The Wolf-Krugman Exchange: AI hype vs reality

You can enable subtitles (captions) in the video player

MARTIN WOLF: I hope you've been enjoying these, by the way.

PAUL KRUGMAN: Oh, it's been fun.

MARTIN WOLF: I don't think I've ever met anyone who, certainly not in his 70s, who sustained the work output that you have in the last few months.

PAUL KRUGMAN: It's true that there were a few periods when I kind of now regret spending quite so much time on the newsletter and less time drinking Aperol spritzes in the Piazza when we were in Italy.

MARTIN WOLF: So let's go and talk about artificial intelligence, which is a nice change. This is the fourth in our series, The Wolf-Krugman Exchange. I'm Martin Wolf, Chief Economics Commentator at the Financial Times.

PAUL KRUGMAN: And I'm Paul Krugman, professor at the City University of New York, author of an independent Substack newsletter. Today's episode is being recorded on Friday, June 20 at 10:30 in Massachusetts, which I'm actually not in New York right now, which is 3:30 in the afternoon over in London.

MARTIN WOLF: And we're recording it on Friday because next week I'm going to be in India. And the only rational explanation for this, I'm going to be in Delhi, is that the 30 degrees centigrade-plus temperatures that we're experiencing here in London will-- there, instead, I'll get a properly hot day, I imagine, around about 45.

PAUL KRUGMAN: Yeah, well, I think I'll keep busy. And I am someplace which is marginally cooler.

MARTIN WOLF: So we decided that this week we would look at artificial intelligence, partly because it allows us not to spend our whole time talking about what's going on in the US right now. And so we will look at artificial intelligence itself but also how its impact is beginning to spread through our economies and our lives, and what its longer term implications might possibly be. It's certainly the most interesting technological change we can see right now.

So, Paul, when you think of what is now being called artificial intelligence, or as I read today in the FT, one expert refers to these technologies not as artificial intelligence but stochastic parrots, which I think is a lovely description. Anyway, whichever description you want, what excites you, what disturbs you about this phenomenon?

PAUL KRUGMAN: What we're calling artificial intelligence really isn't at this point intelligence. There's an endless dispute about whether it may be about to become something that you might really call that. But really, this is an evolution of large language models, of basically taking in tonnes and tonnes of data, applying very clever algorithms, so clever we don't quite understand how they work, to be able to answer in natural language questions posed in natural language. And it's not a minor thing. There's a bunch of areas where we used to joke about how bad attempts to automate thinking, or something that looked thinking really were.

Translation was a joke. I don't know if-- the old anecdote about the supposed Russian-English translation programme that took the spirit was strong, but the flesh was weak. And it came back as the vodka was good, but the meat was spoiled. So it used to be the translation was a joke. Now it's actually quite good. You can actually-- I can read foreign language news articles, and they may be slightly stilted. But they're really very, very good. Recognition of speech is quite good. So we certainly have achieved something major. But whether it is truly revolutionary, what it's going to do, that's up in the air.

MARTIN WOLF: So I have a couple of reactions to that. I remember one very much supporting you, that there's this famous notion of the Turing test, obviously, from Alan Turing, the great theoretician of computing from the 1930s, with John Von Neumann, a sort of father of computing. And he argued, we think that computing was intelligent if we could have an exchange with that computer.

And it sounds like a human being, it feels like a human being, so that people are fooled into thinking it's a human being. And as far as I can see, they've passed that test. So that's something quite significant. But I would also say it's quite interesting, this question of how we define intelligence. To me, the most exciting use, and I get this very much from somebody I've got to quite well in this area, Demis Hassabis, who's at Google DeepMind.

What he is excited by is the capacity of computer programmes now to do really deep scientific work. And they got the Nobel Prize for the ability, which is a computational problem, but obviously, fantastically complex one, to work out all the different ways proteins could be folded. I don't know whether that's intelligent, but at least it's very, very easy to see. It's very, very powerful and useful. And human beings just couldn't do it on their own.

