limited amount there was a limited amount of such policies in the US in response to the China shock. um And that might make you pessimistic that the future, there there could be i similar amounts of redistribution and that you could see skyrocketing inequality.
00:59:59
Speaker
I think it's it's hard to get away from the idea that there will be skyrocketing inequality in a truly transformative AI scenario. But skyrocketing inequality might still be consistent with everyone being better off.
01:00:14
Speaker
yeah Just because 30% GDP growth, as discussed earlier, is such a massive growth rate, it's hard to have 30% growth in the economy overall and leave the vast majority of people or even large chunk of people worse off.
01:00:30
Speaker
Yeah. So something like perhaps Sam Altman is a multi-trillionaire, but the the average person is ah is a millionaire. And so even though we have extreme inequality, we we you know what we actually care about is is is more how people, you know the welfare of the population at large.
01:00:49
Speaker
Yeah. And I think the China experience here is a good example where if I'm recalling my numbers correctly, like inequality has certainly skyrocketed in China over the last 40, 45 years.
01:01:00
Speaker
But poverty has been reduced so much. The average person, their their their life has been improved so substantially that concerns about inequality are are less of an issue in China than in the West.
01:01:15
Speaker
Yeah. Yeah. Before we we move on to other topics, and we talked about interest rates as ah as a great indicator of what's going going to happen with with AI. Are there any other indicators that might be interesting to look at?
01:01:29
Speaker
So I wrote some someone down or some of them down that I could think of. Maybe maybe CapEx, capital expenditure from from the large corporations, maybe patent filings.
01:01:40
Speaker
perhaps papers published by the AI corporations. what what do you What do you think of other indicators? We've discussed stock prices as perhaps not, I mean, somewhat useful, but but perhaps not as useful as as interest rates. is there If you were to write another paper on on on any other indicator, what what would you choose?
01:02:00
Speaker
So if I to choose an indicator for forecasting, it would be equities because stock prices are forward looking. Yeah. Capital expenditure is also forward looking to some extent, but it's less likely to be made on third year horizons. yeah i Wages again are contemporaneous. So I think there's a lot of good work that still could be done with stock prices and understanding and ah Why is Nvidia's market cap so high?
01:02:26
Speaker
How much of that is future profits ah because of high markups versus actual quantities are going to be high? ah Why has Nvidia increased so much more than TSMC has?
01:02:38
Speaker
Like there's a lot that could be done with stock prices that hasn't been done. I would love for someone to write that paper. yeah um more More broadly about economic indicators for interpreting ai the big ones I think are wages.
01:02:53
Speaker
interest rates, the labor share the economy, so how much of national income is going to workers versus going to capital owners and other, and something like unemployment or ah labor force participation, so how many people can find jobs and want jobs, or how many people just are working in the economy as a whole.
01:03:14
Speaker
Those prices and quantities, I think, would really reflect the different scenarios that people talk about for the possible implications of AI. And you can also look at those sector in the economy, in particular to see if AI is affecting the economy in a heterogeneous way.
01:03:32
Speaker
For example, automating all of white collar work, whereas while making blue collar work more relatively valuable. ah So you might see skyrocketing employment in factory workers.
01:03:44
Speaker
and rapidly declining employment in software engineers. aye ay so So those four or five big indicators for the economy as a whole and by sector.
01:03:55
Speaker
Yeah, yeah. When you think of the labor share, um Yeah, of the economy. but So how do you define that? Because I'm guessing most people, most individuals are both um kind of capital owners and ah workers.
01:04:13
Speaker
and And so they maybe maybe people own some assets, they own they have some retirement funds, they own a house maybe, and they also go to a job. How do you define the labor share versus the the kind of capital share of the economy?
01:04:26
Speaker
Yeah, this this terminology that I and economists and others use, I think it's actually really bad of workers versus capitalists. It's just a Marxian framing when exactly as you say, most people have some mix of those, certainly at any point in time, and then even more so over the life cycle where yeah Yeah, yeah, I won't belabor that, pun intended.
01:04:49
Speaker
um So what is the labor share of technical definition is the total wage income earned by workers divided by GDP. ah So that reflects in in in in the world as we know it today, at least that reflects the how to put this.
