Lunch Money with Paul Krugman and Heather Cox Richardson

Añadido: 2026-05-30

[00:00:02]Hey everybody. I've missed you uh as I've been traveling and we have here South Minneapolis and the Hudson Valley, New York. Hot and sunny there. It's hot and sunny here. It was cold in Maine just two days ago and today it's 90°.

[00:00:16]Phoenix, Arizona. You know, 90°.

[00:00:19]Rochester, New York. Reading, Connecticut. Milo, Washington. You got me on that one. Uh Prospect, Connecticut. Vienna, Austria.

[00:00:29]Kristen is here from Facebook. So, hi Kristen. And Facebook is on board. Um, how you doing, Professor Krugman? I know you're on vacation.

[00:00:38]>> Yeah, I I'm uh like I wrote the other day, I'm in Europe, which means I don't have to think about Trump 100% of the time. Only about 90%. So, um that's that's a little bit of release psychologically.

[00:00:49]>> It's really astonishing, isn't it?

[00:00:51]>> Yeah. So, yeah.

[00:00:54]>> Anyway, but uh hopefully we don't talk entirely about him today.

[00:00:58]Well, no. So, so I actually am really interested and would love to hear what you have to say about artificial intelligence, not itself as an entity, but as a factor in the economy because boy, it sure looks to me like we are way overinvested in AI. I think the growth on the stock market is basically AI companies. We know now that AI is more um there's more construction in AI data centers than there is in commercial real estate. Now there's a lot going on in commercial real estate. Um and I'm wondering I just can can we just talk about that and you walk us through what this looks like because everybody keeps saying oh it's a bubble like you know the housing bubble or like you know the dot bubble and I'm looking at it and saying boy it looks to me like it's deeper in our economy even than that. So what do you think about it? Okay. So I mean obviously we history history history is mostly what we have to go on because we and there had been many um bubbles like this. I mean the um um s it does look there are clear resemblances to the dotcom bubble which was really more of a comm bubble because telecoms were a big part of it too. Um and uh it but it looks like the railroad bubble. uh it looks like the canal bubble in uh in England and in even earlier than the railroad bubble. Um so there's a there's a lot of examples of stuff like this that have happened. Um, as always, it's it's, you know, that there's there's some broad similarities and and these keep, you know, bubbles, most of the bubbles were pretty clearly bubbles at the time, and uh, that was certainly true for, which I still remember in real time, but this one is I'm finding that the contrasts with the late 90s bubble are are really illuminating. This one is really uh you know obviously it's technology lots of investment there's an enormous enthusiasm of a kind but in other ways it's it's quite different so I don't know if where you want I mean >> so let's start with this I I'm sorry but what exactly is a bubble >> yeah um it's always a question but a bubble is you more or less it means that people are investing in something that isn't really going to has no realistic chance of paying off not socially but just commercially to to an extent that justifies the amount of money being thrown at it. And crucially pretty much a bubble is something that people do because everyone else is doing it. So I mean they uh Bob Schiller the great bubble theorist of and of modern economics said that bubble is a natural Ponzi scheme. It's something where you get in and you make money because other people get in and they keep people keep on coming in because everybody before them made money, but in the end it's it's a game that there the money isn't really there. It's all depends on fresh crops of suckers coming in and at some point you run out of suckers. So, but and that in a Ponzi scheme somebody a Ponzi uh somebody or a Bernie Maidoff does it deliberately in a in a bubble it happens u naturally. There is there was nobody orchestrating it but nonetheless uh it it the logic of it is the same as a Ponzi scheme. And so basically no it's there's a lot of pornography test here.

[00:04:22]you know it when you see it. But the uh the um the bubbles are it it's not just the fact that people are wrong, but people are wrong in a kind of what should have been a predictable way and where it's really something that is sustained by the by by the sustained by the momentum by the fact that other people keep on coming in until they don't.

[00:04:44]>> Okay. So when historians talk about this the word that you know we'll be talking and talking and somebody will just say tulips um is that is that something that you can explain to people that that reference because it's kind of a cool story and when you take it out of the economic system that we understand now it's easier to see I think.