PAUL KRUGMAN: Yeah, there's a lot of tasks that we have regarded as being tasks that required very smart people, that were very highly paid and they were really difficult that can now be done by whatever it is we're calling this thing, generative AI or whatever, stochastic parrots, but they're stochastic parrots that produce really useful stuff. And this is significant.

Now, whether it's the same thing as what-- it's clear that the Turing test, Turing screwed up. Hard to say that, but Alan Turing was kind of wrong about what would be involved because we clearly now have programmes that pass the Turing test quite easily. And yet, we don't think that they're people. Nobody really thinks that they're people yet.

On the other hand-- well, I'm not sure how many hands I've already used here, but radical improvements in productivity in something. That's an old story that's happened repeatedly in many parts of the economy. And is this one really different, or is this just hitting an area that has not previously been much touched by technology?

MARTIN WOLF: Well, this is obviously the big question. I think it's an interesting question of whether Turing got it wrong or whether he actually had a perfectly plausible view. But we don't actually feel that the machines that can do this are, in fact, people. So we're even more suspicious of machines than he thought we would be. But anyway, let's leave that to one side. Let's go back to this history you talked about because, as you said, going back to the Luddites in the early 19th century, the Luddites were a movement of workers against the introduction of machines in the early 19th century.

And their skill was weaving using power looms. And they were seeing that replaced by new machines, pretty well every major technological revolution. In that case, machinery really was dramatic, what it could do and how many jobs it got rid of. And if you think of the history of machinery and other innovations, every time people have said, well, all the jobs will be destroyed, we'll have mass unemployment.

And after a while, we've had an adjustment processes. We found new ways to spend our incomes in different areas. And it ends up with just a whole new set of jobs, which nobody imagined. So if you told anyone in 1800 that nobody would work on farms, essentially, when that was overwhelmingly the biggest industry in the world, they would have said, what? So what did you all do? Well, if we listed all the jobs we now do, they would have no idea what they were.

PAUL KRUGMAN: That's right. Or they're things that people did but that were marginal. But as you get richer and as you can do the old stuff very efficiently, you discover that, well, OK. Let's do more of those other things. We have an awful lot of people employed in health care now. I don't actually what the numbers are, but we may very well have more yoga instructors than coal miners at this point in America. So we do different things. And the history of predicting mass unemployment from technology is very, very long. I've been even personally watched repeated episodes.

There was a whole stretch in the '90s when everyone was sure that mass unemployment was just around the corner because we were deindustrializing. There was a lot of predictions of mass unemployment in the early 2010s and constant disbelief that periods of high unemployment could actually be just because we have insufficient aggregate demand. Macroeconomics just doesn't appeal to people intuitively, and technological unemployment does. And yet, it never really seems to happen except on a very localised basis.

MARTIN WOLF: I think this is clearly right. In the '50s, roughly 40% of the British labour force was employed in industry, overwhelmingly manufacturing. Now it's about 10%. And if you told them that, this is not so long ago, that could happen and they would actually be employing a higher proportion of the overall population because all the women are working too, they wouldn't have believed it. So let me, however, play devil's advocate just to see how this works out.

If these new programmes are able, and that's a very big question, to do a huge proportion of the analytical thinking work that we now do, which one would also might think is the core activity of human beings, at least for human beings like ourselves, but thinking, creating launches so much of our activities.

If machines do all this basic analysis, maybe we'll decide that actually it would really be much better to have a computer as a judge in a court because computers are completely reliable. They're not going to be emotional. There's not going to be the famous effect, which is explored in social science, that judges in the morning behave quite differently from judges in the afternoon.

So we could imagine a world in which we decide, well, really, wouldn't we rather our president were a computer? So many mistakes will be avoided. We are, in addition, having clearly a very significant robotic revolution underway. Isn't it possible that what we're going to lose here-- first of all, there's really going to be a vast amount of employment that's going to be effective. So even if we do find jobs, that's unimaginably different. And isn't it possible also that we're going to find the marginal product of a very large part of the labour force isn't much above subsistence because we don't really want them for anything?