01:05:09
Speaker
um It reflects what share of output is attributable to labor in in in a specific sense under specific assumptions that apply pretty well, we think, to today's world.
01:05:23
Speaker
Yeah. Okay. Interesting. So, so it, it, you can't really look at papers directly or look at patent filings directly or something like that.
01:05:34
Speaker
I'm just wondering whether there's some like intellectual output from the companies that would be worth looking into a measuring. and is it the case that whenever there's a new paper or whenever there's a new patent filing, that is immediately incorporated into the price of of the stock? Or how should I think about this?
01:05:53
Speaker
Yeah, so I should have said all your suggestions were also great. Someone should like make a dashboard of all these things. ah Capital expenditures in particular, I'll note, like this this gets a lot of discussion, but I think it's still underrated that hyperscaler CapEx, that's like Google, Meta, et cetera, their are capital expenditure is like 1% of GDP.
01:06:11
Speaker
yes That's comparable to the height of the dot-com boom. Not clear that we're at the height of anything. ah the The numbers I've seen for real... The the railway boom in the UK in mid 19th century is like 2% sustained for 25 years of GDP on railway investment.
01:06:28
Speaker
ah So 1% is getting up there. Of course, I should say the railway boom. era, that 2% wasn't constant over time. It was higher and lower. yeah um So there's a lot of capital expenditure. I just moved to Virginia.
01:06:43
Speaker
Basically, the ground's covered in wires from all the data centers. Obviously, that's exaggeration, but lot of data centers here. yeah um You bring up the impact of patents on stock prices. There's very interesting work that's been done by folks looking at historically when patents were granted.
01:07:00
Speaker
If you look at the stock price of firms who were granted that patent around the day the patent was granted, ah you can see a large impact of patents on stock prices.
01:07:14
Speaker
And you can like accumulate that, add that up over all companies in the country or the world over the course of a year. And you can sort of see a measure of innovations, i value captured by company profits as reflected in stock prices.
01:07:30
Speaker
And you can like see how that varies over time. a Leona Kogan and many co-authors have have done this very cool work. um So something like that could could be done today as well.
01:07:45
Speaker
AI, it's a bit trickier because there's less that gets patented. Again, many of these companies aren't directly listed, though, again, they they often have publicly listed affiliates i like Microsoft, et cetera, Google.
01:07:59
Speaker
The moonshot economic indicator that I think of is... Benchmarks are this huge thing in the AI world. We want to look at how good is AI or new AI models at well-defined tasks ah that are scorable in automatic way, or like the math Olympiad recently. yeah i In economics, there's this database produced by the government in in the US, ah the O-Net database that has this list of 19,000 tasks in the economy performed by American workers.
01:08:32
Speaker
And It would be amazing if we could have ah some very expensive program, because it would be very expensive to do, measuring what fraction of tasks can models perform themselves, or how much do ah models improve human productivity on each task, and keep track of that over time by task.
01:08:54
Speaker
Then you could do things like what share of tasks has gone up over time or what share of task is automatable and how has that changed over time? You could do the standard thing in AI and take a trend and just extrapolate it. You could try and see when will we have 100% automation of the economy of 2025, your tasks, new tasks are constantly being invented.
01:09:16
Speaker
um that That I think would be the moonshot for ah if if the government decided and wanted to throw a lot of money. at understanding the economic implications of AI, that that would be what I would suggest.
01:09:27
Speaker
yeah That would be extremely valuable data to have. that That would be super interesting, but also quite complex, as you mentioned. yeah so how would um How would these tasks change over time? so I would imagine as AI gets better, some tasks are now automatable.
01:09:45
Speaker
Those tasks now make up a smaller percentage of what people actually do. so the The task list from 2024 is not relevant is not as relevant in anymore.
01:09:57
Speaker
So I guess the that what you're tracking is also how which tasks people are spending their time doing. Yes. So the way I put this is that the second moonshot I wish the government would find would be keeping track of exactly this, where this ONET database of tasks is static over decades. It gets very occasional updates.
01:10:21
Speaker
So number one, like we could really do with an update in the year 2025. That'd be great. Or potentially just sort of a new way of thinking about this. Like, should we be using technology to keep track of what people are doing minute by minute of the workday?