[00:05:04]>> Yeah. I mean the okay I I'm not really fond of the tulips analogy but sort of the first thing that that people think of as being something like a modern bubble was the tulip mania in the Netherlands I guess 17th century uh you uh Netherlands was the you know sort of not quite the first modern economy because they weren't quite modern but they were on that way they were commercialized they were banking and people were speculating in tulips now tulip bulbs which was we're were in fact valuable investments but crazy people were were buying tulip bulbs and the prices went up because people were buying them and then their prices went up further and um so I mean the let me I'm not really fond of it I mean you can see the financial logic there but there wasn't a whole lot of you know real investment people weren't building um uh tuna infra tulip tuna t people weren't building tulip infrastructure they were uh but I guess in terms of the psychology, the market logic, it was not that different from uh from railroad shares or or do shares. So yeah, and it it's it is telling you I guess the fact that you know this is the this is the the Holland of of Rembrandt and uh and there was not only wasn't there an internet, there wasn't even telephones and yet the psychological logic was the same. And that's kind of telling you that that in some ways there's a kind of universality about bubbles. But >> well, so so when we look at AI now, am I correct that there are two super companies in which the majority of AI money is invested?

[00:06:44]>> Yeah, I mean there's uh I would have said there are two big ones. There's OpenAI and there's um Anthropic and um who are the the the big players. But you know it's it's an industry. Um it's not just these are the two biggest AI models. So you're either talking to chatb GPT or to claude which are the two companies leaders but then Google has its own model which is Gemini and then Elon Musk has a a really bad one uh Grock and um the um um and then there's a bunch of Chinese versions which are where they've taken a very different strategy. So it's a little bit more complicated than that. And then there there's this network. So a lot in a lot of ways you want to think of this whole AI boom bubble as being a little bit another parallel, but it's a little bit like the the California gold rush where the the modelers are, you know, pe the people who are selling uh you know anthropic and and open AI are like are like miners, prospectors looking for gold. And um all we know in California in the you know 1840s was that the people looking for gold mostly ended up bust. But the people who made money was basically the Levi Strausses. You didn't make money by finding gold. You made money by selling equipment uh by selling jeans and and picks and shovels and also uh brothel and and liquor to to the prospectors. And um so there is the the equivalent of that now are companies like Nvidian which is selling the the specialized chips that go into this and there's a bunch of other they're companies making a lot of money basically renting out um computational capacity. Uh, so now we're starting to see at least a little bit of money being made by by Anthropic.

[00:08:41]Anyway, Claude, I all of my friends are playing with Claude and I I just can't get myself up to do it. But the uh um which is the vibe coding which lets you do programming without un knowing how to program. And so there is actually they're actually making some money because people are subscribing to that service. But it's at this point it's mostly most of the money being made is from people basically selling equipment selling uh the the suppliers to this thing. Um and you know the so the the question from a kind of economic point of view is whether there will ever be enough revenue uh whether people actually end up paying enough for for AI this thing that we call AI um to to justify all of the money being thrown at the industry. And you know history would suggest there's a very good chance that you know the most likely outcome is no.

[00:09:36]The most likely outcome is that it will end up being a waste.

[00:09:39]Uh it again history doesn't always repeat. So maybe maybe this pays off but I don't think that explains the enthusiasm.

[00:09:48]>> Well, it's interesting because one of the things that you're seeing lately is the the the changing model for paying AI. That is most of the use of AI currently is subsidized really quite heavily. For every dollar of computing power that people use, it's subsidized between three and $25 at the minimum.

[00:10:07]And the idea that people are actually going to pay the extraordinary costs that certainly right now it it would warrant doesn't seem like it's going to happen.

[00:10:17]>> Well, there's a question. Let let me give you let me play devil's advocate here for a second. um when the internet when the dotcom bubble happened and when there was all of this stuff people were offering you know all these services on the internet that were really they had no way people weren't willing to pay remotely enough to justify the money that was being thrown at it but what eventually happened was that a few companies managed to create walled gardens managed to create enclaves I mean essentially um uh Facebook is a w garden where people pay and or or or watch ads or whatever. Uh Google basically ended up being a kind of w garden. I mean the the search was free.

[00:11:02]Uh but the Google was making money out of out of pushing targeted ads. um the uh Amazon eventually, you know, we we used to joke I I'm old enough to remember when Amazon was famously unprofitable and was never going to be profitable, but turns out that well, in the end, Jeff Bezos created um somewhat he built a moat with the with all of the infrastructure, the distribution centers. And so now Amazon is is a a huge money maker and and um uh and and evil, but that's another story. and the um um and what's happening with the AI is to a certain extent they're building walled gardens from the beginning. So I I know people who've been using Claude uh or have been playing with Claude I think would be a better description and the results have been terrible and it turns out that the results are terrible unless you pay by a higher tier of service. Now even there it's not remotely enough to justify the expense but clearly Anthropic is trying to create a situation which people get hooked on vibe coding and then end up well they're you know they're they're addicted and they're they're going to end up shelling out large amounts of money to to have the the claw that works and something like that you can already see the outlines at least of the how the industry intends to make money. Now I I um history suggests that's that well that that usually there are only a few winners. Although one thing that's also different sorry I'm I'm going on but uh from the the the dotcom bubble is in the docom bubble there were hundreds of players trying to uh succeed and in the end um just a few uh highly profitable corporations survived. This is not like that. This industry at least on the US side is just a handful of players. So the chance that that um that one or two or maybe three big AI models will end up becoming u uh highly profitable monopolies is it's not that remote. So you know as I say things things tend to be somewhat different.