PAUL KRUGMAN: All of that is possible. History would say probably not because it just hasn't happened before. And again, this stuff goes back forever and ever. Ricardo, in the third edition of his Principles of Political Economy, worried about unemployment due to machinery. And this is 1819 or thereabouts.

MARTIN WOLF: Famously.

PAUL KRUGMAN: So right now we think of the important stuff, the stuff that-- the really good jobs. Is analytical thinking judgement? Maybe-- I could be wrong, but I think we're quite a ways from robot plumbers, that we're quite a ways from having a lot of what we think of right now as being relatively mundane things that require no more than common sense. But common sense is actually one of the things that AI appears to be quite bad at. And it's something that people are quite good at.

So I can argue this either way. One version of what we're calling AI is it's just a souped up version of autocorrect. It's filling things in based upon what other people have done. And then you can say, yeah, but aren't an awful lot of jobs that real people do and earn fairly high salaries doing basically souped up autocorrect? Which is also true. History always says that we find other stuff to do. And so far, the successful applications of AI are fairly limited. So far, we're certainly not seeing a productivity surge commensurate with what people are saying.

MARTIN WOLF: So just focus on what we know from past experience about the adjustment process. So there's very, very famous work, which I think you've cited frequently and others have going back to the introduction of electricity, which was obviously one of the great general purpose technologies of the second Industrial Revolution. Changed everything, really everything. And the case turned out to be that it was transformative. It did change everything, but it took about 40 years to do so before it got into the factories.

They redesigned factories. They started developing all the clever motors you could put in everything that would refrigerate the houses and do the washing and all the rest of it. It's just a long, slow process in which the adjustment in the labour force and the new jobs come along. And the fact that we're seeing this implemented quite slowly, but it's still very early stages, might suggest we're going through a similar process. And the effects will be very large, but they will be bigger than we think, as it were, for many things. But they will also take longer than we think. Do you think that's a plausible way of thinking about the future?

PAUL KRUGMAN: In principle, that should be my view. There's a wonderful paper, old paper by Paul David--

MARTIN WOLF: Indeed

PAUL KRUGMAN: --about why was information technology not showing up in the productivity numbers? I think it was called The Computer and the Dynamo. And his point was that it actually did take around 40 years for businesses to figure out what to do with electricity because it requires-- it's not just understanding. But you actually have to redesign the way you do work.

MARTIN WOLF: Yes.

PAUL KRUGMAN: An old-style factory is a six-story tall mill with a steam engine in the basement and very cramped corridors because you're trying to minimise power loss. And it's actually very awkward to work in. And you replace that with electric motors with a big, sprawling one-story building with wide aisles. But you have to change everything. You have to change where you locate, how you organise work. Everything gets affected. So this is the story, and many of us have invoked that story to explain why technologies don't transform things as fast as you think they're going to.

I have to say that my personal impression is that we're actually seeing on AI is not that story. What we're actually seeing is a rush to implement AI before actually it's been proved that it's useful, that there's this enormous fashionability of putting AI. I'm finding that stuff that I use routinely, search engines, have actually been degraded because the companies involved are so eager to be there on the AI. And I have to put in extra work to turn the damn stuff off--

MARTIN WOLF: Indeed.

PAUL KRUGMAN: --so I can just get a plain, ordinary search result. So I wonder whether this time around we're not seeing instead something like a rush to be part of the wave of the future before we're actually even sure that it really is the wave of the future.

MARTIN WOLF: You might argue that in the cases you mention, electricity, but even with the computer, originally, initially it was a work tool. Businesses reorganised themselves. It's a bit closer to AI. But it involved a lot of reorganisation, quite deep reorganisation to make the mainframe replace all your clocks, for example. You had to think about work processes in a profound way. And that may happen here. And with electricity, as you point out, it meant changing every factory. But here people, I think, are thinking it's cheap, from our point of view. It's there.