01:10:34
Speaker
And even using AI to classify what people are working on or something more dynamically. So like maybe this whole static database structure needs a rethink. I don't think that sounds like it'd be really hard for one individual researcher to try and do on their own, ah but the government might be able to do it.
01:10:51
Speaker
ae so So a one-time big update would be nice. Even better would be real time updating every year whatever, this list of tasks. how people How is people's time allocated across these tasks, changing over time?
01:11:05
Speaker
What new tasks are added versus deleted that humans do versus machines? That would all be great. we Yeah. to my knowledge, at least, don't have that information. Yeah, we would have a much more granular view of what's happening. I think when you're looking at the unemployment rate or the labor participation rate, it's it's it's it's all condensed into one number and you don't really know what's causing what and it's Of course, as you can see that there's not a massive effect on of AI on unemployment yet.
01:11:31
Speaker
But it will be certainly it would be great to see ah whether some industries are being affected. This could be happening right now. and And maybe we don't have great insight just because it's not moving the needle on the one number that we might be looking at.
01:11:46
Speaker
Yeah. And of course, these numbers are certainly the US, I'm i'm sure elsewhere, broken down by industry. ah ae There's surveys that are pretty frequent ah that get you richer demographic information. So we could see things like our younger coders losing jobs. That would be super interesting.
01:12:07
Speaker
But those surveys, while in the tens of thousands, provide very nice data for lots of purposes. The data is pretty noisy, is pretty small. What you kind of want to go down to for new graduates working in CS, yeah how is their employment looking? hi Expanding those or using private sector data, which I think there's there's work that should come out soon on this topic. i ah that That would be really useful to look at.
01:12:34
Speaker
Yeah, yeah. Here's a question I've asked a a bunch of guests. It's about the difference in economic effects the economic effects of AI and then what we see on benchmarks.
01:12:46
Speaker
It is surprising to me, and I think to to ah other people also, that we are seeing this very strong performance on a bunch of benchmarks. So AIs are now incredible at passing college exams, and they there they can pass the bar exam, they can do all...
01:13:03
Speaker
They can score highly on medical kind of examinations. they can They can do especially well on on coding and and math and so on. and how is it How is it the case that I am in daily dialogue with a chatbot that is better at me at math and and and coding, and but it is not yet it is not yet it it doesn't ah have massive economic effects yet?
01:13:30
Speaker
Yeah, so like this is just one of the greatest questions of our time. So i can offer some speculations or a couple of different answers. One is like maybe capabilities really are this good and it just hasn't, the technology has not diffused through the economy.
01:13:46
Speaker
So firm managers are old, crusty. They don't want to adopt ah AI to replace workers or workers don't want to adopt AI because they don't know about it.
01:13:58
Speaker
Things like that. That's one possibility. Slow diffusion in general is a major lesson lesson of economic history that technology takes time to diffuse for various reasons. um i i I don't think that's sort of the main reason just that ay i people in San Francisco want to fantasize or something that ah old fusty people don't want to adopt new technologies.
01:14:23
Speaker
I don't even think many people believe that because it's so a priority hard to believe. Yeah, it it doesn't seem super plausible to me either just because, i mean, it is quite easy to incorporate these models into your workflow.
01:14:35
Speaker
And you can i mean yeah it's something that's available to to individual ah workers that they can use for themselves. And they have incentives to try to use these models to make their their lives easier and so on.
01:14:48
Speaker
It is something that is yeah much much more easy to implement in the economy than than, say, if you had to spread physical hardware to do something better. It's definitely true, but I think will lead to or i want to disambiguate two things where what one is pure diffusion, aye just have people learned about this technology or something like that.
01:15:10
Speaker
Another is incorporation to workflows. Let me drill in on that where ah workflows involving other people can be hard to change around. And so there's a Microsoft study where they experimentally rolled out copilots to some workers versus others.
01:15:29
Speaker
And they found that workers who started using copilot really reduced amount of time they spent on email a lot because copilot was helping write emails faster. um But the time spent in meetings didn't change at all ah because the coordination with other people is not something that...