[00:13:13]Now my guess is that that probably I mean that we don't want to talk about what AI is exactly but it's you know it's it's um there are some I think inherent weaknesses of it. I mean it a technology where you cannot predict exactly what the tools will do and you cannot know when they're going to betray you, right? When they're going to uh deliver hallucinations instead of actual um actual true results. That's weird. I don't know if there's anything been anything like that. And there there has to be some you have to wonder just how widely how much can we rely will our society be willing to rely on stuff that every once in a while just decides to uh to go crazy or um or you know basically turn Frankenstein's monster on you. So it it's so that would be my guess. But it's not it's not as if there's no possible way this could these guys could make money. Well, but there is something interesting in it as well that I think you've identified that many of the things that we're identifying as bubbles actually start with a product that people want.

[00:14:19]>> Yeah.

[00:14:19]>> You know, they don't have to create their own their own markets. And the other piece of that is I certainly have heard people say exactly what you're saying that there will be a a fallout where, you know, we'll get a few good ideas out of where we are and then you can have your walled gardens around those things. But it's rare. I mean, I can think of an occasion for it. I mean, we get the Union Pacific Railroad in the in the 1860s because Congress recognizes that people actually would like to get to California, but if you actually wait for there to be enough of a market in the planes to get those railroads going all the way to California, you're going to be waiting a very long time. So, they put the money up to to create a market for those railroads. But then very quickly you get all these branch roads that lead to nowhere and end up feeding that dot that that uh that railroad boom in the 1870s that collapses. So I it does feel to me like we're with something different. You're not getting those walled gardens right now where people say, "Yeah, yeah, yeah. I really would want to get into that and I'm willing to pay for it." The way we were with iPhones, for example, or the way we were with the I remember the first time I turned on the internet. I was at teaching at MIT and they had us they made us take seminars so that we understood this new technology and I can still remember going home and saying oh my god I cannot but my my world just changed because I can do all this research and this is in the very early days. You look at the AI stuff and you think I have yet and I was an agnostic.

[00:15:48]I actually usually like new technologies. I took one look at it.

[00:15:51]I've been using it. I I started using it pretty heavily just to to see what it would do and I have become completely against it because so far I haven't seen anything that isn't crap. I am willing to admit that there are places where it is probably a good thing that you know checking um engineering plans for example. We know that there are ways in which mixing cement for example and in construction projects are much more efficient if in fact you use AI. But like you say, right now I don't see it taking off.

[00:16:23]>> Well, okay, you and I are of course not typical. Um, and let's I'll I'll break that down in a second because I think there's there's a important distinction here, but the um um what it's, you know, what I actually am using a little bit of AI for is actually um producing transcripts of videos. you you run at the video through AI to produce a a transcript, which is often, you know, hilarious in detail, but you can fix that. And um um you wouldn't believe what what AI was making of um of the word vibe session. Uh but anyway the um um but it can do certain things and and uh I I do find that there's a actually economic history often there are a lot of papers that I can that have tables and charts and um I can feed them into a sort of low-grade AI model um as a PDF and and and get the numbers out instead of having to type the numbers from the old papers. So, so you know there are there are uses even for someone like me but the um what is really and and sorry and in a lot of ways I mean the AI is it it is kind of awesome how much it manages to produce intelligible if sometimes dishonest responses to plain language questions.