It seems to be able to answer the sorts of questions we used to ask our law clerks or consultants of a medium grade. So why not ask them these questions? They do quite well. So we are rushing into it. But it doesn't seem yet, maybe it's just very early days, that we are seeing mass unemployment-- I don't know whether it's different in the US. I haven't looked so closely-- of the people who are working in these sorts of activities. Though, I do hear, and I have read, that there's a very significant reduction in quite a number of economies in graduate recruitment, which might be affected by this. I don't know.

PAUL KRUGMAN: I have actually just written about it just before we had this conversation. There has been a dramatic fall off in job opportunities for new college graduates.

MARTIN WOLF: It seems to be happening in China too. I don't know whether it has anything to do with this.

PAUL KRUGMAN: Yeah, and the trouble is, it is so sudden. This is really-- there's been a downward trend in this employment advantage of having a college degree that's been going on for some years. But this abrupt surge in unemployment among recent college graduates and this apparent virtual collapse of jobs this year makes you wonder, is that really the technology, or is it something else? And unfortunately, there are a few other things going on in the world, like the US going wild on tariff policy, that are also probably affecting this.

So we don't know. But for there to be significant dislocations, and even quite quick dislocations, is certainly possible. That doesn't tell you very much about what the long-term effects are, but the idea that we could be seeing a really rapid change in elimination of whole categories of jobs on a fairly short time frame, maybe. Although, again, I read the news stories, and I never quite how much is hype and how much is reality.

MARTIN WOLF: I think that's right. And the processes of this kind-- AI is, after all, really quite new. The businesses are very excited about it, as you rightly said. But I think most of them don't really know what to do with it and how far they should trust it. So there seemed to be two sorts of views out there. One is that it's going to end up as a complement to skilled people. It may remove some of the middle grade analysts and so forth that we've had. But AI plus highly skilled humans will still be the best way of operating.

It will change the structure of employment, but human beings will be very actively involved in most of the tasks they still do. Or actually, we will find over time that if you're going to be treated by a doctor, well, the principal analyst of what's going on and diagnostician and all the rest of it is actually going to be one of these AI programmes. And that would reduce, or at least profoundly change, the relationship of us to the people in charge, as it were. But my impression at the moment is there are quite a lot of very different views among experts on how it will play out.

PAUL KRUGMAN: Yeah, this has been one of those subjects where I've tried to talk to people, people who really do pay attention in a way that I can't, and have come to the conclusion that anything that I want to believe about the prospects of AI and its economic effects, all I need to do is do a little searching, and I can find some expert who will tell me whatever it is I want to believe. It's one of those situations where there's just such a range of possible interpretations. And I'm not saying that these people are dishonest or anything. It's just that it's really unknown at this point. And there's so little actual experience.

MARTIN WOLF: One of the problems that I do have some connection with, partly because I have grandchildren and partly because of what my wife does-- she's an expert on skills policy and universities, academic expert on these things-- is to question, OK. We don't know what's going to happen. But should we be already thinking about how we should teach people, what children should learn? Because I suppose you might feel, well, they should learn something different.

Or is it the case-- I think we obviously don't know-- but is it the case that, actually, the sorts of things people learned how to do because it was just a way of developing the mind, writing essays, doing analytical work, doing equations, all the rest of it is still the best way of training human beings. Then we'll see what happens when it comes along.

PAUL KRUGMAN: Yeah, a few years ago, the slogan for young people was learn to code because that was clearly the future.

MARTIN WOLF: That's not very good advice anymore, is it?

PAUL KRUGMAN: No, it turns out that one of the things that AI is pretty good at is writing code, not presumably the highest end, most sophisticated, but basic code that could get stuff done is one of those things that you can turn over to the software. And so that was a really bad advice. Other things we don't know. It's kind of wild. I actually-- let me give you an idea of the kinds of things that make me sceptical.