01:15:46
Speaker
ah AIs as they exist today in ah helpful, harmless chatbots can can really help with. And work by people like Eric Merylson and others has emphasized that to fully gain the benefits of a new technology, firms will often typically historically have to reorganize their internal processes to take advantage of these new technologies.
01:16:13
Speaker
The most famous historical example being the adoption of electricity, initial factories in the 19th century, sort of just ae taking existing workflows, sort of plugging electricity in, not changing anything else.
01:16:27
Speaker
But then over the course of actual decades, really changing to ah the sort of Model T setup that completely changed these internal processes to best harness the new technology.
01:16:39
Speaker
And it's very plausible that those sorts of intangibles, these internal processes, are are slow ah to figure out and not something directly improved by AI.
01:16:50
Speaker
The counterpoint there is that maybe we'll have drop-in remote workers in in two years or whatever. And those are drop-in remote workers. You don't need to change from internal processes at all. Yeah. So these drop-in remote workers will be able to understand the context they're working in. They would they would function like remote workers, basically.
01:17:07
Speaker
They would understand the context. They would understand the code base. They would read the internal documents, emails, and so on. And the first part of that argument is is actually quite plausible to me that that we are not taking as much as advantage of these models as we could. as we could We're not properly integrating them. We're not kind of pushing pushing to get as much out of them as we could because it takes time. It's difficult to integrate ah ai into into all of our processes.
01:17:38
Speaker
Yeah, totally. so So these two arguments, diffusion and reorganizing internal processes, yeah sort of take the idea that these models are really good. They're just other things that need to change yeah before the effects diffuse through the economy.
01:17:52
Speaker
But it's also possible that benchmarks just aren't representative of economic tasks. Yeah. sort of as hinted at earlier, benchmarks need to be things where are the output's verifiable.
01:18:03
Speaker
Uh, math Olympiad is something where you, you, you can check whether you have the right answer or not. Uh, maybe takes some time to verify a proof or something, but it it is sort of checkable. Whereas writing an economics paper, how do you verify whether you did a good job on that or not?
01:18:18
Speaker
Much harder. yeah Um, so benchmarks are a certain kind of task, uh, To my knowledge, there's not so much work investigating how is progress advancing on tasks by characteristic.
01:18:34
Speaker
So there's this famous meter study, meter being model evaluation and threat research, a sort of think tank that ah studies, evaluates new LLM models.
01:18:47
Speaker
new LLMs, they have the same study ah showing that the horizon of tasks that LLMs can do has been increasing, sort of doubling every seven months or something like that.
01:18:59
Speaker
yeah and So GPT-2 could do like ah one second task, GPT-5 can do a two hour and 15 minute task, something like that. On a narrow set of tasks that they they have, these HCAST machine learning or software engineering type of tasks.
01:19:19
Speaker
That's amazing research, a super important data point. And in the paper, they look at if you have messy tasks versus non-messy tasks, do we see faster slower progress on messy tasks versus non-messy tasks?
01:19:33
Speaker
They don't actually have very many, very messy tasks or any maximally messy tasks at all. They they don't find differential progress i in messy versus non-messy tasks.
01:19:46
Speaker
They do find like a lower level of progress, but the slope is the same. yeah I would still be interested in seeing for sort of O-Net as a whole, for example, does that sort of result hold up? um Because I think it's very plausible that benchmarks are these narrowly defined tasks that don't really capture the breadth of what a worker does every day.
01:20:08
Speaker
Like work is pretty complicated. Yeah, yeah. and And actual workers tend to to carry out plans over weeks or months, perhaps years and so on. But still, I mean, what does it even mean? What does it even mean for a task to take a week or a month or something? That is something that i that I'm kind of interested in, because we are We are kind of on this curve, according to the meter study, where the models in two or three years, I think, will be able to do a month's work, um on a task that would that would take a human a month with 50% success rate. and what is What is a month-long task? I can't actually kind of...
01:20:50
Speaker
perhaps conceptualize what it would mean to work on one task for a month. ah Perhaps I just lack focus or something, but do have like do you think it's plausible that throughout the economy there are tasks that are very long term?