[00:17:45]That is that is that is awesome given where we used to be but it's not reliable. But the main thing is that a lot of the AI and certainly what is likely to be the paying uses of AI is not coming from individuals. It's not coming from me or you or some middle manager deciding hey maybe I can use AI to do this better or maybe I I'm just going to have some fun with it. that you know there's I do know people slightly scary who are developing relationships with chat GDP GPT but the um um but the it's mostly coming from people working at businesses and large organizations who are being told you must use AI and this is something I've never seen before this is kind of coercive technology adoption where the the big money is telling workers that you must use this technology and you know one think you remember the early days of the internet the um it was joyful people loved the internet people hate AI we're now having a regular pattern of at at college commencements of speakers who start talking about AI and all the students start booing because everybody hates this and the question is how far can you go with a technology that everybody hates um so that's that's one of the things that really drives me a little you know it this is unprecedented I don't you know people the the the uh the pe people whose jobs were displaced by power looms the lites um okay they they hated the technology because they didn't like what it was doing to their jobs but people hate using AI they hate the fact that other people are using it but they are to a large extent being draoned into doing it and that is I'm not sure that I can think of a historical example like that and it doesn't seem like it's a very sustainable path forward >> so Um, Henry Ford would have something to say about trying to force people to take on new technologies. I actually saw an Edsil a few years ago. I'd never seen one. I'd always just heard about how much they were rejected. And I saw it and I'm like, that's it. They just didn't like the front grill. They're like, yeah, people just didn't like the front grill and they they wouldn't do it. Um, but that brings up another question for me. Um you're hearing a lot there were stories out today about uh companies cutting thousands of jobs because people were being replaced by AI. And I have a question for you about that. And I do I I actually then want to end with like what what does this look like for the entire society? But it certainly looks to me that as the economy slows down, that it's certainly possible that companies are letting workers go, saying it's AI and what they're really doing is they're cutting they're reducing their forces. Is it right that AI is is possibly simply being a cover for people who wanted to downsize anyway?

[00:20:40]>> Well, there's some of of both. I mean, if if you are a company that wants to lay off workers and actually if you're a company that wants to in effect um uh increase the workload on on a smaller number of workers, then uh AI is a great cover story. You can say, "Oh, we're doing this because of the wonders of modern technology, and by the way, uh we expect you to in effect uh put in 10-hour days." Uh so but I think this is real but the question really is more uh we keep on getting stories of companies that have laid off a lot of workers um saying that AI can do the job and then finding out that no it can't. And now I I don't think this is all fake. I mean um if we say that AI is mostly just kind of doing routine stuff or it's delivering you know in effect um um some some of my uh some of my friends who actually work on this Henry Farrell they they say that AI is a social technology.

[00:21:41]It's really what it's doing is it's it's basically elomerating what lots of people have said and it's it's delivering back to you what a lot of people uh who know something about a subject would say if asked the question you ask and it's not re it's not understanding there's no there's no mind there but it is kind of delivering a kind of aggregated standard response and a lot of jobs are like that you know if you're if you're talking to uh to the help desk at a call center somewhere uh thousands of miles away. It's the person that you're talking to if it is a per if there's an actual person may very well be there with a three- ring binder looking for what what am I supposed to say? And AI can replace that job. Uh AI is basically doing this much the same. Yeah, a lot of people are doing fairly routinized standardized just the common opinion of common opinion um responses to things as as a way of doing their jobs. So that that's real. So it's not just that AI is an excuse but again the excuse I mean there's nothing uh we always see this right businesses uh to the extent that businesses care either what their workers feel or what their customers feel um stuff happening external excuses or is a big I mean this is this is the story of greedflation you know that companies um may raise prices when there's um an energy crisis not because their actual costs have gone up but because whatever everybody raising prices who will notice if I get greedy.

[00:23:11]So there's something like that on AI and jobs as well don't know. I mean and again we there are enough stories now of companies that have laid off all of their experienced professionals for AI and it turns out well you know the experienced professionals uh could actually deal with questions that were not routine could actually didn't didn't hallucinate and um and so they're they're finding that they made a mistake but this is it's amazing how how little we know about how this works. I I don't you know I don't reme I I don't remember there being so much uncertainty about what you could actually do with the internet. Uh and of course I don't have any memory of what people thought you could do with railroads. But I don't think that I think this is kind of unprecedented as a as a um this massive technology that we're investing trillions of dollars in and still nobody quite knows how it works or what it will do. Well, and that's, you know, I want to end with with what my real question is, that is, um, if I'm correct and and these people I'm reading are correct about it looking like a real bubble.

[00:24:21]>> Yeah. um to what degree and again the reason I use the comparison of the 19th century railroads and or you could do the 19 the 20 1920s with cars the investment in AI in data centers in um in hiring practices certainly in investments certainly in Nvidia and all these different places that are tied into that specific technology what does it look like if that bubble bursts Now, I've heard, you know, from a lot of people, you included, that we're going to get some good technologies out of it no matter what happens. And I agree with that. We always do. But with all the pressures that are on the American economy right now, um I'm worried. And should I be?