So there was a big announcement by Amazon that it expects to get rid of a lot of workers thanks to AI, which sounds fine, except that I have actually done a little bit of work on Amazon as a business. And Amazon is one of those things. There's an illusion that it's untouched by human hands. You just click on something, and stuff magically appears at your door.

MARTIN WOLF: [INAUDIBLE]

PAUL KRUGMAN: And of course, what it really has is it has 1.1 million workers, mostly in distribution centres and warehouses, moving stuff around. And how is AI going to replace-- eventually maybe, if we have robots who can do that, maybe. But at the moment, it's not at all clear how ChatGPT or something like that is going to replace those. So is this just hype? Is this like-- there was a period a few years ago when everybody out there was putting blockchain in their name as a way of making them seem cutting edge. And is this comparable just hype rather than reality?

MARTIN WOLF: I think that's a really, really interesting question. I was always intensely suspicious of blockchain and cryptocurrencies and all those things. I wrote about it. So I'm very cheered up that it doesn't seem to have amounted to much. Though, it does seem to have played a big part in buying the US presidency. But anyway, we should leave that for the moment. Let's think of some of the more concrete aspects. Let's suppose there is a sizable labour market adjustment.

Yeah, one of the points that David Autor made in the podcast I had with him, which I thought was a very good one, is that, unlike what he called the China shock, which basically just the rapid collapse of quite a number of manufacturing businesses, and therefore factories, located in very specific locations across industrial countries, which obviously created a big adjustment problem because it tended to create a big shock to very specific locations. And they lost their tradables output. And we've discussed that already. The good thing about AI is it looks as though it's the technology, like the use of computers, which mostly will have a broad effect but not a very concentrated effect. So it should be one, in principle, we can adjust to relatively easily.

PAUL KRUGMAN: It is difficult to come up with examples of highly concentrated, geographically concentrated industries where AI will take away the jobs. You might worry a little bit about, actually, of all things, New York and London. How many of the jobs being done by people in the finance industry can be automated? So the localities at risk might actually not be some small town producing furniture but some major financial centre that really doesn't need all of these guys poring over spreadsheets anymore. But it's probably-- this is much more of a general purpose technology.

Although, even there, going back, we were talking about electricity. One of the effects of electrification was that factories, when you shifted to sprawling one-story factories, they moved out of city centres. And that was actually quite disruptive. It did eliminate a lot of the blue collar jobs that used to serve people in the inner cities. But yeah, this is probably-- we're all speculating. Everything's speculation. But it's probably not something where you say, oh, everybody who works in Bradford, Yorkshire is going to lose their jobs to AI.

MARTIN WOLF: I think you've got there something quite important because if you do think about it and if you think who might be most affected plausibly-- and I leave aside the robotic side. So I've convinced myself that the safest job in the world is probably gardener. But assume that we really do displace a lot of white collar jobs, a lot of the sorts of jobs that young graduates do, the sorts of jobs done by legal assistants, even junior lawyers. This is broadly the group of people whose jobs have expanded enormously in the last three or four decades.

And that's partly why we've had this enormous expansion in universities, less so in the US because you already had such a huge university system. But in Britain I often mention this. When I went to university, 5% of the generation went to university. Now it's 40%. And it's because these jobs have expanded so much. So if you have a lot of very, very unhappy, educated people expecting a better life than they're going to have, and many of them are already pretty unhappy, it does seem to me that this could have quite-- if it happens-- really quite difficult social and political effects in societies that are already suffering from those effects.

PAUL KRUGMAN: Yeah, the Luddites, always worth remembering, the Luddites were not unskilled day labourers.

MARTIN WOLF: They were elite. They were the skilled elite, absolutely.

PAUL KRUGMAN: Although, can I say there-- give a slightly, maybe slightly optimistic take. While yes, we're going to have a lot of unhappy people. On the other hand, in some ways AI may be an equalising force. I grew up in the US of the '60s, where skilled, blue collar workers earned incomes, it seemed like they earned incomes not very different from, say, middle managers.