01:21:05
Speaker
and And what would be examples there? So there definitely are tasks that are very long-term and they're very painful. I see that as a researcher where I, to me, having worked in industry before going to grad school, like research just has such a long cycle before you put out a paper or whatever.
01:21:22
Speaker
It's much more painful. Feedback loops are much slower. So those tasks do exist. But exactly as you said, I think that's not a huge share of the economy. the The way... the the The framing of this that I find convincing, I'm totally recapitulating argument from Toby Ord here, is that so if a model can do a one minute task, why can't it do two one minute tasks in a row? And that that means it can do a two minute task.
01:21:49
Speaker
yeah And the argument he makes is that
01:21:54
Speaker
If you can do a one minute task with 50% probability, as you said, this is this is what the meter study is looking at, then that means doing a two minute task. i Those two tasks are in chain together at 50% probability each.
01:22:08
Speaker
If those were say independent probabilities, then it would be a 25% probability of succeeding on the two minute task. Hence why models are worse at longer horizon in tasks than shorter horizon tasks. I agree without sort of decomposing it that way, it's it's it's not obvious.
01:22:23
Speaker
what what distinguishes the short horizon in task versus a long horizon in task. i it It does feel like there is something i that maybe we just haven't captured. Otherwise, this TopiGuard framework is is what I find most useful for thinking about it.
01:22:38
Speaker
Yeah, yeah. This is, I think, also what you see in coding that if a model is unable to fix its own mistake, those mistakes accumulate over time such that the output is no longer useful beyond a certain kind of ah time horizon of a task.
01:22:55
Speaker
um I do still, i mean, perhaps this is getting too philosophical, but it's just, even if you're even if you've written a paper and you're now you're now trying to incorporate feedback on the paper, and this is and this tedious and slow process that that might be necessary for good research, you are not spending you know you're not spending all your time on this one task for six months, say, right?
01:23:18
Speaker
So, Yeah, I mean, even even book writing or something very very concentrated. Okay, my question is, what do you think is the upper limit for how humans can for how much a time humans can spend on a task? And could the AI be approaching that limit?
01:23:36
Speaker
yeah That's a very interesting question. Upper limit. ah ah the The sort of rule of thumb I have in my head for when would I be satisfied that we have AGI is when any task that takes a month.
01:23:47
Speaker
Yeah. AI can do it. That's just a completely made up number. yeah another Another thought i can throw throughout is thinking of decomposing the world into tasks, this task-based framework that has, in economics, become very dominant very quickly in this macro labor area and is also used often in the AI world.
01:24:12
Speaker
Perhaps it's just sort of a conceptual error to decompose the world into individual tasks, as opposed to thinking about how tasks fit together into a broader puzzle. aye And like human civilization as a whole or groups within the human civilization certainly spend, aye hi have have initiatives that last many years to achieve a goal.
01:24:38
Speaker
And maybe there's some category error in to to think of decomposing these into smaller tasks. And they're not separable in some way because you have to hold context in your head. And if you don't have a long enough context window, like possibly these LLMs don't.
01:24:53
Speaker
um You just can't you can't do it, so you you can't decompose it that way. Or the self-correction, like you describe, can't be separated across these tasks. And so don't need to think in terms of like grand projects, grand arcs, grand initiatives.
01:25:08
Speaker
ah But this is me just philosophizing. Though if someone wants to come up with a replacement for the task model in economics, I think there could be something real there. Yeah, interesting. I think we should end by chatting a bit about the most interesting kind of open problems as you see it. We've we've touched a about upon some of them in this conversation, right?
01:25:29
Speaker
The things we would like to know about AI ah that are at the intersection of ai and economics. Are there others that come to mind as like, this is you know this is like the research paper you would love to see or you would love to write?
01:25:43
Speaker
So the big answer is that there's a lot. There's so much low hanging fruit. It's super exciting to work in the economics of AI. I have to advertise it to other economists. You should yeah you should consider transitioning over. Don't ah get hung up on the sunk cost of all the years you've spent studying monetary policy or whatever.
01:26:00
Speaker
and The question is whether being an economist actually makes you better at avoiding the sunk cost fallacy or whether economists are kind of just as human as the rest of us. That would be interesting paper. I would read that paper. um in In terms of most important questions. So some of the most important questions that I think of as like the Hamming question, what's the most important question in your field?