[00:25:07]>> Yeah. Let me give you um good news sort of and bad news definitely. Um the the good news sort of and I say sort of for a reason you'll um is that uh you on the face of it if you just look at the scale of the AI investment that you say that's that's the whole economy that's driving all of our economic growth but it turns out that an awful lot of the AI spending is actually imported tech gear. It's actually imported chips and and computer equipment and so on. So that if the AI bubble bursts, uh, a large part of the bubble bursting will actually be a full in US imports rather than than a a big be a big shock to the domestic economy, but not nearly as much as you might think. That's one of that there's been a back and forth about, you know, how much of recent growth is just AI. And it turns out that that just looking at the raw number misses the high import intensity of the stuff. So it's not in some ways this is a shock to the world economy and not so much to the US economy specifically. So that's I guess that's kind of good news not so good for other countries but you know they Taiwan is experiencing enormous economic growth because of all the chips they're selling to US AI companies. So you know this is there a lot of the bad news will end up showing up in Taiwan rather than in the US. Um the uh the bad news the dotcom again and this would been true of railways as well the dotcom the dotcom bubble in terms of the actual money laid out the really big stuff was telecoms rather than dotcoms it was the the telecommunications companies investing especially in fiber optics putting laying down tremendous amounts of of fiber optic cable which stayed unused for a long time there was lots of dark fiber after after the dotcom bubble burst But it was still there. Fiber optic cable doesn't doesn't depreciate rapidly. It was still there in the ground and eventually got used as we found. So it was a lot of useful investments. As I understand it, these data centers that are being built, the hey investment in chips, the investment in software, um this stuff will depreciate physically pretty fast. It will become outmoded uh pretty fast. So I think there's likely be a much higher proportion of just wasted investment that never finds a use out of this boom than there was out of the last tech boom. So not so great. And especially by the way and just you know last thing the Chinese are um they are taking a very different approach. They're building much more limited models that just don't use as much information, but get a high fraction of the performance, use a lot less energy. If the world ends up going to that model of AI instead of the all-encompassing ones, then we will have just wasted the money.

[00:28:05]We will have spent a lot of money on on building um uh super impressive stuff that nobody actually wants to use. So there's a chance that that we may have an exceptional much more real railroads, you know, obviously rail the railroads still had railroads. Um you could you could use the tracks later on the you could use dark fiber, you could um actually you know the the the sort of the I think the original boom that looks something like this was in fact the British canal boom uh uh around around 1800. Um, all of those left usable legacies and this one might not.

[00:28:43]>> Wow. Fascinating. Absolutely fascinating. Um, thank you. It's always a pleasure to talk to you.

[00:28:49]>> Good to be on. Uh, let's do this again.

[00:28:52]>> Yeah, I think we're on the schedule and well and thank you everyone for coming.

[00:28:56]I I'm afraid I don't know what the schedule is, but we will post it on it.

[00:29:01]>> And enjoy the rest of your vacation.

[00:29:04]>> Thanks.

[00:29:05]When do you come When do you come back?

[00:29:07]When do you go back to work?

[00:29:09]>> Uh well, I'm I'm working.

[00:29:11]>> I know. I know.

[00:29:12]>> Uh no, I'll be back in the in the States in uh early next month >> and but no, it's uh you know, it's academia as you know. It's a I actually have uh Are we We're not live anymore, are we? Yes, we are.

[00:29:29]>> Yes, we are. So, I'll just tell you that I I you know I have I have academic I have academic connections and stuff. Um I'm about to go on uh on Dutch TV. I guess I can tell you that much. And uh so which is why I'm not uh as casual as I usually am. And um the um so I'm not sure we're getting some vacation in. Uh Rob and I are going to be uh being in the Netherlands. We're probably going to be uh going on a bike trip with friends tomorrow. But So there there's there's there's plenty of stuff, but it's uh yeah, it's uh um it's it's nice to have some distance, although I am I'm I'm I am missing New York at this point.

[00:30:12]>> Well, you No, you're not because I was just there. And it's hotter than the shades of Blue Hades. Once it comes once the temperature comes down, you'll like it again. It is beautiful, but boy, it went I I I was in Boston yesterday. It was 96 degrees.

[00:30:26]>> God.

[00:30:26]>> Yeah. But anyway, all right. have have a wonderful rest of your vacation and we'll talk soon and thank you everybody for being There.