In fact, I grew up literally on a street where some of the people on the street were plumbers, and my father was a middle manager. And that completely changed. Maybe we go back to that. Maybe we go back to a situation where people who can actually deal with the material world become appropriately valued again. And people who push symbols around find that, well, computers can also push symbols around.

MARTIN WOLF: Let's talk about just the social and political dimensions. How plausible is it-- I've just been writing about copyright in AI. But this is a broader question. You've written quite a few pieces recently about inequality. We are seeing enormous concentrations of wealth and income in our societies, in the US particularly. And some people have referred to this as techno feudalism.

Do you think that if this AI revolution continues, and that's clearly what the companies that dominate hope, and you mentioned some of them, we're going to find this sort of extreme concentration of power and influence and money in the tech elite that are driving this relatively small number of companies proceed even further? And how worried should we be about that because it doesn't look very healthy to me?

PAUL KRUGMAN: Yeah, I'm actually not quite sure how AI plays into this. And I'm actually, as we speak, working on this. And it seems to me that the defining feature of a lot of the technology-- that the reason that we have these immense fortunes-- and it really is true that at this point the top ranks of wealth are very much dominated by tech bros.

MARTIN WOLF: Yes.

PAUL KRUGMAN: It really is. Basically, there's Warren Buffett, who actually seems to have a genuine unreproducible skill. And everybody else, we try to understand why do we have these giant fortunes in tech. Why are there a handful of tech bros with this enormous amount of money? It really is very much about network externalities, which you can [? almost ?] jargon. But it basically means that you do something or you use something because everybody else does. Everybody uses Amazon because everybody else uses Amazon, and it's very much easier to get regular stuff.

Or for that matter, I'm still doing a lot of work in Excel, which is crazy, but Excel is universal, and everybody knows how to use it. And these things are these self-reinforcing, self-locking in advantages, are the basis really of a different kind of monopoly power, very different from the kind of monopoly power you had in the Gilded Age. But it's monopoly power all the same. And it gives rise to a handful of incredibly large fortunes. It's just very difficult to break into that. And does AI reinforce that tendency? Possibly.

Does it, on the other hand, make it easier-- I don't know. Maybe an AI model will make it easier to get stuff quickly on demand from some smaller-- I actually have no idea. I don't think people have actually tried to work out that consequence. I think people have tended to say, well, it displaces workers. Therefore, it must enhance the power of the corporate bosses, which it might. But we don't know that. I think the corporations like the idea of not having to actually deal with workers. But that may be, again, part of the hype.

MARTIN WOLF: Yes, it's a very interesting question of how that plays out first within the AI sector itself. I was very interested, because it happened very recently, in this sudden emergence of this Chinese company, DeepSeek. And that seemed to suggest that the idea that there were infinite economies of scale and scope in the AI industry itself might not be right.

Obviously, we'll see whether that's true. And then, of course, there's the question of what effect it has on the users, the industries that use it. And it's pretty clear at this stage, we have no idea. But right now, the firms that have the resources to do the colossal investments that at least the Americans are pursuing are relatively limited because the investments are so stupendous.

PAUL KRUGMAN: Yeah, and these things do-- these are not-- AI, what we're calling AI, certainly doesn't function anything like human intelligence. What it does is it scoops up vast quantities of data and does very, very complex calculations on that data, which is a big, big upfront investment. In a peculiar way, it's information, but it actually seems to require a lot of physical capital, giant server farms, huge amounts of power consumption. So this may actually be something that favours not so much technological dominance as it favours basically people with lots of money to invest in largely physical capital.

MARTIN WOLF: But at the moment, for the reasons of history, the people who have the know-how of how to invest it and the money are already established players in the tech industry, relatively limited-- very limited number of major firms, which, interestingly, doesn't include some of them. The most valuable of the tech companies for quite a while was Apple. But it doesn't seem to be a significant player in this at all. And some of them, OpenAI is obviously a new player, but it's got Microsoft linked to it.