01:26:24
Speaker
Yes.
01:26:26
Speaker
I'll distinguish between empirical questions or sort of sort of applied micro empirical questions versus microeconomic theory questions that might have relevance to AI safety, and then grand macro theory, or maybe bit of empirical macro questions.
01:26:44
Speaker
And macro is where I have the most familiarity. So let me start there. and we can talk about the others so ah if if there's time or interest. So within macro, I think the big question is, Will AI lead to a speed up in economic growth or will it get bottlenecked by certain sectors or areas?
01:27:01
Speaker
So those bottlenecks could be things like energy, we just don't have enough, fossil fuels, that's ah that's a limited quantity, that bottlenecks the price of fossil fuels is going to spike. Land, potentially.
01:27:13
Speaker
Land, very plausibly, or certain sectors of the economy, AI just isn't good at. Clearly, we've seen much more progress in the cognitive domain than in the physical domain. Robotics is i has has a lower level of progress.
01:27:28
Speaker
Will we end up all as blue-collar workers in 10 years? so So where will the bottlenecks be? I think it's plausibly the best answer to the Hammond question. yeah The second one is this idea of i automating AI research that you brought up a bit earlier. aye How much, to put it in economic terms, how much dynamic complementarity is there between AI today and AI ah tomorrow? a To spell that out, ay if if we have faster AI progress,
01:28:02
Speaker
Let put this is different way. Going back to IJ Good, at at the very least, there's this idea of recursive self-improvement. yeah That if you have better AI, it can write better it can do better AI research, which will lead to better AI, so on and so forth.
01:28:18
Speaker
And the speculation that's often laid on top of this is that this would lead to an intelligence explosion. that That's IJ Good's terminology. ah A recursive loop like that does not necessarily lead to an explosion does not necessarily lead to a mathematical singularity.
01:28:36
Speaker
It depends on the strength of that feedback loop. So aye to to I'm trying to think about how to say this about drawing graphs in the air. So if the feedback loop is super strong, then you can indeed have a mathematical singularity, infinite growth in finite time.
01:28:52
Speaker
If the feedback loop is only moderately strong, you can just have exponential growth, which is sort of what we're used to in the post-war era or in the last 200 years. If the reinforcement, the feedback loop is too weak, you can even have things leveling off.
01:29:09
Speaker
So you have self-improvement. It leads to faster AI, and it's always leading to a little bit more and more AI improvements, but things level off. yeah So what is the strength of that feedback loop?
01:29:21
Speaker
What is the diminishing returns, the intertemporal diminishing returns to AI progress? ah Or to put another way, ah are ideas getting harder to find in AI? ah Or are they getting easier to find?
01:29:34
Speaker
a In which case you would see a singularity. um there There's a limited amount of work on that. Edgar Erdl and Tamai Beziroglu, I think, i have have some papers on this looking at progress in AI chess and ah maybe one other domain.
01:29:53
Speaker
And anyway, they they have the best work on this. yeah More work on that could be done. That one is really interesting. I mean, big it's it could be it's potentially so consequential for for for the future we're likely to see.
01:30:05
Speaker
um Yeah. Isn't it the case that that in in any domain, ah you will kind of pick the low hanging fruit, but find the ideas that are easiest to find first, and then you will face the kind of, it will be ah more and more difficult to find good ideas for how to improve ah beyond that.
01:30:24
Speaker
or Or is that a misunderstanding of the of the kind of ideas and growth literature in economics? Yeah. So that that's definitely the prior in the literature.
01:30:36
Speaker
I think that's definitely the right prior just thinking about reality. Yeah. One could imagine, though, that it takes a really long time to pick all that low-hanging fruit. and Sufficiently such a long time that...
01:30:48
Speaker
for an extended period, like you know maybe even centuries or whatever, there is a period of increasing returns to scale where getting the low hanging fruit allows you to you know beef up your muscles and pick fruit even faster.
01:31:00
Speaker
yeah Even though eventually you'll have to reach higher up on the tree and those strong muscles won't help you reach the apples to yeah really extend the metaphor. um and and And so if we look today, what do we see?
01:31:16
Speaker
Increasing returns or decreasing returns? um Yeah, so so on on the macro side of things, those are sort of the top two things that I think of. yeah ah on On the micro theory side of things, I will give a pitch for...
01:31:31
Speaker
I think there are a lot of lessons one could take from microeconomic theory for AI safety, for agent foundations work, where like the von Neumann Morgenstern axioms that are sort of often discussed just as one example in the AI safety world, like that is sort of the foundation of modern economics.
01:31:52
Speaker
i want One would hope that there's further lessons from micro theory there. And there has been work done Eric Chen, ah Alexis Gersengren, and Sammy Peterson have very interesting work on ah the AI alignment problem from a microeconomic theory perspective as do a few others.
01:32:10
Speaker
aye And so i I can imagine that in a few years, if this problem seems increasingly serious, that there will be more microeconomic theorists working on this question.
01:32:21
Speaker
Yeah. do Do you think there are other areas of economic theory, perhaps something that's that was conceived way before ai was even a thing that's relevant to to to to thinking about AI?
01:32:34
Speaker
Yeah, great question. So I think there's a lot. I think there are a lot of essays or papers that could be written just applying ideas from econ broadly, in particular, perhaps economic theory to the problem of AI.
01:32:48
Speaker
So for example, after the 2008 financial crisis,
01:32:54
Speaker
Regulators developed this idea of stress tests for financial institutions, going in and doing a simulated scenario of a financial crisis in banks, given their sort of asset holdings, their loans, et cetera, to see if they would survive the financial crisis.
01:33:11
Speaker
that this made up financial crisis. So there's this large econ theory literature for some reason investigating when is this efficient or something like that. That's not so far from the idea of red teaming in AI where ae Anthropic has a team, others have teams i trying to sort of battle test LLMs in the worst case scenario. Yeah, I And in fact, therere there's a paper by Joao Guerrero, Sergio Ribello, and a third co-author, Forgive Me, I'm Forgetting, sort of taking this exact idea to the AI world.
01:33:44
Speaker
And fact, in the first draft of the paper, if I'm not mistaken, they didn't even use the term red teaming because they were economists. They hadn't heard this term. In newer drafts of the paper, like, oh yeah, this is a theory of red teaming, the optimality of red teaming.
01:33:57
Speaker
Or there's a literature on ah insurance against cyber attacks of firms. So firms can take out insurance about getting hacked. ah And there's been discussion, ah Gabriel Weil and others have written nicely about should ai companies be forced to take out liability insurance as a mechanism for...
01:34:17
Speaker
i ah encouraging them to internalize risks from advanced AI. yeah the cyber The cyber insurance risk literature could be adapted there, or I'm sure other parts of the insurance literature. And in fact, ah Gabriel's work is, of course, directly economics related itself.
01:34:35
Speaker
yeah So those those are some miscellaneous examples. Yeah, I think there's a bunch of things that could be done. Yeah, and and I think in in general, and this is just my impression as a non-researcher, but I think there's something about finding an intersection of some area that's been studied deeply and then applying those ideas to a completely different area. So AI and economics would be an example. Perhaps Egeo Contra's report on bio-anchors is an example of studying machine learning by looking at evolution evolution.
01:35:07
Speaker
kind of finding some intersection that's that's that's unexplored is is often fruitful, is my impression. Yeah, totally. So ah to circle back the whole conversation, but the way this paper on AI and interest rates came about is that, again, for for like a decade, I was obsessed with monetary policy. How do we prevent recessions? How do we prevent another 2008?
01:35:27
Speaker
And in the monetary policy world, central bankers are very worried about predicting future inflation. And so this gave this background in forecasting particular. central bankers often look at financial market expectations of future inflation over the next 10 or 30 even years, because there's instruments that directly forecast inflation expectations.
01:35:47
Speaker
hi and And so that sort of market monetarism or market-based perspective ah is is how Trevor, Zach, and I ported this interest rates perspective to the AI forecasting world.
01:35:59
Speaker
Oh, perfect. I think that ties up our conversation and nicely. Basil, thanks a lot for chatting with me. It's it's been great. Thanks very much, Gus, for inviting me on. Super fun conversation.