But it does look at-- this is one of those things in which incumbents seem-- or some incumbents seem to be incredibly well positioned to expand their reach further. And that's why people are concerned about this notion that there will be a sort of feudal lords over us all. They did certainly play some part, or some of them, in this election, in the recent elections. So it links up with this idea that, at least in the US, politics is becoming a plutocratic sport. And so it links up with also future of democracy.

PAUL KRUGMAN: Although, we should say that the really big money, apparently accounting for something like 40% of corporate spending on the election, was crypto.

MARTIN WOLF: Yes.

PAUL KRUGMAN: And I'm highly uncertain about what the economic payoff to AI is, but I'm quite certain about what the payoff to crypto is, which is nothing. But unfortunately, it turns out to be able to buy a government.

MARTIN WOLF: Yes, extraordinary bubbles all of themselves have remarkably distorting effects for a while.

PAUL KRUGMAN: Yeah, and it's going to be an interesting question, actually. How much, though-- coming back to this techno feudalism-- how much is the power of incumbency versus just being able to deploy very large amounts of capital? And I think we're going to find that out.

MARTIN WOLF: So let's do the cultural coda.

PAUL KRUGMAN: OK. So I think I'm the leadoff on the cultural coda. And it is some music. It's Loretta Lynn singing Coal Miner's Daughter, which was also the basis of a wonderful old film, which I think is-- what on Earth does this have to do with it? But in fact, there are basically no coal miner's daughters anymore. Coal mining was more than half a million workers in the United States in the immediate aftermath of World War II. By 2000, coal production was actually higher in 2000 than it had been in the 1940s. But 85% of the workers were gone.

And what was that about? It was all about technology. First, strip mining and then blowing the tops off mountains to get at the coal, which meant that you didn't need a whole lot of workers, which is showing that you can get massive displacements of particular kinds of workers by technology. We did not suffer mass unemployment because of the disappearance of the coal industry.

We did suffer a lot of changes. Some places were hurt, but also, ways of life disappeared. So I think, in some ways, I like coal as an example of just how-- first of all, of how much technology really can change things, but also that the latest, fanciest technology is not the first time we've seen this movie or the second or the third or the fourth. This has been happening again and again over the past couple of centuries.

MARTIN WOLF: My memory is that in Britain, the coal mining industry at its peak employed about a million, which is an extraordinary number for a much smaller country. And my view of the disappearance of coal mining was set by the very famous description of George Orwell of what it's actually like to be a miner. And I decided we should be very happy. Now, my cultural coda, this week, it's a novel. And I think, to me, it's the most important novel of the 20th century, at least the most revealing. It's Thomas Mann's The Magic Mountain.

It was published in 1924, and the '20s and '30s are a period, I think, more and more about. And the core of the book was the intellectual ferment going on in the first half of the 20th century between old fashioned, staid, civilised, liberal humanism, embodied, in this case, in the figure of a man called Settembrini, putting forward the sort of views I hold, and I'm beginning to feel are equally old fashioned. And on the other hand, Naphta, who is a Marxist revolutionary. But actually, when you push him, he turns out to be very similar to the far right revolutionaries.

And really, the difference between Hitler and Stalin turned out to be pretty small when all things are done. And it gets you this sense of profound conflict. And Mann puts forward the idea that the first World War was the beginning of the destruction that followed from this. It was written during the war and published shortly afterwards. So it's extraordinary how a book-- novel, published hundreds years ago, can be so brilliant at describing the sort of things we are seeing right now and the challenges we are now seeing between pretty feeble, liberal, humanist-type people and passionate authoritarians.

So thank you very much for joining us for the part four of the Wolf-Krugman Exchange, something very different on AI. We'll be back with you again next week when we will be discussing the ways in which the economic system has changed as a result of recent events, possibly forever. Is this time, perhaps, really different? And if you're looking for interesting ways to fill your time before then, I've recently put together a selection of what I think are the most interesting economics books for your summer reading, the books published in the first part of this year. And a link to that list will be in the show notes.